Merge branch 'sg/doc-show-branch-typofix'
[git.git] / read-cache.c
blobd57958233e82df6551fdc890f1238ae65beb8940
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
5 */
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
7 #include "cache.h"
8 #include "config.h"
9 #include "diff.h"
10 #include "diffcore.h"
11 #include "tempfile.h"
12 #include "lockfile.h"
13 #include "cache-tree.h"
14 #include "refs.h"
15 #include "dir.h"
16 #include "object-store.h"
17 #include "tree.h"
18 #include "commit.h"
19 #include "blob.h"
20 #include "resolve-undo.h"
21 #include "strbuf.h"
22 #include "varint.h"
23 #include "split-index.h"
24 #include "utf8.h"
25 #include "fsmonitor.h"
26 #include "thread-utils.h"
27 #include "progress.h"
29 /* Mask for the name length in ce_flags in the on-disk index */
31 #define CE_NAMEMASK (0x0fff)
33 /* Index extensions.
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;
72 return ce;
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;
80 return ce;
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;
89 else
90 pool_ptr = &istate->ce_mem_pool;
92 if (!*pool_ptr)
93 mem_pool_init(pool_ptr, 0);
95 return *pool_ptr;
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)
154 int changed = 0;
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;
162 #ifdef USE_NSEC
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;
168 #endif
170 if (check_stat) {
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;
178 #ifdef USE_STDEV
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;
186 #endif
188 if (sd->sd_size != (unsigned int) st->st_size)
189 changed |= DATA_CHANGED;
191 return 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,
214 struct stat *st)
216 int match = -1;
217 int fd = git_open_cloexec(ce->name, O_RDONLY);
219 if (fd >= 0) {
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 */
225 return match;
228 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
230 int match = -1;
231 void *buffer;
232 unsigned long size;
233 enum object_type type;
234 struct strbuf sb = STRBUF_INIT;
236 if (strbuf_readlink(&sb, ce->name, expected_size))
237 return -1;
239 buffer = read_object_file(&ce->oid, &type, &size);
240 if (buffer) {
241 if (size == sb.len)
242 match = memcmp(buffer, sb.buf, size);
243 free(buffer);
245 strbuf_release(&sb);
246 return match;
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
257 * sub-project).
259 * If so, we consider it always to match.
261 if (resolve_gitlink_ref(ce->name, "HEAD", &oid) < 0)
262 return 0;
263 return !oideq(&oid, &ce->oid);
266 static int ce_modified_check_fs(struct index_state *istate,
267 const struct cache_entry *ce,
268 struct stat *st)
270 switch (st->st_mode & S_IFMT) {
271 case S_IFREG:
272 if (ce_compare_data(istate, ce, st))
273 return DATA_CHANGED;
274 break;
275 case S_IFLNK:
276 if (ce_compare_link(ce, xsize_t(st->st_size)))
277 return DATA_CHANGED;
278 break;
279 case S_IFDIR:
280 if (S_ISGITLINK(ce->ce_mode))
281 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
282 /* else fallthrough */
283 default:
284 return TYPE_CHANGED;
286 return 0;
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) {
297 case S_IFREG:
298 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
299 /* We consider only the owner x bit to be relevant for
300 * "mode changes"
302 if (trust_executable_bit &&
303 (0100 & (ce->ce_mode ^ st->st_mode)))
304 changed |= MODE_CHANGED;
305 break;
306 case S_IFLNK:
307 if (!S_ISLNK(st->st_mode) &&
308 (has_symlinks || !S_ISREG(st->st_mode)))
309 changed |= TYPE_CHANGED;
310 break;
311 case S_IFGITLINK:
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;
317 return changed;
318 default:
319 die("internal error: ce_mode is %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;
330 return changed;
333 static int is_racy_stat(const struct index_state *istate,
334 const struct stat_data *sd)
336 return (istate->timestamp.sec &&
337 #ifdef USE_NSEC
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))
342 #else
343 istate->timestamp.sec <= sd->sd_mtime.sec
344 #endif
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))
382 return 0;
383 if (!ignore_valid && (ce->ce_flags & CE_VALID))
384 return 0;
385 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID))
386 return 0;
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:
402 * echo frotz >file
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;
417 else
418 changed |= ce_modified_check_fs(istate, ce, st);
421 return changed;
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);
431 if (!changed)
432 return 0;
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))
438 return 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))
456 return changed;
458 changed_fs = ce_modified_check_fs(istate, ce, st);
459 if (changed_fs)
460 return changed | changed_fs;
461 return 0;
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;
469 int cmp;
471 cmp = memcmp(name1, name2, len);
472 if (cmp)
473 return cmp;
474 c1 = name1[len];
475 c2 = name2[len];
476 if (!c1 && S_ISDIR(mode1))
477 c1 = '/';
478 if (!c2 && S_ISDIR(mode2))
479 c2 = '/';
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);
500 if (cmp)
501 return cmp;
502 /* Directories and files compare equal (same length, same name) */
503 if (len1 == len2)
504 return 0;
505 c1 = name1[len];
506 if (!c1 && S_ISDIR(mode1))
507 c1 = '/';
508 c2 = name2[len];
509 if (!c2 && S_ISDIR(mode2))
510 c2 = '/';
511 if (c1 == '/' && !c2)
512 return 0;
513 if (c2 == '/' && !c1)
514 return 0;
515 return c1 - c2;
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);
522 if (cmp)
523 return cmp;
524 if (len1 < len2)
525 return -1;
526 if (len1 > len2)
527 return 1;
528 return 0;
531 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
533 int cmp;
535 cmp = name_compare(name1, len1, name2, len2);
536 if (cmp)
537 return cmp;
539 if (stage1 < stage2)
540 return -1;
541 if (stage1 > stage2)
542 return 1;
543 return 0;
546 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
548 int first, last;
550 first = 0;
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));
556 if (!cmp)
557 return next;
558 if (cmp < 0) {
559 last = next;
560 continue;
562 first = next+1;
564 return -first-1;
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;
580 istate->cache_nr--;
581 if (pos >= istate->cache_nr)
582 return 0;
583 MOVE_ARRAY(istate->cache + pos, istate->cache + pos + 1,
584 istate->cache_nr - pos);
585 return 1;
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;
596 unsigned int i, j;
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]);
603 else
604 ce_array[j++] = ce_array[i];
606 if (j == istate->cache_nr)
607 return;
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));
615 if (pos < 0)
616 pos = -pos-1;
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);
621 return 0;
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;
635 if (pos >= 0)
636 return pos;
638 /* maybe unmerged? */
639 pos = -1 - pos;
640 if (pos >= istate->cache_nr ||
641 compare_name((ce = istate->cache[pos]), path, namelen))
642 return -1;
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))
648 pos++;
649 return pos;
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)
671 int len;
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)", ce->name, alias->name);
677 /* Ok, create the new entry using the name of the existing alias */
678 len = ce_namelen(alias);
679 new_entry = make_empty_cache_entry(istate, len);
680 memcpy(new_entry->name, alias->name, len);
681 copy_cache_entry(new_entry, ce);
682 save_or_free_index_entry(istate, ce);
683 return new_entry;
686 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
688 struct object_id oid;
689 if (write_object_file("", 0, blob_type, &oid))
690 die("cannot create an empty blob in the object database");
691 oidcpy(&ce->oid, &oid);
694 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
696 int namelen, was_same;
697 mode_t st_mode = st->st_mode;
698 struct cache_entry *ce, *alias = NULL;
699 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
700 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
701 int pretend = flags & ADD_CACHE_PRETEND;
702 int intent_only = flags & ADD_CACHE_INTENT;
703 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
704 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
705 int newflags = HASH_WRITE_OBJECT;
707 if (flags & HASH_RENORMALIZE)
708 newflags |= HASH_RENORMALIZE;
710 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
711 return error("%s: can only add regular files, symbolic links or git-directories", path);
713 namelen = strlen(path);
714 if (S_ISDIR(st_mode)) {
715 while (namelen && path[namelen-1] == '/')
716 namelen--;
718 ce = make_empty_cache_entry(istate, namelen);
719 memcpy(ce->name, path, namelen);
720 ce->ce_namelen = namelen;
721 if (!intent_only)
722 fill_stat_cache_info(ce, st);
723 else
724 ce->ce_flags |= CE_INTENT_TO_ADD;
727 if (trust_executable_bit && has_symlinks) {
728 ce->ce_mode = create_ce_mode(st_mode);
729 } else {
730 /* If there is an existing entry, pick the mode bits and type
731 * from it, otherwise assume unexecutable regular file.
733 struct cache_entry *ent;
734 int pos = index_name_pos_also_unmerged(istate, path, namelen);
736 ent = (0 <= pos) ? istate->cache[pos] : NULL;
737 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
740 /* When core.ignorecase=true, determine if a directory of the same name but differing
741 * case already exists within the Git repository. If it does, ensure the directory
742 * case of the file being added to the repository matches (is folded into) the existing
743 * entry's directory case.
745 if (ignore_case) {
746 adjust_dirname_case(istate, ce->name);
748 if (!(flags & HASH_RENORMALIZE)) {
749 alias = index_file_exists(istate, ce->name,
750 ce_namelen(ce), ignore_case);
751 if (alias &&
752 !ce_stage(alias) &&
753 !ie_match_stat(istate, alias, st, ce_option)) {
754 /* Nothing changed, really */
755 if (!S_ISGITLINK(alias->ce_mode))
756 ce_mark_uptodate(alias);
757 alias->ce_flags |= CE_ADDED;
759 discard_cache_entry(ce);
760 return 0;
763 if (!intent_only) {
764 if (index_path(istate, &ce->oid, path, st, newflags)) {
765 discard_cache_entry(ce);
766 return error("unable to index file %s", path);
768 } else
769 set_object_name_for_intent_to_add_entry(ce);
771 if (ignore_case && alias && different_name(ce, alias))
772 ce = create_alias_ce(istate, ce, alias);
773 ce->ce_flags |= CE_ADDED;
775 /* It was suspected to be racily clean, but it turns out to be Ok */
776 was_same = (alias &&
777 !ce_stage(alias) &&
778 oideq(&alias->oid, &ce->oid) &&
779 ce->ce_mode == alias->ce_mode);
781 if (pretend)
782 discard_cache_entry(ce);
783 else if (add_index_entry(istate, ce, add_option)) {
784 discard_cache_entry(ce);
785 return error("unable to add %s to index", path);
787 if (verbose && !was_same)
788 printf("add '%s'\n", path);
789 return 0;
792 int add_file_to_index(struct index_state *istate, const char *path, int flags)
794 struct stat st;
795 if (lstat(path, &st))
796 die_errno("unable to stat '%s'", path);
797 return add_to_index(istate, path, &st, flags);
800 struct cache_entry *make_empty_cache_entry(struct index_state *istate, size_t len)
802 return mem_pool__ce_calloc(find_mem_pool(istate), len);
805 struct cache_entry *make_empty_transient_cache_entry(size_t len)
807 return xcalloc(1, cache_entry_size(len));
810 struct cache_entry *make_cache_entry(struct index_state *istate,
811 unsigned int mode,
812 const struct object_id *oid,
813 const char *path,
814 int stage,
815 unsigned int refresh_options)
817 struct cache_entry *ce, *ret;
818 int len;
820 if (!verify_path(path, mode)) {
821 error("Invalid path '%s'", path);
822 return NULL;
825 len = strlen(path);
826 ce = make_empty_cache_entry(istate, len);
828 oidcpy(&ce->oid, oid);
829 memcpy(ce->name, path, len);
830 ce->ce_flags = create_ce_flags(stage);
831 ce->ce_namelen = len;
832 ce->ce_mode = create_ce_mode(mode);
834 ret = refresh_cache_entry(istate, ce, refresh_options);
835 if (ret != ce)
836 discard_cache_entry(ce);
837 return ret;
840 struct cache_entry *make_transient_cache_entry(unsigned int mode, const struct object_id *oid,
841 const char *path, int stage)
843 struct cache_entry *ce;
844 int len;
846 if (!verify_path(path, mode)) {
847 error("Invalid path '%s'", path);
848 return NULL;
851 len = strlen(path);
852 ce = make_empty_transient_cache_entry(len);
854 oidcpy(&ce->oid, oid);
855 memcpy(ce->name, path, len);
856 ce->ce_flags = create_ce_flags(stage);
857 ce->ce_namelen = len;
858 ce->ce_mode = create_ce_mode(mode);
860 return ce;
864 * Chmod an index entry with either +x or -x.
