2be1afdde0b1bde19583ad8fa183fb7f2546cc13
[git/git.git] / sha1_file.c
1 /*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
9 #include "cache.h"
10 #include "string-list.h"
11 #include "lockfile.h"
12 #include "delta.h"
13 #include "pack.h"
14 #include "blob.h"
15 #include "commit.h"
16 #include "run-command.h"
17 #include "tag.h"
18 #include "tree.h"
19 #include "tree-walk.h"
20 #include "refs.h"
21 #include "pack-revindex.h"
22 #include "sha1-lookup.h"
23 #include "bulk-checkin.h"
24 #include "streaming.h"
25 #include "dir.h"
26
27 #ifndef O_NOATIME
28 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
29 #define O_NOATIME 01000000
30 #else
31 #define O_NOATIME 0
32 #endif
33 #endif
34
35 #define SZ_FMT PRIuMAX
36 static inline uintmax_t sz_fmt(size_t s) { return s; }
37
38 const unsigned char null_sha1[20];
39
40 /*
41 * This is meant to hold a *small* number of objects that you would
42 * want read_sha1_file() to be able to return, but yet you do not want
43 * to write them into the object store (e.g. a browse-only
44 * application).
45 */
46 static struct cached_object {
47 unsigned char sha1[20];
48 enum object_type type;
49 void *buf;
50 unsigned long size;
51 } *cached_objects;
52 static int cached_object_nr, cached_object_alloc;
53
54 static struct cached_object empty_tree = {
55 EMPTY_TREE_SHA1_BIN_LITERAL,
56 OBJ_TREE,
57 "",
58 0
59 };
60
61 /*
62 * A pointer to the last packed_git in which an object was found.
63 * When an object is sought, we look in this packfile first, because
64 * objects that are looked up at similar times are often in the same
65 * packfile as one another.
66 */
67 static struct packed_git *last_found_pack;
68
69 static struct cached_object *find_cached_object(const unsigned char *sha1)
70 {
71 int i;
72 struct cached_object *co = cached_objects;
73
74 for (i = 0; i < cached_object_nr; i++, co++) {
75 if (!hashcmp(co->sha1, sha1))
76 return co;
77 }
78 if (!hashcmp(sha1, empty_tree.sha1))
79 return &empty_tree;
80 return NULL;
81 }
82
83 int mkdir_in_gitdir(const char *path)
84 {
85 if (mkdir(path, 0777)) {
86 int saved_errno = errno;
87 struct stat st;
88 struct strbuf sb = STRBUF_INIT;
89
90 if (errno != EEXIST)
91 return -1;
92 /*
93 * Are we looking at a path in a symlinked worktree
94 * whose original repository does not yet have it?
95 * e.g. .git/rr-cache pointing at its original
96 * repository in which the user hasn't performed any
97 * conflict resolution yet?
98 */
99 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
100 strbuf_readlink(&sb, path, st.st_size) ||
101 !is_absolute_path(sb.buf) ||
102 mkdir(sb.buf, 0777)) {
103 strbuf_release(&sb);
104 errno = saved_errno;
105 return -1;
106 }
107 strbuf_release(&sb);
108 }
109 return adjust_shared_perm(path);
110 }
111
112 enum scld_error safe_create_leading_directories(char *path)
113 {
114 char *next_component = path + offset_1st_component(path);
115 enum scld_error ret = SCLD_OK;
116
117 while (ret == SCLD_OK && next_component) {
118 struct stat st;
119 char *slash = next_component, slash_character;
120
121 while (*slash && !is_dir_sep(*slash))
122 slash++;
123
124 if (!*slash)
125 break;
126
127 next_component = slash + 1;
128 while (is_dir_sep(*next_component))
129 next_component++;
130 if (!*next_component)
131 break;
132
133 slash_character = *slash;
134 *slash = '\0';
135 if (!stat(path, &st)) {
136 /* path exists */
137 if (!S_ISDIR(st.st_mode))
138 ret = SCLD_EXISTS;
139 } else if (mkdir(path, 0777)) {
140 if (errno == EEXIST &&
141 !stat(path, &st) && S_ISDIR(st.st_mode))
142 ; /* somebody created it since we checked */
143 else if (errno == ENOENT)
144 /*
145 * Either mkdir() failed because
146 * somebody just pruned the containing
147 * directory, or stat() failed because
148 * the file that was in our way was
149 * just removed. Either way, inform
150 * the caller that it might be worth
151 * trying again:
152 */
153 ret = SCLD_VANISHED;
154 else
155 ret = SCLD_FAILED;
156 } else if (adjust_shared_perm(path)) {
157 ret = SCLD_PERMS;
158 }
159 *slash = slash_character;
160 }
161 return ret;
162 }
163
164 enum scld_error safe_create_leading_directories_const(const char *path)
165 {
166 /* path points to cache entries, so xstrdup before messing with it */
167 char *buf = xstrdup(path);
168 enum scld_error result = safe_create_leading_directories(buf);
169 free(buf);
170 return result;
171 }
172
173 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
174 {
175 int i;
176 for (i = 0; i < 20; i++) {
177 static char hex[] = "0123456789abcdef";
178 unsigned int val = sha1[i];
179 char *pos = pathbuf + i*2 + (i > 0);
180 *pos++ = hex[val >> 4];
181 *pos = hex[val & 0xf];
182 }
183 }
184
185 const char *sha1_file_name(const unsigned char *sha1)
186 {
187 static char buf[PATH_MAX];
188 const char *objdir;
189 int len;
190
191 objdir = get_object_directory();
192 len = strlen(objdir);
193
194 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
195 if (len + 43 > PATH_MAX)
196 die("insanely long object directory %s", objdir);
197 memcpy(buf, objdir, len);
198 buf[len] = '/';
199 buf[len+3] = '/';
200 buf[len+42] = '\0';
201 fill_sha1_path(buf + len + 1, sha1);
202 return buf;
203 }
204
205 /*
206 * Return the name of the pack or index file with the specified sha1
207 * in its filename. *base and *name are scratch space that must be
208 * provided by the caller. which should be "pack" or "idx".
209 */
210 static char *sha1_get_pack_name(const unsigned char *sha1,
211 char **name, char **base, const char *which)
212 {
213 static const char hex[] = "0123456789abcdef";
214 char *buf;
215 int i;
216
217 if (!*base) {
218 const char *sha1_file_directory = get_object_directory();
219 int len = strlen(sha1_file_directory);
220 *base = xmalloc(len + 60);
221 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
222 sha1_file_directory, which);
223 *name = *base + len + 11;
224 }
225
226 buf = *name;
227
228 for (i = 0; i < 20; i++) {
229 unsigned int val = *sha1++;
230 *buf++ = hex[val >> 4];
231 *buf++ = hex[val & 0xf];
232 }
233
234 return *base;
235 }
236
237 char *sha1_pack_name(const unsigned char *sha1)
238 {
239 static char *name, *base;
240
241 return sha1_get_pack_name(sha1, &name, &base, "pack");
242 }
243
244 char *sha1_pack_index_name(const unsigned char *sha1)
245 {
246 static char *name, *base;
247
248 return sha1_get_pack_name(sha1, &name, &base, "idx");
249 }
250
251 struct alternate_object_database *alt_odb_list;
252 static struct alternate_object_database **alt_odb_tail;
253
254 /*
255 * Prepare alternate object database registry.
256 *
257 * The variable alt_odb_list points at the list of struct
258 * alternate_object_database. The elements on this list come from
259 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
260 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
261 * whose contents is similar to that environment variable but can be
262 * LF separated. Its base points at a statically allocated buffer that
263 * contains "/the/directory/corresponding/to/.git/objects/...", while
264 * its name points just after the slash at the end of ".git/objects/"
265 * in the example above, and has enough space to hold 40-byte hex
266 * SHA1, an extra slash for the first level indirection, and the
267 * terminating NUL.
268 */
269 static int link_alt_odb_entry(const char *entry, const char *relative_base,
270 int depth, const char *normalized_objdir)
271 {
272 struct alternate_object_database *ent;
273 struct alternate_object_database *alt;
274 int pfxlen, entlen;
275 struct strbuf pathbuf = STRBUF_INIT;
276
277 if (!is_absolute_path(entry) && relative_base) {
278 strbuf_addstr(&pathbuf, real_path(relative_base));
279 strbuf_addch(&pathbuf, '/');
280 }
281 strbuf_addstr(&pathbuf, entry);
282
283 normalize_path_copy(pathbuf.buf, pathbuf.buf);
284
285 pfxlen = strlen(pathbuf.buf);
286
287 /*
288 * The trailing slash after the directory name is given by
289 * this function at the end. Remove duplicates.
290 */
291 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
292 pfxlen -= 1;
293
294 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
295 ent = xmalloc(sizeof(*ent) + entlen);
296 memcpy(ent->base, pathbuf.buf, pfxlen);
297 strbuf_release(&pathbuf);
298
299 ent->name = ent->base + pfxlen + 1;
300 ent->base[pfxlen + 3] = '/';
301 ent->base[pfxlen] = ent->base[entlen-1] = 0;
302
303 /* Detect cases where alternate disappeared */
304 if (!is_directory(ent->base)) {
305 error("object directory %s does not exist; "
306 "check .git/objects/info/alternates.",
307 ent->base);
308 free(ent);
309 return -1;
310 }
311
312 /* Prevent the common mistake of listing the same
313 * thing twice, or object directory itself.
314 */
315 for (alt = alt_odb_list; alt; alt = alt->next) {
316 if (pfxlen == alt->name - alt->base - 1 &&
317 !memcmp(ent->base, alt->base, pfxlen)) {
318 free(ent);
319 return -1;
320 }
321 }
322 if (!strcmp_icase(ent->base, normalized_objdir)) {
323 free(ent);
324 return -1;
325 }
326
327 /* add the alternate entry */
328 *alt_odb_tail = ent;
329 alt_odb_tail = &(ent->next);
330 ent->next = NULL;
331
332 /* recursively add alternates */
333 read_info_alternates(ent->base, depth + 1);
334
335 ent->base[pfxlen] = '/';
336
337 return 0;
338 }
339
340 static void link_alt_odb_entries(const char *alt, int len, int sep,
341 const char *relative_base, int depth)
342 {
343 struct string_list entries = STRING_LIST_INIT_NODUP;
344 char *alt_copy;
345 int i;
346 struct strbuf objdirbuf = STRBUF_INIT;
347
348 if (depth > 5) {
349 error("%s: ignoring alternate object stores, nesting too deep.",
350 relative_base);
351 return;
352 }
353
354 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
355 normalize_path_copy(objdirbuf.buf, objdirbuf.buf);
356
357 alt_copy = xmemdupz(alt, len);
358 string_list_split_in_place(&entries, alt_copy, sep, -1);
359 for (i = 0; i < entries.nr; i++) {
360 const char *entry = entries.items[i].string;
361 if (entry[0] == '\0' || entry[0] == '#')
362 continue;
363 if (!is_absolute_path(entry) && depth) {
364 error("%s: ignoring relative alternate object store %s",
365 relative_base, entry);
366 } else {
367 link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
368 }
369 }
370 string_list_clear(&entries, 0);
371 free(alt_copy);
372 strbuf_release(&objdirbuf);
373 }
374
375 void read_info_alternates(const char * relative_base, int depth)
376 {
377 char *map;
378 size_t mapsz;
379 struct stat st;
380 char *path;
381 int fd;
382
383 path = xstrfmt("%s/info/alternates", relative_base);
384 fd = git_open_noatime(path);
385 free(path);
386 if (fd < 0)
387 return;
388 if (fstat(fd, &st) || (st.st_size == 0)) {
389 close(fd);
390 return;
391 }
392 mapsz = xsize_t(st.st_size);
393 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
394 close(fd);
395
396 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
397
398 munmap(map, mapsz);
399 }
400
401 void add_to_alternates_file(const char *reference)
402 {
403 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
404 char *alts = git_pathdup("objects/info/alternates");
405 FILE *in, *out;
406
407 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
408 out = fdopen_lock_file(lock, "w");
409 if (!out)
410 die_errno("unable to fdopen alternates lockfile");
411
412 in = fopen(alts, "r");
413 if (in) {
414 struct strbuf line = STRBUF_INIT;
415 int found = 0;
416
417 while (strbuf_getline(&line, in, '\n') != EOF) {
418 if (!strcmp(reference, line.buf)) {
419 found = 1;
420 break;
421 }
422 fprintf_or_die(out, "%s\n", line.buf);
423 }
424
425 strbuf_release(&line);
426 fclose(in);
427
428 if (found) {
429 rollback_lock_file(lock);
430 lock = NULL;
431 }
432 }
433 else if (errno != ENOENT)
434 die_errno("unable to read alternates file");
435
436 if (lock) {
437 fprintf_or_die(out, "%s\n", reference);
438 if (commit_lock_file(lock))
439 die_errno("unable to move new alternates file into place");
440 if (alt_odb_tail)
441 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
442 }
443 free(alts);
444 }
445
446 int foreach_alt_odb(alt_odb_fn fn, void *cb)
447 {
448 struct alternate_object_database *ent;
449 int r = 0;
450
451 prepare_alt_odb();
452 for (ent = alt_odb_list; ent; ent = ent->next) {
453 r = fn(ent, cb);
454 if (r)
455 break;
456 }
457 return r;
458 }
459
460 void prepare_alt_odb(void)
461 {
462 const char *alt;
463
464 if (alt_odb_tail)
465 return;
466
467 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
468 if (!alt) alt = "";
469
470 alt_odb_tail = &alt_odb_list;
471 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
472
473 read_info_alternates(get_object_directory(), 0);
474 }
475
476 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
477 static int freshen_file(const char *fn)
478 {
479 struct utimbuf t;
480 t.actime = t.modtime = time(NULL);
481 return !utime(fn, &t);
482 }
483
484 /*
485 * All of the check_and_freshen functions return 1 if the file exists and was
486 * freshened (if freshening was requested), 0 otherwise. If they return
487 * 0, you should not assume that it is safe to skip a write of the object (it
488 * either does not exist on disk, or has a stale mtime and may be subject to
489 * pruning).
