Coverage Report

Created: 2026-03-21 06:46

next uncovered line (L), next uncovered region (R), next uncovered branch (B)
/src/git/read-cache.c
Line
Count
Source
1
/*
2
 * GIT - The information manager from hell
3
 *
4
 * Copyright (C) Linus Torvalds, 2005
5
 */
6
7
#define USE_THE_REPOSITORY_VARIABLE
8
#define DISABLE_SIGN_COMPARE_WARNINGS
9
10
#include "git-compat-util.h"
11
#include "config.h"
12
#include "date.h"
13
#include "diff.h"
14
#include "diffcore.h"
15
#include "hex.h"
16
#include "tempfile.h"
17
#include "lockfile.h"
18
#include "cache-tree.h"
19
#include "refs.h"
20
#include "dir.h"
21
#include "object-file.h"
22
#include "odb.h"
23
#include "oid-array.h"
24
#include "tree.h"
25
#include "commit.h"
26
#include "environment.h"
27
#include "gettext.h"
28
#include "mem-pool.h"
29
#include "name-hash.h"
30
#include "object-name.h"
31
#include "path.h"
32
#include "preload-index.h"
33
#include "read-cache.h"
34
#include "repository.h"
35
#include "resolve-undo.h"
36
#include "revision.h"
37
#include "strbuf.h"
38
#include "trace2.h"
39
#include "varint.h"
40
#include "split-index.h"
41
#include "symlinks.h"
42
#include "utf8.h"
43
#include "fsmonitor.h"
44
#include "thread-utils.h"
45
#include "progress.h"
46
#include "sparse-index.h"
47
#include "csum-file.h"
48
#include "promisor-remote.h"
49
#include "hook.h"
50
#include "submodule.h"
51
#include "submodule-config.h"
52
#include "advice.h"
53
54
/* Mask for the name length in ce_flags in the on-disk index */
55
56
0
#define CE_NAMEMASK  (0x0fff)
57
58
/* Index extensions.
59
 *
60
 * The first letter should be 'A'..'Z' for extensions that are not
61
 * necessary for a correct operation (i.e. optimization data).
62
 * When new extensions are added that _needs_ to be understood in
63
 * order to correctly interpret the index file, pick character that
64
 * is outside the range, to cause the reader to abort.
65
 */
66
67
0
#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
68
0
#define CACHE_EXT_TREE 0x54524545  /* "TREE" */
69
0
#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
70
0
#define CACHE_EXT_LINK 0x6c696e6b    /* "link" */
71
0
#define CACHE_EXT_UNTRACKED 0x554E5452    /* "UNTR" */
72
0
#define CACHE_EXT_FSMONITOR 0x46534D4E    /* "FSMN" */
73
0
#define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945  /* "EOIE" */
74
0
#define CACHE_EXT_INDEXENTRYOFFSETTABLE 0x49454F54 /* "IEOT" */
75
0
#define CACHE_EXT_SPARSE_DIRECTORIES 0x73646972 /* "sdir" */
76
77
/* changes that can be kept in $GIT_DIR/index (basically all extensions) */
78
0
#define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
79
0
     CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
80
0
     SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
81
82
83
/*
84
 * This is an estimate of the pathname length in the index.  We use
85
 * this for V4 index files to guess the un-deltafied size of the index
86
 * in memory because of pathname deltafication.  This is not required
87
 * for V2/V3 index formats because their pathnames are not compressed.
88
 * If the initial amount of memory set aside is not sufficient, the
89
 * mem pool will allocate extra memory.
90
 */
91
0
#define CACHE_ENTRY_PATH_LENGTH 80
92
93
enum index_search_mode {
94
  NO_EXPAND_SPARSE = 0,
95
  EXPAND_SPARSE = 1
96
};
97
98
static inline struct cache_entry *mem_pool__ce_alloc(struct mem_pool *mem_pool, size_t len)
99
0
{
100
0
  struct cache_entry *ce;
101
0
  ce = mem_pool_alloc(mem_pool, cache_entry_size(len));
102
0
  ce->mem_pool_allocated = 1;
103
0
  return ce;
104
0
}
105
106
static inline struct cache_entry *mem_pool__ce_calloc(struct mem_pool *mem_pool, size_t len)
107
0
{
108
0
  struct cache_entry * ce;
109
0
  ce = mem_pool_calloc(mem_pool, 1, cache_entry_size(len));
110
0
  ce->mem_pool_allocated = 1;
111
0
  return ce;
112
0
}
113
114
static struct mem_pool *find_mem_pool(struct index_state *istate)
115
0
{
116
0
  struct mem_pool **pool_ptr;
117
118
0
  if (istate->split_index && istate->split_index->base)
119
0
    pool_ptr = &istate->split_index->base->ce_mem_pool;
120
0
  else
121
0
    pool_ptr = &istate->ce_mem_pool;
122
123
0
  if (!*pool_ptr) {
124
0
    *pool_ptr = xmalloc(sizeof(**pool_ptr));
125
0
    mem_pool_init(*pool_ptr, 0);
126
0
  }
127
128
0
  return *pool_ptr;
129
0
}
130
131
static const char *alternate_index_output;
132
133
static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
134
0
{
135
0
  if (S_ISSPARSEDIR(ce->ce_mode))
136
0
    istate->sparse_index = INDEX_COLLAPSED;
137
138
0
  istate->cache[nr] = ce;
139
0
  add_name_hash(istate, ce);
140
0
}
141
142
static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
143
0
{
144
0
  struct cache_entry *old = istate->cache[nr];
145
146
0
  replace_index_entry_in_base(istate, old, ce);
147
0
  remove_name_hash(istate, old);
148
0
  discard_cache_entry(old);
149
0
  ce->ce_flags &= ~CE_HASHED;
150
0
  set_index_entry(istate, nr, ce);
151
0
  ce->ce_flags |= CE_UPDATE_IN_BASE;
152
0
  mark_fsmonitor_invalid(istate, ce);
153
0
  istate->cache_changed |= CE_ENTRY_CHANGED;
154
0
}
155
156
void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
157
0
{
158
0
  struct cache_entry *old_entry = istate->cache[nr], *new_entry, *refreshed;
159
0
  int namelen = strlen(new_name);
160
161
0
  new_entry = make_empty_cache_entry(istate, namelen);
162
0
  copy_cache_entry(new_entry, old_entry);
163
0
  new_entry->ce_flags &= ~CE_HASHED;
164
0
  new_entry->ce_namelen = namelen;
165
0
  new_entry->index = 0;
166
0
  memcpy(new_entry->name, new_name, namelen + 1);
167
168
0
  cache_tree_invalidate_path(istate, old_entry->name);
169
0
  untracked_cache_remove_from_index(istate, old_entry->name);
170
0
  remove_index_entry_at(istate, nr);
171
172
  /*
173
   * Refresh the new index entry. Using 'refresh_cache_entry' ensures
174
   * we only update stat info if the entry is otherwise up-to-date (i.e.,
175
   * the contents/mode haven't changed). This ensures that we reflect the
176
   * 'ctime' of the rename in the index without (incorrectly) updating
177
   * the cached stat info to reflect unstaged changes on disk.
178
   */
179
0
  refreshed = refresh_cache_entry(istate, new_entry, CE_MATCH_REFRESH);
180
0
  if (refreshed && refreshed != new_entry) {
181
0
    add_index_entry(istate, refreshed, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
182
0
    discard_cache_entry(new_entry);
183
0
  } else
184
0
    add_index_entry(istate, new_entry, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
185
0
}
186
187
/*
188
 * This only updates the "non-critical" parts of the directory
189
 * cache, ie the parts that aren't tracked by GIT, and only used
190
 * to validate the cache.
191
 */
192
void fill_stat_cache_info(struct index_state *istate, struct cache_entry *ce, struct stat *st)
193
0
{
194
0
  fill_stat_data(&ce->ce_stat_data, st);
195
196
0
  if (assume_unchanged)
197
0
    ce->ce_flags |= CE_VALID;
198
199
0
  if (S_ISREG(st->st_mode)) {
200
0
    ce_mark_uptodate(ce);
201
0
    mark_fsmonitor_valid(istate, ce);
202
0
  }
203
0
}
204
205
static unsigned int st_mode_from_ce(const struct cache_entry *ce)
206
0
{
207
0
  extern int trust_executable_bit, has_symlinks;
208
209
0
  switch (ce->ce_mode & S_IFMT) {
210
0
  case S_IFLNK:
211
0
    return has_symlinks ? S_IFLNK : (S_IFREG | 0644);
212
0
  case S_IFREG:
213
0
    return (ce->ce_mode & (trust_executable_bit ? 0755 : 0644)) | S_IFREG;
214
0
  case S_IFGITLINK:
215
0
    return S_IFDIR | 0755;
216
0
  case S_IFDIR:
217
0
    return ce->ce_mode;
218
0
  default:
219
0
    BUG("unsupported ce_mode: %o", ce->ce_mode);
220
0
  }
221
0
}
222
223
int fake_lstat(const struct cache_entry *ce, struct stat *st)
224
0
{
225
0
  fake_lstat_data(&ce->ce_stat_data, st);
226
0
  st->st_mode = st_mode_from_ce(ce);
227
228
  /* always succeed as lstat() replacement */
229
0
  return 0;
230
0
}
231
232
static int ce_compare_data(struct index_state *istate,
233
         const struct cache_entry *ce,
234
         struct stat *st)
235
0
{
236
0
  int match = -1;
237
0
  int fd = git_open_cloexec(ce->name, O_RDONLY);
238
239
0
  if (fd >= 0) {
240
0
    struct object_id oid;
241
0
    if (!index_fd(istate, &oid, fd, st, OBJ_BLOB, ce->name, 0))
242
0
      match = !oideq(&oid, &ce->oid);
243
    /* index_fd() closed the file descriptor already */
244
0
  }
245
0
  return match;
246
0
}
247
248
static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
249
0
{
250
0
  int match = -1;
251
0
  void *buffer;
252
0
  unsigned long size;
253
0
  enum object_type type;
254
0
  struct strbuf sb = STRBUF_INIT;
255
256
0
  if (strbuf_readlink(&sb, ce->name, expected_size))
257
0
    return -1;
258
259
0
  buffer = odb_read_object(the_repository->objects, &ce->oid, &type, &size);
260
0
  if (buffer) {
261
0
    if (size == sb.len)
262
0
      match = memcmp(buffer, sb.buf, size);
263
0
    free(buffer);
264
0
  }
265
0
  strbuf_release(&sb);
266
0
  return match;
267
0
}
268
269
static int ce_compare_gitlink(const struct cache_entry *ce)
270
0
{
271
0
  struct object_id oid;
272
273
  /*
274
   * We don't actually require that the .git directory
275
   * under GITLINK directory be a valid git directory. It
276
   * might even be missing (in case nobody populated that
277
   * sub-project).
278
   *
279
   * If so, we consider it always to match.
280
   */
281
0
  if (repo_resolve_gitlink_ref(the_repository, ce->name,
282
0
             "HEAD", &oid) < 0)
283
0
    return 0;
284
0
  return !oideq(&oid, &ce->oid);
285
0
}
286
287
static int ce_modified_check_fs(struct index_state *istate,
288
        const struct cache_entry *ce,
289
        struct stat *st)
290
0
{
291
0
  switch (st->st_mode & S_IFMT) {
292
0
  case S_IFREG:
293
0
    if (ce_compare_data(istate, ce, st))
294
0
      return DATA_CHANGED;
295
0
    break;
296
0
  case S_IFLNK:
297
0
    if (ce_compare_link(ce, xsize_t(st->st_size)))
298
0
      return DATA_CHANGED;
299
0
    break;
300
0
  case S_IFDIR:
301
0
    if (S_ISGITLINK(ce->ce_mode))
302
0
      return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
303
    /* else fallthrough */
304
0
  default:
305
0
    return TYPE_CHANGED;
306
0
  }
307
0
  return 0;
308
0
}
309
310
static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
311
0
{
312
0
  unsigned int changed = 0;
313
314
0
  if (ce->ce_flags & CE_REMOVE)
315
0
    return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
316
317
0
  switch (ce->ce_mode & S_IFMT) {
318
0
  case S_IFREG:
319
0
    changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
320
    /* We consider only the owner x bit to be relevant for
321
     * "mode changes"
322
     */
323
0
    if (trust_executable_bit &&
324
0
        (0100 & (ce->ce_mode ^ st->st_mode)))
325
0
      changed |= MODE_CHANGED;
326
0
    break;
327
0
  case S_IFLNK:
328
0
    if (!S_ISLNK(st->st_mode) &&
329
0
        (has_symlinks || !S_ISREG(st->st_mode)))
330
0
      changed |= TYPE_CHANGED;
331
0
    break;
332
0
  case S_IFGITLINK:
333
    /* We ignore most of the st_xxx fields for gitlinks */
334
0
    if (!S_ISDIR(st->st_mode))
335
0
      changed |= TYPE_CHANGED;
336
0
    else if (ce_compare_gitlink(ce))
337
0
      changed |= DATA_CHANGED;
338
0
    return changed;
339
0
  default:
340
0
    BUG("unsupported ce_mode: %o", ce->ce_mode);
341
0
  }
342
343
0
  changed |= match_stat_data(&ce->ce_stat_data, st);
344
345
  /* Racily smudged entry? */
346
0
  if (!ce->ce_stat_data.sd_size) {
347
0
    if (!is_empty_blob_oid(&ce->oid, the_repository->hash_algo))
348
0
      changed |= DATA_CHANGED;
349
0
  }
350
351
0
  return changed;
352
0
}
353
354
static int is_racy_stat(const struct index_state *istate,
355
      const struct stat_data *sd)
356
0
{
357
0
  return (istate->timestamp.sec &&
358
#ifdef USE_NSEC
359
     /* nanosecond timestamped files can also be racy! */
360
    (istate->timestamp.sec < sd->sd_mtime.sec ||
361
     (istate->timestamp.sec == sd->sd_mtime.sec &&
362
      istate->timestamp.nsec <= sd->sd_mtime.nsec))
363
#else
364
0
    istate->timestamp.sec <= sd->sd_mtime.sec
365
0
#endif
366
0
    );
367
0
}
368
369
int is_racy_timestamp(const struct index_state *istate,
370
           const struct cache_entry *ce)
371
0
{
372
0
  return (!S_ISGITLINK(ce->ce_mode) &&
373
0
    is_racy_stat(istate, &ce->ce_stat_data));
374
0
}
375
376
int match_stat_data_racy(const struct index_state *istate,
377
       const struct stat_data *sd, struct stat *st)
378
0
{
379
0
  if (is_racy_stat(istate, sd))
380
0
    return MTIME_CHANGED;
381
0
  return match_stat_data(sd, st);
382
0
}
383
384
int ie_match_stat(struct index_state *istate,
385
      const struct cache_entry *ce, struct stat *st,
386
      unsigned int options)
387
0
{
388
0
  unsigned int changed;
389
0
  int ignore_valid = options & CE_MATCH_IGNORE_VALID;
390
0
  int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
391
0
  int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
392
0
  int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
393
394
0
  if (!ignore_fsmonitor)
395
0
    refresh_fsmonitor(istate);
396
  /*
397
   * If it's marked as always valid in the index, it's
398
   * valid whatever the checked-out copy says.
399
   *
400
   * skip-worktree has the same effect with higher precedence
401
   */
402
0
  if (!ignore_skip_worktree && ce_skip_worktree(ce))
403
0
    return 0;
404
0
  if (!ignore_valid && (ce->ce_flags & CE_VALID))
405
0
    return 0;
406
0
  if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID))
407
0
    return 0;
408
409
  /*
410
   * Intent-to-add entries have not been added, so the index entry
411
   * by definition never matches what is in the work tree until it
412
   * actually gets added.
413
   */
414
0
  if (ce_intent_to_add(ce))
415
0
    return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
416
417
0
  changed = ce_match_stat_basic(ce, st);
418
419
  /*
420
   * Within 1 second of this sequence:
421
   *  echo xyzzy >file && git-update-index --add file
422
   * running this command:
423
   *  echo frotz >file
424
   * would give a falsely clean cache entry.  The mtime and
425
   * length match the cache, and other stat fields do not change.
426
   *
427
   * We could detect this at update-index time (the cache entry
428
   * being registered/updated records the same time as "now")
429
   * and delay the return from git-update-index, but that would
430
   * effectively mean we can make at most one commit per second,
431
   * which is not acceptable.  Instead, we check cache entries
432
   * whose mtime are the same as the index file timestamp more
433
   * carefully than others.
434
   */
435
0
  if (!changed && is_racy_timestamp(istate, ce)) {
436
0
    if (assume_racy_is_modified)
437
0
      changed |= DATA_CHANGED;
438
0
    else
439
0
      changed |= ce_modified_check_fs(istate, ce, st);
440
0
  }
441
442
0
  return changed;
443
0
}
444
445
int ie_modified(struct index_state *istate,
446
    const struct cache_entry *ce,
447
    struct stat *st, unsigned int options)
448
0
{
449
0
  int changed, changed_fs;
450
451
0
  changed = ie_match_stat(istate, ce, st, options);
452
0
  if (!changed)
453
0
    return 0;
454
  /*
455
   * If the mode or type has changed, there's no point in trying
456
   * to refresh the entry - it's not going to match
457
   */
458
0
  if (changed & (MODE_CHANGED | TYPE_CHANGED))
459
0
    return changed;
460
461
  /*
462
   * Immediately after read-tree or update-index --cacheinfo,
463
   * the length field is zero, as we have never even read the
464
   * lstat(2) information once, and we cannot trust DATA_CHANGED
465
   * returned by ie_match_stat() which in turn was returned by
466
   * ce_match_stat_basic() to signal that the filesize of the
467
   * blob changed.  We have to actually go to the filesystem to
468
   * see if the contents match, and if so, should answer "unchanged".
469
   *
470
   * The logic does not apply to gitlinks, as ce_match_stat_basic()
471
   * already has checked the actual HEAD from the filesystem in the
472
   * subproject.  If ie_match_stat() already said it is different,
473
   * then we know it is.
474
   */
475
0
  if ((changed & DATA_CHANGED) &&
476
#ifdef GIT_WINDOWS_NATIVE
477
      /*
478
       * Work around Git for Windows v2.27.0 fixing a bug where symlinks'
479
       * target path lengths were not read at all, and instead recorded
480
       * as 4096: now, all symlinks would appear as modified.
481
       *
482
       * So let's just special-case symlinks with a target path length
483
       * (i.e. `sd_size`) of 4096 and force them to be re-checked.
484
       */
485
      (!S_ISLNK(st->st_mode) || ce->ce_stat_data.sd_size != MAX_PATH) &&
486
#endif
487
0
      (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
488
0
    return changed;
489
490
0
  changed_fs = ce_modified_check_fs(istate, ce, st);
491
0
  if (changed_fs)
492
0
    return changed | changed_fs;
493
0
  return 0;
494
0
}
495
496
static int cache_name_stage_compare(const char *name1, int len1, int stage1,
497
            const char *name2, int len2, int stage2)
498
0
{
499
0
  int cmp;
500
501
0
  cmp = name_compare(name1, len1, name2, len2);
502
0
  if (cmp)
503
0
    return cmp;
504
505
0
  if (stage1 < stage2)
506
0
    return -1;
507
0
  if (stage1 > stage2)
508
0
    return 1;
509
0
  return 0;
510
0
}
511
512
int cmp_cache_name_compare(const void *a_, const void *b_)
513
0
{
514
0
  const struct cache_entry *ce1, *ce2;
515
516
0
  ce1 = *((const struct cache_entry **)a_);
517
0
  ce2 = *((const struct cache_entry **)b_);
518
0
  return cache_name_stage_compare(ce1->name, ce1->ce_namelen, ce_stage(ce1),
519
0
          ce2->name, ce2->ce_namelen, ce_stage(ce2));
520
0
}
521
522
static int index_name_stage_pos(struct index_state *istate,
523
        const char *name, int namelen,
524
        int stage,
525
        enum index_search_mode search_mode)
526
0
{
527
0
  int first, last;
528
529
0
  first = 0;
530
0
  last = istate->cache_nr;
531
0
  while (last > first) {
532
0
    int next = first + ((last - first) >> 1);
533
0
    struct cache_entry *ce = istate->cache[next];
534
0
    int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
535
0
    if (!cmp)
536
0
      return next;
537
0
    if (cmp < 0) {
538
0
      last = next;
539
0
      continue;
540
0
    }
541
0
    first = next+1;
542
0
  }
543
544
0
  if (search_mode == EXPAND_SPARSE && istate->sparse_index &&
545
0
      first > 0) {
546
    /* Note: first <= istate->cache_nr */
547
0
    struct cache_entry *ce = istate->cache[first - 1];
548
549
    /*
550
     * If we are in a sparse-index _and_ the entry before the
551
     * insertion position is a sparse-directory entry that is
552
     * an ancestor of 'name', then we need to expand the index
553
     * and search again. This will only trigger once, because
554
     * thereafter the index is fully expanded.
