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"
29 /* Mask for the name length in ce_flags in the on-disk index */
31 #define CE_NAMEMASK (0x0fff)
35 * The first letter should be 'A'..'Z' for extensions that are not
36 * necessary for a correct operation (i.e. optimization data).
37 * When new extensions are added that _needs_ to be understood in
38 * order to correctly interpret the index file, pick character that
39 * is outside the range, to cause the reader to abort.
42 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
43 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
44 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
45 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
46 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
47 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
48 #define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945 /* "EOIE" */
49 #define CACHE_EXT_INDEXENTRYOFFSETTABLE 0x49454F54 /* "IEOT" */
51 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
52 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
53 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
54 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
58 * This is an estimate of the pathname length in the index. We use
59 * this for V4 index files to guess the un-deltafied size of the index
60 * in memory because of pathname deltafication. This is not required
61 * for V2/V3 index formats because their pathnames are not compressed.
62 * If the initial amount of memory set aside is not sufficient, the
63 * mem pool will allocate extra memory.
65 #define CACHE_ENTRY_PATH_LENGTH 80
67 static inline struct cache_entry
*mem_pool__ce_alloc(struct mem_pool
*mem_pool
, size_t len
)
69 struct cache_entry
*ce
;
70 ce
= mem_pool_alloc(mem_pool
, cache_entry_size(len
));
71 ce
->mem_pool_allocated
= 1;
75 static inline struct cache_entry
*mem_pool__ce_calloc(struct mem_pool
*mem_pool
, size_t len
)
77 struct cache_entry
* ce
;
78 ce
= mem_pool_calloc(mem_pool
, 1, cache_entry_size(len
));
79 ce
->mem_pool_allocated
= 1;
83 static struct mem_pool
*find_mem_pool(struct index_state
*istate
)
85 struct mem_pool
**pool_ptr
;
87 if (istate
->split_index
&& istate
->split_index
->base
)
88 pool_ptr
= &istate
->split_index
->base
->ce_mem_pool
;
90 pool_ptr
= &istate
->ce_mem_pool
;
93 mem_pool_init(pool_ptr
, 0);
98 struct index_state the_index
;
99 static const char *alternate_index_output
;
101 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
103 istate
->cache
[nr
] = ce
;
104 add_name_hash(istate
, ce
);
107 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
109 struct cache_entry
*old
= istate
->cache
[nr
];
111 replace_index_entry_in_base(istate
, old
, ce
);
112 remove_name_hash(istate
, old
);
113 discard_cache_entry(old
);
114 ce
->ce_flags
&= ~CE_HASHED
;
115 set_index_entry(istate
, nr
, ce
);
116 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
117 mark_fsmonitor_invalid(istate
, ce
);
118 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
121 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
123 struct cache_entry
*old_entry
= istate
->cache
[nr
], *new_entry
;
124 int namelen
= strlen(new_name
);
126 new_entry
= make_empty_cache_entry(istate
, namelen
);
127 copy_cache_entry(new_entry
, old_entry
);
128 new_entry
->ce_flags
&= ~CE_HASHED
;
129 new_entry
->ce_namelen
= namelen
;
130 new_entry
->index
= 0;
131 memcpy(new_entry
->name
, new_name
, namelen
+ 1);
133 cache_tree_invalidate_path(istate
, old_entry
->name
);
134 untracked_cache_remove_from_index(istate
, old_entry
->name
);
135 remove_index_entry_at(istate
, nr
);
136 add_index_entry(istate
, new_entry
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
139 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
141 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
142 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
143 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
144 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
145 sd
->sd_dev
= st
->st_dev
;
146 sd
->sd_ino
= st
->st_ino
;
147 sd
->sd_uid
= st
->st_uid
;
148 sd
->sd_gid
= st
->st_gid
;
149 sd
->sd_size
= st
->st_size
;
152 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
156 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
157 changed
|= MTIME_CHANGED
;
158 if (trust_ctime
&& check_stat
&&
159 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
160 changed
|= CTIME_CHANGED
;
163 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
164 changed
|= MTIME_CHANGED
;
165 if (trust_ctime
&& check_stat
&&
166 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
167 changed
|= CTIME_CHANGED
;
171 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
172 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
173 changed
|= OWNER_CHANGED
;
174 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
175 changed
|= INODE_CHANGED
;
180 * st_dev breaks on network filesystems where different
181 * clients will have different views of what "device"
182 * the filesystem is on
184 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
185 changed
|= INODE_CHANGED
;
188 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
189 changed
|= DATA_CHANGED
;
195 * This only updates the "non-critical" parts of the directory
196 * cache, ie the parts that aren't tracked by GIT, and only used
197 * to validate the cache.
199 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
201 fill_stat_data(&ce
->ce_stat_data
, st
);
203 if (assume_unchanged
)
204 ce
->ce_flags
|= CE_VALID
;
206 if (S_ISREG(st
->st_mode
)) {
207 ce_mark_uptodate(ce
);
208 mark_fsmonitor_valid(ce
);
212 static int ce_compare_data(struct index_state
*istate
,
213 const struct cache_entry
*ce
,
217 int fd
= git_open_cloexec(ce
->name
, O_RDONLY
);
220 struct object_id oid
;
221 if (!index_fd(istate
, &oid
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
222 match
= !oideq(&oid
, &ce
->oid
);
223 /* index_fd() closed the file descriptor already */
228 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
233 enum object_type type
;
234 struct strbuf sb
= STRBUF_INIT
;
236 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
239 buffer
= read_object_file(&ce
->oid
, &type
, &size
);
242 match
= memcmp(buffer
, sb
.buf
, size
);
249 static int ce_compare_gitlink(const struct cache_entry
*ce
)
251 struct object_id oid
;
254 * We don't actually require that the .git directory
255 * under GITLINK directory be a valid git directory. It
256 * might even be missing (in case nobody populated that
259 * If so, we consider it always to match.
261 if (resolve_gitlink_ref(ce
->name
, "HEAD", &oid
) < 0)
263 return !oideq(&oid
, &ce
->oid
);
266 static int ce_modified_check_fs(struct index_state
*istate
,
267 const struct cache_entry
*ce
,
270 switch (st
->st_mode
& S_IFMT
) {
272 if (ce_compare_data(istate
, ce
, st
))
276 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
280 if (S_ISGITLINK(ce
->ce_mode
))
281 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
282 /* else fallthrough */
289 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
291 unsigned int changed
= 0;
293 if (ce
->ce_flags
& CE_REMOVE
)
294 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
296 switch (ce
->ce_mode
& S_IFMT
) {
298 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
299 /* We consider only the owner x bit to be relevant for
302 if (trust_executable_bit
&&
303 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
304 changed
|= MODE_CHANGED
;
307 if (!S_ISLNK(st
->st_mode
) &&
308 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
309 changed
|= TYPE_CHANGED
;
312 /* We ignore most of the st_xxx fields for gitlinks */
313 if (!S_ISDIR(st
->st_mode
))
314 changed
|= TYPE_CHANGED
;
315 else if (ce_compare_gitlink(ce
))
316 changed
|= DATA_CHANGED
;
319 BUG("unsupported ce_mode: %o", ce
->ce_mode
);
322 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
324 /* Racily smudged entry? */
325 if (!ce
->ce_stat_data
.sd_size
) {
326 if (!is_empty_blob_sha1(ce
->oid
.hash
))
327 changed
|= DATA_CHANGED
;
333 static int is_racy_stat(const struct index_state
*istate
,
334 const struct stat_data
*sd
)
336 return (istate
->timestamp
.sec
&&
338 /* nanosecond timestamped files can also be racy! */
339 (istate
->timestamp
.sec
< sd
->sd_mtime
.sec
||
340 (istate
->timestamp
.sec
== sd
->sd_mtime
.sec
&&
341 istate
->timestamp
.nsec
<= sd
->sd_mtime
.nsec
))
343 istate
->timestamp
.sec
<= sd
->sd_mtime
.sec
348 int is_racy_timestamp(const struct index_state
*istate
,
349 const struct cache_entry
*ce
)
351 return (!S_ISGITLINK(ce
->ce_mode
) &&
352 is_racy_stat(istate
, &ce
->ce_stat_data
));
355 int match_stat_data_racy(const struct index_state
*istate
,
356 const struct stat_data
*sd
, struct stat
*st
)
358 if (is_racy_stat(istate
, sd
))
359 return MTIME_CHANGED
;
360 return match_stat_data(sd
, st
);
363 int ie_match_stat(struct index_state
*istate
,
364 const struct cache_entry
*ce
, struct stat
*st
,
365 unsigned int options
)
367 unsigned int changed
;
368 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
369 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
370 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
371 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
373 if (!ignore_fsmonitor
)
374 refresh_fsmonitor(istate
);
376 * If it's marked as always valid in the index, it's
377 * valid whatever the checked-out copy says.
379 * skip-worktree has the same effect with higher precedence
381 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
383 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
385 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
))
389 * Intent-to-add entries have not been added, so the index entry
390 * by definition never matches what is in the work tree until it
391 * actually gets added.
393 if (ce_intent_to_add(ce
))
394 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
396 changed
= ce_match_stat_basic(ce
, st
);
399 * Within 1 second of this sequence:
400 * echo xyzzy >file && git-update-index --add file
401 * running this command:
403 * would give a falsely clean cache entry. The mtime and
404 * length match the cache, and other stat fields do not change.
406 * We could detect this at update-index time (the cache entry
407 * being registered/updated records the same time as "now")
408 * and delay the return from git-update-index, but that would
409 * effectively mean we can make at most one commit per second,
410 * which is not acceptable. Instead, we check cache entries
411 * whose mtime are the same as the index file timestamp more
412 * carefully than others.
414 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
415 if (assume_racy_is_modified
)
416 changed
|= DATA_CHANGED
;
418 changed
|= ce_modified_check_fs(istate
, ce
, st
);
424 int ie_modified(struct index_state
*istate
,
425 const struct cache_entry
*ce
,
426 struct stat
*st
, unsigned int options
)
428 int changed
, changed_fs
;
430 changed
= ie_match_stat(istate
, ce
, st
, options
);
434 * If the mode or type has changed, there's no point in trying
435 * to refresh the entry - it's not going to match
437 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
441 * Immediately after read-tree or update-index --cacheinfo,
442 * the length field is zero, as we have never even read the
443 * lstat(2) information once, and we cannot trust DATA_CHANGED
444 * returned by ie_match_stat() which in turn was returned by
445 * ce_match_stat_basic() to signal that the filesize of the
446 * blob changed. We have to actually go to the filesystem to
447 * see if the contents match, and if so, should answer "unchanged".
449 * The logic does not apply to gitlinks, as ce_match_stat_basic()
450 * already has checked the actual HEAD from the filesystem in the
451 * subproject. If ie_match_stat() already said it is different,
452 * then we know it is.
454 if ((changed
& DATA_CHANGED
) &&
455 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
458 changed_fs
= ce_modified_check_fs(istate
, ce
, st
);
460 return changed
| changed_fs
;
464 int base_name_compare(const char *name1
, int len1
, int mode1
,
465 const char *name2
, int len2
, int mode2
)
467 unsigned char c1
, c2
;
468 int len
= len1
< len2
? len1
: len2
;
471 cmp
= memcmp(name1
, name2
, len
);
476 if (!c1
&& S_ISDIR(mode1
))
478 if (!c2
&& S_ISDIR(mode2
))
480 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
484 * df_name_compare() is identical to base_name_compare(), except it
485 * compares conflicting directory/file entries as equal. Note that
486 * while a directory name compares as equal to a regular file, they
487 * then individually compare _differently_ to a filename that has
488 * a dot after the basename (because '\0' < '.' < '/').
