2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
12 #include "cache-tree.h"
15 #include "object-store.h"
19 #include "resolve-undo.h"
20 #include "run-command.h"
23 #include "split-index.h"
25 #include "fsmonitor.h"
26 #include "thread-utils.h"
28 #include "sparse-index.h"
29 #include "csum-file.h"
31 /* Mask for the name length in ce_flags in the on-disk index */
33 #define CE_NAMEMASK (0x0fff)
37 * The first letter should be 'A'..'Z' for extensions that are not
38 * necessary for a correct operation (i.e. optimization data).
39 * When new extensions are added that _needs_ to be understood in
40 * order to correctly interpret the index file, pick character that
41 * is outside the range, to cause the reader to abort.
44 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
45 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
46 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
47 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
48 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
49 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
50 #define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945 /* "EOIE" */
51 #define CACHE_EXT_INDEXENTRYOFFSETTABLE 0x49454F54 /* "IEOT" */
52 #define CACHE_EXT_SPARSE_DIRECTORIES 0x73646972 /* "sdir" */
54 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
55 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
56 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
57 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
61 * This is an estimate of the pathname length in the index. We use
62 * this for V4 index files to guess the un-deltafied size of the index
63 * in memory because of pathname deltafication. This is not required
64 * for V2/V3 index formats because their pathnames are not compressed.
65 * If the initial amount of memory set aside is not sufficient, the
66 * mem pool will allocate extra memory.
68 #define CACHE_ENTRY_PATH_LENGTH 80
70 static inline struct cache_entry
*mem_pool__ce_alloc(struct mem_pool
*mem_pool
, size_t len
)
72 struct cache_entry
*ce
;
73 ce
= mem_pool_alloc(mem_pool
, cache_entry_size(len
));
74 ce
->mem_pool_allocated
= 1;
78 static inline struct cache_entry
*mem_pool__ce_calloc(struct mem_pool
*mem_pool
, size_t len
)
80 struct cache_entry
* ce
;
81 ce
= mem_pool_calloc(mem_pool
, 1, cache_entry_size(len
));
82 ce
->mem_pool_allocated
= 1;
86 static struct mem_pool
*find_mem_pool(struct index_state
*istate
)
88 struct mem_pool
**pool_ptr
;
90 if (istate
->split_index
&& istate
->split_index
->base
)
91 pool_ptr
= &istate
->split_index
->base
->ce_mem_pool
;
93 pool_ptr
= &istate
->ce_mem_pool
;
96 *pool_ptr
= xmalloc(sizeof(**pool_ptr
));
97 mem_pool_init(*pool_ptr
, 0);
103 static const char *alternate_index_output
;
105 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
107 if (S_ISSPARSEDIR(ce
->ce_mode
))
108 istate
->sparse_index
= 1;
110 istate
->cache
[nr
] = ce
;
111 add_name_hash(istate
, ce
);
114 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
116 struct cache_entry
*old
= istate
->cache
[nr
];
118 replace_index_entry_in_base(istate
, old
, ce
);
119 remove_name_hash(istate
, old
);
120 discard_cache_entry(old
);
121 ce
->ce_flags
&= ~CE_HASHED
;
122 set_index_entry(istate
, nr
, ce
);
123 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
124 mark_fsmonitor_invalid(istate
, ce
);
125 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
128 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
130 struct cache_entry
*old_entry
= istate
->cache
[nr
], *new_entry
;
131 int namelen
= strlen(new_name
);
133 new_entry
= make_empty_cache_entry(istate
, namelen
);
134 copy_cache_entry(new_entry
, old_entry
);
135 new_entry
->ce_flags
&= ~CE_HASHED
;
136 new_entry
->ce_namelen
= namelen
;
137 new_entry
->index
= 0;
138 memcpy(new_entry
->name
, new_name
, namelen
+ 1);
140 cache_tree_invalidate_path(istate
, old_entry
->name
);
141 untracked_cache_remove_from_index(istate
, old_entry
->name
);
142 remove_index_entry_at(istate
, nr
);
143 add_index_entry(istate
, new_entry
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
146 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
148 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
149 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
150 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
151 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
152 sd
->sd_dev
= st
->st_dev
;
153 sd
->sd_ino
= st
->st_ino
;
154 sd
->sd_uid
= st
->st_uid
;
155 sd
->sd_gid
= st
->st_gid
;
156 sd
->sd_size
= st
->st_size
;
159 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
163 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
164 changed
|= MTIME_CHANGED
;
165 if (trust_ctime
&& check_stat
&&
166 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
167 changed
|= CTIME_CHANGED
;
170 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
171 changed
|= MTIME_CHANGED
;
172 if (trust_ctime
&& check_stat
&&
173 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
174 changed
|= CTIME_CHANGED
;
178 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
179 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
180 changed
|= OWNER_CHANGED
;
181 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
182 changed
|= INODE_CHANGED
;
187 * st_dev breaks on network filesystems where different
188 * clients will have different views of what "device"
189 * the filesystem is on
191 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
192 changed
|= INODE_CHANGED
;
195 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
196 changed
|= DATA_CHANGED
;
202 * This only updates the "non-critical" parts of the directory
203 * cache, ie the parts that aren't tracked by GIT, and only used
204 * to validate the cache.
206 void fill_stat_cache_info(struct index_state
*istate
, struct cache_entry
*ce
, struct stat
*st
)
208 fill_stat_data(&ce
->ce_stat_data
, st
);
210 if (assume_unchanged
)
211 ce
->ce_flags
|= CE_VALID
;
213 if (S_ISREG(st
->st_mode
)) {
214 ce_mark_uptodate(ce
);
215 mark_fsmonitor_valid(istate
, ce
);
219 static int ce_compare_data(struct index_state
*istate
,
220 const struct cache_entry
*ce
,
224 int fd
= git_open_cloexec(ce
->name
, O_RDONLY
);
227 struct object_id oid
;
228 if (!index_fd(istate
, &oid
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
229 match
= !oideq(&oid
, &ce
->oid
);
230 /* index_fd() closed the file descriptor already */
235 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
240 enum object_type type
;
241 struct strbuf sb
= STRBUF_INIT
;
243 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
246 buffer
= read_object_file(&ce
->oid
, &type
, &size
);
249 match
= memcmp(buffer
, sb
.buf
, size
);
256 static int ce_compare_gitlink(const struct cache_entry
*ce
)
258 struct object_id oid
;
261 * We don't actually require that the .git directory
262 * under GITLINK directory be a valid git directory. It
263 * might even be missing (in case nobody populated that
266 * If so, we consider it always to match.
268 if (resolve_gitlink_ref(ce
->name
, "HEAD", &oid
) < 0)
270 return !oideq(&oid
, &ce
->oid
);
273 static int ce_modified_check_fs(struct index_state
*istate
,
274 const struct cache_entry
*ce
,
277 switch (st
->st_mode
& S_IFMT
) {
279 if (ce_compare_data(istate
, ce
, st
))
283 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
287 if (S_ISGITLINK(ce
->ce_mode
))
288 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
289 /* else fallthrough */
296 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
298 unsigned int changed
= 0;
300 if (ce
->ce_flags
& CE_REMOVE
)
301 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
303 switch (ce
->ce_mode
& S_IFMT
) {
305 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
306 /* We consider only the owner x bit to be relevant for
309 if (trust_executable_bit
&&
310 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
311 changed
|= MODE_CHANGED
;
314 if (!S_ISLNK(st
->st_mode
) &&
315 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
316 changed
|= TYPE_CHANGED
;
319 /* We ignore most of the st_xxx fields for gitlinks */
320 if (!S_ISDIR(st
->st_mode
))
321 changed
|= TYPE_CHANGED
;
322 else if (ce_compare_gitlink(ce
))
323 changed
|= DATA_CHANGED
;
326 BUG("unsupported ce_mode: %o", ce
->ce_mode
);
329 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
331 /* Racily smudged entry? */
332 if (!ce
->ce_stat_data
.sd_size
) {
333 if (!is_empty_blob_sha1(ce
->oid
.hash
))
334 changed
|= DATA_CHANGED
;
340 static int is_racy_stat(const struct index_state
*istate
,
341 const struct stat_data
*sd
)
343 return (istate
->timestamp
.sec
&&
345 /* nanosecond timestamped files can also be racy! */
346 (istate
->timestamp
.sec
< sd
->sd_mtime
.sec
||
347 (istate
->timestamp
.sec
== sd
->sd_mtime
.sec
&&
348 istate
->timestamp
.nsec
<= sd
->sd_mtime
.nsec
))
350 istate
->timestamp
.sec
<= sd
->sd_mtime
.sec
355 int is_racy_timestamp(const struct index_state
*istate
,
356 const struct cache_entry
*ce
)
358 return (!S_ISGITLINK(ce
->ce_mode
) &&
359 is_racy_stat(istate
, &ce
->ce_stat_data
));
362 int match_stat_data_racy(const struct index_state
*istate
,
363 const struct stat_data
*sd
, struct stat
*st
)
365 if (is_racy_stat(istate
, sd
))
366 return MTIME_CHANGED
;
367 return match_stat_data(sd
, st
);
370 int ie_match_stat(struct index_state
*istate
,
371 const struct cache_entry
*ce
, struct stat
*st
,
372 unsigned int options
)
374 unsigned int changed
;
375 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
376 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
377 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
378 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
380 if (!ignore_fsmonitor
)
381 refresh_fsmonitor(istate
);
383 * If it's marked as always valid in the index, it's
384 * valid whatever the checked-out copy says.
386 * skip-worktree has the same effect with higher precedence
388 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
390 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
392 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
))
396 * Intent-to-add entries have not been added, so the index entry
397 * by definition never matches what is in the work tree until it
398 * actually gets added.
400 if (ce_intent_to_add(ce
))
401 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
403 changed
= ce_match_stat_basic(ce
, st
);
406 * Within 1 second of this sequence:
407 * echo xyzzy >file && git-update-index --add file
408 * running this command:
410 * would give a falsely clean cache entry. The mtime and
411 * length match the cache, and other stat fields do not change.
413 * We could detect this at update-index time (the cache entry
414 * being registered/updated records the same time as "now")
415 * and delay the return from git-update-index, but that would
416 * effectively mean we can make at most one commit per second,
417 * which is not acceptable. Instead, we check cache entries
418 * whose mtime are the same as the index file timestamp more
419 * carefully than others.
421 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
422 if (assume_racy_is_modified
)
423 changed
|= DATA_CHANGED
;
425 changed
|= ce_modified_check_fs(istate
, ce
, st
);
431 int ie_modified(struct index_state
*istate
,
432 const struct cache_entry
*ce
,
433 struct stat
*st
, unsigned int options
)
435 int changed
, changed_fs
;
437 changed
= ie_match_stat(istate
, ce
, st
, options
);
441 * If the mode or type has changed, there's no point in trying
442 * to refresh the entry - it's not going to match
444 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
448 * Immediately after read-tree or update-index --cacheinfo,
449 * the length field is zero, as we have never even read the
450 * lstat(2) information once, and we cannot trust DATA_CHANGED
451 * returned by ie_match_stat() which in turn was returned by
452 * ce_match_stat_basic() to signal that the filesize of the
453 * blob changed. We have to actually go to the filesystem to
454 * see if the contents match, and if so, should answer "unchanged".
456 * The logic does not apply to gitlinks, as ce_match_stat_basic()
457 * already has checked the actual HEAD from the filesystem in the
458 * subproject. If ie_match_stat() already said it is different,
459 * then we know it is.
461 if ((changed
& DATA_CHANGED
) &&
462 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
465 changed_fs
= ce_modified_check_fs(istate
, ce
, st
);
467 return changed
| changed_fs
;
471 int base_name_compare(const char *name1
, int len1
, int mode1
,
472 const char *name2
, int len2
, int mode2
)
474 unsigned char c1
, c2
;
475 int len
= len1
< len2
? len1
: len2
;
478 cmp
= memcmp(name1
, name2
, len
);
483 if (!c1
&& S_ISDIR(mode1
))
485 if (!c2
&& S_ISDIR(mode2
))
487 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
491 * df_name_compare() is identical to base_name_compare(), except it
492 * compares conflicting directory/file entries as equal. Note that
493 * while a directory name compares as equal to a regular file, they
494 * then individually compare _differently_ to a filename that has
495 * a dot after the basename (because '\0' < '.' < '/').