490 */
491 static int check_and_freshen_file(const char *fn, int freshen)
492 {
493 if (access(fn, F_OK))
494 return 0;
495 if (freshen && !freshen_file(fn))
496 return 0;
497 return 1;
498 }
499
500 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
501 {
502 return check_and_freshen_file(sha1_file_name(sha1), freshen);
503 }
504
505 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
506 {
507 struct alternate_object_database *alt;
508 prepare_alt_odb();
509 for (alt = alt_odb_list; alt; alt = alt->next) {
510 fill_sha1_path(alt->name, sha1);
511 if (check_and_freshen_file(alt->base, freshen))
512 return 1;
513 }
514 return 0;
515 }
516
517 static int check_and_freshen(const unsigned char *sha1, int freshen)
518 {
519 return check_and_freshen_local(sha1, freshen) ||
520 check_and_freshen_nonlocal(sha1, freshen);
521 }
522
523 int has_loose_object_nonlocal(const unsigned char *sha1)
524 {
525 return check_and_freshen_nonlocal(sha1, 0);
526 }
527
528 static int has_loose_object(const unsigned char *sha1)
529 {
530 return check_and_freshen(sha1, 0);
531 }
532
533 static unsigned int pack_used_ctr;
534 static unsigned int pack_mmap_calls;
535 static unsigned int peak_pack_open_windows;
536 static unsigned int pack_open_windows;
537 static unsigned int pack_open_fds;
538 static unsigned int pack_max_fds;
539 static size_t peak_pack_mapped;
540 static size_t pack_mapped;
541 struct packed_git *packed_git;
542
543 void pack_report(void)
544 {
545 fprintf(stderr,
546 "pack_report: getpagesize() = %10" SZ_FMT "\n"
547 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
548 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
549 sz_fmt(getpagesize()),
550 sz_fmt(packed_git_window_size),
551 sz_fmt(packed_git_limit));
552 fprintf(stderr,
553 "pack_report: pack_used_ctr = %10u\n"
554 "pack_report: pack_mmap_calls = %10u\n"
555 "pack_report: pack_open_windows = %10u / %10u\n"
556 "pack_report: pack_mapped = "
557 "%10" SZ_FMT " / %10" SZ_FMT "\n",
558 pack_used_ctr,
559 pack_mmap_calls,
560 pack_open_windows, peak_pack_open_windows,
561 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
562 }
563
564 /*
565 * Open and mmap the index file at path, perform a couple of
566 * consistency checks, then record its information to p. Return 0 on
567 * success.
568 */
569 static int check_packed_git_idx(const char *path, struct packed_git *p)
570 {
571 void *idx_map;
572 struct pack_idx_header *hdr;
573 size_t idx_size;
574 uint32_t version, nr, i, *index;
575 int fd = git_open_noatime(path);
576 struct stat st;
577
578 if (fd < 0)
579 return -1;
580 if (fstat(fd, &st)) {
581 close(fd);
582 return -1;
583 }
584 idx_size = xsize_t(st.st_size);
585 if (idx_size < 4 * 256 + 20 + 20) {
586 close(fd);
587 return error("index file %s is too small", path);
588 }
589 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
590 close(fd);
591
592 hdr = idx_map;
593 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
594 version = ntohl(hdr->idx_version);
595 if (version < 2 || version > 2) {
596 munmap(idx_map, idx_size);
597 return error("index file %s is version %"PRIu32
598 " and is not supported by this binary"
599 " (try upgrading GIT to a newer version)",
600 path, version);
601 }
602 } else
603 version = 1;
604
605 nr = 0;
606 index = idx_map;
607 if (version > 1)
608 index += 2; /* skip index header */
609 for (i = 0; i < 256; i++) {
610 uint32_t n = ntohl(index[i]);
611 if (n < nr) {
612 munmap(idx_map, idx_size);
613 return error("non-monotonic index %s", path);
614 }
615 nr = n;
616 }
617
618 if (version == 1) {
619 /*
620 * Total size:
621 * - 256 index entries 4 bytes each
622 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
623 * - 20-byte SHA1 of the packfile
624 * - 20-byte SHA1 file checksum
625 */
626 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
627 munmap(idx_map, idx_size);
628 return error("wrong index v1 file size in %s", path);
629 }
630 } else if (version == 2) {
631 /*
632 * Minimum size:
633 * - 8 bytes of header
634 * - 256 index entries 4 bytes each
635 * - 20-byte sha1 entry * nr
636 * - 4-byte crc entry * nr
637 * - 4-byte offset entry * nr
638 * - 20-byte SHA1 of the packfile
639 * - 20-byte SHA1 file checksum
640 * And after the 4-byte offset table might be a
641 * variable sized table containing 8-byte entries
642 * for offsets larger than 2^31.
643 */
644 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
645 unsigned long max_size = min_size;
646 if (nr)
647 max_size += (nr - 1)*8;
648 if (idx_size < min_size || idx_size > max_size) {
649 munmap(idx_map, idx_size);
650 return error("wrong index v2 file size in %s", path);
651 }
652 if (idx_size != min_size &&
653 /*
654 * make sure we can deal with large pack offsets.
655 * 31-bit signed offset won't be enough, neither
656 * 32-bit unsigned one will be.
657 */
658 (sizeof(off_t) <= 4)) {
659 munmap(idx_map, idx_size);
660 return error("pack too large for current definition of off_t in %s", path);
661 }
662 }
663
664 p->index_version = version;
665 p->index_data = idx_map;
666 p->index_size = idx_size;
667 p->num_objects = nr;
668 return 0;
669 }
670
671 int open_pack_index(struct packed_git *p)
672 {
673 char *idx_name;
674 size_t len;
675 int ret;
676
677 if (p->index_data)
678 return 0;
679
680 if (!strip_suffix(p->pack_name, ".pack", &len))
681 die("BUG: pack_name does not end in .pack");
682 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
683 ret = check_packed_git_idx(idx_name, p);
684 free(idx_name);
685 return ret;
686 }
687
688 static void scan_windows(struct packed_git *p,
689 struct packed_git **lru_p,
690 struct pack_window **lru_w,
691 struct pack_window **lru_l)
692 {
693 struct pack_window *w, *w_l;
694
695 for (w_l = NULL, w = p->windows; w; w = w->next) {
696 if (!w->inuse_cnt) {
697 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
698 *lru_p = p;
699 *lru_w = w;
700 *lru_l = w_l;
701 }
702 }
703 w_l = w;
704 }
705 }
706
707 static int unuse_one_window(struct packed_git *current)
708 {
709 struct packed_git *p, *lru_p = NULL;
710 struct pack_window *lru_w = NULL, *lru_l = NULL;
711
712 if (current)
713 scan_windows(current, &lru_p, &lru_w, &lru_l);
714 for (p = packed_git; p; p = p->next)
715 scan_windows(p, &lru_p, &lru_w, &lru_l);
716 if (lru_p) {
717 munmap(lru_w->base, lru_w->len);
718 pack_mapped -= lru_w->len;
719 if (lru_l)
720 lru_l->next = lru_w->next;
721 else
722 lru_p->windows = lru_w->next;
723 free(lru_w);
724 pack_open_windows--;
725 return 1;
726 }
727 return 0;
728 }
729
730 void release_pack_memory(size_t need)
731 {
732 size_t cur = pack_mapped;
733 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
734 ; /* nothing */
735 }
736
737 static void mmap_limit_check(size_t length)
738 {
739 static size_t limit = 0;
740 if (!limit) {
741 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
742 if (!limit)
743 limit = SIZE_MAX;
744 }
745 if (length > limit)
746 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
747 (uintmax_t)length, (uintmax_t)limit);
748 }
749
750 void *xmmap_gently(void *start, size_t length,
751 int prot, int flags, int fd, off_t offset)
752 {
753 void *ret;
754
755 mmap_limit_check(length);
756 ret = mmap(start, length, prot, flags, fd, offset);
757 if (ret == MAP_FAILED) {
758 if (!length)
759 return NULL;
760 release_pack_memory(length);
761 ret = mmap(start, length, prot, flags, fd, offset);
762 }
763 return ret;
764 }
765
766 void *xmmap(void *start, size_t length,
767 int prot, int flags, int fd, off_t offset)
768 {
769 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
770 if (ret == MAP_FAILED)
771 die_errno("mmap failed");
772 return ret;
773 }
774
775 void close_pack_windows(struct packed_git *p)
776 {
777 while (p->windows) {
778 struct pack_window *w = p->windows;
779
780 if (w->inuse_cnt)
781 die("pack '%s' still has open windows to it",
782 p->pack_name);
783 munmap(w->base, w->len);
784 pack_mapped -= w->len;
785 pack_open_windows--;
786 p->windows = w->next;
787 free(w);
788 }
789 }
790
791 /*
792 * The LRU pack is the one with the oldest MRU window, preferring packs
793 * with no used windows, or the oldest mtime if it has no windows allocated.
794 */
795 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
796 {
797 struct pack_window *w, *this_mru_w;
798 int has_windows_inuse = 0;
799
800 /*
801 * Reject this pack if it has windows and the previously selected
802 * one does not. If this pack does not have windows, reject
803 * it if the pack file is newer than the previously selected one.
804 */
805 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
806 return;
807
808 for (w = this_mru_w = p->windows; w; w = w->next) {
809 /*
810 * Reject this pack if any of its windows are in use,
811 * but the previously selected pack did not have any
812 * inuse windows. Otherwise, record that this pack
813 * has windows in use.
814 */
815 if (w->inuse_cnt) {
816 if (*accept_windows_inuse)
817 has_windows_inuse = 1;
818 else
819 return;
820 }
821
822 if (w->last_used > this_mru_w->last_used)
823 this_mru_w = w;
824
825 /*
826 * Reject this pack if it has windows that have been
827 * used more recently than the previously selected pack.
828 * If the previously selected pack had windows inuse and
829 * we have not encountered a window in this pack that is
830 * inuse, skip this check since we prefer a pack with no
831 * inuse windows to one that has inuse windows.
832 */
833 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
834 this_mru_w->last_used > (*mru_w)->last_used)
835 return;
836 }
837
838 /*
839 * Select this pack.