555
     */
556
0
    if (S_ISSPARSEDIR(ce->ce_mode) &&
557
0
        ce_namelen(ce) < namelen &&
558
0
        !strncmp(name, ce->name, ce_namelen(ce))) {
559
0
      ensure_full_index(istate);
560
0
      return index_name_stage_pos(istate, name, namelen, stage, search_mode);
561
0
    }
562
0
  }
563
564
0
  return -first-1;
565
0
}
566
567
int index_name_pos(struct index_state *istate, const char *name, int namelen)
568
0
{
569
0
  return index_name_stage_pos(istate, name, namelen, 0, EXPAND_SPARSE);
570
0
}
571
572
int index_name_pos_sparse(struct index_state *istate, const char *name, int namelen)
573
0
{
574
0
  return index_name_stage_pos(istate, name, namelen, 0, NO_EXPAND_SPARSE);
575
0
}
576
577
int index_entry_exists(struct index_state *istate, const char *name, int namelen)
578
0
{
579
0
  return index_name_stage_pos(istate, name, namelen, 0, NO_EXPAND_SPARSE) >= 0;
580
0
}
581
582
int remove_index_entry_at(struct index_state *istate, int pos)
583
0
{
584
0
  struct cache_entry *ce = istate->cache[pos];
585
586
0
  record_resolve_undo(istate, ce);
587
0
  remove_name_hash(istate, ce);
588
0
  save_or_free_index_entry(istate, ce);
589
0
  istate->cache_changed |= CE_ENTRY_REMOVED;
590
0
  istate->cache_nr--;
591
0
  if (pos >= istate->cache_nr)
592
0
    return 0;
593
0
  MOVE_ARRAY(istate->cache + pos, istate->cache + pos + 1,
594
0
       istate->cache_nr - pos);
595
0
  return 1;
596
0
}
597
598
/*
599
 * Remove all cache entries marked for removal, that is where
600
 * CE_REMOVE is set in ce_flags.  This is much more effective than
601
 * calling remove_index_entry_at() for each entry to be removed.
602
 */
603
void remove_marked_cache_entries(struct index_state *istate, int invalidate)
604
0
{
605
0
  struct cache_entry **ce_array = istate->cache;
606
0
  unsigned int i, j;
607
608
0
  for (i = j = 0; i < istate->cache_nr; i++) {
609
0
    if (ce_array[i]->ce_flags & CE_REMOVE) {
610
0
      if (invalidate) {
611
0
        cache_tree_invalidate_path(istate,
612
0
                 ce_array[i]->name);
613
0
        untracked_cache_remove_from_index(istate,
614
0
                  ce_array[i]->name);
615
0
      }
616
0
      remove_name_hash(istate, ce_array[i]);
617
0
      save_or_free_index_entry(istate, ce_array[i]);
618
0
    }
619
0
    else
620
0
      ce_array[j++] = ce_array[i];
621
0
  }
622
0
  if (j == istate->cache_nr)
623
0
    return;
624
0
  istate->cache_changed |= CE_ENTRY_REMOVED;
625
0
  istate->cache_nr = j;
626
0
}
627
628
int remove_file_from_index(struct index_state *istate, const char *path)
629
0
{
630
0
  int pos = index_name_pos(istate, path, strlen(path));
631
0
  if (pos < 0)
632
0
    pos = -pos-1;
633
0
  cache_tree_invalidate_path(istate, path);
634
0
  untracked_cache_remove_from_index(istate, path);
635
0
  while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
636
0
    remove_index_entry_at(istate, pos);
637
0
  return 0;
638
0
}
639
640
static int compare_name(struct cache_entry *ce, const char *path, int namelen)
641
0
{
642
0
  return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
643
0
}
644
645
static int index_name_pos_also_unmerged(struct index_state *istate,
646
  const char *path, int namelen)
647
0
{
648
0
  int pos = index_name_pos(istate, path, namelen);
649
0
  struct cache_entry *ce;
650
651
0
  if (pos >= 0)
652
0
    return pos;
653
654
  /* maybe unmerged? */
655
0
  pos = -1 - pos;
656
0
  if (pos >= istate->cache_nr ||
657
0
      compare_name((ce = istate->cache[pos]), path, namelen))
658
0
    return -1;
659
660
  /* order of preference: stage 2, 1, 3 */
661
0
  if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
662
0
      ce_stage((ce = istate->cache[pos + 1])) == 2 &&
663
0
      !compare_name(ce, path, namelen))
664
0
    pos++;
665
0
  return pos;
666
0
}
667
668
static int different_name(struct cache_entry *ce, struct cache_entry *alias)
669
0
{
670
0
  int len = ce_namelen(ce);
671
0
  return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
672
0
}
673
674
/*
675
 * If we add a filename that aliases in the cache, we will use the
676
 * name that we already have - but we don't want to update the same
677
 * alias twice, because that implies that there were actually two
678
 * different files with aliasing names!
679
 *
680
 * So we use the CE_ADDED flag to verify that the alias was an old
681
 * one before we accept it as
682
 */
683
static struct cache_entry *create_alias_ce(struct index_state *istate,
684
             struct cache_entry *ce,
685
             struct cache_entry *alias)
686
0
{
687
0
  int len;
688
0
  struct cache_entry *new_entry;
689
690
0
  if (alias->ce_flags & CE_ADDED)
691
0
    die(_("will not add file alias '%s' ('%s' already exists in index)"),
692
0
        ce->name, alias->name);
693
694
  /* Ok, create the new entry using the name of the existing alias */
695
0
  len = ce_namelen(alias);
696
0
  new_entry = make_empty_cache_entry(istate, len);
697
0
  memcpy(new_entry->name, alias->name, len);
698
0
  copy_cache_entry(new_entry, ce);
699
0
  save_or_free_index_entry(istate, ce);
700
0
  return new_entry;
701
0
}
702
703
void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
704
0
{
705
0
  struct object_id oid;
706
0
  if (odb_write_object(the_repository->objects, "", 0, OBJ_BLOB, &oid))
707
0
    die(_("cannot create an empty blob in the object database"));
708
0
  oidcpy(&ce->oid, &oid);
709
0
}
710
711
int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
712
0
{
713
0
  int namelen, was_same;
714
0
  mode_t st_mode = st->st_mode;
715
0
  struct cache_entry *ce, *alias = NULL;
716
0
  unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
717
0
  int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
718
0
  int pretend = flags & ADD_CACHE_PRETEND;
719
0
  int intent_only = flags & ADD_CACHE_INTENT;
720
0
  int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
721
0
        (intent_only ? ADD_CACHE_NEW_ONLY : 0));
722
0
  unsigned hash_flags = pretend ? 0 : INDEX_WRITE_OBJECT;
723
724
0
  if (flags & ADD_CACHE_RENORMALIZE)
725
0
    hash_flags |= INDEX_RENORMALIZE;
726
727
0
  if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
728
0
    return error(_("%s: can only add regular files, symbolic links or git-directories"), path);
729
730
0
  namelen = strlen(path);
731
0
  if (S_ISDIR(st_mode)) {
732
0
    while (namelen && path[namelen-1] == '/')
733
0
      namelen--;
734
0
  }
735
0
  ce = make_empty_cache_entry(istate, namelen);
736
0
  memcpy(ce->name, path, namelen);
737
0
  ce->ce_namelen = namelen;
738
0
  if (!intent_only)
739
0
    fill_stat_cache_info(istate, ce, st);
740
0
  else
741
0
    ce->ce_flags |= CE_INTENT_TO_ADD;
742
743
744
0
  if (trust_executable_bit && has_symlinks) {
745
0
    ce->ce_mode = create_ce_mode(st_mode);
746
0
  } else {
747
    /* If there is an existing entry, pick the mode bits and type
748
     * from it, otherwise assume unexecutable regular file.
749
     */
750
0
    struct cache_entry *ent;
751
0
    int pos = index_name_pos_also_unmerged(istate, path, namelen);
752
753
0
    ent = (0 <= pos) ? istate->cache[pos] : NULL;
754
0
    ce->ce_mode = ce_mode_from_stat(ent, st_mode);
755
0
  }
756
757
  /* When core.ignorecase=true, determine if a directory of the same name but differing
758
   * case already exists within the Git repository.  If it does, ensure the directory
759
   * case of the file being added to the repository matches (is folded into) the existing
760
   * entry's directory case.
761
   */
762
0
  if (ignore_case) {
763
0
    adjust_dirname_case(istate, ce->name);
764
0
  }
765
0
  if (!(flags & ADD_CACHE_RENORMALIZE)) {
766
0
    alias = index_file_exists(istate, ce->name,
767
0
            ce_namelen(ce), ignore_case);
768
0
    if (alias &&
769
0
        !ce_stage(alias) &&
770
0
        !ie_match_stat(istate, alias, st, ce_option)) {
771
      /* Nothing changed, really */
772
0
      if (!S_ISGITLINK(alias->ce_mode))
773
0
        ce_mark_uptodate(alias);
774
0
      alias->ce_flags |= CE_ADDED;
775
776
0
      discard_cache_entry(ce);
777
0
      return 0;
778
0
    }
779
0
  }
780
0
  if (!intent_only) {
781
0
    if (index_path(istate, &ce->oid, path, st, hash_flags)) {
782
0
      discard_cache_entry(ce);
783
0
      return error(_("unable to index file '%s'"), path);
784
0
    }
785
0
  } else
786
0
    set_object_name_for_intent_to_add_entry(ce);
787
788
0
  if (ignore_case && alias && different_name(ce, alias))
789
0
    ce = create_alias_ce(istate, ce, alias);
790
0
  ce->ce_flags |= CE_ADDED;
791
792
  /* It was suspected to be racily clean, but it turns out to be Ok */
793
0
  was_same = (alias &&
794
0
        !ce_stage(alias) &&
795
0
        oideq(&alias->oid, &ce->oid) &&
796
0
        ce->ce_mode == alias->ce_mode);
797
798
0
  if (pretend)
799
0
    discard_cache_entry(ce);
800
0
  else if (add_index_entry(istate, ce, add_option)) {
801
0
    discard_cache_entry(ce);
802
0
    return error(_("unable to add '%s' to index"), path);
803
0
  }
804
0
  if (verbose && !was_same)
805
0
    printf("add '%s'\n", path);
806
0
  return 0;
807
0
}
808
809
int add_file_to_index(struct index_state *istate, const char *path, int flags)
810
0
{
811
0
  struct stat st;
812
0
  if (lstat(path, &st))
813
0
    die_errno(_("unable to stat '%s'"), path);
814
0
  return add_to_index(istate, path, &st, flags);
815
0
}
816
817
struct cache_entry *make_empty_cache_entry(struct index_state *istate, size_t len)
818
0
{
819
0
  return mem_pool__ce_calloc(find_mem_pool(istate), len);
820
0
}
821
822
struct cache_entry *make_empty_transient_cache_entry(size_t len,
823
                 struct mem_pool *ce_mem_pool)
824
0
{
825
0
  if (ce_mem_pool)
826
0
    return mem_pool__ce_calloc(ce_mem_pool, len);
827
0
  return xcalloc(1, cache_entry_size(len));
828
0
}
829
830
enum verify_path_result {
831
  PATH_OK,
832
  PATH_INVALID,
833
  PATH_DIR_WITH_SEP,
834
};
835
836
static enum verify_path_result verify_path_internal(const char *, unsigned);
837
838
int verify_path(const char *path, unsigned mode)
839
0
{
840
0
  return verify_path_internal(path, mode) == PATH_OK;
841
0
}
842
843
struct cache_entry *make_cache_entry(struct index_state *istate,
844
             unsigned int mode,
845
             const struct object_id *oid,
846
             const char *path,
847
             int stage,
848
             unsigned int refresh_options)
849
0
{
850
0
  struct cache_entry *ce, *ret;
851
0
  int len;
852
853
0
  if (verify_path_internal(path, mode) == PATH_INVALID) {
854
0
    error(_("invalid path '%s'"), path);
855
0
    return NULL;
856
0
  }
857
858
0
  len = strlen(path);
859
0
  ce = make_empty_cache_entry(istate, len);
860
861
0
  oidcpy(&ce->oid, oid);
862
0
  memcpy(ce->name, path, len);
863
0
  ce->ce_flags = create_ce_flags(stage);
864
0
  ce->ce_namelen = len;
865
0
  ce->ce_mode = create_ce_mode(mode);
866
867
0
  ret = refresh_cache_entry(istate, ce, refresh_options);
868
0
  if (ret != ce)
869
0
    discard_cache_entry(ce);
870
0
  return ret;
871
0
}
872
873
struct cache_entry *make_transient_cache_entry(unsigned int mode,
874
                 const struct object_id *oid,
875
                 const char *path,
876
                 int stage,
877
                 struct mem_pool *ce_mem_pool)
878
0
{
879
0
  struct cache_entry *ce;
880
0
  int len;
881
882
0
  if (!verify_path(path, mode)) {
883
0
    error(_("invalid path '%s'"), path);
884
0
    return NULL;
885
0
  }
886
887
0
  len = strlen(path);
888
0
  ce = make_empty_transient_cache_entry(len, ce_mem_pool);
889
890
0
  oidcpy(&ce->oid, oid);
891
0
  memcpy(ce->name, path, len);
892
0
  ce->ce_flags = create_ce_flags(stage);
893
0
  ce->ce_namelen = len;
894
0
  ce->ce_mode = create_ce_mode(mode);
895
896
0
  return ce;
897
0
}
898
899
/*
900
 * Chmod an index entry with either +x or -x.
901
 *
902
 * Returns -1 if the chmod for the particular cache entry failed (if it's
903
 * not a regular file), -2 if an invalid flip argument is passed in, 0
904
 * otherwise.
905
 */
906
int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
907
          char flip)
908
0
{
909
0
  if (!S_ISREG(ce->ce_mode))
910
0
    return -1;
911
0
  switch (flip) {
912
0
  case '+':
913
0
    ce->ce_mode |= 0111;
914
0
    break;
915
0
  case '-':
916
0
    ce->ce_mode &= ~0111;
917
0
    break;
918
0
  default:
919
0
    return -2;
920
0
  }
921
0
  cache_tree_invalidate_path(istate, ce->name);
922
0
  ce->ce_flags |= CE_UPDATE_IN_BASE;
923
0
  mark_fsmonitor_invalid(istate, ce);
924
0
  istate->cache_changed |= CE_ENTRY_CHANGED;
925
926
0
  return 0;
927
0
}
928
929
int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
930
0
{
931
0
  int len = ce_namelen(a);
932
0
  return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
933
0
}
934
935
/*
936
 * We fundamentally don't like some paths: we don't want
937
 * dot or dot-dot anywhere, and for obvious reasons don't
938
 * want to recurse into ".git" either.
939
 *
940
 * Also, we don't want double slashes or slashes at the
941
 * end that can make pathnames ambiguous.
942
 */
943
static int verify_dotfile(const char *rest, unsigned mode)
944
0
{
945
  /*
946
   * The first character was '.', but that
947
   * has already been discarded, we now test
948
   * the rest.
949
   */
950
951
  /* "." is not allowed */
952
0
  if (*rest == '\0' || is_dir_sep(*rest))
953
0
    return 0;
954
955
0
  switch (*rest) {
956
  /*
957
   * ".git" followed by NUL or slash is bad. Note that we match
958
   * case-insensitively here, even if ignore_case is not set.
959
   * This outlaws ".GIT" everywhere out of an abundance of caution,
960
   * since there's really no good reason to allow it.
961
   *
962
   * Once we've seen ".git", we can also find ".gitmodules", etc (also
963
   * case-insensitively).
964
   */
965
0
  case 'g':
966
0
  case 'G':
967
0
    if (rest[1] != 'i' && rest[1] != 'I')
968
0
      break;
969
0
    if (rest[2] != 't' && rest[2] != 'T')
970
0
      break;
971
0
    if (rest[3] == '\0' || is_dir_sep(rest[3]))
972
0
      return 0;
973
0
    if (S_ISLNK(mode)) {
974
0
      rest += 3;
975
0
      if (skip_iprefix(rest, "modules", &rest) &&
976
0
          (*rest == '\0' || is_dir_sep(*rest)))
977
0
        return 0;
978
0
    }
979
0
    break;
980
0
  case '.':
981
0
    if (rest[1] == '\0' || is_dir_sep(rest[1]))
982
0
      return 0;
983
0
  }
984
0
  return 1;
985
0
}
986
987
static enum verify_path_result verify_path_internal(const char *path,
988
                unsigned mode)
989
0
{
990
0
  char c = 0;
991
992
0
  if (has_dos_drive_prefix(path))
993
0
    return PATH_INVALID;
994
995
0
  if (!is_valid_path(path))
996
0
    return PATH_INVALID;
997
998
0
  goto inside;
999
0
  for (;;) {
1000
0
    if (!c)
1001
0
      return PATH_OK;
1002
0
    if (is_dir_sep(c)) {
1003
0
inside:
1004
0
      if (protect_hfs) {
1005
1006
0
        if (is_hfs_dotgit(path))
1007
0
          return PATH_INVALID;
1008
0
        if (S_ISLNK(mode)) {
1009
0
          if (is_hfs_dotgitmodules(path))
1010
0
            return PATH_INVALID;
1011
0
        }
1012
0
      }
1013
0
      if (protect_ntfs) {
1014
#if defined GIT_WINDOWS_NATIVE || defined __CYGWIN__
1015
        if (c == '\\')
1016
          return PATH_INVALID;
1017
#endif
1018
0
        if (is_ntfs_dotgit(path))
1019
0
          return PATH_INVALID;
1020
0
        if (S_ISLNK(mode)) {
1021
0
          if (is_ntfs_dotgitmodules(path))
1022
0
            return PATH_INVALID;
1023
0
        }
1024
0
      }
1025
1026
0
      c = *path++;
1027
0
      if ((c == '.' && !verify_dotfile(path, mode)) ||
1028
0
          is_dir_sep(c))
1029
0
        return PATH_INVALID;
1030
      /*
1031
       * allow terminating directory separators for
1032
       * sparse directory entries.
1033
       */
1034
0
      if (c == '\0')
1035
0
        return S_ISDIR(mode) ? PATH_DIR_WITH_SEP :
1036
0
                   PATH_INVALID;
1037
0
    } else if (c == '\\' && protect_ntfs) {
1038
0
      if (is_ntfs_dotgit(path))
1039
0
        return PATH_INVALID;
1040
0
      if (S_ISLNK(mode)) {
1041
0
        if (is_ntfs_dotgitmodules(path))
1042
0
          return PATH_INVALID;
1043
0
      }
1044
0
    }
1045
1046
0
    c = *path++;
1047
0
  }
1048
0
}
1049
1050
/*
1051
 * Do we have another file that has the beginning components being a
1052
 * proper superset of the name we're trying to add?
1053
 */
1054
static int has_file_name(struct index_state *istate,
1055
       const struct cache_entry *ce, int pos, int ok_to_replace)
1056
0
{
1057
0
  int retval = 0;
1058
0
  int len = ce_namelen(ce);
1059
0
  int stage = ce_stage(ce);
1060
0
  const char *name = ce->name;
1061
1062
0
  while (pos < istate->cache_nr) {
1063
0
    struct cache_entry *p = istate->cache[pos++];
1064
1065
0
    if (len >= ce_namelen(p))
1066
0
      break;
1067
0
    if (memcmp(name, p->name, len))
1068
0
      break;
1069
0
    if (ce_stage(p) != stage)
1070
0
      continue;
1071
0
    if (p->name[len] != '/')
1072
0
      continue;
1073
0
    if (p->ce_flags & CE_REMOVE)
1074
0
      continue;
1075
0
    retval = -1;
1076
0
    if (!ok_to_replace)
1077
0
      break;
1078
0
    remove_index_entry_at(istate, --pos);
1079
0
  }
1080
0
  return retval;
1081
0
}
1082
1083
1084
/*
1085
 * Like strcmp(), but also return the offset of the first change.