866 * Returns -1 if the chmod for the particular cache entry failed (if it's
867 * not a regular file), -2 if an invalid flip argument is passed in, 0
868 * otherwise.
870 int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
871 char flip)
873 if (!S_ISREG(ce->ce_mode))
874 return -1;
875 switch (flip) {
876 case '+':
877 ce->ce_mode |= 0111;
878 break;
879 case '-':
880 ce->ce_mode &= ~0111;
881 break;
882 default:
883 return -2;
885 cache_tree_invalidate_path(istate, ce->name);
886 ce->ce_flags |= CE_UPDATE_IN_BASE;
887 mark_fsmonitor_invalid(istate, ce);
888 istate->cache_changed |= CE_ENTRY_CHANGED;
890 return 0;
893 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
895 int len = ce_namelen(a);
896 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
900 * We fundamentally don't like some paths: we don't want
901 * dot or dot-dot anywhere, and for obvious reasons don't
902 * want to recurse into ".git" either.
904 * Also, we don't want double slashes or slashes at the
905 * end that can make pathnames ambiguous.
907 static int verify_dotfile(const char *rest, unsigned mode)
910 * The first character was '.', but that
911 * has already been discarded, we now test
912 * the rest.
915 /* "." is not allowed */
916 if (*rest == '\0' || is_dir_sep(*rest))
917 return 0;
919 switch (*rest) {
921 * ".git" followed by NUL or slash is bad. Note that we match
922 * case-insensitively here, even if ignore_case is not set.
923 * This outlaws ".GIT" everywhere out of an abundance of caution,
924 * since there's really no good reason to allow it.
926 * Once we've seen ".git", we can also find ".gitmodules", etc (also
927 * case-insensitively).
929 case 'g':
930 case 'G':
931 if (rest[1] != 'i' && rest[1] != 'I')
932 break;
933 if (rest[2] != 't' && rest[2] != 'T')
934 break;
935 if (rest[3] == '\0' || is_dir_sep(rest[3]))
936 return 0;
937 if (S_ISLNK(mode)) {
938 rest += 3;
939 if (skip_iprefix(rest, "modules", &rest) &&
940 (*rest == '\0' || is_dir_sep(*rest)))
941 return 0;
943 break;
944 case '.':
945 if (rest[1] == '\0' || is_dir_sep(rest[1]))
946 return 0;
948 return 1;
951 int verify_path(const char *path, unsigned mode)
953 char c;
955 if (has_dos_drive_prefix(path))
956 return 0;
958 goto inside;
959 for (;;) {
960 if (!c)
961 return 1;
962 if (is_dir_sep(c)) {
963 inside:
964 if (protect_hfs) {
965 if (is_hfs_dotgit(path))
966 return 0;
967 if (S_ISLNK(mode)) {
968 if (is_hfs_dotgitmodules(path))
969 return 0;
972 if (protect_ntfs) {
973 if (is_ntfs_dotgit(path))
974 return 0;
975 if (S_ISLNK(mode)) {
976 if (is_ntfs_dotgitmodules(path))
977 return 0;
981 c = *path++;
982 if ((c == '.' && !verify_dotfile(path, mode)) ||
983 is_dir_sep(c) || c == '\0')
984 return 0;
986 c = *path++;
991 * Do we have another file that has the beginning components being a
992 * proper superset of the name we're trying to add?
994 static int has_file_name(struct index_state *istate,
995 const struct cache_entry *ce, int pos, int ok_to_replace)
997 int retval = 0;
998 int len = ce_namelen(ce);
999 int stage = ce_stage(ce);
1000 const char *name = ce->name;
1002 while (pos < istate->cache_nr) {
1003 struct cache_entry *p = istate->cache[pos++];
1005 if (len >= ce_namelen(p))
1006 break;
1007 if (memcmp(name, p->name, len))
1008 break;
1009 if (ce_stage(p) != stage)
1010 continue;
1011 if (p->name[len] != '/')
1012 continue;
1013 if (p->ce_flags & CE_REMOVE)
1014 continue;
1015 retval = -1;
1016 if (!ok_to_replace)
1017 break;
1018 remove_index_entry_at(istate, --pos);
1020 return retval;
1025 * Like strcmp(), but also return the offset of the first change.
1026 * If strings are equal, return the length.
1028 int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
1030 size_t k;
1032 if (!first_change)
1033 return strcmp(s1, s2);
1035 for (k = 0; s1[k] == s2[k]; k++)
1036 if (s1[k] == '\0')
1037 break;
1039 *first_change = k;
1040 return (unsigned char)s1[k] - (unsigned char)s2[k];
1044 * Do we have another file with a pathname that is a proper
1045 * subset of the name we're trying to add?
1047 * That is, is there another file in the index with a path
1048 * that matches a sub-directory in the given entry?
1050 static int has_dir_name(struct index_state *istate,
1051 const struct cache_entry *ce, int pos, int ok_to_replace)
1053 int retval = 0;
1054 int stage = ce_stage(ce);
1055 const char *name = ce->name;
1056 const char *slash = name + ce_namelen(ce);
1057 size_t len_eq_last;
1058 int cmp_last = 0;
1061 * We are frequently called during an iteration on a sorted
1062 * list of pathnames and while building a new index. Therefore,
1063 * there is a high probability that this entry will eventually
1064 * be appended to the index, rather than inserted in the middle.
1065 * If we can confirm that, we can avoid binary searches on the
1066 * components of the pathname.
1068 * Compare the entry's full path with the last path in the index.
1070 if (istate->cache_nr > 0) {
1071 cmp_last = strcmp_offset(name,
1072 istate->cache[istate->cache_nr - 1]->name,
1073 &len_eq_last);
1074 if (cmp_last > 0) {
1075 if (len_eq_last == 0) {
1077 * The entry sorts AFTER the last one in the
1078 * index and their paths have no common prefix,
1079 * so there cannot be a F/D conflict.
1081 return retval;
1082 } else {
1084 * The entry sorts AFTER the last one in the
1085 * index, but has a common prefix. Fall through
1086 * to the loop below to disect the entry's path
1087 * and see where the difference is.
1090 } else if (cmp_last == 0) {
1092 * The entry exactly matches the last one in the
1093 * index, but because of multiple stage and CE_REMOVE
1094 * items, we fall through and let the regular search
1095 * code handle it.
1100 for (;;) {
1101 size_t len;
1103 for (;;) {
1104 if (*--slash == '/')
1105 break;
1106 if (slash <= ce->name)
1107 return retval;
1109 len = slash - name;
1111 if (cmp_last > 0) {
1113 * (len + 1) is a directory boundary (including
1114 * the trailing slash). And since the loop is
1115 * decrementing "slash", the first iteration is
1116 * the longest directory prefix; subsequent
1117 * iterations consider parent directories.
1120 if (len + 1 <= len_eq_last) {
1122 * The directory prefix (including the trailing
1123 * slash) also appears as a prefix in the last
1124 * entry, so the remainder cannot collide (because
1125 * strcmp said the whole path was greater).
1127 * EQ: last: xxx/A
1128 * this: xxx/B
1130 * LT: last: xxx/file_A
1131 * this: xxx/file_B
1133 return retval;
1136 if (len > len_eq_last) {
1138 * This part of the directory prefix (excluding
1139 * the trailing slash) is longer than the known
1140 * equal portions, so this sub-directory cannot
1141 * collide with a file.
1143 * GT: last: xxxA
1144 * this: xxxB/file
1146 return retval;
1149 if (istate->cache_nr > 0 &&
1150 ce_namelen(istate->cache[istate->cache_nr - 1]) > len) {
1152 * The directory prefix lines up with part of
1153 * a longer file or directory name, but sorts
1154 * after it, so this sub-directory cannot
1155 * collide with a file.
1157 * last: xxx/yy-file (because '-' sorts before '/')
1158 * this: xxx/yy/abc
1160 return retval;
1164 * This is a possible collision. Fall through and
1165 * let the regular search code handle it.
1167 * last: xxx
1168 * this: xxx/file
1172 pos = index_name_stage_pos(istate, name, len, stage);
1173 if (pos >= 0) {
1175 * Found one, but not so fast. This could
1176 * be a marker that says "I was here, but
1177 * I am being removed". Such an entry is
1178 * not a part of the resulting tree, and
1179 * it is Ok to have a directory at the same
1180 * path.
1182 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
1183 retval = -1;
1184 if (!ok_to_replace)
1185 break;
1186 remove_index_entry_at(istate, pos);
1187 continue;
1190 else
1191 pos = -pos-1;
1194 * Trivial optimization: if we find an entry that
1195 * already matches the sub-directory, then we know
1196 * we're ok, and we can exit.
1198 while (pos < istate->cache_nr) {
1199 struct cache_entry *p = istate->cache[pos];
1200 if ((ce_namelen(p) <= len) ||
1201 (p->name[len] != '/') ||
1202 memcmp(p->name, name, len))
1203 break; /* not our subdirectory */
1204 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1206 * p is at the same stage as our entry, and
1207 * is a subdirectory of what we are looking
1208 * at, so we cannot have conflicts at our
1209 * level or anything shorter.
1211 return retval;
1212 pos++;
1215 return retval;
1218 /* We may be in a situation where we already have path/file and path
1219 * is being added, or we already have path and path/file is being
1220 * added. Either one would result in a nonsense tree that has path
1221 * twice when git-write-tree tries to write it out. Prevent it.
1223 * If ok-to-replace is specified, we remove the conflicting entries
1224 * from the cache so the caller should recompute the insert position.
1225 * When this happens, we return non-zero.
1227 static int check_file_directory_conflict(struct index_state *istate,
1228 const struct cache_entry *ce,
1229 int pos, int ok_to_replace)
1231 int retval;
1234 * When ce is an "I am going away" entry, we allow it to be added
1236 if (ce->ce_flags & CE_REMOVE)
1237 return 0;
1240 * We check if the path is a sub-path of a subsequent pathname
1241 * first, since removing those will not change the position
1242 * in the array.
1244 retval = has_file_name(istate, ce, pos, ok_to_replace);
1247 * Then check if the path might have a clashing sub-directory
1248 * before it.
1250 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
1253 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
1255 int pos;
1256 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1257 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1258 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1259 int new_only = option & ADD_CACHE_NEW_ONLY;
1261 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1262 cache_tree_invalidate_path(istate, ce->name);
1265 * If this entry's path sorts after the last entry in the index,
1266 * we can avoid searching for it.
1268 if (istate->cache_nr > 0 &&
1269 strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1270 pos = -istate->cache_nr - 1;
1271 else
1272 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1274 /* existing match? Just replace it. */
1275 if (pos >= 0) {
1276 if (!new_only)
1277 replace_index_entry(istate, pos, ce);
1278 return 0;
1280 pos = -pos-1;
1282 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1283 untracked_cache_add_to_index(istate, ce->name);
1286 * Inserting a merged entry ("stage 0") into the index
1287 * will always replace all non-merged entries..