840 */
841 *mru_w = this_mru_w;
842 *lru_p = p;
843 *accept_windows_inuse = has_windows_inuse;
844 }
845
846 static int close_one_pack(void)
847 {
848 struct packed_git *p, *lru_p = NULL;
849 struct pack_window *mru_w = NULL;
850 int accept_windows_inuse = 1;
851
852 for (p = packed_git; p; p = p->next) {
853 if (p->pack_fd == -1)
854 continue;
855 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
856 }
857
858 if (lru_p) {
859 close(lru_p->pack_fd);
860 pack_open_fds--;
861 lru_p->pack_fd = -1;
862 return 1;
863 }
864
865 return 0;
866 }
867
868 void unuse_pack(struct pack_window **w_cursor)
869 {
870 struct pack_window *w = *w_cursor;
871 if (w) {
872 w->inuse_cnt--;
873 *w_cursor = NULL;
874 }
875 }
876
877 void close_pack_index(struct packed_git *p)
878 {
879 if (p->index_data) {
880 munmap((void *)p->index_data, p->index_size);
881 p->index_data = NULL;
882 }
883 }
884
885 /*
886 * This is used by git-repack in case a newly created pack happens to
887 * contain the same set of objects as an existing one. In that case
888 * the resulting file might be different even if its name would be the
889 * same. It is best to close any reference to the old pack before it is
890 * replaced on disk. Of course no index pointers or windows for given pack
891 * must subsist at this point. If ever objects from this pack are requested
892 * again, the new version of the pack will be reinitialized through
893 * reprepare_packed_git().
894 */
895 void free_pack_by_name(const char *pack_name)
896 {
897 struct packed_git *p, **pp = &packed_git;
898
899 while (*pp) {
900 p = *pp;
901 if (strcmp(pack_name, p->pack_name) == 0) {
902 clear_delta_base_cache();
903 close_pack_windows(p);
904 if (p->pack_fd != -1) {
905 close(p->pack_fd);
906 pack_open_fds--;
907 }
908 close_pack_index(p);
909 free(p->bad_object_sha1);
910 *pp = p->next;
911 if (last_found_pack == p)
912 last_found_pack = NULL;
913 free(p);
914 return;
915 }
916 pp = &p->next;
917 }
918 }
919
920 static unsigned int get_max_fd_limit(void)
921 {
922 #ifdef RLIMIT_NOFILE
923 {
924 struct rlimit lim;
925
926 if (!getrlimit(RLIMIT_NOFILE, &lim))
927 return lim.rlim_cur;
928 }
929 #endif
930
931 #ifdef _SC_OPEN_MAX
932 {
933 long open_max = sysconf(_SC_OPEN_MAX);
934 if (0 < open_max)
935 return open_max;
936 /*
937 * Otherwise, we got -1 for one of the two
938 * reasons:
939 *
940 * (1) sysconf() did not understand _SC_OPEN_MAX
941 * and signaled an error with -1; or
942 * (2) sysconf() said there is no limit.
943 *
944 * We _could_ clear errno before calling sysconf() to
945 * tell these two cases apart and return a huge number
946 * in the latter case to let the caller cap it to a
947 * value that is not so selfish, but letting the
948 * fallback OPEN_MAX codepath take care of these cases
949 * is a lot simpler.
950 */
951 }
952 #endif
953
954 #ifdef OPEN_MAX
955 return OPEN_MAX;
956 #else
957 return 1; /* see the caller ;-) */
958 #endif
959 }
960
961 /*
962 * Do not call this directly as this leaks p->pack_fd on error return;
963 * call open_packed_git() instead.
964 */
965 static int open_packed_git_1(struct packed_git *p)
966 {
967 struct stat st;
968 struct pack_header hdr;
969 unsigned char sha1[20];
970 unsigned char *idx_sha1;
971 long fd_flag;
972
973 if (!p->index_data && open_pack_index(p))
974 return error("packfile %s index unavailable", p->pack_name);
975
976 if (!pack_max_fds) {
977 unsigned int max_fds = get_max_fd_limit();
978
979 /* Save 3 for stdin/stdout/stderr, 22 for work */
980 if (25 < max_fds)
981 pack_max_fds = max_fds - 25;
982 else
983 pack_max_fds = 1;
984 }
985
986 while (pack_max_fds <= pack_open_fds && close_one_pack())
987 ; /* nothing */
988
989 p->pack_fd = git_open_noatime(p->pack_name);
990 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
991 return -1;
992 pack_open_fds++;
993
994 /* If we created the struct before we had the pack we lack size. */
995 if (!p->pack_size) {
996 if (!S_ISREG(st.st_mode))
997 return error("packfile %s not a regular file", p->pack_name);
998 p->pack_size = st.st_size;
999 } else if (p->pack_size != st.st_size)
1000 return error("packfile %s size changed", p->pack_name);
1001
1002 /* We leave these file descriptors open with sliding mmap;
1003 * there is no point keeping them open across exec(), though.
1004 */
1005 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
1006 if (fd_flag < 0)
1007 return error("cannot determine file descriptor flags");
1008 fd_flag |= FD_CLOEXEC;
1009 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
1010 return error("cannot set FD_CLOEXEC");
1011
1012 /* Verify we recognize this pack file format. */
1013 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
1014 return error("file %s is far too short to be a packfile", p->pack_name);
1015 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
1016 return error("file %s is not a GIT packfile", p->pack_name);
1017 if (!pack_version_ok(hdr.hdr_version))
1018 return error("packfile %s is version %"PRIu32" and not"
1019 " supported (try upgrading GIT to a newer version)",
1020 p->pack_name, ntohl(hdr.hdr_version));
1021
1022 /* Verify the pack matches its index. */
1023 if (p->num_objects != ntohl(hdr.hdr_entries))
1024 return error("packfile %s claims to have %"PRIu32" objects"
1025 " while index indicates %"PRIu32" objects",
1026 p->pack_name, ntohl(hdr.hdr_entries),
1027 p->num_objects);
1028 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
1029 return error("end of packfile %s is unavailable", p->pack_name);
1030 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
1031 return error("packfile %s signature is unavailable", p->pack_name);
1032 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
1033 if (hashcmp(sha1, idx_sha1))
1034 return error("packfile %s does not match index", p->pack_name);
1035 return 0;
1036 }
1037
1038 static int open_packed_git(struct packed_git *p)
1039 {
1040 if (!open_packed_git_1(p))
1041 return 0;
1042 if (p->pack_fd != -1) {
1043 close(p->pack_fd);
1044 pack_open_fds--;
1045 p->pack_fd = -1;
1046 }
1047 return -1;
1048 }
1049
1050 static int in_window(struct pack_window *win, off_t offset)
1051 {
1052 /* We must promise at least 20 bytes (one hash) after the
1053 * offset is available from this window, otherwise the offset
1054 * is not actually in this window and a different window (which
1055 * has that one hash excess) must be used. This is to support
1056 * the object header and delta base parsing routines below.
1057 */
1058 off_t win_off = win->offset;
1059 return win_off <= offset
1060 && (offset + 20) <= (win_off + win->len);
1061 }
1062
1063 unsigned char *use_pack(struct packed_git *p,
1064 struct pack_window **w_cursor,
1065 off_t offset,
1066 unsigned long *left)
1067 {
1068 struct pack_window *win = *w_cursor;
1069
1070 /* Since packfiles end in a hash of their content and it's
1071 * pointless to ask for an offset into the middle of that
1072 * hash, and the in_window function above wouldn't match
1073 * don't allow an offset too close to the end of the file.
1074 */
1075 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1076 die("packfile %s cannot be accessed", p->pack_name);
1077 if (offset > (p->pack_size - 20))
1078 die("offset beyond end of packfile (truncated pack?)");
1079
1080 if (!win || !in_window(win, offset)) {
1081 if (win)
1082 win->inuse_cnt--;
1083 for (win = p->windows; win; win = win->next) {
1084 if (in_window(win, offset))
1085 break;
1086 }
1087 if (!win) {
1088 size_t window_align = packed_git_window_size / 2;
1089 off_t len;
1090
1091 if (p->pack_fd == -1 && open_packed_git(p))
1092 die("packfile %s cannot be accessed", p->pack_name);
1093
1094 win = xcalloc(1, sizeof(*win));
1095 win->offset = (offset / window_align) * window_align;
1096 len = p->pack_size - win->offset;
1097 if (len > packed_git_window_size)
1098 len = packed_git_window_size;
1099 win->len = (size_t)len;
1100 pack_mapped += win->len;
1101 while (packed_git_limit < pack_mapped
1102 && unuse_one_window(p))
1103 ; /* nothing */
1104 win->base = xmmap(NULL, win->len,
1105 PROT_READ, MAP_PRIVATE,
1106 p->pack_fd, win->offset);
1107 if (win->base == MAP_FAILED)
1108 die("packfile %s cannot be mapped: %s",
1109 p->pack_name,
1110 strerror(errno));
1111 if (!win->offset && win->len == p->pack_size
1112 && !p->do_not_close) {
1113 close(p->pack_fd);
1114 pack_open_fds--;
1115 p->pack_fd = -1;
1116 }
1117 pack_mmap_calls++;
1118 pack_open_windows++;
1119 if (pack_mapped > peak_pack_mapped)
1120 peak_pack_mapped = pack_mapped;
1121 if (pack_open_windows > peak_pack_open_windows)
1122 peak_pack_open_windows = pack_open_windows;
1123 win->next = p->windows;
1124 p->windows = win;
1125 }
1126 }
1127 if (win != *w_cursor) {
1128 win->last_used = pack_used_ctr++;
1129 win->inuse_cnt++;
1130 *w_cursor = win;
1131 }
1132 offset -= win->offset;
1133 if (left)
1134 *left = win->len - xsize_t(offset);
1135 return win->base + offset;
1136 }
1137
1138 static struct packed_git *alloc_packed_git(int extra)
1139 {
1140 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1141 memset(p, 0, sizeof(*p));
1142 p->pack_fd = -1;
1143 return p;
1144 }
1145
1146 static void try_to_free_pack_memory(size_t size)
1147 {
1148 release_pack_memory(size);
1149 }
1150
1151 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
1152 {
1153 static int have_set_try_to_free_routine;
1154 struct stat st;
1155 size_t alloc;
1156 struct packed_git *p;
1157
1158 if (!have_set_try_to_free_routine) {
1159 have_set_try_to_free_routine = 1;
1160 set_try_to_free_routine(try_to_free_pack_memory);
1161 }
1162
1163 /*
1164 * Make sure a corresponding .pack file exists and that
1165 * the index looks sane.
1166 */
1167 if (!strip_suffix_mem(path, &path_len, ".idx"))
1168 return NULL;
1169
1170 /*
1171 * ".pack" is long enough to hold any suffix we're adding (and
1172 * the use xsnprintf double-checks that)
1173 */
1174 alloc = path_len + strlen(".pack") + 1;
1175 p = alloc_packed_git(alloc);
1176 memcpy(p->pack_name, path, path_len);
1177
1178 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
1179 if (!access(p->pack_name, F_OK))
1180 p->pack_keep = 1;
1181
1182 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
1183 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1184 free(p);
1185 return NULL;
1186 }
1187
1188 /* ok, it looks sane as far as we can check without
1189 * actually mapping the pack file.