497 * This is used by routines that want to traverse the git namespace
498 * but then handle conflicting entries together when possible.
500 int df_name_compare(const char *name1
, int len1
, int mode1
,
501 const char *name2
, int len2
, int mode2
)
503 int len
= len1
< len2
? len1
: len2
, cmp
;
504 unsigned char c1
, c2
;
506 cmp
= memcmp(name1
, name2
, len
);
509 /* Directories and files compare equal (same length, same name) */
513 if (!c1
&& S_ISDIR(mode1
))
516 if (!c2
&& S_ISDIR(mode2
))
518 if (c1
== '/' && !c2
)
520 if (c2
== '/' && !c1
)
525 int name_compare(const char *name1
, size_t len1
, const char *name2
, size_t len2
)
527 size_t min_len
= (len1
< len2
) ? len1
: len2
;
528 int cmp
= memcmp(name1
, name2
, min_len
);
538 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
542 cmp
= name_compare(name1
, len1
, name2
, len2
);
553 static int index_name_stage_pos(struct index_state
*istate
, const char *name
, int namelen
, int stage
)
558 last
= istate
->cache_nr
;
559 while (last
> first
) {
560 int next
= first
+ ((last
- first
) >> 1);
561 struct cache_entry
*ce
= istate
->cache
[next
];
562 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
572 if (istate
->sparse_index
&&
574 /* Note: first <= istate->cache_nr */
575 struct cache_entry
*ce
= istate
->cache
[first
- 1];
578 * If we are in a sparse-index _and_ the entry before the
579 * insertion position is a sparse-directory entry that is
580 * an ancestor of 'name', then we need to expand the index
581 * and search again. This will only trigger once, because
582 * thereafter the index is fully expanded.
584 if (S_ISSPARSEDIR(ce
->ce_mode
) &&
585 ce_namelen(ce
) < namelen
&&
586 !strncmp(name
, ce
->name
, ce_namelen(ce
))) {
587 ensure_full_index(istate
);
588 return index_name_stage_pos(istate
, name
, namelen
, stage
);
595 int index_name_pos(struct index_state
*istate
, const char *name
, int namelen
)
597 return index_name_stage_pos(istate
, name
, namelen
, 0);
600 int remove_index_entry_at(struct index_state
*istate
, int pos
)
602 struct cache_entry
*ce
= istate
->cache
[pos
];
604 record_resolve_undo(istate
, ce
);
605 remove_name_hash(istate
, ce
);
606 save_or_free_index_entry(istate
, ce
);
607 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
609 if (pos
>= istate
->cache_nr
)
611 MOVE_ARRAY(istate
->cache
+ pos
, istate
->cache
+ pos
+ 1,
612 istate
->cache_nr
- pos
);
617 * Remove all cache entries marked for removal, that is where
618 * CE_REMOVE is set in ce_flags. This is much more effective than
619 * calling remove_index_entry_at() for each entry to be removed.
621 void remove_marked_cache_entries(struct index_state
*istate
, int invalidate
)
623 struct cache_entry
**ce_array
= istate
->cache
;
626 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
627 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
629 cache_tree_invalidate_path(istate
,
631 untracked_cache_remove_from_index(istate
,
634 remove_name_hash(istate
, ce_array
[i
]);
635 save_or_free_index_entry(istate
, ce_array
[i
]);
638 ce_array
[j
++] = ce_array
[i
];
640 if (j
== istate
->cache_nr
)
642 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
643 istate
->cache_nr
= j
;
646 int remove_file_from_index(struct index_state
*istate
, const char *path
)
648 int pos
= index_name_pos(istate
, path
, strlen(path
));
651 cache_tree_invalidate_path(istate
, path
);
652 untracked_cache_remove_from_index(istate
, path
);
653 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
654 remove_index_entry_at(istate
, pos
);
658 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
660 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
663 static int index_name_pos_also_unmerged(struct index_state
*istate
,
664 const char *path
, int namelen
)
666 int pos
= index_name_pos(istate
, path
, namelen
);
667 struct cache_entry
*ce
;
672 /* maybe unmerged? */
674 if (pos
>= istate
->cache_nr
||
675 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
678 /* order of preference: stage 2, 1, 3 */
679 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
680 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
681 !compare_name(ce
, path
, namelen
))
686 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
688 int len
= ce_namelen(ce
);
689 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
693 * If we add a filename that aliases in the cache, we will use the
694 * name that we already have - but we don't want to update the same
695 * alias twice, because that implies that there were actually two
696 * different files with aliasing names!
698 * So we use the CE_ADDED flag to verify that the alias was an old
699 * one before we accept it as
701 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
702 struct cache_entry
*ce
,
703 struct cache_entry
*alias
)
706 struct cache_entry
*new_entry
;
708 if (alias
->ce_flags
& CE_ADDED
)
709 die(_("will not add file alias '%s' ('%s' already exists in index)"),
710 ce
->name
, alias
->name
);
712 /* Ok, create the new entry using the name of the existing alias */
713 len
= ce_namelen(alias
);
714 new_entry
= make_empty_cache_entry(istate
, len
);
715 memcpy(new_entry
->name
, alias
->name
, len
);
716 copy_cache_entry(new_entry
, ce
);
717 save_or_free_index_entry(istate
, ce
);
721 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
723 struct object_id oid
;
724 if (write_object_file("", 0, blob_type
, &oid
))
725 die(_("cannot create an empty blob in the object database"));
726 oidcpy(&ce
->oid
, &oid
);
729 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
731 int namelen
, was_same
;
732 mode_t st_mode
= st
->st_mode
;
733 struct cache_entry
*ce
, *alias
= NULL
;
734 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
735 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
736 int pretend
= flags
& ADD_CACHE_PRETEND
;
737 int intent_only
= flags
& ADD_CACHE_INTENT
;
738 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
739 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
740 int hash_flags
= HASH_WRITE_OBJECT
;
741 struct object_id oid
;
743 if (flags
& ADD_CACHE_RENORMALIZE
)
744 hash_flags
|= HASH_RENORMALIZE
;
746 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
747 return error(_("%s: can only add regular files, symbolic links or git-directories"), path
);
749 namelen
= strlen(path
);
750 if (S_ISDIR(st_mode
)) {
751 if (resolve_gitlink_ref(path
, "HEAD", &oid
) < 0)
752 return error(_("'%s' does not have a commit checked out"), path
);
753 while (namelen
&& path
[namelen
-1] == '/')
756 ce
= make_empty_cache_entry(istate
, namelen
);
757 memcpy(ce
->name
, path
, namelen
);
758 ce
->ce_namelen
= namelen
;
760 fill_stat_cache_info(istate
, ce
, st
);
762 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
765 if (trust_executable_bit
&& has_symlinks
) {
766 ce
->ce_mode
= create_ce_mode(st_mode
);
768 /* If there is an existing entry, pick the mode bits and type
769 * from it, otherwise assume unexecutable regular file.
771 struct cache_entry
*ent
;
772 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
774 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
775 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
778 /* When core.ignorecase=true, determine if a directory of the same name but differing
779 * case already exists within the Git repository. If it does, ensure the directory
780 * case of the file being added to the repository matches (is folded into) the existing
781 * entry's directory case.
784 adjust_dirname_case(istate
, ce
->name
);
786 if (!(flags
& ADD_CACHE_RENORMALIZE
)) {
787 alias
= index_file_exists(istate
, ce
->name
,
788 ce_namelen(ce
), ignore_case
);
791 !ie_match_stat(istate
, alias
, st
, ce_option
)) {
792 /* Nothing changed, really */
793 if (!S_ISGITLINK(alias
->ce_mode
))
794 ce_mark_uptodate(alias
);
795 alias
->ce_flags
|= CE_ADDED
;
797 discard_cache_entry(ce
);
802 if (index_path(istate
, &ce
->oid
, path
, st
, hash_flags
)) {
803 discard_cache_entry(ce
);
804 return error(_("unable to index file '%s'"), path
);
807 set_object_name_for_intent_to_add_entry(ce
);
809 if (ignore_case
&& alias
&& different_name(ce
, alias
))
810 ce
= create_alias_ce(istate
, ce
, alias
);
811 ce
->ce_flags
|= CE_ADDED
;
813 /* It was suspected to be racily clean, but it turns out to be Ok */
816 oideq(&alias
->oid
, &ce
->oid
) &&
817 ce
->ce_mode
== alias
->ce_mode
);
820 discard_cache_entry(ce
);
821 else if (add_index_entry(istate
, ce
, add_option
)) {
822 discard_cache_entry(ce
);
823 return error(_("unable to add '%s' to index"), path
);
825 if (verbose
&& !was_same
)
826 printf("add '%s'\n", path
);
830 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
833 if (lstat(path
, &st
))
834 die_errno(_("unable to stat '%s'"), path
);
835 return add_to_index(istate
, path
, &st
, flags
);
838 struct cache_entry
*make_empty_cache_entry(struct index_state
*istate
, size_t len
)
840 return mem_pool__ce_calloc(find_mem_pool(istate
), len
);
843 struct cache_entry
*make_empty_transient_cache_entry(size_t len
,
844 struct mem_pool
*ce_mem_pool
)
847 return mem_pool__ce_calloc(ce_mem_pool
, len
);
848 return xcalloc(1, cache_entry_size(len
));
851 struct cache_entry
*make_cache_entry(struct index_state
*istate
,
853 const struct object_id
*oid
,
856 unsigned int refresh_options
)
858 struct cache_entry
*ce
, *ret
;
861 if (!verify_path(path
, mode
)) {
862 error(_("invalid path '%s'"), path
);
867 ce
= make_empty_cache_entry(istate
, len
);
869 oidcpy(&ce
->oid
, oid
);
870 memcpy(ce
->name
, path
, len
);
871 ce
->ce_flags
= create_ce_flags(stage
);
872 ce
->ce_namelen
= len
;
873 ce
->ce_mode
= create_ce_mode(mode
);
875 ret
= refresh_cache_entry(istate
, ce
, refresh_options
);
877 discard_cache_entry(ce
);
881 struct cache_entry
*make_transient_cache_entry(unsigned int mode
,
882 const struct object_id
*oid
,
885 struct mem_pool
*ce_mem_pool
)
887 struct cache_entry
*ce
;
890 if (!verify_path(path
, mode
)) {
891 error(_("invalid path '%s'"), path
);
896 ce
= make_empty_transient_cache_entry(len
, ce_mem_pool
);
898 oidcpy(&ce
->oid
, oid
);
899 memcpy(ce
->name
, path
, len
);
900 ce
->ce_flags
= create_ce_flags(stage
);
901 ce
->ce_namelen
= len
;
902 ce
->ce_mode
= create_ce_mode(mode
);
908 * Chmod an index entry with either +x or -x.
910 * Returns -1 if the chmod for the particular cache entry failed (if it's
911 * not a regular file), -2 if an invalid flip argument is passed in, 0
914 int chmod_index_entry(struct index_state
*istate
, struct cache_entry
*ce
,
917 if (!S_ISREG(ce
->ce_mode
))
924 ce
->ce_mode
&= ~0111;
929 cache_tree_invalidate_path(istate
, ce
->name
);
930 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
931 mark_fsmonitor_invalid(istate
, ce
);
932 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
937 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
939 int len
= ce_namelen(a
);
940 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
944 * We fundamentally don't like some paths: we don't want
945 * dot or dot-dot anywhere, and for obvious reasons don't
946 * want to recurse into ".git" either.
948 * Also, we don't want double slashes or slashes at the
949 * end that can make pathnames ambiguous.
951 static int verify_dotfile(const char *rest
, unsigned mode
)
954 * The first character was '.', but that
955 * has already been discarded, we now test
959 /* "." is not allowed */
960 if (*rest
== '\0' || is_dir_sep(*rest
))
965 * ".git" followed by NUL or slash is bad. Note that we match
966 * case-insensitively here, even if ignore_case is not set.
967 * This outlaws ".GIT" everywhere out of an abundance of caution,
968 * since there's really no good reason to allow it.
970 * Once we've seen ".git", we can also find ".gitmodules", etc (also
971 * case-insensitively).