1289 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1290 while (ce_same_name(istate->cache[pos], ce)) {
1291 ok_to_add = 1;
1292 if (!remove_index_entry_at(istate, pos))
1293 break;
1297 if (!ok_to_add)
1298 return -1;
1299 if (!verify_path(ce->name, ce->ce_mode))
1300 return error("Invalid path '%s'", ce->name);
1302 if (!skip_df_check &&
1303 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1304 if (!ok_to_replace)
1305 return error("'%s' appears as both a file and as a directory",
1306 ce->name);
1307 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1308 pos = -pos-1;
1310 return pos + 1;
1313 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1315 int pos;
1317 if (option & ADD_CACHE_JUST_APPEND)
1318 pos = istate->cache_nr;
1319 else {
1320 int ret;
1321 ret = add_index_entry_with_check(istate, ce, option);
1322 if (ret <= 0)
1323 return ret;
1324 pos = ret - 1;
1327 /* Make sure the array is big enough .. */
1328 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1330 /* Add it in.. */
1331 istate->cache_nr++;
1332 if (istate->cache_nr > pos + 1)
1333 MOVE_ARRAY(istate->cache + pos + 1, istate->cache + pos,
1334 istate->cache_nr - pos - 1);
1335 set_index_entry(istate, pos, ce);
1336 istate->cache_changed |= CE_ENTRY_ADDED;
1337 return 0;
1341 * "refresh" does not calculate a new sha1 file or bring the
1342 * cache up-to-date for mode/content changes. But what it
1343 * _does_ do is to "re-match" the stat information of a file
1344 * with the cache, so that you can refresh the cache for a
1345 * file that hasn't been changed but where the stat entry is
1346 * out of date.
1348 * For example, you'd want to do this after doing a "git-read-tree",
1349 * to link up the stat cache details with the proper files.
1351 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1352 struct cache_entry *ce,
1353 unsigned int options, int *err,
1354 int *changed_ret)
1356 struct stat st;
1357 struct cache_entry *updated;
1358 int changed;
1359 int refresh = options & CE_MATCH_REFRESH;
1360 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1361 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1362 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1363 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
1365 if (!refresh || ce_uptodate(ce))
1366 return ce;
1368 if (!ignore_fsmonitor)
1369 refresh_fsmonitor(istate);
1371 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1372 * that the change to the work tree does not matter and told
1373 * us not to worry.
1375 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1376 ce_mark_uptodate(ce);
1377 return ce;
1379 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1380 ce_mark_uptodate(ce);
1381 return ce;
1383 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {
1384 ce_mark_uptodate(ce);
1385 return ce;
1388 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1389 if (ignore_missing)
1390 return ce;
1391 if (err)
1392 *err = ENOENT;
1393 return NULL;
1396 if (lstat(ce->name, &st) < 0) {
1397 if (ignore_missing && errno == ENOENT)
1398 return ce;
1399 if (err)
1400 *err = errno;
1401 return NULL;
1404 changed = ie_match_stat(istate, ce, &st, options);
1405 if (changed_ret)
1406 *changed_ret = changed;
1407 if (!changed) {
1409 * The path is unchanged. If we were told to ignore
1410 * valid bit, then we did the actual stat check and
1411 * found that the entry is unmodified. If the entry
1412 * is not marked VALID, this is the place to mark it
1413 * valid again, under "assume unchanged" mode.
1415 if (ignore_valid && assume_unchanged &&
1416 !(ce->ce_flags & CE_VALID))
1417 ; /* mark this one VALID again */
1418 else {
1420 * We do not mark the index itself "modified"
1421 * because CE_UPTODATE flag is in-core only;
1422 * we are not going to write this change out.
1424 if (!S_ISGITLINK(ce->ce_mode)) {
1425 ce_mark_uptodate(ce);
1426 mark_fsmonitor_valid(ce);
1428 return ce;
1432 if (ie_modified(istate, ce, &st, options)) {
1433 if (err)
1434 *err = EINVAL;
1435 return NULL;
1438 updated = make_empty_cache_entry(istate, ce_namelen(ce));
1439 copy_cache_entry(updated, ce);
1440 memcpy(updated->name, ce->name, ce->ce_namelen + 1);
1441 fill_stat_cache_info(updated, &st);
1443 * If ignore_valid is not set, we should leave CE_VALID bit
1444 * alone. Otherwise, paths marked with --no-assume-unchanged
1445 * (i.e. things to be edited) will reacquire CE_VALID bit
1446 * automatically, which is not really what we want.
1448 if (!ignore_valid && assume_unchanged &&
1449 !(ce->ce_flags & CE_VALID))
1450 updated->ce_flags &= ~CE_VALID;
1452 /* istate->cache_changed is updated in the caller */
1453 return updated;
1456 static void show_file(const char * fmt, const char * name, int in_porcelain,
1457 int * first, const char *header_msg)
1459 if (in_porcelain && *first && header_msg) {
1460 printf("%s\n", header_msg);
1461 *first = 0;
1463 printf(fmt, name);
1466 int refresh_index(struct index_state *istate, unsigned int flags,
1467 const struct pathspec *pathspec,
1468 char *seen, const char *header_msg)
1470 int i;
1471 int has_errors = 0;
1472 int really = (flags & REFRESH_REALLY) != 0;
1473 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1474 int quiet = (flags & REFRESH_QUIET) != 0;
1475 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1476 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1477 int first = 1;
1478 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1479 unsigned int options = (CE_MATCH_REFRESH |
1480 (really ? CE_MATCH_IGNORE_VALID : 0) |
1481 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1482 const char *modified_fmt;
1483 const char *deleted_fmt;
1484 const char *typechange_fmt;
1485 const char *added_fmt;
1486 const char *unmerged_fmt;
1487 struct progress *progress = NULL;
1489 if (flags & REFRESH_PROGRESS && isatty(2))
1490 progress = start_delayed_progress(_("Refresh index"),
1491 istate->cache_nr);
1493 trace_performance_enter();
1494 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1495 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1496 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1497 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1498 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1499 for (i = 0; i < istate->cache_nr; i++) {
1500 struct cache_entry *ce, *new_entry;
1501 int cache_errno = 0;
1502 int changed = 0;
1503 int filtered = 0;
1505 ce = istate->cache[i];
1506 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1507 continue;
1509 if (pathspec && !ce_path_match(istate, ce, pathspec, seen))
1510 filtered = 1;
1512 if (ce_stage(ce)) {
1513 while ((i < istate->cache_nr) &&
1514 ! strcmp(istate->cache[i]->name, ce->name))
1515 i++;
1516 i--;
1517 if (allow_unmerged)
1518 continue;
1519 if (!filtered)
1520 show_file(unmerged_fmt, ce->name, in_porcelain,
1521 &first, header_msg);
1522 has_errors = 1;
1523 continue;
1526 if (filtered)
1527 continue;
1529 new_entry = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1530 if (new_entry == ce)
1531 continue;
1532 if (progress)
1533 display_progress(progress, i);
1534 if (!new_entry) {
1535 const char *fmt;
1537 if (really && cache_errno == EINVAL) {
1538 /* If we are doing --really-refresh that
1539 * means the index is not valid anymore.
1541 ce->ce_flags &= ~CE_VALID;
1542 ce->ce_flags |= CE_UPDATE_IN_BASE;
1543 mark_fsmonitor_invalid(istate, ce);
1544 istate->cache_changed |= CE_ENTRY_CHANGED;
1546 if (quiet)
1547 continue;
1549 if (cache_errno == ENOENT)
1550 fmt = deleted_fmt;
1551 else if (ce_intent_to_add(ce))
1552 fmt = added_fmt; /* must be before other checks */
1553 else if (changed & TYPE_CHANGED)
1554 fmt = typechange_fmt;
1555 else
1556 fmt = modified_fmt;
1557 show_file(fmt,
1558 ce->name, in_porcelain, &first, header_msg);
1559 has_errors = 1;
1560 continue;
1563 replace_index_entry(istate, i, new_entry);
1565 if (progress) {
1566 display_progress(progress, istate->cache_nr);
1567 stop_progress(&progress);
1569 trace_performance_leave("refresh index");
1570 return has_errors;
1573 struct cache_entry *refresh_cache_entry(struct index_state *istate,
1574 struct cache_entry *ce,
1575 unsigned int options)
1577 return refresh_cache_ent(istate, ce, options, NULL, NULL);
1581 /*****************************************************************
1582 * Index File I/O
1583 *****************************************************************/
1585 #define INDEX_FORMAT_DEFAULT 3
1587 static unsigned int get_index_format_default(void)
1589 char *envversion = getenv("GIT_INDEX_VERSION");
1590 char *endp;
1591 int value;
1592 unsigned int version = INDEX_FORMAT_DEFAULT;
1594 if (!envversion) {
1595 if (!git_config_get_int("index.version", &value))
1596 version = value;
1597 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1598 warning(_("index.version set, but the value is invalid.\n"
1599 "Using version %i"), INDEX_FORMAT_DEFAULT);
1600 return INDEX_FORMAT_DEFAULT;
1602 return version;
1605 version = strtoul(envversion, &endp, 10);
1606 if (*endp ||
1607 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1608 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1609 "Using version %i"), INDEX_FORMAT_DEFAULT);
1610 version = INDEX_FORMAT_DEFAULT;
1612 return version;
1616 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1617 * Again - this is just a (very strong in practice) heuristic that
1618 * the inode hasn't changed.
1620 * We save the fields in big-endian order to allow using the
1621 * index file over NFS transparently.