1190 */
1191 p->pack_size = st.st_size;
1192 p->pack_local = local;
1193 p->mtime = st.st_mtime;
1194 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1195 hashclr(p->sha1);
1196 return p;
1197 }
1198
1199 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1200 {
1201 const char *path = sha1_pack_name(sha1);
1202 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
1203
1204 strcpy(p->pack_name, path);
1205 hashcpy(p->sha1, sha1);
1206 if (check_packed_git_idx(idx_path, p)) {
1207 free(p);
1208 return NULL;
1209 }
1210
1211 return p;
1212 }
1213
1214 void install_packed_git(struct packed_git *pack)
1215 {
1216 if (pack->pack_fd != -1)
1217 pack_open_fds++;
1218
1219 pack->next = packed_git;
1220 packed_git = pack;
1221 }
1222
1223 void (*report_garbage)(const char *desc, const char *path);
1224
1225 static void report_helper(const struct string_list *list,
1226 int seen_bits, int first, int last)
1227 {
1228 const char *msg;
1229 switch (seen_bits) {
1230 case 0:
1231 msg = "no corresponding .idx or .pack";
1232 break;
1233 case 1:
1234 msg = "no corresponding .idx";
1235 break;
1236 case 2:
1237 msg = "no corresponding .pack";
1238 break;
1239 default:
1240 return;
1241 }
1242 for (; first < last; first++)
1243 report_garbage(msg, list->items[first].string);
1244 }
1245
1246 static void report_pack_garbage(struct string_list *list)
1247 {
1248 int i, baselen = -1, first = 0, seen_bits = 0;
1249
1250 if (!report_garbage)
1251 return;
1252
1253 string_list_sort(list);
1254
1255 for (i = 0; i < list->nr; i++) {
1256 const char *path = list->items[i].string;
1257 if (baselen != -1 &&
1258 strncmp(path, list->items[first].string, baselen)) {
1259 report_helper(list, seen_bits, first, i);
1260 baselen = -1;
1261 seen_bits = 0;
1262 }
1263 if (baselen == -1) {
1264 const char *dot = strrchr(path, '.');
1265 if (!dot) {
1266 report_garbage("garbage found", path);
1267 continue;
1268 }
1269 baselen = dot - path + 1;
1270 first = i;
1271 }
1272 if (!strcmp(path + baselen, "pack"))
1273 seen_bits |= 1;
1274 else if (!strcmp(path + baselen, "idx"))
1275 seen_bits |= 2;
1276 }
1277 report_helper(list, seen_bits, first, list->nr);
1278 }
1279
1280 static void prepare_packed_git_one(char *objdir, int local)
1281 {
1282 struct strbuf path = STRBUF_INIT;
1283 size_t dirnamelen;
1284 DIR *dir;
1285 struct dirent *de;
1286 struct string_list garbage = STRING_LIST_INIT_DUP;
1287
1288 strbuf_addstr(&path, objdir);
1289 strbuf_addstr(&path, "/pack");
1290 dir = opendir(path.buf);
1291 if (!dir) {
1292 if (errno != ENOENT)
1293 error("unable to open object pack directory: %s: %s",
1294 path.buf, strerror(errno));
1295 strbuf_release(&path);
1296 return;
1297 }
1298 strbuf_addch(&path, '/');
1299 dirnamelen = path.len;
1300 while ((de = readdir(dir)) != NULL) {
1301 struct packed_git *p;
1302 size_t base_len;
1303
1304 if (is_dot_or_dotdot(de->d_name))
1305 continue;
1306
1307 strbuf_setlen(&path, dirnamelen);
1308 strbuf_addstr(&path, de->d_name);
1309
1310 base_len = path.len;
1311 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1312 /* Don't reopen a pack we already have. */
1313 for (p = packed_git; p; p = p->next) {
1314 size_t len;
1315 if (strip_suffix(p->pack_name, ".pack", &len) &&
1316 len == base_len &&
1317 !memcmp(p->pack_name, path.buf, len))
1318 break;
1319 }
1320 if (p == NULL &&
1321 /*
1322 * See if it really is a valid .idx file with
1323 * corresponding .pack file that we can map.
1324 */
1325 (p = add_packed_git(path.buf, path.len, local)) != NULL)
1326 install_packed_git(p);
1327 }
1328
1329 if (!report_garbage)
1330 continue;
1331
1332 if (ends_with(de->d_name, ".idx") ||
1333 ends_with(de->d_name, ".pack") ||
1334 ends_with(de->d_name, ".bitmap") ||
1335 ends_with(de->d_name, ".keep"))
1336 string_list_append(&garbage, path.buf);
1337 else
1338 report_garbage("garbage found", path.buf);
1339 }
1340 closedir(dir);
1341 report_pack_garbage(&garbage);
1342 string_list_clear(&garbage, 0);
1343 strbuf_release(&path);
1344 }
1345
1346 static int sort_pack(const void *a_, const void *b_)
1347 {
1348 struct packed_git *a = *((struct packed_git **)a_);
1349 struct packed_git *b = *((struct packed_git **)b_);
1350 int st;
1351
1352 /*
1353 * Local packs tend to contain objects specific to our
1354 * variant of the project than remote ones. In addition,
1355 * remote ones could be on a network mounted filesystem.
1356 * Favor local ones for these reasons.
1357 */
1358 st = a->pack_local - b->pack_local;
1359 if (st)
1360 return -st;
1361
1362 /*
1363 * Younger packs tend to contain more recent objects,
1364 * and more recent objects tend to get accessed more
1365 * often.
1366 */
1367 if (a->mtime < b->mtime)
1368 return 1;
1369 else if (a->mtime == b->mtime)
1370 return 0;
1371 return -1;
1372 }
1373
1374 static void rearrange_packed_git(void)
1375 {
1376 struct packed_git **ary, *p;
1377 int i, n;
1378
1379 for (n = 0, p = packed_git; p; p = p->next)
1380 n++;
1381 if (n < 2)
1382 return;
1383
1384 /* prepare an array of packed_git for easier sorting */
1385 ary = xcalloc(n, sizeof(struct packed_git *));
1386 for (n = 0, p = packed_git; p; p = p->next)
1387 ary[n++] = p;
1388
1389 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1390
1391 /* link them back again */
1392 for (i = 0; i < n - 1; i++)
1393 ary[i]->next = ary[i + 1];
1394 ary[n - 1]->next = NULL;
1395 packed_git = ary[0];
1396
1397 free(ary);
1398 }
1399
1400 static int prepare_packed_git_run_once = 0;
1401 void prepare_packed_git(void)
1402 {
1403 struct alternate_object_database *alt;
1404
1405 if (prepare_packed_git_run_once)
1406 return;
1407 prepare_packed_git_one(get_object_directory(), 1);
1408 prepare_alt_odb();
1409 for (alt = alt_odb_list; alt; alt = alt->next) {
1410 alt->name[-1] = 0;
1411 prepare_packed_git_one(alt->base, 0);
1412 alt->name[-1] = '/';
1413 }
1414 rearrange_packed_git();
1415 prepare_packed_git_run_once = 1;
1416 }
1417
1418 void reprepare_packed_git(void)
1419 {
1420 prepare_packed_git_run_once = 0;
1421 prepare_packed_git();
1422 }
1423
1424 static void mark_bad_packed_object(struct packed_git *p,
1425 const unsigned char *sha1)
1426 {
1427 unsigned i;
1428 for (i = 0; i < p->num_bad_objects; i++)
1429 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1430 return;
1431 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1432 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1433 p->num_bad_objects++;
1434 }
1435
1436 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1437 {
1438 struct packed_git *p;
1439 unsigned i;
1440
1441 for (p = packed_git; p; p = p->next)
1442 for (i = 0; i < p->num_bad_objects; i++)
1443 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1444 return p;
1445 return NULL;
1446 }
1447
1448 /*
1449 * With an in-core object data in "map", rehash it to make sure the
1450 * object name actually matches "sha1" to detect object corruption.
1451 * With "map" == NULL, try reading the object named with "sha1" using
1452 * the streaming interface and rehash it to do the same.
1453 */
1454 int check_sha1_signature(const unsigned char *sha1, void *map,
1455 unsigned long size, const char *type)
1456 {
1457 unsigned char real_sha1[20];
1458 enum object_type obj_type;
1459 struct git_istream *st;
1460 git_SHA_CTX c;
1461 char hdr[32];
1462 int hdrlen;
1463
1464 if (map) {
1465 hash_sha1_file(map, size, type, real_sha1);
1466 return hashcmp(sha1, real_sha1) ? -1 : 0;
1467 }
1468
1469 st = open_istream(sha1, &obj_type, &size, NULL);
1470 if (!st)
1471 return -1;
1472
1473 /* Generate the header */
1474 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
1475
1476 /* Sha1.. */
1477 git_SHA1_Init(&c);
1478 git_SHA1_Update(&c, hdr, hdrlen);
1479 for (;;) {
1480 char buf[1024 * 16];
1481 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1482
1483 if (readlen < 0) {
1484 close_istream(st);
1485 return -1;
1486 }
1487 if (!readlen)
1488 break;
1489 git_SHA1_Update(&c, buf, readlen);
1490 }
1491 git_SHA1_Final(real_sha1, &c);
1492 close_istream(st);
1493 return hashcmp(sha1, real_sha1) ? -1 : 0;
1494 }
1495
1496 int git_open_noatime(const char *name)
1497 {
1498 static int sha1_file_open_flag = O_NOATIME;
1499
1500 for (;;) {
1501 int fd;
1502
1503 errno = 0;
1504 fd = open(name, O_RDONLY | sha1_file_open_flag);
1505 if (fd >= 0)
1506 return fd;
1507
1508 /* Might the failure be due to O_NOATIME? */
1509 if (errno != ENOENT && sha1_file_open_flag) {
1510 sha1_file_open_flag = 0;
1511 continue;
1512 }
1513
1514 return -1;
1515 }
1516 }
1517
1518 static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1519 {
1520 struct alternate_object_database *alt;
1521
1522 if (!lstat(sha1_file_name(sha1), st))
1523 return 0;
1524
1525 prepare_alt_odb();
1526 errno = ENOENT;
1527 for (alt = alt_odb_list; alt; alt = alt->next) {
1528 fill_sha1_path(alt->name, sha1);
1529 if (!lstat(alt->base, st))
1530 return 0;
1531 }
1532
1533 return -1;
1534 }
1535
1536 static int open_sha1_file(const unsigned char *sha1)
1537 {
1538 int fd;
1539 struct alternate_object_database *alt;
1540 int most_interesting_errno;
1541
1542 fd = git_open_noatime(sha1_file_name(sha1));
1543 if (fd >= 0)
1544 return fd;
1545 most_interesting_errno = errno;
1546
1547 prepare_alt_odb();
1548 for (alt = alt_odb_list; alt; alt = alt->next) {
1549 fill_sha1_path(alt->name, sha1);
1550 fd = git_open_noatime(alt->base);
1551 if (fd >= 0)
1552 return fd;
1553 if (most_interesting_errno == ENOENT)
1554 most_interesting_errno = errno;
1555 }
1556 errno = most_interesting_errno;
1557 return -1;
1558 }
1559
1560 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1561 {
1562 void *map;
1563 int fd;
1564
1565 fd = open_sha1_file(sha1);
1566 map = NULL;
1567 if (fd >= 0) {
1568 struct stat st;
1569
1570 if (!fstat(fd, &st)) {
1571 *size = xsize_t(st.st_size);
1572 if (!*size) {
1573 /* mmap() is forbidden on empty files */
1574 error("object file %s is empty", sha1_file_name(sha1));
1575 return NULL;
1576 }
1577 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1578 }
1579 close(fd);
1580 }
1581 return map;
1582 }
1583
1584 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1585 unsigned long len, enum object_type *type, unsigned long *sizep)
1586 {
1587 unsigned shift;
1588 unsigned long size, c;
1589 unsigned long used = 0;
1590
1591 c = buf[used++];
1592 *type = (c >> 4) & 7;
1593 size = c & 15;
1594 shift = 4;
1595 while (c & 0x80) {
1596 if (len <= used || bitsizeof(long) <= shift) {
1597 error("bad object header");
1598 size = used = 0;
1599 break;
1600 }
1601 c = buf[used++];
1602 size += (c & 0x7f) << shift;
1603 shift += 7;
1604 }
1605 *sizep = size;
1606 return used;
1607 }
1608
1609 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1610 {
1611 /* Get the data stream */
1612 memset(stream, 0, sizeof(*stream));
1613 stream->next_in = map;
1614 stream->avail_in = mapsize;
1615 stream->next_out = buffer;
1616 stream->avail_out = bufsiz;
1617
1618 git_inflate_init(stream);
1619 return git_inflate(stream, 0);
1620 }
1621
1622 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1623 unsigned long mapsize, void *buffer,
1624 unsigned long bufsiz, struct strbuf *header)
1625 {
1626 int status;
1627
1628 status = unpack_sha1_header(stream, map, mapsize, buffer, bufsiz);
1629
1630 /*
1631 * Check if entire header is unpacked in the first iteration.
1632 */
1633 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1634 return 0;
1635
1636 /*
1637 * buffer[0..bufsiz] was not large enough. Copy the partial
1638 * result out to header, and then append the result of further
1639 * reading the stream.
1640 */
1641 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1642 stream->next_out = buffer;
1643 stream->avail_out = bufsiz;
1644
1645 do {
1646 status = git_inflate(stream, 0);
1647 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1648 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1649 return 0;
1650 stream->next_out = buffer;
1651 stream->avail_out = bufsiz;
1652 } while (status != Z_STREAM_END);
1653 return -1;
1654 }
1655
1656 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1657 {
1658 int bytes = strlen(buffer) + 1;
1659 unsigned char *buf = xmallocz(size);
1660 unsigned long n;
1661 int status = Z_OK;
1662
1663 n = stream->total_out - bytes;
1664 if (n > size)
1665 n = size;
1666 memcpy(buf, (char *) buffer + bytes, n);
1667 bytes = n;
1668 if (bytes <= size) {
1669 /*
1670 * The above condition must be (bytes <= size), not
1671 * (bytes < size). In other words, even though we
1672 * expect no more output and set avail_out to zero,
1673 * the input zlib stream may have bytes that express
1674 * "this concludes the stream", and we *do* want to
1675 * eat that input.