975 if (rest
[1] != 'i' && rest
[1] != 'I')
977 if (rest
[2] != 't' && rest
[2] != 'T')
979 if (rest
[3] == '\0' || is_dir_sep(rest
[3]))
983 if (skip_iprefix(rest
, "modules", &rest
) &&
984 (*rest
== '\0' || is_dir_sep(*rest
)))
989 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
995 int verify_path(const char *path
, unsigned mode
)
999 if (has_dos_drive_prefix(path
))
1002 if (!is_valid_path(path
))
1009 if (is_dir_sep(c
)) {
1013 if (is_hfs_dotgit(path
))
1015 if (S_ISLNK(mode
)) {
1016 if (is_hfs_dotgitmodules(path
))
1021 #if defined GIT_WINDOWS_NATIVE || defined __CYGWIN__
1025 if (is_ntfs_dotgit(path
))
1027 if (S_ISLNK(mode
)) {
1028 if (is_ntfs_dotgitmodules(path
))
1034 if ((c
== '.' && !verify_dotfile(path
, mode
)) ||
1038 * allow terminating directory separators for
1039 * sparse directory entries.
1042 return S_ISDIR(mode
);
1043 } else if (c
== '\\' && protect_ntfs
) {
1044 if (is_ntfs_dotgit(path
))
1046 if (S_ISLNK(mode
)) {
1047 if (is_ntfs_dotgitmodules(path
))
1057 * Do we have another file that has the beginning components being a
1058 * proper superset of the name we're trying to add?
1060 static int has_file_name(struct index_state
*istate
,
1061 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1064 int len
= ce_namelen(ce
);
1065 int stage
= ce_stage(ce
);
1066 const char *name
= ce
->name
;
1068 while (pos
< istate
->cache_nr
) {
1069 struct cache_entry
*p
= istate
->cache
[pos
++];
1071 if (len
>= ce_namelen(p
))
1073 if (memcmp(name
, p
->name
, len
))
1075 if (ce_stage(p
) != stage
)
1077 if (p
->name
[len
] != '/')
1079 if (p
->ce_flags
& CE_REMOVE
)
1084 remove_index_entry_at(istate
, --pos
);
1091 * Like strcmp(), but also return the offset of the first change.
1092 * If strings are equal, return the length.
1094 int strcmp_offset(const char *s1
, const char *s2
, size_t *first_change
)
1099 return strcmp(s1
, s2
);
1101 for (k
= 0; s1
[k
] == s2
[k
]; k
++)
1106 return (unsigned char)s1
[k
] - (unsigned char)s2
[k
];
1110 * Do we have another file with a pathname that is a proper
1111 * subset of the name we're trying to add?
1113 * That is, is there another file in the index with a path
1114 * that matches a sub-directory in the given entry?
1116 static int has_dir_name(struct index_state
*istate
,
1117 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1120 int stage
= ce_stage(ce
);
1121 const char *name
= ce
->name
;
1122 const char *slash
= name
+ ce_namelen(ce
);
1127 * We are frequently called during an iteration on a sorted
1128 * list of pathnames and while building a new index. Therefore,
1129 * there is a high probability that this entry will eventually
1130 * be appended to the index, rather than inserted in the middle.
1131 * If we can confirm that, we can avoid binary searches on the
1132 * components of the pathname.
1134 * Compare the entry's full path with the last path in the index.
1136 if (istate
->cache_nr
> 0) {
1137 cmp_last
= strcmp_offset(name
,
1138 istate
->cache
[istate
->cache_nr
- 1]->name
,
1141 if (len_eq_last
== 0) {
1143 * The entry sorts AFTER the last one in the
1144 * index and their paths have no common prefix,
1145 * so there cannot be a F/D conflict.
1150 * The entry sorts AFTER the last one in the
1151 * index, but has a common prefix. Fall through
1152 * to the loop below to disect the entry's path
1153 * and see where the difference is.
1156 } else if (cmp_last
== 0) {
1158 * The entry exactly matches the last one in the
1159 * index, but because of multiple stage and CE_REMOVE
1160 * items, we fall through and let the regular search
1170 if (*--slash
== '/')
1172 if (slash
<= ce
->name
)
1179 * (len + 1) is a directory boundary (including
1180 * the trailing slash). And since the loop is
1181 * decrementing "slash", the first iteration is
1182 * the longest directory prefix; subsequent
1183 * iterations consider parent directories.
1186 if (len
+ 1 <= len_eq_last
) {
1188 * The directory prefix (including the trailing
1189 * slash) also appears as a prefix in the last
1190 * entry, so the remainder cannot collide (because
1191 * strcmp said the whole path was greater).
1196 * LT: last: xxx/file_A
1202 if (len
> len_eq_last
) {
1204 * This part of the directory prefix (excluding
1205 * the trailing slash) is longer than the known
1206 * equal portions, so this sub-directory cannot
1207 * collide with a file.
1216 * This is a possible collision. Fall through and
1217 * let the regular search code handle it.
1224 pos
= index_name_stage_pos(istate
, name
, len
, stage
);
1227 * Found one, but not so fast. This could
1228 * be a marker that says "I was here, but
1229 * I am being removed". Such an entry is
1230 * not a part of the resulting tree, and
1231 * it is Ok to have a directory at the same
1234 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
1238 remove_index_entry_at(istate
, pos
);
1246 * Trivial optimization: if we find an entry that
1247 * already matches the sub-directory, then we know
1248 * we're ok, and we can exit.
1250 while (pos
< istate
->cache_nr
) {
1251 struct cache_entry
*p
= istate
->cache
[pos
];
1252 if ((ce_namelen(p
) <= len
) ||
1253 (p
->name
[len
] != '/') ||
1254 memcmp(p
->name
, name
, len
))
1255 break; /* not our subdirectory */
1256 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
1258 * p is at the same stage as our entry, and
1259 * is a subdirectory of what we are looking
1260 * at, so we cannot have conflicts at our
1261 * level or anything shorter.
1270 /* We may be in a situation where we already have path/file and path
1271 * is being added, or we already have path and path/file is being
1272 * added. Either one would result in a nonsense tree that has path
1273 * twice when git-write-tree tries to write it out. Prevent it.
1275 * If ok-to-replace is specified, we remove the conflicting entries
1276 * from the cache so the caller should recompute the insert position.
1277 * When this happens, we return non-zero.
1279 static int check_file_directory_conflict(struct index_state
*istate
,
1280 const struct cache_entry
*ce
,
1281 int pos
, int ok_to_replace
)
1286 * When ce is an "I am going away" entry, we allow it to be added
1288 if (ce
->ce_flags
& CE_REMOVE
)
1292 * We check if the path is a sub-path of a subsequent pathname
1293 * first, since removing those will not change the position
1296 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
1299 * Then check if the path might have a clashing sub-directory
1302 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
1305 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1308 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
1309 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
1310 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
1311 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
1313 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1314 cache_tree_invalidate_path(istate
, ce
->name
);
1317 * If this entry's path sorts after the last entry in the index,
1318 * we can avoid searching for it.
1320 if (istate
->cache_nr
> 0 &&
1321 strcmp(ce
->name
, istate
->cache
[istate
->cache_nr
- 1]->name
) > 0)
1322 pos
= index_pos_to_insert_pos(istate
->cache_nr
);
1324 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1326 /* existing match? Just replace it. */
1329 replace_index_entry(istate
, pos
, ce
);
1334 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1335 untracked_cache_add_to_index(istate
, ce
->name
);
1338 * Inserting a merged entry ("stage 0") into the index
1339 * will always replace all non-merged entries..
1341 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
1342 while (ce_same_name(istate
->cache
[pos
], ce
)) {
1344 if (!remove_index_entry_at(istate
, pos
))
1351 if (!verify_path(ce
->name
, ce
->ce_mode
))
1352 return error(_("invalid path '%s'"), ce
->name
);
1354 if (!skip_df_check
&&
1355 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1357 return error(_("'%s' appears as both a file and as a directory"),
1359 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1365 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1369 if (option
& ADD_CACHE_JUST_APPEND
)
1370 pos
= istate
->cache_nr
;
1373 ret
= add_index_entry_with_check(istate
, ce
, option
);
1379 /* Make sure the array is big enough .. */
1380 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1384 if (istate
->cache_nr
> pos
+ 1)
1385 MOVE_ARRAY(istate
->cache
+ pos
+ 1, istate
->cache
+ pos
,
1386 istate
->cache_nr
- pos
- 1);
1387 set_index_entry(istate
, pos
, ce
);
1388 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1393 * "refresh" does not calculate a new sha1 file or bring the
1394 * cache up-to-date for mode/content changes. But what it
1395 * _does_ do is to "re-match" the stat information of a file
1396 * with the cache, so that you can refresh the cache for a
1397 * file that hasn't been changed but where the stat entry is
1400 * For example, you'd want to do this after doing a "git-read-tree",
1401 * to link up the stat cache details with the proper files.
1403 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1404 struct cache_entry
*ce
,
1405 unsigned int options
, int *err
,
1411 struct cache_entry
*updated
;
1413 int refresh
= options
& CE_MATCH_REFRESH
;
1414 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1415 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1416 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1417 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
1419 if (!refresh
|| ce_uptodate(ce
))
1422 if (!ignore_fsmonitor
)
1423 refresh_fsmonitor(istate
);
1425 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1426 * that the change to the work tree does not matter and told
1429 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1430 ce_mark_uptodate(ce
);
1433 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1434 ce_mark_uptodate(ce
);
1437 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
)) {
1438 ce_mark_uptodate(ce
);
1442 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1452 if (lstat(ce
->name
, &st
) < 0) {
1453 if (ignore_missing
&& errno
== ENOENT
)
1460 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1462 *changed_ret
= changed
;
1465 * The path is unchanged. If we were told to ignore
1466 * valid bit, then we did the actual stat check and
1467 * found that the entry is unmodified. If the entry
1468 * is not marked VALID, this is the place to mark it
1469 * valid again, under "assume unchanged" mode.
1471 if (ignore_valid
&& assume_unchanged
&&
1472 !(ce
->ce_flags
& CE_VALID
))
1473 ; /* mark this one VALID again */
1476 * We do not mark the index itself "modified"
1477 * because CE_UPTODATE flag is in-core only;
1478 * we are not going to write this change out.
1480 if (!S_ISGITLINK(ce
->ce_mode
)) {
1481 ce_mark_uptodate(ce
);
1482 mark_fsmonitor_valid(istate
, ce
);
1490 if (ie_modified(istate
, ce
, &st
, options
)) {
1496 updated
= make_empty_cache_entry(istate
, ce_namelen(ce
));
1497 copy_cache_entry(updated
, ce
);
1498 memcpy(updated
->name
, ce
->name
, ce
->ce_namelen
+ 1);
1499 fill_stat_cache_info(istate
, updated
, &st
);
1501 * If ignore_valid is not set, we should leave CE_VALID bit
1502 * alone. Otherwise, paths marked with --no-assume-unchanged
1503 * (i.e. things to be edited) will reacquire CE_VALID bit
1504 * automatically, which is not really what we want.