1623 struct ondisk_cache_entry {
1624 struct cache_time ctime;
1625 struct cache_time mtime;
1626 uint32_t dev;
1627 uint32_t ino;
1628 uint32_t mode;
1629 uint32_t uid;
1630 uint32_t gid;
1631 uint32_t size;
1632 unsigned char sha1[20];
1633 uint16_t flags;
1634 char name[FLEX_ARRAY]; /* more */
1638 * This struct is used when CE_EXTENDED bit is 1
1639 * The struct must match ondisk_cache_entry exactly from
1640 * ctime till flags
1642 struct ondisk_cache_entry_extended {
1643 struct cache_time ctime;
1644 struct cache_time mtime;
1645 uint32_t dev;
1646 uint32_t ino;
1647 uint32_t mode;
1648 uint32_t uid;
1649 uint32_t gid;
1650 uint32_t size;
1651 unsigned char sha1[20];
1652 uint16_t flags;
1653 uint16_t flags2;
1654 char name[FLEX_ARRAY]; /* more */
1657 /* These are only used for v3 or lower */
1658 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1659 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1660 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1661 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1662 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1663 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1664 ondisk_cache_entry_size(ce_namelen(ce)))
1666 /* Allow fsck to force verification of the index checksum. */
1667 int verify_index_checksum;
1669 /* Allow fsck to force verification of the cache entry order. */
1670 int verify_ce_order;
1672 static int verify_hdr(const struct cache_header *hdr, unsigned long size)
1674 git_hash_ctx c;
1675 unsigned char hash[GIT_MAX_RAWSZ];
1676 int hdr_version;
1678 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1679 return error("bad signature");
1680 hdr_version = ntohl(hdr->hdr_version);
1681 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1682 return error("bad index version %d", hdr_version);
1684 if (!verify_index_checksum)
1685 return 0;
1687 the_hash_algo->init_fn(&c);
1688 the_hash_algo->update_fn(&c, hdr, size - the_hash_algo->rawsz);
1689 the_hash_algo->final_fn(hash, &c);
1690 if (!hasheq(hash, (unsigned char *)hdr + size - the_hash_algo->rawsz))
1691 return error("bad index file sha1 signature");
1692 return 0;
1695 static int read_index_extension(struct index_state *istate,
1696 const char *ext, const char *data, unsigned long sz)
1698 switch (CACHE_EXT(ext)) {
1699 case CACHE_EXT_TREE:
1700 istate->cache_tree = cache_tree_read(data, sz);
1701 break;
1702 case CACHE_EXT_RESOLVE_UNDO:
1703 istate->resolve_undo = resolve_undo_read(data, sz);
1704 break;
1705 case CACHE_EXT_LINK:
1706 if (read_link_extension(istate, data, sz))
1707 return -1;
1708 break;
1709 case CACHE_EXT_UNTRACKED:
1710 istate->untracked = read_untracked_extension(data, sz);
1711 break;
1712 case CACHE_EXT_FSMONITOR:
1713 read_fsmonitor_extension(istate, data, sz);
1714 break;
1715 case CACHE_EXT_ENDOFINDEXENTRIES:
1716 case CACHE_EXT_INDEXENTRYOFFSETTABLE:
1717 /* already handled in do_read_index() */
1718 break;
1719 default:
1720 if (*ext < 'A' || 'Z' < *ext)
1721 return error("index uses %.4s extension, which we do not understand",
1722 ext);
1723 fprintf(stderr, "ignoring %.4s extension\n", ext);
1724 break;
1726 return 0;
1729 int hold_locked_index(struct lock_file *lk, int lock_flags)
1731 return hold_lock_file_for_update(lk, get_index_file(), lock_flags);
1734 int read_index(struct index_state *istate)
1736 return read_index_from(istate, get_index_file(), get_git_dir());
1739 static struct cache_entry *create_from_disk(struct mem_pool *ce_mem_pool,
1740 unsigned int version,
1741 struct ondisk_cache_entry *ondisk,
1742 unsigned long *ent_size,
1743 const struct cache_entry *previous_ce)
1745 struct cache_entry *ce;
1746 size_t len;
1747 const char *name;
1748 unsigned int flags;
1749 size_t copy_len;
1751 * Adjacent cache entries tend to share the leading paths, so it makes
1752 * sense to only store the differences in later entries. In the v4
1753 * on-disk format of the index, each on-disk cache entry stores the
1754 * number of bytes to be stripped from the end of the previous name,
1755 * and the bytes to append to the result, to come up with its name.
1757 int expand_name_field = version == 4;
1759 /* On-disk flags are just 16 bits */
1760 flags = get_be16(&ondisk->flags);
1761 len = flags & CE_NAMEMASK;
1763 if (flags & CE_EXTENDED) {
1764 struct ondisk_cache_entry_extended *ondisk2;
1765 int extended_flags;
1766 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1767 extended_flags = get_be16(&ondisk2->flags2) << 16;
1768 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1769 if (extended_flags & ~CE_EXTENDED_FLAGS)
1770 die("Unknown index entry format %08x", extended_flags);
1771 flags |= extended_flags;
1772 name = ondisk2->name;
1774 else
1775 name = ondisk->name;
1777 if (expand_name_field) {
1778 const unsigned char *cp = (const unsigned char *)name;
1779 size_t strip_len, previous_len;
1781 /* If we're at the begining of a block, ignore the previous name */
1782 strip_len = decode_varint(&cp);
1783 if (previous_ce) {
1784 previous_len = previous_ce->ce_namelen;
1785 if (previous_len < strip_len)
1786 die(_("malformed name field in the index, near path '%s'"),
1787 previous_ce->name);
1788 copy_len = previous_len - strip_len;
1789 } else {
1790 copy_len = 0;
1792 name = (const char *)cp;
1795 if (len == CE_NAMEMASK) {
1796 len = strlen(name);
1797 if (expand_name_field)
1798 len += copy_len;
1801 ce = mem_pool__ce_alloc(ce_mem_pool, len);
1803 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1804 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1805 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1806 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1807 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1808 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1809 ce->ce_mode = get_be32(&ondisk->mode);
1810 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1811 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1812 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1813 ce->ce_flags = flags & ~CE_NAMEMASK;
1814 ce->ce_namelen = len;
1815 ce->index = 0;
1816 hashcpy(ce->oid.hash, ondisk->sha1);
1818 if (expand_name_field) {
1819 if (copy_len)
1820 memcpy(ce->name, previous_ce->name, copy_len);
1821 memcpy(ce->name + copy_len, name, len + 1 - copy_len);
1822 *ent_size = (name - ((char *)ondisk)) + len + 1 - copy_len;
1823 } else {
1824 memcpy(ce->name, name, len + 1);
1825 *ent_size = ondisk_ce_size(ce);
1827 return ce;
1830 static void check_ce_order(struct index_state *istate)
1832 unsigned int i;
1834 if (!verify_ce_order)
1835 return;
1837 for (i = 1; i < istate->cache_nr; i++) {
1838 struct cache_entry *ce = istate->cache[i - 1];
1839 struct cache_entry *next_ce = istate->cache[i];
1840 int name_compare = strcmp(ce->name, next_ce->name);
1842 if (0 < name_compare)
1843 die("unordered stage entries in index");
1844 if (!name_compare) {
1845 if (!ce_stage(ce))
1846 die("multiple stage entries for merged file '%s'",
1847 ce->name);
1848 if (ce_stage(ce) > ce_stage(next_ce))
1849 die("unordered stage entries for '%s'",
1850 ce->name);
1855 static void tweak_untracked_cache(struct index_state *istate)
1857 switch (git_config_get_untracked_cache()) {
1858 case -1: /* keep: do nothing */
1859 break;
1860 case 0: /* false */
1861 remove_untracked_cache(istate);
1862 break;
1863 case 1: /* true */
1864 add_untracked_cache(istate);
1865 break;
1866 default: /* unknown value: do nothing */
1867 break;
1871 static void tweak_split_index(struct index_state *istate)
1873 switch (git_config_get_split_index()) {
1874 case -1: /* unset: do nothing */
1875 break;
1876 case 0: /* false */
1877 remove_split_index(istate);
1878 break;
1879 case 1: /* true */
1880 add_split_index(istate);
1881 break;
1882 default: /* unknown value: do nothing */
1883 break;
1887 static void post_read_index_from(struct index_state *istate)
1889 check_ce_order(istate);
1890 tweak_untracked_cache(istate);
1891 tweak_split_index(istate);
1892 tweak_fsmonitor(istate);
1895 static size_t estimate_cache_size_from_compressed(unsigned int entries)
1897 return entries * (sizeof(struct cache_entry) + CACHE_ENTRY_PATH_LENGTH);
1900 static size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1902 long per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1905 * Account for potential alignment differences.
1907 per_entry += align_padding_size(sizeof(struct cache_entry), -sizeof(struct ondisk_cache_entry));
1908 return ondisk_size + entries * per_entry;
1911 struct index_entry_offset
1913 /* starting byte offset into index file, count of index entries in this block */
1914 int offset, nr;
1917 struct index_entry_offset_table
1919 int nr;
1920 struct index_entry_offset entries[FLEX_ARRAY];
1923 #ifndef NO_PTHREADS
1924 static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset);
1925 static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot);
1926 #endif
1928 static size_t read_eoie_extension(const char *mmap, size_t mmap_size);
1929 static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset);
1931 struct load_index_extensions
1933 #ifndef NO_PTHREADS
1934 pthread_t pthread;
1935 #endif
1936 struct index_state *istate;
1937 const char *mmap;
1938 size_t mmap_size;
1939 unsigned long src_offset;
1942 static void *load_index_extensions(void *_data)
1944 struct load_index_extensions *p = _data;
1945 unsigned long src_offset = p->src_offset;
1947 while (src_offset <= p->mmap_size - the_hash_algo->rawsz - 8) {
1948 /* After an array of active_nr index entries,
1949 * there can be arbitrary number of extended
1950 * sections, each of which is prefixed with
1951 * extension name (4-byte) and section length
1952 * in 4-byte network byte order.
1954 uint32_t extsize = get_be32(p->mmap + src_offset + 4);
1955 if (read_index_extension(p->istate,
1956 p->mmap + src_offset,
1957 p->mmap + src_offset + 8,
1958 extsize) < 0) {
1959 munmap((void *)p->mmap, p->mmap_size);
1960 die(_("index file corrupt"));
1962 src_offset += 8;
1963 src_offset += extsize;
1966 return NULL;
1970 * A helper function that will load the specified range of cache entries
1971 * from the memory mapped file and add them to the given index.
1973 static unsigned long load_cache_entry_block(struct index_state *istate,
1974 struct mem_pool *ce_mem_pool, int offset, int nr, const char *mmap,
1975 unsigned long start_offset, const struct cache_entry *previous_ce)
1977 int i;
1978 unsigned long src_offset = start_offset;
1980 for (i = offset; i < offset + nr; i++) {
1981 struct ondisk_cache_entry *disk_ce;
1982 struct cache_entry *ce;
1983 unsigned long consumed;
1985 disk_ce = (struct ondisk_cache_entry *)(mmap + src_offset);
1986 ce = create_from_disk(ce_mem_pool, istate->version, disk_ce, &consumed, previous_ce);
1987 set_index_entry(istate, i, ce);
1989 src_offset += consumed;
1990 previous_ce = ce;
1992 return src_offset - start_offset;
1995 static unsigned long load_all_cache_entries(struct index_state *istate,
1996 const char *mmap, size_t mmap_size, unsigned long src_offset)
1998 unsigned long consumed;
2000 if (istate->version == 4) {
2001 mem_pool_init(&istate->ce_mem_pool,
2002 estimate_cache_size_from_compressed(istate->cache_nr));
2003 } else {
2004 mem_pool_init(&istate->ce_mem_pool,
2005 estimate_cache_size(mmap_size, istate->cache_nr));
2008 consumed = load_cache_entry_block(istate, istate->ce_mem_pool,
2009 0, istate->cache_nr, mmap, src_offset, NULL);
2010 return consumed;
2013 #ifndef NO_PTHREADS
2016 * Mostly randomly chosen maximum thread counts: we
2017 * cap the parallelism to online_cpus() threads, and we want
2018 * to have at least 10000 cache entries per thread for it to
2019 * be worth starting a thread.
2022 #define THREAD_COST (10000)
2024 struct load_cache_entries_thread_data
2026 pthread_t pthread;
2027 struct index_state *istate;
2028 struct mem_pool *ce_mem_pool;
2029 int offset;
2030 const char *mmap;
2031 struct index_entry_offset_table *ieot;
2032 int ieot_start; /* starting index into the ieot array */
2033 int ieot_blocks; /* count of ieot entries to process */
2034 unsigned long consumed; /* return # of bytes in index file processed */
2038 * A thread proc to run the load_cache_entries() computation
2039 * across multiple background threads.
