2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
13 #include "cache-tree.h"
16 #include "object-store.h"
20 #include "resolve-undo.h"
23 #include "split-index.h"
25 #include "fsmonitor.h"
26 #include "thread-utils.h"
28 /* Mask for the name length in ce_flags in the on-disk index */
30 #define CE_NAMEMASK (0x0fff)
34 * The first letter should be 'A'..'Z' for extensions that are not
35 * necessary for a correct operation (i.e. optimization data).
36 * When new extensions are added that _needs_ to be understood in
37 * order to correctly interpret the index file, pick character that
38 * is outside the range, to cause the reader to abort.
41 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
42 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
43 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
44 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
45 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
46 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
47 #define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945 /* "EOIE" */
48 #define CACHE_EXT_INDEXENTRYOFFSETTABLE 0x49454F54 /* "IEOT" */
50 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
51 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
52 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
53 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
57 * This is an estimate of the pathname length in the index. We use
58 * this for V4 index files to guess the un-deltafied size of the index
59 * in memory because of pathname deltafication. This is not required
60 * for V2/V3 index formats because their pathnames are not compressed.
61 * If the initial amount of memory set aside is not sufficient, the
62 * mem pool will allocate extra memory.
64 #define CACHE_ENTRY_PATH_LENGTH 80
66 static inline struct cache_entry
*mem_pool__ce_alloc(struct mem_pool
*mem_pool
, size_t len
)
68 struct cache_entry
*ce
;
69 ce
= mem_pool_alloc(mem_pool
, cache_entry_size(len
));
70 ce
->mem_pool_allocated
= 1;
74 static inline struct cache_entry
*mem_pool__ce_calloc(struct mem_pool
*mem_pool
, size_t len
)
76 struct cache_entry
* ce
;
77 ce
= mem_pool_calloc(mem_pool
, 1, cache_entry_size(len
));
78 ce
->mem_pool_allocated
= 1;
82 static struct mem_pool
*find_mem_pool(struct index_state
*istate
)
84 struct mem_pool
**pool_ptr
;
86 if (istate
->split_index
&& istate
->split_index
->base
)
87 pool_ptr
= &istate
->split_index
->base
->ce_mem_pool
;
89 pool_ptr
= &istate
->ce_mem_pool
;
92 mem_pool_init(pool_ptr
, 0);
97 struct index_state the_index
;
98 static const char *alternate_index_output
;
100 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
102 istate
->cache
[nr
] = ce
;
103 add_name_hash(istate
, ce
);
106 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
108 struct cache_entry
*old
= istate
->cache
[nr
];
110 replace_index_entry_in_base(istate
, old
, ce
);
111 remove_name_hash(istate
, old
);
112 discard_cache_entry(old
);
113 ce
->ce_flags
&= ~CE_HASHED
;
114 set_index_entry(istate
, nr
, ce
);
115 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
116 mark_fsmonitor_invalid(istate
, ce
);
117 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
120 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
122 struct cache_entry
*old_entry
= istate
->cache
[nr
], *new_entry
;
123 int namelen
= strlen(new_name
);
125 new_entry
= make_empty_cache_entry(istate
, namelen
);
126 copy_cache_entry(new_entry
, old_entry
);
127 new_entry
->ce_flags
&= ~CE_HASHED
;
128 new_entry
->ce_namelen
= namelen
;
129 new_entry
->index
= 0;
130 memcpy(new_entry
->name
, new_name
, namelen
+ 1);
132 cache_tree_invalidate_path(istate
, old_entry
->name
);
133 untracked_cache_remove_from_index(istate
, old_entry
->name
);
134 remove_index_entry_at(istate
, nr
);
135 add_index_entry(istate
, new_entry
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
138 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
140 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
141 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
142 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
143 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
144 sd
->sd_dev
= st
->st_dev
;
145 sd
->sd_ino
= st
->st_ino
;
146 sd
->sd_uid
= st
->st_uid
;
147 sd
->sd_gid
= st
->st_gid
;
148 sd
->sd_size
= st
->st_size
;
151 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
155 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
156 changed
|= MTIME_CHANGED
;
157 if (trust_ctime
&& check_stat
&&
158 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
159 changed
|= CTIME_CHANGED
;
162 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
163 changed
|= MTIME_CHANGED
;
164 if (trust_ctime
&& check_stat
&&
165 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
166 changed
|= CTIME_CHANGED
;
170 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
171 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
172 changed
|= OWNER_CHANGED
;
173 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
174 changed
|= INODE_CHANGED
;
179 * st_dev breaks on network filesystems where different
180 * clients will have different views of what "device"
181 * the filesystem is on
183 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
184 changed
|= INODE_CHANGED
;
187 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
188 changed
|= DATA_CHANGED
;
194 * This only updates the "non-critical" parts of the directory
195 * cache, ie the parts that aren't tracked by GIT, and only used
196 * to validate the cache.
198 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
200 fill_stat_data(&ce
->ce_stat_data
, st
);
202 if (assume_unchanged
)
203 ce
->ce_flags
|= CE_VALID
;
205 if (S_ISREG(st
->st_mode
)) {
206 ce_mark_uptodate(ce
);
207 mark_fsmonitor_valid(ce
);
211 static int ce_compare_data(const struct cache_entry
*ce
, struct stat
*st
)
214 int fd
= git_open_cloexec(ce
->name
, O_RDONLY
);
217 struct object_id oid
;
218 if (!index_fd(&oid
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
219 match
= !oideq(&oid
, &ce
->oid
);
220 /* index_fd() closed the file descriptor already */
225 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
230 enum object_type type
;
231 struct strbuf sb
= STRBUF_INIT
;
233 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
236 buffer
= read_object_file(&ce
->oid
, &type
, &size
);
239 match
= memcmp(buffer
, sb
.buf
, size
);
246 static int ce_compare_gitlink(const struct cache_entry
*ce
)
248 struct object_id oid
;
251 * We don't actually require that the .git directory
252 * under GITLINK directory be a valid git directory. It
253 * might even be missing (in case nobody populated that
256 * If so, we consider it always to match.
258 if (resolve_gitlink_ref(ce
->name
, "HEAD", &oid
) < 0)
260 return !oideq(&oid
, &ce
->oid
);
263 static int ce_modified_check_fs(const struct cache_entry
*ce
, struct stat
*st
)
265 switch (st
->st_mode
& S_IFMT
) {
267 if (ce_compare_data(ce
, st
))
271 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
275 if (S_ISGITLINK(ce
->ce_mode
))
276 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
277 /* else fallthrough */
284 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
286 unsigned int changed
= 0;
288 if (ce
->ce_flags
& CE_REMOVE
)
289 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
291 switch (ce
->ce_mode
& S_IFMT
) {
293 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
294 /* We consider only the owner x bit to be relevant for
297 if (trust_executable_bit
&&
298 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
299 changed
|= MODE_CHANGED
;
302 if (!S_ISLNK(st
->st_mode
) &&
303 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
304 changed
|= TYPE_CHANGED
;
307 /* We ignore most of the st_xxx fields for gitlinks */
308 if (!S_ISDIR(st
->st_mode
))
309 changed
|= TYPE_CHANGED
;
310 else if (ce_compare_gitlink(ce
))
311 changed
|= DATA_CHANGED
;
314 die("internal error: ce_mode is %o", ce
->ce_mode
);
317 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
319 /* Racily smudged entry? */
320 if (!ce
->ce_stat_data
.sd_size
) {
321 if (!is_empty_blob_sha1(ce
->oid
.hash
))
322 changed
|= DATA_CHANGED
;
328 static int is_racy_stat(const struct index_state
*istate
,
329 const struct stat_data
*sd
)
331 return (istate
->timestamp
.sec
&&
333 /* nanosecond timestamped files can also be racy! */
334 (istate
->timestamp
.sec
< sd
->sd_mtime
.sec
||
335 (istate
->timestamp
.sec
== sd
->sd_mtime
.sec
&&
336 istate
->timestamp
.nsec
<= sd
->sd_mtime
.nsec
))
338 istate
->timestamp
.sec
<= sd
->sd_mtime
.sec
343 static int is_racy_timestamp(const struct index_state
*istate
,
344 const struct cache_entry
*ce
)
346 return (!S_ISGITLINK(ce
->ce_mode
) &&
347 is_racy_stat(istate
, &ce
->ce_stat_data
));
350 int match_stat_data_racy(const struct index_state
*istate
,
351 const struct stat_data
*sd
, struct stat
*st
)
353 if (is_racy_stat(istate
, sd
))
354 return MTIME_CHANGED
;
355 return match_stat_data(sd
, st
);
358 int ie_match_stat(struct index_state
*istate
,
359 const struct cache_entry
*ce
, struct stat
*st
,
360 unsigned int options
)
362 unsigned int changed
;
363 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
364 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
365 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
366 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
368 if (!ignore_fsmonitor
)
369 refresh_fsmonitor(istate
);
371 * If it's marked as always valid in the index, it's
372 * valid whatever the checked-out copy says.
374 * skip-worktree has the same effect with higher precedence
376 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
378 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
380 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
))
384 * Intent-to-add entries have not been added, so the index entry
385 * by definition never matches what is in the work tree until it
386 * actually gets added.
388 if (ce_intent_to_add(ce
))
389 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
391 changed
= ce_match_stat_basic(ce
, st
);
394 * Within 1 second of this sequence:
395 * echo xyzzy >file && git-update-index --add file
396 * running this command:
398 * would give a falsely clean cache entry. The mtime and
399 * length match the cache, and other stat fields do not change.
401 * We could detect this at update-index time (the cache entry
402 * being registered/updated records the same time as "now")
403 * and delay the return from git-update-index, but that would
404 * effectively mean we can make at most one commit per second,
405 * which is not acceptable. Instead, we check cache entries
406 * whose mtime are the same as the index file timestamp more
407 * carefully than others.
409 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
410 if (assume_racy_is_modified
)
411 changed
|= DATA_CHANGED
;
413 changed
|= ce_modified_check_fs(ce
, st
);
419 int ie_modified(struct index_state
*istate
,
420 const struct cache_entry
*ce
,
421 struct stat
*st
, unsigned int options
)
423 int changed
, changed_fs
;
425 changed
= ie_match_stat(istate
, ce
, st
, options
);
429 * If the mode or type has changed, there's no point in trying
430 * to refresh the entry - it's not going to match
432 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
436 * Immediately after read-tree or update-index --cacheinfo,
437 * the length field is zero, as we have never even read the
438 * lstat(2) information once, and we cannot trust DATA_CHANGED
439 * returned by ie_match_stat() which in turn was returned by
440 * ce_match_stat_basic() to signal that the filesize of the
441 * blob changed. We have to actually go to the filesystem to
442 * see if the contents match, and if so, should answer "unchanged".
444 * The logic does not apply to gitlinks, as ce_match_stat_basic()
445 * already has checked the actual HEAD from the filesystem in the
446 * subproject. If ie_match_stat() already said it is different,
447 * then we know it is.
449 if ((changed
& DATA_CHANGED
) &&
450 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
453 changed_fs
= ce_modified_check_fs(ce
, st
);
455 return changed
| changed_fs
;
459 int base_name_compare(const char *name1
, int len1
, int mode1
,
460 const char *name2
, int len2
, int mode2
)
462 unsigned char c1
, c2
;
463 int len
= len1
< len2
? len1
: len2
;
466 cmp
= memcmp(name1
, name2
, len
);
471 if (!c1
&& S_ISDIR(mode1
))
473 if (!c2
&& S_ISDIR(mode2
))
475 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
479 * df_name_compare() is identical to base_name_compare(), except it
480 * compares conflicting directory/file entries as equal. Note that
481 * while a directory name compares as equal to a regular file, they
482 * then individually compare _differently_ to a filename that has
483 * a dot after the basename (because '\0' < '.' < '/').