1506 if (!ignore_valid
&& assume_unchanged
&&
1507 !(ce
->ce_flags
& CE_VALID
))
1508 updated
->ce_flags
&= ~CE_VALID
;
1510 /* istate->cache_changed is updated in the caller */
1514 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1515 int * first
, const char *header_msg
)
1517 if (in_porcelain
&& *first
&& header_msg
) {
1518 printf("%s\n", header_msg
);
1524 int repo_refresh_and_write_index(struct repository
*repo
,
1525 unsigned int refresh_flags
,
1526 unsigned int write_flags
,
1528 const struct pathspec
*pathspec
,
1529 char *seen
, const char *header_msg
)
1531 struct lock_file lock_file
= LOCK_INIT
;
1534 fd
= repo_hold_locked_index(repo
, &lock_file
, 0);
1535 if (!gentle
&& fd
< 0)
1537 if (refresh_index(repo
->index
, refresh_flags
, pathspec
, seen
, header_msg
))
1539 if (0 <= fd
&& write_locked_index(repo
->index
, &lock_file
, COMMIT_LOCK
| write_flags
))
1545 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1546 const struct pathspec
*pathspec
,
1547 char *seen
, const char *header_msg
)
1551 int really
= (flags
& REFRESH_REALLY
) != 0;
1552 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1553 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1554 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1555 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1556 int ignore_skip_worktree
= (flags
& REFRESH_IGNORE_SKIP_WORKTREE
) != 0;
1558 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1559 unsigned int options
= (CE_MATCH_REFRESH
|
1560 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1561 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1562 const char *modified_fmt
;
1563 const char *deleted_fmt
;
1564 const char *typechange_fmt
;
1565 const char *added_fmt
;
1566 const char *unmerged_fmt
;
1567 struct progress
*progress
= NULL
;
1568 int t2_sum_lstat
= 0;
1569 int t2_sum_scan
= 0;
1571 if (flags
& REFRESH_PROGRESS
&& isatty(2))
1572 progress
= start_delayed_progress(_("Refresh index"),
1575 trace_performance_enter();
1576 modified_fmt
= in_porcelain
? "M\t%s\n" : "%s: needs update\n";
1577 deleted_fmt
= in_porcelain
? "D\t%s\n" : "%s: needs update\n";
1578 typechange_fmt
= in_porcelain
? "T\t%s\n" : "%s: needs update\n";
1579 added_fmt
= in_porcelain
? "A\t%s\n" : "%s: needs update\n";
1580 unmerged_fmt
= in_porcelain
? "U\t%s\n" : "%s: needs merge\n";
1582 * Use the multi-threaded preload_index() to refresh most of the
1583 * cache entries quickly then in the single threaded loop below,
1584 * we only have to do the special cases that are left.
1586 preload_index(istate
, pathspec
, 0);
1587 trace2_region_enter("index", "refresh", NULL
);
1589 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1590 struct cache_entry
*ce
, *new_entry
;
1591 int cache_errno
= 0;
1594 int t2_did_lstat
= 0;
1595 int t2_did_scan
= 0;
1597 ce
= istate
->cache
[i
];
1598 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1600 if (ignore_skip_worktree
&& ce_skip_worktree(ce
))
1604 * If this entry is a sparse directory, then there isn't
1605 * any stat() information to update. Ignore the entry.
1607 if (S_ISSPARSEDIR(ce
->ce_mode
))
1610 if (pathspec
&& !ce_path_match(istate
, ce
, pathspec
, seen
))
1614 while ((i
< istate
->cache_nr
) &&
1615 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1621 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1622 &first
, header_msg
);
1630 new_entry
= refresh_cache_ent(istate
, ce
, options
,
1631 &cache_errno
, &changed
,
1632 &t2_did_lstat
, &t2_did_scan
);
1633 t2_sum_lstat
+= t2_did_lstat
;
1634 t2_sum_scan
+= t2_did_scan
;
1635 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
);
1672 display_progress(progress
, istate
->cache_nr
);
1673 stop_progress(&progress
);
1674 trace_performance_leave("refresh index");
1678 struct cache_entry
*refresh_cache_entry(struct index_state
*istate
,
1679 struct cache_entry
*ce
,
1680 unsigned int options
)
1682 return refresh_cache_ent(istate
, ce
, options
, NULL
, NULL
, NULL
, NULL
);
1686 /*****************************************************************
1688 *****************************************************************/
1690 #define INDEX_FORMAT_DEFAULT 3
1692 static unsigned int get_index_format_default(struct repository
*r
)
1694 char *envversion
= getenv("GIT_INDEX_VERSION");
1696 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1699 prepare_repo_settings(r
);
1701 if (r
->settings
.index_version
>= 0)
1702 version
= r
->settings
.index_version
;
1703 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1704 warning(_("index.version set, but the value is invalid.\n"
1705 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1706 return INDEX_FORMAT_DEFAULT
;
1711 version
= strtoul(envversion
, &endp
, 10);
1713 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1714 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1715 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1716 version
= INDEX_FORMAT_DEFAULT
;
1722 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1723 * Again - this is just a (very strong in practice) heuristic that
1724 * the inode hasn't changed.
1726 * We save the fields in big-endian order to allow using the
1727 * index file over NFS transparently.
1729 struct ondisk_cache_entry
{
1730 struct cache_time ctime
;
1731 struct cache_time mtime
;
1739 * unsigned char hash[hashsz];
1741 * if (flags & CE_EXTENDED)
1744 unsigned char data
[GIT_MAX_RAWSZ
+ 2 * sizeof(uint16_t)];
1745 char name
[FLEX_ARRAY
];
1748 /* These are only used for v3 or lower */
1749 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1750 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,data) + (len) + 8) & ~7)
1751 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1752 #define ondisk_data_size(flags, len) (the_hash_algo->rawsz + \
1753 ((flags & CE_EXTENDED) ? 2 : 1) * sizeof(uint16_t) + len)
1754 #define ondisk_data_size_max(len) (ondisk_data_size(CE_EXTENDED, len))
1755 #define ondisk_ce_size(ce) (ondisk_cache_entry_size(ondisk_data_size((ce)->ce_flags, ce_namelen(ce))))
1757 /* Allow fsck to force verification of the index checksum. */
1758 int verify_index_checksum
;
1760 /* Allow fsck to force verification of the cache entry order. */
1761 int verify_ce_order
;
1763 static int verify_hdr(const struct cache_header
*hdr
, unsigned long size
)
1766 unsigned char hash
[GIT_MAX_RAWSZ
];
1769 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1770 return error(_("bad signature 0x%08x"), hdr
->hdr_signature
);
1771 hdr_version
= ntohl(hdr
->hdr_version
);
1772 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1773 return error(_("bad index version %d"), hdr_version
);
1775 if (!verify_index_checksum
)
1778 the_hash_algo
->init_fn(&c
);
1779 the_hash_algo
->update_fn(&c
, hdr
, size
- the_hash_algo
->rawsz
);
1780 the_hash_algo
->final_fn(hash
, &c
);
1781 if (!hasheq(hash
, (unsigned char *)hdr
+ size
- the_hash_algo
->rawsz
))
1782 return error(_("bad index file sha1 signature"));
1786 static int read_index_extension(struct index_state
*istate
,
1787 const char *ext
, const char *data
, unsigned long sz
)
1789 switch (CACHE_EXT(ext
)) {
1790 case CACHE_EXT_TREE
:
1791 istate
->cache_tree
= cache_tree_read(data
, sz
);
1793 case CACHE_EXT_RESOLVE_UNDO
:
1794 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1796 case CACHE_EXT_LINK
:
1797 if (read_link_extension(istate
, data
, sz
))
1800 case CACHE_EXT_UNTRACKED
:
1801 istate
->untracked
= read_untracked_extension(data
, sz
);
1803 case CACHE_EXT_FSMONITOR
:
1804 read_fsmonitor_extension(istate
, data
, sz
);
1806 case CACHE_EXT_ENDOFINDEXENTRIES
:
1807 case CACHE_EXT_INDEXENTRYOFFSETTABLE
:
1808 /* already handled in do_read_index() */
1810 case CACHE_EXT_SPARSE_DIRECTORIES
:
1811 /* no content, only an indicator */
1812 istate
->sparse_index
= 1;
1815 if (*ext
< 'A' || 'Z' < *ext
)
1816 return error(_("index uses %.4s extension, which we do not understand"),
1818 fprintf_ln(stderr
, _("ignoring %.4s extension"), ext
);
1824 static struct cache_entry
*create_from_disk(struct mem_pool
*ce_mem_pool
,
1825 unsigned int version
,
1826 struct ondisk_cache_entry
*ondisk
,
1827 unsigned long *ent_size
,
1828 const struct cache_entry
*previous_ce
)
1830 struct cache_entry
*ce
;
1833 const unsigned hashsz
= the_hash_algo
->rawsz
;
1834 const uint16_t *flagsp
= (const uint16_t *)(ondisk
->data
+ hashsz
);
1836 size_t copy_len
= 0;
1838 * Adjacent cache entries tend to share the leading paths, so it makes
1839 * sense to only store the differences in later entries. In the v4
1840 * on-disk format of the index, each on-disk cache entry stores the
1841 * number of bytes to be stripped from the end of the previous name,
1842 * and the bytes to append to the result, to come up with its name.
1844 int expand_name_field
= version
== 4;
1846 /* On-disk flags are just 16 bits */
1847 flags
= get_be16(flagsp
);
1848 len
= flags
& CE_NAMEMASK
;
1850 if (flags
& CE_EXTENDED
) {
1852 extended_flags
= get_be16(flagsp
+ 1) << 16;
1853 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1854 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1855 die(_("unknown index entry format 0x%08x"), extended_flags
);
1856 flags
|= extended_flags
;
1857 name
= (const char *)(flagsp
+ 2);
1860 name
= (const char *)(flagsp
+ 1);
1862 if (expand_name_field
) {
1863 const unsigned char *cp
= (const unsigned char *)name
;
1864 size_t strip_len
, previous_len
;
1866 /* If we're at the beginning of a block, ignore the previous name */
1867 strip_len
= decode_varint(&cp
);
1869 previous_len
= previous_ce
->ce_namelen
;
1870 if (previous_len
< strip_len
)
1871 die(_("malformed name field in the index, near path '%s'"),
1873 copy_len
= previous_len
- strip_len
;
1875 name
= (const char *)cp
;
1878 if (len
== CE_NAMEMASK
) {
1880 if (expand_name_field
)
1884 ce
= mem_pool__ce_alloc(ce_mem_pool
, len
);
1886 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1887 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1888 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1889 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1890 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1891 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1892 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1893 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1894 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1895 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1896 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1897 ce
->ce_namelen
= len
;
1899 oidread(&ce
->oid
, ondisk
->data
);
1900 memcpy(ce
->name
, name
, len
);
1901 ce
->name
[len
] = '\0';
1903 if (expand_name_field
) {
1905 memcpy(ce
->name
, previous_ce
->name
, copy_len
);
1906 memcpy(ce
->name
+ copy_len
, name
, len
+ 1 - copy_len
);
1907 *ent_size
= (name
- ((char *)ondisk
)) + len
+ 1 - copy_len
;
1909 memcpy(ce
->name
, name
, len
+ 1);
1910 *ent_size
= ondisk_ce_size(ce
);
1915 static void check_ce_order(struct index_state
*istate
)
1919 if (!verify_ce_order
)
1922 for (i
= 1; i
< istate
->cache_nr
; i
++) {
1923 struct cache_entry
*ce
= istate
->cache
[i
- 1];
1924 struct cache_entry
*next_ce
= istate
->cache
[i
];
1925 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1927 if (0 < name_compare
)
1928 die(_("unordered stage entries in index"));
1929 if (!name_compare
) {
1931 die(_("multiple stage entries for merged file '%s'"),
1933 if (ce_stage(ce
) > ce_stage(next_ce
))
1934 die(_("unordered stage entries for '%s'"),
1940 static void tweak_untracked_cache(struct index_state
*istate
)
1942 struct repository
*r
= the_repository
;
1944 prepare_repo_settings(r
);
1946 if (r
->settings
.core_untracked_cache
== UNTRACKED_CACHE_REMOVE
) {
1947 remove_untracked_cache(istate
);
1951 if (r
->settings
.core_untracked_cache
== UNTRACKED_CACHE_WRITE
)
1952 add_untracked_cache(istate
);
1955 static void tweak_split_index(struct index_state
*istate
)
1957 switch (git_config_get_split_index()) {
1958 case -1: /* unset: do nothing */
1961 remove_split_index(istate
);
1964 add_split_index(istate
);
1966 default: /* unknown value: do nothing */
1971 static void post_read_index_from(struct index_state
*istate
)
1973 check_ce_order(istate
);
1974 tweak_untracked_cache(istate
);
1975 tweak_split_index(istate
);
1976 tweak_fsmonitor(istate
);
1979 static size_t estimate_cache_size_from_compressed(unsigned int entries
)
1981 return entries
* (sizeof(struct cache_entry
) + CACHE_ENTRY_PATH_LENGTH
);
1984 static size_t estimate_cache_size(size_t ondisk_size
, unsigned int entries
)
1986 long per_entry
= sizeof(struct cache_entry
) - sizeof(struct ondisk_cache_entry
);
1989 * Account for potential alignment differences.