2041 static void *load_cache_entries_thread(void *_data)
2043 struct load_cache_entries_thread_data *p = _data;
2044 int i;
2046 /* iterate across all ieot blocks assigned to this thread */
2047 for (i = p->ieot_start; i < p->ieot_start + p->ieot_blocks; i++) {
2048 p->consumed += load_cache_entry_block(p->istate, p->ce_mem_pool,
2049 p->offset, p->ieot->entries[i].nr, p->mmap, p->ieot->entries[i].offset, NULL);
2050 p->offset += p->ieot->entries[i].nr;
2052 return NULL;
2055 static unsigned long load_cache_entries_threaded(struct index_state *istate, const char *mmap, size_t mmap_size,
2056 unsigned long src_offset, int nr_threads, struct index_entry_offset_table *ieot)
2058 int i, offset, ieot_blocks, ieot_start, err;
2059 struct load_cache_entries_thread_data *data;
2060 unsigned long consumed = 0;
2062 /* a little sanity checking */
2063 if (istate->name_hash_initialized)
2064 BUG("the name hash isn't thread safe");
2066 mem_pool_init(&istate->ce_mem_pool, 0);
2068 /* ensure we have no more threads than we have blocks to process */
2069 if (nr_threads > ieot->nr)
2070 nr_threads = ieot->nr;
2071 data = xcalloc(nr_threads, sizeof(*data));
2073 offset = ieot_start = 0;
2074 ieot_blocks = DIV_ROUND_UP(ieot->nr, nr_threads);
2075 for (i = 0; i < nr_threads; i++) {
2076 struct load_cache_entries_thread_data *p = &data[i];
2077 int nr, j;
2079 if (ieot_start + ieot_blocks > ieot->nr)
2080 ieot_blocks = ieot->nr - ieot_start;
2082 p->istate = istate;
2083 p->offset = offset;
2084 p->mmap = mmap;
2085 p->ieot = ieot;
2086 p->ieot_start = ieot_start;
2087 p->ieot_blocks = ieot_blocks;
2089 /* create a mem_pool for each thread */
2090 nr = 0;
2091 for (j = p->ieot_start; j < p->ieot_start + p->ieot_blocks; j++)
2092 nr += p->ieot->entries[j].nr;
2093 if (istate->version == 4) {
2094 mem_pool_init(&p->ce_mem_pool,
2095 estimate_cache_size_from_compressed(nr));
2096 } else {
2097 mem_pool_init(&p->ce_mem_pool,
2098 estimate_cache_size(mmap_size, nr));
2101 err = pthread_create(&p->pthread, NULL, load_cache_entries_thread, p);
2102 if (err)
2103 die(_("unable to create load_cache_entries thread: %s"), strerror(err));
2105 /* increment by the number of cache entries in the ieot block being processed */
2106 for (j = 0; j < ieot_blocks; j++)
2107 offset += ieot->entries[ieot_start + j].nr;
2108 ieot_start += ieot_blocks;
2111 for (i = 0; i < nr_threads; i++) {
2112 struct load_cache_entries_thread_data *p = &data[i];
2114 err = pthread_join(p->pthread, NULL);
2115 if (err)
2116 die(_("unable to join load_cache_entries thread: %s"), strerror(err));
2117 mem_pool_combine(istate->ce_mem_pool, p->ce_mem_pool);
2118 consumed += p->consumed;
2121 free(data);
2123 return consumed;
2125 #endif
2127 /* remember to discard_cache() before reading a different cache! */
2128 int do_read_index(struct index_state *istate, const char *path, int must_exist)
2130 int fd;
2131 struct stat st;
2132 unsigned long src_offset;
2133 const struct cache_header *hdr;
2134 const char *mmap;
2135 size_t mmap_size;
2136 struct load_index_extensions p;
2137 size_t extension_offset = 0;
2138 #ifndef NO_PTHREADS
2139 int nr_threads, cpus;
2140 struct index_entry_offset_table *ieot = NULL;
2141 #endif
2143 if (istate->initialized)
2144 return istate->cache_nr;
2146 istate->timestamp.sec = 0;
2147 istate->timestamp.nsec = 0;
2148 fd = open(path, O_RDONLY);
2149 if (fd < 0) {
2150 if (!must_exist && errno == ENOENT)
2151 return 0;
2152 die_errno("%s: index file open failed", path);
2155 if (fstat(fd, &st))
2156 die_errno("cannot stat the open index");
2158 mmap_size = xsize_t(st.st_size);
2159 if (mmap_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2160 die("index file smaller than expected");
2162 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
2163 if (mmap == MAP_FAILED)
2164 die_errno("unable to map index file");
2165 close(fd);
2167 hdr = (const struct cache_header *)mmap;
2168 if (verify_hdr(hdr, mmap_size) < 0)
2169 goto unmap;
2171 hashcpy(istate->oid.hash, (const unsigned char *)hdr + mmap_size - the_hash_algo->rawsz);
2172 istate->version = ntohl(hdr->hdr_version);
2173 istate->cache_nr = ntohl(hdr->hdr_entries);
2174 istate->cache_alloc = alloc_nr(istate->cache_nr);
2175 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
2176 istate->initialized = 1;
2178 p.istate = istate;
2179 p.mmap = mmap;
2180 p.mmap_size = mmap_size;
2182 src_offset = sizeof(*hdr);
2184 #ifndef NO_PTHREADS
2185 nr_threads = git_config_get_index_threads();
2187 /* TODO: does creating more threads than cores help? */
2188 if (!nr_threads) {
2189 nr_threads = istate->cache_nr / THREAD_COST;
2190 cpus = online_cpus();
2191 if (nr_threads > cpus)
2192 nr_threads = cpus;
2195 if (nr_threads > 1) {
2196 extension_offset = read_eoie_extension(mmap, mmap_size);
2197 if (extension_offset) {
2198 int err;
2200 p.src_offset = extension_offset;
2201 err = pthread_create(&p.pthread, NULL, load_index_extensions, &p);
2202 if (err)
2203 die(_("unable to create load_index_extensions thread: %s"), strerror(err));
2205 nr_threads--;
2210 * Locate and read the index entry offset table so that we can use it
2211 * to multi-thread the reading of the cache entries.
2213 if (extension_offset && nr_threads > 1)
2214 ieot = read_ieot_extension(mmap, mmap_size, extension_offset);
2216 if (ieot) {
2217 src_offset += load_cache_entries_threaded(istate, mmap, mmap_size, src_offset, nr_threads, ieot);
2218 free(ieot);
2219 } else {
2220 src_offset += load_all_cache_entries(istate, mmap, mmap_size, src_offset);
2222 #else
2223 src_offset += load_all_cache_entries(istate, mmap, mmap_size, src_offset);
2224 #endif
2226 istate->timestamp.sec = st.st_mtime;
2227 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2229 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2230 #ifndef NO_PTHREADS
2231 if (extension_offset) {
2232 int ret = pthread_join(p.pthread, NULL);
2233 if (ret)
2234 die(_("unable to join load_index_extensions thread: %s"), strerror(ret));
2236 #endif
2237 if (!extension_offset) {
2238 p.src_offset = src_offset;
2239 load_index_extensions(&p);
2241 munmap((void *)mmap, mmap_size);
2242 return istate->cache_nr;
2244 unmap:
2245 munmap((void *)mmap, mmap_size);
2246 die("index file corrupt");
2250 * Signal that the shared index is used by updating its mtime.
2252 * This way, shared index can be removed if they have not been used
2253 * for some time.
2255 static void freshen_shared_index(const char *shared_index, int warn)
2257 if (!check_and_freshen_file(shared_index, 1) && warn)
2258 warning("could not freshen shared index '%s'", shared_index);
2261 int read_index_from(struct index_state *istate, const char *path,
2262 const char *gitdir)
2264 struct split_index *split_index;
2265 int ret;
2266 char *base_oid_hex;
2267 char *base_path;
2269 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2270 if (istate->initialized)
2271 return istate->cache_nr;
2273 trace_performance_enter();
2274 ret = do_read_index(istate, path, 0);
2275 trace_performance_leave("read cache %s", path);
2277 split_index = istate->split_index;
2278 if (!split_index || is_null_oid(&split_index->base_oid)) {
2279 post_read_index_from(istate);
2280 return ret;
2283 trace_performance_enter();
2284 if (split_index->base)
2285 discard_index(split_index->base);
2286 else
2287 split_index->base = xcalloc(1, sizeof(*split_index->base));
2289 base_oid_hex = oid_to_hex(&split_index->base_oid);
2290 base_path = xstrfmt("%s/sharedindex.%s", gitdir, base_oid_hex);
2291 ret = do_read_index(split_index->base, base_path, 1);
2292 if (!oideq(&split_index->base_oid, &split_index->base->oid))
2293 die("broken index, expect %s in %s, got %s",
2294 base_oid_hex, base_path,
2295 oid_to_hex(&split_index->base->oid));
2297 freshen_shared_index(base_path, 0);
2298 merge_base_index(istate);
2299 post_read_index_from(istate);
2300 free(base_path);
2301 trace_performance_leave("read cache %s", base_path);
2302 return ret;
2305 int is_index_unborn(struct index_state *istate)
2307 return (!istate->cache_nr && !istate->timestamp.sec);
2310 int discard_index(struct index_state *istate)
2313 * Cache entries in istate->cache[] should have been allocated
2314 * from the memory pool associated with this index, or from an
2315 * associated split_index. There is no need to free individual
2316 * cache entries. validate_cache_entries can detect when this
2317 * assertion does not hold.
2319 validate_cache_entries(istate);
2321 resolve_undo_clear_index(istate);
2322 istate->cache_nr = 0;
2323 istate->cache_changed = 0;
2324 istate->timestamp.sec = 0;
2325 istate->timestamp.nsec = 0;
2326 free_name_hash(istate);
2327 cache_tree_free(&(istate->cache_tree));
2328 istate->initialized = 0;
2329 FREE_AND_NULL(istate->cache);
2330 istate->cache_alloc = 0;
2331 discard_split_index(istate);
2332 free_untracked_cache(istate->untracked);
2333 istate->untracked = NULL;
2335 if (istate->ce_mem_pool) {
2336 mem_pool_discard(istate->ce_mem_pool, should_validate_cache_entries());
2337 istate->ce_mem_pool = NULL;
2340 return 0;
2344 * Validate the cache entries of this index.
2345 * All cache entries associated with this index
2346 * should have been allocated by the memory pool
2347 * associated with this index, or by a referenced
2348 * split index.