485 * This is used by routines that want to traverse the git namespace
486 * but then handle conflicting entries together when possible.
488 int df_name_compare(const char *name1
, int len1
, int mode1
,
489 const char *name2
, int len2
, int mode2
)
491 int len
= len1
< len2
? len1
: len2
, cmp
;
492 unsigned char c1
, c2
;
494 cmp
= memcmp(name1
, name2
, len
);
497 /* Directories and files compare equal (same length, same name) */
501 if (!c1
&& S_ISDIR(mode1
))
504 if (!c2
&& S_ISDIR(mode2
))
506 if (c1
== '/' && !c2
)
508 if (c2
== '/' && !c1
)
513 int name_compare(const char *name1
, size_t len1
, const char *name2
, size_t len2
)
515 size_t min_len
= (len1
< len2
) ? len1
: len2
;
516 int cmp
= memcmp(name1
, name2
, min_len
);
526 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
530 cmp
= name_compare(name1
, len1
, name2
, len2
);
541 static int index_name_stage_pos(const struct index_state
*istate
, const char *name
, int namelen
, int stage
)
546 last
= istate
->cache_nr
;
547 while (last
> first
) {
548 int next
= (last
+ first
) >> 1;
549 struct cache_entry
*ce
= istate
->cache
[next
];
550 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
562 int index_name_pos(const struct index_state
*istate
, const char *name
, int namelen
)
564 return index_name_stage_pos(istate
, name
, namelen
, 0);
567 int remove_index_entry_at(struct index_state
*istate
, int pos
)
569 struct cache_entry
*ce
= istate
->cache
[pos
];
571 record_resolve_undo(istate
, ce
);
572 remove_name_hash(istate
, ce
);
573 save_or_free_index_entry(istate
, ce
);
574 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
576 if (pos
>= istate
->cache_nr
)
578 MOVE_ARRAY(istate
->cache
+ pos
, istate
->cache
+ pos
+ 1,
579 istate
->cache_nr
- pos
);
584 * Remove all cache entries marked for removal, that is where
585 * CE_REMOVE is set in ce_flags. This is much more effective than
586 * calling remove_index_entry_at() for each entry to be removed.
588 void remove_marked_cache_entries(struct index_state
*istate
)
590 struct cache_entry
**ce_array
= istate
->cache
;
593 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
594 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
595 remove_name_hash(istate
, ce_array
[i
]);
596 save_or_free_index_entry(istate
, ce_array
[i
]);
599 ce_array
[j
++] = ce_array
[i
];
601 if (j
== istate
->cache_nr
)
603 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
604 istate
->cache_nr
= j
;
607 int remove_file_from_index(struct index_state
*istate
, const char *path
)
609 int pos
= index_name_pos(istate
, path
, strlen(path
));
612 cache_tree_invalidate_path(istate
, path
);
613 untracked_cache_remove_from_index(istate
, path
);
614 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
615 remove_index_entry_at(istate
, pos
);
619 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
621 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
624 static int index_name_pos_also_unmerged(struct index_state
*istate
,
625 const char *path
, int namelen
)
627 int pos
= index_name_pos(istate
, path
, namelen
);
628 struct cache_entry
*ce
;
633 /* maybe unmerged? */
635 if (pos
>= istate
->cache_nr
||
636 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
639 /* order of preference: stage 2, 1, 3 */
640 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
641 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
642 !compare_name(ce
, path
, namelen
))
647 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
649 int len
= ce_namelen(ce
);
650 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
654 * If we add a filename that aliases in the cache, we will use the
655 * name that we already have - but we don't want to update the same
656 * alias twice, because that implies that there were actually two
657 * different files with aliasing names!
659 * So we use the CE_ADDED flag to verify that the alias was an old
660 * one before we accept it as
662 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
663 struct cache_entry
*ce
,
664 struct cache_entry
*alias
)
667 struct cache_entry
*new_entry
;
669 if (alias
->ce_flags
& CE_ADDED
)
670 die("Will not add file alias '%s' ('%s' already exists in index)", ce
->name
, alias
->name
);
672 /* Ok, create the new entry using the name of the existing alias */
673 len
= ce_namelen(alias
);
674 new_entry
= make_empty_cache_entry(istate
, len
);
675 memcpy(new_entry
->name
, alias
->name
, len
);
676 copy_cache_entry(new_entry
, ce
);
677 save_or_free_index_entry(istate
, ce
);
681 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
683 struct object_id oid
;
684 if (write_object_file("", 0, blob_type
, &oid
))
685 die("cannot create an empty blob in the object database");
686 oidcpy(&ce
->oid
, &oid
);
689 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
691 int namelen
, was_same
;
692 mode_t st_mode
= st
->st_mode
;
693 struct cache_entry
*ce
, *alias
= NULL
;
694 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
695 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
696 int pretend
= flags
& ADD_CACHE_PRETEND
;
697 int intent_only
= flags
& ADD_CACHE_INTENT
;
698 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
699 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
700 int newflags
= HASH_WRITE_OBJECT
;
702 if (flags
& HASH_RENORMALIZE
)
703 newflags
|= HASH_RENORMALIZE
;
705 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
706 return error("%s: can only add regular files, symbolic links or git-directories", path
);
708 namelen
= strlen(path
);
709 if (S_ISDIR(st_mode
)) {
710 while (namelen
&& path
[namelen
-1] == '/')
713 ce
= make_empty_cache_entry(istate
, namelen
);
714 memcpy(ce
->name
, path
, namelen
);
715 ce
->ce_namelen
= namelen
;
717 fill_stat_cache_info(ce
, st
);
719 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
722 if (trust_executable_bit
&& has_symlinks
) {
723 ce
->ce_mode
= create_ce_mode(st_mode
);
725 /* If there is an existing entry, pick the mode bits and type
726 * from it, otherwise assume unexecutable regular file.
728 struct cache_entry
*ent
;
729 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
731 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
732 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
735 /* When core.ignorecase=true, determine if a directory of the same name but differing
736 * case already exists within the Git repository. If it does, ensure the directory
737 * case of the file being added to the repository matches (is folded into) the existing
738 * entry's directory case.
741 adjust_dirname_case(istate
, ce
->name
);
743 if (!(flags
& HASH_RENORMALIZE
)) {
744 alias
= index_file_exists(istate
, ce
->name
,
745 ce_namelen(ce
), ignore_case
);
748 !ie_match_stat(istate
, alias
, st
, ce_option
)) {
749 /* Nothing changed, really */
750 if (!S_ISGITLINK(alias
->ce_mode
))
751 ce_mark_uptodate(alias
);
752 alias
->ce_flags
|= CE_ADDED
;
754 discard_cache_entry(ce
);
759 if (index_path(&ce
->oid
, path
, st
, newflags
)) {
760 discard_cache_entry(ce
);
761 return error("unable to index file %s", path
);
764 set_object_name_for_intent_to_add_entry(ce
);
766 if (ignore_case
&& alias
&& different_name(ce
, alias
))
767 ce
= create_alias_ce(istate
, ce
, alias
);
768 ce
->ce_flags
|= CE_ADDED
;
770 /* It was suspected to be racily clean, but it turns out to be Ok */
773 oideq(&alias
->oid
, &ce
->oid
) &&
774 ce
->ce_mode
== alias
->ce_mode
);
777 discard_cache_entry(ce
);
778 else if (add_index_entry(istate
, ce
, add_option
)) {
779 discard_cache_entry(ce
);
780 return error("unable to add %s to index", path
);
782 if (verbose
&& !was_same
)
783 printf("add '%s'\n", path
);
787 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
790 if (lstat(path
, &st
))
791 die_errno("unable to stat '%s'", path
);
792 return add_to_index(istate
, path
, &st
, flags
);
795 struct cache_entry
*make_empty_cache_entry(struct index_state
*istate
, size_t len
)
797 return mem_pool__ce_calloc(find_mem_pool(istate
), len
);
800 struct cache_entry
*make_empty_transient_cache_entry(size_t len
)
802 return xcalloc(1, cache_entry_size(len
));
805 struct cache_entry
*make_cache_entry(struct index_state
*istate
,
807 const struct object_id
*oid
,
810 unsigned int refresh_options
)
812 struct cache_entry
*ce
, *ret
;
815 if (!verify_path(path
, mode
)) {
816 error("Invalid path '%s'", path
);
821 ce
= make_empty_cache_entry(istate
, len
);
823 oidcpy(&ce
->oid
, oid
);
824 memcpy(ce
->name
, path
, len
);
825 ce
->ce_flags
= create_ce_flags(stage
);
826 ce
->ce_namelen
= len
;
827 ce
->ce_mode
= create_ce_mode(mode
);
829 ret
= refresh_cache_entry(&the_index
, ce
, refresh_options
);
831 discard_cache_entry(ce
);
835 struct cache_entry
*make_transient_cache_entry(unsigned int mode
, const struct object_id
*oid
,
836 const char *path
, int stage
)
838 struct cache_entry
*ce
;
841 if (!verify_path(path
, mode
)) {
842 error("Invalid path '%s'", path
);
847 ce
= make_empty_transient_cache_entry(len
);
849 oidcpy(&ce
->oid
, oid
);
850 memcpy(ce
->name
, path
, len
);
851 ce
->ce_flags
= create_ce_flags(stage
);
852 ce
->ce_namelen
= len
;
853 ce
->ce_mode
= create_ce_mode(mode
);
859 * Chmod an index entry with either +x or -x.
861 * Returns -1 if the chmod for the particular cache entry failed (if it's
862 * not a regular file), -2 if an invalid flip argument is passed in, 0
865 int chmod_index_entry(struct index_state
*istate
, struct cache_entry
*ce
,
868 if (!S_ISREG(ce
->ce_mode
))
875 ce
->ce_mode
&= ~0111;
880 cache_tree_invalidate_path(istate
, ce
->name
);
881 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
882 mark_fsmonitor_invalid(istate
, ce
);
883 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
888 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
890 int len
= ce_namelen(a
);
891 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
895 * We fundamentally don't like some paths: we don't want
896 * dot or dot-dot anywhere, and for obvious reasons don't
897 * want to recurse into ".git" either.
899 * Also, we don't want double slashes or slashes at the
900 * end that can make pathnames ambiguous.
902 static int verify_dotfile(const char *rest
, unsigned mode
)
905 * The first character was '.', but that
906 * has already been discarded, we now test
910 /* "." is not allowed */
911 if (*rest
== '\0' || is_dir_sep(*rest
))
916 * ".git" followed by NUL or slash is bad. Note that we match
917 * case-insensitively here, even if ignore_case is not set.
918 * This outlaws ".GIT" everywhere out of an abundance of caution,
919 * since there's really no good reason to allow it.
921 * Once we've seen ".git", we can also find ".gitmodules", etc (also
922 * case-insensitively).