1086
 * If strings are equal, return the length.
1087
 */
1088
int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
1089
0
{
1090
0
  size_t k;
1091
1092
0
  if (!first_change)
1093
0
    return strcmp(s1, s2);
1094
1095
0
  for (k = 0; s1[k] == s2[k]; k++)
1096
0
    if (s1[k] == '\0')
1097
0
      break;
1098
1099
0
  *first_change = k;
1100
0
  return (unsigned char)s1[k] - (unsigned char)s2[k];
1101
0
}
1102
1103
/*
1104
 * Do we have another file with a pathname that is a proper
1105
 * subset of the name we're trying to add?
1106
 *
1107
 * That is, is there another file in the index with a path
1108
 * that matches a sub-directory in the given entry?
1109
 */
1110
static int has_dir_name(struct index_state *istate,
1111
      const struct cache_entry *ce, int pos, int ok_to_replace)
1112
0
{
1113
0
  int retval = 0;
1114
0
  int stage = ce_stage(ce);
1115
0
  const char *name = ce->name;
1116
0
  const char *slash = name + ce_namelen(ce);
1117
0
  size_t len_eq_last;
1118
0
  int cmp_last = 0;
1119
1120
  /*
1121
   * We are frequently called during an iteration on a sorted
1122
   * list of pathnames and while building a new index.  Therefore,
1123
   * there is a high probability that this entry will eventually
1124
   * be appended to the index, rather than inserted in the middle.
1125
   * If we can confirm that, we can avoid binary searches on the
1126
   * components of the pathname.
1127
   *
1128
   * Compare the entry's full path with the last path in the index.
1129
   */
1130
0
  if (!istate->cache_nr)
1131
0
    return 0;
1132
1133
0
  cmp_last = strcmp_offset(name,
1134
0
         istate->cache[istate->cache_nr - 1]->name,
1135
0
         &len_eq_last);
1136
0
  if (cmp_last > 0 && name[len_eq_last] != '/')
1137
    /*
1138
     * The entry sorts AFTER the last one in the
1139
     * index and their paths have no common prefix,
1140
     * so there cannot be a F/D conflict.
1141
     */
1142
0
    return 0;
1143
1144
0
  for (;;) {
1145
0
    size_t len;
1146
1147
0
    for (;;) {
1148
0
      if (*--slash == '/')
1149
0
        break;
1150
0
      if (slash <= ce->name)
1151
0
        return retval;
1152
0
    }
1153
0
    len = slash - name;
1154
1155
0
    pos = index_name_stage_pos(istate, name, len, stage, EXPAND_SPARSE);
1156
0
    if (pos >= 0) {
1157
      /*
1158
       * Found one, but not so fast.  This could
1159
       * be a marker that says "I was here, but
1160
       * I am being removed".  Such an entry is
1161
       * not a part of the resulting tree, and
1162
       * it is Ok to have a directory at the same
1163
       * path.
1164
       */
1165
0
      if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
1166
0
        retval = -1;
1167
0
        if (!ok_to_replace)
1168
0
          break;
1169
0
        remove_index_entry_at(istate, pos);
1170
0
        continue;
1171
0
      }
1172
0
    }
1173
0
    else
1174
0
      pos = -pos-1;
1175
1176
    /*
1177
     * Trivial optimization: if we find an entry that
1178
     * already matches the sub-directory, then we know
1179
     * we're ok, and we can exit.
1180
     */
1181
0
    while (pos < istate->cache_nr) {
1182
0
      struct cache_entry *p = istate->cache[pos];
1183
0
      if ((ce_namelen(p) <= len) ||
1184
0
          (p->name[len] != '/') ||
1185
0
          memcmp(p->name, name, len))
1186
0
        break; /* not our subdirectory */
1187
0
      if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1188
        /*
1189
         * p is at the same stage as our entry, and
1190
         * is a subdirectory of what we are looking
1191
         * at, so we cannot have conflicts at our
1192
         * level or anything shorter.
1193
         */
1194
0
        return retval;
1195
0
      pos++;
1196
0
    }
1197
0
  }
1198
0
  return retval;
1199
0
}
1200
1201
/* We may be in a situation where we already have path/file and path
1202
 * is being added, or we already have path and path/file is being
1203
 * added.  Either one would result in a nonsense tree that has path
1204
 * twice when git-write-tree tries to write it out.  Prevent it.
1205
 *
1206
 * If ok-to-replace is specified, we remove the conflicting entries
1207
 * from the cache so the caller should recompute the insert position.
1208
 * When this happens, we return non-zero.
1209
 */
1210
static int check_file_directory_conflict(struct index_state *istate,
1211
           const struct cache_entry *ce,
1212
           int pos, int ok_to_replace)
1213
0
{
1214
0
  int retval;
1215
1216
  /*
1217
   * When ce is an "I am going away" entry, we allow it to be added
1218
   */
1219
0
  if (ce->ce_flags & CE_REMOVE)
1220
0
    return 0;
1221
1222
  /*
1223
   * We check if the path is a sub-path of a subsequent pathname
1224
   * first, since removing those will not change the position
1225
   * in the array.
1226
   */
1227
0
  retval = has_file_name(istate, ce, pos, ok_to_replace);
1228
1229
  /*
1230
   * Then check if the path might have a clashing sub-directory
1231
   * before it.
1232
   */
1233
0
  return retval + has_dir_name(istate, ce, pos, ok_to_replace);
1234
0
}
1235
1236
static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
1237
0
{
1238
0
  int pos;
1239
0
  int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1240
0
  int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1241
0
  int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1242
0
  int new_only = option & ADD_CACHE_NEW_ONLY;
1243
1244
  /*
1245
   * If this entry's path sorts after the last entry in the index,
1246
   * we can avoid searching for it.
1247
   */
1248
0
  if (istate->cache_nr > 0 &&
1249
0
    strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1250
0
    pos = index_pos_to_insert_pos(istate->cache_nr);
1251
0
  else
1252
0
    pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce), EXPAND_SPARSE);
1253
1254
  /*
1255
   * Cache tree path should be invalidated only after index_name_stage_pos,
1256
   * in case it expands a sparse index.
1257
   */
1258
0
  if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1259
0
    cache_tree_invalidate_path(istate, ce->name);
1260
1261
  /* existing match? Just replace it. */
1262
0
  if (pos >= 0) {
1263
0
    if (!new_only)
1264
0
      replace_index_entry(istate, pos, ce);
1265
0
    return 0;
1266
0
  }
1267
0
  pos = -pos-1;
1268
1269
0
  if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1270
0
    untracked_cache_add_to_index(istate, ce->name);
1271
1272
  /*
1273
   * Inserting a merged entry ("stage 0") into the index
1274
   * will always replace all non-merged entries..
1275
   */
1276
0
  if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1277
0
    while (ce_same_name(istate->cache[pos], ce)) {
1278
0
      ok_to_add = 1;
1279
0
      if (!remove_index_entry_at(istate, pos))
1280
0
        break;
1281
0
    }
1282
0
  }
1283
1284
0
  if (!ok_to_add)
1285
0
    return -1;
1286
0
  if (verify_path_internal(ce->name, ce->ce_mode) == PATH_INVALID)
1287
0
    return error(_("invalid path '%s'"), ce->name);
1288
1289
0
  if (!skip_df_check &&
1290
0
      check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1291
0
    if (!ok_to_replace)
1292
0
      return error(_("'%s' appears as both a file and as a directory"),
1293
0
             ce->name);
1294
0
    pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce), EXPAND_SPARSE);
1295
0
    pos = -pos-1;
1296
0
  }
1297
0
  return pos + 1;
1298
0
}
1299
1300
int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1301
0
{
1302
0
  int pos;
1303
1304
0
  if (option & ADD_CACHE_JUST_APPEND)
1305
0
    pos = istate->cache_nr;
1306
0
  else {
1307
0
    int ret;
1308
0
    ret = add_index_entry_with_check(istate, ce, option);
1309
0
    if (ret <= 0)
1310
0
      return ret;
1311
0
    pos = ret - 1;
1312
0
  }
1313
1314
  /* Make sure the array is big enough .. */
1315
0
  ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1316
1317
  /* Add it in.. */
1318
0
  istate->cache_nr++;
1319
0
  if (istate->cache_nr > pos + 1)
1320
0
    MOVE_ARRAY(istate->cache + pos + 1, istate->cache + pos,
1321
0
         istate->cache_nr - pos - 1);
1322
0
  set_index_entry(istate, pos, ce);
1323
0
  istate->cache_changed |= CE_ENTRY_ADDED;
1324
0
  return 0;
1325
0
}
1326
1327
/*
1328
 * "refresh" does not calculate a new sha1 file or bring the
1329
 * cache up-to-date for mode/content changes. But what it
1330
 * _does_ do is to "re-match" the stat information of a file
1331
 * with the cache, so that you can refresh the cache for a
1332
 * file that hasn't been changed but where the stat entry is
1333
 * out of date.
1334
 *
1335
 * For example, you'd want to do this after doing a "git-read-tree",
1336
 * to link up the stat cache details with the proper files.
1337
 */
1338
static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1339
               struct cache_entry *ce,
1340
               unsigned int options, int *err,
1341
               int *changed_ret,
1342
               int *t2_did_lstat,
1343
               int *t2_did_scan)
1344
0
{
1345
0
  struct stat st;
1346
0
  struct cache_entry *updated;
1347
0
  int changed;
1348
0
  int refresh = options & CE_MATCH_REFRESH;
1349
0
  int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1350
0
  int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1351
0
  int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1352
0
  int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
1353
1354
0
  if (!refresh || ce_uptodate(ce))
1355
0
    return ce;
1356
1357
0
  if (!ignore_fsmonitor)
1358
0
    refresh_fsmonitor(istate);
1359
  /*
1360
   * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1361
   * that the change to the work tree does not matter and told
1362
   * us not to worry.
1363
   */
1364
0
  if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1365
0
    ce_mark_uptodate(ce);
1366
0
    return ce;
1367
0
  }
1368
0
  if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1369
0
    ce_mark_uptodate(ce);
1370
0
    return ce;
1371
0
  }
1372
0
  if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {
1373
0
    ce_mark_uptodate(ce);
1374
0
    return ce;
1375
0
  }
1376
1377
0
  if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1378
0
    if (ignore_missing)
1379
0
      return ce;
1380
0
    if (err)
1381
0
      *err = ENOENT;
1382
0
    return NULL;
1383
0
  }
1384
1385
0
  if (t2_did_lstat)
1386
0
    *t2_did_lstat = 1;
1387
0
  if (lstat(ce->name, &st) < 0) {
1388
0
    if (ignore_missing && errno == ENOENT)
1389
0
      return ce;
1390
0
    if (err)
1391
0
      *err = errno;
1392
0
    return NULL;
1393
0
  }
1394
1395
0
  changed = ie_match_stat(istate, ce, &st, options);
1396
0
  if (changed_ret)
1397
0
    *changed_ret = changed;
1398
0
  if (!changed) {
1399
    /*
1400
     * The path is unchanged.  If we were told to ignore
1401
     * valid bit, then we did the actual stat check and
1402
     * found that the entry is unmodified.  If the entry
1403
     * is not marked VALID, this is the place to mark it
1404
     * valid again, under "assume unchanged" mode.
1405
     */
1406
0
    if (ignore_valid && assume_unchanged &&
1407
0
        !(ce->ce_flags & CE_VALID))
1408
0
      ; /* mark this one VALID again */
1409
0
    else {
1410
      /*
1411
       * We do not mark the index itself "modified"
1412
       * because CE_UPTODATE flag is in-core only;
1413
       * we are not going to write this change out.
1414
       */
1415
0
      if (!S_ISGITLINK(ce->ce_mode)) {
1416
0
        ce_mark_uptodate(ce);
1417
0
        mark_fsmonitor_valid(istate, ce);
1418
0
      }
1419
0
      return ce;
1420
0
    }
1421
0
  }
1422
1423
0
  if (t2_did_scan)
1424
0
    *t2_did_scan = 1;
1425
0
  if (ie_modified(istate, ce, &st, options)) {
1426
0
    if (err)
1427
0
      *err = EINVAL;
1428
0
    return NULL;
1429
0
  }
1430
1431
0
  updated = make_empty_cache_entry(istate, ce_namelen(ce));
1432
0
  copy_cache_entry(updated, ce);
1433
0
  memcpy(updated->name, ce->name, ce->ce_namelen + 1);
1434
0
  fill_stat_cache_info(istate, updated, &st);
1435
  /*
1436
   * If ignore_valid is not set, we should leave CE_VALID bit
1437
   * alone.  Otherwise, paths marked with --no-assume-unchanged
1438
   * (i.e. things to be edited) will reacquire CE_VALID bit
1439
   * automatically, which is not really what we want.
1440
   */
1441
0
  if (!ignore_valid && assume_unchanged &&
1442
0
      !(ce->ce_flags & CE_VALID))
1443
0
    updated->ce_flags &= ~CE_VALID;
1444
1445
  /* istate->cache_changed is updated in the caller */
1446
0
  return updated;
1447
0
}
1448
1449
static void show_file(const char * fmt, const char * name, int in_porcelain,
1450
          int * first, const char *header_msg)
1451
0
{
1452
0
  if (in_porcelain && *first && header_msg) {
1453
0
    printf("%s\n", header_msg);
1454
0
    *first = 0;
1455
0
  }
1456
0
  printf(fmt, name);
1457
0
}
1458
1459
int repo_refresh_and_write_index(struct repository *repo,
1460
         unsigned int refresh_flags,
1461
         unsigned int write_flags,
1462
         int gentle,
1463
         const struct pathspec *pathspec,
1464
         char *seen, const char *header_msg)
1465
0
{
1466
0
  struct lock_file lock_file = LOCK_INIT;
1467
0
  int fd, ret = 0;
1468
1469
0
  fd = repo_hold_locked_index(repo, &lock_file,
1470
0
            gentle ? 0 : LOCK_REPORT_ON_ERROR);
1471
0
  if (!gentle && fd < 0)
1472
0
    return -1;
1473
0
  if (refresh_index(repo->index, refresh_flags, pathspec, seen, header_msg))
1474
0
    ret = 1;
1475
0
  if (0 <= fd && write_locked_index(repo->index, &lock_file, COMMIT_LOCK | write_flags))
1476
0
    ret = -1;
1477
0
  return ret;
1478
0
}
1479
1480
1481
int refresh_index(struct index_state *istate, unsigned int flags,
1482
      const struct pathspec *pathspec,
1483
      char *seen, const char *header_msg)
1484
0
{
1485
0
  int i;
1486
0
  int has_errors = 0;
1487
0
  int really = (flags & REFRESH_REALLY) != 0;
1488
0
  int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1489
0
  int quiet = (flags & REFRESH_QUIET) != 0;
1490
0
  int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1491
0
  int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1492
0
  int ignore_skip_worktree = (flags & REFRESH_IGNORE_SKIP_WORKTREE) != 0;
1493
0
  int first = 1;
1494
0
  int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1495
0
  unsigned int options = (CE_MATCH_REFRESH |
1496
0
        (really ? CE_MATCH_IGNORE_VALID : 0) |
1497
0
        (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1498
0
  const char *modified_fmt;
1499
0
  const char *deleted_fmt;
1500
0
  const char *typechange_fmt;
1501
0
  const char *added_fmt;
1502
0
  const char *unmerged_fmt;
1503
0
  struct progress *progress = NULL;
1504
0
  int t2_sum_lstat = 0;
1505
0
  int t2_sum_scan = 0;
1506
1507
0
  if (flags & REFRESH_PROGRESS && isatty(2))
1508
0
    progress = start_delayed_progress(the_repository,
1509
0
              _("Refresh index"),
1510
0
              istate->cache_nr);
1511
1512
0
  trace_performance_enter();
1513
0
  modified_fmt   = in_porcelain ? "M\t%s\n" : "%s: needs update\n";
1514
0
  deleted_fmt    = in_porcelain ? "D\t%s\n" : "%s: needs update\n";
1515
0
  typechange_fmt = in_porcelain ? "T\t%s\n" : "%s: needs update\n";
1516
0
  added_fmt      = in_porcelain ? "A\t%s\n" : "%s: needs update\n";
1517
0
  unmerged_fmt   = in_porcelain ? "U\t%s\n" : "%s: needs merge\n";
1518
  /*
1519
   * Use the multi-threaded preload_index() to refresh most of the
1520
   * cache entries quickly then in the single threaded loop below,
1521
   * we only have to do the special cases that are left.
1522
   */
1523
0
  preload_index(istate, pathspec, 0);
1524
0
  trace2_region_enter("index", "refresh", NULL);
1525
1526
0
  for (i = 0; i < istate->cache_nr; i++) {
1527
0
    struct cache_entry *ce, *new_entry;
1528
0
    int cache_errno = 0;
1529
0
    int changed = 0;
1530
0
    int filtered = 0;
1531
0
    int t2_did_lstat = 0;
1532
0
    int t2_did_scan = 0;
1533
1534
0
    ce = istate->cache[i];
1535
0
    if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1536
0
      continue;
1537
0
    if (ignore_skip_worktree && ce_skip_worktree(ce))
1538
0
      continue;
1539
1540
    /*
1541
     * If this entry is a sparse directory, then there isn't
1542
     * any stat() information to update. Ignore the entry.
1543
     */
1544
0
    if (S_ISSPARSEDIR(ce->ce_mode))
1545
0
      continue;
1546
1547
0
    if (pathspec && !ce_path_match(istate, ce, pathspec, seen))
1548
0
      filtered = 1;
1549
1550
0
    if (ce_stage(ce)) {
1551
0
      while ((i < istate->cache_nr) &&
1552
0
             ! strcmp(istate->cache[i]->name, ce->name))
1553
0
        i++;
1554
0
      i--;
1555
0
      if (allow_unmerged)
1556
0
        continue;
1557
0
      if (!filtered)
1558
0
        show_file(unmerged_fmt, ce->name, in_porcelain,
1559
0
            &first, header_msg);
1560
0
      has_errors = 1;
1561
0
      continue;
1562
0
    }
1563
1564
0
    if (filtered)
1565
0
      continue;
1566
1567
0
    new_entry = refresh_cache_ent(istate, ce, options,
1568
0
                &cache_errno, &changed,
1569
0
                &t2_did_lstat, &t2_did_scan);
1570
0
    t2_sum_lstat += t2_did_lstat;
1571
0
    t2_sum_scan += t2_did_scan;
1572
0
    if (new_entry == ce)
1573
0
      continue;
1574
0
    display_progress(progress, i);
1575
0
    if (!new_entry) {
1576
0
      const char *fmt;
1577
1578
0
      if (really && cache_errno == EINVAL) {
1579
        /* If we are doing --really-refresh that
1580
         * means the index is not valid anymore.