2350 void validate_cache_entries(const struct index_state *istate)
2352 int i;
2354 if (!should_validate_cache_entries() ||!istate || !istate->initialized)
2355 return;
2357 for (i = 0; i < istate->cache_nr; i++) {
2358 if (!istate) {
2359 die("internal error: cache entry is not allocated from expected memory pool");
2360 } else if (!istate->ce_mem_pool ||
2361 !mem_pool_contains(istate->ce_mem_pool, istate->cache[i])) {
2362 if (!istate->split_index ||
2363 !istate->split_index->base ||
2364 !istate->split_index->base->ce_mem_pool ||
2365 !mem_pool_contains(istate->split_index->base->ce_mem_pool, istate->cache[i])) {
2366 die("internal error: cache entry is not allocated from expected memory pool");
2371 if (istate->split_index)
2372 validate_cache_entries(istate->split_index->base);
2375 int unmerged_index(const struct index_state *istate)
2377 int i;
2378 for (i = 0; i < istate->cache_nr; i++) {
2379 if (ce_stage(istate->cache[i]))
2380 return 1;
2382 return 0;
2385 int index_has_changes(struct index_state *istate,
2386 struct tree *tree,
2387 struct strbuf *sb)
2389 struct object_id cmp;
2390 int i;
2392 if (istate != &the_index) {
2393 BUG("index_has_changes cannot yet accept istate != &the_index; do_diff_cache needs updating first.");
2395 if (tree)
2396 cmp = tree->object.oid;
2397 if (tree || !get_oid_tree("HEAD", &cmp)) {
2398 struct diff_options opt;
2400 repo_diff_setup(the_repository, &opt);
2401 opt.flags.exit_with_status = 1;
2402 if (!sb)
2403 opt.flags.quick = 1;
2404 do_diff_cache(&cmp, &opt);
2405 diffcore_std(&opt);
2406 for (i = 0; sb && i < diff_queued_diff.nr; i++) {
2407 if (i)
2408 strbuf_addch(sb, ' ');
2409 strbuf_addstr(sb, diff_queued_diff.queue[i]->two->path);
2411 diff_flush(&opt);
2412 return opt.flags.has_changes != 0;
2413 } else {
2414 for (i = 0; sb && i < istate->cache_nr; i++) {
2415 if (i)
2416 strbuf_addch(sb, ' ');
2417 strbuf_addstr(sb, istate->cache[i]->name);
2419 return !!istate->cache_nr;
2423 #define WRITE_BUFFER_SIZE 8192
2424 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
2425 static unsigned long write_buffer_len;
2427 static int ce_write_flush(git_hash_ctx *context, int fd)
2429 unsigned int buffered = write_buffer_len;
2430 if (buffered) {
2431 the_hash_algo->update_fn(context, write_buffer, buffered);
2432 if (write_in_full(fd, write_buffer, buffered) < 0)
2433 return -1;
2434 write_buffer_len = 0;
2436 return 0;
2439 static int ce_write(git_hash_ctx *context, int fd, void *data, unsigned int len)
2441 while (len) {
2442 unsigned int buffered = write_buffer_len;
2443 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
2444 if (partial > len)
2445 partial = len;
2446 memcpy(write_buffer + buffered, data, partial);
2447 buffered += partial;
2448 if (buffered == WRITE_BUFFER_SIZE) {
2449 write_buffer_len = buffered;
2450 if (ce_write_flush(context, fd))
2451 return -1;
2452 buffered = 0;
2454 write_buffer_len = buffered;
2455 len -= partial;
2456 data = (char *) data + partial;
2458 return 0;
2461 static int write_index_ext_header(git_hash_ctx *context, git_hash_ctx *eoie_context,
2462 int fd, unsigned int ext, unsigned int sz)
2464 ext = htonl(ext);
2465 sz = htonl(sz);
2466 if (eoie_context) {
2467 the_hash_algo->update_fn(eoie_context, &ext, 4);
2468 the_hash_algo->update_fn(eoie_context, &sz, 4);
2470 return ((ce_write(context, fd, &ext, 4) < 0) ||
2471 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
2474 static int ce_flush(git_hash_ctx *context, int fd, unsigned char *hash)
2476 unsigned int left = write_buffer_len;
2478 if (left) {
2479 write_buffer_len = 0;
2480 the_hash_algo->update_fn(context, write_buffer, left);
2483 /* Flush first if not enough space for hash signature */
2484 if (left + the_hash_algo->rawsz > WRITE_BUFFER_SIZE) {
2485 if (write_in_full(fd, write_buffer, left) < 0)
2486 return -1;
2487 left = 0;
2490 /* Append the hash signature at the end */
2491 the_hash_algo->final_fn(write_buffer + left, context);
2492 hashcpy(hash, write_buffer + left);
2493 left += the_hash_algo->rawsz;
2494 return (write_in_full(fd, write_buffer, left) < 0) ? -1 : 0;
2497 static void ce_smudge_racily_clean_entry(struct index_state *istate,
2498 struct cache_entry *ce)
2501 * The only thing we care about in this function is to smudge the
2502 * falsely clean entry due to touch-update-touch race, so we leave
2503 * everything else as they are. We are called for entries whose
2504 * ce_stat_data.sd_mtime match the index file mtime.
2506 * Note that this actually does not do much for gitlinks, for
2507 * which ce_match_stat_basic() always goes to the actual
2508 * contents. The caller checks with is_racy_timestamp() which
2509 * always says "no" for gitlinks, so we are not called for them ;-)
2511 struct stat st;
2513 if (lstat(ce->name, &st) < 0)
2514 return;
2515 if (ce_match_stat_basic(ce, &st))
2516 return;
2517 if (ce_modified_check_fs(istate, ce, &st)) {
2518 /* This is "racily clean"; smudge it. Note that this
2519 * is a tricky code. At first glance, it may appear
2520 * that it can break with this sequence:
2522 * $ echo xyzzy >frotz
2523 * $ git-update-index --add frotz
2524 * $ : >frotz
2525 * $ sleep 3
2526 * $ echo filfre >nitfol
2527 * $ git-update-index --add nitfol
2529 * but it does not. When the second update-index runs,
2530 * it notices that the entry "frotz" has the same timestamp
2531 * as index, and if we were to smudge it by resetting its
2532 * size to zero here, then the object name recorded
2533 * in index is the 6-byte file but the cached stat information
2534 * becomes zero --- which would then match what we would
2535 * obtain from the filesystem next time we stat("frotz").
2537 * However, the second update-index, before calling
2538 * this function, notices that the cached size is 6
2539 * bytes and what is on the filesystem is an empty
2540 * file, and never calls us, so the cached size information
2541 * for "frotz" stays 6 which does not match the filesystem.
2543 ce->ce_stat_data.sd_size = 0;
2547 /* Copy miscellaneous fields but not the name */
2548 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
2549 struct cache_entry *ce)
2551 short flags;
2553 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2554 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2555 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2556 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2557 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
2558 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
2559 ondisk->mode = htonl(ce->ce_mode);
2560 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
2561 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
2562 ondisk->size = htonl(ce->ce_stat_data.sd_size);
2563 hashcpy(ondisk->sha1, ce->oid.hash);
2565 flags = ce->ce_flags & ~CE_NAMEMASK;
2566 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2567 ondisk->flags = htons(flags);
2568 if (ce->ce_flags & CE_EXTENDED) {
2569 struct ondisk_cache_entry_extended *ondisk2;
2570 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
2571 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
2575 static int ce_write_entry(git_hash_ctx *c, int fd, struct cache_entry *ce,
2576 struct strbuf *previous_name, struct ondisk_cache_entry *ondisk)
2578 int size;
2579 int result;
2580 unsigned int saved_namelen;
2581 int stripped_name = 0;
2582 static unsigned char padding[8] = { 0x00 };
2584 if (ce->ce_flags & CE_STRIP_NAME) {
2585 saved_namelen = ce_namelen(ce);
2586 ce->ce_namelen = 0;
2587 stripped_name = 1;
2590 if (ce->ce_flags & CE_EXTENDED)
2591 size = offsetof(struct ondisk_cache_entry_extended, name);
2592 else
2593 size = offsetof(struct ondisk_cache_entry, name);
2595 if (!previous_name) {
2596 int len = ce_namelen(ce);
2597 copy_cache_entry_to_ondisk(ondisk, ce);
2598 result = ce_write(c, fd, ondisk, size);
2599 if (!result)
2600 result = ce_write(c, fd, ce->name, len);
2601 if (!result)
2602 result = ce_write(c, fd, padding, align_padding_size(size, len));
2603 } else {
2604 int common, to_remove, prefix_size;
2605 unsigned char to_remove_vi[16];
2606 for (common = 0;
2607 (ce->name[common] &&
2608 common < previous_name->len &&
2609 ce->name[common] == previous_name->buf[common]);
2610 common++)
2611 ; /* still matching */
2612 to_remove = previous_name->len - common;
2613 prefix_size = encode_varint(to_remove, to_remove_vi);
2615 copy_cache_entry_to_ondisk(ondisk, ce);
2616 result = ce_write(c, fd, ondisk, size);
2617 if (!result)
2618 result = ce_write(c, fd, to_remove_vi, prefix_size);
2619 if (!result)
2620 result = ce_write(c, fd, ce->name + common, ce_namelen(ce) - common);
2621 if (!result)
2622 result = ce_write(c, fd, padding, 1);
2624 strbuf_splice(previous_name, common, to_remove,
2625 ce->name + common, ce_namelen(ce) - common);
2627 if (stripped_name) {
2628 ce->ce_namelen = saved_namelen;
2629 ce->ce_flags &= ~CE_STRIP_NAME;
2632 return result;
2636 * This function verifies if index_state has the correct sha1 of the
2637 * index file. Don't die if we have any other failure, just return 0.
2639 static int verify_index_from(const struct index_state *istate, const char *path)
2641 int fd;
2642 ssize_t n;
2643 struct stat st;
2644 unsigned char hash[GIT_MAX_RAWSZ];
2646 if (!istate->initialized)
2647 return 0;
2649 fd = open(path, O_RDONLY);
2650 if (fd < 0)
2651 return 0;
2653 if (fstat(fd, &st))
2654 goto out;
2656 if (st.st_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2657 goto out;
2659 n = pread_in_full(fd, hash, the_hash_algo->rawsz, st.st_size - the_hash_algo->rawsz);
2660 if (n != the_hash_algo->rawsz)
2661 goto out;
2663 if (!hasheq(istate->oid.hash, hash))
2664 goto out;
2666 close(fd);
2667 return 1;
2669 out:
2670 close(fd);
2671 return 0;
2674 static int verify_index(const struct index_state *istate)
2676 return verify_index_from(istate, get_index_file());
2679 static int has_racy_timestamp(struct index_state *istate)
2681 int entries = istate->cache_nr;
2682 int i;
2684 for (i = 0; i < entries; i++) {
2685 struct cache_entry *ce = istate->cache[i];
2686 if (is_racy_timestamp(istate, ce))
2687 return 1;
2689 return 0;
2692 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
2694 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
2695 verify_index(istate))
2696 write_locked_index(istate, lockfile, COMMIT_LOCK);
2697 else
2698 rollback_lock_file(lockfile);
2702 * On success, `tempfile` is closed. If it is the temporary file
2703 * of a `struct lock_file`, we will therefore effectively perform
2704 * a 'close_lock_file_gently()`. Since that is an implementation
2705 * detail of lockfiles, callers of `do_write_index()` should not
2706 * rely on it.
2708 static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
2709 int strip_extensions)
2711 uint64_t start = getnanotime();
2712 int newfd = tempfile->fd;
2713 git_hash_ctx c, eoie_c;
2714 struct cache_header hdr;
2715 int i, err = 0, removed, extended, hdr_version;
2716 struct cache_entry **cache = istate->cache;
2717 int entries = istate->cache_nr;
2718 struct stat st;
2719 struct ondisk_cache_entry_extended ondisk;
2720 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2721 int drop_cache_tree = istate->drop_cache_tree;
2722 off_t offset;
2723 int ieot_entries = 1;
2724 struct index_entry_offset_table *ieot = NULL;
2725 int nr, nr_threads;
2727 for (i = removed = extended = 0; i < entries; i++) {
2728 if (cache[i]->ce_flags & CE_REMOVE)
2729 removed++;
2731 /* reduce extended entries if possible */
2732 cache[i]->ce_flags &= ~CE_EXTENDED;
2733 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2734 extended++;
2735 cache[i]->ce_flags |= CE_EXTENDED;
2739 if (!istate->version) {
2740 istate->version = get_index_format_default();
2741 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
2742 init_split_index(istate);
2745 /* demote version 3 to version 2 when the latter suffices */
2746 if (istate->version == 3 || istate->version == 2)
2747 istate->version = extended ? 3 : 2;
2749 hdr_version = istate->version;
2751 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2752 hdr.hdr_version = htonl(hdr_version);
2753 hdr.hdr_entries = htonl(entries - removed);
2755 the_hash_algo->init_fn(&c);
2756 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2757 return -1;
2759 #ifndef NO_PTHREADS
2760 nr_threads = git_config_get_index_threads();
2761 if (nr_threads != 1) {
2762 int ieot_blocks, cpus;
2765 * ensure default number of ieot blocks maps evenly to the
2766 * default number of threads that will process them leaving
2767 * room for the thread to load the index extensions.