490 * This is used by routines that want to traverse the git namespace
491 * but then handle conflicting entries together when possible.
493 int df_name_compare(const char *name1
, int len1
, int mode1
,
494 const char *name2
, int len2
, int mode2
)
496 int len
= len1
< len2
? len1
: len2
, cmp
;
497 unsigned char c1
, c2
;
499 cmp
= memcmp(name1
, name2
, len
);
502 /* Directories and files compare equal (same length, same name) */
506 if (!c1
&& S_ISDIR(mode1
))
509 if (!c2
&& S_ISDIR(mode2
))
511 if (c1
== '/' && !c2
)
513 if (c2
== '/' && !c1
)
518 int name_compare(const char *name1
, size_t len1
, const char *name2
, size_t len2
)
520 size_t min_len
= (len1
< len2
) ? len1
: len2
;
521 int cmp
= memcmp(name1
, name2
, min_len
);
531 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
535 cmp
= name_compare(name1
, len1
, name2
, len2
);
546 static int index_name_stage_pos(const struct index_state
*istate
, const char *name
, int namelen
, int stage
)
551 last
= istate
->cache_nr
;
552 while (last
> first
) {
553 int next
= (last
+ first
) >> 1;
554 struct cache_entry
*ce
= istate
->cache
[next
];
555 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
567 int index_name_pos(const struct index_state
*istate
, const char *name
, int namelen
)
569 return index_name_stage_pos(istate
, name
, namelen
, 0);
572 int remove_index_entry_at(struct index_state
*istate
, int pos
)
574 struct cache_entry
*ce
= istate
->cache
[pos
];
576 record_resolve_undo(istate
, ce
);
577 remove_name_hash(istate
, ce
);
578 save_or_free_index_entry(istate
, ce
);
579 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
581 if (pos
>= istate
->cache_nr
)
583 MOVE_ARRAY(istate
->cache
+ pos
, istate
->cache
+ pos
+ 1,
584 istate
->cache_nr
- pos
);
589 * Remove all cache entries marked for removal, that is where
590 * CE_REMOVE is set in ce_flags. This is much more effective than
591 * calling remove_index_entry_at() for each entry to be removed.
593 void remove_marked_cache_entries(struct index_state
*istate
)
595 struct cache_entry
**ce_array
= istate
->cache
;
598 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
599 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
600 remove_name_hash(istate
, ce_array
[i
]);
601 save_or_free_index_entry(istate
, ce_array
[i
]);
604 ce_array
[j
++] = ce_array
[i
];
606 if (j
== istate
->cache_nr
)
608 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
609 istate
->cache_nr
= j
;
612 int remove_file_from_index(struct index_state
*istate
, const char *path
)
614 int pos
= index_name_pos(istate
, path
, strlen(path
));
617 cache_tree_invalidate_path(istate
, path
);
618 untracked_cache_remove_from_index(istate
, path
);
619 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
620 remove_index_entry_at(istate
, pos
);
624 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
626 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
629 static int index_name_pos_also_unmerged(struct index_state
*istate
,
630 const char *path
, int namelen
)
632 int pos
= index_name_pos(istate
, path
, namelen
);
633 struct cache_entry
*ce
;
638 /* maybe unmerged? */
640 if (pos
>= istate
->cache_nr
||
641 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
644 /* order of preference: stage 2, 1, 3 */
645 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
646 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
647 !compare_name(ce
, path
, namelen
))
652 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
654 int len
= ce_namelen(ce
);
655 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
659 * If we add a filename that aliases in the cache, we will use the
660 * name that we already have - but we don't want to update the same
661 * alias twice, because that implies that there were actually two
662 * different files with aliasing names!
664 * So we use the CE_ADDED flag to verify that the alias was an old
665 * one before we accept it as
667 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
668 struct cache_entry
*ce
,
669 struct cache_entry
*alias
)
672 struct cache_entry
*new_entry
;
674 if (alias
->ce_flags
& CE_ADDED
)
675 die(_("will not add file alias '%s' ('%s' already exists in index)"),
676 ce
->name
, alias
->name
);
678 /* Ok, create the new entry using the name of the existing alias */
679 len
= ce_namelen(alias
);
680 new_entry
= make_empty_cache_entry(istate
, len
);
681 memcpy(new_entry
->name
, alias
->name
, len
);
682 copy_cache_entry(new_entry
, ce
);
683 save_or_free_index_entry(istate
, ce
);
687 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
689 struct object_id oid
;
690 if (write_object_file("", 0, blob_type
, &oid
))
691 die(_("cannot create an empty blob in the object database"));
692 oidcpy(&ce
->oid
, &oid
);
695 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
697 int namelen
, was_same
;
698 mode_t st_mode
= st
->st_mode
;
699 struct cache_entry
*ce
, *alias
= NULL
;
700 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
701 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
702 int pretend
= flags
& ADD_CACHE_PRETEND
;
703 int intent_only
= flags
& ADD_CACHE_INTENT
;
704 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
705 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
706 int newflags
= HASH_WRITE_OBJECT
;
708 if (flags
& HASH_RENORMALIZE
)
709 newflags
|= HASH_RENORMALIZE
;
711 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
712 return error(_("%s: can only add regular files, symbolic links or git-directories"), path
);
714 namelen
= strlen(path
);
715 if (S_ISDIR(st_mode
)) {
716 while (namelen
&& path
[namelen
-1] == '/')
719 ce
= make_empty_cache_entry(istate
, namelen
);
720 memcpy(ce
->name
, path
, namelen
);
721 ce
->ce_namelen
= namelen
;
723 fill_stat_cache_info(ce
, st
);
725 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
728 if (trust_executable_bit
&& has_symlinks
) {
729 ce
->ce_mode
= create_ce_mode(st_mode
);
731 /* If there is an existing entry, pick the mode bits and type
732 * from it, otherwise assume unexecutable regular file.
734 struct cache_entry
*ent
;
735 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
737 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
738 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
741 /* When core.ignorecase=true, determine if a directory of the same name but differing
742 * case already exists within the Git repository. If it does, ensure the directory
743 * case of the file being added to the repository matches (is folded into) the existing
744 * entry's directory case.
747 adjust_dirname_case(istate
, ce
->name
);
749 if (!(flags
& HASH_RENORMALIZE
)) {
750 alias
= index_file_exists(istate
, ce
->name
,
751 ce_namelen(ce
), ignore_case
);
754 !ie_match_stat(istate
, alias
, st
, ce_option
)) {
755 /* Nothing changed, really */
756 if (!S_ISGITLINK(alias
->ce_mode
))
757 ce_mark_uptodate(alias
);
758 alias
->ce_flags
|= CE_ADDED
;
760 discard_cache_entry(ce
);
765 if (index_path(istate
, &ce
->oid
, path
, st
, newflags
)) {
766 discard_cache_entry(ce
);
767 return error(_("unable to index file '%s'"), path
);
770 set_object_name_for_intent_to_add_entry(ce
);
772 if (ignore_case
&& alias
&& different_name(ce
, alias
))
773 ce
= create_alias_ce(istate
, ce
, alias
);
774 ce
->ce_flags
|= CE_ADDED
;
776 /* It was suspected to be racily clean, but it turns out to be Ok */
779 oideq(&alias
->oid
, &ce
->oid
) &&
780 ce
->ce_mode
== alias
->ce_mode
);
783 discard_cache_entry(ce
);
784 else if (add_index_entry(istate
, ce
, add_option
)) {
785 discard_cache_entry(ce
);
786 return error(_("unable to add '%s' to index"), path
);
788 if (verbose
&& !was_same
)
789 printf("add '%s'\n", path
);
793 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
796 if (lstat(path
, &st
))
797 die_errno(_("unable to stat '%s'"), path
);
798 return add_to_index(istate
, path
, &st
, flags
);
801 struct cache_entry
*make_empty_cache_entry(struct index_state
*istate
, size_t len
)
803 return mem_pool__ce_calloc(find_mem_pool(istate
), len
);
806 struct cache_entry
*make_empty_transient_cache_entry(size_t len
)
808 return xcalloc(1, cache_entry_size(len
));
811 struct cache_entry
*make_cache_entry(struct index_state
*istate
,
813 const struct object_id
*oid
,
816 unsigned int refresh_options
)
818 struct cache_entry
*ce
, *ret
;
821 if (!verify_path(path
, mode
)) {
822 error(_("invalid path '%s'"), path
);
827 ce
= make_empty_cache_entry(istate
, len
);
829 oidcpy(&ce
->oid
, oid
);
830 memcpy(ce
->name
, path
, len
);
831 ce
->ce_flags
= create_ce_flags(stage
);
832 ce
->ce_namelen
= len
;
833 ce
->ce_mode
= create_ce_mode(mode
);
835 ret
= refresh_cache_entry(istate
, ce
, refresh_options
);
837 discard_cache_entry(ce
);
841 struct cache_entry
*make_transient_cache_entry(unsigned int mode
, const struct object_id
*oid
,
842 const char *path
, int stage
)
844 struct cache_entry
*ce
;
847 if (!verify_path(path
, mode
)) {
848 error(_("invalid path '%s'"), path
);
853 ce
= make_empty_transient_cache_entry(len
);
855 oidcpy(&ce
->oid
, oid
);
856 memcpy(ce
->name
, path
, len
);
857 ce
->ce_flags
= create_ce_flags(stage
);
858 ce
->ce_namelen
= len
;
859 ce
->ce_mode
= create_ce_mode(mode
);
865 * Chmod an index entry with either +x or -x.
867 * Returns -1 if the chmod for the particular cache entry failed (if it's
868 * not a regular file), -2 if an invalid flip argument is passed in, 0
871 int chmod_index_entry(struct index_state
*istate
, struct cache_entry
*ce
,
874 if (!S_ISREG(ce
->ce_mode
))
881 ce
->ce_mode
&= ~0111;
886 cache_tree_invalidate_path(istate
, ce
->name
);
887 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
888 mark_fsmonitor_invalid(istate
, ce
);
889 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
894 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
896 int len
= ce_namelen(a
);
897 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
901 * We fundamentally don't like some paths: we don't want
902 * dot or dot-dot anywhere, and for obvious reasons don't
903 * want to recurse into ".git" either.
905 * Also, we don't want double slashes or slashes at the
906 * end that can make pathnames ambiguous.
908 static int verify_dotfile(const char *rest
, unsigned mode
)
911 * The first character was '.', but that
912 * has already been discarded, we now test
916 /* "." is not allowed */
917 if (*rest
== '\0' || is_dir_sep(*rest
))
922 * ".git" followed by NUL or slash is bad. Note that we match
923 * case-insensitively here, even if ignore_case is not set.
924 * This outlaws ".GIT" everywhere out of an abundance of caution,
925 * since there's really no good reason to allow it.
927 * Once we've seen ".git", we can also find ".gitmodules", etc (also
928 * case-insensitively).
932 if (rest
[1] != 'i' && rest
[1] != 'I')
934 if (rest
[2] != 't' && rest
[2] != 'T')
936 if (rest
[3] == '\0' || is_dir_sep(rest
[3]))
940 if (skip_iprefix(rest
, "modules", &rest
) &&
941 (*rest
== '\0' || is_dir_sep(*rest
)))
946 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
952 int verify_path(const char *path
, unsigned mode
)
956 if (has_dos_drive_prefix(path
))
966 if (is_hfs_dotgit(path
))
969 if (is_hfs_dotgitmodules(path
))
974 if (is_ntfs_dotgit(path
))
977 if (is_ntfs_dotgitmodules(path
))
983 if ((c
== '.' && !verify_dotfile(path
, mode
)) ||
984 is_dir_sep(c
) || c
== '\0')
992 * Do we have another file that has the beginning components being a
993 * proper superset of the name we're trying to add?