926 if (rest
[1] != 'i' && rest
[1] != 'I')
928 if (rest
[2] != 't' && rest
[2] != 'T')
930 if (rest
[3] == '\0' || is_dir_sep(rest
[3]))
934 if (skip_iprefix(rest
, "modules", &rest
) &&
935 (*rest
== '\0' || is_dir_sep(*rest
)))
940 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
946 int verify_path(const char *path
, unsigned mode
)
950 if (has_dos_drive_prefix(path
))
960 if (is_hfs_dotgit(path
))
963 if (is_hfs_dotgitmodules(path
))
968 if (is_ntfs_dotgit(path
))
971 if (is_ntfs_dotgitmodules(path
))
977 if ((c
== '.' && !verify_dotfile(path
, mode
)) ||
978 is_dir_sep(c
) || c
== '\0')
986 * Do we have another file that has the beginning components being a
987 * proper superset of the name we're trying to add?
989 static int has_file_name(struct index_state
*istate
,
990 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
993 int len
= ce_namelen(ce
);
994 int stage
= ce_stage(ce
);
995 const char *name
= ce
->name
;
997 while (pos
< istate
->cache_nr
) {
998 struct cache_entry
*p
= istate
->cache
[pos
++];
1000 if (len
>= ce_namelen(p
))
1002 if (memcmp(name
, p
->name
, len
))
1004 if (ce_stage(p
) != stage
)
1006 if (p
->name
[len
] != '/')
1008 if (p
->ce_flags
& CE_REMOVE
)
1013 remove_index_entry_at(istate
, --pos
);
1020 * Like strcmp(), but also return the offset of the first change.
1021 * If strings are equal, return the length.
1023 int strcmp_offset(const char *s1
, const char *s2
, size_t *first_change
)
1028 return strcmp(s1
, s2
);
1030 for (k
= 0; s1
[k
] == s2
[k
]; k
++)
1035 return (unsigned char)s1
[k
] - (unsigned char)s2
[k
];
1039 * Do we have another file with a pathname that is a proper
1040 * subset of the name we're trying to add?
1042 * That is, is there another file in the index with a path
1043 * that matches a sub-directory in the given entry?
1045 static int has_dir_name(struct index_state
*istate
,
1046 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1049 int stage
= ce_stage(ce
);
1050 const char *name
= ce
->name
;
1051 const char *slash
= name
+ ce_namelen(ce
);
1056 * We are frequently called during an iteration on a sorted
1057 * list of pathnames and while building a new index. Therefore,
1058 * there is a high probability that this entry will eventually
1059 * be appended to the index, rather than inserted in the middle.
1060 * If we can confirm that, we can avoid binary searches on the
1061 * components of the pathname.
1063 * Compare the entry's full path with the last path in the index.
1065 if (istate
->cache_nr
> 0) {
1066 cmp_last
= strcmp_offset(name
,
1067 istate
->cache
[istate
->cache_nr
- 1]->name
,
1070 if (len_eq_last
== 0) {
1072 * The entry sorts AFTER the last one in the
1073 * index and their paths have no common prefix,
1074 * so there cannot be a F/D conflict.
1079 * The entry sorts AFTER the last one in the
1080 * index, but has a common prefix. Fall through
1081 * to the loop below to disect the entry's path
1082 * and see where the difference is.
1085 } else if (cmp_last
== 0) {
1087 * The entry exactly matches the last one in the
1088 * index, but because of multiple stage and CE_REMOVE
1089 * items, we fall through and let the regular search
1099 if (*--slash
== '/')
1101 if (slash
<= ce
->name
)
1108 * (len + 1) is a directory boundary (including
1109 * the trailing slash). And since the loop is
1110 * decrementing "slash", the first iteration is
1111 * the longest directory prefix; subsequent
1112 * iterations consider parent directories.
1115 if (len
+ 1 <= len_eq_last
) {
1117 * The directory prefix (including the trailing
1118 * slash) also appears as a prefix in the last
1119 * entry, so the remainder cannot collide (because
1120 * strcmp said the whole path was greater).
1125 * LT: last: xxx/file_A
1131 if (len
> len_eq_last
) {
1133 * This part of the directory prefix (excluding
1134 * the trailing slash) is longer than the known
1135 * equal portions, so this sub-directory cannot
1136 * collide with a file.
1144 if (istate
->cache_nr
> 0 &&
1145 ce_namelen(istate
->cache
[istate
->cache_nr
- 1]) > len
) {
1147 * The directory prefix lines up with part of
1148 * a longer file or directory name, but sorts
1149 * after it, so this sub-directory cannot
1150 * collide with a file.
1152 * last: xxx/yy-file (because '-' sorts before '/')
1159 * This is a possible collision. Fall through and
1160 * let the regular search code handle it.
1167 pos
= index_name_stage_pos(istate
, name
, len
, stage
);
1170 * Found one, but not so fast. This could
1171 * be a marker that says "I was here, but
1172 * I am being removed". Such an entry is
1173 * not a part of the resulting tree, and
1174 * it is Ok to have a directory at the same
1177 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
1181 remove_index_entry_at(istate
, pos
);
1189 * Trivial optimization: if we find an entry that
1190 * already matches the sub-directory, then we know
1191 * we're ok, and we can exit.
1193 while (pos
< istate
->cache_nr
) {
1194 struct cache_entry
*p
= istate
->cache
[pos
];
1195 if ((ce_namelen(p
) <= len
) ||
1196 (p
->name
[len
] != '/') ||
1197 memcmp(p
->name
, name
, len
))
1198 break; /* not our subdirectory */
1199 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
1201 * p is at the same stage as our entry, and
1202 * is a subdirectory of what we are looking
1203 * at, so we cannot have conflicts at our
1204 * level or anything shorter.
1213 /* We may be in a situation where we already have path/file and path
1214 * is being added, or we already have path and path/file is being
1215 * added. Either one would result in a nonsense tree that has path
1216 * twice when git-write-tree tries to write it out. Prevent it.
1218 * If ok-to-replace is specified, we remove the conflicting entries
1219 * from the cache so the caller should recompute the insert position.
1220 * When this happens, we return non-zero.
1222 static int check_file_directory_conflict(struct index_state
*istate
,
1223 const struct cache_entry
*ce
,
1224 int pos
, int ok_to_replace
)
1229 * When ce is an "I am going away" entry, we allow it to be added
1231 if (ce
->ce_flags
& CE_REMOVE
)
1235 * We check if the path is a sub-path of a subsequent pathname
1236 * first, since removing those will not change the position
1239 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
1242 * Then check if the path might have a clashing sub-directory
1245 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
1248 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1251 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
1252 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
1253 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
1254 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
1256 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1257 cache_tree_invalidate_path(istate
, ce
->name
);
1260 * If this entry's path sorts after the last entry in the index,
1261 * we can avoid searching for it.
1263 if (istate
->cache_nr
> 0 &&
1264 strcmp(ce
->name
, istate
->cache
[istate
->cache_nr
- 1]->name
) > 0)
1265 pos
= -istate
->cache_nr
- 1;
1267 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1269 /* existing match? Just replace it. */
1272 replace_index_entry(istate
, pos
, ce
);
1277 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1278 untracked_cache_add_to_index(istate
, ce
->name
);
1281 * Inserting a merged entry ("stage 0") into the index
1282 * will always replace all non-merged entries..
1284 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
1285 while (ce_same_name(istate
->cache
[pos
], ce
)) {
1287 if (!remove_index_entry_at(istate
, pos
))
1294 if (!verify_path(ce
->name
, ce
->ce_mode
))
1295 return error("Invalid path '%s'", ce
->name
);
1297 if (!skip_df_check
&&
1298 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1300 return error("'%s' appears as both a file and as a directory",
1302 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1308 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1312 if (option
& ADD_CACHE_JUST_APPEND
)
1313 pos
= istate
->cache_nr
;
1316 ret
= add_index_entry_with_check(istate
, ce
, option
);
1322 /* Make sure the array is big enough .. */
1323 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1327 if (istate
->cache_nr
> pos
+ 1)
1328 MOVE_ARRAY(istate
->cache
+ pos
+ 1, istate
->cache
+ pos
,
1329 istate
->cache_nr
- pos
- 1);
1330 set_index_entry(istate
, pos
, ce
);
1331 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1336 * "refresh" does not calculate a new sha1 file or bring the
1337 * cache up-to-date for mode/content changes. But what it
1338 * _does_ do is to "re-match" the stat information of a file
1339 * with the cache, so that you can refresh the cache for a
1340 * file that hasn't been changed but where the stat entry is
1343 * For example, you'd want to do this after doing a "git-read-tree",
1344 * to link up the stat cache details with the proper files.
1346 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1347 struct cache_entry
*ce
,
1348 unsigned int options
, int *err
,
1352 struct cache_entry
*updated
;
1354 int refresh
= options
& CE_MATCH_REFRESH
;
1355 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1356 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1357 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1358 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
1360 if (!refresh
|| ce_uptodate(ce
))
1363 if (!ignore_fsmonitor
)
1364 refresh_fsmonitor(istate
);
1366 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1367 * that the change to the work tree does not matter and told
1370 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1371 ce_mark_uptodate(ce
);
1374 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1375 ce_mark_uptodate(ce
);
1378 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
)) {
1379 ce_mark_uptodate(ce
);
1383 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1391 if (lstat(ce
->name
, &st
) < 0) {
1392 if (ignore_missing
&& errno
== ENOENT
)
1399 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1401 *changed_ret
= changed
;
1404 * The path is unchanged. If we were told to ignore
1405 * valid bit, then we did the actual stat check and
1406 * found that the entry is unmodified. If the entry
1407 * is not marked VALID, this is the place to mark it
1408 * valid again, under "assume unchanged" mode.
1410 if (ignore_valid
&& assume_unchanged
&&
1411 !(ce
->ce_flags
& CE_VALID
))
1412 ; /* mark this one VALID again */
1415 * We do not mark the index itself "modified"
1416 * because CE_UPTODATE flag is in-core only;
1417 * we are not going to write this change out.
1419 if (!S_ISGITLINK(ce
->ce_mode
)) {
1420 ce_mark_uptodate(ce
);
1421 mark_fsmonitor_valid(ce
);
1427 if (ie_modified(istate
, ce
, &st
, options
)) {
1433 updated
= make_empty_cache_entry(istate
, ce_namelen(ce
));
1434 copy_cache_entry(updated
, ce
);
1435 memcpy(updated
->name
, ce
->name
, ce
->ce_namelen
+ 1);
1436 fill_stat_cache_info(updated
, &st
);
1438 * If ignore_valid is not set, we should leave CE_VALID bit
1439 * alone. Otherwise, paths marked with --no-assume-unchanged
1440 * (i.e. things to be edited) will reacquire CE_VALID bit
1441 * automatically, which is not really what we want.