2769 if (!nr_threads) {
2770 ieot_blocks = istate->cache_nr / THREAD_COST;
2771 cpus = online_cpus();
2772 if (ieot_blocks > cpus - 1)
2773 ieot_blocks = cpus - 1;
2774 } else {
2775 ieot_blocks = nr_threads;
2776 if (ieot_blocks > istate->cache_nr)
2777 ieot_blocks = istate->cache_nr;
2781 * no reason to write out the IEOT extension if we don't
2782 * have enough blocks to utilize multi-threading
2784 if (ieot_blocks > 1) {
2785 ieot = xcalloc(1, sizeof(struct index_entry_offset_table)
2786 + (ieot_blocks * sizeof(struct index_entry_offset)));
2787 ieot_entries = DIV_ROUND_UP(entries, ieot_blocks);
2790 #endif
2792 offset = lseek(newfd, 0, SEEK_CUR);
2793 if (offset < 0) {
2794 free(ieot);
2795 return -1;
2797 offset += write_buffer_len;
2798 nr = 0;
2799 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2801 for (i = 0; i < entries; i++) {
2802 struct cache_entry *ce = cache[i];
2803 if (ce->ce_flags & CE_REMOVE)
2804 continue;
2805 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2806 ce_smudge_racily_clean_entry(istate, ce);
2807 if (is_null_oid(&ce->oid)) {
2808 static const char msg[] = "cache entry has null sha1: %s";
2809 static int allow = -1;
2811 if (allow < 0)
2812 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2813 if (allow)
2814 warning(msg, ce->name);
2815 else
2816 err = error(msg, ce->name);
2818 drop_cache_tree = 1;
2820 if (ieot && i && (i % ieot_entries == 0)) {
2821 ieot->entries[ieot->nr].nr = nr;
2822 ieot->entries[ieot->nr].offset = offset;
2823 ieot->nr++;
2825 * If we have a V4 index, set the first byte to an invalid
2826 * character to ensure there is nothing common with the previous
2827 * entry
2829 if (previous_name)
2830 previous_name->buf[0] = 0;
2831 nr = 0;
2832 offset = lseek(newfd, 0, SEEK_CUR);
2833 if (offset < 0) {
2834 free(ieot);
2835 return -1;
2837 offset += write_buffer_len;
2839 if (ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
2840 err = -1;
2842 if (err)
2843 break;
2844 nr++;
2846 if (ieot && nr) {
2847 ieot->entries[ieot->nr].nr = nr;
2848 ieot->entries[ieot->nr].offset = offset;
2849 ieot->nr++;
2851 strbuf_release(&previous_name_buf);
2853 if (err) {
2854 free(ieot);
2855 return err;
2858 /* Write extension data here */
2859 offset = lseek(newfd, 0, SEEK_CUR);
2860 if (offset < 0) {
2861 free(ieot);
2862 return -1;
2864 offset += write_buffer_len;
2865 the_hash_algo->init_fn(&eoie_c);
2868 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2869 * can minimize the number of extensions we have to scan through to
2870 * find it during load. Write it out regardless of the
2871 * strip_extensions parameter as we need it when loading the shared
2872 * index.
2874 #ifndef NO_PTHREADS
2875 if (ieot) {
2876 struct strbuf sb = STRBUF_INIT;
2878 write_ieot_extension(&sb, ieot);
2879 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_INDEXENTRYOFFSETTABLE, sb.len) < 0
2880 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2881 strbuf_release(&sb);
2882 free(ieot);
2883 if (err)
2884 return -1;
2886 #endif
2888 if (!strip_extensions && istate->split_index) {
2889 struct strbuf sb = STRBUF_INIT;
2891 err = write_link_extension(&sb, istate) < 0 ||
2892 write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_LINK,
2893 sb.len) < 0 ||
2894 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2895 strbuf_release(&sb);
2896 if (err)
2897 return -1;
2899 if (!strip_extensions && !drop_cache_tree && istate->cache_tree) {
2900 struct strbuf sb = STRBUF_INIT;
2902 cache_tree_write(&sb, istate->cache_tree);
2903 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_TREE, sb.len) < 0
2904 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2905 strbuf_release(&sb);
2906 if (err)
2907 return -1;
2909 if (!strip_extensions && istate->resolve_undo) {
2910 struct strbuf sb = STRBUF_INIT;
2912 resolve_undo_write(&sb, istate->resolve_undo);
2913 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_RESOLVE_UNDO,
2914 sb.len) < 0
2915 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2916 strbuf_release(&sb);
2917 if (err)
2918 return -1;
2920 if (!strip_extensions && istate->untracked) {
2921 struct strbuf sb = STRBUF_INIT;
2923 write_untracked_extension(&sb, istate->untracked);
2924 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_UNTRACKED,
2925 sb.len) < 0 ||
2926 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2927 strbuf_release(&sb);
2928 if (err)
2929 return -1;
2931 if (!strip_extensions && istate->fsmonitor_last_update) {
2932 struct strbuf sb = STRBUF_INIT;
2934 write_fsmonitor_extension(&sb, istate);
2935 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_FSMONITOR, sb.len) < 0
2936 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2937 strbuf_release(&sb);
2938 if (err)
2939 return -1;
2943 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
2944 * so that it can be found and processed before all the index entries are
2945 * read. Write it out regardless of the strip_extensions parameter as we need it
2946 * when loading the shared index.
2948 if (offset) {
2949 struct strbuf sb = STRBUF_INIT;
2951 write_eoie_extension(&sb, &eoie_c, offset);
2952 err = write_index_ext_header(&c, NULL, newfd, CACHE_EXT_ENDOFINDEXENTRIES, sb.len) < 0
2953 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2954 strbuf_release(&sb);
2955 if (err)
2956 return -1;
2959 if (ce_flush(&c, newfd, istate->oid.hash))
2960 return -1;
2961 if (close_tempfile_gently(tempfile)) {
2962 error(_("could not close '%s'"), tempfile->filename.buf);
2963 return -1;
2965 if (stat(tempfile->filename.buf, &st))
2966 return -1;
2967 istate->timestamp.sec = (unsigned int)st.st_mtime;
2968 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2969 trace_performance_since(start, "write index, changed mask = %x", istate->cache_changed);
2970 return 0;
2973 void set_alternate_index_output(const char *name)
2975 alternate_index_output = name;
2978 static int commit_locked_index(struct lock_file *lk)
2980 if (alternate_index_output)
2981 return commit_lock_file_to(lk, alternate_index_output);
2982 else
2983 return commit_lock_file(lk);
2986 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2987 unsigned flags)
2989 int ret = do_write_index(istate, lock->tempfile, 0);
2990 if (ret)
2991 return ret;
2992 if (flags & COMMIT_LOCK)
2993 return commit_locked_index(lock);
2994 return close_lock_file_gently(lock);
2997 static int write_split_index(struct index_state *istate,
2998 struct lock_file *lock,
2999 unsigned flags)
3001 int ret;
3002 prepare_to_write_split_index(istate);
3003 ret = do_write_locked_index(istate, lock, flags);
3004 finish_writing_split_index(istate);
3005 return ret;
3008 static const char *shared_index_expire = "2.weeks.ago";
3010 static unsigned long get_shared_index_expire_date(void)
3012 static unsigned long shared_index_expire_date;
3013 static int shared_index_expire_date_prepared;
3015 if (!shared_index_expire_date_prepared) {
3016 git_config_get_expiry("splitindex.sharedindexexpire",
3017 &shared_index_expire);
3018 shared_index_expire_date = approxidate(shared_index_expire);
3019 shared_index_expire_date_prepared = 1;
3022 return shared_index_expire_date;
3025 static int should_delete_shared_index(const char *shared_index_path)
3027 struct stat st;
3028 unsigned long expiration;
3030 /* Check timestamp */
3031 expiration = get_shared_index_expire_date();
3032 if (!expiration)
3033 return 0;
3034 if (stat(shared_index_path, &st))
3035 return error_errno(_("could not stat '%s'"), shared_index_path);
3036 if (st.st_mtime > expiration)
3037 return 0;
3039 return 1;
3042 static int clean_shared_index_files(const char *current_hex)
3044 struct dirent *de;
3045 DIR *dir = opendir(get_git_dir());
3047 if (!dir)
3048 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3050 while ((de = readdir(dir)) != NULL) {
3051 const char *sha1_hex;
3052 const char *shared_index_path;
3053 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
3054 continue;
3055 if (!strcmp(sha1_hex, current_hex))
3056 continue;
3057 shared_index_path = git_path("%s", de->d_name);
3058 if (should_delete_shared_index(shared_index_path) > 0 &&
3059 unlink(shared_index_path))
3060 warning_errno(_("unable to unlink: %s"), shared_index_path);
3062 closedir(dir);
3064 return 0;
3067 static int write_shared_index(struct index_state *istate,
3068 struct tempfile **temp)
3070 struct split_index *si = istate->split_index;
3071 int ret;
3073 move_cache_to_base_index(istate);
3074 ret = do_write_index(si->base, *temp, 1);
3075 if (ret)
3076 return ret;
3077 ret = adjust_shared_perm(get_tempfile_path(*temp));
3078 if (ret) {
3079 error("cannot fix permission bits on %s", get_tempfile_path(*temp));
3080 return ret;
3082 ret = rename_tempfile(temp,
3083 git_path("sharedindex.%s", oid_to_hex(&si->base->oid)));
3084 if (!ret) {
3085 oidcpy(&si->base_oid, &si->base->oid);
3086 clean_shared_index_files(oid_to_hex(&si->base->oid));
3089 return ret;
3092 static const int default_max_percent_split_change = 20;
3094 static int too_many_not_shared_entries(struct index_state *istate)
3096 int i, not_shared = 0;
3097 int max_split = git_config_get_max_percent_split_change();
3099 switch (max_split) {
3100 case -1:
3101 /* not or badly configured: use the default value */
3102 max_split = default_max_percent_split_change;
3103 break;
3104 case 0:
3105 return 1; /* 0% means always write a new shared index */
3106 case 100:
3107 return 0; /* 100% means never write a new shared index */
3108 default:
3109 break; /* just use the configured value */
3112 /* Count not shared entries */
3113 for (i = 0; i < istate->cache_nr; i++) {
3114 struct cache_entry *ce = istate->cache[i];
3115 if (!ce->index)
3116 not_shared++;
3119 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
3122 int write_locked_index(struct index_state *istate, struct lock_file *lock,
3123 unsigned flags)
3125 int new_shared_index, ret;
3126 struct split_index *si = istate->split_index;
3128 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3129 cache_tree_verify(istate);
3131 if ((flags & SKIP_IF_UNCHANGED) && !istate->cache_changed) {
3132 if (flags & COMMIT_LOCK)
3133 rollback_lock_file(lock);
3134 return 0;
3137 if (istate->fsmonitor_last_update)
3138 fill_fsmonitor_bitmap(istate);
3140 if (!si || alternate_index_output ||
3141 (istate->cache_changed & ~EXTMASK)) {
3142 if (si)
3143 oidclr(&si->base_oid);
3144 ret = do_write_locked_index(istate, lock, flags);
3145 goto out;
3148 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
3149 int v = si->base_oid.hash[0];
3150 if ((v & 15) < 6)
3151 istate->cache_changed |= SPLIT_INDEX_ORDERED;
3153 if (too_many_not_shared_entries(istate))
3154 istate->cache_changed |= SPLIT_INDEX_ORDERED;
3156 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
3158 if (new_shared_index) {
3159 struct tempfile *temp;
3160 int saved_errno;
3162 temp = mks_tempfile(git_path("sharedindex_XXXXXX"));
3163 if (!temp) {
3164 oidclr(&si->base_oid);
3165 ret = do_write_locked_index(istate, lock, flags);
3166 goto out;
3168 ret = write_shared_index(istate, &temp);
3170 saved_errno = errno;
3171 if (is_tempfile_active(temp))
3172 delete_tempfile(&temp);
3173 errno = saved_errno;
3175 if (ret)
3176 goto out;
3179 ret = write_split_index(istate, lock, flags);
3181 /* Freshen the shared index only if the split-index was written */
3182 if (!ret && !new_shared_index) {
3183 const char *shared_index = git_path("sharedindex.%s",
3184 oid_to_hex(&si->base_oid));
3185 freshen_shared_index(shared_index, 1);
3188 out:
3189 if (flags & COMMIT_LOCK)
3190 rollback_lock_file(lock);
3191 return ret;
3195 * Read the index file that is potentially unmerged into given
3196 * index_state, dropping any unmerged entries to stage #0 (potentially
3197 * resulting in a path appearing as both a file and a directory in the
3198 * index; the caller is responsible to clear out the extra entries
3199 * before writing the index to a tree). Returns true if the index is
3200 * unmerged. Callers who want to refuse to work from an unmerged
3201 * state can call this and check its return value, instead of calling
3202 * read_cache().