995 static int has_file_name(struct index_state
*istate
,
996 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
999 int len
= ce_namelen(ce
);
1000 int stage
= ce_stage(ce
);
1001 const char *name
= ce
->name
;
1003 while (pos
< istate
->cache_nr
) {
1004 struct cache_entry
*p
= istate
->cache
[pos
++];
1006 if (len
>= ce_namelen(p
))
1008 if (memcmp(name
, p
->name
, len
))
1010 if (ce_stage(p
) != stage
)
1012 if (p
->name
[len
] != '/')
1014 if (p
->ce_flags
& CE_REMOVE
)
1019 remove_index_entry_at(istate
, --pos
);
1026 * Like strcmp(), but also return the offset of the first change.
1027 * If strings are equal, return the length.
1029 int strcmp_offset(const char *s1
, const char *s2
, size_t *first_change
)
1034 return strcmp(s1
, s2
);
1036 for (k
= 0; s1
[k
] == s2
[k
]; k
++)
1041 return (unsigned char)s1
[k
] - (unsigned char)s2
[k
];
1045 * Do we have another file with a pathname that is a proper
1046 * subset of the name we're trying to add?
1048 * That is, is there another file in the index with a path
1049 * that matches a sub-directory in the given entry?
1051 static int has_dir_name(struct index_state
*istate
,
1052 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1055 int stage
= ce_stage(ce
);
1056 const char *name
= ce
->name
;
1057 const char *slash
= name
+ ce_namelen(ce
);
1062 * We are frequently called during an iteration on a sorted
1063 * list of pathnames and while building a new index. Therefore,
1064 * there is a high probability that this entry will eventually
1065 * be appended to the index, rather than inserted in the middle.
1066 * If we can confirm that, we can avoid binary searches on the
1067 * components of the pathname.
1069 * Compare the entry's full path with the last path in the index.
1071 if (istate
->cache_nr
> 0) {
1072 cmp_last
= strcmp_offset(name
,
1073 istate
->cache
[istate
->cache_nr
- 1]->name
,
1076 if (len_eq_last
== 0) {
1078 * The entry sorts AFTER the last one in the
1079 * index and their paths have no common prefix,
1080 * so there cannot be a F/D conflict.
1085 * The entry sorts AFTER the last one in the
1086 * index, but has a common prefix. Fall through
1087 * to the loop below to disect the entry's path
1088 * and see where the difference is.
1091 } else if (cmp_last
== 0) {
1093 * The entry exactly matches the last one in the
1094 * index, but because of multiple stage and CE_REMOVE
1095 * items, we fall through and let the regular search
1105 if (*--slash
== '/')
1107 if (slash
<= ce
->name
)
1114 * (len + 1) is a directory boundary (including
1115 * the trailing slash). And since the loop is
1116 * decrementing "slash", the first iteration is
1117 * the longest directory prefix; subsequent
1118 * iterations consider parent directories.
1121 if (len
+ 1 <= len_eq_last
) {
1123 * The directory prefix (including the trailing
1124 * slash) also appears as a prefix in the last
1125 * entry, so the remainder cannot collide (because
1126 * strcmp said the whole path was greater).
1131 * LT: last: xxx/file_A
1137 if (len
> len_eq_last
) {
1139 * This part of the directory prefix (excluding
1140 * the trailing slash) is longer than the known
1141 * equal portions, so this sub-directory cannot
1142 * collide with a file.
1150 if (istate
->cache_nr
> 0 &&
1151 ce_namelen(istate
->cache
[istate
->cache_nr
- 1]) > len
) {
1153 * The directory prefix lines up with part of
1154 * a longer file or directory name, but sorts
1155 * after it, so this sub-directory cannot
1156 * collide with a file.
1158 * last: xxx/yy-file (because '-' sorts before '/')
1165 * This is a possible collision. Fall through and
1166 * let the regular search code handle it.
1173 pos
= index_name_stage_pos(istate
, name
, len
, stage
);
1176 * Found one, but not so fast. This could
1177 * be a marker that says "I was here, but
1178 * I am being removed". Such an entry is
1179 * not a part of the resulting tree, and
1180 * it is Ok to have a directory at the same
1183 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
1187 remove_index_entry_at(istate
, pos
);
1195 * Trivial optimization: if we find an entry that
1196 * already matches the sub-directory, then we know
1197 * we're ok, and we can exit.
1199 while (pos
< istate
->cache_nr
) {
1200 struct cache_entry
*p
= istate
->cache
[pos
];
1201 if ((ce_namelen(p
) <= len
) ||
1202 (p
->name
[len
] != '/') ||
1203 memcmp(p
->name
, name
, len
))
1204 break; /* not our subdirectory */
1205 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
1207 * p is at the same stage as our entry, and
1208 * is a subdirectory of what we are looking
1209 * at, so we cannot have conflicts at our
1210 * level or anything shorter.
1219 /* We may be in a situation where we already have path/file and path
1220 * is being added, or we already have path and path/file is being
1221 * added. Either one would result in a nonsense tree that has path
1222 * twice when git-write-tree tries to write it out. Prevent it.
1224 * If ok-to-replace is specified, we remove the conflicting entries
1225 * from the cache so the caller should recompute the insert position.
1226 * When this happens, we return non-zero.
1228 static int check_file_directory_conflict(struct index_state
*istate
,
1229 const struct cache_entry
*ce
,
1230 int pos
, int ok_to_replace
)
1235 * When ce is an "I am going away" entry, we allow it to be added
1237 if (ce
->ce_flags
& CE_REMOVE
)
1241 * We check if the path is a sub-path of a subsequent pathname
1242 * first, since removing those will not change the position
1245 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
1248 * Then check if the path might have a clashing sub-directory
1251 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
1254 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1257 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
1258 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
1259 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
1260 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
1262 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1263 cache_tree_invalidate_path(istate
, ce
->name
);
1266 * If this entry's path sorts after the last entry in the index,
1267 * we can avoid searching for it.
1269 if (istate
->cache_nr
> 0 &&
1270 strcmp(ce
->name
, istate
->cache
[istate
->cache_nr
- 1]->name
) > 0)
1271 pos
= -istate
->cache_nr
- 1;
1273 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1275 /* existing match? Just replace it. */
1278 replace_index_entry(istate
, pos
, ce
);
1283 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1284 untracked_cache_add_to_index(istate
, ce
->name
);
1287 * Inserting a merged entry ("stage 0") into the index
1288 * will always replace all non-merged entries..
1290 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
1291 while (ce_same_name(istate
->cache
[pos
], ce
)) {
1293 if (!remove_index_entry_at(istate
, pos
))
1300 if (!verify_path(ce
->name
, ce
->ce_mode
))
1301 return error(_("invalid path '%s'"), ce
->name
);
1303 if (!skip_df_check
&&
1304 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1306 return error(_("'%s' appears as both a file and as a directory"),
1308 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1314 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1318 if (option
& ADD_CACHE_JUST_APPEND
)
1319 pos
= istate
->cache_nr
;
1322 ret
= add_index_entry_with_check(istate
, ce
, option
);
1328 /* Make sure the array is big enough .. */
1329 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1333 if (istate
->cache_nr
> pos
+ 1)
1334 MOVE_ARRAY(istate
->cache
+ pos
+ 1, istate
->cache
+ pos
,
1335 istate
->cache_nr
- pos
- 1);
1336 set_index_entry(istate
, pos
, ce
);
1337 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1342 * "refresh" does not calculate a new sha1 file or bring the
1343 * cache up-to-date for mode/content changes. But what it
1344 * _does_ do is to "re-match" the stat information of a file
1345 * with the cache, so that you can refresh the cache for a
1346 * file that hasn't been changed but where the stat entry is
1349 * For example, you'd want to do this after doing a "git-read-tree",
1350 * to link up the stat cache details with the proper files.
1352 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1353 struct cache_entry
*ce
,
1354 unsigned int options
, int *err
,
1358 struct cache_entry
*updated
;
1360 int refresh
= options
& CE_MATCH_REFRESH
;
1361 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1362 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1363 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1364 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
1366 if (!refresh
|| ce_uptodate(ce
))
1369 if (!ignore_fsmonitor
)
1370 refresh_fsmonitor(istate
);
1372 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1373 * that the change to the work tree does not matter and told
1376 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1377 ce_mark_uptodate(ce
);
1380 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1381 ce_mark_uptodate(ce
);
1384 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
)) {
1385 ce_mark_uptodate(ce
);
1389 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1397 if (lstat(ce
->name
, &st
) < 0) {
1398 if (ignore_missing
&& errno
== ENOENT
)
1405 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1407 *changed_ret
= changed
;
1410 * The path is unchanged. If we were told to ignore
1411 * valid bit, then we did the actual stat check and
1412 * found that the entry is unmodified. If the entry
1413 * is not marked VALID, this is the place to mark it
1414 * valid again, under "assume unchanged" mode.
1416 if (ignore_valid
&& assume_unchanged
&&
1417 !(ce
->ce_flags
& CE_VALID
))
1418 ; /* mark this one VALID again */
1421 * We do not mark the index itself "modified"
1422 * because CE_UPTODATE flag is in-core only;
1423 * we are not going to write this change out.
1425 if (!S_ISGITLINK(ce
->ce_mode
)) {
1426 ce_mark_uptodate(ce
);
1427 mark_fsmonitor_valid(ce
);
1433 if (ie_modified(istate
, ce
, &st
, options
)) {
1439 updated
= make_empty_cache_entry(istate
, ce_namelen(ce
));
1440 copy_cache_entry(updated
, ce
);
1441 memcpy(updated
->name
, ce
->name
, ce
->ce_namelen
+ 1);
1442 fill_stat_cache_info(updated
, &st
);
1444 * If ignore_valid is not set, we should leave CE_VALID bit
1445 * alone. Otherwise, paths marked with --no-assume-unchanged
1446 * (i.e. things to be edited) will reacquire CE_VALID bit
1447 * automatically, which is not really what we want.
1449 if (!ignore_valid
&& assume_unchanged
&&
1450 !(ce
->ce_flags
& CE_VALID
))
1451 updated
->ce_flags
&= ~CE_VALID
;
1453 /* istate->cache_changed is updated in the caller */
1457 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1458 int * first
, const char *header_msg
)
1460 if (in_porcelain
&& *first
&& header_msg
) {
1461 printf("%s\n", header_msg
);
1467 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1468 const struct pathspec
*pathspec
,
1469 char *seen
, const char *header_msg
)
1473 int really
= (flags
& REFRESH_REALLY
) != 0;
1474 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1475 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1476 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1477 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1479 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1480 unsigned int options
= (CE_MATCH_REFRESH
|
1481 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1482 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1483 const char *modified_fmt
;
1484 const char *deleted_fmt
;
1485 const char *typechange_fmt
;
1486 const char *added_fmt
;
1487 const char *unmerged_fmt
;
1488 struct progress
*progress
= NULL
;
1490 if (flags
& REFRESH_PROGRESS
&& isatty(2))
1491 progress
= start_delayed_progress(_("Refresh index"),
1494 trace_performance_enter();
1495 modified_fmt
= in_porcelain
? "M\t%s\n" : "%s: needs update\n";
1496 deleted_fmt
= in_porcelain
? "D\t%s\n" : "%s: needs update\n";
1497 typechange_fmt
= in_porcelain
? "T\t%s\n" : "%s: needs update\n";
1498 added_fmt
= in_porcelain
? "A\t%s\n" : "%s: needs update\n";
1499 unmerged_fmt
= in_porcelain
? "U\t%s\n" : "%s: needs merge\n";
1501 * Use the multi-threaded preload_index() to refresh most of the
1502 * cache entries quickly then in the single threaded loop below,
1503 * we only have to do the special cases that are left.