1676 *
1677 * Otherwise we would not be able to test that we
1678 * consumed all the input to reach the expected size;
1679 * we also want to check that zlib tells us that all
1680 * went well with status == Z_STREAM_END at the end.
1681 */
1682 stream->next_out = buf + bytes;
1683 stream->avail_out = size - bytes;
1684 while (status == Z_OK)
1685 status = git_inflate(stream, Z_FINISH);
1686 }
1687 if (status == Z_STREAM_END && !stream->avail_in) {
1688 git_inflate_end(stream);
1689 return buf;
1690 }
1691
1692 if (status < 0)
1693 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1694 else if (stream->avail_in)
1695 error("garbage at end of loose object '%s'",
1696 sha1_to_hex(sha1));
1697 free(buf);
1698 return NULL;
1699 }
1700
1701 /*
1702 * We used to just use "sscanf()", but that's actually way
1703 * too permissive for what we want to check. So do an anal
1704 * object header parse by hand.
1705 */
1706 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1707 unsigned int flags)
1708 {
1709 const char *type_buf = hdr;
1710 unsigned long size;
1711 int type, type_len = 0;
1712
1713 /*
1714 * The type can be of any size but is followed by
1715 * a space.
1716 */
1717 for (;;) {
1718 char c = *hdr++;
1719 if (c == ' ')
1720 break;
1721 type_len++;
1722 }
1723
1724 type = type_from_string_gently(type_buf, type_len, 1);
1725 if (oi->typename)
1726 strbuf_add(oi->typename, type_buf, type_len);
1727 /*
1728 * Set type to 0 if its an unknown object and
1729 * we're obtaining the type using '--allow-unkown-type'
1730 * option.
1731 */
1732 if ((flags & LOOKUP_UNKNOWN_OBJECT) && (type < 0))
1733 type = 0;
1734 else if (type < 0)
1735 die("invalid object type");
1736 if (oi->typep)
1737 *oi->typep = type;
1738
1739 /*
1740 * The length must follow immediately, and be in canonical
1741 * decimal format (ie "010" is not valid).
1742 */
1743 size = *hdr++ - '0';
1744 if (size > 9)
1745 return -1;
1746 if (size) {
1747 for (;;) {
1748 unsigned long c = *hdr - '0';
1749 if (c > 9)
1750 break;
1751 hdr++;
1752 size = size * 10 + c;
1753 }
1754 }
1755
1756 if (oi->sizep)
1757 *oi->sizep = size;
1758
1759 /*
1760 * The length must be followed by a zero byte
1761 */
1762 return *hdr ? -1 : type;
1763 }
1764
1765 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1766 {
1767 struct object_info oi;
1768
1769 oi.sizep = sizep;
1770 oi.typename = NULL;
1771 oi.typep = NULL;
1772 return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
1773 }
1774
1775 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1776 {
1777 int ret;
1778 git_zstream stream;
1779 char hdr[8192];
1780
1781 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1782 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1783 return NULL;
1784
1785 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1786 }
1787
1788 unsigned long get_size_from_delta(struct packed_git *p,
1789 struct pack_window **w_curs,
1790 off_t curpos)
1791 {
1792 const unsigned char *data;
1793 unsigned char delta_head[20], *in;
1794 git_zstream stream;
1795 int st;
1796
1797 memset(&stream, 0, sizeof(stream));
1798 stream.next_out = delta_head;
1799 stream.avail_out = sizeof(delta_head);
1800
1801 git_inflate_init(&stream);
1802 do {
1803 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1804 stream.next_in = in;
1805 st = git_inflate(&stream, Z_FINISH);
1806 curpos += stream.next_in - in;
1807 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1808 stream.total_out < sizeof(delta_head));
1809 git_inflate_end(&stream);
1810 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1811 error("delta data unpack-initial failed");
1812 return 0;
1813 }
1814
1815 /* Examine the initial part of the delta to figure out
1816 * the result size.
1817 */
1818 data = delta_head;
1819
1820 /* ignore base size */
1821 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1822
1823 /* Read the result size */
1824 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1825 }
1826
1827 static off_t get_delta_base(struct packed_git *p,
1828 struct pack_window **w_curs,
1829 off_t *curpos,
1830 enum object_type type,
1831 off_t delta_obj_offset)
1832 {
1833 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1834 off_t base_offset;
1835
1836 /* use_pack() assured us we have [base_info, base_info + 20)
1837 * as a range that we can look at without walking off the
1838 * end of the mapped window. Its actually the hash size
1839 * that is assured. An OFS_DELTA longer than the hash size
1840 * is stupid, as then a REF_DELTA would be smaller to store.
1841 */
1842 if (type == OBJ_OFS_DELTA) {
1843 unsigned used = 0;
1844 unsigned char c = base_info[used++];
1845 base_offset = c & 127;
1846 while (c & 128) {
1847 base_offset += 1;
1848 if (!base_offset || MSB(base_offset, 7))
1849 return 0; /* overflow */
1850 c = base_info[used++];
1851 base_offset = (base_offset << 7) + (c & 127);
1852 }
1853 base_offset = delta_obj_offset - base_offset;
1854 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1855 return 0; /* out of bound */
1856 *curpos += used;
1857 } else if (type == OBJ_REF_DELTA) {
1858 /* The base entry _must_ be in the same pack */
1859 base_offset = find_pack_entry_one(base_info, p);
1860 *curpos += 20;
1861 } else
1862 die("I am totally screwed");
1863 return base_offset;
1864 }
1865
1866 /*
1867 * Like get_delta_base above, but we return the sha1 instead of the pack
1868 * offset. This means it is cheaper for REF deltas (we do not have to do
1869 * the final object lookup), but more expensive for OFS deltas (we
1870 * have to load the revidx to convert the offset back into a sha1).
1871 */
1872 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1873 struct pack_window **w_curs,
1874 off_t curpos,
1875 enum object_type type,
1876 off_t delta_obj_offset)
1877 {
1878 if (type == OBJ_REF_DELTA) {
1879 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1880 return base;
1881 } else if (type == OBJ_OFS_DELTA) {
1882 struct revindex_entry *revidx;
1883 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1884 type, delta_obj_offset);
1885
1886 if (!base_offset)
1887 return NULL;
1888
1889 revidx = find_pack_revindex(p, base_offset);
1890 if (!revidx)
1891 return NULL;
1892
1893 return nth_packed_object_sha1(p, revidx->nr);
1894 } else
1895 return NULL;
1896 }
1897
1898 int unpack_object_header(struct packed_git *p,
1899 struct pack_window **w_curs,
1900 off_t *curpos,
1901 unsigned long *sizep)
1902 {
1903 unsigned char *base;
1904 unsigned long left;
1905 unsigned long used;
1906 enum object_type type;
1907
1908 /* use_pack() assures us we have [base, base + 20) available
1909 * as a range that we can look at. (Its actually the hash
1910 * size that is assured.) With our object header encoding
1911 * the maximum deflated object size is 2^137, which is just
1912 * insane, so we know won't exceed what we have been given.
1913 */
1914 base = use_pack(p, w_curs, *curpos, &left);
1915 used = unpack_object_header_buffer(base, left, &type, sizep);
1916 if (!used) {
1917 type = OBJ_BAD;
1918 } else
1919 *curpos += used;
1920
1921 return type;
1922 }
1923
1924 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1925 {
1926 int type;
1927 struct revindex_entry *revidx;
1928 const unsigned char *sha1;
1929 revidx = find_pack_revindex(p, obj_offset);
1930 if (!revidx)
1931 return OBJ_BAD;
1932 sha1 = nth_packed_object_sha1(p, revidx->nr);
1933 mark_bad_packed_object(p, sha1);
1934 type = sha1_object_info(sha1, NULL);
1935 if (type <= OBJ_NONE)
1936 return OBJ_BAD;
1937 return type;
1938 }
1939
1940 #define POI_STACK_PREALLOC 64
1941
1942 static enum object_type packed_to_object_type(struct packed_git *p,
1943 off_t obj_offset,
1944 enum object_type type,
1945 struct pack_window **w_curs,
1946 off_t curpos)
1947 {
1948 off_t small_poi_stack[POI_STACK_PREALLOC];
1949 off_t *poi_stack = small_poi_stack;
1950 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1951
1952 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1953 off_t base_offset;
1954 unsigned long size;
1955 /* Push the object we're going to leave behind */
1956 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1957 poi_stack_alloc = alloc_nr(poi_stack_nr);
1958 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1959 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1960 } else {
1961 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1962 }
1963 poi_stack[poi_stack_nr++] = obj_offset;
1964 /* If parsing the base offset fails, just unwind */
1965 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1966 if (!base_offset)
1967 goto unwind;
1968 curpos = obj_offset = base_offset;
1969 type = unpack_object_header(p, w_curs, &curpos, &size);
1970 if (type <= OBJ_NONE) {
1971 /* If getting the base itself fails, we first
1972 * retry the base, otherwise unwind */
1973 type = retry_bad_packed_offset(p, base_offset);
1974 if (type > OBJ_NONE)
1975 goto out;
1976 goto unwind;
1977 }
1978 }
1979
1980 switch (type) {
1981 case OBJ_BAD:
1982 case OBJ_COMMIT:
1983 case OBJ_TREE:
1984 case OBJ_BLOB:
1985 case OBJ_TAG:
1986 break;
1987 default:
1988 error("unknown object type %i at offset %"PRIuMAX" in %s",
1989 type, (uintmax_t)obj_offset, p->pack_name);
1990 type = OBJ_BAD;
1991 }
1992
1993 out:
1994 if (poi_stack != small_poi_stack)
1995 free(poi_stack);
1996 return type;
1997
1998 unwind:
1999 while (poi_stack_nr) {
2000 obj_offset = poi_stack[--poi_stack_nr];
2001 type = retry_bad_packed_offset(p, obj_offset);
2002 if (type > OBJ_NONE)
2003 goto out;
2004 }
2005 type = OBJ_BAD;
2006 goto out;
2007 }
2008
2009 static int packed_object_info(struct packed_git *p, off_t obj_offset,
2010 struct object_info *oi)
2011 {
2012 struct pack_window *w_curs = NULL;
2013 unsigned long size;
2014 off_t curpos = obj_offset;
2015 enum object_type type;
2016
2017 /*
2018 * We always get the representation type, but only convert it to
2019 * a "real" type later if the caller is interested.