1443 if (!ignore_valid
&& assume_unchanged
&&
1444 !(ce
->ce_flags
& CE_VALID
))
1445 updated
->ce_flags
&= ~CE_VALID
;
1447 /* istate->cache_changed is updated in the caller */
1451 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1452 int * first
, const char *header_msg
)
1454 if (in_porcelain
&& *first
&& header_msg
) {
1455 printf("%s\n", header_msg
);
1461 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1462 const struct pathspec
*pathspec
,
1463 char *seen
, const char *header_msg
)
1467 int really
= (flags
& REFRESH_REALLY
) != 0;
1468 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1469 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1470 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1471 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1473 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1474 unsigned int options
= (CE_MATCH_REFRESH
|
1475 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1476 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1477 const char *modified_fmt
;
1478 const char *deleted_fmt
;
1479 const char *typechange_fmt
;
1480 const char *added_fmt
;
1481 const char *unmerged_fmt
;
1483 trace_performance_enter();
1484 modified_fmt
= (in_porcelain
? "M\t%s\n" : "%s: needs update\n");
1485 deleted_fmt
= (in_porcelain
? "D\t%s\n" : "%s: needs update\n");
1486 typechange_fmt
= (in_porcelain
? "T\t%s\n" : "%s needs update\n");
1487 added_fmt
= (in_porcelain
? "A\t%s\n" : "%s needs update\n");
1488 unmerged_fmt
= (in_porcelain
? "U\t%s\n" : "%s: needs merge\n");
1489 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1490 struct cache_entry
*ce
, *new_entry
;
1491 int cache_errno
= 0;
1495 ce
= istate
->cache
[i
];
1496 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1499 if (pathspec
&& !ce_path_match(&the_index
, ce
, pathspec
, seen
))
1503 while ((i
< istate
->cache_nr
) &&
1504 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1510 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1511 &first
, header_msg
);
1519 new_entry
= refresh_cache_ent(istate
, ce
, options
, &cache_errno
, &changed
);
1520 if (new_entry
== ce
)
1525 if (really
&& cache_errno
== EINVAL
) {
1526 /* If we are doing --really-refresh that
1527 * means the index is not valid anymore.
1529 ce
->ce_flags
&= ~CE_VALID
;
1530 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1531 mark_fsmonitor_invalid(istate
, ce
);
1532 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1537 if (cache_errno
== ENOENT
)
1539 else if (ce_intent_to_add(ce
))
1540 fmt
= added_fmt
; /* must be before other checks */
1541 else if (changed
& TYPE_CHANGED
)
1542 fmt
= typechange_fmt
;
1546 ce
->name
, in_porcelain
, &first
, header_msg
);
1551 replace_index_entry(istate
, i
, new_entry
);
1553 trace_performance_leave("refresh index");
1557 struct cache_entry
*refresh_cache_entry(struct index_state
*istate
,
1558 struct cache_entry
*ce
,
1559 unsigned int options
)
1561 return refresh_cache_ent(istate
, ce
, options
, NULL
, NULL
);
1565 /*****************************************************************
1567 *****************************************************************/
1569 #define INDEX_FORMAT_DEFAULT 3
1571 static unsigned int get_index_format_default(void)
1573 char *envversion
= getenv("GIT_INDEX_VERSION");
1576 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1579 if (!git_config_get_int("index.version", &value
))
1581 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1582 warning(_("index.version set, but the value is invalid.\n"
1583 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1584 return INDEX_FORMAT_DEFAULT
;
1589 version
= strtoul(envversion
, &endp
, 10);
1591 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1592 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1593 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1594 version
= INDEX_FORMAT_DEFAULT
;
1600 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1601 * Again - this is just a (very strong in practice) heuristic that
1602 * the inode hasn't changed.
1604 * We save the fields in big-endian order to allow using the
1605 * index file over NFS transparently.
1607 struct ondisk_cache_entry
{
1608 struct cache_time ctime
;
1609 struct cache_time mtime
;
1616 unsigned char sha1
[20];
1618 char name
[FLEX_ARRAY
]; /* more */
1622 * This struct is used when CE_EXTENDED bit is 1
1623 * The struct must match ondisk_cache_entry exactly from
1626 struct ondisk_cache_entry_extended
{
1627 struct cache_time ctime
;
1628 struct cache_time mtime
;
1635 unsigned char sha1
[20];
1638 char name
[FLEX_ARRAY
]; /* more */
1641 /* These are only used for v3 or lower */
1642 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1643 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1644 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1645 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1646 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1647 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1648 ondisk_cache_entry_size(ce_namelen(ce)))
1650 /* Allow fsck to force verification of the index checksum. */
1651 int verify_index_checksum
;
1653 /* Allow fsck to force verification of the cache entry order. */
1654 int verify_ce_order
;
1656 static int verify_hdr(const struct cache_header
*hdr
, unsigned long size
)
1659 unsigned char hash
[GIT_MAX_RAWSZ
];
1662 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1663 return error("bad signature");
1664 hdr_version
= ntohl(hdr
->hdr_version
);
1665 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1666 return error("bad index version %d", hdr_version
);
1668 if (!verify_index_checksum
)
1671 the_hash_algo
->init_fn(&c
);
1672 the_hash_algo
->update_fn(&c
, hdr
, size
- the_hash_algo
->rawsz
);
1673 the_hash_algo
->final_fn(hash
, &c
);
1674 if (!hasheq(hash
, (unsigned char *)hdr
+ size
- the_hash_algo
->rawsz
))
1675 return error("bad index file sha1 signature");
1679 static int read_index_extension(struct index_state
*istate
,
1680 const char *ext
, const char *data
, unsigned long sz
)
1682 switch (CACHE_EXT(ext
)) {
1683 case CACHE_EXT_TREE
:
1684 istate
->cache_tree
= cache_tree_read(data
, sz
);
1686 case CACHE_EXT_RESOLVE_UNDO
:
1687 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1689 case CACHE_EXT_LINK
:
1690 if (read_link_extension(istate
, data
, sz
))
1693 case CACHE_EXT_UNTRACKED
:
1694 istate
->untracked
= read_untracked_extension(data
, sz
);
1696 case CACHE_EXT_FSMONITOR
:
1697 read_fsmonitor_extension(istate
, data
, sz
);
1699 case CACHE_EXT_ENDOFINDEXENTRIES
:
1700 case CACHE_EXT_INDEXENTRYOFFSETTABLE
:
1701 /* already handled in do_read_index() */
1704 if (*ext
< 'A' || 'Z' < *ext
)
1705 return error("index uses %.4s extension, which we do not understand",
1707 fprintf(stderr
, "ignoring %.4s extension\n", ext
);
1713 int hold_locked_index(struct lock_file
*lk
, int lock_flags
)
1715 return hold_lock_file_for_update(lk
, get_index_file(), lock_flags
);
1718 int read_index(struct index_state
*istate
)
1720 return read_index_from(istate
, get_index_file(), get_git_dir());
1723 static struct cache_entry
*create_from_disk(struct mem_pool
*ce_mem_pool
,
1724 unsigned int version
,
1725 struct ondisk_cache_entry
*ondisk
,
1726 unsigned long *ent_size
,
1727 const struct cache_entry
*previous_ce
)
1729 struct cache_entry
*ce
;
1735 * Adjacent cache entries tend to share the leading paths, so it makes
1736 * sense to only store the differences in later entries. In the v4
1737 * on-disk format of the index, each on-disk cache entry stores the
1738 * number of bytes to be stripped from the end of the previous name,
1739 * and the bytes to append to the result, to come up with its name.
1741 int expand_name_field
= version
== 4;
1743 /* On-disk flags are just 16 bits */
1744 flags
= get_be16(&ondisk
->flags
);
1745 len
= flags
& CE_NAMEMASK
;
1747 if (flags
& CE_EXTENDED
) {
1748 struct ondisk_cache_entry_extended
*ondisk2
;
1750 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1751 extended_flags
= get_be16(&ondisk2
->flags2
) << 16;
1752 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1753 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1754 die("Unknown index entry format %08x", extended_flags
);
1755 flags
|= extended_flags
;
1756 name
= ondisk2
->name
;
1759 name
= ondisk
->name
;
1761 if (expand_name_field
) {
1762 const unsigned char *cp
= (const unsigned char *)name
;
1763 size_t strip_len
, previous_len
;
1765 /* If we're at the begining of a block, ignore the previous name */
1766 strip_len
= decode_varint(&cp
);
1768 previous_len
= previous_ce
->ce_namelen
;
1769 if (previous_len
< strip_len
)
1770 die(_("malformed name field in the index, near path '%s'"),
1772 copy_len
= previous_len
- strip_len
;
1776 name
= (const char *)cp
;
1779 if (len
== CE_NAMEMASK
) {
1781 if (expand_name_field
)
1785 ce
= mem_pool__ce_alloc(ce_mem_pool
, len
);
1787 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1788 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1789 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1790 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1791 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1792 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1793 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1794 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1795 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1796 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1797 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1798 ce
->ce_namelen
= len
;
1800 hashcpy(ce
->oid
.hash
, ondisk
->sha1
);
1802 if (expand_name_field
) {
1804 memcpy(ce
->name
, previous_ce
->name
, copy_len
);
1805 memcpy(ce
->name
+ copy_len
, name
, len
+ 1 - copy_len
);
1806 *ent_size
= (name
- ((char *)ondisk
)) + len
+ 1 - copy_len
;
1808 memcpy(ce
->name
, name
, len
+ 1);
1809 *ent_size
= ondisk_ce_size(ce
);
1814 static void check_ce_order(struct index_state
*istate
)
1818 if (!verify_ce_order
)
1821 for (i
= 1; i
< istate
->cache_nr
; i
++) {
1822 struct cache_entry
*ce
= istate
->cache
[i
- 1];
1823 struct cache_entry
*next_ce
= istate
->cache
[i
];
1824 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1826 if (0 < name_compare
)
1827 die("unordered stage entries in index");
1828 if (!name_compare
) {
1830 die("multiple stage entries for merged file '%s'",
1832 if (ce_stage(ce
) > ce_stage(next_ce
))
1833 die("unordered stage entries for '%s'",
1839 static void tweak_untracked_cache(struct index_state
*istate
)
1841 switch (git_config_get_untracked_cache()) {
1842 case -1: /* keep: do nothing */
1845 remove_untracked_cache(istate
);
1848 add_untracked_cache(istate
);
1850 default: /* unknown value: do nothing */
1855 static void tweak_split_index(struct index_state
*istate
)
1857 switch (git_config_get_split_index()) {
1858 case -1: /* unset: do nothing */
1861 remove_split_index(istate
);
1864 add_split_index(istate
);
1866 default: /* unknown value: do nothing */
1871 static void post_read_index_from(struct index_state
*istate
)
1873 check_ce_order(istate
);
1874 tweak_untracked_cache(istate
);
1875 tweak_split_index(istate
);
1876 tweak_fsmonitor(istate
);
1879 static size_t estimate_cache_size_from_compressed(unsigned int entries
)
1881 return entries
* (sizeof(struct cache_entry
) + CACHE_ENTRY_PATH_LENGTH
);
1884 static size_t estimate_cache_size(size_t ondisk_size
, unsigned int entries
)
1886 long per_entry
= sizeof(struct cache_entry
) - sizeof(struct ondisk_cache_entry
);
1889 * Account for potential alignment differences.
1891 per_entry
+= align_padding_size(sizeof(struct cache_entry
), -sizeof(struct ondisk_cache_entry
));
1892 return ondisk_size
+ entries
* per_entry
;
1895 struct index_entry_offset
1897 /* starting byte offset into index file, count of index entries in this block */
1901 struct index_entry_offset_table
1904 struct index_entry_offset entries
[FLEX_ARRAY
];
1908 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
);
1909 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
);
1912 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
);
1913 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
);
1915 struct load_index_extensions
1920 struct index_state
*istate
;
1923 unsigned long src_offset
;
1926 static void *load_index_extensions(void *_data
)
1928 struct load_index_extensions
*p
= _data
;
1929 unsigned long src_offset
= p
->src_offset
;
1931 while (src_offset
<= p
->mmap_size
- the_hash_algo
->rawsz
- 8) {
1932 /* After an array of active_nr index entries,
1933 * there can be arbitrary number of extended
1934 * sections, each of which is prefixed with
1935 * extension name (4-byte) and section length
1936 * in 4-byte network byte order.