3204 int read_index_unmerged(struct index_state *istate)
3206 int i;
3207 int unmerged = 0;
3209 read_index(istate);
3210 for (i = 0; i < istate->cache_nr; i++) {
3211 struct cache_entry *ce = istate->cache[i];
3212 struct cache_entry *new_ce;
3213 int len;
3215 if (!ce_stage(ce))
3216 continue;
3217 unmerged = 1;
3218 len = ce_namelen(ce);
3219 new_ce = make_empty_cache_entry(istate, len);
3220 memcpy(new_ce->name, ce->name, len);
3221 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
3222 new_ce->ce_namelen = len;
3223 new_ce->ce_mode = ce->ce_mode;
3224 if (add_index_entry(istate, new_ce, ADD_CACHE_SKIP_DFCHECK))
3225 return error("%s: cannot drop to stage #0",
3226 new_ce->name);
3228 return unmerged;
3232 * Returns 1 if the path is an "other" path with respect to
3233 * the index; that is, the path is not mentioned in the index at all,
3234 * either as a file, a directory with some files in the index,
3235 * or as an unmerged entry.
3237 * We helpfully remove a trailing "/" from directories so that
3238 * the output of read_directory can be used as-is.
3240 int index_name_is_other(const struct index_state *istate, const char *name,
3241 int namelen)
3243 int pos;
3244 if (namelen && name[namelen - 1] == '/')
3245 namelen--;
3246 pos = index_name_pos(istate, name, namelen);
3247 if (0 <= pos)
3248 return 0; /* exact match */
3249 pos = -pos - 1;
3250 if (pos < istate->cache_nr) {
3251 struct cache_entry *ce = istate->cache[pos];
3252 if (ce_namelen(ce) == namelen &&
3253 !memcmp(ce->name, name, namelen))
3254 return 0; /* Yup, this one exists unmerged */
3256 return 1;
3259 void *read_blob_data_from_index(const struct index_state *istate,
3260 const char *path, unsigned long *size)
3262 int pos, len;
3263 unsigned long sz;
3264 enum object_type type;
3265 void *data;
3267 len = strlen(path);
3268 pos = index_name_pos(istate, path, len);
3269 if (pos < 0) {
3271 * We might be in the middle of a merge, in which
3272 * case we would read stage #2 (ours).
3274 int i;
3275 for (i = -pos - 1;
3276 (pos < 0 && i < istate->cache_nr &&
3277 !strcmp(istate->cache[i]->name, path));
3278 i++)
3279 if (ce_stage(istate->cache[i]) == 2)
3280 pos = i;
3282 if (pos < 0)
3283 return NULL;
3284 data = read_object_file(&istate->cache[pos]->oid, &type, &sz);
3285 if (!data || type != OBJ_BLOB) {
3286 free(data);
3287 return NULL;
3289 if (size)
3290 *size = sz;
3291 return data;
3294 void stat_validity_clear(struct stat_validity *sv)
3296 FREE_AND_NULL(sv->sd);
3299 int stat_validity_check(struct stat_validity *sv, const char *path)
3301 struct stat st;
3303 if (stat(path, &st) < 0)
3304 return sv->sd == NULL;
3305 if (!sv->sd)
3306 return 0;
3307 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
3310 void stat_validity_update(struct stat_validity *sv, int fd)
3312 struct stat st;
3314 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
3315 stat_validity_clear(sv);
3316 else {
3317 if (!sv->sd)
3318 sv->sd = xcalloc(1, sizeof(struct stat_data));
3319 fill_stat_data(sv->sd, &st);
3323 void move_index_extensions(struct index_state *dst, struct index_state *src)
3325 dst->untracked = src->untracked;
3326 src->untracked = NULL;
3327 dst->cache_tree = src->cache_tree;
3328 src->cache_tree = NULL;
3331 struct cache_entry *dup_cache_entry(const struct cache_entry *ce,
3332 struct index_state *istate)
3334 unsigned int size = ce_size(ce);
3335 int mem_pool_allocated;
3336 struct cache_entry *new_entry = make_empty_cache_entry(istate, ce_namelen(ce));
3337 mem_pool_allocated = new_entry->mem_pool_allocated;
3339 memcpy(new_entry, ce, size);
3340 new_entry->mem_pool_allocated = mem_pool_allocated;
3341 return new_entry;
3344 void discard_cache_entry(struct cache_entry *ce)
3346 if (ce && should_validate_cache_entries())
3347 memset(ce, 0xCD, cache_entry_size(ce->ce_namelen));
3349 if (ce && ce->mem_pool_allocated)
3350 return;
3352 free(ce);
3355 int should_validate_cache_entries(void)
3357 static int validate_index_cache_entries = -1;
3359 if (validate_index_cache_entries < 0) {
3360 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3361 validate_index_cache_entries = 1;
3362 else
3363 validate_index_cache_entries = 0;
3366 return validate_index_cache_entries;
3369 #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3370 #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3372 static size_t read_eoie_extension(const char *mmap, size_t mmap_size)
3375 * The end of index entries (EOIE) extension is guaranteed to be last
3376 * so that it can be found by scanning backwards from the EOF.
3378 * "EOIE"
3379 * <4-byte length>
3380 * <4-byte offset>
3381 * <20-byte hash>
3383 const char *index, *eoie;
3384 uint32_t extsize;
3385 size_t offset, src_offset;
3386 unsigned char hash[GIT_MAX_RAWSZ];
3387 git_hash_ctx c;
3389 /* ensure we have an index big enough to contain an EOIE extension */
3390 if (mmap_size < sizeof(struct cache_header) + EOIE_SIZE_WITH_HEADER + the_hash_algo->rawsz)
3391 return 0;
3393 /* validate the extension signature */
3394 index = eoie = mmap + mmap_size - EOIE_SIZE_WITH_HEADER - the_hash_algo->rawsz;
3395 if (CACHE_EXT(index) != CACHE_EXT_ENDOFINDEXENTRIES)
3396 return 0;
3397 index += sizeof(uint32_t);
3399 /* validate the extension size */
3400 extsize = get_be32(index);
3401 if (extsize != EOIE_SIZE)
3402 return 0;
3403 index += sizeof(uint32_t);
3406 * Validate the offset we're going to look for the first extension
3407 * signature is after the index header and before the eoie extension.
3409 offset = get_be32(index);
3410 if (mmap + offset < mmap + sizeof(struct cache_header))
3411 return 0;
3412 if (mmap + offset >= eoie)
3413 return 0;
3414 index += sizeof(uint32_t);
3417 * The hash is computed over extension types and their sizes (but not
3418 * their contents). E.g. if we have "TREE" extension that is N-bytes
3419 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3420 * then the hash would be:
3422 * SHA-1("TREE" + <binary representation of N> +
3423 * "REUC" + <binary representation of M>)
3425 src_offset = offset;
3426 the_hash_algo->init_fn(&c);
3427 while (src_offset < mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER) {
3428 /* After an array of active_nr index entries,
3429 * there can be arbitrary number of extended
3430 * sections, each of which is prefixed with
3431 * extension name (4-byte) and section length
3432 * in 4-byte network byte order.
3434 uint32_t extsize;
3435 memcpy(&extsize, mmap + src_offset + 4, 4);
3436 extsize = ntohl(extsize);
3438 /* verify the extension size isn't so large it will wrap around */
3439 if (src_offset + 8 + extsize < src_offset)
3440 return 0;
3442 the_hash_algo->update_fn(&c, mmap + src_offset, 8);
3444 src_offset += 8;
3445 src_offset += extsize;
3447 the_hash_algo->final_fn(hash, &c);
3448 if (!hasheq(hash, (const unsigned char *)index))
3449 return 0;
3451 /* Validate that the extension offsets returned us back to the eoie extension. */
3452 if (src_offset != mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER)
3453 return 0;
3455 return offset;
3458 static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset)
3460 uint32_t buffer;
3461 unsigned char hash[GIT_MAX_RAWSZ];
3463 /* offset */
3464 put_be32(&buffer, offset);
3465 strbuf_add(sb, &buffer, sizeof(uint32_t));
3467 /* hash */
3468 the_hash_algo->final_fn(hash, eoie_context);
3469 strbuf_add(sb, hash, the_hash_algo->rawsz);
3472 #ifndef NO_PTHREADS
3473 #define IEOT_VERSION (1)
3475 static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset)
3477 const char *index = NULL;
3478 uint32_t extsize, ext_version;
3479 struct index_entry_offset_table *ieot;
3480 int i, nr;
3482 /* find the IEOT extension */
3483 if (!offset)
3484 return NULL;
3485 while (offset <= mmap_size - the_hash_algo->rawsz - 8) {
3486 extsize = get_be32(mmap + offset + 4);
3487 if (CACHE_EXT((mmap + offset)) == CACHE_EXT_INDEXENTRYOFFSETTABLE) {
3488 index = mmap + offset + 4 + 4;
3489 break;
3491 offset += 8;
3492 offset += extsize;
3494 if (!index)
3495 return NULL;
3497 /* validate the version is IEOT_VERSION */
3498 ext_version = get_be32(index);
3499 if (ext_version != IEOT_VERSION) {
3500 error("invalid IEOT version %d", ext_version);
3501 return NULL;
3503 index += sizeof(uint32_t);
3505 /* extension size - version bytes / bytes per entry */
3506 nr = (extsize - sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3507 if (!nr) {
3508 error("invalid number of IEOT entries %d", nr);
3509 return NULL;
3511 ieot = xmalloc(sizeof(struct index_entry_offset_table)
3512 + (nr * sizeof(struct index_entry_offset)));
3513 ieot->nr = nr;
3514 for (i = 0; i < nr; i++) {
3515 ieot->entries[i].offset = get_be32(index);
3516 index += sizeof(uint32_t);
3517 ieot->entries[i].nr = get_be32(index);
3518 index += sizeof(uint32_t);
3521 return ieot;
3524 static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot)
3526 uint32_t buffer;
3527 int i;
3529 /* version */
3530 put_be32(&buffer, IEOT_VERSION);
3531 strbuf_add(sb, &buffer, sizeof(uint32_t));
3533 /* ieot */
3534 for (i = 0; i < ieot->nr; i++) {
3536 /* offset */
3537 put_be32(&buffer, ieot->entries[i].offset);
3538 strbuf_add(sb, &buffer, sizeof(uint32_t));
3540 /* count */
3541 put_be32(&buffer, ieot->entries[i].nr);
3542 strbuf_add(sb, &buffer, sizeof(uint32_t));
3545 #endif