1505 preload_index(istate
, pathspec
, 0);
1506 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1507 struct cache_entry
*ce
, *new_entry
;
1508 int cache_errno
= 0;
1512 ce
= istate
->cache
[i
];
1513 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1516 if (pathspec
&& !ce_path_match(istate
, ce
, pathspec
, seen
))
1520 while ((i
< istate
->cache_nr
) &&
1521 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1527 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1528 &first
, header_msg
);
1536 new_entry
= refresh_cache_ent(istate
, ce
, options
, &cache_errno
, &changed
);
1537 if (new_entry
== ce
)
1540 display_progress(progress
, i
);
1544 if (really
&& cache_errno
== EINVAL
) {
1545 /* If we are doing --really-refresh that
1546 * means the index is not valid anymore.
1548 ce
->ce_flags
&= ~CE_VALID
;
1549 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1550 mark_fsmonitor_invalid(istate
, ce
);
1551 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1556 if (cache_errno
== ENOENT
)
1558 else if (ce_intent_to_add(ce
))
1559 fmt
= added_fmt
; /* must be before other checks */
1560 else if (changed
& TYPE_CHANGED
)
1561 fmt
= typechange_fmt
;
1565 ce
->name
, in_porcelain
, &first
, header_msg
);
1570 replace_index_entry(istate
, i
, new_entry
);
1573 display_progress(progress
, istate
->cache_nr
);
1574 stop_progress(&progress
);
1576 trace_performance_leave("refresh index");
1580 struct cache_entry
*refresh_cache_entry(struct index_state
*istate
,
1581 struct cache_entry
*ce
,
1582 unsigned int options
)
1584 return refresh_cache_ent(istate
, ce
, options
, NULL
, NULL
);
1588 /*****************************************************************
1590 *****************************************************************/
1592 #define INDEX_FORMAT_DEFAULT 3
1594 static unsigned int get_index_format_default(void)
1596 char *envversion
= getenv("GIT_INDEX_VERSION");
1599 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1602 if (!git_config_get_int("index.version", &value
))
1604 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1605 warning(_("index.version set, but the value is invalid.\n"
1606 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1607 return INDEX_FORMAT_DEFAULT
;
1612 version
= strtoul(envversion
, &endp
, 10);
1614 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1615 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1616 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1617 version
= INDEX_FORMAT_DEFAULT
;
1623 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1624 * Again - this is just a (very strong in practice) heuristic that
1625 * the inode hasn't changed.
1627 * We save the fields in big-endian order to allow using the
1628 * index file over NFS transparently.
1630 struct ondisk_cache_entry
{
1631 struct cache_time ctime
;
1632 struct cache_time mtime
;
1639 unsigned char sha1
[20];
1641 char name
[FLEX_ARRAY
]; /* more */
1645 * This struct is used when CE_EXTENDED bit is 1
1646 * The struct must match ondisk_cache_entry exactly from
1649 struct ondisk_cache_entry_extended
{
1650 struct cache_time ctime
;
1651 struct cache_time mtime
;
1658 unsigned char sha1
[20];
1661 char name
[FLEX_ARRAY
]; /* more */
1664 /* These are only used for v3 or lower */
1665 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1666 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1667 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1668 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1669 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1670 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1671 ondisk_cache_entry_size(ce_namelen(ce)))
1673 /* Allow fsck to force verification of the index checksum. */
1674 int verify_index_checksum
;
1676 /* Allow fsck to force verification of the cache entry order. */
1677 int verify_ce_order
;
1679 static int verify_hdr(const struct cache_header
*hdr
, unsigned long size
)
1682 unsigned char hash
[GIT_MAX_RAWSZ
];
1685 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1686 return error(_("bad signature 0x%08x"), hdr
->hdr_signature
);
1687 hdr_version
= ntohl(hdr
->hdr_version
);
1688 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1689 return error(_("bad index version %d"), hdr_version
);
1691 if (!verify_index_checksum
)
1694 the_hash_algo
->init_fn(&c
);
1695 the_hash_algo
->update_fn(&c
, hdr
, size
- the_hash_algo
->rawsz
);
1696 the_hash_algo
->final_fn(hash
, &c
);
1697 if (!hasheq(hash
, (unsigned char *)hdr
+ size
- the_hash_algo
->rawsz
))
1698 return error(_("bad index file sha1 signature"));
1702 static int read_index_extension(struct index_state
*istate
,
1703 const char *ext
, const char *data
, unsigned long sz
)
1705 switch (CACHE_EXT(ext
)) {
1706 case CACHE_EXT_TREE
:
1707 istate
->cache_tree
= cache_tree_read(data
, sz
);
1709 case CACHE_EXT_RESOLVE_UNDO
:
1710 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1712 case CACHE_EXT_LINK
:
1713 if (read_link_extension(istate
, data
, sz
))
1716 case CACHE_EXT_UNTRACKED
:
1717 istate
->untracked
= read_untracked_extension(data
, sz
);
1719 case CACHE_EXT_FSMONITOR
:
1720 read_fsmonitor_extension(istate
, data
, sz
);
1722 case CACHE_EXT_ENDOFINDEXENTRIES
:
1723 case CACHE_EXT_INDEXENTRYOFFSETTABLE
:
1724 /* already handled in do_read_index() */
1727 if (*ext
< 'A' || 'Z' < *ext
)
1728 return error(_("index uses %.4s extension, which we do not understand"),
1730 fprintf_ln(stderr
, _("ignoring %.4s extension"), ext
);
1736 static struct cache_entry
*create_from_disk(struct mem_pool
*ce_mem_pool
,
1737 unsigned int version
,
1738 struct ondisk_cache_entry
*ondisk
,
1739 unsigned long *ent_size
,
1740 const struct cache_entry
*previous_ce
)
1742 struct cache_entry
*ce
;
1746 size_t copy_len
= 0;
1748 * Adjacent cache entries tend to share the leading paths, so it makes
1749 * sense to only store the differences in later entries. In the v4
1750 * on-disk format of the index, each on-disk cache entry stores the
1751 * number of bytes to be stripped from the end of the previous name,
1752 * and the bytes to append to the result, to come up with its name.
1754 int expand_name_field
= version
== 4;
1756 /* On-disk flags are just 16 bits */
1757 flags
= get_be16(&ondisk
->flags
);
1758 len
= flags
& CE_NAMEMASK
;
1760 if (flags
& CE_EXTENDED
) {
1761 struct ondisk_cache_entry_extended
*ondisk2
;
1763 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1764 extended_flags
= get_be16(&ondisk2
->flags2
) << 16;
1765 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1766 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1767 die(_("unknown index entry format 0x%08x"), extended_flags
);
1768 flags
|= extended_flags
;
1769 name
= ondisk2
->name
;
1772 name
= ondisk
->name
;
1774 if (expand_name_field
) {
1775 const unsigned char *cp
= (const unsigned char *)name
;
1776 size_t strip_len
, previous_len
;
1778 /* If we're at the begining of a block, ignore the previous name */
1779 strip_len
= decode_varint(&cp
);
1781 previous_len
= previous_ce
->ce_namelen
;
1782 if (previous_len
< strip_len
)
1783 die(_("malformed name field in the index, near path '%s'"),
1785 copy_len
= previous_len
- strip_len
;
1787 name
= (const char *)cp
;
1790 if (len
== CE_NAMEMASK
) {
1792 if (expand_name_field
)
1796 ce
= mem_pool__ce_alloc(ce_mem_pool
, len
);
1798 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1799 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1800 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1801 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1802 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1803 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1804 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1805 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1806 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1807 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1808 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1809 ce
->ce_namelen
= len
;
1811 hashcpy(ce
->oid
.hash
, ondisk
->sha1
);
1813 if (expand_name_field
) {
1815 memcpy(ce
->name
, previous_ce
->name
, copy_len
);
1816 memcpy(ce
->name
+ copy_len
, name
, len
+ 1 - copy_len
);
1817 *ent_size
= (name
- ((char *)ondisk
)) + len
+ 1 - copy_len
;
1819 memcpy(ce
->name
, name
, len
+ 1);
1820 *ent_size
= ondisk_ce_size(ce
);
1825 static void check_ce_order(struct index_state
*istate
)
1829 if (!verify_ce_order
)
1832 for (i
= 1; i
< istate
->cache_nr
; i
++) {
1833 struct cache_entry
*ce
= istate
->cache
[i
- 1];
1834 struct cache_entry
*next_ce
= istate
->cache
[i
];
1835 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1837 if (0 < name_compare
)
1838 die(_("unordered stage entries in index"));
1839 if (!name_compare
) {
1841 die(_("multiple stage entries for merged file '%s'"),
1843 if (ce_stage(ce
) > ce_stage(next_ce
))
1844 die(_("unordered stage entries for '%s'"),
1850 static void tweak_untracked_cache(struct index_state
*istate
)
1852 switch (git_config_get_untracked_cache()) {
1853 case -1: /* keep: do nothing */
1856 remove_untracked_cache(istate
);
1859 add_untracked_cache(istate
);
1861 default: /* unknown value: do nothing */
1866 static void tweak_split_index(struct index_state
*istate
)
1868 switch (git_config_get_split_index()) {
1869 case -1: /* unset: do nothing */
1872 remove_split_index(istate
);
1875 add_split_index(istate
);
1877 default: /* unknown value: do nothing */
1882 static void post_read_index_from(struct index_state
*istate
)
1884 check_ce_order(istate
);
1885 tweak_untracked_cache(istate
);
1886 tweak_split_index(istate
);
1887 tweak_fsmonitor(istate
);
1890 static size_t estimate_cache_size_from_compressed(unsigned int entries
)
1892 return entries
* (sizeof(struct cache_entry
) + CACHE_ENTRY_PATH_LENGTH
);
1895 static size_t estimate_cache_size(size_t ondisk_size
, unsigned int entries
)
1897 long per_entry
= sizeof(struct cache_entry
) - sizeof(struct ondisk_cache_entry
);
1900 * Account for potential alignment differences.
1902 per_entry
+= align_padding_size(sizeof(struct cache_entry
), -sizeof(struct ondisk_cache_entry
));
1903 return ondisk_size
+ entries
* per_entry
;
1906 struct index_entry_offset
1908 /* starting byte offset into index file, count of index entries in this block */
1912 struct index_entry_offset_table
1915 struct index_entry_offset entries
[FLEX_ARRAY
];
1918 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
);
1919 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
);
1921 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
);
1922 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
);
1924 struct load_index_extensions
1927 struct index_state
*istate
;
1930 unsigned long src_offset
;
1933 static void *load_index_extensions(void *_data
)
1935 struct load_index_extensions
*p
= _data
;
1936 unsigned long src_offset
= p
->src_offset
;
1938 while (src_offset
<= p
->mmap_size
- the_hash_algo
->rawsz
- 8) {
1939 /* After an array of active_nr index entries,
1940 * there can be arbitrary number of extended
1941 * sections, each of which is prefixed with
1942 * extension name (4-byte) and section length
1943 * in 4-byte network byte order.