1938 uint32_t extsize
= get_be32(p
->mmap
+ src_offset
+ 4);
1939 if (read_index_extension(p
->istate
,
1940 p
->mmap
+ src_offset
,
1941 p
->mmap
+ src_offset
+ 8,
1943 munmap((void *)p
->mmap
, p
->mmap_size
);
1944 die(_("index file corrupt"));
1947 src_offset
+= extsize
;
1954 * A helper function that will load the specified range of cache entries
1955 * from the memory mapped file and add them to the given index.
1957 static unsigned long load_cache_entry_block(struct index_state
*istate
,
1958 struct mem_pool
*ce_mem_pool
, int offset
, int nr
, const char *mmap
,
1959 unsigned long start_offset
, const struct cache_entry
*previous_ce
)
1962 unsigned long src_offset
= start_offset
;
1964 for (i
= offset
; i
< offset
+ nr
; i
++) {
1965 struct ondisk_cache_entry
*disk_ce
;
1966 struct cache_entry
*ce
;
1967 unsigned long consumed
;
1969 disk_ce
= (struct ondisk_cache_entry
*)(mmap
+ src_offset
);
1970 ce
= create_from_disk(ce_mem_pool
, istate
->version
, disk_ce
, &consumed
, previous_ce
);
1971 set_index_entry(istate
, i
, ce
);
1973 src_offset
+= consumed
;
1976 return src_offset
- start_offset
;
1979 static unsigned long load_all_cache_entries(struct index_state
*istate
,
1980 const char *mmap
, size_t mmap_size
, unsigned long src_offset
)
1982 unsigned long consumed
;
1984 if (istate
->version
== 4) {
1985 mem_pool_init(&istate
->ce_mem_pool
,
1986 estimate_cache_size_from_compressed(istate
->cache_nr
));
1988 mem_pool_init(&istate
->ce_mem_pool
,
1989 estimate_cache_size(mmap_size
, istate
->cache_nr
));
1992 consumed
= load_cache_entry_block(istate
, istate
->ce_mem_pool
,
1993 0, istate
->cache_nr
, mmap
, src_offset
, NULL
);
2000 * Mostly randomly chosen maximum thread counts: we
2001 * cap the parallelism to online_cpus() threads, and we want
2002 * to have at least 10000 cache entries per thread for it to
2003 * be worth starting a thread.
2006 #define THREAD_COST (10000)
2008 struct load_cache_entries_thread_data
2011 struct index_state
*istate
;
2012 struct mem_pool
*ce_mem_pool
;
2015 struct index_entry_offset_table
*ieot
;
2016 int ieot_start
; /* starting index into the ieot array */
2017 int ieot_blocks
; /* count of ieot entries to process */
2018 unsigned long consumed
; /* return # of bytes in index file processed */
2022 * A thread proc to run the load_cache_entries() computation
2023 * across multiple background threads.
2025 static void *load_cache_entries_thread(void *_data
)
2027 struct load_cache_entries_thread_data
*p
= _data
;
2030 /* iterate across all ieot blocks assigned to this thread */
2031 for (i
= p
->ieot_start
; i
< p
->ieot_start
+ p
->ieot_blocks
; i
++) {
2032 p
->consumed
+= load_cache_entry_block(p
->istate
, p
->ce_mem_pool
,
2033 p
->offset
, p
->ieot
->entries
[i
].nr
, p
->mmap
, p
->ieot
->entries
[i
].offset
, NULL
);
2034 p
->offset
+= p
->ieot
->entries
[i
].nr
;
2039 static unsigned long load_cache_entries_threaded(struct index_state
*istate
, const char *mmap
, size_t mmap_size
,
2040 unsigned long src_offset
, int nr_threads
, struct index_entry_offset_table
*ieot
)
2042 int i
, offset
, ieot_blocks
, ieot_start
, err
;
2043 struct load_cache_entries_thread_data
*data
;
2044 unsigned long consumed
= 0;
2046 /* a little sanity checking */
2047 if (istate
->name_hash_initialized
)
2048 BUG("the name hash isn't thread safe");
2050 mem_pool_init(&istate
->ce_mem_pool
, 0);
2052 /* ensure we have no more threads than we have blocks to process */
2053 if (nr_threads
> ieot
->nr
)
2054 nr_threads
= ieot
->nr
;
2055 data
= xcalloc(nr_threads
, sizeof(*data
));
2057 offset
= ieot_start
= 0;
2058 ieot_blocks
= DIV_ROUND_UP(ieot
->nr
, nr_threads
);
2059 for (i
= 0; i
< nr_threads
; i
++) {
2060 struct load_cache_entries_thread_data
*p
= &data
[i
];
2063 if (ieot_start
+ ieot_blocks
> ieot
->nr
)
2064 ieot_blocks
= ieot
->nr
- ieot_start
;
2070 p
->ieot_start
= ieot_start
;
2071 p
->ieot_blocks
= ieot_blocks
;
2073 /* create a mem_pool for each thread */
2075 for (j
= p
->ieot_start
; j
< p
->ieot_start
+ p
->ieot_blocks
; j
++)
2076 nr
+= p
->ieot
->entries
[j
].nr
;
2077 if (istate
->version
== 4) {
2078 mem_pool_init(&p
->ce_mem_pool
,
2079 estimate_cache_size_from_compressed(nr
));
2081 mem_pool_init(&p
->ce_mem_pool
,
2082 estimate_cache_size(mmap_size
, nr
));
2085 err
= pthread_create(&p
->pthread
, NULL
, load_cache_entries_thread
, p
);
2087 die(_("unable to create load_cache_entries thread: %s"), strerror(err
));
2089 /* increment by the number of cache entries in the ieot block being processed */
2090 for (j
= 0; j
< ieot_blocks
; j
++)
2091 offset
+= ieot
->entries
[ieot_start
+ j
].nr
;
2092 ieot_start
+= ieot_blocks
;
2095 for (i
= 0; i
< nr_threads
; i
++) {
2096 struct load_cache_entries_thread_data
*p
= &data
[i
];
2098 err
= pthread_join(p
->pthread
, NULL
);
2100 die(_("unable to join load_cache_entries thread: %s"), strerror(err
));
2101 mem_pool_combine(istate
->ce_mem_pool
, p
->ce_mem_pool
);
2102 consumed
+= p
->consumed
;
2111 /* remember to discard_cache() before reading a different cache! */
2112 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
2116 unsigned long src_offset
;
2117 const struct cache_header
*hdr
;
2120 struct load_index_extensions p
;
2121 size_t extension_offset
= 0;
2123 int nr_threads
, cpus
;
2124 struct index_entry_offset_table
*ieot
= NULL
;
2127 if (istate
->initialized
)
2128 return istate
->cache_nr
;
2130 istate
->timestamp
.sec
= 0;
2131 istate
->timestamp
.nsec
= 0;
2132 fd
= open(path
, O_RDONLY
);
2134 if (!must_exist
&& errno
== ENOENT
)
2136 die_errno("%s: index file open failed", path
);
2140 die_errno("cannot stat the open index");
2142 mmap_size
= xsize_t(st
.st_size
);
2143 if (mmap_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2144 die("index file smaller than expected");
2146 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2147 if (mmap
== MAP_FAILED
)
2148 die_errno("unable to map index file");
2151 hdr
= (const struct cache_header
*)mmap
;
2152 if (verify_hdr(hdr
, mmap_size
) < 0)
2155 hashcpy(istate
->oid
.hash
, (const unsigned char *)hdr
+ mmap_size
- the_hash_algo
->rawsz
);
2156 istate
->version
= ntohl(hdr
->hdr_version
);
2157 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
2158 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
2159 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(*istate
->cache
));
2160 istate
->initialized
= 1;
2164 p
.mmap_size
= mmap_size
;
2166 src_offset
= sizeof(*hdr
);
2169 nr_threads
= git_config_get_index_threads();
2171 /* TODO: does creating more threads than cores help? */
2173 nr_threads
= istate
->cache_nr
/ THREAD_COST
;
2174 cpus
= online_cpus();
2175 if (nr_threads
> cpus
)
2179 if (nr_threads
> 1) {
2180 extension_offset
= read_eoie_extension(mmap
, mmap_size
);
2181 if (extension_offset
) {
2184 p
.src_offset
= extension_offset
;
2185 err
= pthread_create(&p
.pthread
, NULL
, load_index_extensions
, &p
);
2187 die(_("unable to create load_index_extensions thread: %s"), strerror(err
));
2194 * Locate and read the index entry offset table so that we can use it
2195 * to multi-thread the reading of the cache entries.
2197 if (extension_offset
&& nr_threads
> 1)
2198 ieot
= read_ieot_extension(mmap
, mmap_size
, extension_offset
);
2201 src_offset
+= load_cache_entries_threaded(istate
, mmap
, mmap_size
, src_offset
, nr_threads
, ieot
);
2204 src_offset
+= load_all_cache_entries(istate
, mmap
, mmap_size
, src_offset
);
2207 src_offset
+= load_all_cache_entries(istate
, mmap
, mmap_size
, src_offset
);
2210 istate
->timestamp
.sec
= st
.st_mtime
;
2211 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2213 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2215 if (extension_offset
) {
2216 int ret
= pthread_join(p
.pthread
, NULL
);
2218 die(_("unable to join load_index_extensions thread: %s"), strerror(ret
));
2221 if (!extension_offset
) {
2222 p
.src_offset
= src_offset
;
2223 load_index_extensions(&p
);
2225 munmap((void *)mmap
, mmap_size
);
2226 return istate
->cache_nr
;
2229 munmap((void *)mmap
, mmap_size
);
2230 die("index file corrupt");
2234 * Signal that the shared index is used by updating its mtime.
2236 * This way, shared index can be removed if they have not been used
2239 static void freshen_shared_index(const char *shared_index
, int warn
)
2241 if (!check_and_freshen_file(shared_index
, 1) && warn
)
2242 warning("could not freshen shared index '%s'", shared_index
);
2245 int read_index_from(struct index_state
*istate
, const char *path
,
2248 struct split_index
*split_index
;
2253 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2254 if (istate
->initialized
)
2255 return istate
->cache_nr
;
2257 trace_performance_enter();
2258 ret
= do_read_index(istate
, path
, 0);
2259 trace_performance_leave("read cache %s", path
);
2261 split_index
= istate
->split_index
;
2262 if (!split_index
|| is_null_oid(&split_index
->base_oid
)) {
2263 post_read_index_from(istate
);
2267 trace_performance_enter();
2268 if (split_index
->base
)
2269 discard_index(split_index
->base
);
2271 split_index
->base
= xcalloc(1, sizeof(*split_index
->base
));
2273 base_oid_hex
= oid_to_hex(&split_index
->base_oid
);
2274 base_path
= xstrfmt("%s/sharedindex.%s", gitdir
, base_oid_hex
);
2275 ret
= do_read_index(split_index
->base
, base_path
, 1);
2276 if (!oideq(&split_index
->base_oid
, &split_index
->base
->oid
))
2277 die("broken index, expect %s in %s, got %s",
2278 base_oid_hex
, base_path
,
2279 oid_to_hex(&split_index
->base
->oid
));
2281 freshen_shared_index(base_path
, 0);
2282 merge_base_index(istate
);
2283 post_read_index_from(istate
);
2285 trace_performance_leave("read cache %s", base_path
);
2289 int is_index_unborn(struct index_state
*istate
)
2291 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
2294 int discard_index(struct index_state
*istate
)
2297 * Cache entries in istate->cache[] should have been allocated
2298 * from the memory pool associated with this index, or from an
2299 * associated split_index. There is no need to free individual
2300 * cache entries. validate_cache_entries can detect when this
2301 * assertion does not hold.