1581
         */
1582
0
        ce->ce_flags &= ~CE_VALID;
1583
0
        ce->ce_flags |= CE_UPDATE_IN_BASE;
1584
0
        mark_fsmonitor_invalid(istate, ce);
1585
0
        istate->cache_changed |= CE_ENTRY_CHANGED;
1586
0
      }
1587
0
      if (quiet)
1588
0
        continue;
1589
1590
0
      if (cache_errno == ENOENT)
1591
0
        fmt = deleted_fmt;
1592
0
      else if (ce_intent_to_add(ce))
1593
0
        fmt = added_fmt; /* must be before other checks */
1594
0
      else if (changed & TYPE_CHANGED)
1595
0
        fmt = typechange_fmt;
1596
0
      else
1597
0
        fmt = modified_fmt;
1598
0
      show_file(fmt,
1599
0
          ce->name, in_porcelain, &first, header_msg);
1600
0
      has_errors = 1;
1601
0
      continue;
1602
0
    }
1603
1604
0
    replace_index_entry(istate, i, new_entry);
1605
0
  }
1606
0
  trace2_data_intmax("index", NULL, "refresh/sum_lstat", t2_sum_lstat);
1607
0
  trace2_data_intmax("index", NULL, "refresh/sum_scan", t2_sum_scan);
1608
0
  trace2_region_leave("index", "refresh", NULL);
1609
0
  display_progress(progress, istate->cache_nr);
1610
0
  stop_progress(&progress);
1611
0
  trace_performance_leave("refresh index");
1612
0
  return has_errors;
1613
0
}
1614
1615
struct cache_entry *refresh_cache_entry(struct index_state *istate,
1616
          struct cache_entry *ce,
1617
          unsigned int options)
1618
0
{
1619
0
  return refresh_cache_ent(istate, ce, options, NULL, NULL, NULL, NULL);
1620
0
}
1621
1622
1623
/*****************************************************************
1624
 * Index File I/O
1625
 *****************************************************************/
1626
1627
0
#define INDEX_FORMAT_DEFAULT 3
1628
1629
static unsigned int get_index_format_default(struct repository *r)
1630
0
{
1631
0
  char *envversion = getenv("GIT_INDEX_VERSION");
1632
0
  char *endp;
1633
0
  unsigned int version = INDEX_FORMAT_DEFAULT;
1634
1635
0
  if (!envversion) {
1636
0
    prepare_repo_settings(r);
1637
1638
0
    if (r->settings.index_version >= 0)
1639
0
      version = r->settings.index_version;
1640
0
    if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1641
0
      warning(_("index.version set, but the value is invalid.\n"
1642
0
          "Using version %i"), INDEX_FORMAT_DEFAULT);
1643
0
      return INDEX_FORMAT_DEFAULT;
1644
0
    }
1645
0
    return version;
1646
0
  }
1647
1648
0
  version = strtoul(envversion, &endp, 10);
1649
0
  if (*endp ||
1650
0
      version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1651
0
    warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1652
0
        "Using version %i"), INDEX_FORMAT_DEFAULT);
1653
0
    version = INDEX_FORMAT_DEFAULT;
1654
0
  }
1655
0
  return version;
1656
0
}
1657
1658
/*
1659
 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1660
 * Again - this is just a (very strong in practice) heuristic that
1661
 * the inode hasn't changed.
1662
 *
1663
 * We save the fields in big-endian order to allow using the
1664
 * index file over NFS transparently.
1665
 */
1666
struct ondisk_cache_entry {
1667
  struct cache_time ctime;
1668
  struct cache_time mtime;
1669
  uint32_t dev;
1670
  uint32_t ino;
1671
  uint32_t mode;
1672
  uint32_t uid;
1673
  uint32_t gid;
1674
  uint32_t size;
1675
  /*
1676
   * unsigned char hash[hashsz];
1677
   * uint16_t flags;
1678
   * if (flags & CE_EXTENDED)
1679
   *  uint16_t flags2;
1680
   */
1681
  unsigned char data[GIT_MAX_RAWSZ + 2 * sizeof(uint16_t)];
1682
  char name[FLEX_ARRAY];
1683
};
1684
1685
/* These are only used for v3 or lower */
1686
0
#define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1687
0
#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,data) + (len) + 8) & ~7)
1688
0
#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1689
0
#define ondisk_data_size(flags, len) (the_hash_algo->rawsz + \
1690
0
             ((flags & CE_EXTENDED) ? 2 : 1) * sizeof(uint16_t) + len)
1691
#define ondisk_data_size_max(len) (ondisk_data_size(CE_EXTENDED, len))
1692
0
#define ondisk_ce_size(ce) (ondisk_cache_entry_size(ondisk_data_size((ce)->ce_flags, ce_namelen(ce))))
1693
1694
/* Allow fsck to force verification of the index checksum. */
1695
int verify_index_checksum;
1696
1697
/* Allow fsck to force verification of the cache entry order. */
1698
int verify_ce_order;
1699
1700
static int verify_hdr(const struct cache_header *hdr, unsigned long size)
1701
0
{
1702
0
  struct git_hash_ctx c;
1703
0
  unsigned char hash[GIT_MAX_RAWSZ];
1704
0
  int hdr_version;
1705
0
  unsigned char *start, *end;
1706
0
  struct object_id oid;
1707
1708
0
  if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1709
0
    return error(_("bad signature 0x%08x"), hdr->hdr_signature);
1710
0
  hdr_version = ntohl(hdr->hdr_version);
1711
0
  if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1712
0
    return error(_("bad index version %d"), hdr_version);
1713
1714
0
  if (!verify_index_checksum)
1715
0
    return 0;
1716
1717
0
  end = (unsigned char *)hdr + size;
1718
0
  start = end - the_hash_algo->rawsz;
1719
0
  oidread(&oid, start, the_repository->hash_algo);
1720
0
  if (oideq(&oid, null_oid(the_hash_algo)))
1721
0
    return 0;
1722
1723
0
  the_hash_algo->init_fn(&c);
1724
0
  git_hash_update(&c, hdr, size - the_hash_algo->rawsz);
1725
0
  git_hash_final(hash, &c);
1726
0
  if (!hasheq(hash, start, the_repository->hash_algo))
1727
0
    return error(_("bad index file sha1 signature"));
1728
0
  return 0;
1729
0
}
1730
1731
static int read_index_extension(struct index_state *istate,
1732
        const char *ext, const char *data, unsigned long sz)
1733
0
{
1734
0
  switch (CACHE_EXT(ext)) {
1735
0
  case CACHE_EXT_TREE:
1736
0
    istate->cache_tree = cache_tree_read(data, sz);
1737
0
    break;
1738
0
  case CACHE_EXT_RESOLVE_UNDO:
1739
0
    istate->resolve_undo = resolve_undo_read(data, sz, the_hash_algo);
1740
0
    break;
1741
0
  case CACHE_EXT_LINK:
1742
0
    if (read_link_extension(istate, data, sz))
1743
0
      return -1;
1744
0
    break;
1745
0
  case CACHE_EXT_UNTRACKED:
1746
0
    istate->untracked = read_untracked_extension(data, sz);
1747
0
    break;
1748
0
  case CACHE_EXT_FSMONITOR:
1749
0
    read_fsmonitor_extension(istate, data, sz);
1750
0
    break;
1751
0
  case CACHE_EXT_ENDOFINDEXENTRIES:
1752
0
  case CACHE_EXT_INDEXENTRYOFFSETTABLE:
1753
    /* already handled in do_read_index() */
1754
0
    break;
1755
0
  case CACHE_EXT_SPARSE_DIRECTORIES:
1756
    /* no content, only an indicator */
1757
0
    istate->sparse_index = INDEX_COLLAPSED;
1758
0
    break;
1759
0
  default:
1760
0
    if (*ext < 'A' || 'Z' < *ext)
1761
0
      return error(_("index uses %.4s extension, which we do not understand"),
1762
0
             ext);
1763
0
    fprintf_ln(stderr, _("ignoring %.4s extension"), ext);
1764
0
    break;
1765
0
  }
1766
0
  return 0;
1767
0
}
1768
1769
/*
1770
 * Parses the contents of the cache entry contained within the 'ondisk' buffer
1771
 * into a new incore 'cache_entry'.
1772
 *
1773
 * Note that 'char *ondisk' may not be aligned to a 4-byte address interval in
1774
 * index v4, so we cannot cast it to 'struct ondisk_cache_entry *' and access
1775
 * its members. Instead, we use the byte offsets of members within the struct to
1776
 * identify where 'get_be16()', 'get_be32()', and 'oidread()' (which can all
1777
 * read from an unaligned memory buffer) should read from the 'ondisk' buffer
1778
 * into the corresponding incore 'cache_entry' members.
1779
 */
1780
static struct cache_entry *create_from_disk(struct mem_pool *ce_mem_pool,
1781
              unsigned int version,
1782
              const char *ondisk,
1783
              unsigned long *ent_size,
1784
              const struct cache_entry *previous_ce)
1785
0
{
1786
0
  struct cache_entry *ce;
1787
0
  size_t len;
1788
0
  const char *name;
1789
0
  const unsigned hashsz = the_hash_algo->rawsz;
1790
0
  const char *flagsp = ondisk + offsetof(struct ondisk_cache_entry, data) + hashsz;
1791
0
  unsigned int flags;
1792
0
  size_t copy_len = 0;
1793
  /*
1794
   * Adjacent cache entries tend to share the leading paths, so it makes
1795
   * sense to only store the differences in later entries.  In the v4
1796
   * on-disk format of the index, each on-disk cache entry stores the
1797
   * number of bytes to be stripped from the end of the previous name,
1798
   * and the bytes to append to the result, to come up with its name.
1799
   */
1800
0
  int expand_name_field = version == 4;
1801
1802
  /* On-disk flags are just 16 bits */
1803
0
  flags = get_be16(flagsp);
1804
0
  len = flags & CE_NAMEMASK;
1805
1806
0
  if (flags & CE_EXTENDED) {
1807
0
    int extended_flags;
1808
0
    extended_flags = get_be16(flagsp + sizeof(uint16_t)) << 16;
1809
    /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1810
0
    if (extended_flags & ~CE_EXTENDED_FLAGS)
1811
0
      die(_("unknown index entry format 0x%08x"), extended_flags);
1812
0
    flags |= extended_flags;
1813
0
    name = (const char *)(flagsp + 2 * sizeof(uint16_t));
1814
0
  }
1815
0
  else
1816
0
    name = (const char *)(flagsp + sizeof(uint16_t));
1817
1818
0
  if (expand_name_field) {
1819
0
    const unsigned char *cp = (const unsigned char *)name;
1820
0
    uint64_t strip_len, previous_len;
1821
1822
    /* If we're at the beginning of a block, ignore the previous name */
1823
0
    strip_len = decode_varint(&cp);
1824
0
    if (previous_ce) {
1825
0
      previous_len = previous_ce->ce_namelen;
1826
0
      if (previous_len < strip_len)
1827
0
        die(_("malformed name field in the index, near path '%s'"),
1828
0
          previous_ce->name);
1829
0
      copy_len = previous_len - strip_len;
1830
0
    }
1831
0
    name = (const char *)cp;
1832
0
  }
1833
1834
0
  if (len == CE_NAMEMASK) {
1835
0
    len = strlen(name);
1836
0
    if (expand_name_field)
1837
0
      len += copy_len;
1838
0
  }
1839
1840
0
  ce = mem_pool__ce_alloc(ce_mem_pool, len);
1841
1842
  /*
1843
   * NEEDSWORK: using 'offsetof()' is cumbersome and should be replaced
1844
   * with something more akin to 'load_bitmap_entries_v1()'s use of
1845
   * 'read_be16'/'read_be32'. For consistency with the corresponding
1846
   * ondisk entry write function ('copy_cache_entry_to_ondisk()'), this
1847
   * should be done at the same time as removing references to
1848
   * 'ondisk_cache_entry' there.
1849
   */
1850
0
  ce->ce_stat_data.sd_ctime.sec = get_be32(ondisk + offsetof(struct ondisk_cache_entry, ctime)
1851
0
              + offsetof(struct cache_time, sec));
1852
0
  ce->ce_stat_data.sd_mtime.sec = get_be32(ondisk + offsetof(struct ondisk_cache_entry, mtime)
1853
0
              + offsetof(struct cache_time, sec));
1854
0
  ce->ce_stat_data.sd_ctime.nsec = get_be32(ondisk + offsetof(struct ondisk_cache_entry, ctime)
1855
0
               + offsetof(struct cache_time, nsec));
1856
0
  ce->ce_stat_data.sd_mtime.nsec = get_be32(ondisk + offsetof(struct ondisk_cache_entry, mtime)
1857
0
               + offsetof(struct cache_time, nsec));
1858
0
  ce->ce_stat_data.sd_dev   = get_be32(ondisk + offsetof(struct ondisk_cache_entry, dev));
1859
0
  ce->ce_stat_data.sd_ino   = get_be32(ondisk + offsetof(struct ondisk_cache_entry, ino));
1860
0
  ce->ce_mode  = get_be32(ondisk + offsetof(struct ondisk_cache_entry, mode));
1861
0
  ce->ce_stat_data.sd_uid   = get_be32(ondisk + offsetof(struct ondisk_cache_entry, uid));
1862
0
  ce->ce_stat_data.sd_gid   = get_be32(ondisk + offsetof(struct ondisk_cache_entry, gid));
1863
0
  ce->ce_stat_data.sd_size  = get_be32(ondisk + offsetof(struct ondisk_cache_entry, size));
1864
0
  ce->ce_flags = flags & ~CE_NAMEMASK;
1865
0
  ce->ce_namelen = len;
1866
0
  ce->index = 0;
1867
0
  oidread(&ce->oid, (const unsigned char *)ondisk + offsetof(struct ondisk_cache_entry, data),
1868
0
    the_repository->hash_algo);
1869
1870
0
  if (expand_name_field) {
1871
0
    if (copy_len)
1872
0
      memcpy(ce->name, previous_ce->name, copy_len);
1873
0
    memcpy(ce->name + copy_len, name, len + 1 - copy_len);
1874
0
    *ent_size = (name - ((char *)ondisk)) + len + 1 - copy_len;
1875
0
  } else {
1876
0
    memcpy(ce->name, name, len + 1);
1877
0
    *ent_size = ondisk_ce_size(ce);
1878
0
  }
1879
0
  return ce;
1880
0
}
1881
1882
static void check_ce_order(struct index_state *istate)
1883
0
{
1884
0
  unsigned int i;
1885
1886
0
  if (!verify_ce_order)
1887
0
    return;
1888
1889
0
  for (i = 1; i < istate->cache_nr; i++) {
1890
0
    struct cache_entry *ce = istate->cache[i - 1];
1891
0
    struct cache_entry *next_ce = istate->cache[i];
1892
0
    int name_compare = strcmp(ce->name, next_ce->name);
1893
1894
0
    if (0 < name_compare)
1895
0
      die(_("unordered stage entries in index"));
1896
0
    if (!name_compare) {
1897
0
      if (!ce_stage(ce))
1898
0
        die(_("multiple stage entries for merged file '%s'"),
1899
0
            ce->name);
1900
0
      if (ce_stage(ce) > ce_stage(next_ce))
1901
0
        die(_("unordered stage entries for '%s'"),
1902
0
            ce->name);
1903
0
    }
1904
0
  }
1905
0
}
1906
1907
static void tweak_untracked_cache(struct index_state *istate)
1908
0
{
1909
0
  struct repository *r = the_repository;
1910
1911
0
  prepare_repo_settings(r);
1912
1913
0
  switch (r->settings.core_untracked_cache) {
1914
0
  case UNTRACKED_CACHE_REMOVE:
1915
0
    remove_untracked_cache(istate);
1916
0
    break;
1917
0
  case UNTRACKED_CACHE_WRITE:
1918
0
    add_untracked_cache(istate);
1919
0
    break;
1920
0
  case UNTRACKED_CACHE_KEEP:
1921
    /*
1922
     * Either an explicit "core.untrackedCache=keep", the
1923
     * default if "core.untrackedCache" isn't configured,
1924
     * or a fallback on an unknown "core.untrackedCache"
1925
     * value.
1926
     */
1927
0
    break;
1928
0
  }
1929
0
}
1930
1931
static void tweak_split_index(struct index_state *istate)
1932
0
{
1933
0
  switch (repo_config_get_split_index(the_repository)) {
1934
0
  case -1: /* unset: do nothing */
1935
0
    break;
1936
0
  case 0: /* false */
1937
0
    remove_split_index(istate);
1938
0
    break;
1939
0
  case 1: /* true */
1940
0
    add_split_index(istate);
1941
0
    break;
1942
0
  default: /* unknown value: do nothing */
1943
0
    break;
1944
0
  }
1945
0
}
1946
1947
static void post_read_index_from(struct index_state *istate)
1948
0
{
1949
0
  check_ce_order(istate);
1950
0
  tweak_untracked_cache(istate);
1951
0
  tweak_split_index(istate);
1952
0
  tweak_fsmonitor(istate);
1953
0
}
1954
1955
static size_t estimate_cache_size_from_compressed(unsigned int entries)
1956
0
{
1957
0
  return entries * (sizeof(struct cache_entry) + CACHE_ENTRY_PATH_LENGTH);
1958
0
}
1959
1960
static size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1961
0
{
1962
0
  long per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1963
1964
  /*
1965
   * Account for potential alignment differences.
1966
   */
1967
0
  per_entry += align_padding_size(per_entry, 0);
1968
0
  return ondisk_size + entries * per_entry;
1969
0
}
1970
1971
struct index_entry_offset
1972
{
1973
  /* starting byte offset into index file, count of index entries in this block */
1974
  int offset, nr;
1975
};
1976
1977
struct index_entry_offset_table
1978
{
1979
  int nr;
1980
  struct index_entry_offset entries[FLEX_ARRAY];
1981
};
1982
1983
static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset);
1984
static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot);
1985
1986
static size_t read_eoie_extension(const char *mmap, size_t mmap_size);
1987
static void write_eoie_extension(struct strbuf *sb, struct git_hash_ctx *eoie_context, size_t offset);
1988
1989
struct load_index_extensions
1990
{
1991
  pthread_t pthread;
1992
  struct index_state *istate;
1993
  const char *mmap;
1994
  size_t mmap_size;
1995
  unsigned long src_offset;
1996
};
1997
1998
static void *load_index_extensions(void *_data)
1999
0
{
2000
0
  struct load_index_extensions *p = _data;
2001
0
  unsigned long src_offset = p->src_offset;
2002
2003
0
  while (src_offset <= p->mmap_size - the_hash_algo->rawsz - 8) {
2004
    /* After an array of active_nr index entries,
2005
     * there can be arbitrary number of extended
2006
     * sections, each of which is prefixed with
2007
     * extension name (4-byte) and section length
2008
     * in 4-byte network byte order.
2009
     */
2010
0
    uint32_t extsize = get_be32(p->mmap + src_offset + 4);
2011
0
    if (read_index_extension(p->istate,
2012
0
           p->mmap + src_offset,
2013
0
           p->mmap + src_offset + 8,
2014
0
           extsize) < 0) {
2015
0
      munmap((void *)p->mmap, p->mmap_size);
2016
0
      die(_("index file corrupt"));
2017
0
    }
2018
0
    src_offset += 8;
2019
0
    src_offset += extsize;
2020
0
  }
2021
2022
0
  return NULL;
2023
0
}
2024
2025
/*
2026
 * A helper function that will load the specified range of cache entries
2027
 * from the memory mapped file and add them to the given index.
2028
 */
2029
static unsigned long load_cache_entry_block(struct index_state *istate,
2030
      struct mem_pool *ce_mem_pool, int offset, int nr, const char *mmap,
2031
      unsigned long start_offset, const struct cache_entry *previous_ce)
2032
0
{
2033
0
  int i;
2034
0
  unsigned long src_offset = start_offset;
2035
2036
0
  for (i = offset; i < offset + nr; i++) {
2037
0
    struct cache_entry *ce;
2038
0
    unsigned long consumed;
2039
2040
0
    ce = create_from_disk(ce_mem_pool, istate->version,
2041
0
              mmap + src_offset,
2042
0
              &consumed, previous_ce);
2043
0
    set_index_entry(istate, i, ce);
2044
2045
0
    src_offset += consumed;
2046
0
    previous_ce = ce;
2047
0
  }
2048
0
  return src_offset - start_offset;
2049
0
}
2050
2051
static unsigned long load_all_cache_entries(struct index_state *istate,
2052
      const char *mmap, size_t mmap_size, unsigned long src_offset)
2053
0
{
2054
0
  unsigned long consumed;
2055
2056
0
  istate->ce_mem_pool = xmalloc(sizeof(*istate->ce_mem_pool));
2057
0
  if (istate->version == 4) {
2058
0
    mem_pool_init(istate->ce_mem_pool,
2059
0
        estimate_cache_size_from_compressed(istate->cache_nr));
2060
0
  } else {
2061
0
    mem_pool_init(istate->ce_mem_pool,
2062
0
        estimate_cache_size(mmap_size, istate->cache_nr));
2063
0
  }
2064
2065
0
  consumed = load_cache_entry_block(istate, istate->ce_mem_pool,
2066
0
          0, istate->cache_nr, mmap, src_offset, NULL);
2067
0
  return consumed;
2068
0
}
2069
2070
/*
2071
 * Mostly randomly chosen maximum thread counts: we
2072
 * cap the parallelism to online_cpus() threads, and we want
2073
 * to have at least 10000 cache entries per thread for it to
2074
 * be worth starting a thread.