1991 per_entry
+= align_padding_size(per_entry
, 0);
1992 return ondisk_size
+ entries
* per_entry
;
1995 struct index_entry_offset
1997 /* starting byte offset into index file, count of index entries in this block */
2001 struct index_entry_offset_table
2004 struct index_entry_offset entries
[FLEX_ARRAY
];
2007 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
);
2008 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
);
2010 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
);
2011 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
);
2013 struct load_index_extensions
2016 struct index_state
*istate
;
2019 unsigned long src_offset
;
2022 static void *load_index_extensions(void *_data
)
2024 struct load_index_extensions
*p
= _data
;
2025 unsigned long src_offset
= p
->src_offset
;
2027 while (src_offset
<= p
->mmap_size
- the_hash_algo
->rawsz
- 8) {
2028 /* After an array of active_nr index entries,
2029 * there can be arbitrary number of extended
2030 * sections, each of which is prefixed with
2031 * extension name (4-byte) and section length
2032 * in 4-byte network byte order.
2034 uint32_t extsize
= get_be32(p
->mmap
+ src_offset
+ 4);
2035 if (read_index_extension(p
->istate
,
2036 p
->mmap
+ src_offset
,
2037 p
->mmap
+ src_offset
+ 8,
2039 munmap((void *)p
->mmap
, p
->mmap_size
);
2040 die(_("index file corrupt"));
2043 src_offset
+= extsize
;
2050 * A helper function that will load the specified range of cache entries
2051 * from the memory mapped file and add them to the given index.
2053 static unsigned long load_cache_entry_block(struct index_state
*istate
,
2054 struct mem_pool
*ce_mem_pool
, int offset
, int nr
, const char *mmap
,
2055 unsigned long start_offset
, const struct cache_entry
*previous_ce
)
2058 unsigned long src_offset
= start_offset
;
2060 for (i
= offset
; i
< offset
+ nr
; i
++) {
2061 struct ondisk_cache_entry
*disk_ce
;
2062 struct cache_entry
*ce
;
2063 unsigned long consumed
;
2065 disk_ce
= (struct ondisk_cache_entry
*)(mmap
+ src_offset
);
2066 ce
= create_from_disk(ce_mem_pool
, istate
->version
, disk_ce
, &consumed
, previous_ce
);
2067 set_index_entry(istate
, i
, ce
);
2069 src_offset
+= consumed
;
2072 return src_offset
- start_offset
;
2075 static unsigned long load_all_cache_entries(struct index_state
*istate
,
2076 const char *mmap
, size_t mmap_size
, unsigned long src_offset
)
2078 unsigned long consumed
;
2080 istate
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2081 if (istate
->version
== 4) {
2082 mem_pool_init(istate
->ce_mem_pool
,
2083 estimate_cache_size_from_compressed(istate
->cache_nr
));
2085 mem_pool_init(istate
->ce_mem_pool
,
2086 estimate_cache_size(mmap_size
, istate
->cache_nr
));
2089 consumed
= load_cache_entry_block(istate
, istate
->ce_mem_pool
,
2090 0, istate
->cache_nr
, mmap
, src_offset
, NULL
);
2095 * Mostly randomly chosen maximum thread counts: we
2096 * cap the parallelism to online_cpus() threads, and we want
2097 * to have at least 10000 cache entries per thread for it to
2098 * be worth starting a thread.
2101 #define THREAD_COST (10000)
2103 struct load_cache_entries_thread_data
2106 struct index_state
*istate
;
2107 struct mem_pool
*ce_mem_pool
;
2110 struct index_entry_offset_table
*ieot
;
2111 int ieot_start
; /* starting index into the ieot array */
2112 int ieot_blocks
; /* count of ieot entries to process */
2113 unsigned long consumed
; /* return # of bytes in index file processed */
2117 * A thread proc to run the load_cache_entries() computation
2118 * across multiple background threads.
2120 static void *load_cache_entries_thread(void *_data
)
2122 struct load_cache_entries_thread_data
*p
= _data
;
2125 /* iterate across all ieot blocks assigned to this thread */
2126 for (i
= p
->ieot_start
; i
< p
->ieot_start
+ p
->ieot_blocks
; i
++) {
2127 p
->consumed
+= load_cache_entry_block(p
->istate
, p
->ce_mem_pool
,
2128 p
->offset
, p
->ieot
->entries
[i
].nr
, p
->mmap
, p
->ieot
->entries
[i
].offset
, NULL
);
2129 p
->offset
+= p
->ieot
->entries
[i
].nr
;
2134 static unsigned long load_cache_entries_threaded(struct index_state
*istate
, const char *mmap
, size_t mmap_size
,
2135 int nr_threads
, struct index_entry_offset_table
*ieot
)
2137 int i
, offset
, ieot_blocks
, ieot_start
, err
;
2138 struct load_cache_entries_thread_data
*data
;
2139 unsigned long consumed
= 0;
2141 /* a little sanity checking */
2142 if (istate
->name_hash_initialized
)
2143 BUG("the name hash isn't thread safe");
2145 istate
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2146 mem_pool_init(istate
->ce_mem_pool
, 0);
2148 /* ensure we have no more threads than we have blocks to process */
2149 if (nr_threads
> ieot
->nr
)
2150 nr_threads
= ieot
->nr
;
2151 CALLOC_ARRAY(data
, nr_threads
);
2153 offset
= ieot_start
= 0;
2154 ieot_blocks
= DIV_ROUND_UP(ieot
->nr
, nr_threads
);
2155 for (i
= 0; i
< nr_threads
; i
++) {
2156 struct load_cache_entries_thread_data
*p
= &data
[i
];
2159 if (ieot_start
+ ieot_blocks
> ieot
->nr
)
2160 ieot_blocks
= ieot
->nr
- ieot_start
;
2166 p
->ieot_start
= ieot_start
;
2167 p
->ieot_blocks
= ieot_blocks
;
2169 /* create a mem_pool for each thread */
2171 for (j
= p
->ieot_start
; j
< p
->ieot_start
+ p
->ieot_blocks
; j
++)
2172 nr
+= p
->ieot
->entries
[j
].nr
;
2173 p
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2174 if (istate
->version
== 4) {
2175 mem_pool_init(p
->ce_mem_pool
,
2176 estimate_cache_size_from_compressed(nr
));
2178 mem_pool_init(p
->ce_mem_pool
,
2179 estimate_cache_size(mmap_size
, nr
));
2182 err
= pthread_create(&p
->pthread
, NULL
, load_cache_entries_thread
, p
);
2184 die(_("unable to create load_cache_entries thread: %s"), strerror(err
));
2186 /* increment by the number of cache entries in the ieot block being processed */
2187 for (j
= 0; j
< ieot_blocks
; j
++)
2188 offset
+= ieot
->entries
[ieot_start
+ j
].nr
;
2189 ieot_start
+= ieot_blocks
;
2192 for (i
= 0; i
< nr_threads
; i
++) {
2193 struct load_cache_entries_thread_data
*p
= &data
[i
];
2195 err
= pthread_join(p
->pthread
, NULL
);
2197 die(_("unable to join load_cache_entries thread: %s"), strerror(err
));
2198 mem_pool_combine(istate
->ce_mem_pool
, p
->ce_mem_pool
);
2199 consumed
+= p
->consumed
;
2207 /* remember to discard_cache() before reading a different cache! */
2208 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
2212 unsigned long src_offset
;
2213 const struct cache_header
*hdr
;
2216 struct load_index_extensions p
;
2217 size_t extension_offset
= 0;
2218 int nr_threads
, cpus
;
2219 struct index_entry_offset_table
*ieot
= NULL
;
2221 if (istate
->initialized
)
2222 return istate
->cache_nr
;
2224 istate
->timestamp
.sec
= 0;
2225 istate
->timestamp
.nsec
= 0;
2226 fd
= open(path
, O_RDONLY
);
2228 if (!must_exist
&& errno
== ENOENT
)
2230 die_errno(_("%s: index file open failed"), path
);
2234 die_errno(_("%s: cannot stat the open index"), path
);
2236 mmap_size
= xsize_t(st
.st_size
);
2237 if (mmap_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2238 die(_("%s: index file smaller than expected"), path
);
2240 mmap
= xmmap_gently(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2241 if (mmap
== MAP_FAILED
)
2242 die_errno(_("%s: unable to map index file%s"), path
,
2246 hdr
= (const struct cache_header
*)mmap
;
2247 if (verify_hdr(hdr
, mmap_size
) < 0)
2250 oidread(&istate
->oid
, (const unsigned char *)hdr
+ mmap_size
- the_hash_algo
->rawsz
);
2251 istate
->version
= ntohl(hdr
->hdr_version
);
2252 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
2253 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
2254 CALLOC_ARRAY(istate
->cache
, istate
->cache_alloc
);
2255 istate
->initialized
= 1;
2259 p
.mmap_size
= mmap_size
;
2261 src_offset
= sizeof(*hdr
);
2263 if (git_config_get_index_threads(&nr_threads
))
2266 /* TODO: does creating more threads than cores help? */
2268 nr_threads
= istate
->cache_nr
/ THREAD_COST
;
2269 cpus
= online_cpus();
2270 if (nr_threads
> cpus
)
2277 if (nr_threads
> 1) {
2278 extension_offset
= read_eoie_extension(mmap
, mmap_size
);
2279 if (extension_offset
) {
2282 p
.src_offset
= extension_offset
;
2283 err
= pthread_create(&p
.pthread
, NULL
, load_index_extensions
, &p
);
2285 die(_("unable to create load_index_extensions thread: %s"), strerror(err
));
2292 * Locate and read the index entry offset table so that we can use it
2293 * to multi-thread the reading of the cache entries.
2295 if (extension_offset
&& nr_threads
> 1)
2296 ieot
= read_ieot_extension(mmap
, mmap_size
, extension_offset
);
2299 src_offset
+= load_cache_entries_threaded(istate
, mmap
, mmap_size
, nr_threads
, ieot
);
2302 src_offset
+= load_all_cache_entries(istate
, mmap
, mmap_size
, src_offset
);
2305 istate
->timestamp
.sec
= st
.st_mtime
;
2306 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2308 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2309 if (extension_offset
) {
2310 int ret
= pthread_join(p
.pthread
, NULL
);
2312 die(_("unable to join load_index_extensions thread: %s"), strerror(ret
));
2314 p
.src_offset
= src_offset
;
2315 load_index_extensions(&p
);
2317 munmap((void *)mmap
, mmap_size
);
2320 * TODO trace2: replace "the_repository" with the actual repo instance
2321 * that is associated with the given "istate".
2323 trace2_data_intmax("index", the_repository
, "read/version",
2325 trace2_data_intmax("index", the_repository
, "read/cache_nr",
2329 istate
->repo
= the_repository
;
2330 prepare_repo_settings(istate
->repo
);
2331 if (istate
->repo
->settings
.command_requires_full_index
)
2332 ensure_full_index(istate
);
2334 return istate
->cache_nr
;
2337 munmap((void *)mmap
, mmap_size
);
2338 die(_("index file corrupt"));
2342 * Signal that the shared index is used by updating its mtime.
2344 * This way, shared index can be removed if they have not been used
2347 static void freshen_shared_index(const char *shared_index
, int warn
)
2349 if (!check_and_freshen_file(shared_index
, 1) && warn
)
2350 warning(_("could not freshen shared index '%s'"), shared_index
);
2353 int read_index_from(struct index_state
*istate
, const char *path
,
2356 struct split_index
*split_index
;
2361 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2362 if (istate
->initialized
)
2363 return istate
->cache_nr
;
2366 * TODO trace2: replace "the_repository" with the actual repo instance
2367 * that is associated with the given "istate".