2020 */
2021 type = unpack_object_header(p, &w_curs, &curpos, &size);
2022
2023 if (oi->sizep) {
2024 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2025 off_t tmp_pos = curpos;
2026 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
2027 type, obj_offset);
2028 if (!base_offset) {
2029 type = OBJ_BAD;
2030 goto out;
2031 }
2032 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2033 if (*oi->sizep == 0) {
2034 type = OBJ_BAD;
2035 goto out;
2036 }
2037 } else {
2038 *oi->sizep = size;
2039 }
2040 }
2041
2042 if (oi->disk_sizep) {
2043 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2044 *oi->disk_sizep = revidx[1].offset - obj_offset;
2045 }
2046
2047 if (oi->typep) {
2048 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
2049 if (*oi->typep < 0) {
2050 type = OBJ_BAD;
2051 goto out;
2052 }
2053 }
2054
2055 if (oi->delta_base_sha1) {
2056 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2057 const unsigned char *base;
2058
2059 base = get_delta_base_sha1(p, &w_curs, curpos,
2060 type, obj_offset);
2061 if (!base) {
2062 type = OBJ_BAD;
2063 goto out;
2064 }
2065
2066 hashcpy(oi->delta_base_sha1, base);
2067 } else
2068 hashclr(oi->delta_base_sha1);
2069 }
2070
2071 out:
2072 unuse_pack(&w_curs);
2073 return type;
2074 }
2075
2076 static void *unpack_compressed_entry(struct packed_git *p,
2077 struct pack_window **w_curs,
2078 off_t curpos,
2079 unsigned long size)
2080 {
2081 int st;
2082 git_zstream stream;
2083 unsigned char *buffer, *in;
2084
2085 buffer = xmallocz_gently(size);
2086 if (!buffer)
2087 return NULL;
2088 memset(&stream, 0, sizeof(stream));
2089 stream.next_out = buffer;
2090 stream.avail_out = size + 1;
2091
2092 git_inflate_init(&stream);
2093 do {
2094 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2095 stream.next_in = in;
2096 st = git_inflate(&stream, Z_FINISH);
2097 if (!stream.avail_out)
2098 break; /* the payload is larger than it should be */
2099 curpos += stream.next_in - in;
2100 } while (st == Z_OK || st == Z_BUF_ERROR);
2101 git_inflate_end(&stream);
2102 if ((st != Z_STREAM_END) || stream.total_out != size) {
2103 free(buffer);
2104 return NULL;
2105 }
2106
2107 return buffer;
2108 }
2109
2110 #define MAX_DELTA_CACHE (256)
2111
2112 static size_t delta_base_cached;
2113
2114 static struct delta_base_cache_lru_list {
2115 struct delta_base_cache_lru_list *prev;
2116 struct delta_base_cache_lru_list *next;
2117 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
2118
2119 static struct delta_base_cache_entry {
2120 struct delta_base_cache_lru_list lru;
2121 void *data;
2122 struct packed_git *p;
2123 off_t base_offset;
2124 unsigned long size;
2125 enum object_type type;
2126 } delta_base_cache[MAX_DELTA_CACHE];
2127
2128 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
2129 {
2130 unsigned long hash;
2131
2132 hash = (unsigned long)p + (unsigned long)base_offset;
2133 hash += (hash >> 8) + (hash >> 16);
2134 return hash % MAX_DELTA_CACHE;
2135 }
2136
2137 static struct delta_base_cache_entry *
2138 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2139 {
2140 unsigned long hash = pack_entry_hash(p, base_offset);
2141 return delta_base_cache + hash;
2142 }
2143
2144 static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
2145 struct packed_git *p, off_t base_offset)
2146 {
2147 return (ent->data && ent->p == p && ent->base_offset == base_offset);
2148 }
2149
2150 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2151 {
2152 struct delta_base_cache_entry *ent;
2153 ent = get_delta_base_cache_entry(p, base_offset);
2154 return eq_delta_base_cache_entry(ent, p, base_offset);
2155 }
2156
2157 static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2158 {
2159 ent->data = NULL;
2160 ent->lru.next->prev = ent->lru.prev;
2161 ent->lru.prev->next = ent->lru.next;
2162 delta_base_cached -= ent->size;
2163 }
2164
2165 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2166 unsigned long *base_size, enum object_type *type, int keep_cache)
2167 {
2168 struct delta_base_cache_entry *ent;
2169 void *ret;
2170
2171 ent = get_delta_base_cache_entry(p, base_offset);
2172
2173 if (!eq_delta_base_cache_entry(ent, p, base_offset))
2174 return unpack_entry(p, base_offset, type, base_size);
2175
2176 ret = ent->data;
2177
2178 if (!keep_cache)
2179 clear_delta_base_cache_entry(ent);
2180 else
2181 ret = xmemdupz(ent->data, ent->size);
2182 *type = ent->type;
2183 *base_size = ent->size;
2184 return ret;
2185 }
2186
2187 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2188 {
2189 if (ent->data) {
2190 free(ent->data);
2191 ent->data = NULL;
2192 ent->lru.next->prev = ent->lru.prev;
2193 ent->lru.prev->next = ent->lru.next;
2194 delta_base_cached -= ent->size;
2195 }
2196 }
2197
2198 void clear_delta_base_cache(void)
2199 {
2200 unsigned long p;
2201 for (p = 0; p < MAX_DELTA_CACHE; p++)
2202 release_delta_base_cache(&delta_base_cache[p]);
2203 }
2204
2205 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2206 void *base, unsigned long base_size, enum object_type type)
2207 {
2208 unsigned long hash = pack_entry_hash(p, base_offset);
2209 struct delta_base_cache_entry *ent = delta_base_cache + hash;
2210 struct delta_base_cache_lru_list *lru;
2211
2212 release_delta_base_cache(ent);
2213 delta_base_cached += base_size;
2214
2215 for (lru = delta_base_cache_lru.next;
2216 delta_base_cached > delta_base_cache_limit
2217 && lru != &delta_base_cache_lru;
2218 lru = lru->next) {
2219 struct delta_base_cache_entry *f = (void *)lru;
2220 if (f->type == OBJ_BLOB)
2221 release_delta_base_cache(f);
2222 }
2223 for (lru = delta_base_cache_lru.next;
2224 delta_base_cached > delta_base_cache_limit
2225 && lru != &delta_base_cache_lru;
2226 lru = lru->next) {
2227 struct delta_base_cache_entry *f = (void *)lru;
2228 release_delta_base_cache(f);
2229 }
2230
2231 ent->p = p;
2232 ent->base_offset = base_offset;
2233 ent->type = type;
2234 ent->data = base;
2235 ent->size = base_size;
2236 ent->lru.next = &delta_base_cache_lru;
2237 ent->lru.prev = delta_base_cache_lru.prev;
2238 delta_base_cache_lru.prev->next = &ent->lru;
2239 delta_base_cache_lru.prev = &ent->lru;
2240 }
2241
2242 static void *read_object(const unsigned char *sha1, enum object_type *type,
2243 unsigned long *size);
2244
2245 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2246 {
2247 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2248 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2249 p->pack_name, (uintmax_t)obj_offset);
2250 }
2251
2252 int do_check_packed_object_crc;
2253
2254 #define UNPACK_ENTRY_STACK_PREALLOC 64
2255 struct unpack_entry_stack_ent {
2256 off_t obj_offset;
2257 off_t curpos;
2258 unsigned long size;
2259 };
2260
2261 void *unpack_entry(struct packed_git *p, off_t obj_offset,
2262 enum object_type *final_type, unsigned long *final_size)
2263 {
2264 struct pack_window *w_curs = NULL;
2265 off_t curpos = obj_offset;
2266 void *data = NULL;
2267 unsigned long size;
2268 enum object_type type;
2269 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2270 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2271 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2272 int base_from_cache = 0;
2273
2274 write_pack_access_log(p, obj_offset);
2275
2276 /* PHASE 1: drill down to the innermost base object */
2277 for (;;) {
2278 off_t base_offset;
2279 int i;
2280 struct delta_base_cache_entry *ent;
2281
2282 ent = get_delta_base_cache_entry(p, curpos);
2283 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2284 type = ent->type;
2285 data = ent->data;
2286 size = ent->size;
2287 clear_delta_base_cache_entry(ent);
2288 base_from_cache = 1;
2289 break;
2290 }
2291
2292 if (do_check_packed_object_crc && p->index_version > 1) {
2293 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2294 unsigned long len = revidx[1].offset - obj_offset;
2295 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2296 const unsigned char *sha1 =
2297 nth_packed_object_sha1(p, revidx->nr);
2298 error("bad packed object CRC for %s",
2299 sha1_to_hex(sha1));
2300 mark_bad_packed_object(p, sha1);
2301 unuse_pack(&w_curs);
2302 return NULL;
2303 }
2304 }
2305
2306 type = unpack_object_header(p, &w_curs, &curpos, &size);
2307 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2308 break;
2309
2310 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2311 if (!base_offset) {
2312 error("failed to validate delta base reference "
2313 "at offset %"PRIuMAX" from %s",
2314 (uintmax_t)curpos, p->pack_name);
2315 /* bail to phase 2, in hopes of recovery */
2316 data = NULL;
2317 break;
2318 }
2319
2320 /* push object, proceed to base */
2321 if (delta_stack_nr >= delta_stack_alloc
2322 && delta_stack == small_delta_stack) {
2323 delta_stack_alloc = alloc_nr(delta_stack_nr);
2324 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2325 memcpy(delta_stack, small_delta_stack,
2326 sizeof(*delta_stack)*delta_stack_nr);
2327 } else {
2328 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2329 }
2330 i = delta_stack_nr++;
2331 delta_stack[i].obj_offset = obj_offset;
2332 delta_stack[i].curpos = curpos;
2333 delta_stack[i].size = size;
2334
2335 curpos = obj_offset = base_offset;
2336 }
2337
2338 /* PHASE 2: handle the base */
2339 switch (type) {
2340 case OBJ_OFS_DELTA:
2341 case OBJ_REF_DELTA:
2342 if (data)
2343 die("BUG in unpack_entry: left loop at a valid delta");
2344 break;
2345 case OBJ_COMMIT:
2346 case OBJ_TREE:
2347 case OBJ_BLOB:
2348 case OBJ_TAG:
2349 if (!base_from_cache)
2350 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2351 break;
2352 default:
2353 data = NULL;
2354 error("unknown object type %i at offset %"PRIuMAX" in %s",
2355 type, (uintmax_t)obj_offset, p->pack_name);
2356 }
2357
2358 /* PHASE 3: apply deltas in order */
2359
2360 /* invariants:
2361 * 'data' holds the base data, or NULL if there was corruption
2362 */
2363 while (delta_stack_nr) {
2364 void *delta_data;
2365 void *base = data;
2366 unsigned long delta_size, base_size = size;
2367 int i;
2368
2369 data = NULL;
2370
2371 if (base)
2372 add_delta_base_cache(p, obj_offset, base, base_size, type);
2373
2374 if (!base) {
2375 /*
2376 * We're probably in deep shit, but let's try to fetch
2377 * the required base anyway from another pack or loose.
2378 * This is costly but should happen only in the presence
2379 * of a corrupted pack, and is better than failing outright.
2380 */
2381 struct revindex_entry *revidx;
2382 const unsigned char *base_sha1;
2383 revidx = find_pack_revindex(p, obj_offset);
2384 if (revidx) {
2385 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2386 error("failed to read delta base object %s"
2387 " at offset %"PRIuMAX" from %s",
2388 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2389 p->pack_name);
2390 mark_bad_packed_object(p, base_sha1);
2391 base = read_object(base_sha1, &type, &base_size);
2392 }
2393 }
2394
2395 i = --delta_stack_nr;
2396 obj_offset = delta_stack[i].obj_offset;
2397 curpos = delta_stack[i].curpos;
2398 delta_size = delta_stack[i].size;
2399
2400 if (!base)
2401 continue;
2402
2403 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2404
2405 if (!delta_data) {
2406 error("failed to unpack compressed delta "
2407 "at offset %"PRIuMAX" from %s",
2408 (uintmax_t)curpos, p->pack_name);
2409 data = NULL;
2410 continue;
2411 }
2412
2413 data = patch_delta(base, base_size,
2414 delta_data, delta_size,
2415 &size);
2416
2417 /*
2418 * We could not apply the delta; warn the user, but keep going.
2419 * Our failure will be noticed either in the next iteration of
2420 * the loop, or if this is the final delta, in the caller when
2421 * we return NULL. Those code paths will take care of making
2422 * a more explicit warning and retrying with another copy of
2423 * the object.
2424 */
2425 if (!data)
2426 error("failed to apply delta");
2427
2428 free(delta_data);
2429 }
2430
2431 *final_type = type;
2432 *final_size = size;
2433
2434 unuse_pack(&w_curs);
2435
2436 if (delta_stack != small_delta_stack)
2437 free(delta_stack);
2438
2439 return data;
2440 }
2441
2442 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2443 uint32_t n)
2444 {
2445 const unsigned char *index = p->index_data;
2446 if (!index) {
2447 if (open_pack_index(p))
2448 return NULL;
2449 index = p->index_data;
2450 }
2451 if (n >= p->num_objects)
2452 return NULL;
2453 index += 4 * 256;
2454 if (p->index_version == 1) {
2455 return index + 24 * n + 4;
2456 } else {
2457 index += 8;
2458 return index + 20 * n;
2459 }
2460 }
2461
2462 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2463 {
2464 const unsigned char *index = p->index_data;
2465 index += 4 * 256;
2466 if (p->index_version == 1) {
2467 return ntohl(*((uint32_t *)(index + 24 * n)));
2468 } else {
2469 uint32_t off;
2470 index += 8 + p->num_objects * (20 + 4);
2471 off = ntohl(*((uint32_t *)(index + 4 * n)));
2472 if (!(off & 0x80000000))
2473 return off;
2474 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2475 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2476 ntohl(*((uint32_t *)(index + 4)));
2477 }
2478 }
2479
2480 off_t find_pack_entry_one(const unsigned char *sha1,
2481 struct packed_git *p)
2482 {
2483 const uint32_t *level1_ofs = p->index_data;
2484 const unsigned char *index = p->index_data;
2485 unsigned hi, lo, stride;
2486 static int use_lookup = -1;
2487 static int debug_lookup = -1;
2488
2489 if (debug_lookup < 0)
2490 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2491
2492 if (!index) {
2493 if (open_pack_index(p))
2494 return 0;
2495 level1_ofs = p->index_data;
2496 index = p->index_data;
2497 }
2498 if (p->index_version > 1) {
2499 level1_ofs += 2;
2500 index += 8;
2501 }
2502 index += 4 * 256;
2503 hi = ntohl(level1_ofs[*sha1]);
2504 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2505 if (p->index_version > 1) {
2506 stride = 20;
2507 } else {
2508 stride = 24;
2509 index += 4;
2510 }
2511
2512 if (debug_lookup)
2513 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2514 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2515
2516 if (use_lookup < 0)
2517 use_lookup = !!getenv("GIT_USE_LOOKUP");
2518 if (use_lookup) {
2519 int pos = sha1_entry_pos(index, stride, 0,
2520 lo, hi, p->num_objects, sha1);
2521 if (pos < 0)
2522 return 0;
2523 return nth_packed_object_offset(p, pos);
2524 }
2525
2526 do {
2527 unsigned mi = (lo + hi) / 2;
2528 int cmp = hashcmp(index + mi * stride, sha1);
2529
2530 if (debug_lookup)
2531 printf("lo %u hi %u rg %u mi %u\n",
2532 lo, hi, hi - lo, mi);
2533 if (!cmp)
2534 return nth_packed_object_offset(p, mi);
2535 if (cmp > 0)
2536 hi = mi;
2537 else
2538 lo = mi+1;
2539 } while (lo < hi);
2540 return 0;
2541 }
2542
2543 int is_pack_valid(struct packed_git *p)
2544 {
2545 /* An already open pack is known to be valid. */
2546 if (p->pack_fd != -1)
2547 return 1;
2548
2549 /* If the pack has one window completely covering the
2550 * file size, the pack is known to be valid even if
2551 * the descriptor is not currently open.