2075
 */
2076
2077
0
#define THREAD_COST   (10000)
2078
2079
struct load_cache_entries_thread_data
2080
{
2081
  pthread_t pthread;
2082
  struct index_state *istate;
2083
  struct mem_pool *ce_mem_pool;
2084
  int offset;
2085
  const char *mmap;
2086
  struct index_entry_offset_table *ieot;
2087
  int ieot_start;   /* starting index into the ieot array */
2088
  int ieot_blocks;  /* count of ieot entries to process */
2089
  unsigned long consumed; /* return # of bytes in index file processed */
2090
};
2091
2092
/*
2093
 * A thread proc to run the load_cache_entries() computation
2094
 * across multiple background threads.
2095
 */
2096
static void *load_cache_entries_thread(void *_data)
2097
0
{
2098
0
  struct load_cache_entries_thread_data *p = _data;
2099
0
  int i;
2100
2101
  /* iterate across all ieot blocks assigned to this thread */
2102
0
  for (i = p->ieot_start; i < p->ieot_start + p->ieot_blocks; i++) {
2103
0
    p->consumed += load_cache_entry_block(p->istate, p->ce_mem_pool,
2104
0
      p->offset, p->ieot->entries[i].nr, p->mmap, p->ieot->entries[i].offset, NULL);
2105
0
    p->offset += p->ieot->entries[i].nr;
2106
0
  }
2107
0
  return NULL;
2108
0
}
2109
2110
static unsigned long load_cache_entries_threaded(struct index_state *istate, const char *mmap, size_t mmap_size,
2111
             int nr_threads, struct index_entry_offset_table *ieot)
2112
0
{
2113
0
  int i, offset, ieot_blocks, ieot_start, err;
2114
0
  struct load_cache_entries_thread_data *data;
2115
0
  unsigned long consumed = 0;
2116
2117
  /* a little sanity checking */
2118
0
  if (istate->name_hash_initialized)
2119
0
    BUG("the name hash isn't thread safe");
2120
2121
0
  istate->ce_mem_pool = xmalloc(sizeof(*istate->ce_mem_pool));
2122
0
  mem_pool_init(istate->ce_mem_pool, 0);
2123
2124
  /* ensure we have no more threads than we have blocks to process */
2125
0
  if (nr_threads > ieot->nr)
2126
0
    nr_threads = ieot->nr;
2127
0
  CALLOC_ARRAY(data, nr_threads);
2128
2129
0
  offset = ieot_start = 0;
2130
0
  ieot_blocks = DIV_ROUND_UP(ieot->nr, nr_threads);
2131
0
  for (i = 0; i < nr_threads; i++) {
2132
0
    struct load_cache_entries_thread_data *p = &data[i];
2133
0
    int nr, j;
2134
2135
0
    if (ieot_start + ieot_blocks > ieot->nr)
2136
0
      ieot_blocks = ieot->nr - ieot_start;
2137
2138
0
    p->istate = istate;
2139
0
    p->offset = offset;
2140
0
    p->mmap = mmap;
2141
0
    p->ieot = ieot;
2142
0
    p->ieot_start = ieot_start;
2143
0
    p->ieot_blocks = ieot_blocks;
2144
2145
    /* create a mem_pool for each thread */
2146
0
    nr = 0;
2147
0
    for (j = p->ieot_start; j < p->ieot_start + p->ieot_blocks; j++)
2148
0
      nr += p->ieot->entries[j].nr;
2149
0
    p->ce_mem_pool = xmalloc(sizeof(*istate->ce_mem_pool));
2150
0
    if (istate->version == 4) {
2151
0
      mem_pool_init(p->ce_mem_pool,
2152
0
        estimate_cache_size_from_compressed(nr));
2153
0
    } else {
2154
0
      mem_pool_init(p->ce_mem_pool,
2155
0
        estimate_cache_size(mmap_size, nr));
2156
0
    }
2157
2158
0
    err = pthread_create(&p->pthread, NULL, load_cache_entries_thread, p);
2159
0
    if (err)
2160
0
      die(_("unable to create load_cache_entries thread: %s"), strerror(err));
2161
2162
    /* increment by the number of cache entries in the ieot block being processed */
2163
0
    for (j = 0; j < ieot_blocks; j++)
2164
0
      offset += ieot->entries[ieot_start + j].nr;
2165
0
    ieot_start += ieot_blocks;
2166
0
  }
2167
2168
0
  for (i = 0; i < nr_threads; i++) {
2169
0
    struct load_cache_entries_thread_data *p = &data[i];
2170
2171
0
    err = pthread_join(p->pthread, NULL);
2172
0
    if (err)
2173
0
      die(_("unable to join load_cache_entries thread: %s"), strerror(err));
2174
0
    mem_pool_combine(istate->ce_mem_pool, p->ce_mem_pool);
2175
0
    free(p->ce_mem_pool);
2176
0
    consumed += p->consumed;
2177
0
  }
2178
2179
0
  free(data);
2180
2181
0
  return consumed;
2182
0
}
2183
2184
static void set_new_index_sparsity(struct index_state *istate)
2185
0
{
2186
  /*
2187
   * If the index's repo exists, mark it sparse according to
2188
   * repo settings.
2189
   */
2190
0
  prepare_repo_settings(istate->repo);
2191
0
  if (!istate->repo->settings.command_requires_full_index &&
2192
0
      is_sparse_index_allowed(istate, 0))
2193
0
    istate->sparse_index = 1;
2194
0
}
2195
2196
/* remember to discard_cache() before reading a different cache! */
2197
int do_read_index(struct index_state *istate, const char *path, int must_exist)
2198
0
{
2199
0
  int fd;
2200
0
  struct stat st;
2201
0
  unsigned long src_offset;
2202
0
  const struct cache_header *hdr;
2203
0
  const char *mmap;
2204
0
  size_t mmap_size;
2205
0
  struct load_index_extensions p;
2206
0
  size_t extension_offset = 0;
2207
0
  int nr_threads, cpus;
2208
0
  struct index_entry_offset_table *ieot = NULL;
2209
2210
0
  if (istate->initialized)
2211
0
    return istate->cache_nr;
2212
2213
0
  istate->timestamp.sec = 0;
2214
0
  istate->timestamp.nsec = 0;
2215
0
  fd = open(path, O_RDONLY);
2216
0
  if (fd < 0) {
2217
0
    if (!must_exist && errno == ENOENT) {
2218
0
      set_new_index_sparsity(istate);
2219
0
      istate->initialized = 1;
2220
0
      return 0;
2221
0
    }
2222
0
    die_errno(_("%s: index file open failed"), path);
2223
0
  }
2224
2225
0
  if (fstat(fd, &st))
2226
0
    die_errno(_("%s: cannot stat the open index"), path);
2227
2228
0
  mmap_size = xsize_t(st.st_size);
2229
0
  if (mmap_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2230
0
    die(_("%s: index file smaller than expected"), path);
2231
2232
0
  mmap = xmmap_gently(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
2233
0
  if (mmap == MAP_FAILED)
2234
0
    die_errno(_("%s: unable to map index file%s"), path,
2235
0
      mmap_os_err());
2236
0
  close(fd);
2237
2238
0
  hdr = (const struct cache_header *)mmap;
2239
0
  if (verify_hdr(hdr, mmap_size) < 0)
2240
0
    goto unmap;
2241
2242
0
  oidread(&istate->oid, (const unsigned char *)hdr + mmap_size - the_hash_algo->rawsz,
2243
0
    the_repository->hash_algo);
2244
0
  istate->version = ntohl(hdr->hdr_version);
2245
0
  istate->cache_nr = ntohl(hdr->hdr_entries);
2246
0
  istate->cache_alloc = alloc_nr(istate->cache_nr);
2247
0
  CALLOC_ARRAY(istate->cache, istate->cache_alloc);
2248
0
  istate->initialized = 1;
2249
2250
0
  p.istate = istate;
2251
0
  p.mmap = mmap;
2252
0
  p.mmap_size = mmap_size;
2253
2254
0
  src_offset = sizeof(*hdr);
2255
2256
0
  if (repo_config_get_index_threads(the_repository, &nr_threads))
2257
0
    nr_threads = 1;
2258
2259
  /* TODO: does creating more threads than cores help? */
2260
0
  if (!nr_threads) {
2261
0
    nr_threads = istate->cache_nr / THREAD_COST;
2262
0
    cpus = online_cpus();
2263
0
    if (nr_threads > cpus)
2264
0
      nr_threads = cpus;
2265
0
  }
2266
2267
0
  if (!HAVE_THREADS)
2268
0
    nr_threads = 1;
2269
2270
0
  if (nr_threads > 1) {
2271
0
    extension_offset = read_eoie_extension(mmap, mmap_size);
2272
0
    if (extension_offset) {
2273
0
      int err;
2274
2275
0
      p.src_offset = extension_offset;
2276
0
      err = pthread_create(&p.pthread, NULL, load_index_extensions, &p);
2277
0
      if (err)
2278
0
        die(_("unable to create load_index_extensions thread: %s"), strerror(err));
2279
2280
0
      nr_threads--;
2281
0
    }
2282
0
  }
2283
2284
  /*
2285
   * Locate and read the index entry offset table so that we can use it
2286
   * to multi-thread the reading of the cache entries.
2287
   */
2288
0
  if (extension_offset && nr_threads > 1)
2289
0
    ieot = read_ieot_extension(mmap, mmap_size, extension_offset);
2290
2291
0
  if (ieot) {
2292
0
    src_offset += load_cache_entries_threaded(istate, mmap, mmap_size, nr_threads, ieot);
2293
0
    free(ieot);
2294
0
  } else {
2295
0
    src_offset += load_all_cache_entries(istate, mmap, mmap_size, src_offset);
2296
0
  }
2297
2298
0
  istate->timestamp.sec = st.st_mtime;
2299
0
  istate->timestamp.nsec = ST_MTIME_NSEC(st);
2300
2301
  /* if we created a thread, join it otherwise load the extensions on the primary thread */
2302
0
  if (extension_offset) {
2303
0
    int ret = pthread_join(p.pthread, NULL);
2304
0
    if (ret)
2305
0
      die(_("unable to join load_index_extensions thread: %s"), strerror(ret));
2306
0
  } else {
2307
0
    p.src_offset = src_offset;
2308
0
    load_index_extensions(&p);
2309
0
  }
2310
0
  munmap((void *)mmap, mmap_size);
2311
2312
  /*
2313
   * TODO trace2: replace "the_repository" with the actual repo instance
2314
   * that is associated with the given "istate".
2315
   */
2316
0
  trace2_data_intmax("index", the_repository, "read/version",
2317
0
         istate->version);
2318
0
  trace2_data_intmax("index", the_repository, "read/cache_nr",
2319
0
         istate->cache_nr);
2320
2321
  /*
2322
   * If the command explicitly requires a full index, force it
2323
   * to be full. Otherwise, correct the sparsity based on repository
2324
   * settings and other properties of the index (if necessary).
2325
   */
2326
0
  prepare_repo_settings(istate->repo);
2327
0
  if (istate->repo->settings.command_requires_full_index)
2328
0
    ensure_full_index(istate);
2329
0
  else
2330
0
    ensure_correct_sparsity(istate);
2331
2332
0
  return istate->cache_nr;
2333
2334
0
unmap:
2335
0
  munmap((void *)mmap, mmap_size);
2336
0
  die(_("index file corrupt"));
2337
0
}
2338
2339
/*
2340
 * Signal that the shared index is used by updating its mtime.
2341
 *
2342
 * This way, shared index can be removed if they have not been used
2343
 * for some time.
2344
 */
2345
static void freshen_shared_index(const char *shared_index, int warn)
2346
0
{
2347
0
  if (!check_and_freshen_file(shared_index, 1) && warn)
2348
0
    warning(_("could not freshen shared index '%s'"), shared_index);
2349
0
}
2350
2351
int read_index_from(struct index_state *istate, const char *path,
2352
        const char *gitdir)
2353
0
{
2354
0
  struct split_index *split_index;
2355
0
  int ret;
2356
0
  char *base_oid_hex;
2357
0
  char *base_path;
2358
2359
  /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2360
0
  if (istate->initialized)
2361
0
    return istate->cache_nr;
2362
2363
  /*
2364
   * TODO trace2: replace "the_repository" with the actual repo instance
2365
   * that is associated with the given "istate".
2366
   */
2367
0
  trace2_region_enter_printf("index", "do_read_index", the_repository,
2368
0
           "%s", path);
2369
0
  trace_performance_enter();
2370
0
  ret = do_read_index(istate, path, 0);
2371
0
  trace_performance_leave("read cache %s", path);
2372
0
  trace2_region_leave_printf("index", "do_read_index", the_repository,
2373
0
           "%s", path);
2374
2375
0
  split_index = istate->split_index;
2376
0
  if (!split_index || is_null_oid(&split_index->base_oid)) {
2377
0
    post_read_index_from(istate);
2378
0
    return ret;
2379
0
  }
2380
2381
0
  trace_performance_enter();
2382
0
  if (split_index->base)
2383
0
    release_index(split_index->base);
2384
0
  else
2385
0
    ALLOC_ARRAY(split_index->base, 1);
2386
0
  index_state_init(split_index->base, istate->repo);
2387
2388
0
  base_oid_hex = oid_to_hex(&split_index->base_oid);
2389
0
  base_path = xstrfmt("%s/sharedindex.%s", gitdir, base_oid_hex);
2390
0
  if (file_exists(base_path)) {
2391
0
    trace2_region_enter_printf("index", "shared/do_read_index",
2392
0
          the_repository, "%s", base_path);
2393
2394
0
    ret = do_read_index(split_index->base, base_path, 0);
2395
0
    trace2_region_leave_printf("index", "shared/do_read_index",
2396
0
          the_repository, "%s", base_path);
2397
0
  } else {
2398
0
    char *path_copy = xstrdup(path);
2399
0
    char *base_path2 = xstrfmt("%s/sharedindex.%s",
2400
0
             dirname(path_copy), base_oid_hex);
2401
0
    free(path_copy);
2402
0
    trace2_region_enter_printf("index", "shared/do_read_index",
2403
0
             the_repository, "%s", base_path2);
2404
0
    ret = do_read_index(split_index->base, base_path2, 1);
2405
0
    trace2_region_leave_printf("index", "shared/do_read_index",
2406
0
             the_repository, "%s", base_path2);
2407
0
    free(base_path2);
2408
0
  }
2409
0
  if (!oideq(&split_index->base_oid, &split_index->base->oid))
2410
0
    die(_("broken index, expect %s in %s, got %s"),
2411
0
        base_oid_hex, base_path,
2412
0
        oid_to_hex(&split_index->base->oid));
2413
2414
0
  freshen_shared_index(base_path, 0);
2415
0
  merge_base_index(istate);
2416
0
  post_read_index_from(istate);
2417
0
  trace_performance_leave("read cache %s", base_path);
2418
0
  free(base_path);
2419
0
  return ret;
2420
0
}
2421
2422
int is_index_unborn(struct index_state *istate)
2423
0
{
2424
0
  return (!istate->cache_nr && !istate->timestamp.sec);
2425
0
}
2426
2427
void index_state_init(struct index_state *istate, struct repository *r)
2428
0
{
2429
0
  struct index_state blank = INDEX_STATE_INIT(r);
2430
0
  memcpy(istate, &blank, sizeof(*istate));
2431
0
}
2432
2433
void release_index(struct index_state *istate)
2434
0
{
2435
  /*
2436
   * Cache entries in istate->cache[] should have been allocated
2437
   * from the memory pool associated with this index, or from an
2438
   * associated split_index. There is no need to free individual
2439
   * cache entries. validate_cache_entries can detect when this
2440
   * assertion does not hold.
2441
   */
2442
0
  validate_cache_entries(istate);
2443
2444
0
  resolve_undo_clear_index(istate);
2445
0
  free_name_hash(istate);
2446
0
  cache_tree_free(&(istate->cache_tree));
2447
0
  free(istate->fsmonitor_last_update);
2448
0
  free(istate->cache);
2449
0
  discard_split_index(istate);
2450
0
  free_untracked_cache(istate->untracked);
2451
2452
0
  if (istate->sparse_checkout_patterns) {
2453
0
    clear_pattern_list(istate->sparse_checkout_patterns);
2454
0
    FREE_AND_NULL(istate->sparse_checkout_patterns);
2455
0
  }
2456
2457
0
  if (istate->ce_mem_pool) {
2458
0
    mem_pool_discard(istate->ce_mem_pool, should_validate_cache_entries());
2459
0
    FREE_AND_NULL(istate->ce_mem_pool);
2460
0
  }
2461
0
}
2462
2463
void discard_index(struct index_state *istate)
2464
0
{
2465
0
  release_index(istate);
2466
0
  index_state_init(istate, istate->repo);
2467
0
}
2468
2469
/*
2470
 * Validate the cache entries of this index.
2471
 * All cache entries associated with this index
2472
 * should have been allocated by the memory pool
2473
 * associated with this index, or by a referenced
2474
 * split index.
2475
 */
2476
void validate_cache_entries(const struct index_state *istate)
2477
0
{
2478
0
  int i;
2479
2480
0
  if (!should_validate_cache_entries() ||!istate || !istate->initialized)
2481
0
    return;
2482
2483
0
  for (i = 0; i < istate->cache_nr; i++) {
2484
0
    if (!istate) {
2485
0
      BUG("cache entry is not allocated from expected memory pool");
2486
0
    } else if (!istate->ce_mem_pool ||
2487
0
      !mem_pool_contains(istate->ce_mem_pool, istate->cache[i])) {
2488
0
      if (!istate->split_index ||
2489
0
        !istate->split_index->base ||
2490
0
        !istate->split_index->base->ce_mem_pool ||
2491
0
        !mem_pool_contains(istate->split_index->base->ce_mem_pool, istate->cache[i])) {
2492
0
        BUG("cache entry is not allocated from expected memory pool");
2493
0
      }
2494
0
    }
2495
0
  }
2496
2497
0
  if (istate->split_index)
2498
0
    validate_cache_entries(istate->split_index->base);
2499
0
}
2500
2501
int unmerged_index(const struct index_state *istate)
2502
0
{
2503
0
  int i;
2504
0
  for (i = 0; i < istate->cache_nr; i++) {
2505
0
    if (ce_stage(istate->cache[i]))
2506
0
      return 1;
2507
0
  }
2508
0
  return 0;
2509
0
}
2510
2511
int repo_index_has_changes(struct repository *repo,
2512
         struct tree *tree,
2513
         struct strbuf *sb)
2514
0
{
2515
0
  struct index_state *istate = repo->index;
2516
0
  struct object_id cmp;
2517
0
  int i;
2518
2519
0
  if (tree)
2520
0
    cmp = tree->object.oid;
2521
0
  if (tree || !repo_get_oid_tree(repo, "HEAD", &cmp)) {
2522
0
    struct diff_options opt;
2523
2524
0
    repo_diff_setup(repo, &opt);
2525
0
    opt.flags.exit_with_status = 1;
2526
0
    if (!sb)
2527
0
      opt.flags.quick = 1;
2528
0
    diff_setup_done(&opt);
2529
0
    do_diff_cache(&cmp, &opt);
2530
0
    diffcore_std(&opt);
2531
0
    for (i = 0; sb && i < diff_queued_diff.nr; i++) {
2532
0
      if (i)
2533
0
        strbuf_addch(sb, ' ');
2534
0
      strbuf_addstr(sb, diff_queued_diff.queue[i]->two->path);
2535
0
    }
2536
0
    diff_flush(&opt);
2537
0
    return opt.flags.has_changes != 0;
2538
0
  } else {
2539
    /* TODO: audit for interaction with sparse-index. */
2540
0
    ensure_full_index(istate);
2541
0
    for (i = 0; sb && i < istate->cache_nr; i++) {
2542
0
      if (i)
2543
0
        strbuf_addch(sb, ' ');
2544
0
      strbuf_addstr(sb, istate->cache[i]->name);
2545
0
    }
2546
0
    return !!istate->cache_nr;
2547
0
  }
2548
0
}
2549
2550
static int write_index_ext_header(struct hashfile *f,
2551
          struct git_hash_ctx *eoie_f,
2552
          unsigned int ext,
2553
          unsigned int sz)
2554
0
{
2555
0
  hashwrite_be32(f, ext);
2556
0
  hashwrite_be32(f, sz);
2557
2558
0
  if (eoie_f) {
2559
0
    ext = htonl(ext);
2560
0
    sz = htonl(sz);
2561
0
    git_hash_update(eoie_f, &ext, sizeof(ext));
2562
0
    git_hash_update(eoie_f, &sz, sizeof(sz));
2563
0
  }
2564
0
  return 0;
2565
0
}
2566
2567
static void ce_smudge_racily_clean_entry(struct index_state *istate,
2568
           struct cache_entry *ce)
2569
0
{
2570
  /*
2571
   * The only thing we care about in this function is to smudge the
2572
   * falsely clean entry due to touch-update-touch race, so we leave
2573
   * everything else as they are.  We are called for entries whose
2574
   * ce_stat_data.sd_mtime match the index file mtime.