1945 uint32_t extsize
= get_be32(p
->mmap
+ src_offset
+ 4);
1946 if (read_index_extension(p
->istate
,
1947 p
->mmap
+ src_offset
,
1948 p
->mmap
+ src_offset
+ 8,
1950 munmap((void *)p
->mmap
, p
->mmap_size
);
1951 die(_("index file corrupt"));
1954 src_offset
+= extsize
;
1961 * A helper function that will load the specified range of cache entries
1962 * from the memory mapped file and add them to the given index.
1964 static unsigned long load_cache_entry_block(struct index_state
*istate
,
1965 struct mem_pool
*ce_mem_pool
, int offset
, int nr
, const char *mmap
,
1966 unsigned long start_offset
, const struct cache_entry
*previous_ce
)
1969 unsigned long src_offset
= start_offset
;
1971 for (i
= offset
; i
< offset
+ nr
; i
++) {
1972 struct ondisk_cache_entry
*disk_ce
;
1973 struct cache_entry
*ce
;
1974 unsigned long consumed
;
1976 disk_ce
= (struct ondisk_cache_entry
*)(mmap
+ src_offset
);
1977 ce
= create_from_disk(ce_mem_pool
, istate
->version
, disk_ce
, &consumed
, previous_ce
);
1978 set_index_entry(istate
, i
, ce
);
1980 src_offset
+= consumed
;
1983 return src_offset
- start_offset
;
1986 static unsigned long load_all_cache_entries(struct index_state
*istate
,
1987 const char *mmap
, size_t mmap_size
, unsigned long src_offset
)
1989 unsigned long consumed
;
1991 if (istate
->version
== 4) {
1992 mem_pool_init(&istate
->ce_mem_pool
,
1993 estimate_cache_size_from_compressed(istate
->cache_nr
));
1995 mem_pool_init(&istate
->ce_mem_pool
,
1996 estimate_cache_size(mmap_size
, istate
->cache_nr
));
1999 consumed
= load_cache_entry_block(istate
, istate
->ce_mem_pool
,
2000 0, istate
->cache_nr
, mmap
, src_offset
, NULL
);
2005 * Mostly randomly chosen maximum thread counts: we
2006 * cap the parallelism to online_cpus() threads, and we want
2007 * to have at least 10000 cache entries per thread for it to
2008 * be worth starting a thread.
2011 #define THREAD_COST (10000)
2013 struct load_cache_entries_thread_data
2016 struct index_state
*istate
;
2017 struct mem_pool
*ce_mem_pool
;
2020 struct index_entry_offset_table
*ieot
;
2021 int ieot_start
; /* starting index into the ieot array */
2022 int ieot_blocks
; /* count of ieot entries to process */
2023 unsigned long consumed
; /* return # of bytes in index file processed */
2027 * A thread proc to run the load_cache_entries() computation
2028 * across multiple background threads.
2030 static void *load_cache_entries_thread(void *_data
)
2032 struct load_cache_entries_thread_data
*p
= _data
;
2035 /* iterate across all ieot blocks assigned to this thread */
2036 for (i
= p
->ieot_start
; i
< p
->ieot_start
+ p
->ieot_blocks
; i
++) {
2037 p
->consumed
+= load_cache_entry_block(p
->istate
, p
->ce_mem_pool
,
2038 p
->offset
, p
->ieot
->entries
[i
].nr
, p
->mmap
, p
->ieot
->entries
[i
].offset
, NULL
);
2039 p
->offset
+= p
->ieot
->entries
[i
].nr
;
2044 static unsigned long load_cache_entries_threaded(struct index_state
*istate
, const char *mmap
, size_t mmap_size
,
2045 unsigned long src_offset
, int nr_threads
, struct index_entry_offset_table
*ieot
)
2047 int i
, offset
, ieot_blocks
, ieot_start
, err
;
2048 struct load_cache_entries_thread_data
*data
;
2049 unsigned long consumed
= 0;
2051 /* a little sanity checking */
2052 if (istate
->name_hash_initialized
)
2053 BUG("the name hash isn't thread safe");
2055 mem_pool_init(&istate
->ce_mem_pool
, 0);
2057 /* ensure we have no more threads than we have blocks to process */
2058 if (nr_threads
> ieot
->nr
)
2059 nr_threads
= ieot
->nr
;
2060 data
= xcalloc(nr_threads
, sizeof(*data
));
2062 offset
= ieot_start
= 0;
2063 ieot_blocks
= DIV_ROUND_UP(ieot
->nr
, nr_threads
);
2064 for (i
= 0; i
< nr_threads
; i
++) {
2065 struct load_cache_entries_thread_data
*p
= &data
[i
];
2068 if (ieot_start
+ ieot_blocks
> ieot
->nr
)
2069 ieot_blocks
= ieot
->nr
- ieot_start
;
2075 p
->ieot_start
= ieot_start
;
2076 p
->ieot_blocks
= ieot_blocks
;
2078 /* create a mem_pool for each thread */
2080 for (j
= p
->ieot_start
; j
< p
->ieot_start
+ p
->ieot_blocks
; j
++)
2081 nr
+= p
->ieot
->entries
[j
].nr
;
2082 if (istate
->version
== 4) {
2083 mem_pool_init(&p
->ce_mem_pool
,
2084 estimate_cache_size_from_compressed(nr
));
2086 mem_pool_init(&p
->ce_mem_pool
,
2087 estimate_cache_size(mmap_size
, nr
));
2090 err
= pthread_create(&p
->pthread
, NULL
, load_cache_entries_thread
, p
);
2092 die(_("unable to create load_cache_entries thread: %s"), strerror(err
));
2094 /* increment by the number of cache entries in the ieot block being processed */
2095 for (j
= 0; j
< ieot_blocks
; j
++)
2096 offset
+= ieot
->entries
[ieot_start
+ j
].nr
;
2097 ieot_start
+= ieot_blocks
;
2100 for (i
= 0; i
< nr_threads
; i
++) {
2101 struct load_cache_entries_thread_data
*p
= &data
[i
];
2103 err
= pthread_join(p
->pthread
, NULL
);
2105 die(_("unable to join load_cache_entries thread: %s"), strerror(err
));
2106 mem_pool_combine(istate
->ce_mem_pool
, p
->ce_mem_pool
);
2107 consumed
+= p
->consumed
;
2115 /* remember to discard_cache() before reading a different cache! */
2116 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
2120 unsigned long src_offset
;
2121 const struct cache_header
*hdr
;
2124 struct load_index_extensions p
;
2125 size_t extension_offset
= 0;
2126 int nr_threads
, cpus
;
2127 struct index_entry_offset_table
*ieot
= NULL
;
2129 if (istate
->initialized
)
2130 return istate
->cache_nr
;
2132 istate
->timestamp
.sec
= 0;
2133 istate
->timestamp
.nsec
= 0;
2134 fd
= open(path
, O_RDONLY
);
2136 if (!must_exist
&& errno
== ENOENT
)
2138 die_errno(_("%s: index file open failed"), path
);
2142 die_errno(_("%s: cannot stat the open index"), path
);
2144 mmap_size
= xsize_t(st
.st_size
);
2145 if (mmap_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2146 die(_("%s: index file smaller than expected"), path
);
2148 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2149 if (mmap
== MAP_FAILED
)
2150 die_errno(_("%s: unable to map index file"), path
);
2153 hdr
= (const struct cache_header
*)mmap
;
2154 if (verify_hdr(hdr
, mmap_size
) < 0)
2157 hashcpy(istate
->oid
.hash
, (const unsigned char *)hdr
+ mmap_size
- the_hash_algo
->rawsz
);
2158 istate
->version
= ntohl(hdr
->hdr_version
);
2159 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
2160 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
2161 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(*istate
->cache
));
2162 istate
->initialized
= 1;
2166 p
.mmap_size
= mmap_size
;
2168 src_offset
= sizeof(*hdr
);
2170 if (git_config_get_index_threads(&nr_threads
))
2173 /* TODO: does creating more threads than cores help? */
2175 nr_threads
= istate
->cache_nr
/ THREAD_COST
;
2176 cpus
= online_cpus();
2177 if (nr_threads
> cpus
)
2184 if (nr_threads
> 1) {
2185 extension_offset
= read_eoie_extension(mmap
, mmap_size
);
2186 if (extension_offset
) {
2189 p
.src_offset
= extension_offset
;
2190 err
= pthread_create(&p
.pthread
, NULL
, load_index_extensions
, &p
);
2192 die(_("unable to create load_index_extensions thread: %s"), strerror(err
));
2199 * Locate and read the index entry offset table so that we can use it
2200 * to multi-thread the reading of the cache entries.
2202 if (extension_offset
&& nr_threads
> 1)
2203 ieot
= read_ieot_extension(mmap
, mmap_size
, extension_offset
);
2206 src_offset
+= load_cache_entries_threaded(istate
, mmap
, mmap_size
, src_offset
, nr_threads
, ieot
);
2209 src_offset
+= load_all_cache_entries(istate
, mmap
, mmap_size
, src_offset
);
2212 istate
->timestamp
.sec
= st
.st_mtime
;
2213 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2215 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2216 if (extension_offset
) {
2217 int ret
= pthread_join(p
.pthread
, NULL
);
2219 die(_("unable to join load_index_extensions thread: %s"), strerror(ret
));
2221 p
.src_offset
= src_offset
;
2222 load_index_extensions(&p
);
2224 munmap((void *)mmap
, mmap_size
);
2225 return istate
->cache_nr
;
2228 munmap((void *)mmap
, mmap_size
);
2229 die(_("index file corrupt"));
2233 * Signal that the shared index is used by updating its mtime.
2235 * This way, shared index can be removed if they have not been used
2238 static void freshen_shared_index(const char *shared_index
, int warn
)
2240 if (!check_and_freshen_file(shared_index
, 1) && warn
)
2241 warning(_("could not freshen shared index '%s'"), shared_index
);
2244 int read_index_from(struct index_state
*istate
, const char *path
,
2247 struct split_index
*split_index
;
2252 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2253 if (istate
->initialized
)
2254 return istate
->cache_nr
;
2256 trace_performance_enter();
2257 ret
= do_read_index(istate
, path
, 0);
2258 trace_performance_leave("read cache %s", path
);
2260 split_index
= istate
->split_index
;
2261 if (!split_index
|| is_null_oid(&split_index
->base_oid
)) {
2262 post_read_index_from(istate
);
2266 trace_performance_enter();
2267 if (split_index
->base
)
2268 discard_index(split_index
->base
);
2270 split_index
->base
= xcalloc(1, sizeof(*split_index
->base
));
2272 base_oid_hex
= oid_to_hex(&split_index
->base_oid
);
2273 base_path
= xstrfmt("%s/sharedindex.%s", gitdir
, base_oid_hex
);
2274 ret
= do_read_index(split_index
->base
, base_path
, 1);
2275 if (!oideq(&split_index
->base_oid
, &split_index
->base
->oid
))
2276 die(_("broken index, expect %s in %s, got %s"),
2277 base_oid_hex
, base_path
,
2278 oid_to_hex(&split_index
->base
->oid
));
2280 freshen_shared_index(base_path
, 0);
2281 merge_base_index(istate
);
2282 post_read_index_from(istate
);
2283 trace_performance_leave("read cache %s", base_path
);
2288 int is_index_unborn(struct index_state
*istate
)
2290 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
2293 int discard_index(struct index_state
*istate
)
2296 * Cache entries in istate->cache[] should have been allocated
2297 * from the memory pool associated with this index, or from an
2298 * associated split_index. There is no need to free individual
2299 * cache entries. validate_cache_entries can detect when this
2300 * assertion does not hold.