2369 trace2_region_enter_printf("index", "do_read_index", the_repository
,
2371 trace_performance_enter();
2372 ret
= do_read_index(istate
, path
, 0);
2373 trace_performance_leave("read cache %s", path
);
2374 trace2_region_leave_printf("index", "do_read_index", the_repository
,
2377 split_index
= istate
->split_index
;
2378 if (!split_index
|| is_null_oid(&split_index
->base_oid
)) {
2379 post_read_index_from(istate
);
2383 trace_performance_enter();
2384 if (split_index
->base
)
2385 discard_index(split_index
->base
);
2387 CALLOC_ARRAY(split_index
->base
, 1);
2389 base_oid_hex
= oid_to_hex(&split_index
->base_oid
);
2390 base_path
= xstrfmt("%s/sharedindex.%s", gitdir
, base_oid_hex
);
2391 trace2_region_enter_printf("index", "shared/do_read_index",
2392 the_repository
, "%s", base_path
);
2393 ret
= do_read_index(split_index
->base
, base_path
, 1);
2394 trace2_region_leave_printf("index", "shared/do_read_index",
2395 the_repository
, "%s", base_path
);
2396 if (!oideq(&split_index
->base_oid
, &split_index
->base
->oid
))
2397 die(_("broken index, expect %s in %s, got %s"),
2398 base_oid_hex
, base_path
,
2399 oid_to_hex(&split_index
->base
->oid
));
2401 freshen_shared_index(base_path
, 0);
2402 merge_base_index(istate
);
2403 post_read_index_from(istate
);
2404 trace_performance_leave("read cache %s", base_path
);
2409 int is_index_unborn(struct index_state
*istate
)
2411 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
2414 int discard_index(struct index_state
*istate
)
2417 * Cache entries in istate->cache[] should have been allocated
2418 * from the memory pool associated with this index, or from an
2419 * associated split_index. There is no need to free individual
2420 * cache entries. validate_cache_entries can detect when this
2421 * assertion does not hold.
2423 validate_cache_entries(istate
);
2425 resolve_undo_clear_index(istate
);
2426 istate
->cache_nr
= 0;
2427 istate
->cache_changed
= 0;
2428 istate
->timestamp
.sec
= 0;
2429 istate
->timestamp
.nsec
= 0;
2430 free_name_hash(istate
);
2431 cache_tree_free(&(istate
->cache_tree
));
2432 istate
->initialized
= 0;
2433 istate
->fsmonitor_has_run_once
= 0;
2434 FREE_AND_NULL(istate
->fsmonitor_last_update
);
2435 FREE_AND_NULL(istate
->cache
);
2436 istate
->cache_alloc
= 0;
2437 discard_split_index(istate
);
2438 free_untracked_cache(istate
->untracked
);
2439 istate
->untracked
= NULL
;
2441 if (istate
->ce_mem_pool
) {
2442 mem_pool_discard(istate
->ce_mem_pool
, should_validate_cache_entries());
2443 FREE_AND_NULL(istate
->ce_mem_pool
);
2450 * Validate the cache entries of this index.
2451 * All cache entries associated with this index
2452 * should have been allocated by the memory pool
2453 * associated with this index, or by a referenced
2456 void validate_cache_entries(const struct index_state
*istate
)
2460 if (!should_validate_cache_entries() ||!istate
|| !istate
->initialized
)
2463 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2465 BUG("cache entry is not allocated from expected memory pool");
2466 } else if (!istate
->ce_mem_pool
||
2467 !mem_pool_contains(istate
->ce_mem_pool
, istate
->cache
[i
])) {
2468 if (!istate
->split_index
||
2469 !istate
->split_index
->base
||
2470 !istate
->split_index
->base
->ce_mem_pool
||
2471 !mem_pool_contains(istate
->split_index
->base
->ce_mem_pool
, istate
->cache
[i
])) {
2472 BUG("cache entry is not allocated from expected memory pool");
2477 if (istate
->split_index
)
2478 validate_cache_entries(istate
->split_index
->base
);
2481 int unmerged_index(const struct index_state
*istate
)
2484 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2485 if (ce_stage(istate
->cache
[i
]))
2491 int repo_index_has_changes(struct repository
*repo
,
2495 struct index_state
*istate
= repo
->index
;
2496 struct object_id cmp
;
2500 cmp
= tree
->object
.oid
;
2501 if (tree
|| !get_oid_tree("HEAD", &cmp
)) {
2502 struct diff_options opt
;
2504 repo_diff_setup(repo
, &opt
);
2505 opt
.flags
.exit_with_status
= 1;
2507 opt
.flags
.quick
= 1;
2508 do_diff_cache(&cmp
, &opt
);
2510 for (i
= 0; sb
&& i
< diff_queued_diff
.nr
; i
++) {
2512 strbuf_addch(sb
, ' ');
2513 strbuf_addstr(sb
, diff_queued_diff
.queue
[i
]->two
->path
);
2516 return opt
.flags
.has_changes
!= 0;
2518 /* TODO: audit for interaction with sparse-index. */
2519 ensure_full_index(istate
);
2520 for (i
= 0; sb
&& i
< istate
->cache_nr
; i
++) {
2522 strbuf_addch(sb
, ' ');
2523 strbuf_addstr(sb
, istate
->cache
[i
]->name
);
2525 return !!istate
->cache_nr
;
2529 static int write_index_ext_header(struct hashfile
*f
,
2530 git_hash_ctx
*eoie_f
,
2534 hashwrite_be32(f
, ext
);
2535 hashwrite_be32(f
, sz
);
2540 the_hash_algo
->update_fn(eoie_f
, &ext
, sizeof(ext
));
2541 the_hash_algo
->update_fn(eoie_f
, &sz
, sizeof(sz
));
2546 static void ce_smudge_racily_clean_entry(struct index_state
*istate
,
2547 struct cache_entry
*ce
)
2550 * The only thing we care about in this function is to smudge the
2551 * falsely clean entry due to touch-update-touch race, so we leave
2552 * everything else as they are. We are called for entries whose
2553 * ce_stat_data.sd_mtime match the index file mtime.
2555 * Note that this actually does not do much for gitlinks, for
2556 * which ce_match_stat_basic() always goes to the actual
2557 * contents. The caller checks with is_racy_timestamp() which
2558 * always says "no" for gitlinks, so we are not called for them ;-)
2562 if (lstat(ce
->name
, &st
) < 0)
2564 if (ce_match_stat_basic(ce
, &st
))
2566 if (ce_modified_check_fs(istate
, ce
, &st
)) {
2567 /* This is "racily clean"; smudge it. Note that this
2568 * is a tricky code. At first glance, it may appear
2569 * that it can break with this sequence:
2571 * $ echo xyzzy >frotz
2572 * $ git-update-index --add frotz
2575 * $ echo filfre >nitfol
2576 * $ git-update-index --add nitfol
2578 * but it does not. When the second update-index runs,
2579 * it notices that the entry "frotz" has the same timestamp
2580 * as index, and if we were to smudge it by resetting its
2581 * size to zero here, then the object name recorded
2582 * in index is the 6-byte file but the cached stat information
2583 * becomes zero --- which would then match what we would
2584 * obtain from the filesystem next time we stat("frotz").
2586 * However, the second update-index, before calling
2587 * this function, notices that the cached size is 6
2588 * bytes and what is on the filesystem is an empty
2589 * file, and never calls us, so the cached size information
2590 * for "frotz" stays 6 which does not match the filesystem.
2592 ce
->ce_stat_data
.sd_size
= 0;
2596 /* Copy miscellaneous fields but not the name */
2597 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
2598 struct cache_entry
*ce
)
2601 const unsigned hashsz
= the_hash_algo
->rawsz
;
2602 uint16_t *flagsp
= (uint16_t *)(ondisk
->data
+ hashsz
);
2604 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
2605 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
2606 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
2607 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
2608 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
2609 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
2610 ondisk
->mode
= htonl(ce
->ce_mode
);
2611 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
2612 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
2613 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
2614 hashcpy(ondisk
->data
, ce
->oid
.hash
);
2616 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
2617 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
2618 flagsp
[0] = htons(flags
);
2619 if (ce
->ce_flags
& CE_EXTENDED
) {
2620 flagsp
[1] = htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
2624 static int ce_write_entry(struct hashfile
*f
, struct cache_entry
*ce
,
2625 struct strbuf
*previous_name
, struct ondisk_cache_entry
*ondisk
)
2628 unsigned int saved_namelen
;
2629 int stripped_name
= 0;
2630 static unsigned char padding
[8] = { 0x00 };
2632 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2633 saved_namelen
= ce_namelen(ce
);
2638 size
= offsetof(struct ondisk_cache_entry
,data
) + ondisk_data_size(ce
->ce_flags
, 0);
2640 if (!previous_name
) {
2641 int len
= ce_namelen(ce
);
2642 copy_cache_entry_to_ondisk(ondisk
, ce
);
2643 hashwrite(f
, ondisk
, size
);
2644 hashwrite(f
, ce
->name
, len
);
2645 hashwrite(f
, padding
, align_padding_size(size
, len
));
2647 int common
, to_remove
, prefix_size
;
2648 unsigned char to_remove_vi
[16];
2650 (ce
->name
[common
] &&
2651 common
< previous_name
->len
&&
2652 ce
->name
[common
] == previous_name
->buf
[common
]);
2654 ; /* still matching */
2655 to_remove
= previous_name
->len
- common
;
2656 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
2658 copy_cache_entry_to_ondisk(ondisk
, ce
);
2659 hashwrite(f
, ondisk
, size
);
2660 hashwrite(f
, to_remove_vi
, prefix_size
);
2661 hashwrite(f
, ce
->name
+ common
, ce_namelen(ce
) - common
);
2662 hashwrite(f
, padding
, 1);
2664 strbuf_splice(previous_name
, common
, to_remove
,
2665 ce
->name
+ common
, ce_namelen(ce
) - common
);
2667 if (stripped_name
) {
2668 ce
->ce_namelen
= saved_namelen
;
2669 ce
->ce_flags
&= ~CE_STRIP_NAME
;
2676 * This function verifies if index_state has the correct sha1 of the
2677 * index file. Don't die if we have any other failure, just return 0.
2679 static int verify_index_from(const struct index_state
*istate
, const char *path
)
2684 unsigned char hash
[GIT_MAX_RAWSZ
];
2686 if (!istate
->initialized
)
2689 fd
= open(path
, O_RDONLY
);
2696 if (st
.st_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2699 n
= pread_in_full(fd
, hash
, the_hash_algo
->rawsz
, st
.st_size
- the_hash_algo
->rawsz
);
2700 if (n
!= the_hash_algo
->rawsz
)
2703 if (!hasheq(istate
->oid
.hash
, hash
))
2714 static int repo_verify_index(struct repository
*repo
)
2716 return verify_index_from(repo
->index
, repo
->index_file
);
2719 static int has_racy_timestamp(struct index_state
*istate
)
2721 int entries
= istate
->cache_nr
;
2724 for (i
= 0; i
< entries
; i
++) {
2725 struct cache_entry
*ce
= istate
->cache
[i
];
2726 if (is_racy_timestamp(istate
, ce
))
2732 void repo_update_index_if_able(struct repository
*repo
,
2733 struct lock_file
*lockfile
)
2735 if ((repo
->index
->cache_changed
||
2736 has_racy_timestamp(repo
->index
)) &&
2737 repo_verify_index(repo
))
2738 write_locked_index(repo
->index
, lockfile
, COMMIT_LOCK
);
2740 rollback_lock_file(lockfile
);
2743 static int record_eoie(void)
2747 if (!git_config_get_bool("index.recordendofindexentries", &val
))
2751 * As a convenience, the end of index entries extension
2752 * used for threading is written by default if the user
2753 * explicitly requested threaded index reads.
2755 return !git_config_get_index_threads(&val
) && val
!= 1;
2758 static int record_ieot(void)
2762 if (!git_config_get_bool("index.recordoffsettable", &val
))
2766 * As a convenience, the offset table used for threading is
2767 * written by default if the user explicitly requested
2768 * threaded index reads.