2575
   *
2576
   * Note that this actually does not do much for gitlinks, for
2577
   * which ce_match_stat_basic() always goes to the actual
2578
   * contents.  The caller checks with is_racy_timestamp() which
2579
   * always says "no" for gitlinks, so we are not called for them ;-)
2580
   */
2581
0
  struct stat st;
2582
2583
0
  if (lstat(ce->name, &st) < 0)
2584
0
    return;
2585
0
  if (ce_match_stat_basic(ce, &st))
2586
0
    return;
2587
0
  if (ce_modified_check_fs(istate, ce, &st)) {
2588
    /* This is "racily clean"; smudge it.  Note that this
2589
     * is a tricky code.  At first glance, it may appear
2590
     * that it can break with this sequence:
2591
     *
2592
     * $ echo xyzzy >frotz
2593
     * $ git-update-index --add frotz
2594
     * $ : >frotz
2595
     * $ sleep 3
2596
     * $ echo filfre >nitfol
2597
     * $ git-update-index --add nitfol
2598
     *
2599
     * but it does not.  When the second update-index runs,
2600
     * it notices that the entry "frotz" has the same timestamp
2601
     * as index, and if we were to smudge it by resetting its
2602
     * size to zero here, then the object name recorded
2603
     * in index is the 6-byte file but the cached stat information
2604
     * becomes zero --- which would then match what we would
2605
     * obtain from the filesystem next time we stat("frotz").
2606
     *
2607
     * However, the second update-index, before calling
2608
     * this function, notices that the cached size is 6
2609
     * bytes and what is on the filesystem is an empty
2610
     * file, and never calls us, so the cached size information
2611
     * for "frotz" stays 6 which does not match the filesystem.
2612
     */
2613
0
    ce->ce_stat_data.sd_size = 0;
2614
0
  }
2615
0
}
2616
2617
/* Copy miscellaneous fields but not the name */
2618
static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
2619
               struct cache_entry *ce)
2620
0
{
2621
0
  short flags;
2622
0
  const unsigned hashsz = the_hash_algo->rawsz;
2623
0
  uint16_t *flagsp = (uint16_t *)(ondisk->data + hashsz);
2624
2625
0
  ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2626
0
  ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2627
0
  ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2628
0
  ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2629
0
  ondisk->dev  = htonl(ce->ce_stat_data.sd_dev);
2630
0
  ondisk->ino  = htonl(ce->ce_stat_data.sd_ino);
2631
0
  ondisk->mode = htonl(ce->ce_mode);
2632
0
  ondisk->uid  = htonl(ce->ce_stat_data.sd_uid);
2633
0
  ondisk->gid  = htonl(ce->ce_stat_data.sd_gid);
2634
0
  ondisk->size = htonl(ce->ce_stat_data.sd_size);
2635
0
  hashcpy(ondisk->data, ce->oid.hash, the_repository->hash_algo);
2636
2637
0
  flags = ce->ce_flags & ~CE_NAMEMASK;
2638
0
  flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2639
0
  flagsp[0] = htons(flags);
2640
0
  if (ce->ce_flags & CE_EXTENDED) {
2641
0
    flagsp[1] = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
2642
0
  }
2643
0
}
2644
2645
static int ce_write_entry(struct hashfile *f, struct cache_entry *ce,
2646
        struct strbuf *previous_name, struct ondisk_cache_entry *ondisk)
2647
0
{
2648
0
  int size;
2649
0
  unsigned int saved_namelen;
2650
0
  int stripped_name = 0;
2651
0
  static unsigned char padding[8] = { 0x00 };
2652
2653
0
  if (ce->ce_flags & CE_STRIP_NAME) {
2654
0
    saved_namelen = ce_namelen(ce);
2655
0
    ce->ce_namelen = 0;
2656
0
    stripped_name = 1;
2657
0
  }
2658
2659
0
  size = offsetof(struct ondisk_cache_entry,data) + ondisk_data_size(ce->ce_flags, 0);
2660
2661
0
  if (!previous_name) {
2662
0
    int len = ce_namelen(ce);
2663
0
    copy_cache_entry_to_ondisk(ondisk, ce);
2664
0
    hashwrite(f, ondisk, size);
2665
0
    hashwrite(f, ce->name, len);
2666
0
    hashwrite(f, padding, align_padding_size(size, len));
2667
0
  } else {
2668
0
    int common, to_remove;
2669
0
    uint8_t prefix_size;
2670
0
    unsigned char to_remove_vi[16];
2671
2672
0
    for (common = 0;
2673
0
         (common < previous_name->len &&
2674
0
          ce->name[common] &&
2675
0
          ce->name[common] == previous_name->buf[common]);
2676
0
         common++)
2677
0
      ; /* still matching */
2678
0
    to_remove = previous_name->len - common;
2679
0
    prefix_size = encode_varint(to_remove, to_remove_vi);
2680
2681
0
    copy_cache_entry_to_ondisk(ondisk, ce);
2682
0
    hashwrite(f, ondisk, size);
2683
0
    hashwrite(f, to_remove_vi, prefix_size);
2684
0
    hashwrite(f, ce->name + common, ce_namelen(ce) - common);
2685
0
    hashwrite(f, padding, 1);
2686
2687
0
    strbuf_splice(previous_name, common, to_remove,
2688
0
            ce->name + common, ce_namelen(ce) - common);
2689
0
  }
2690
0
  if (stripped_name) {
2691
0
    ce->ce_namelen = saved_namelen;
2692
0
    ce->ce_flags &= ~CE_STRIP_NAME;
2693
0
  }
2694
2695
0
  return 0;
2696
0
}
2697
2698
/*
2699
 * This function verifies if index_state has the correct sha1 of the
2700
 * index file.  Don't die if we have any other failure, just return 0.
2701
 */
2702
static int verify_index_from(const struct index_state *istate, const char *path)
2703
0
{
2704
0
  int fd;
2705
0
  ssize_t n;
2706
0
  struct stat st;
2707
0
  unsigned char hash[GIT_MAX_RAWSZ];
2708
2709
0
  if (!istate->initialized)
2710
0
    return 0;
2711
2712
0
  fd = open(path, O_RDONLY);
2713
0
  if (fd < 0)
2714
0
    return 0;
2715
2716
0
  if (fstat(fd, &st))
2717
0
    goto out;
2718
2719
0
  if (st.st_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2720
0
    goto out;
2721
2722
0
  n = pread_in_full(fd, hash, the_hash_algo->rawsz, st.st_size - the_hash_algo->rawsz);
2723
0
  if (n != the_hash_algo->rawsz)
2724
0
    goto out;
2725
2726
0
  if (!hasheq(istate->oid.hash, hash, the_repository->hash_algo))
2727
0
    goto out;
2728
2729
0
  close(fd);
2730
0
  return 1;
2731
2732
0
out:
2733
0
  close(fd);
2734
0
  return 0;
2735
0
}
2736
2737
static int repo_verify_index(struct repository *repo)
2738
0
{
2739
0
  return verify_index_from(repo->index, repo->index_file);
2740
0
}
2741
2742
int has_racy_timestamp(struct index_state *istate)
2743
0
{
2744
0
  int entries = istate->cache_nr;
2745
0
  int i;
2746
2747
0
  for (i = 0; i < entries; i++) {
2748
0
    struct cache_entry *ce = istate->cache[i];
2749
0
    if (is_racy_timestamp(istate, ce))
2750
0
      return 1;
2751
0
  }
2752
0
  return 0;
2753
0
}
2754
2755
void repo_update_index_if_able(struct repository *repo,
2756
             struct lock_file *lockfile)
2757
0
{
2758
0
  if ((repo->index->cache_changed ||
2759
0
       has_racy_timestamp(repo->index)) &&
2760
0
      repo_verify_index(repo))
2761
0
    write_locked_index(repo->index, lockfile, COMMIT_LOCK);
2762
0
  else
2763
0
    rollback_lock_file(lockfile);
2764
0
}
2765
2766
static int record_eoie(void)
2767
0
{
2768
0
  int val;
2769
2770
0
  if (!repo_config_get_bool(the_repository, "index.recordendofindexentries", &val))
2771
0
    return val;
2772
2773
  /*
2774
   * As a convenience, the end of index entries extension
2775
   * used for threading is written by default if the user
2776
   * explicitly requested threaded index reads.
2777
   */
2778
0
  return !repo_config_get_index_threads(the_repository, &val) && val != 1;
2779
0
}
2780
2781
static int record_ieot(void)
2782
0
{
2783
0
  int val;
2784
2785
0
  if (!repo_config_get_bool(the_repository, "index.recordoffsettable", &val))
2786
0
    return val;
2787
2788
  /*
2789
   * As a convenience, the offset table used for threading is
2790
   * written by default if the user explicitly requested
2791
   * threaded index reads.
2792
   */
2793
0
  return !repo_config_get_index_threads(the_repository, &val) && val != 1;
2794
0
}
2795
2796
enum write_extensions {
2797
  WRITE_NO_EXTENSION =              0,
2798
  WRITE_SPLIT_INDEX_EXTENSION =     1<<0,
2799
  WRITE_CACHE_TREE_EXTENSION =      1<<1,
2800
  WRITE_RESOLVE_UNDO_EXTENSION =    1<<2,
2801
  WRITE_UNTRACKED_CACHE_EXTENSION = 1<<3,
2802
  WRITE_FSMONITOR_EXTENSION =       1<<4,
2803
};
2804
0
#define WRITE_ALL_EXTENSIONS ((enum write_extensions)-1)
2805
2806
/*
2807
 * On success, `tempfile` is closed. If it is the temporary file
2808
 * of a `struct lock_file`, we will therefore effectively perform
2809
 * a 'close_lock_file_gently()`. Since that is an implementation
2810
 * detail of lockfiles, callers of `do_write_index()` should not
2811
 * rely on it.
2812
 */
2813
static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
2814
        enum write_extensions write_extensions, unsigned flags)
2815
0
{
2816
0
  uint64_t start = getnanotime();
2817
0
  struct hashfile *f;
2818
0
  struct git_hash_ctx *eoie_c = NULL;
2819
0
  struct cache_header hdr;
2820
0
  int i, err = 0, removed, extended, hdr_version;
2821
0
  struct cache_entry **cache = istate->cache;
2822
0
  int entries = istate->cache_nr;
2823
0
  struct stat st;
2824
0
  struct ondisk_cache_entry ondisk;
2825
0
  struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2826
0
  int drop_cache_tree = istate->drop_cache_tree;
2827
0
  off_t offset;
2828
0
  int csum_fsync_flag;
2829
0
  int ieot_entries = 1;
2830
0
  struct index_entry_offset_table *ieot = NULL;
2831
0
  struct repository *r = istate->repo;
2832
0
  struct strbuf sb = STRBUF_INIT;
2833
0
  int nr, nr_threads, ret;
2834
2835
0
  f = hashfd(the_repository->hash_algo, tempfile->fd, tempfile->filename.buf);
2836
2837
0
  prepare_repo_settings(r);
2838
0
  f->skip_hash = r->settings.index_skip_hash;
2839
2840
0
  for (i = removed = extended = 0; i < entries; i++) {
2841
0
    if (cache[i]->ce_flags & CE_REMOVE)
2842
0
      removed++;
2843
2844
    /* reduce extended entries if possible */
2845
0
    cache[i]->ce_flags &= ~CE_EXTENDED;
2846
0
    if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2847
0
      extended++;
2848
0
      cache[i]->ce_flags |= CE_EXTENDED;
2849
0
    }
2850
0
  }
2851
2852
0
  if (!istate->version)
2853
0
    istate->version = get_index_format_default(r);
2854
2855
  /* demote version 3 to version 2 when the latter suffices */
2856
0
  if (istate->version == 3 || istate->version == 2)
2857
0
    istate->version = extended ? 3 : 2;
2858
2859
0
  hdr_version = istate->version;
2860
2861
0
  hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2862
0
  hdr.hdr_version = htonl(hdr_version);
2863
0
  hdr.hdr_entries = htonl(entries - removed);
2864
2865
0
  hashwrite(f, &hdr, sizeof(hdr));
2866
2867
0
  if (!HAVE_THREADS || repo_config_get_index_threads(the_repository, &nr_threads))
2868
0
    nr_threads = 1;
2869
2870
0
  if (nr_threads != 1 && record_ieot()) {
2871
0
    int ieot_blocks, cpus;
2872
2873
    /*
2874
     * ensure default number of ieot blocks maps evenly to the
2875
     * default number of threads that will process them leaving
2876
     * room for the thread to load the index extensions.
2877
     */
2878
0
    if (!nr_threads) {
2879
0
      ieot_blocks = istate->cache_nr / THREAD_COST;
2880
0
      cpus = online_cpus();
2881
0
      if (ieot_blocks > cpus - 1)
2882
0
        ieot_blocks = cpus - 1;
2883
0
    } else {
2884
0
      ieot_blocks = nr_threads;
2885
0
      if (ieot_blocks > istate->cache_nr)
2886
0
        ieot_blocks = istate->cache_nr;
2887
0
    }
2888
2889
    /*
2890
     * no reason to write out the IEOT extension if we don't
2891
     * have enough blocks to utilize multi-threading
2892
     */
2893
0
    if (ieot_blocks > 1) {
2894
0
      ieot = xcalloc(1, sizeof(struct index_entry_offset_table)
2895
0
        + (ieot_blocks * sizeof(struct index_entry_offset)));
2896
0
      ieot_entries = DIV_ROUND_UP(entries, ieot_blocks);
2897
0
    }
2898
0
  }
2899
2900
0
  offset = hashfile_total(f);
2901
2902
0
  nr = 0;
2903
0
  previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2904
2905
0
  for (i = 0; i < entries; i++) {
2906
0
    struct cache_entry *ce = cache[i];
2907
0
    if (ce->ce_flags & CE_REMOVE)
2908
0
      continue;
2909
0
    if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2910
0
      ce_smudge_racily_clean_entry(istate, ce);
2911
0
    if (is_null_oid(&ce->oid)) {
2912
0
      static const char msg[] = "cache entry has null sha1: %s";
2913
0
      static int allow = -1;
2914
2915
0
      if (allow < 0)
2916
0
        allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2917
0
      if (allow)
2918
0
        warning(msg, ce->name);
2919
0
      else
2920
0
        err = error(msg, ce->name);
2921
2922
0
      drop_cache_tree = 1;
2923
0
    }
2924
0
    if (ieot && i && (i % ieot_entries == 0)) {
2925
0
      ieot->entries[ieot->nr].nr = nr;
2926
0
      ieot->entries[ieot->nr].offset = offset;
2927
0
      ieot->nr++;
2928
      /*
2929
       * If we have a V4 index, set the first byte to an invalid
2930
       * character to ensure there is nothing common with the previous
2931
       * entry
2932
       */
2933
0
      if (previous_name)
2934
0
        previous_name->buf[0] = 0;
2935
0
      nr = 0;
2936
2937
0
      offset = hashfile_total(f);
2938
0
    }
2939
0
    if (ce_write_entry(f, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
2940
0
      err = -1;
2941
2942
0
    if (err)
2943
0
      break;
2944
0
    nr++;
2945
0
  }
2946
0
  if (ieot && nr) {
2947
0
    ieot->entries[ieot->nr].nr = nr;
2948
0
    ieot->entries[ieot->nr].offset = offset;
2949
0
    ieot->nr++;
2950
0
  }
2951
0
  strbuf_release(&previous_name_buf);
2952
2953
0
  if (err) {
2954
0
    ret = err;
2955
0
    goto out;
2956
0
  }
2957
2958
0
  offset = hashfile_total(f);
2959
2960
  /*
2961
   * The extension headers must be hashed on their own for the
2962
   * EOIE extension. Create a hashfile here to compute that hash.
2963
   */
2964
0
  if (offset && record_eoie()) {
2965
0
    CALLOC_ARRAY(eoie_c, 1);
2966
0
    the_hash_algo->init_fn(eoie_c);
2967
0
  }
2968
2969
  /*
2970
   * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2971
   * can minimize the number of extensions we have to scan through to
2972
   * find it during load.  Write it out regardless of the
2973
   * strip_extensions parameter as we need it when loading the shared
2974
   * index.
2975
   */
2976
0
  if (ieot) {
2977
0
    strbuf_reset(&sb);
2978
2979
0
    write_ieot_extension(&sb, ieot);
2980
0
    err = write_index_ext_header(f, eoie_c, CACHE_EXT_INDEXENTRYOFFSETTABLE, sb.len) < 0;
2981
0
    hashwrite(f, sb.buf, sb.len);
2982
0
    if (err) {
2983
0
      ret = -1;
2984
0
      goto out;
2985
0
    }
2986
0
  }
2987
2988
0
  if (write_extensions & WRITE_SPLIT_INDEX_EXTENSION &&
2989
0
      istate->split_index) {
2990
0
    strbuf_reset(&sb);
2991
2992
0
    if (istate->sparse_index)
2993
0
      die(_("cannot write split index for a sparse index"));
2994
2995
0
    err = write_link_extension(&sb, istate) < 0 ||
2996
0
      write_index_ext_header(f, eoie_c, CACHE_EXT_LINK,
2997
0
                 sb.len) < 0;
2998
0
    hashwrite(f, sb.buf, sb.len);
2999
0
    if (err) {
3000
0
      ret = -1;
3001
0
      goto out;
3002
0
    }
3003
0
  }
3004
0
  if (write_extensions & WRITE_CACHE_TREE_EXTENSION &&
3005
0
      !drop_cache_tree && istate->cache_tree) {
3006
0
    strbuf_reset(&sb);
3007
3008
0
    cache_tree_write(&sb, istate->cache_tree);
3009
0
    err = write_index_ext_header(f, eoie_c, CACHE_EXT_TREE, sb.len) < 0;
3010
0
    hashwrite(f, sb.buf, sb.len);
3011
0
    if (err) {
3012
0
      ret = -1;
3013
0
      goto out;
3014
0
    }
3015
0
  }
3016
0
  if (write_extensions & WRITE_RESOLVE_UNDO_EXTENSION &&
3017
0
      istate->resolve_undo) {
3018
0
    strbuf_reset(&sb);
3019
3020
0
    resolve_undo_write(&sb, istate->resolve_undo, the_hash_algo);
3021
0
    err = write_index_ext_header(f, eoie_c, CACHE_EXT_RESOLVE_UNDO,
3022
0
               sb.len) < 0;
3023
0
    hashwrite(f, sb.buf, sb.len);
3024
0
    if (err) {
3025
0
      ret = -1;
3026
0
      goto out;
3027
0
    }
3028
0
  }
3029
0
  if (write_extensions & WRITE_UNTRACKED_CACHE_EXTENSION &&
3030
0
      istate->untracked) {
3031
0
    strbuf_reset(&sb);
3032
3033
0
    write_untracked_extension(&sb, istate->untracked);
3034
0
    err = write_index_ext_header(f, eoie_c, CACHE_EXT_UNTRACKED,
3035
0
               sb.len) < 0;
3036
0
    hashwrite(f, sb.buf, sb.len);
3037
0
    if (err) {
3038
0
      ret = -1;
3039
0
      goto out;
3040
0
    }
3041
0
  }
3042
0
  if (write_extensions & WRITE_FSMONITOR_EXTENSION &&
3043
0
      istate->fsmonitor_last_update) {
3044
0
    strbuf_reset(&sb);
3045
3046
0
    write_fsmonitor_extension(&sb, istate);
3047
0
    err = write_index_ext_header(f, eoie_c, CACHE_EXT_FSMONITOR, sb.len) < 0;
3048
0
    hashwrite(f, sb.buf, sb.len);
3049
0
    if (err) {
3050
0
      ret = -1;
3051
0
      goto out;
3052
0
    }
3053
0
  }
3054
0
  if (istate->sparse_index) {
3055
0
    if (write_index_ext_header(f, eoie_c, CACHE_EXT_SPARSE_DIRECTORIES, 0) < 0) {
3056
0
      ret = -1;
3057
0
      goto out;
3058
0
    }
3059
0
  }
3060
3061
  /*
3062
   * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
3063
   * so that it can be found and processed before all the index entries are
3064
   * read.  Write it out regardless of the strip_extensions parameter as we need it
3065
   * when loading the shared index.