2303 validate_cache_entries(istate
);
2305 resolve_undo_clear_index(istate
);
2306 istate
->cache_nr
= 0;
2307 istate
->cache_changed
= 0;
2308 istate
->timestamp
.sec
= 0;
2309 istate
->timestamp
.nsec
= 0;
2310 free_name_hash(istate
);
2311 cache_tree_free(&(istate
->cache_tree
));
2312 istate
->initialized
= 0;
2313 FREE_AND_NULL(istate
->cache
);
2314 istate
->cache_alloc
= 0;
2315 discard_split_index(istate
);
2316 free_untracked_cache(istate
->untracked
);
2317 istate
->untracked
= NULL
;
2319 if (istate
->ce_mem_pool
) {
2320 mem_pool_discard(istate
->ce_mem_pool
, should_validate_cache_entries());
2321 istate
->ce_mem_pool
= NULL
;
2328 * Validate the cache entries of this index.
2329 * All cache entries associated with this index
2330 * should have been allocated by the memory pool
2331 * associated with this index, or by a referenced
2334 void validate_cache_entries(const struct index_state
*istate
)
2338 if (!should_validate_cache_entries() ||!istate
|| !istate
->initialized
)
2341 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2343 die("internal error: cache entry is not allocated from expected memory pool");
2344 } else if (!istate
->ce_mem_pool
||
2345 !mem_pool_contains(istate
->ce_mem_pool
, istate
->cache
[i
])) {
2346 if (!istate
->split_index
||
2347 !istate
->split_index
->base
||
2348 !istate
->split_index
->base
->ce_mem_pool
||
2349 !mem_pool_contains(istate
->split_index
->base
->ce_mem_pool
, istate
->cache
[i
])) {
2350 die("internal error: cache entry is not allocated from expected memory pool");
2355 if (istate
->split_index
)
2356 validate_cache_entries(istate
->split_index
->base
);
2359 int unmerged_index(const struct index_state
*istate
)
2362 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2363 if (ce_stage(istate
->cache
[i
]))
2369 int index_has_changes(const struct index_state
*istate
,
2373 struct object_id cmp
;
2376 if (istate
!= &the_index
) {
2377 BUG("index_has_changes cannot yet accept istate != &the_index; do_diff_cache needs updating first.");
2380 cmp
= tree
->object
.oid
;
2381 if (tree
|| !get_oid_tree("HEAD", &cmp
)) {
2382 struct diff_options opt
;
2385 opt
.flags
.exit_with_status
= 1;
2387 opt
.flags
.quick
= 1;
2388 do_diff_cache(&cmp
, &opt
);
2390 for (i
= 0; sb
&& i
< diff_queued_diff
.nr
; i
++) {
2392 strbuf_addch(sb
, ' ');
2393 strbuf_addstr(sb
, diff_queued_diff
.queue
[i
]->two
->path
);
2396 return opt
.flags
.has_changes
!= 0;
2398 for (i
= 0; sb
&& i
< istate
->cache_nr
; i
++) {
2400 strbuf_addch(sb
, ' ');
2401 strbuf_addstr(sb
, istate
->cache
[i
]->name
);
2403 return !!istate
->cache_nr
;
2407 #define WRITE_BUFFER_SIZE 8192
2408 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
2409 static unsigned long write_buffer_len
;
2411 static int ce_write_flush(git_hash_ctx
*context
, int fd
)
2413 unsigned int buffered
= write_buffer_len
;
2415 the_hash_algo
->update_fn(context
, write_buffer
, buffered
);
2416 if (write_in_full(fd
, write_buffer
, buffered
) < 0)
2418 write_buffer_len
= 0;
2423 static int ce_write(git_hash_ctx
*context
, int fd
, void *data
, unsigned int len
)
2426 unsigned int buffered
= write_buffer_len
;
2427 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
2430 memcpy(write_buffer
+ buffered
, data
, partial
);
2431 buffered
+= partial
;
2432 if (buffered
== WRITE_BUFFER_SIZE
) {
2433 write_buffer_len
= buffered
;
2434 if (ce_write_flush(context
, fd
))
2438 write_buffer_len
= buffered
;
2440 data
= (char *) data
+ partial
;
2445 static int write_index_ext_header(git_hash_ctx
*context
, git_hash_ctx
*eoie_context
,
2446 int fd
, unsigned int ext
, unsigned int sz
)
2451 the_hash_algo
->update_fn(eoie_context
, &ext
, 4);
2452 the_hash_algo
->update_fn(eoie_context
, &sz
, 4);
2454 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
2455 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
2458 static int ce_flush(git_hash_ctx
*context
, int fd
, unsigned char *hash
)
2460 unsigned int left
= write_buffer_len
;
2463 write_buffer_len
= 0;
2464 the_hash_algo
->update_fn(context
, write_buffer
, left
);
2467 /* Flush first if not enough space for hash signature */
2468 if (left
+ the_hash_algo
->rawsz
> WRITE_BUFFER_SIZE
) {
2469 if (write_in_full(fd
, write_buffer
, left
) < 0)
2474 /* Append the hash signature at the end */
2475 the_hash_algo
->final_fn(write_buffer
+ left
, context
);
2476 hashcpy(hash
, write_buffer
+ left
);
2477 left
+= the_hash_algo
->rawsz
;
2478 return (write_in_full(fd
, write_buffer
, left
) < 0) ? -1 : 0;
2481 static void ce_smudge_racily_clean_entry(struct cache_entry
*ce
)
2484 * The only thing we care about in this function is to smudge the
2485 * falsely clean entry due to touch-update-touch race, so we leave
2486 * everything else as they are. We are called for entries whose
2487 * ce_stat_data.sd_mtime match the index file mtime.
2489 * Note that this actually does not do much for gitlinks, for
2490 * which ce_match_stat_basic() always goes to the actual
2491 * contents. The caller checks with is_racy_timestamp() which
2492 * always says "no" for gitlinks, so we are not called for them ;-)
2496 if (lstat(ce
->name
, &st
) < 0)
2498 if (ce_match_stat_basic(ce
, &st
))
2500 if (ce_modified_check_fs(ce
, &st
)) {
2501 /* This is "racily clean"; smudge it. Note that this
2502 * is a tricky code. At first glance, it may appear
2503 * that it can break with this sequence:
2505 * $ echo xyzzy >frotz
2506 * $ git-update-index --add frotz
2509 * $ echo filfre >nitfol
2510 * $ git-update-index --add nitfol
2512 * but it does not. When the second update-index runs,
2513 * it notices that the entry "frotz" has the same timestamp
2514 * as index, and if we were to smudge it by resetting its
2515 * size to zero here, then the object name recorded
2516 * in index is the 6-byte file but the cached stat information
2517 * becomes zero --- which would then match what we would
2518 * obtain from the filesystem next time we stat("frotz").
2520 * However, the second update-index, before calling
2521 * this function, notices that the cached size is 6
2522 * bytes and what is on the filesystem is an empty
2523 * file, and never calls us, so the cached size information
2524 * for "frotz" stays 6 which does not match the filesystem.
2526 ce
->ce_stat_data
.sd_size
= 0;
2530 /* Copy miscellaneous fields but not the name */
2531 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
2532 struct cache_entry
*ce
)
2536 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
2537 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
2538 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
2539 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
2540 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
2541 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
2542 ondisk
->mode
= htonl(ce
->ce_mode
);
2543 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
2544 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
2545 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
2546 hashcpy(ondisk
->sha1
, ce
->oid
.hash
);
2548 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
2549 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
2550 ondisk
->flags
= htons(flags
);
2551 if (ce
->ce_flags
& CE_EXTENDED
) {
2552 struct ondisk_cache_entry_extended
*ondisk2
;
2553 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
2554 ondisk2
->flags2
= htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
2558 static int ce_write_entry(git_hash_ctx
*c
, int fd
, struct cache_entry
*ce
,
2559 struct strbuf
*previous_name
, struct ondisk_cache_entry
*ondisk
)
2563 unsigned int saved_namelen
;
2564 int stripped_name
= 0;
2565 static unsigned char padding
[8] = { 0x00 };
2567 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2568 saved_namelen
= ce_namelen(ce
);
2573 if (ce
->ce_flags
& CE_EXTENDED
)
2574 size
= offsetof(struct ondisk_cache_entry_extended
, name
);
2576 size
= offsetof(struct ondisk_cache_entry
, name
);
2578 if (!previous_name
) {
2579 int len
= ce_namelen(ce
);
2580 copy_cache_entry_to_ondisk(ondisk
, ce
);
2581 result
= ce_write(c
, fd
, ondisk
, size
);
2583 result
= ce_write(c
, fd
, ce
->name
, len
);
2585 result
= ce_write(c
, fd
, padding
, align_padding_size(size
, len
));
2587 int common
, to_remove
, prefix_size
;
2588 unsigned char to_remove_vi
[16];
2590 (ce
->name
[common
] &&
2591 common
< previous_name
->len
&&
2592 ce
->name
[common
] == previous_name
->buf
[common
]);
2594 ; /* still matching */
2595 to_remove
= previous_name
->len
- common
;
2596 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
2598 copy_cache_entry_to_ondisk(ondisk
, ce
);
2599 result
= ce_write(c
, fd
, ondisk
, size
);
2601 result
= ce_write(c
, fd
, to_remove_vi
, prefix_size
);
2603 result
= ce_write(c
, fd
, ce
->name
+ common
, ce_namelen(ce
) - common
);
2605 result
= ce_write(c
, fd
, padding
, 1);
2607 strbuf_splice(previous_name
, common
, to_remove
,
2608 ce
->name
+ common
, ce_namelen(ce
) - common
);
2610 if (stripped_name
) {
2611 ce
->ce_namelen
= saved_namelen
;
2612 ce
->ce_flags
&= ~CE_STRIP_NAME
;
2619 * This function verifies if index_state has the correct sha1 of the
2620 * index file. Don't die if we have any other failure, just return 0.