2302 validate_cache_entries(istate
);
2304 resolve_undo_clear_index(istate
);
2305 istate
->cache_nr
= 0;
2306 istate
->cache_changed
= 0;
2307 istate
->timestamp
.sec
= 0;
2308 istate
->timestamp
.nsec
= 0;
2309 free_name_hash(istate
);
2310 cache_tree_free(&(istate
->cache_tree
));
2311 istate
->initialized
= 0;
2312 FREE_AND_NULL(istate
->cache
);
2313 istate
->cache_alloc
= 0;
2314 discard_split_index(istate
);
2315 free_untracked_cache(istate
->untracked
);
2316 istate
->untracked
= NULL
;
2318 if (istate
->ce_mem_pool
) {
2319 mem_pool_discard(istate
->ce_mem_pool
, should_validate_cache_entries());
2320 istate
->ce_mem_pool
= NULL
;
2327 * Validate the cache entries of this index.
2328 * All cache entries associated with this index
2329 * should have been allocated by the memory pool
2330 * associated with this index, or by a referenced
2333 void validate_cache_entries(const struct index_state
*istate
)
2337 if (!should_validate_cache_entries() ||!istate
|| !istate
->initialized
)
2340 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2342 BUG("cache entry is not allocated from expected memory pool");
2343 } else if (!istate
->ce_mem_pool
||
2344 !mem_pool_contains(istate
->ce_mem_pool
, istate
->cache
[i
])) {
2345 if (!istate
->split_index
||
2346 !istate
->split_index
->base
||
2347 !istate
->split_index
->base
->ce_mem_pool
||
2348 !mem_pool_contains(istate
->split_index
->base
->ce_mem_pool
, istate
->cache
[i
])) {
2349 BUG("cache entry is not allocated from expected memory pool");
2354 if (istate
->split_index
)
2355 validate_cache_entries(istate
->split_index
->base
);
2358 int unmerged_index(const struct index_state
*istate
)
2361 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2362 if (ce_stage(istate
->cache
[i
]))
2368 int repo_index_has_changes(struct repository
*repo
,
2372 struct index_state
*istate
= repo
->index
;
2373 struct object_id cmp
;
2377 cmp
= tree
->object
.oid
;
2378 if (tree
|| !get_oid_tree("HEAD", &cmp
)) {
2379 struct diff_options opt
;
2381 repo_diff_setup(repo
, &opt
);
2382 opt
.flags
.exit_with_status
= 1;
2384 opt
.flags
.quick
= 1;
2385 do_diff_cache(&cmp
, &opt
);
2387 for (i
= 0; sb
&& i
< diff_queued_diff
.nr
; i
++) {
2389 strbuf_addch(sb
, ' ');
2390 strbuf_addstr(sb
, diff_queued_diff
.queue
[i
]->two
->path
);
2393 return opt
.flags
.has_changes
!= 0;
2395 for (i
= 0; sb
&& i
< istate
->cache_nr
; i
++) {
2397 strbuf_addch(sb
, ' ');
2398 strbuf_addstr(sb
, istate
->cache
[i
]->name
);
2400 return !!istate
->cache_nr
;
2404 #define WRITE_BUFFER_SIZE 8192
2405 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
2406 static unsigned long write_buffer_len
;
2408 static int ce_write_flush(git_hash_ctx
*context
, int fd
)
2410 unsigned int buffered
= write_buffer_len
;
2412 the_hash_algo
->update_fn(context
, write_buffer
, buffered
);
2413 if (write_in_full(fd
, write_buffer
, buffered
) < 0)
2415 write_buffer_len
= 0;
2420 static int ce_write(git_hash_ctx
*context
, int fd
, void *data
, unsigned int len
)
2423 unsigned int buffered
= write_buffer_len
;
2424 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
2427 memcpy(write_buffer
+ buffered
, data
, partial
);
2428 buffered
+= partial
;
2429 if (buffered
== WRITE_BUFFER_SIZE
) {
2430 write_buffer_len
= buffered
;
2431 if (ce_write_flush(context
, fd
))
2435 write_buffer_len
= buffered
;
2437 data
= (char *) data
+ partial
;
2442 static int write_index_ext_header(git_hash_ctx
*context
, git_hash_ctx
*eoie_context
,
2443 int fd
, unsigned int ext
, unsigned int sz
)
2448 the_hash_algo
->update_fn(eoie_context
, &ext
, 4);
2449 the_hash_algo
->update_fn(eoie_context
, &sz
, 4);
2451 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
2452 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
2455 static int ce_flush(git_hash_ctx
*context
, int fd
, unsigned char *hash
)
2457 unsigned int left
= write_buffer_len
;
2460 write_buffer_len
= 0;
2461 the_hash_algo
->update_fn(context
, write_buffer
, left
);
2464 /* Flush first if not enough space for hash signature */
2465 if (left
+ the_hash_algo
->rawsz
> WRITE_BUFFER_SIZE
) {
2466 if (write_in_full(fd
, write_buffer
, left
) < 0)
2471 /* Append the hash signature at the end */
2472 the_hash_algo
->final_fn(write_buffer
+ left
, context
);
2473 hashcpy(hash
, write_buffer
+ left
);
2474 left
+= the_hash_algo
->rawsz
;
2475 return (write_in_full(fd
, write_buffer
, left
) < 0) ? -1 : 0;
2478 static void ce_smudge_racily_clean_entry(struct index_state
*istate
,
2479 struct cache_entry
*ce
)
2482 * The only thing we care about in this function is to smudge the
2483 * falsely clean entry due to touch-update-touch race, so we leave
2484 * everything else as they are. We are called for entries whose
2485 * ce_stat_data.sd_mtime match the index file mtime.
2487 * Note that this actually does not do much for gitlinks, for
2488 * which ce_match_stat_basic() always goes to the actual
2489 * contents. The caller checks with is_racy_timestamp() which
2490 * always says "no" for gitlinks, so we are not called for them ;-)
2494 if (lstat(ce
->name
, &st
) < 0)
2496 if (ce_match_stat_basic(ce
, &st
))
2498 if (ce_modified_check_fs(istate
, ce
, &st
)) {
2499 /* This is "racily clean"; smudge it. Note that this
2500 * is a tricky code. At first glance, it may appear
2501 * that it can break with this sequence:
2503 * $ echo xyzzy >frotz
2504 * $ git-update-index --add frotz
2507 * $ echo filfre >nitfol
2508 * $ git-update-index --add nitfol
2510 * but it does not. When the second update-index runs,
2511 * it notices that the entry "frotz" has the same timestamp
2512 * as index, and if we were to smudge it by resetting its
2513 * size to zero here, then the object name recorded
2514 * in index is the 6-byte file but the cached stat information
2515 * becomes zero --- which would then match what we would
2516 * obtain from the filesystem next time we stat("frotz").
2518 * However, the second update-index, before calling
2519 * this function, notices that the cached size is 6
2520 * bytes and what is on the filesystem is an empty
2521 * file, and never calls us, so the cached size information
2522 * for "frotz" stays 6 which does not match the filesystem.
2524 ce
->ce_stat_data
.sd_size
= 0;
2528 /* Copy miscellaneous fields but not the name */
2529 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
2530 struct cache_entry
*ce
)
2534 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
2535 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
2536 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
2537 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
2538 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
2539 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
2540 ondisk
->mode
= htonl(ce
->ce_mode
);
2541 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
2542 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
2543 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
2544 hashcpy(ondisk
->sha1
, ce
->oid
.hash
);
2546 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
2547 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
2548 ondisk
->flags
= htons(flags
);
2549 if (ce
->ce_flags
& CE_EXTENDED
) {
2550 struct ondisk_cache_entry_extended
*ondisk2
;
2551 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
2552 ondisk2
->flags2
= htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
2556 static int ce_write_entry(git_hash_ctx
*c
, int fd
, struct cache_entry
*ce
,
2557 struct strbuf
*previous_name
, struct ondisk_cache_entry
*ondisk
)
2561 unsigned int saved_namelen
;
2562 int stripped_name
= 0;
2563 static unsigned char padding
[8] = { 0x00 };
2565 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2566 saved_namelen
= ce_namelen(ce
);
2571 if (ce
->ce_flags
& CE_EXTENDED
)
2572 size
= offsetof(struct ondisk_cache_entry_extended
, name
);
2574 size
= offsetof(struct ondisk_cache_entry
, name
);
2576 if (!previous_name
) {
2577 int len
= ce_namelen(ce
);
2578 copy_cache_entry_to_ondisk(ondisk
, ce
);
2579 result
= ce_write(c
, fd
, ondisk
, size
);
2581 result
= ce_write(c
, fd
, ce
->name
, len
);
2583 result
= ce_write(c
, fd
, padding
, align_padding_size(size
, len
));
2585 int common
, to_remove
, prefix_size
;
2586 unsigned char to_remove_vi
[16];
2588 (ce
->name
[common
] &&
2589 common
< previous_name
->len
&&
2590 ce
->name
[common
] == previous_name
->buf
[common
]);
2592 ; /* still matching */
2593 to_remove
= previous_name
->len
- common
;
2594 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
2596 copy_cache_entry_to_ondisk(ondisk
, ce
);
2597 result
= ce_write(c
, fd
, ondisk
, size
);
2599 result
= ce_write(c
, fd
, to_remove_vi
, prefix_size
);
2601 result
= ce_write(c
, fd
, ce
->name
+ common
, ce_namelen(ce
) - common
);
2603 result
= ce_write(c
, fd
, padding
, 1);
2605 strbuf_splice(previous_name
, common
, to_remove
,
2606 ce
->name
+ common
, ce_namelen(ce
) - common
);
2608 if (stripped_name
) {
2609 ce
->ce_namelen
= saved_namelen
;
2610 ce
->ce_flags
&= ~CE_STRIP_NAME
;
2617 * This function verifies if index_state has the correct sha1 of the
2618 * index file. Don't die if we have any other failure, just return 0.
2620 static int verify_index_from(const struct index_state
*istate
, const char *path
)
2625 unsigned char hash
[GIT_MAX_RAWSZ
];
2627 if (!istate
->initialized
)
2630 fd
= open(path
, O_RDONLY
);
2637 if (st
.st_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2640 n
= pread_in_full(fd
, hash
, the_hash_algo
->rawsz
, st
.st_size
- the_hash_algo
->rawsz
);
2641 if (n
!= the_hash_algo
->rawsz
)
2644 if (!hasheq(istate
->oid
.hash
, hash
))
2655 static int repo_verify_index(struct repository
*repo
)
2657 return verify_index_from(repo
->index
, repo
->index_file
);
2660 static int has_racy_timestamp(struct index_state
*istate
)
2662 int entries
= istate
->cache_nr
;
2665 for (i
= 0; i
< entries
; i
++) {
2666 struct cache_entry
*ce
= istate
->cache
[i
];
2667 if (is_racy_timestamp(istate
, ce
))
2673 void repo_update_index_if_able(struct repository
*repo
,
2674 struct lock_file
*lockfile
)
2676 if ((repo
->index
->cache_changed
||
2677 has_racy_timestamp(repo
->index
)) &&
2678 repo_verify_index(repo
))
2679 write_locked_index(repo
->index
, lockfile
, COMMIT_LOCK
);
2681 rollback_lock_file(lockfile
);
2684 static int record_eoie(void)
2688 if (!git_config_get_bool("index.recordendofindexentries", &val
))
2692 * As a convenience, the end of index entries extension
2693 * used for threading is written by default if the user
2694 * explicitly requested threaded index reads.