2552 */
2553 if (p->windows) {
2554 struct pack_window *w = p->windows;
2555
2556 if (!w->offset && w->len == p->pack_size)
2557 return 1;
2558 }
2559
2560 /* Force the pack to open to prove its valid. */
2561 return !open_packed_git(p);
2562 }
2563
2564 static int fill_pack_entry(const unsigned char *sha1,
2565 struct pack_entry *e,
2566 struct packed_git *p)
2567 {
2568 off_t offset;
2569
2570 if (p->num_bad_objects) {
2571 unsigned i;
2572 for (i = 0; i < p->num_bad_objects; i++)
2573 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2574 return 0;
2575 }
2576
2577 offset = find_pack_entry_one(sha1, p);
2578 if (!offset)
2579 return 0;
2580
2581 /*
2582 * We are about to tell the caller where they can locate the
2583 * requested object. We better make sure the packfile is
2584 * still here and can be accessed before supplying that
2585 * answer, as it may have been deleted since the index was
2586 * loaded!
2587 */
2588 if (!is_pack_valid(p))
2589 return 0;
2590 e->offset = offset;
2591 e->p = p;
2592 hashcpy(e->sha1, sha1);
2593 return 1;
2594 }
2595
2596 /*
2597 * Iff a pack file contains the object named by sha1, return true and
2598 * store its location to e.
2599 */
2600 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2601 {
2602 struct packed_git *p;
2603
2604 prepare_packed_git();
2605 if (!packed_git)
2606 return 0;
2607
2608 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2609 return 1;
2610
2611 for (p = packed_git; p; p = p->next) {
2612 if (p == last_found_pack)
2613 continue; /* we already checked this one */
2614
2615 if (fill_pack_entry(sha1, e, p)) {
2616 last_found_pack = p;
2617 return 1;
2618 }
2619 }
2620 return 0;
2621 }
2622
2623 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2624 struct packed_git *packs)
2625 {
2626 struct packed_git *p;
2627
2628 for (p = packs; p; p = p->next) {
2629 if (find_pack_entry_one(sha1, p))
2630 return p;
2631 }
2632 return NULL;
2633
2634 }
2635
2636 static int sha1_loose_object_info(const unsigned char *sha1,
2637 struct object_info *oi,
2638 int flags)
2639 {
2640 int status = 0;
2641 unsigned long mapsize;
2642 void *map;
2643 git_zstream stream;
2644 char hdr[32];
2645 struct strbuf hdrbuf = STRBUF_INIT;
2646
2647 if (oi->delta_base_sha1)
2648 hashclr(oi->delta_base_sha1);
2649
2650 /*
2651 * If we don't care about type or size, then we don't
2652 * need to look inside the object at all. Note that we
2653 * do not optimize out the stat call, even if the
2654 * caller doesn't care about the disk-size, since our
2655 * return value implicitly indicates whether the
2656 * object even exists.
2657 */
2658 if (!oi->typep && !oi->typename && !oi->sizep) {
2659 struct stat st;
2660 if (stat_sha1_file(sha1, &st) < 0)
2661 return -1;
2662 if (oi->disk_sizep)
2663 *oi->disk_sizep = st.st_size;
2664 return 0;
2665 }
2666
2667 map = map_sha1_file(sha1, &mapsize);
2668 if (!map)
2669 return -1;
2670 if (oi->disk_sizep)
2671 *oi->disk_sizep = mapsize;
2672 if ((flags & LOOKUP_UNKNOWN_OBJECT)) {
2673 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2674 status = error("unable to unpack %s header with --allow-unknown-type",
2675 sha1_to_hex(sha1));
2676 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2677 status = error("unable to unpack %s header",
2678 sha1_to_hex(sha1));
2679 if (status < 0)
2680 ; /* Do nothing */
2681 else if (hdrbuf.len) {
2682 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2683 status = error("unable to parse %s header with --allow-unknown-type",
2684 sha1_to_hex(sha1));
2685 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
2686 status = error("unable to parse %s header", sha1_to_hex(sha1));
2687 git_inflate_end(&stream);
2688 munmap(map, mapsize);
2689 if (status && oi->typep)
2690 *oi->typep = status;
2691 strbuf_release(&hdrbuf);
2692 return 0;
2693 }
2694
2695 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2696 {
2697 struct cached_object *co;
2698 struct pack_entry e;
2699 int rtype;
2700 enum object_type real_type;
2701 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2702
2703 co = find_cached_object(real);
2704 if (co) {
2705 if (oi->typep)
2706 *(oi->typep) = co->type;
2707 if (oi->sizep)
2708 *(oi->sizep) = co->size;
2709 if (oi->disk_sizep)
2710 *(oi->disk_sizep) = 0;
2711 if (oi->delta_base_sha1)
2712 hashclr(oi->delta_base_sha1);
2713 if (oi->typename)
2714 strbuf_addstr(oi->typename, typename(co->type));
2715 oi->whence = OI_CACHED;
2716 return 0;
2717 }
2718
2719 if (!find_pack_entry(real, &e)) {
2720 /* Most likely it's a loose object. */
2721 if (!sha1_loose_object_info(real, oi, flags)) {
2722 oi->whence = OI_LOOSE;
2723 return 0;
2724 }
2725
2726 /* Not a loose object; someone else may have just packed it. */
2727 reprepare_packed_git();
2728 if (!find_pack_entry(real, &e))
2729 return -1;
2730 }
2731
2732 /*
2733 * packed_object_info() does not follow the delta chain to
2734 * find out the real type, unless it is given oi->typep.
2735 */
2736 if (oi->typename && !oi->typep)
2737 oi->typep = &real_type;
2738
2739 rtype = packed_object_info(e.p, e.offset, oi);
2740 if (rtype < 0) {
2741 mark_bad_packed_object(e.p, real);
2742 if (oi->typep == &real_type)
2743 oi->typep = NULL;
2744 return sha1_object_info_extended(real, oi, 0);
2745 } else if (in_delta_base_cache(e.p, e.offset)) {
2746 oi->whence = OI_DBCACHED;
2747 } else {
2748 oi->whence = OI_PACKED;
2749 oi->u.packed.offset = e.offset;
2750 oi->u.packed.pack = e.p;
2751 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2752 rtype == OBJ_OFS_DELTA);
2753 }
2754 if (oi->typename)
2755 strbuf_addstr(oi->typename, typename(*oi->typep));
2756 if (oi->typep == &real_type)
2757 oi->typep = NULL;
2758
2759 return 0;
2760 }
2761
2762 /* returns enum object_type or negative */
2763 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2764 {
2765 enum object_type type;
2766 struct object_info oi = {NULL};
2767
2768 oi.typep = &type;
2769 oi.sizep = sizep;
2770 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2771 return -1;
2772 return type;
2773 }
2774
2775 static void *read_packed_sha1(const unsigned char *sha1,
2776 enum object_type *type, unsigned long *size)
2777 {
2778 struct pack_entry e;
2779 void *data;
2780
2781 if (!find_pack_entry(sha1, &e))
2782 return NULL;
2783 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2784 if (!data) {
2785 /*
2786 * We're probably in deep shit, but let's try to fetch
2787 * the required object anyway from another pack or loose.
2788 * This should happen only in the presence of a corrupted
2789 * pack, and is better than failing outright.
2790 */
2791 error("failed to read object %s at offset %"PRIuMAX" from %s",
2792 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2793 mark_bad_packed_object(e.p, sha1);
2794 data = read_object(sha1, type, size);
2795 }
2796 return data;
2797 }
2798
2799 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2800 unsigned char *sha1)
2801 {
2802 struct cached_object *co;
2803
2804 hash_sha1_file(buf, len, typename(type), sha1);
2805 if (has_sha1_file(sha1) || find_cached_object(sha1))
2806 return 0;
2807 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2808 co = &cached_objects[cached_object_nr++];
2809 co->size = len;
2810 co->type = type;
2811 co->buf = xmalloc(len);
2812 memcpy(co->buf, buf, len);
2813 hashcpy(co->sha1, sha1);
2814 return 0;
2815 }
2816
2817 static void *read_object(const unsigned char *sha1, enum object_type *type,
2818 unsigned long *size)
2819 {
2820 unsigned long mapsize;
2821 void *map, *buf;
2822 struct cached_object *co;
2823
2824 co = find_cached_object(sha1);
2825 if (co) {
2826 *type = co->type;
2827 *size = co->size;
2828 return xmemdupz(co->buf, co->size);
2829 }
2830
2831 buf = read_packed_sha1(sha1, type, size);
2832 if (buf)
2833 return buf;
2834 map = map_sha1_file(sha1, &mapsize);
2835 if (map) {
2836 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2837 munmap(map, mapsize);
2838 return buf;
2839 }
2840 reprepare_packed_git();
2841 return read_packed_sha1(sha1, type, size);
2842 }
2843
2844 /*
2845 * This function dies on corrupt objects; the callers who want to
2846 * deal with them should arrange to call read_object() and give error
2847 * messages themselves.
2848 */
2849 void *read_sha1_file_extended(const unsigned char *sha1,
2850 enum object_type *type,
2851 unsigned long *size,
2852 unsigned flag)
2853 {
2854 void *data;
2855 const struct packed_git *p;
2856 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
2857
2858 errno = 0;
2859 data = read_object(repl, type, size);
2860 if (data)
2861 return data;
2862
2863 if (errno && errno != ENOENT)
2864 die_errno("failed to read object %s", sha1_to_hex(sha1));
2865
2866 /* die if we replaced an object with one that does not exist */
2867 if (repl != sha1)
2868 die("replacement %s not found for %s",
2869 sha1_to_hex(repl), sha1_to_hex(sha1));
2870
2871 if (has_loose_object(repl)) {
2872 const char *path = sha1_file_name(sha1);
2873
2874 die("loose object %s (stored in %s) is corrupt",
2875 sha1_to_hex(repl), path);
2876 }
2877
2878 if ((p = has_packed_and_bad(repl)) != NULL)
2879 die("packed object %s (stored in %s) is corrupt",
2880 sha1_to_hex(repl), p->pack_name);
2881
2882 return NULL;
2883 }
2884
2885 void *read_object_with_reference(const unsigned char *sha1,
2886 const char *required_type_name,
2887 unsigned long *size,
2888 unsigned char *actual_sha1_return)
2889 {
2890 enum object_type type, required_type;
2891 void *buffer;
2892 unsigned long isize;
2893 unsigned char actual_sha1[20];
2894
2895 required_type = type_from_string(required_type_name);
2896 hashcpy(actual_sha1, sha1);
2897 while (1) {
2898 int ref_length = -1;
2899 const char *ref_type = NULL;
2900
2901 buffer = read_sha1_file(actual_sha1, &type, &isize);
2902 if (!buffer)
2903 return NULL;
2904 if (type == required_type) {
2905 *size = isize;
2906 if (actual_sha1_return)
2907 hashcpy(actual_sha1_return, actual_sha1);
2908 return buffer;
2909 }
2910 /* Handle references */
2911 else if (type == OBJ_COMMIT)
2912 ref_type = "tree ";
2913 else if (type == OBJ_TAG)
2914 ref_type = "object ";
2915 else {
2916 free(buffer);
2917 return NULL;
2918 }
2919 ref_length = strlen(ref_type);
2920
2921 if (ref_length + 40 > isize ||
2922 memcmp(buffer, ref_type, ref_length) ||
2923 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2924 free(buffer);
2925 return NULL;
2926 }
2927 free(buffer);
2928 /* Now we have the ID of the referred-to object in
2929 * actual_sha1. Check again. */
2930 }
2931 }
2932
2933 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2934 const char *type, unsigned char *sha1,
2935 char *hdr, int *hdrlen)
2936 {
2937 git_SHA_CTX c;
2938
2939 /* Generate the header */
2940 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
2941
2942 /* Sha1.. */
2943 git_SHA1_Init(&c);
2944 git_SHA1_Update(&c, hdr, *hdrlen);
2945 git_SHA1_Update(&c, buf, len);
2946 git_SHA1_Final(sha1, &c);
2947 }
2948
2949 /*
2950 * Move the just written object into its final resting place.