2622 static int verify_index_from(const struct index_state
*istate
, const char *path
)
2627 unsigned char hash
[GIT_MAX_RAWSZ
];
2629 if (!istate
->initialized
)
2632 fd
= open(path
, O_RDONLY
);
2639 if (st
.st_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2642 n
= pread_in_full(fd
, hash
, the_hash_algo
->rawsz
, st
.st_size
- the_hash_algo
->rawsz
);
2643 if (n
!= the_hash_algo
->rawsz
)
2646 if (!hasheq(istate
->oid
.hash
, hash
))
2657 static int verify_index(const struct index_state
*istate
)
2659 return verify_index_from(istate
, get_index_file());
2662 static int has_racy_timestamp(struct index_state
*istate
)
2664 int entries
= istate
->cache_nr
;
2667 for (i
= 0; i
< entries
; i
++) {
2668 struct cache_entry
*ce
= istate
->cache
[i
];
2669 if (is_racy_timestamp(istate
, ce
))
2675 void update_index_if_able(struct index_state
*istate
, struct lock_file
*lockfile
)
2677 if ((istate
->cache_changed
|| has_racy_timestamp(istate
)) &&
2678 verify_index(istate
))
2679 write_locked_index(istate
, lockfile
, COMMIT_LOCK
);
2681 rollback_lock_file(lockfile
);
2685 * On success, `tempfile` is closed. If it is the temporary file
2686 * of a `struct lock_file`, we will therefore effectively perform
2687 * a 'close_lock_file_gently()`. Since that is an implementation
2688 * detail of lockfiles, callers of `do_write_index()` should not
2691 static int do_write_index(struct index_state
*istate
, struct tempfile
*tempfile
,
2692 int strip_extensions
)
2694 uint64_t start
= getnanotime();
2695 int newfd
= tempfile
->fd
;
2696 git_hash_ctx c
, eoie_c
;
2697 struct cache_header hdr
;
2698 int i
, err
= 0, removed
, extended
, hdr_version
;
2699 struct cache_entry
**cache
= istate
->cache
;
2700 int entries
= istate
->cache_nr
;
2702 struct ondisk_cache_entry_extended ondisk
;
2703 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
2704 int drop_cache_tree
= istate
->drop_cache_tree
;
2706 int ieot_entries
= 1;
2707 struct index_entry_offset_table
*ieot
= NULL
;
2710 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
2711 if (cache
[i
]->ce_flags
& CE_REMOVE
)
2714 /* reduce extended entries if possible */
2715 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
2716 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
2718 cache
[i
]->ce_flags
|= CE_EXTENDED
;
2722 if (!istate
->version
) {
2723 istate
->version
= get_index_format_default();
2724 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
2725 init_split_index(istate
);
2728 /* demote version 3 to version 2 when the latter suffices */
2729 if (istate
->version
== 3 || istate
->version
== 2)
2730 istate
->version
= extended
? 3 : 2;
2732 hdr_version
= istate
->version
;
2734 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2735 hdr
.hdr_version
= htonl(hdr_version
);
2736 hdr
.hdr_entries
= htonl(entries
- removed
);
2738 the_hash_algo
->init_fn(&c
);
2739 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
2743 nr_threads
= git_config_get_index_threads();
2744 if (nr_threads
!= 1) {
2745 int ieot_blocks
, cpus
;
2748 * ensure default number of ieot blocks maps evenly to the
2749 * default number of threads that will process them leaving
2750 * room for the thread to load the index extensions.
2753 ieot_blocks
= istate
->cache_nr
/ THREAD_COST
;
2754 cpus
= online_cpus();
2755 if (ieot_blocks
> cpus
- 1)
2756 ieot_blocks
= cpus
- 1;
2758 ieot_blocks
= nr_threads
;
2759 if (ieot_blocks
> istate
->cache_nr
)
2760 ieot_blocks
= istate
->cache_nr
;
2764 * no reason to write out the IEOT extension if we don't
2765 * have enough blocks to utilize multi-threading
2767 if (ieot_blocks
> 1) {
2768 ieot
= xcalloc(1, sizeof(struct index_entry_offset_table
)
2769 + (ieot_blocks
* sizeof(struct index_entry_offset
)));
2770 ieot_entries
= DIV_ROUND_UP(entries
, ieot_blocks
);
2775 offset
= lseek(newfd
, 0, SEEK_CUR
);
2780 offset
+= write_buffer_len
;
2782 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
2784 for (i
= 0; i
< entries
; i
++) {
2785 struct cache_entry
*ce
= cache
[i
];
2786 if (ce
->ce_flags
& CE_REMOVE
)
2788 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2789 ce_smudge_racily_clean_entry(ce
);
2790 if (is_null_oid(&ce
->oid
)) {
2791 static const char msg
[] = "cache entry has null sha1: %s";
2792 static int allow
= -1;
2795 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2797 warning(msg
, ce
->name
);
2799 err
= error(msg
, ce
->name
);
2801 drop_cache_tree
= 1;
2803 if (ieot
&& i
&& (i
% ieot_entries
== 0)) {
2804 ieot
->entries
[ieot
->nr
].nr
= nr
;
2805 ieot
->entries
[ieot
->nr
].offset
= offset
;
2808 * If we have a V4 index, set the first byte to an invalid
2809 * character to ensure there is nothing common with the previous
2813 previous_name
->buf
[0] = 0;
2815 offset
= lseek(newfd
, 0, SEEK_CUR
);
2820 offset
+= write_buffer_len
;
2822 if (ce_write_entry(&c
, newfd
, ce
, previous_name
, (struct ondisk_cache_entry
*)&ondisk
) < 0)
2830 ieot
->entries
[ieot
->nr
].nr
= nr
;
2831 ieot
->entries
[ieot
->nr
].offset
= offset
;
2834 strbuf_release(&previous_name_buf
);
2841 /* Write extension data here */
2842 offset
= lseek(newfd
, 0, SEEK_CUR
);
2847 offset
+= write_buffer_len
;
2848 the_hash_algo
->init_fn(&eoie_c
);
2851 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2852 * can minimize the number of extensions we have to scan through to
2853 * find it during load. Write it out regardless of the
2854 * strip_extensions parameter as we need it when loading the shared
2859 struct strbuf sb
= STRBUF_INIT
;
2861 write_ieot_extension(&sb
, ieot
);
2862 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_INDEXENTRYOFFSETTABLE
, sb
.len
) < 0
2863 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2864 strbuf_release(&sb
);
2871 if (!strip_extensions
&& istate
->split_index
) {
2872 struct strbuf sb
= STRBUF_INIT
;
2874 err
= write_link_extension(&sb
, istate
) < 0 ||
2875 write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_LINK
,
2877 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2878 strbuf_release(&sb
);
2882 if (!strip_extensions
&& !drop_cache_tree
&& istate
->cache_tree
) {
2883 struct strbuf sb
= STRBUF_INIT
;
2885 cache_tree_write(&sb
, istate
->cache_tree
);
2886 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
2887 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2888 strbuf_release(&sb
);
2892 if (!strip_extensions
&& istate
->resolve_undo
) {
2893 struct strbuf sb
= STRBUF_INIT
;
2895 resolve_undo_write(&sb
, istate
->resolve_undo
);
2896 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_RESOLVE_UNDO
,
2898 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2899 strbuf_release(&sb
);
2903 if (!strip_extensions
&& istate
->untracked
) {
2904 struct strbuf sb
= STRBUF_INIT
;
2906 write_untracked_extension(&sb
, istate
->untracked
);
2907 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_UNTRACKED
,
2909 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2910 strbuf_release(&sb
);
2914 if (!strip_extensions
&& istate
->fsmonitor_last_update
) {
2915 struct strbuf sb
= STRBUF_INIT
;
2917 write_fsmonitor_extension(&sb
, istate
);
2918 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_FSMONITOR
, sb
.len
) < 0
2919 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2920 strbuf_release(&sb
);
2926 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
2927 * so that it can be found and processed before all the index entries are
2928 * read. Write it out regardless of the strip_extensions parameter as we need it
2929 * when loading the shared index.
2932 struct strbuf sb
= STRBUF_INIT
;
2934 write_eoie_extension(&sb
, &eoie_c
, offset
);
2935 err
= write_index_ext_header(&c
, NULL
, newfd
, CACHE_EXT_ENDOFINDEXENTRIES
, sb
.len
) < 0
2936 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2937 strbuf_release(&sb
);
2942 if (ce_flush(&c
, newfd
, istate
->oid
.hash
))
2944 if (close_tempfile_gently(tempfile
)) {
2945 error(_("could not close '%s'"), tempfile
->filename
.buf
);
2948 if (stat(tempfile
->filename
.buf
, &st
))
2950 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
2951 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2952 trace_performance_since(start
, "write index, changed mask = %x", istate
->cache_changed
);
2956 void set_alternate_index_output(const char *name
)
2958 alternate_index_output
= name
;
2961 static int commit_locked_index(struct lock_file
*lk
)
2963 if (alternate_index_output
)
2964 return commit_lock_file_to(lk
, alternate_index_output
);
2966 return commit_lock_file(lk
);
2969 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2972 int ret
= do_write_index(istate
, lock
->tempfile
, 0);
2975 if (flags
& COMMIT_LOCK
)
2976 return commit_locked_index(lock
);
2977 return close_lock_file_gently(lock
);
2980 static int write_split_index(struct index_state
*istate
,
2981 struct lock_file
*lock
,
2985 prepare_to_write_split_index(istate
);
2986 ret
= do_write_locked_index(istate
, lock
, flags
);
2987 finish_writing_split_index(istate
);
2991 static const char *shared_index_expire
= "2.weeks.ago";
2993 static unsigned long get_shared_index_expire_date(void)
2995 static unsigned long shared_index_expire_date
;
2996 static int shared_index_expire_date_prepared
;
2998 if (!shared_index_expire_date_prepared
) {
2999 git_config_get_expiry("splitindex.sharedindexexpire",
3000 &shared_index_expire
);
3001 shared_index_expire_date
= approxidate(shared_index_expire
);
3002 shared_index_expire_date_prepared
= 1;
3005 return shared_index_expire_date
;
3008 static int should_delete_shared_index(const char *shared_index_path
)
3011 unsigned long expiration
;
3013 /* Check timestamp */
3014 expiration
= get_shared_index_expire_date();
3017 if (stat(shared_index_path
, &st
))
3018 return error_errno(_("could not stat '%s'"), shared_index_path
);
3019 if (st
.st_mtime
> expiration
)
3025 static int clean_shared_index_files(const char *current_hex
)
3028 DIR *dir
= opendir(get_git_dir());
3031 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3033 while ((de
= readdir(dir
)) != NULL
) {
3034 const char *sha1_hex
;
3035 const char *shared_index_path
;
3036 if (!skip_prefix(de
->d_name
, "sharedindex.", &sha1_hex
))
3038 if (!strcmp(sha1_hex
, current_hex
))
3040 shared_index_path
= git_path("%s", de
->d_name
);
3041 if (should_delete_shared_index(shared_index_path
) > 0 &&
3042 unlink(shared_index_path
))
3043 warning_errno(_("unable to unlink: %s"), shared_index_path
);
3050 static int write_shared_index(struct index_state
*istate
,
3051 struct tempfile
**temp
)
3053 struct split_index
*si
= istate
->split_index
;
3056 move_cache_to_base_index(istate
);
3057 ret
= do_write_index(si
->base
, *temp
, 1);
3060 ret
= adjust_shared_perm(get_tempfile_path(*temp
));
3062 error("cannot fix permission bits on %s", get_tempfile_path(*temp
));
3065 ret
= rename_tempfile(temp
,
3066 git_path("sharedindex.%s", oid_to_hex(&si
->base
->oid
)));
3068 oidcpy(&si
->base_oid
, &si
->base
->oid
);
3069 clean_shared_index_files(oid_to_hex(&si
->base
->oid
));
3075 static const int default_max_percent_split_change
= 20;
3077 static int too_many_not_shared_entries(struct index_state
*istate
)
3079 int i
, not_shared
= 0;
3080 int max_split
= git_config_get_max_percent_split_change();
3082 switch (max_split
) {
3084 /* not or badly configured: use the default value */
3085 max_split
= default_max_percent_split_change
;
3088 return 1; /* 0% means always write a new shared index */
3090 return 0; /* 100% means never write a new shared index */
3092 break; /* just use the configured value */
3095 /* Count not shared entries */
3096 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3097 struct cache_entry
*ce
= istate
->cache
[i
];
3102 return (int64_t)istate
->cache_nr
* max_split
< (int64_t)not_shared
* 100;
3105 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3108 int new_shared_index
, ret
;
3109 struct split_index
*si
= istate
->split_index
;
3111 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3112 cache_tree_verify(istate
);
3114 if ((flags
& SKIP_IF_UNCHANGED
) && !istate
->cache_changed
) {
3115 if (flags
& COMMIT_LOCK
)
3116 rollback_lock_file(lock
);
3120 if (istate
->fsmonitor_last_update
)
3121 fill_fsmonitor_bitmap(istate
);
3123 if (!si
|| alternate_index_output
||
3124 (istate
->cache_changed
& ~EXTMASK
)) {
3126 oidclr(&si
->base_oid
);
3127 ret
= do_write_locked_index(istate
, lock
, flags
);
3131 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
3132 int v
= si
->base_oid
.hash
[0];
3134 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3136 if (too_many_not_shared_entries(istate
))
3137 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3139 new_shared_index
= istate
->cache_changed
& SPLIT_INDEX_ORDERED
;
3141 if (new_shared_index
) {
3142 struct tempfile
*temp
;
3145 temp
= mks_tempfile(git_path("sharedindex_XXXXXX"));
3147 oidclr(&si
->base_oid
);
3148 ret
= do_write_locked_index(istate
, lock
, flags
);
3151 ret
= write_shared_index(istate
, &temp
);
3153 saved_errno
= errno
;
3154 if (is_tempfile_active(temp
))
3155 delete_tempfile(&temp
);
3156 errno
= saved_errno
;
3162 ret
= write_split_index(istate
, lock
, flags
);
3164 /* Freshen the shared index only if the split-index was written */
3165 if (!ret
&& !new_shared_index
) {
3166 const char *shared_index
= git_path("sharedindex.%s",
3167 oid_to_hex(&si
->base_oid
));
3168 freshen_shared_index(shared_index
, 1);
3172 if (flags
& COMMIT_LOCK
)
3173 rollback_lock_file(lock
);
3178 * Read the index file that is potentially unmerged into given
3179 * index_state, dropping any unmerged entries to stage #0 (potentially
3180 * resulting in a path appearing as both a file and a directory in the
3181 * index; the caller is responsible to clear out the extra entries
3182 * before writing the index to a tree). Returns true if the index is
3183 * unmerged. Callers who want to refuse to work from an unmerged
3184 * state can call this and check its return value, instead of calling
3187 int read_index_unmerged(struct index_state
*istate
)
3193 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3194 struct cache_entry
*ce
= istate
->cache
[i
];
3195 struct cache_entry
*new_ce
;
3201 len
= ce_namelen(ce
);
3202 new_ce
= make_empty_cache_entry(istate
, len
);
3203 memcpy(new_ce
->name
, ce
->name
, len
);
3204 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
3205 new_ce
->ce_namelen
= len
;
3206 new_ce
->ce_mode
= ce
->ce_mode
;
3207 if (add_index_entry(istate
, new_ce
, ADD_CACHE_SKIP_DFCHECK
))
3208 return error("%s: cannot drop to stage #0",
3215 * Returns 1 if the path is an "other" path with respect to
3216 * the index; that is, the path is not mentioned in the index at all,
3217 * either as a file, a directory with some files in the index,
3218 * or as an unmerged entry.