3066
   */
3067
0
  if (eoie_c) {
3068
0
    strbuf_reset(&sb);
3069
3070
0
    write_eoie_extension(&sb, eoie_c, offset);
3071
0
    err = write_index_ext_header(f, NULL, CACHE_EXT_ENDOFINDEXENTRIES, sb.len) < 0;
3072
0
    hashwrite(f, sb.buf, sb.len);
3073
0
    if (err) {
3074
0
      ret = -1;
3075
0
      goto out;
3076
0
    }
3077
0
  }
3078
3079
0
  csum_fsync_flag = 0;
3080
0
  if (!alternate_index_output && (flags & COMMIT_LOCK))
3081
0
    csum_fsync_flag = CSUM_FSYNC;
3082
3083
0
  finalize_hashfile(f, istate->oid.hash, FSYNC_COMPONENT_INDEX,
3084
0
        CSUM_HASH_IN_STREAM | csum_fsync_flag);
3085
0
  f = NULL;
3086
3087
0
  if (close_tempfile_gently(tempfile)) {
3088
0
    ret = error(_("could not close '%s'"), get_tempfile_path(tempfile));
3089
0
    goto out;
3090
0
  }
3091
0
  if (stat(get_tempfile_path(tempfile), &st)) {
3092
0
    ret = -1;
3093
0
    goto out;
3094
0
  }
3095
0
  istate->timestamp.sec = (unsigned int)st.st_mtime;
3096
0
  istate->timestamp.nsec = ST_MTIME_NSEC(st);
3097
0
  trace_performance_since(start, "write index, changed mask = %x", istate->cache_changed);
3098
3099
  /*
3100
   * TODO trace2: replace "the_repository" with the actual repo instance
3101
   * that is associated with the given "istate".
3102
   */
3103
0
  trace2_data_intmax("index", the_repository, "write/version",
3104
0
         istate->version);
3105
0
  trace2_data_intmax("index", the_repository, "write/cache_nr",
3106
0
         istate->cache_nr);
3107
3108
0
  ret = 0;
3109
3110
0
out:
3111
0
  if (f)
3112
0
    free_hashfile(f);
3113
0
  strbuf_release(&sb);
3114
0
  free(eoie_c);
3115
0
  free(ieot);
3116
0
  return ret;
3117
0
}
3118
3119
void set_alternate_index_output(const char *name)
3120
0
{
3121
0
  alternate_index_output = name;
3122
0
}
3123
3124
static int commit_locked_index(struct lock_file *lk)
3125
0
{
3126
0
  if (alternate_index_output)
3127
0
    return commit_lock_file_to(lk, alternate_index_output);
3128
0
  else
3129
0
    return commit_lock_file(lk);
3130
0
}
3131
3132
static int do_write_locked_index(struct index_state *istate,
3133
         struct lock_file *lock,
3134
         unsigned flags,
3135
         enum write_extensions write_extensions)
3136
0
{
3137
0
  int ret;
3138
0
  int was_full = istate->sparse_index == INDEX_EXPANDED;
3139
3140
0
  ret = convert_to_sparse(istate, 0);
3141
3142
0
  if (ret) {
3143
0
    warning(_("failed to convert to a sparse-index"));
3144
0
    return ret;
3145
0
  }
3146
3147
  /*
3148
   * TODO trace2: replace "the_repository" with the actual repo instance
3149
   * that is associated with the given "istate".
3150
   */
3151
0
  trace2_region_enter_printf("index", "do_write_index", the_repository,
3152
0
           "%s", get_lock_file_path(lock));
3153
0
  ret = do_write_index(istate, lock->tempfile, write_extensions, flags);
3154
0
  trace2_region_leave_printf("index", "do_write_index", the_repository,
3155
0
           "%s", get_lock_file_path(lock));
3156
3157
0
  if (was_full)
3158
0
    ensure_full_index(istate);
3159
3160
0
  if (ret)
3161
0
    return ret;
3162
0
  if (flags & COMMIT_LOCK)
3163
0
    ret = commit_locked_index(lock);
3164
0
  else
3165
0
    ret = close_lock_file_gently(lock);
3166
3167
0
  run_hooks_l(the_repository, "post-index-change",
3168
0
        istate->updated_workdir ? "1" : "0",
3169
0
        istate->updated_skipworktree ? "1" : "0", NULL);
3170
0
  istate->updated_workdir = 0;
3171
0
  istate->updated_skipworktree = 0;
3172
3173
0
  return ret;
3174
0
}
3175
3176
static int write_split_index(struct index_state *istate,
3177
           struct lock_file *lock,
3178
           unsigned flags)
3179
0
{
3180
0
  int ret;
3181
0
  prepare_to_write_split_index(istate);
3182
0
  ret = do_write_locked_index(istate, lock, flags, WRITE_ALL_EXTENSIONS);
3183
0
  finish_writing_split_index(istate);
3184
0
  return ret;
3185
0
}
3186
3187
static unsigned long get_shared_index_expire_date(void)
3188
0
{
3189
0
  static unsigned long shared_index_expire_date;
3190
0
  static int shared_index_expire_date_prepared;
3191
3192
0
  if (!shared_index_expire_date_prepared) {
3193
0
    const char *shared_index_expire = "2.weeks.ago";
3194
0
    char *value = NULL;
3195
3196
0
    repo_config_get_expiry(the_repository, "splitindex.sharedindexexpire",
3197
0
               &value);
3198
0
    if (value)
3199
0
      shared_index_expire = value;
3200
3201
0
    shared_index_expire_date = approxidate(shared_index_expire);
3202
0
    shared_index_expire_date_prepared = 1;
3203
3204
0
    free(value);
3205
0
  }
3206
3207
0
  return shared_index_expire_date;
3208
0
}
3209
3210
static int should_delete_shared_index(const char *shared_index_path)
3211
0
{
3212
0
  struct stat st;
3213
0
  unsigned long expiration;
3214
3215
  /* Check timestamp */
3216
0
  expiration = get_shared_index_expire_date();
3217
0
  if (!expiration)
3218
0
    return 0;
3219
0
  if (stat(shared_index_path, &st))
3220
0
    return error_errno(_("could not stat '%s'"), shared_index_path);
3221
0
  if (st.st_mtime > expiration)
3222
0
    return 0;
3223
3224
0
  return 1;
3225
0
}
3226
3227
static int clean_shared_index_files(const char *current_hex)
3228
0
{
3229
0
  struct dirent *de;
3230
0
  DIR *dir = opendir(repo_get_git_dir(the_repository));
3231
3232
0
  if (!dir)
3233
0
    return error_errno(_("unable to open git dir: %s"),
3234
0
           repo_get_git_dir(the_repository));
3235
3236
0
  while ((de = readdir(dir)) != NULL) {
3237
0
    const char *sha1_hex;
3238
0
    char *shared_index_path;
3239
0
    if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
3240
0
      continue;
3241
0
    if (!strcmp(sha1_hex, current_hex))
3242
0
      continue;
3243
3244
0
    shared_index_path = repo_git_path(the_repository, "%s", de->d_name);
3245
0
    if (should_delete_shared_index(shared_index_path) > 0 &&
3246
0
        unlink(shared_index_path))
3247
0
      warning_errno(_("unable to unlink: %s"), shared_index_path);
3248
3249
0
    free(shared_index_path);
3250
0
  }
3251
0
  closedir(dir);
3252
3253
0
  return 0;
3254
0
}
3255
3256
static int write_shared_index(struct index_state *istate,
3257
            struct tempfile **temp, unsigned flags)
3258
0
{
3259
0
  struct split_index *si = istate->split_index;
3260
0
  int ret, was_full = !istate->sparse_index;
3261
0
  char *path;
3262
3263
0
  move_cache_to_base_index(istate);
3264
0
  convert_to_sparse(istate, 0);
3265
3266
0
  trace2_region_enter_printf("index", "shared/do_write_index",
3267
0
           the_repository, "%s", get_tempfile_path(*temp));
3268
0
  ret = do_write_index(si->base, *temp, WRITE_NO_EXTENSION, flags);
3269
0
  trace2_region_leave_printf("index", "shared/do_write_index",
3270
0
           the_repository, "%s", get_tempfile_path(*temp));
3271
3272
0
  if (was_full)
3273
0
    ensure_full_index(istate);
3274
3275
0
  if (ret)
3276
0
    return ret;
3277
0
  ret = adjust_shared_perm(the_repository, get_tempfile_path(*temp));
3278
0
  if (ret) {
3279
0
    error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp));
3280
0
    return ret;
3281
0
  }
3282
3283
0
  path = repo_git_path(the_repository, "sharedindex.%s", oid_to_hex(&si->base->oid));
3284
0
  ret = rename_tempfile(temp, path);
3285
0
  if (!ret) {
3286
0
    oidcpy(&si->base_oid, &si->base->oid);
3287
0
    clean_shared_index_files(oid_to_hex(&si->base->oid));
3288
0
  }
3289
3290
0
  free(path);
3291
0
  return ret;
3292
0
}
3293
3294
static const int default_max_percent_split_change = 20;
3295
3296
static int too_many_not_shared_entries(struct index_state *istate)
3297
0
{
3298
0
  int i, not_shared = 0;
3299
0
  int max_split = repo_config_get_max_percent_split_change(the_repository);
3300
3301
0
  switch (max_split) {
3302
0
  case -1:
3303
    /* not or badly configured: use the default value */
3304
0
    max_split = default_max_percent_split_change;
3305
0
    break;
3306
0
  case 0:
3307
0
    return 1; /* 0% means always write a new shared index */
3308
0
  case 100:
3309
0
    return 0; /* 100% means never write a new shared index */
3310
0
  default:
3311
0
    break; /* just use the configured value */
3312
0
  }
3313
3314
  /* Count not shared entries */
3315
0
  for (i = 0; i < istate->cache_nr; i++) {
3316
0
    struct cache_entry *ce = istate->cache[i];
3317
0
    if (!ce->index)
3318
0
      not_shared++;
3319
0
  }
3320
3321
0
  return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
3322
0
}
3323
3324
int write_locked_index(struct index_state *istate, struct lock_file *lock,
3325
           unsigned flags)
3326
0
{
3327
0
  int new_shared_index, ret, test_split_index_env;
3328
0
  struct split_index *si = istate->split_index;
3329
3330
0
  if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0) &&
3331
0
      cache_tree_verify(the_repository, istate) < 0)
3332
0
    return -1;
3333
3334
0
  if ((flags & SKIP_IF_UNCHANGED) && !istate->cache_changed) {
3335
0
    if (flags & COMMIT_LOCK)
3336
0
      rollback_lock_file(lock);
3337
0
    return 0;
3338
0
  }
3339
3340
0
  if (istate->fsmonitor_last_update)
3341
0
    fill_fsmonitor_bitmap(istate);
3342
3343
0
  test_split_index_env = git_env_bool("GIT_TEST_SPLIT_INDEX", 0);
3344
3345
0
  if ((!si && !test_split_index_env) ||
3346
0
      alternate_index_output ||
3347
0
      (istate->cache_changed & ~EXTMASK)) {
3348
0
    ret = do_write_locked_index(istate, lock, flags,
3349
0
              ~WRITE_SPLIT_INDEX_EXTENSION);
3350
0
    goto out;
3351
0
  }
3352
3353
0
  if (test_split_index_env) {
3354
0
    if (!si) {
3355
0
      si = init_split_index(istate);
3356
0
      istate->cache_changed |= SPLIT_INDEX_ORDERED;
3357
0
    } else {
3358
0
      int v = si->base_oid.hash[0];
3359
0
      if ((v & 15) < 6)
3360
0
        istate->cache_changed |= SPLIT_INDEX_ORDERED;
3361
0
    }
3362
0
  }
3363
0
  if (too_many_not_shared_entries(istate))
3364
0
    istate->cache_changed |= SPLIT_INDEX_ORDERED;
3365
3366
0
  new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
3367
3368
0
  if (new_shared_index) {
3369
0
    struct tempfile *temp;
3370
0
    int saved_errno;
3371
0
    char *path;
3372
3373
    /* Same initial permissions as the main .git/index file */
3374
0
    path = repo_git_path(the_repository, "sharedindex_XXXXXX");
3375
0
    temp = mks_tempfile_sm(path, 0, 0666);
3376
0
    free(path);
3377
0
    if (!temp) {
3378
0
      ret = do_write_locked_index(istate, lock, flags,
3379
0
                ~WRITE_SPLIT_INDEX_EXTENSION);
3380
0
      goto out;
3381
0
    }
3382
0
    ret = write_shared_index(istate, &temp, flags);
3383
3384
0
    saved_errno = errno;
3385
0
    if (is_tempfile_active(temp))
3386
0
      delete_tempfile(&temp);
3387
0
    errno = saved_errno;
3388
3389
0
    if (ret)
3390
0
      goto out;
3391
0
  }
3392
3393
0
  ret = write_split_index(istate, lock, flags);
3394
3395
  /* Freshen the shared index only if the split-index was written */
3396
0
  if (!ret && !new_shared_index && !is_null_oid(&si->base_oid)) {
3397
0
    char *shared_index = repo_git_path(the_repository, "sharedindex.%s",
3398
0
               oid_to_hex(&si->base_oid));
3399
0
    freshen_shared_index(shared_index, 1);
3400
0
    free(shared_index);
3401
0
  }
3402
3403
0
out:
3404
0
  if (flags & COMMIT_LOCK)
3405
0
    rollback_lock_file(lock);
3406
0
  return ret;
3407
0
}
3408
3409
/*
3410
 * Read the index file that is potentially unmerged into given
3411
 * index_state, dropping any unmerged entries to stage #0 (potentially
3412
 * resulting in a path appearing as both a file and a directory in the
3413
 * index; the caller is responsible to clear out the extra entries
3414
 * before writing the index to a tree).  Returns true if the index is
3415
 * unmerged.  Callers who want to refuse to work from an unmerged
3416
 * state can call this and check its return value, instead of calling
3417
 * read_cache().
3418
 */
3419
int repo_read_index_unmerged(struct repository *repo)
3420
0
{
3421
0
  struct index_state *istate;
3422
0
  int i;
3423
0
  int unmerged = 0;
3424
3425
0
  repo_read_index(repo);
3426
0
  istate = repo->index;
3427
0
  for (i = 0; i < istate->cache_nr; i++) {
3428
0
    struct cache_entry *ce = istate->cache[i];
3429
0
    struct cache_entry *new_ce;
3430
0
    int len;
3431
3432
0
    if (!ce_stage(ce))
3433
0
      continue;
3434
0
    unmerged = 1;
3435
0
    len = ce_namelen(ce);
3436
0
    new_ce = make_empty_cache_entry(istate, len);
3437
0
    memcpy(new_ce->name, ce->name, len);
3438
0
    new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
3439
0
    new_ce->ce_namelen = len;
3440
0
    new_ce->ce_mode = ce->ce_mode;
3441
0
    if (add_index_entry(istate, new_ce, ADD_CACHE_SKIP_DFCHECK))
3442
0
      return error(_("%s: cannot drop to stage #0"),
3443
0
             new_ce->name);
3444
0
  }
3445
0
  return unmerged;
3446
0
}
3447
3448
/*
3449
 * Returns 1 if the path is an "other" path with respect to
3450
 * the index; that is, the path is not mentioned in the index at all,
3451
 * either as a file, a directory with some files in the index,
3452
 * or as an unmerged entry.
3453
 *
3454
 * We helpfully remove a trailing "/" from directories so that
3455
 * the output of read_directory can be used as-is.
3456
 */
3457
int index_name_is_other(struct index_state *istate, const char *name,
3458
      int namelen)
3459
0
{
3460
0
  int pos;
3461
0
  if (namelen && name[namelen - 1] == '/')
3462
0
    namelen--;
3463
0
  pos = index_name_pos(istate, name, namelen);
3464
0
  if (0 <= pos)
3465
0
    return 0; /* exact match */
3466
0
  pos = -pos - 1;
3467
0
  if (pos < istate->cache_nr) {
3468
0
    struct cache_entry *ce = istate->cache[pos];
3469
0
    if (ce_namelen(ce) == namelen &&
3470
0
        !memcmp(ce->name, name, namelen))
3471
0
      return 0; /* Yup, this one exists unmerged */
3472
0
  }
3473
0
  return 1;
3474
0
}
3475
3476
void *read_blob_data_from_index(struct index_state *istate,
3477
        const char *path, unsigned long *size)
3478
0
{
3479
0
  int pos, len;
3480
0
  unsigned long sz;
3481
0
  enum object_type type;
3482
0
  void *data;
3483
3484
0
  len = strlen(path);
3485
0
  pos = index_name_pos(istate, path, len);
3486
0
  if (pos < 0) {
3487
    /*
3488
     * We might be in the middle of a merge, in which
3489
     * case we would read stage #2 (ours).
3490
     */
3491
0
    int i;
3492
0
    for (i = -pos - 1;
3493
0
         (pos < 0 && i < istate->cache_nr &&
3494
0
          !strcmp(istate->cache[i]->name, path));
3495
0
         i++)
3496
0
      if (ce_stage(istate->cache[i]) == 2)
3497
0
        pos = i;
3498
0
  }
3499
0
  if (pos < 0)
3500
0
    return NULL;
3501
0
  data = odb_read_object(the_repository->objects, &istate->cache[pos]->oid,
3502
0
             &type, &sz);
3503
0
  if (!data || type != OBJ_BLOB) {
3504
0
    free(data);
3505
0
    return NULL;
3506
0
  }
3507
0
  if (size)
3508
0
    *size = sz;
3509
0
  return data;
3510
0
}
3511
3512
void move_index_extensions(struct index_state *dst, struct index_state *src)
3513
0
{
3514
0
  dst->untracked = src->untracked;
3515
0
  src->untracked = NULL;
3516
0
  dst->cache_tree = src->cache_tree;
3517
0
  src->cache_tree = NULL;
3518
0
}
3519
3520
struct cache_entry *dup_cache_entry(const struct cache_entry *ce,
3521
            struct index_state *istate)
3522
0
{
3523
0
  unsigned int size = ce_size(ce);
3524
0
  int mem_pool_allocated;
3525
0
  struct cache_entry *new_entry = make_empty_cache_entry(istate, ce_namelen(ce));
3526
0
  mem_pool_allocated = new_entry->mem_pool_allocated;
3527
3528
0
  memcpy(new_entry, ce, size);
3529
0
  new_entry->mem_pool_allocated = mem_pool_allocated;
3530
0
  return new_entry;
3531
0
}
3532
3533
void discard_cache_entry(struct cache_entry *ce)
3534
0
{
3535
0
  if (ce && should_validate_cache_entries())
3536
0
    memset(ce, 0xCD, cache_entry_size(ce->ce_namelen));
3537
3538
0
  if (ce && ce->mem_pool_allocated)
3539
0
    return;
3540
3541
0
  free(ce);
3542
0
}
3543
3544
int should_validate_cache_entries(void)
3545
0
{
3546
0
  static int validate_index_cache_entries = -1;
3547
3548
0
  if (validate_index_cache_entries < 0) {
3549
0
    if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3550
0
      validate_index_cache_entries = 1;
3551
0
    else
3552
0
      validate_index_cache_entries = 0;
3553
0
  }
3554
3555
0
  return validate_index_cache_entries;
3556
0
}
3557
3558
0
#define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3559
0
#define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3560
3561
static size_t read_eoie_extension(const char *mmap, size_t mmap_size)
3562
0
{
3563
  /*
3564
   * The end of index entries (EOIE) extension is guaranteed to be last
3565
   * so that it can be found by scanning backwards from the EOF.