2696 return !git_config_get_index_threads(&val
) && val
!= 1;
2699 static int record_ieot(void)
2703 if (!git_config_get_bool("index.recordoffsettable", &val
))
2707 * As a convenience, the offset table used for threading is
2708 * written by default if the user explicitly requested
2709 * threaded index reads.
2711 return !git_config_get_index_threads(&val
) && val
!= 1;
2715 * On success, `tempfile` is closed. If it is the temporary file
2716 * of a `struct lock_file`, we will therefore effectively perform
2717 * a 'close_lock_file_gently()`. Since that is an implementation
2718 * detail of lockfiles, callers of `do_write_index()` should not
2721 static int do_write_index(struct index_state
*istate
, struct tempfile
*tempfile
,
2722 int strip_extensions
)
2724 uint64_t start
= getnanotime();
2725 int newfd
= tempfile
->fd
;
2726 git_hash_ctx c
, eoie_c
;
2727 struct cache_header hdr
;
2728 int i
, err
= 0, removed
, extended
, hdr_version
;
2729 struct cache_entry
**cache
= istate
->cache
;
2730 int entries
= istate
->cache_nr
;
2732 struct ondisk_cache_entry_extended ondisk
;
2733 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
2734 int drop_cache_tree
= istate
->drop_cache_tree
;
2736 int ieot_entries
= 1;
2737 struct index_entry_offset_table
*ieot
= NULL
;
2740 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
2741 if (cache
[i
]->ce_flags
& CE_REMOVE
)
2744 /* reduce extended entries if possible */
2745 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
2746 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
2748 cache
[i
]->ce_flags
|= CE_EXTENDED
;
2752 if (!istate
->version
) {
2753 istate
->version
= get_index_format_default();
2754 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
2755 init_split_index(istate
);
2758 /* demote version 3 to version 2 when the latter suffices */
2759 if (istate
->version
== 3 || istate
->version
== 2)
2760 istate
->version
= extended
? 3 : 2;
2762 hdr_version
= istate
->version
;
2764 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2765 hdr
.hdr_version
= htonl(hdr_version
);
2766 hdr
.hdr_entries
= htonl(entries
- removed
);
2768 the_hash_algo
->init_fn(&c
);
2769 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
2772 if (!HAVE_THREADS
|| git_config_get_index_threads(&nr_threads
))
2775 if (nr_threads
!= 1 && record_ieot()) {
2776 int ieot_blocks
, cpus
;
2779 * ensure default number of ieot blocks maps evenly to the
2780 * default number of threads that will process them leaving
2781 * room for the thread to load the index extensions.
2784 ieot_blocks
= istate
->cache_nr
/ THREAD_COST
;
2785 cpus
= online_cpus();
2786 if (ieot_blocks
> cpus
- 1)
2787 ieot_blocks
= cpus
- 1;
2789 ieot_blocks
= nr_threads
;
2790 if (ieot_blocks
> istate
->cache_nr
)
2791 ieot_blocks
= istate
->cache_nr
;
2795 * no reason to write out the IEOT extension if we don't
2796 * have enough blocks to utilize multi-threading
2798 if (ieot_blocks
> 1) {
2799 ieot
= xcalloc(1, sizeof(struct index_entry_offset_table
)
2800 + (ieot_blocks
* sizeof(struct index_entry_offset
)));
2801 ieot_entries
= DIV_ROUND_UP(entries
, ieot_blocks
);
2805 offset
= lseek(newfd
, 0, SEEK_CUR
);
2810 offset
+= write_buffer_len
;
2812 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
2814 for (i
= 0; i
< entries
; i
++) {
2815 struct cache_entry
*ce
= cache
[i
];
2816 if (ce
->ce_flags
& CE_REMOVE
)
2818 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2819 ce_smudge_racily_clean_entry(istate
, ce
);
2820 if (is_null_oid(&ce
->oid
)) {
2821 static const char msg
[] = "cache entry has null sha1: %s";
2822 static int allow
= -1;
2825 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2827 warning(msg
, ce
->name
);
2829 err
= error(msg
, ce
->name
);
2831 drop_cache_tree
= 1;
2833 if (ieot
&& i
&& (i
% ieot_entries
== 0)) {
2834 ieot
->entries
[ieot
->nr
].nr
= nr
;
2835 ieot
->entries
[ieot
->nr
].offset
= offset
;
2838 * If we have a V4 index, set the first byte to an invalid
2839 * character to ensure there is nothing common with the previous
2843 previous_name
->buf
[0] = 0;
2845 offset
= lseek(newfd
, 0, SEEK_CUR
);
2850 offset
+= write_buffer_len
;
2852 if (ce_write_entry(&c
, newfd
, ce
, previous_name
, (struct ondisk_cache_entry
*)&ondisk
) < 0)
2860 ieot
->entries
[ieot
->nr
].nr
= nr
;
2861 ieot
->entries
[ieot
->nr
].offset
= offset
;
2864 strbuf_release(&previous_name_buf
);
2871 /* Write extension data here */
2872 offset
= lseek(newfd
, 0, SEEK_CUR
);
2877 offset
+= write_buffer_len
;
2878 the_hash_algo
->init_fn(&eoie_c
);
2881 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2882 * can minimize the number of extensions we have to scan through to
2883 * find it during load. Write it out regardless of the
2884 * strip_extensions parameter as we need it when loading the shared
2888 struct strbuf sb
= STRBUF_INIT
;
2890 write_ieot_extension(&sb
, ieot
);
2891 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_INDEXENTRYOFFSETTABLE
, sb
.len
) < 0
2892 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2893 strbuf_release(&sb
);
2899 if (!strip_extensions
&& istate
->split_index
) {
2900 struct strbuf sb
= STRBUF_INIT
;
2902 err
= write_link_extension(&sb
, istate
) < 0 ||
2903 write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_LINK
,
2905 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2906 strbuf_release(&sb
);
2910 if (!strip_extensions
&& !drop_cache_tree
&& istate
->cache_tree
) {
2911 struct strbuf sb
= STRBUF_INIT
;
2913 cache_tree_write(&sb
, istate
->cache_tree
);
2914 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
2915 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2916 strbuf_release(&sb
);
2920 if (!strip_extensions
&& istate
->resolve_undo
) {
2921 struct strbuf sb
= STRBUF_INIT
;
2923 resolve_undo_write(&sb
, istate
->resolve_undo
);
2924 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_RESOLVE_UNDO
,
2926 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2927 strbuf_release(&sb
);
2931 if (!strip_extensions
&& istate
->untracked
) {
2932 struct strbuf sb
= STRBUF_INIT
;
2934 write_untracked_extension(&sb
, istate
->untracked
);
2935 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_UNTRACKED
,
2937 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2938 strbuf_release(&sb
);
2942 if (!strip_extensions
&& istate
->fsmonitor_last_update
) {
2943 struct strbuf sb
= STRBUF_INIT
;
2945 write_fsmonitor_extension(&sb
, istate
);
2946 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_FSMONITOR
, sb
.len
) < 0
2947 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2948 strbuf_release(&sb
);
2954 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
2955 * so that it can be found and processed before all the index entries are
2956 * read. Write it out regardless of the strip_extensions parameter as we need it
2957 * when loading the shared index.
2959 if (offset
&& record_eoie()) {
2960 struct strbuf sb
= STRBUF_INIT
;
2962 write_eoie_extension(&sb
, &eoie_c
, offset
);
2963 err
= write_index_ext_header(&c
, NULL
, newfd
, CACHE_EXT_ENDOFINDEXENTRIES
, sb
.len
) < 0
2964 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2965 strbuf_release(&sb
);
2970 if (ce_flush(&c
, newfd
, istate
->oid
.hash
))
2972 if (close_tempfile_gently(tempfile
)) {
2973 error(_("could not close '%s'"), tempfile
->filename
.buf
);
2976 if (stat(tempfile
->filename
.buf
, &st
))
2978 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
2979 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2980 trace_performance_since(start
, "write index, changed mask = %x", istate
->cache_changed
);
2984 void set_alternate_index_output(const char *name
)
2986 alternate_index_output
= name
;
2989 static int commit_locked_index(struct lock_file
*lk
)
2991 if (alternate_index_output
)
2992 return commit_lock_file_to(lk
, alternate_index_output
);
2994 return commit_lock_file(lk
);
2997 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3000 int ret
= do_write_index(istate
, lock
->tempfile
, 0);
3003 if (flags
& COMMIT_LOCK
)
3004 return commit_locked_index(lock
);
3005 return close_lock_file_gently(lock
);
3008 static int write_split_index(struct index_state
*istate
,
3009 struct lock_file
*lock
,
3013 prepare_to_write_split_index(istate
);
3014 ret
= do_write_locked_index(istate
, lock
, flags
);
3015 finish_writing_split_index(istate
);
3019 static const char *shared_index_expire
= "2.weeks.ago";
3021 static unsigned long get_shared_index_expire_date(void)
3023 static unsigned long shared_index_expire_date
;
3024 static int shared_index_expire_date_prepared
;
3026 if (!shared_index_expire_date_prepared
) {
3027 git_config_get_expiry("splitindex.sharedindexexpire",
3028 &shared_index_expire
);
3029 shared_index_expire_date
= approxidate(shared_index_expire
);
3030 shared_index_expire_date_prepared
= 1;
3033 return shared_index_expire_date
;
3036 static int should_delete_shared_index(const char *shared_index_path
)
3039 unsigned long expiration
;
3041 /* Check timestamp */
3042 expiration
= get_shared_index_expire_date();
3045 if (stat(shared_index_path
, &st
))
3046 return error_errno(_("could not stat '%s'"), shared_index_path
);
3047 if (st
.st_mtime
> expiration
)
3053 static int clean_shared_index_files(const char *current_hex
)
3056 DIR *dir
= opendir(get_git_dir());
3059 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3061 while ((de
= readdir(dir
)) != NULL
) {
3062 const char *sha1_hex
;
3063 const char *shared_index_path
;
3064 if (!skip_prefix(de
->d_name
, "sharedindex.", &sha1_hex
))
3066 if (!strcmp(sha1_hex
, current_hex
))
3068 shared_index_path
= git_path("%s", de
->d_name
);
3069 if (should_delete_shared_index(shared_index_path
) > 0 &&
3070 unlink(shared_index_path
))
3071 warning_errno(_("unable to unlink: %s"), shared_index_path
);
3078 static int write_shared_index(struct index_state
*istate
,
3079 struct tempfile
**temp
)
3081 struct split_index
*si
= istate
->split_index
;
3084 move_cache_to_base_index(istate
);
3085 ret
= do_write_index(si
->base
, *temp
, 1);
3088 ret
= adjust_shared_perm(get_tempfile_path(*temp
));
3090 error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp
));
3093 ret
= rename_tempfile(temp
,
3094 git_path("sharedindex.%s", oid_to_hex(&si
->base
->oid
)));
3096 oidcpy(&si
->base_oid
, &si
->base
->oid
);
3097 clean_shared_index_files(oid_to_hex(&si
->base
->oid
));
3103 static const int default_max_percent_split_change
= 20;
3105 static int too_many_not_shared_entries(struct index_state
*istate
)
3107 int i
, not_shared
= 0;
3108 int max_split
= git_config_get_max_percent_split_change();
3110 switch (max_split
) {
3112 /* not or badly configured: use the default value */
3113 max_split
= default_max_percent_split_change
;