3220 * We helpfully remove a trailing "/" from directories so that
3221 * the output of read_directory can be used as-is.
3223 int index_name_is_other(const struct index_state
*istate
, const char *name
,
3227 if (namelen
&& name
[namelen
- 1] == '/')
3229 pos
= index_name_pos(istate
, name
, namelen
);
3231 return 0; /* exact match */
3233 if (pos
< istate
->cache_nr
) {
3234 struct cache_entry
*ce
= istate
->cache
[pos
];
3235 if (ce_namelen(ce
) == namelen
&&
3236 !memcmp(ce
->name
, name
, namelen
))
3237 return 0; /* Yup, this one exists unmerged */
3242 void *read_blob_data_from_index(const struct index_state
*istate
,
3243 const char *path
, unsigned long *size
)
3247 enum object_type type
;
3251 pos
= index_name_pos(istate
, path
, len
);
3254 * We might be in the middle of a merge, in which
3255 * case we would read stage #2 (ours).
3259 (pos
< 0 && i
< istate
->cache_nr
&&
3260 !strcmp(istate
->cache
[i
]->name
, path
));
3262 if (ce_stage(istate
->cache
[i
]) == 2)
3267 data
= read_object_file(&istate
->cache
[pos
]->oid
, &type
, &sz
);
3268 if (!data
|| type
!= OBJ_BLOB
) {
3277 void stat_validity_clear(struct stat_validity
*sv
)
3279 FREE_AND_NULL(sv
->sd
);
3282 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
3286 if (stat(path
, &st
) < 0)
3287 return sv
->sd
== NULL
;
3290 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
3293 void stat_validity_update(struct stat_validity
*sv
, int fd
)
3297 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
3298 stat_validity_clear(sv
);
3301 sv
->sd
= xcalloc(1, sizeof(struct stat_data
));
3302 fill_stat_data(sv
->sd
, &st
);
3306 void move_index_extensions(struct index_state
*dst
, struct index_state
*src
)
3308 dst
->untracked
= src
->untracked
;
3309 src
->untracked
= NULL
;
3310 dst
->cache_tree
= src
->cache_tree
;
3311 src
->cache_tree
= NULL
;
3314 struct cache_entry
*dup_cache_entry(const struct cache_entry
*ce
,
3315 struct index_state
*istate
)
3317 unsigned int size
= ce_size(ce
);
3318 int mem_pool_allocated
;
3319 struct cache_entry
*new_entry
= make_empty_cache_entry(istate
, ce_namelen(ce
));
3320 mem_pool_allocated
= new_entry
->mem_pool_allocated
;
3322 memcpy(new_entry
, ce
, size
);
3323 new_entry
->mem_pool_allocated
= mem_pool_allocated
;
3327 void discard_cache_entry(struct cache_entry
*ce
)
3329 if (ce
&& should_validate_cache_entries())
3330 memset(ce
, 0xCD, cache_entry_size(ce
->ce_namelen
));
3332 if (ce
&& ce
->mem_pool_allocated
)
3338 int should_validate_cache_entries(void)
3340 static int validate_index_cache_entries
= -1;
3342 if (validate_index_cache_entries
< 0) {
3343 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3344 validate_index_cache_entries
= 1;
3346 validate_index_cache_entries
= 0;
3349 return validate_index_cache_entries
;
3352 #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3353 #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3355 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
)
3358 * The end of index entries (EOIE) extension is guaranteed to be last
3359 * so that it can be found by scanning backwards from the EOF.
3366 const char *index
, *eoie
;
3368 size_t offset
, src_offset
;
3369 unsigned char hash
[GIT_MAX_RAWSZ
];
3372 /* ensure we have an index big enough to contain an EOIE extension */
3373 if (mmap_size
< sizeof(struct cache_header
) + EOIE_SIZE_WITH_HEADER
+ the_hash_algo
->rawsz
)
3376 /* validate the extension signature */
3377 index
= eoie
= mmap
+ mmap_size
- EOIE_SIZE_WITH_HEADER
- the_hash_algo
->rawsz
;
3378 if (CACHE_EXT(index
) != CACHE_EXT_ENDOFINDEXENTRIES
)
3380 index
+= sizeof(uint32_t);
3382 /* validate the extension size */
3383 extsize
= get_be32(index
);
3384 if (extsize
!= EOIE_SIZE
)
3386 index
+= sizeof(uint32_t);
3389 * Validate the offset we're going to look for the first extension
3390 * signature is after the index header and before the eoie extension.
3392 offset
= get_be32(index
);
3393 if (mmap
+ offset
< mmap
+ sizeof(struct cache_header
))
3395 if (mmap
+ offset
>= eoie
)
3397 index
+= sizeof(uint32_t);
3400 * The hash is computed over extension types and their sizes (but not
3401 * their contents). E.g. if we have "TREE" extension that is N-bytes
3402 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3403 * then the hash would be:
3405 * SHA-1("TREE" + <binary representation of N> +
3406 * "REUC" + <binary representation of M>)
3408 src_offset
= offset
;
3409 the_hash_algo
->init_fn(&c
);
3410 while (src_offset
< mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
) {
3411 /* After an array of active_nr index entries,
3412 * there can be arbitrary number of extended
3413 * sections, each of which is prefixed with
3414 * extension name (4-byte) and section length
3415 * in 4-byte network byte order.
3418 memcpy(&extsize
, mmap
+ src_offset
+ 4, 4);
3419 extsize
= ntohl(extsize
);
3421 /* verify the extension size isn't so large it will wrap around */
3422 if (src_offset
+ 8 + extsize
< src_offset
)
3425 the_hash_algo
->update_fn(&c
, mmap
+ src_offset
, 8);
3428 src_offset
+= extsize
;
3430 the_hash_algo
->final_fn(hash
, &c
);
3431 if (!hasheq(hash
, (const unsigned char *)index
))
3434 /* Validate that the extension offsets returned us back to the eoie extension. */
3435 if (src_offset
!= mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
)
3441 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
)
3444 unsigned char hash
[GIT_MAX_RAWSZ
];
3447 put_be32(&buffer
, offset
);
3448 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3451 the_hash_algo
->final_fn(hash
, eoie_context
);
3452 strbuf_add(sb
, hash
, the_hash_algo
->rawsz
);
3456 #define IEOT_VERSION (1)
3458 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
)
3460 const char *index
= NULL
;
3461 uint32_t extsize
, ext_version
;
3462 struct index_entry_offset_table
*ieot
;
3465 /* find the IEOT extension */
3468 while (offset
<= mmap_size
- the_hash_algo
->rawsz
- 8) {
3469 extsize
= get_be32(mmap
+ offset
+ 4);
3470 if (CACHE_EXT((mmap
+ offset
)) == CACHE_EXT_INDEXENTRYOFFSETTABLE
) {
3471 index
= mmap
+ offset
+ 4 + 4;
3480 /* validate the version is IEOT_VERSION */
3481 ext_version
= get_be32(index
);
3482 if (ext_version
!= IEOT_VERSION
) {
3483 error("invalid IEOT version %d", ext_version
);
3486 index
+= sizeof(uint32_t);
3488 /* extension size - version bytes / bytes per entry */
3489 nr
= (extsize
- sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3491 error("invalid number of IEOT entries %d", nr
);
3494 ieot
= xmalloc(sizeof(struct index_entry_offset_table
)
3495 + (nr
* sizeof(struct index_entry_offset
)));
3497 for (i
= 0; i
< nr
; i
++) {
3498 ieot
->entries
[i
].offset
= get_be32(index
);
3499 index
+= sizeof(uint32_t);
3500 ieot
->entries
[i
].nr
= get_be32(index
);
3501 index
+= sizeof(uint32_t);
3507 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
)
3513 put_be32(&buffer
, IEOT_VERSION
);
3514 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3517 for (i
= 0; i
< ieot
->nr
; i
++) {
3520 put_be32(&buffer
, ieot
->entries
[i
].offset
);
3521 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3524 put_be32(&buffer
, ieot
->entries
[i
].nr
);
3525 strbuf_add(sb
, &buffer
, sizeof(uint32_t));