3566
   *
3567
   * "EOIE"
3568
   * <4-byte length>
3569
   * <4-byte offset>
3570
   * <20-byte hash>
3571
   */
3572
0
  const char *index, *eoie;
3573
0
  uint32_t extsize;
3574
0
  size_t offset, src_offset;
3575
0
  unsigned char hash[GIT_MAX_RAWSZ];
3576
0
  struct git_hash_ctx c;
3577
3578
  /* ensure we have an index big enough to contain an EOIE extension */
3579
0
  if (mmap_size < sizeof(struct cache_header) + EOIE_SIZE_WITH_HEADER + the_hash_algo->rawsz)
3580
0
    return 0;
3581
3582
  /* validate the extension signature */
3583
0
  index = eoie = mmap + mmap_size - EOIE_SIZE_WITH_HEADER - the_hash_algo->rawsz;
3584
0
  if (CACHE_EXT(index) != CACHE_EXT_ENDOFINDEXENTRIES)
3585
0
    return 0;
3586
0
  index += sizeof(uint32_t);
3587
3588
  /* validate the extension size */
3589
0
  extsize = get_be32(index);
3590
0
  if (extsize != EOIE_SIZE)
3591
0
    return 0;
3592
0
  index += sizeof(uint32_t);
3593
3594
  /*
3595
   * Validate the offset we're going to look for the first extension
3596
   * signature is after the index header and before the eoie extension.
3597
   */
3598
0
  offset = get_be32(index);
3599
0
  if (mmap + offset < mmap + sizeof(struct cache_header))
3600
0
    return 0;
3601
0
  if (mmap + offset >= eoie)
3602
0
    return 0;
3603
0
  index += sizeof(uint32_t);
3604
3605
  /*
3606
   * The hash is computed over extension types and their sizes (but not
3607
   * their contents).  E.g. if we have "TREE" extension that is N-bytes
3608
   * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3609
   * then the hash would be:
3610
   *
3611
   * SHA-1("TREE" + <binary representation of N> +
3612
   *   "REUC" + <binary representation of M>)
3613
   */
3614
0
  src_offset = offset;
3615
0
  the_hash_algo->init_fn(&c);
3616
0
  while (src_offset < mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER) {
3617
    /* After an array of active_nr index entries,
3618
     * there can be arbitrary number of extended
3619
     * sections, each of which is prefixed with
3620
     * extension name (4-byte) and section length
3621
     * in 4-byte network byte order.
3622
     */
3623
0
    uint32_t extsize;
3624
0
    memcpy(&extsize, mmap + src_offset + 4, 4);
3625
0
    extsize = ntohl(extsize);
3626
3627
    /* verify the extension size isn't so large it will wrap around */
3628
0
    if (src_offset + 8 + extsize < src_offset)
3629
0
      return 0;
3630
3631
0
    git_hash_update(&c, mmap + src_offset, 8);
3632
3633
0
    src_offset += 8;
3634
0
    src_offset += extsize;
3635
0
  }
3636
0
  git_hash_final(hash, &c);
3637
0
  if (!hasheq(hash, (const unsigned char *)index, the_repository->hash_algo))
3638
0
    return 0;
3639
3640
  /* Validate that the extension offsets returned us back to the eoie extension. */
3641
0
  if (src_offset != mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER)
3642
0
    return 0;
3643
3644
0
  return offset;
3645
0
}
3646
3647
static void write_eoie_extension(struct strbuf *sb, struct git_hash_ctx *eoie_context, size_t offset)
3648
0
{
3649
0
  uint32_t buffer;
3650
0
  unsigned char hash[GIT_MAX_RAWSZ];
3651
3652
  /* offset */
3653
0
  put_be32(&buffer, offset);
3654
0
  strbuf_add(sb, &buffer, sizeof(uint32_t));
3655
3656
  /* hash */
3657
0
  git_hash_final(hash, eoie_context);
3658
0
  strbuf_add(sb, hash, the_hash_algo->rawsz);
3659
0
}
3660
3661
0
#define IEOT_VERSION  (1)
3662
3663
static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset)
3664
0
{
3665
0
  const char *index = NULL;
3666
0
  uint32_t extsize, ext_version;
3667
0
  struct index_entry_offset_table *ieot;
3668
0
  int i, nr;
3669
3670
  /* find the IEOT extension */
3671
0
  if (!offset)
3672
0
    return NULL;
3673
0
  while (offset <= mmap_size - the_hash_algo->rawsz - 8) {
3674
0
    extsize = get_be32(mmap + offset + 4);
3675
0
    if (CACHE_EXT((mmap + offset)) == CACHE_EXT_INDEXENTRYOFFSETTABLE) {
3676
0
      index = mmap + offset + 4 + 4;
3677
0
      break;
3678
0
    }
3679
0
    offset += 8;
3680
0
    offset += extsize;
3681
0
  }
3682
0
  if (!index)
3683
0
    return NULL;
3684
3685
  /* validate the version is IEOT_VERSION */
3686
0
  ext_version = get_be32(index);
3687
0
  if (ext_version != IEOT_VERSION) {
3688
0
    error("invalid IEOT version %d", ext_version);
3689
0
    return NULL;
3690
0
  }
3691
0
  index += sizeof(uint32_t);
3692
3693
  /* extension size - version bytes / bytes per entry */
3694
0
  nr = (extsize - sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3695
0
  if (!nr) {
3696
0
    error("invalid number of IEOT entries %d", nr);
3697
0
    return NULL;
3698
0
  }
3699
0
  ieot = xmalloc(sizeof(struct index_entry_offset_table)
3700
0
           + (nr * sizeof(struct index_entry_offset)));
3701
0
  ieot->nr = nr;
3702
0
  for (i = 0; i < nr; i++) {
3703
0
    ieot->entries[i].offset = get_be32(index);
3704
0
    index += sizeof(uint32_t);
3705
0
    ieot->entries[i].nr = get_be32(index);
3706
0
    index += sizeof(uint32_t);
3707
0
  }
3708
3709
0
  return ieot;
3710
0
}
3711
3712
static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot)
3713
0
{
3714
0
  uint32_t buffer;
3715
0
  int i;
3716
3717
  /* version */
3718
0
  put_be32(&buffer, IEOT_VERSION);
3719
0
  strbuf_add(sb, &buffer, sizeof(uint32_t));
3720
3721
  /* ieot */
3722
0
  for (i = 0; i < ieot->nr; i++) {
3723
3724
    /* offset */
3725
0
    put_be32(&buffer, ieot->entries[i].offset);
3726
0
    strbuf_add(sb, &buffer, sizeof(uint32_t));
3727
3728
    /* count */
3729
0
    put_be32(&buffer, ieot->entries[i].nr);
3730
0
    strbuf_add(sb, &buffer, sizeof(uint32_t));
3731
0
  }
3732
0
}
3733
3734
void prefetch_cache_entries(const struct index_state *istate,
3735
          must_prefetch_predicate must_prefetch)
3736
0
{
3737
0
  int i;
3738
0
  struct oid_array to_fetch = OID_ARRAY_INIT;
3739
3740
0
  for (i = 0; i < istate->cache_nr; i++) {
3741
0
    struct cache_entry *ce = istate->cache[i];
3742
3743
0
    if (S_ISGITLINK(ce->ce_mode) || !must_prefetch(ce))
3744
0
      continue;
3745
0
    if (!odb_read_object_info_extended(the_repository->objects,
3746
0
               &ce->oid, NULL,
3747
0
               OBJECT_INFO_FOR_PREFETCH))
3748
0
      continue;
3749
0
    oid_array_append(&to_fetch, &ce->oid);
3750
0
  }
3751
0
  promisor_remote_get_direct(the_repository,
3752
0
           to_fetch.oid, to_fetch.nr);
3753
0
  oid_array_clear(&to_fetch);
3754
0
}
3755
3756
static int read_one_entry_opt(struct index_state *istate,
3757
            const struct object_id *oid,
3758
            struct strbuf *base,
3759
            const char *pathname,
3760
            unsigned mode, int opt)
3761
0
{
3762
0
  int len;
3763
0
  struct cache_entry *ce;
3764
3765
0
  if (S_ISDIR(mode))
3766
0
    return READ_TREE_RECURSIVE;
3767
3768
0
  len = strlen(pathname);
3769
0
  ce = make_empty_cache_entry(istate, base->len + len);
3770
3771
0
  ce->ce_mode = create_ce_mode(mode);
3772
0
  ce->ce_flags = create_ce_flags(1);
3773
0
  ce->ce_namelen = base->len + len;
3774
0
  memcpy(ce->name, base->buf, base->len);
3775
0
  memcpy(ce->name + base->len, pathname, len+1);
3776
0
  oidcpy(&ce->oid, oid);
3777
0
  return add_index_entry(istate, ce, opt);
3778
0
}
3779
3780
static int read_one_entry(const struct object_id *oid, struct strbuf *base,
3781
        const char *pathname, unsigned mode,
3782
        void *context)
3783
0
{
3784
0
  struct index_state *istate = context;
3785
0
  return read_one_entry_opt(istate, oid, base, pathname,
3786
0
          mode,
3787
0
          ADD_CACHE_OK_TO_ADD|ADD_CACHE_SKIP_DFCHECK);
3788
0
}
3789
3790
/*
3791
 * This is used when the caller knows there is no existing entries at
3792
 * the stage that will conflict with the entry being added.
3793
 */
3794
static int read_one_entry_quick(const struct object_id *oid, struct strbuf *base,
3795
        const char *pathname, unsigned mode,
3796
        void *context)
3797
0
{
3798
0
  struct index_state *istate = context;
3799
0
  return read_one_entry_opt(istate, oid, base, pathname,
3800
0
          mode, ADD_CACHE_JUST_APPEND);
3801
0
}
3802
3803
/*
3804
 * Read the tree specified with --with-tree option
3805
 * (typically, HEAD) into stage #1 and then
3806
 * squash them down to stage #0.  This is used for
3807
 * --error-unmatch to list and check the path patterns
3808
 * that were given from the command line.  We are not
3809
 * going to write this index out.
3810
 */
3811
void overlay_tree_on_index(struct index_state *istate,
3812
         const char *tree_name, const char *prefix)
3813
0
{
3814
0
  struct tree *tree;
3815
0
  struct object_id oid;
3816
0
  struct pathspec pathspec;
3817
0
  struct cache_entry *last_stage0 = NULL;
3818
0
  int i;
3819
0
  read_tree_fn_t fn = NULL;
3820
0
  int err;
3821
3822
0
  if (repo_get_oid(the_repository, tree_name, &oid))
3823
0
    die("tree-ish %s not found.", tree_name);
3824
0
  tree = repo_parse_tree_indirect(the_repository, &oid);
3825
0
  if (!tree)
3826
0
    die("bad tree-ish %s", tree_name);
3827
3828
  /* Hoist the unmerged entries up to stage #3 to make room */
3829
  /* TODO: audit for interaction with sparse-index. */
3830
0
  ensure_full_index(istate);
3831
0
  for (i = 0; i < istate->cache_nr; i++) {
3832
0
    struct cache_entry *ce = istate->cache[i];
3833
0
    if (!ce_stage(ce))
3834
0
      continue;
3835
0
    ce->ce_flags |= CE_STAGEMASK;
3836
0
  }
3837
3838
0
  if (prefix) {
3839
0
    static const char *(matchbuf[1]);
3840
0
    matchbuf[0] = NULL;
3841
0
    parse_pathspec(&pathspec, PATHSPEC_ALL_MAGIC,
3842
0
             PATHSPEC_PREFER_CWD, prefix, matchbuf);
3843
0
  } else
3844
0
    memset(&pathspec, 0, sizeof(pathspec));
3845
3846
  /*
3847
   * See if we have cache entry at the stage.  If so,
3848
   * do it the original slow way, otherwise, append and then
3849
   * sort at the end.
3850
   */
3851
0
  for (i = 0; !fn && i < istate->cache_nr; i++) {
3852
0
    const struct cache_entry *ce = istate->cache[i];
3853
0
    if (ce_stage(ce) == 1)
3854
0
      fn = read_one_entry;
3855
0
  }
3856
3857
0
  if (!fn)
3858
0
    fn = read_one_entry_quick;
3859
0
  err = read_tree(the_repository, tree, &pathspec, fn, istate);
3860
0
  clear_pathspec(&pathspec);
3861
0
  if (err)
3862
0
    die("unable to read tree entries %s", tree_name);
3863
3864
  /*
3865
   * Sort the cache entry -- we need to nuke the cache tree, though.
3866
   */
3867
0
  if (fn == read_one_entry_quick) {
3868
0
    cache_tree_free(&istate->cache_tree);
3869
0
    QSORT(istate->cache, istate->cache_nr, cmp_cache_name_compare);
3870
0
  }
3871
3872
0
  for (i = 0; i < istate->cache_nr; i++) {
3873
0
    struct cache_entry *ce = istate->cache[i];
3874
0
    switch (ce_stage(ce)) {
3875
0
    case 0:
3876
0
      last_stage0 = ce;
3877
      /* fallthru */
3878
0
    default:
3879
0
      continue;
3880
0
    case 1:
3881
      /*
3882
       * If there is stage #0 entry for this, we do not
3883
       * need to show it.  We use CE_UPDATE bit to mark
3884
       * such an entry.
3885
       */
3886
0
      if (last_stage0 &&
3887
0
          !strcmp(last_stage0->name, ce->name))
3888
0
        ce->ce_flags |= CE_UPDATE;
3889
0
    }
3890
0
  }
3891
0
}
3892
3893
struct update_callback_data {
3894
  struct index_state *index;
3895
  struct repository *repo;
3896
  struct pathspec *pathspec;
3897
  int include_sparse;
3898
  int flags;
3899
  int add_errors;
3900
  int ignored_too;
3901
};
3902
3903
static int fix_unmerged_status(struct diff_filepair *p,
3904
             struct update_callback_data *data)
3905
0
{
3906
0
  if (p->status != DIFF_STATUS_UNMERGED)
3907
0
    return p->status;
3908
0
  if (!(data->flags & ADD_CACHE_IGNORE_REMOVAL) && !p->two->mode)
3909
    /*
3910
     * This is not an explicit add request, and the
3911
     * path is missing from the working tree (deleted)
3912
     */
3913
0
    return DIFF_STATUS_DELETED;
3914
0
  else
3915
    /*
3916
     * Either an explicit add request, or path exists
3917
     * in the working tree.  An attempt to explicitly
3918
     * add a path that does not exist in the working tree
3919
     * will be caught as an error by the caller immediately.
3920
     */
3921
0
    return DIFF_STATUS_MODIFIED;
3922
0
}
3923
3924
static int skip_submodule(const char *path,
3925
            struct repository *repo,
3926
            struct pathspec *pathspec,
3927
            int ignored_too)
3928
0
{
3929
0
    struct stat st;
3930
0
    const struct submodule *sub;
3931
0
    int pathspec_matches = 0;
3932
0
    int ps_i;
3933
0
    char *norm_pathspec = NULL;
3934
3935
    /* Only consider if path is a directory */
3936
0
    if (lstat(path, &st) || !S_ISDIR(st.st_mode))
3937
0
    return 0;
3938
3939
    /* Check if it's a submodule with ignore=all */
3940
0
    sub = submodule_from_path(repo, null_oid(the_hash_algo), path);
3941
0
    if (!sub || !sub->name || !sub->ignore || strcmp(sub->ignore, "all"))
3942
0
    return 0;
3943
3944
0
    trace_printf("ignore=all: %s\n", path);
3945
0
    trace_printf("pathspec %s\n", (pathspec && pathspec->nr)
3946
0
                  ? "has pathspec"
3947
0
                  : "no pathspec");
3948
3949
    /* Check if submodule path is explicitly mentioned in pathspec */
3950
0
    if (pathspec) {
3951
0
    for (ps_i = 0; ps_i < pathspec->nr; ps_i++) {
3952
0
      const char *m = pathspec->items[ps_i].match;
3953
0
      if (!m)
3954
0
        continue;
3955
0
      norm_pathspec = xstrdup(m);
3956
0
      strip_dir_trailing_slashes(norm_pathspec);
3957
0
      if (!strcmp(path, norm_pathspec)) {
3958
0
        pathspec_matches = 1;
3959
0
        FREE_AND_NULL(norm_pathspec);
3960
0
        break;
3961
0
      }
3962
0
      FREE_AND_NULL(norm_pathspec);
3963
0
    }
3964
0
    }
3965
3966
    /* If explicitly matched and forced, allow adding */
3967
0
    if (pathspec_matches) {
3968
0
    if (ignored_too && ignored_too > 0) {
3969
0
      trace_printf("Add submodule due to --force: %s\n", path);
3970
0
      return 0;
3971
0
    } else {
3972
0
      advise_if_enabled(ADVICE_ADD_IGNORED_FILE,
3973
0
        _("Skipping submodule due to ignore=all: %s\n"
3974
0
          "Use --force if you really want to add the submodule."), path);
3975
0
      return 1;
3976
0
    }
3977
0
    }
3978
3979
    /* No explicit pathspec match -> skip silently */
3980
0
    trace_printf("Pathspec to submodule does not match explicitly: %s\n", path);
3981
0
    return 1;
3982
0
}
3983
3984
static void update_callback(struct diff_queue_struct *q,
3985
              struct diff_options *opt UNUSED, void *cbdata)
3986
0
{
3987
0
  int i;
3988
0
  struct update_callback_data *data = cbdata;
3989
3990
0
  for (i = 0; i < q->nr; i++) {
3991
0
    struct diff_filepair *p = q->queue[i];
3992
0
    const char *path = p->one->path;
3993
3994
0
    if (!data->include_sparse &&
3995
0
      !path_in_sparse_checkout(path, data->index))
3996
0
      continue;
3997
3998
0
    switch (fix_unmerged_status(p, data)) {
3999
0
    default:
4000
0
      die(_("unexpected diff status %c"), p->status);
4001
0
    case DIFF_STATUS_MODIFIED:
4002
0
    case DIFF_STATUS_TYPE_CHANGED:
4003
0
      if (skip_submodule(path, data->repo,
4004
0
                data->pathspec,
4005
0
                data->ignored_too))
4006
0
        continue;
4007
4008
0
      if (add_file_to_index(data->index, path, data->flags)) {
4009
0
        if (!(data->flags & ADD_CACHE_IGNORE_ERRORS))
4010
0
          die(_("updating files failed"));
4011
0
        data->add_errors++;
4012
0
      }
4013
0
      break;
4014
0
    case DIFF_STATUS_DELETED:
4015
0
      if (data->flags & ADD_CACHE_IGNORE_REMOVAL)
4016
0
        break;
4017
0
      if (!(data->flags & ADD_CACHE_PRETEND))
4018
0
        remove_file_from_index(data->index, path);
4019
0
      if (data->flags & (ADD_CACHE_PRETEND|ADD_CACHE_VERBOSE))
4020
0
        printf(_("remove '%s'\n"), path);
4021
0
      break;
4022
0
    }
4023
0
  }
4024
0
}
4025
4026
int add_files_to_cache(struct repository *repo, const char *prefix,
4027
           const struct pathspec *pathspec, char *ps_matched,
4028
           int include_sparse, int flags, int ignored_too )
4029
0
{
4030
0
  struct odb_transaction *transaction;
4031
0
  struct update_callback_data data;
4032
0
  struct rev_info rev;
4033
4034
0
  memset(&data, 0, sizeof(data));
4035
0
  data.index = repo->index;
4036
0
  data.include_sparse = include_sparse;
4037
0
  data.flags = flags;
4038
0
  data.repo = repo;
4039
0
  data.ignored_too = ignored_too;
4040
0
  data.pathspec = (struct pathspec *)pathspec;
4041
4042
0
  repo_init_revisions(repo, &rev, prefix);
4043
0
  setup_revisions(0, NULL, &rev, NULL);
4044
0
  if (pathspec) {
4045
0
    copy_pathspec(&rev.prune_data, pathspec);
4046
0
    rev.ps_matched = ps_matched;
4047
0
  }
4048
0
  rev.diffopt.output_format = DIFF_FORMAT_CALLBACK;
4049
0
  rev.diffopt.format_callback = update_callback;
4050
0
  rev.diffopt.format_callback_data = &data;
4051
0
  rev.diffopt.flags.override_submodule_config = 1;
4052
0
  rev.max_count = 0; /* do not compare unmerged paths with stage #2 */
4053
4054
  /*
4055
   * Use an ODB transaction to optimize adding multiple objects.
4056
   * This function is invoked from commands other than 'add', which
4057
   * may not have their own transaction active.
4058
   */
4059
0
  transaction = odb_transaction_begin(repo->objects);
4060
0
  run_diff_files(&rev, DIFF_RACY_IS_MODIFIED);
4061
0
  odb_transaction_commit(transaction);
4062
4063
0
  release_revisions(&rev);
4064
0
  return !!data.add_errors;
4065
0
}