2770 return !git_config_get_index_threads(&val
) && val
!= 1;
2774 * On success, `tempfile` is closed. If it is the temporary file
2775 * of a `struct lock_file`, we will therefore effectively perform
2776 * a 'close_lock_file_gently()`. Since that is an implementation
2777 * detail of lockfiles, callers of `do_write_index()` should not
2780 static int do_write_index(struct index_state
*istate
, struct tempfile
*tempfile
,
2781 int strip_extensions
)
2783 uint64_t start
= getnanotime();
2785 git_hash_ctx
*eoie_c
= NULL
;
2786 struct cache_header hdr
;
2787 int i
, err
= 0, removed
, extended
, hdr_version
;
2788 struct cache_entry
**cache
= istate
->cache
;
2789 int entries
= istate
->cache_nr
;
2791 struct ondisk_cache_entry ondisk
;
2792 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
2793 int drop_cache_tree
= istate
->drop_cache_tree
;
2795 int ieot_entries
= 1;
2796 struct index_entry_offset_table
*ieot
= NULL
;
2799 f
= hashfd(tempfile
->fd
, tempfile
->filename
.buf
);
2801 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
2802 if (cache
[i
]->ce_flags
& CE_REMOVE
)
2805 /* reduce extended entries if possible */
2806 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
2807 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
2809 cache
[i
]->ce_flags
|= CE_EXTENDED
;
2813 if (!istate
->version
) {
2814 istate
->version
= get_index_format_default(the_repository
);
2815 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
2816 init_split_index(istate
);
2819 /* demote version 3 to version 2 when the latter suffices */
2820 if (istate
->version
== 3 || istate
->version
== 2)
2821 istate
->version
= extended
? 3 : 2;
2823 hdr_version
= istate
->version
;
2825 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2826 hdr
.hdr_version
= htonl(hdr_version
);
2827 hdr
.hdr_entries
= htonl(entries
- removed
);
2829 hashwrite(f
, &hdr
, sizeof(hdr
));
2831 if (!HAVE_THREADS
|| git_config_get_index_threads(&nr_threads
))
2834 if (nr_threads
!= 1 && record_ieot()) {
2835 int ieot_blocks
, cpus
;
2838 * ensure default number of ieot blocks maps evenly to the
2839 * default number of threads that will process them leaving
2840 * room for the thread to load the index extensions.
2843 ieot_blocks
= istate
->cache_nr
/ THREAD_COST
;
2844 cpus
= online_cpus();
2845 if (ieot_blocks
> cpus
- 1)
2846 ieot_blocks
= cpus
- 1;
2848 ieot_blocks
= nr_threads
;
2849 if (ieot_blocks
> istate
->cache_nr
)
2850 ieot_blocks
= istate
->cache_nr
;
2854 * no reason to write out the IEOT extension if we don't
2855 * have enough blocks to utilize multi-threading
2857 if (ieot_blocks
> 1) {
2858 ieot
= xcalloc(1, sizeof(struct index_entry_offset_table
)
2859 + (ieot_blocks
* sizeof(struct index_entry_offset
)));
2860 ieot_entries
= DIV_ROUND_UP(entries
, ieot_blocks
);
2864 offset
= hashfile_total(f
);
2867 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
2869 for (i
= 0; i
< entries
; i
++) {
2870 struct cache_entry
*ce
= cache
[i
];
2871 if (ce
->ce_flags
& CE_REMOVE
)
2873 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2874 ce_smudge_racily_clean_entry(istate
, ce
);
2875 if (is_null_oid(&ce
->oid
)) {
2876 static const char msg
[] = "cache entry has null sha1: %s";
2877 static int allow
= -1;
2880 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2882 warning(msg
, ce
->name
);
2884 err
= error(msg
, ce
->name
);
2886 drop_cache_tree
= 1;
2888 if (ieot
&& i
&& (i
% ieot_entries
== 0)) {
2889 ieot
->entries
[ieot
->nr
].nr
= nr
;
2890 ieot
->entries
[ieot
->nr
].offset
= offset
;
2893 * If we have a V4 index, set the first byte to an invalid
2894 * character to ensure there is nothing common with the previous
2898 previous_name
->buf
[0] = 0;
2901 offset
= hashfile_total(f
);
2903 if (ce_write_entry(f
, ce
, previous_name
, (struct ondisk_cache_entry
*)&ondisk
) < 0)
2911 ieot
->entries
[ieot
->nr
].nr
= nr
;
2912 ieot
->entries
[ieot
->nr
].offset
= offset
;
2915 strbuf_release(&previous_name_buf
);
2922 offset
= hashfile_total(f
);
2925 * The extension headers must be hashed on their own for the
2926 * EOIE extension. Create a hashfile here to compute that hash.
2928 if (offset
&& record_eoie()) {
2929 CALLOC_ARRAY(eoie_c
, 1);
2930 the_hash_algo
->init_fn(eoie_c
);
2934 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2935 * can minimize the number of extensions we have to scan through to
2936 * find it during load. Write it out regardless of the
2937 * strip_extensions parameter as we need it when loading the shared
2941 struct strbuf sb
= STRBUF_INIT
;
2943 write_ieot_extension(&sb
, ieot
);
2944 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_INDEXENTRYOFFSETTABLE
, sb
.len
) < 0;
2945 hashwrite(f
, sb
.buf
, sb
.len
);
2946 strbuf_release(&sb
);
2952 if (!strip_extensions
&& istate
->split_index
&&
2953 !is_null_oid(&istate
->split_index
->base_oid
)) {
2954 struct strbuf sb
= STRBUF_INIT
;
2956 err
= write_link_extension(&sb
, istate
) < 0 ||
2957 write_index_ext_header(f
, eoie_c
, CACHE_EXT_LINK
,
2959 hashwrite(f
, sb
.buf
, sb
.len
);
2960 strbuf_release(&sb
);
2964 if (!strip_extensions
&& !drop_cache_tree
&& istate
->cache_tree
) {
2965 struct strbuf sb
= STRBUF_INIT
;
2967 cache_tree_write(&sb
, istate
->cache_tree
);
2968 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_TREE
, sb
.len
) < 0;
2969 hashwrite(f
, sb
.buf
, sb
.len
);
2970 strbuf_release(&sb
);
2974 if (!strip_extensions
&& istate
->resolve_undo
) {
2975 struct strbuf sb
= STRBUF_INIT
;
2977 resolve_undo_write(&sb
, istate
->resolve_undo
);
2978 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_RESOLVE_UNDO
,
2980 hashwrite(f
, sb
.buf
, sb
.len
);
2981 strbuf_release(&sb
);
2985 if (!strip_extensions
&& istate
->untracked
) {
2986 struct strbuf sb
= STRBUF_INIT
;
2988 write_untracked_extension(&sb
, istate
->untracked
);
2989 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_UNTRACKED
,
2991 hashwrite(f
, sb
.buf
, sb
.len
);
2992 strbuf_release(&sb
);
2996 if (!strip_extensions
&& istate
->fsmonitor_last_update
) {
2997 struct strbuf sb
= STRBUF_INIT
;
2999 write_fsmonitor_extension(&sb
, istate
);
3000 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_FSMONITOR
, sb
.len
) < 0;
3001 hashwrite(f
, sb
.buf
, sb
.len
);
3002 strbuf_release(&sb
);
3006 if (istate
->sparse_index
) {
3007 if (write_index_ext_header(f
, eoie_c
, CACHE_EXT_SPARSE_DIRECTORIES
, 0) < 0)
3012 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
3013 * so that it can be found and processed before all the index entries are
3014 * read. Write it out regardless of the strip_extensions parameter as we need it
3015 * when loading the shared index.
3018 struct strbuf sb
= STRBUF_INIT
;
3020 write_eoie_extension(&sb
, eoie_c
, offset
);
3021 err
= write_index_ext_header(f
, NULL
, CACHE_EXT_ENDOFINDEXENTRIES
, sb
.len
) < 0;
3022 hashwrite(f
, sb
.buf
, sb
.len
);
3023 strbuf_release(&sb
);
3028 finalize_hashfile(f
, istate
->oid
.hash
, CSUM_HASH_IN_STREAM
);
3029 if (close_tempfile_gently(tempfile
)) {
3030 error(_("could not close '%s'"), get_tempfile_path(tempfile
));
3033 if (stat(get_tempfile_path(tempfile
), &st
))
3035 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
3036 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
3037 trace_performance_since(start
, "write index, changed mask = %x", istate
->cache_changed
);
3040 * TODO trace2: replace "the_repository" with the actual repo instance
3041 * that is associated with the given "istate".
3043 trace2_data_intmax("index", the_repository
, "write/version",
3045 trace2_data_intmax("index", the_repository
, "write/cache_nr",
3051 void set_alternate_index_output(const char *name
)
3053 alternate_index_output
= name
;
3056 static int commit_locked_index(struct lock_file
*lk
)
3058 if (alternate_index_output
)
3059 return commit_lock_file_to(lk
, alternate_index_output
);
3061 return commit_lock_file(lk
);
3064 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3068 int was_full
= !istate
->sparse_index
;
3070 ret
= convert_to_sparse(istate
);
3073 warning(_("failed to convert to a sparse-index"));
3078 * TODO trace2: replace "the_repository" with the actual repo instance
3079 * that is associated with the given "istate".