3116 return 1; /* 0% means always write a new shared index */
3118 return 0; /* 100% means never write a new shared index */
3120 break; /* just use the configured value */
3123 /* Count not shared entries */
3124 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3125 struct cache_entry
*ce
= istate
->cache
[i
];
3130 return (int64_t)istate
->cache_nr
* max_split
< (int64_t)not_shared
* 100;
3133 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3136 int new_shared_index
, ret
;
3137 struct split_index
*si
= istate
->split_index
;
3139 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3140 cache_tree_verify(the_repository
, istate
);
3142 if ((flags
& SKIP_IF_UNCHANGED
) && !istate
->cache_changed
) {
3143 if (flags
& COMMIT_LOCK
)
3144 rollback_lock_file(lock
);
3148 if (istate
->fsmonitor_last_update
)
3149 fill_fsmonitor_bitmap(istate
);
3151 if (!si
|| alternate_index_output
||
3152 (istate
->cache_changed
& ~EXTMASK
)) {
3154 oidclr(&si
->base_oid
);
3155 ret
= do_write_locked_index(istate
, lock
, flags
);
3159 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
3160 int v
= si
->base_oid
.hash
[0];
3162 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3164 if (too_many_not_shared_entries(istate
))
3165 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3167 new_shared_index
= istate
->cache_changed
& SPLIT_INDEX_ORDERED
;
3169 if (new_shared_index
) {
3170 struct tempfile
*temp
;
3173 /* Same initial permissions as the main .git/index file */
3174 temp
= mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);
3176 oidclr(&si
->base_oid
);
3177 ret
= do_write_locked_index(istate
, lock
, flags
);
3180 ret
= write_shared_index(istate
, &temp
);
3182 saved_errno
= errno
;
3183 if (is_tempfile_active(temp
))
3184 delete_tempfile(&temp
);
3185 errno
= saved_errno
;
3191 ret
= write_split_index(istate
, lock
, flags
);
3193 /* Freshen the shared index only if the split-index was written */
3194 if (!ret
&& !new_shared_index
) {
3195 const char *shared_index
= git_path("sharedindex.%s",
3196 oid_to_hex(&si
->base_oid
));
3197 freshen_shared_index(shared_index
, 1);
3201 if (flags
& COMMIT_LOCK
)
3202 rollback_lock_file(lock
);
3207 * Read the index file that is potentially unmerged into given
3208 * index_state, dropping any unmerged entries to stage #0 (potentially
3209 * resulting in a path appearing as both a file and a directory in the
3210 * index; the caller is responsible to clear out the extra entries
3211 * before writing the index to a tree). Returns true if the index is
3212 * unmerged. Callers who want to refuse to work from an unmerged
3213 * state can call this and check its return value, instead of calling
3216 int repo_read_index_unmerged(struct repository
*repo
)
3218 struct index_state
*istate
;
3222 repo_read_index(repo
);
3223 istate
= repo
->index
;
3224 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3225 struct cache_entry
*ce
= istate
->cache
[i
];
3226 struct cache_entry
*new_ce
;
3232 len
= ce_namelen(ce
);
3233 new_ce
= make_empty_cache_entry(istate
, len
);
3234 memcpy(new_ce
->name
, ce
->name
, len
);
3235 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
3236 new_ce
->ce_namelen
= len
;
3237 new_ce
->ce_mode
= ce
->ce_mode
;
3238 if (add_index_entry(istate
, new_ce
, ADD_CACHE_SKIP_DFCHECK
))
3239 return error(_("%s: cannot drop to stage #0"),
3246 * Returns 1 if the path is an "other" path with respect to
3247 * the index; that is, the path is not mentioned in the index at all,
3248 * either as a file, a directory with some files in the index,
3249 * or as an unmerged entry.
3251 * We helpfully remove a trailing "/" from directories so that
3252 * the output of read_directory can be used as-is.
3254 int index_name_is_other(const struct index_state
*istate
, const char *name
,
3258 if (namelen
&& name
[namelen
- 1] == '/')
3260 pos
= index_name_pos(istate
, name
, namelen
);
3262 return 0; /* exact match */
3264 if (pos
< istate
->cache_nr
) {
3265 struct cache_entry
*ce
= istate
->cache
[pos
];
3266 if (ce_namelen(ce
) == namelen
&&
3267 !memcmp(ce
->name
, name
, namelen
))
3268 return 0; /* Yup, this one exists unmerged */
3273 void *read_blob_data_from_index(const struct index_state
*istate
,
3274 const char *path
, unsigned long *size
)
3278 enum object_type type
;
3282 pos
= index_name_pos(istate
, path
, len
);
3285 * We might be in the middle of a merge, in which
3286 * case we would read stage #2 (ours).
3290 (pos
< 0 && i
< istate
->cache_nr
&&
3291 !strcmp(istate
->cache
[i
]->name
, path
));
3293 if (ce_stage(istate
->cache
[i
]) == 2)
3298 data
= read_object_file(&istate
->cache
[pos
]->oid
, &type
, &sz
);
3299 if (!data
|| type
!= OBJ_BLOB
) {
3308 void stat_validity_clear(struct stat_validity
*sv
)
3310 FREE_AND_NULL(sv
->sd
);
3313 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
3317 if (stat(path
, &st
) < 0)
3318 return sv
->sd
== NULL
;
3321 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
3324 void stat_validity_update(struct stat_validity
*sv
, int fd
)
3328 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
3329 stat_validity_clear(sv
);
3332 sv
->sd
= xcalloc(1, sizeof(struct stat_data
));
3333 fill_stat_data(sv
->sd
, &st
);
3337 void move_index_extensions(struct index_state
*dst
, struct index_state
*src
)
3339 dst
->untracked
= src
->untracked
;
3340 src
->untracked
= NULL
;
3341 dst
->cache_tree
= src
->cache_tree
;
3342 src
->cache_tree
= NULL
;
3345 struct cache_entry
*dup_cache_entry(const struct cache_entry
*ce
,
3346 struct index_state
*istate
)
3348 unsigned int size
= ce_size(ce
);
3349 int mem_pool_allocated
;
3350 struct cache_entry
*new_entry
= make_empty_cache_entry(istate
, ce_namelen(ce
));
3351 mem_pool_allocated
= new_entry
->mem_pool_allocated
;
3353 memcpy(new_entry
, ce
, size
);
3354 new_entry
->mem_pool_allocated
= mem_pool_allocated
;
3358 void discard_cache_entry(struct cache_entry
*ce
)
3360 if (ce
&& should_validate_cache_entries())
3361 memset(ce
, 0xCD, cache_entry_size(ce
->ce_namelen
));
3363 if (ce
&& ce
->mem_pool_allocated
)
3369 int should_validate_cache_entries(void)
3371 static int validate_index_cache_entries
= -1;
3373 if (validate_index_cache_entries
< 0) {
3374 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3375 validate_index_cache_entries
= 1;
3377 validate_index_cache_entries
= 0;
3380 return validate_index_cache_entries
;
3383 #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3384 #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3386 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
)
3389 * The end of index entries (EOIE) extension is guaranteed to be last
3390 * so that it can be found by scanning backwards from the EOF.
3397 const char *index
, *eoie
;
3399 size_t offset
, src_offset
;
3400 unsigned char hash
[GIT_MAX_RAWSZ
];
3403 /* ensure we have an index big enough to contain an EOIE extension */
3404 if (mmap_size
< sizeof(struct cache_header
) + EOIE_SIZE_WITH_HEADER
+ the_hash_algo
->rawsz
)
3407 /* validate the extension signature */
3408 index
= eoie
= mmap
+ mmap_size
- EOIE_SIZE_WITH_HEADER
- the_hash_algo
->rawsz
;
3409 if (CACHE_EXT(index
) != CACHE_EXT_ENDOFINDEXENTRIES
)
3411 index
+= sizeof(uint32_t);
3413 /* validate the extension size */
3414 extsize
= get_be32(index
);
3415 if (extsize
!= EOIE_SIZE
)
3417 index
+= sizeof(uint32_t);
3420 * Validate the offset we're going to look for the first extension
3421 * signature is after the index header and before the eoie extension.
3423 offset
= get_be32(index
);
3424 if (mmap
+ offset
< mmap
+ sizeof(struct cache_header
))
3426 if (mmap
+ offset
>= eoie
)
3428 index
+= sizeof(uint32_t);
3431 * The hash is computed over extension types and their sizes (but not
3432 * their contents). E.g. if we have "TREE" extension that is N-bytes
3433 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3434 * then the hash would be:
3436 * SHA-1("TREE" + <binary representation of N> +
3437 * "REUC" + <binary representation of M>)
3439 src_offset
= offset
;
3440 the_hash_algo
->init_fn(&c
);
3441 while (src_offset
< mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
) {
3442 /* After an array of active_nr index entries,
3443 * there can be arbitrary number of extended
3444 * sections, each of which is prefixed with
3445 * extension name (4-byte) and section length
3446 * in 4-byte network byte order.
3449 memcpy(&extsize
, mmap
+ src_offset
+ 4, 4);
3450 extsize
= ntohl(extsize
);
3452 /* verify the extension size isn't so large it will wrap around */
3453 if (src_offset
+ 8 + extsize
< src_offset
)
3456 the_hash_algo
->update_fn(&c
, mmap
+ src_offset
, 8);
3459 src_offset
+= extsize
;
3461 the_hash_algo
->final_fn(hash
, &c
);
3462 if (!hasheq(hash
, (const unsigned char *)index
))
3465 /* Validate that the extension offsets returned us back to the eoie extension. */
3466 if (src_offset
!= mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
)
3472 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
)
3475 unsigned char hash
[GIT_MAX_RAWSZ
];
3478 put_be32(&buffer
, offset
);
3479 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3482 the_hash_algo
->final_fn(hash
, eoie_context
);
3483 strbuf_add(sb
, hash
, the_hash_algo
->rawsz
);
3486 #define IEOT_VERSION (1)
3488 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
)
3490 const char *index
= NULL
;
3491 uint32_t extsize
, ext_version
;
3492 struct index_entry_offset_table
*ieot
;
3495 /* find the IEOT extension */
3498 while (offset
<= mmap_size
- the_hash_algo
->rawsz
- 8) {
3499 extsize
= get_be32(mmap
+ offset
+ 4);
3500 if (CACHE_EXT((mmap
+ offset
)) == CACHE_EXT_INDEXENTRYOFFSETTABLE
) {
3501 index
= mmap
+ offset
+ 4 + 4;
3510 /* validate the version is IEOT_VERSION */
3511 ext_version
= get_be32(index
);
3512 if (ext_version
!= IEOT_VERSION
) {
3513 error("invalid IEOT version %d", ext_version
);
3516 index
+= sizeof(uint32_t);
3518 /* extension size - version bytes / bytes per entry */
3519 nr
= (extsize
- sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3521 error("invalid number of IEOT entries %d", nr
);
3524 ieot
= xmalloc(sizeof(struct index_entry_offset_table
)
3525 + (nr
* sizeof(struct index_entry_offset
)));
3527 for (i
= 0; i
< nr
; i
++) {
3528 ieot
->entries
[i
].offset
= get_be32(index
);
3529 index
+= sizeof(uint32_t);
3530 ieot
->entries
[i
].nr
= get_be32(index
);
3531 index
+= sizeof(uint32_t);
3537 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
)
3543 put_be32(&buffer
, IEOT_VERSION
);
3544 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3547 for (i
= 0; i
< ieot
->nr
; i
++) {
3550 put_be32(&buffer
, ieot
->entries
[i
].offset
);
3551 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3554 put_be32(&buffer
, ieot
->entries
[i
].nr
);
3555 strbuf_add(sb
, &buffer
, sizeof(uint32_t));