3081 trace2_region_enter_printf("index", "do_write_index", the_repository
,
3082 "%s", get_lock_file_path(lock
));
3083 ret
= do_write_index(istate
, lock
->tempfile
, 0);
3084 trace2_region_leave_printf("index", "do_write_index", the_repository
,
3085 "%s", get_lock_file_path(lock
));
3088 ensure_full_index(istate
);
3092 if (flags
& COMMIT_LOCK
)
3093 ret
= commit_locked_index(lock
);
3095 ret
= close_lock_file_gently(lock
);
3097 run_hook_le(NULL
, "post-index-change",
3098 istate
->updated_workdir
? "1" : "0",
3099 istate
->updated_skipworktree
? "1" : "0", NULL
);
3100 istate
->updated_workdir
= 0;
3101 istate
->updated_skipworktree
= 0;
3106 static int write_split_index(struct index_state
*istate
,
3107 struct lock_file
*lock
,
3111 prepare_to_write_split_index(istate
);
3112 ret
= do_write_locked_index(istate
, lock
, flags
);
3113 finish_writing_split_index(istate
);
3117 static const char *shared_index_expire
= "2.weeks.ago";
3119 static unsigned long get_shared_index_expire_date(void)
3121 static unsigned long shared_index_expire_date
;
3122 static int shared_index_expire_date_prepared
;
3124 if (!shared_index_expire_date_prepared
) {
3125 git_config_get_expiry("splitindex.sharedindexexpire",
3126 &shared_index_expire
);
3127 shared_index_expire_date
= approxidate(shared_index_expire
);
3128 shared_index_expire_date_prepared
= 1;
3131 return shared_index_expire_date
;
3134 static int should_delete_shared_index(const char *shared_index_path
)
3137 unsigned long expiration
;
3139 /* Check timestamp */
3140 expiration
= get_shared_index_expire_date();
3143 if (stat(shared_index_path
, &st
))
3144 return error_errno(_("could not stat '%s'"), shared_index_path
);
3145 if (st
.st_mtime
> expiration
)
3151 static int clean_shared_index_files(const char *current_hex
)
3154 DIR *dir
= opendir(get_git_dir());
3157 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3159 while ((de
= readdir(dir
)) != NULL
) {
3160 const char *sha1_hex
;
3161 const char *shared_index_path
;
3162 if (!skip_prefix(de
->d_name
, "sharedindex.", &sha1_hex
))
3164 if (!strcmp(sha1_hex
, current_hex
))
3166 shared_index_path
= git_path("%s", de
->d_name
);
3167 if (should_delete_shared_index(shared_index_path
) > 0 &&
3168 unlink(shared_index_path
))
3169 warning_errno(_("unable to unlink: %s"), shared_index_path
);
3176 static int write_shared_index(struct index_state
*istate
,
3177 struct tempfile
**temp
)
3179 struct split_index
*si
= istate
->split_index
;
3180 int ret
, was_full
= !istate
->sparse_index
;
3182 move_cache_to_base_index(istate
);
3183 convert_to_sparse(istate
);
3185 trace2_region_enter_printf("index", "shared/do_write_index",
3186 the_repository
, "%s", get_tempfile_path(*temp
));
3187 ret
= do_write_index(si
->base
, *temp
, 1);
3188 trace2_region_leave_printf("index", "shared/do_write_index",
3189 the_repository
, "%s", get_tempfile_path(*temp
));
3192 ensure_full_index(istate
);
3196 ret
= adjust_shared_perm(get_tempfile_path(*temp
));
3198 error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp
));
3201 ret
= rename_tempfile(temp
,
3202 git_path("sharedindex.%s", oid_to_hex(&si
->base
->oid
)));
3204 oidcpy(&si
->base_oid
, &si
->base
->oid
);
3205 clean_shared_index_files(oid_to_hex(&si
->base
->oid
));
3211 static const int default_max_percent_split_change
= 20;
3213 static int too_many_not_shared_entries(struct index_state
*istate
)
3215 int i
, not_shared
= 0;
3216 int max_split
= git_config_get_max_percent_split_change();
3218 switch (max_split
) {
3220 /* not or badly configured: use the default value */
3221 max_split
= default_max_percent_split_change
;
3224 return 1; /* 0% means always write a new shared index */
3226 return 0; /* 100% means never write a new shared index */
3228 break; /* just use the configured value */
3231 /* Count not shared entries */
3232 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3233 struct cache_entry
*ce
= istate
->cache
[i
];
3238 return (int64_t)istate
->cache_nr
* max_split
< (int64_t)not_shared
* 100;
3241 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3244 int new_shared_index
, ret
;
3245 struct split_index
*si
= istate
->split_index
;
3247 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3248 cache_tree_verify(the_repository
, istate
);
3250 if ((flags
& SKIP_IF_UNCHANGED
) && !istate
->cache_changed
) {
3251 if (flags
& COMMIT_LOCK
)
3252 rollback_lock_file(lock
);
3256 if (istate
->fsmonitor_last_update
)
3257 fill_fsmonitor_bitmap(istate
);
3259 if (!si
|| alternate_index_output
||
3260 (istate
->cache_changed
& ~EXTMASK
)) {
3262 oidclr(&si
->base_oid
);
3263 ret
= do_write_locked_index(istate
, lock
, flags
);
3267 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
3268 int v
= si
->base_oid
.hash
[0];
3270 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3272 if (too_many_not_shared_entries(istate
))
3273 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3275 new_shared_index
= istate
->cache_changed
& SPLIT_INDEX_ORDERED
;
3277 if (new_shared_index
) {
3278 struct tempfile
*temp
;
3281 /* Same initial permissions as the main .git/index file */
3282 temp
= mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);
3284 oidclr(&si
->base_oid
);
3285 ret
= do_write_locked_index(istate
, lock
, flags
);
3288 ret
= write_shared_index(istate
, &temp
);
3290 saved_errno
= errno
;
3291 if (is_tempfile_active(temp
))
3292 delete_tempfile(&temp
);
3293 errno
= saved_errno
;
3299 ret
= write_split_index(istate
, lock
, flags
);
3301 /* Freshen the shared index only if the split-index was written */
3302 if (!ret
&& !new_shared_index
&& !is_null_oid(&si
->base_oid
)) {
3303 const char *shared_index
= git_path("sharedindex.%s",
3304 oid_to_hex(&si
->base_oid
));
3305 freshen_shared_index(shared_index
, 1);
3309 if (flags
& COMMIT_LOCK
)
3310 rollback_lock_file(lock
);
3315 * Read the index file that is potentially unmerged into given
3316 * index_state, dropping any unmerged entries to stage #0 (potentially
3317 * resulting in a path appearing as both a file and a directory in the
3318 * index; the caller is responsible to clear out the extra entries
3319 * before writing the index to a tree). Returns true if the index is
3320 * unmerged. Callers who want to refuse to work from an unmerged
3321 * state can call this and check its return value, instead of calling
3324 int repo_read_index_unmerged(struct repository
*repo
)
3326 struct index_state
*istate
;
3330 repo_read_index(repo
);
3331 istate
= repo
->index
;
3332 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3333 struct cache_entry
*ce
= istate
->cache
[i
];
3334 struct cache_entry
*new_ce
;
3340 len
= ce_namelen(ce
);
3341 new_ce
= make_empty_cache_entry(istate
, len
);
3342 memcpy(new_ce
->name
, ce
->name
, len
);
3343 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
3344 new_ce
->ce_namelen
= len
;
3345 new_ce
->ce_mode
= ce
->ce_mode
;
3346 if (add_index_entry(istate
, new_ce
, ADD_CACHE_SKIP_DFCHECK
))
3347 return error(_("%s: cannot drop to stage #0"),
3354 * Returns 1 if the path is an "other" path with respect to
3355 * the index; that is, the path is not mentioned in the index at all,
3356 * either as a file, a directory with some files in the index,
3357 * or as an unmerged entry.
3359 * We helpfully remove a trailing "/" from directories so that
3360 * the output of read_directory can be used as-is.
3362 int index_name_is_other(struct index_state
*istate
, const char *name
,
3366 if (namelen
&& name
[namelen
- 1] == '/')
3368 pos
= index_name_pos(istate
, name
, namelen
);
3370 return 0; /* exact match */
3372 if (pos
< istate
->cache_nr
) {
3373 struct cache_entry
*ce
= istate
->cache
[pos
];
3374 if (ce_namelen(ce
) == namelen
&&
3375 !memcmp(ce
->name
, name
, namelen
))
3376 return 0; /* Yup, this one exists unmerged */
3381 void *read_blob_data_from_index(struct index_state
*istate
,
3382 const char *path
, unsigned long *size
)
3386 enum object_type type
;
3390 pos
= index_name_pos(istate
, path
, len
);
3393 * We might be in the middle of a merge, in which
3394 * case we would read stage #2 (ours).
3398 (pos
< 0 && i
< istate
->cache_nr
&&
3399 !strcmp(istate
->cache
[i
]->name
, path
));
3401 if (ce_stage(istate
->cache
[i
]) == 2)
3406 data
= read_object_file(&istate
->cache
[pos
]->oid
, &type
, &sz
);
3407 if (!data
|| type
!= OBJ_BLOB
) {
3416 void stat_validity_clear(struct stat_validity
*sv
)
3418 FREE_AND_NULL(sv
->sd
);
3421 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
3425 if (stat(path
, &st
) < 0)
3426 return sv
->sd
== NULL
;
3429 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
3432 void stat_validity_update(struct stat_validity
*sv
, int fd
)
3436 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
3437 stat_validity_clear(sv
);
3440 CALLOC_ARRAY(sv
->sd
, 1);
3441 fill_stat_data(sv
->sd
, &st
);
3445 void move_index_extensions(struct index_state
*dst
, struct index_state
*src
)
3447 dst
->untracked
= src
->untracked
;
3448 src
->untracked
= NULL
;
3449 dst
->cache_tree
= src
->cache_tree
;
3450 src
->cache_tree
= NULL
;
3453 struct cache_entry
*dup_cache_entry(const struct cache_entry
*ce
,
3454 struct index_state
*istate
)
3456 unsigned int size
= ce_size(ce
);
3457 int mem_pool_allocated
;
3458 struct cache_entry
*new_entry
= make_empty_cache_entry(istate
, ce_namelen(ce
));
3459 mem_pool_allocated
= new_entry
->mem_pool_allocated
;
3461 memcpy(new_entry
, ce
, size
);
3462 new_entry
->mem_pool_allocated
= mem_pool_allocated
;
3466 void discard_cache_entry(struct cache_entry
*ce
)
3468 if (ce
&& should_validate_cache_entries())
3469 memset(ce
, 0xCD, cache_entry_size(ce
->ce_namelen
));
3471 if (ce
&& ce
->mem_pool_allocated
)
3477 int should_validate_cache_entries(void)
3479 static int validate_index_cache_entries
= -1;
3481 if (validate_index_cache_entries
< 0) {
3482 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3483 validate_index_cache_entries
= 1;
3485 validate_index_cache_entries
= 0;
3488 return validate_index_cache_entries
;
3491 #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3492 #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3494 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
)
3497 * The end of index entries (EOIE) extension is guaranteed to be last
3498 * so that it can be found by scanning backwards from the EOF.
3505 const char *index
, *eoie
;
3507 size_t offset
, src_offset
;
3508 unsigned char hash
[GIT_MAX_RAWSZ
];
3511 /* ensure we have an index big enough to contain an EOIE extension */
3512 if (mmap_size
< sizeof(struct cache_header
) + EOIE_SIZE_WITH_HEADER
+ the_hash_algo
->rawsz
)
3515 /* validate the extension signature */
3516 index
= eoie
= mmap
+ mmap_size
- EOIE_SIZE_WITH_HEADER
- the_hash_algo
->rawsz
;
3517 if (CACHE_EXT(index
) != CACHE_EXT_ENDOFINDEXENTRIES
)
3519 index
+= sizeof(uint32_t);
3521 /* validate the extension size */
3522 extsize
= get_be32(index
);
3523 if (extsize
!= EOIE_SIZE
)
3525 index
+= sizeof(uint32_t);
3528 * Validate the offset we're going to look for the first extension
3529 * signature is after the index header and before the eoie extension.
3531 offset
= get_be32(index
);
3532 if (mmap
+ offset
< mmap
+ sizeof(struct cache_header
))
3534 if (mmap
+ offset
>= eoie
)
3536 index
+= sizeof(uint32_t);
3539 * The hash is computed over extension types and their sizes (but not
3540 * their contents). E.g. if we have "TREE" extension that is N-bytes
3541 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3542 * then the hash would be:
3544 * SHA-1("TREE" + <binary representation of N> +
3545 * "REUC" + <binary representation of M>)
3547 src_offset
= offset
;
3548 the_hash_algo
->init_fn(&c
);
3549 while (src_offset
< mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
) {
3550 /* After an array of active_nr index entries,
3551 * there can be arbitrary number of extended
3552 * sections, each of which is prefixed with
3553 * extension name (4-byte) and section length
3554 * in 4-byte network byte order.
3557 memcpy(&extsize
, mmap
+ src_offset
+ 4, 4);
3558 extsize
= ntohl(extsize
);
3560 /* verify the extension size isn't so large it will wrap around */
3561 if (src_offset
+ 8 + extsize
< src_offset
)
3564 the_hash_algo
->update_fn(&c
, mmap
+ src_offset
, 8);
3567 src_offset
+= extsize
;
3569 the_hash_algo
->final_fn(hash
, &c
);
3570 if (!hasheq(hash
, (const unsigned char *)index
))
3573 /* Validate that the extension offsets returned us back to the eoie extension. */
3574 if (src_offset
!= mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
)
3580 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
)
3583 unsigned char hash
[GIT_MAX_RAWSZ
];
3586 put_be32(&buffer
, offset
);
3587 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3590 the_hash_algo
->final_fn(hash
, eoie_context
);
3591 strbuf_add(sb
, hash
, the_hash_algo
->rawsz
);
3594 #define IEOT_VERSION (1)
3596 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
)
3598 const char *index
= NULL
;
3599 uint32_t extsize
, ext_version
;
3600 struct index_entry_offset_table
*ieot
;
3603 /* find the IEOT extension */
3606 while (offset
<= mmap_size
- the_hash_algo
->rawsz
- 8) {
3607 extsize
= get_be32(mmap
+ offset
+ 4);
3608 if (CACHE_EXT((mmap
+ offset
)) == CACHE_EXT_INDEXENTRYOFFSETTABLE
) {
3609 index
= mmap
+ offset
+ 4 + 4;
3618 /* validate the version is IEOT_VERSION */
3619 ext_version
= get_be32(index
);
3620 if (ext_version
!= IEOT_VERSION
) {
3621 error("invalid IEOT version %d", ext_version
);
3624 index
+= sizeof(uint32_t);
3626 /* extension size - version bytes / bytes per entry */
3627 nr
= (extsize
- sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3629 error("invalid number of IEOT entries %d", nr
);
3632 ieot
= xmalloc(sizeof(struct index_entry_offset_table
)
3633 + (nr
* sizeof(struct index_entry_offset
)));
3635 for (i
= 0; i
< nr
; i
++) {
3636 ieot
->entries
[i
].offset
= get_be32(index
);
3637 index
+= sizeof(uint32_t);
3638 ieot
->entries
[i
].nr
= get_be32(index
);
3639 index
+= sizeof(uint32_t);
3645 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
)
3651 put_be32(&buffer
, IEOT_VERSION
);
3652 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3655 for (i
= 0; i
< ieot
->nr
; i
++) {
3658 put_be32(&buffer
, ieot
->entries
[i
].offset
);
3659 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3662 put_be32(&buffer
, ieot
->entries
[i
].nr
);
3663 strbuf_add(sb
, &buffer
, sizeof(uint32_t));