2951 */
2952 int finalize_object_file(const char *tmpfile, const char *filename)
2953 {
2954 int ret = 0;
2955
2956 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2957 goto try_rename;
2958 else if (link(tmpfile, filename))
2959 ret = errno;
2960
2961 /*
2962 * Coda hack - coda doesn't like cross-directory links,
2963 * so we fall back to a rename, which will mean that it
2964 * won't be able to check collisions, but that's not a
2965 * big deal.
2966 *
2967 * The same holds for FAT formatted media.
2968 *
2969 * When this succeeds, we just return. We have nothing
2970 * left to unlink.
2971 */
2972 if (ret && ret != EEXIST) {
2973 try_rename:
2974 if (!rename(tmpfile, filename))
2975 goto out;
2976 ret = errno;
2977 }
2978 unlink_or_warn(tmpfile);
2979 if (ret) {
2980 if (ret != EEXIST) {
2981 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2982 }
2983 /* FIXME!!! Collision check here ? */
2984 }
2985
2986 out:
2987 if (adjust_shared_perm(filename))
2988 return error("unable to set permission to '%s'", filename);
2989 return 0;
2990 }
2991
2992 static int write_buffer(int fd, const void *buf, size_t len)
2993 {
2994 if (write_in_full(fd, buf, len) < 0)
2995 return error("file write error (%s)", strerror(errno));
2996 return 0;
2997 }
2998
2999 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
3000 unsigned char *sha1)
3001 {
3002 char hdr[32];
3003 int hdrlen = sizeof(hdr);
3004 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3005 return 0;
3006 }
3007
3008 /* Finalize a file on disk, and close it. */
3009 static void close_sha1_file(int fd)
3010 {
3011 if (fsync_object_files)
3012 fsync_or_die(fd, "sha1 file");
3013 if (close(fd) != 0)
3014 die_errno("error when closing sha1 file");
3015 }
3016
3017 /* Size of directory component, including the ending '/' */
3018 static inline int directory_size(const char *filename)
3019 {
3020 const char *s = strrchr(filename, '/');
3021 if (!s)
3022 return 0;
3023 return s - filename + 1;
3024 }
3025
3026 /*
3027 * This creates a temporary file in the same directory as the final
3028 * 'filename'
3029 *
3030 * We want to avoid cross-directory filename renames, because those
3031 * can have problems on various filesystems (FAT, NFS, Coda).
3032 */
3033 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
3034 {
3035 int fd, dirlen = directory_size(filename);
3036
3037 if (dirlen + 20 > bufsiz) {
3038 errno = ENAMETOOLONG;
3039 return -1;
3040 }
3041 memcpy(buffer, filename, dirlen);
3042 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
3043 fd = git_mkstemp_mode(buffer, 0444);
3044 if (fd < 0 && dirlen && errno == ENOENT) {
3045 /* Make sure the directory exists */
3046 memcpy(buffer, filename, dirlen);
3047 buffer[dirlen-1] = 0;
3048 if (mkdir(buffer, 0777) && errno != EEXIST)
3049 return -1;
3050 if (adjust_shared_perm(buffer))
3051 return -1;
3052
3053 /* Try again */
3054 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
3055 fd = git_mkstemp_mode(buffer, 0444);
3056 }
3057 return fd;
3058 }
3059
3060 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
3061 const void *buf, unsigned long len, time_t mtime)
3062 {
3063 int fd, ret;
3064 unsigned char compressed[4096];
3065 git_zstream stream;
3066 git_SHA_CTX c;
3067 unsigned char parano_sha1[20];
3068 static char tmp_file[PATH_MAX];
3069 const char *filename = sha1_file_name(sha1);
3070
3071 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
3072 if (fd < 0) {
3073 if (errno == EACCES)
3074 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3075 else
3076 return error("unable to create temporary file: %s", strerror(errno));
3077 }
3078
3079 /* Set it up */
3080 git_deflate_init(&stream, zlib_compression_level);
3081 stream.next_out = compressed;
3082 stream.avail_out = sizeof(compressed);
3083 git_SHA1_Init(&c);
3084
3085 /* First header.. */
3086 stream.next_in = (unsigned char *)hdr;
3087 stream.avail_in = hdrlen;
3088 while (git_deflate(&stream, 0) == Z_OK)
3089 ; /* nothing */
3090 git_SHA1_Update(&c, hdr, hdrlen);
3091
3092 /* Then the data itself.. */
3093 stream.next_in = (void *)buf;
3094 stream.avail_in = len;
3095 do {
3096 unsigned char *in0 = stream.next_in;
3097 ret = git_deflate(&stream, Z_FINISH);
3098 git_SHA1_Update(&c, in0, stream.next_in - in0);
3099 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3100 die("unable to write sha1 file");
3101 stream.next_out = compressed;
3102 stream.avail_out = sizeof(compressed);
3103 } while (ret == Z_OK);
3104
3105 if (ret != Z_STREAM_END)
3106 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
3107 ret = git_deflate_end_gently(&stream);
3108 if (ret != Z_OK)
3109 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
3110 git_SHA1_Final(parano_sha1, &c);
3111 if (hashcmp(sha1, parano_sha1) != 0)
3112 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
3113
3114 close_sha1_file(fd);
3115
3116 if (mtime) {
3117 struct utimbuf utb;
3118 utb.actime = mtime;
3119 utb.modtime = mtime;
3120 if (utime(tmp_file, &utb) < 0)
3121 warning("failed utime() on %s: %s",
3122 tmp_file, strerror(errno));
3123 }
3124
3125 return finalize_object_file(tmp_file, filename);
3126 }
3127
3128 static int freshen_loose_object(const unsigned char *sha1)
3129 {
3130 return check_and_freshen(sha1, 1);
3131 }
3132
3133 static int freshen_packed_object(const unsigned char *sha1)
3134 {
3135 struct pack_entry e;
3136 if (!find_pack_entry(sha1, &e))
3137 return 0;
3138 if (e.p->freshened)
3139 return 1;
3140 if (!freshen_file(e.p->pack_name))
3141 return 0;
3142 e.p->freshened = 1;
3143 return 1;
3144 }
3145
3146 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3147 {
3148 char hdr[32];
3149 int hdrlen = sizeof(hdr);
3150
3151 /* Normally if we have it in the pack then we do not bother writing
3152 * it out into .git/objects/??/?{38} file.
3153 */
3154 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3155 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3156 return 0;
3157 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3158 }
3159
3160 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3161 unsigned char *sha1, unsigned flags)
3162 {
3163 char *header;
3164 int hdrlen, status = 0;
3165
3166 /* type string, SP, %lu of the length plus NUL must fit this */
3167 hdrlen = strlen(type) + 32;
3168 header = xmalloc(hdrlen);
3169 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3170
3171 if (!(flags & HASH_WRITE_OBJECT))
3172 goto cleanup;
3173 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3174 goto cleanup;
3175 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3176
3177 cleanup:
3178 free(header);
3179 return status;
3180 }
3181
3182 int force_object_loose(const unsigned char *sha1, time_t mtime)
3183 {
3184 void *buf;
3185 unsigned long len;
3186 enum object_type type;
3187 char hdr[32];
3188 int hdrlen;
3189 int ret;
3190
3191 if (has_loose_object(sha1))
3192 return 0;
3193 buf = read_packed_sha1(sha1, &type, &len);
3194 if (!buf)
3195 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3196 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
3197 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3198 free(buf);
3199
3200 return ret;
3201 }
3202
3203 int has_pack_index(const unsigned char *sha1)
3204 {
3205 struct stat st;
3206 if (stat(sha1_pack_index_name(sha1), &st))
3207 return 0;
3208 return 1;
3209 }
3210
3211 int has_sha1_pack(const unsigned char *sha1)
3212 {
3213 struct pack_entry e;
3214 return find_pack_entry(sha1, &e);
3215 }
3216
3217 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3218 {
3219 struct pack_entry e;
3220
3221 if (find_pack_entry(sha1, &e))
3222 return 1;
3223 if (has_loose_object(sha1))
3224 return 1;
3225 if (flags & HAS_SHA1_QUICK)
3226 return 0;
3227 reprepare_packed_git();
3228 return find_pack_entry(sha1, &e);
3229 }
3230
3231 static void check_tree(const void *buf, size_t size)
3232 {
3233 struct tree_desc desc;
3234 struct name_entry entry;
3235
3236 init_tree_desc(&desc, buf, size);
3237 while (tree_entry(&desc, &entry))
3238 /* do nothing
3239 * tree_entry() will die() on malformed entries */
3240 ;
3241 }
3242
3243 static void check_commit(const void *buf, size_t size)
3244 {
3245 struct commit c;
3246 memset(&c, 0, sizeof(c));
3247 if (parse_commit_buffer(&c, buf, size))
3248 die("corrupt commit");
3249 }
3250
3251 static void check_tag(const void *buf, size_t size)
3252 {
3253 struct tag t;
3254 memset(&t, 0, sizeof(t));
3255 if (parse_tag_buffer(&t, buf, size))
3256 die("corrupt tag");
3257 }
3258
3259 static int index_mem(unsigned char *sha1, void *buf, size_t size,
3260 enum object_type type,
3261 const char *path, unsigned flags)
3262 {
3263 int ret, re_allocated = 0;
3264 int write_object = flags & HASH_WRITE_OBJECT;
3265
3266 if (!type)
3267 type = OBJ_BLOB;
3268
3269 /*
3270 * Convert blobs to git internal format
3271 */
3272 if ((type == OBJ_BLOB) && path) {
3273 struct strbuf nbuf = STRBUF_INIT;
3274 if (convert_to_git(path, buf, size, &nbuf,
3275 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3276 buf = strbuf_detach(&nbuf, &size);
3277 re_allocated = 1;
3278 }
3279 }
3280 if (flags & HASH_FORMAT_CHECK) {
3281 if (type == OBJ_TREE)
3282 check_tree(buf, size);
3283 if (type == OBJ_COMMIT)
3284 check_commit(buf, size);
3285 if (type == OBJ_TAG)
3286 check_tag(buf, size);
3287 }
3288
3289 if (write_object)
3290 ret = write_sha1_file(buf, size, typename(type), sha1);
3291 else
3292 ret = hash_sha1_file(buf, size, typename(type), sha1);
3293 if (re_allocated)
3294 free(buf);
3295 return ret;
3296 }
3297
3298 static int index_stream_convert_blob(unsigned char *sha1, int fd,
3299 const char *path, unsigned flags)
3300 {
3301 int ret;
3302 const int write_object = flags & HASH_WRITE_OBJECT;
3303 struct strbuf sbuf = STRBUF_INIT;
3304
3305 assert(path);
3306 assert(would_convert_to_git_filter_fd(path));
3307
3308 convert_to_git_filter_fd(path, fd, &sbuf,
3309 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3310
3311 if (write_object)
3312 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3313 sha1);
3314 else
3315 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3316 sha1);
3317 strbuf_release(&sbuf);
3318 return ret;
3319 }
3320
3321 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3322 const char *path, unsigned flags)
3323 {
3324 struct strbuf sbuf = STRBUF_INIT;
3325 int ret;
3326
3327 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3328 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3329 else
3330 ret = -1;
3331 strbuf_release(&sbuf);
3332 return ret;
3333 }
3334
3335 #define SMALL_FILE_SIZE (32*1024)
3336
3337 static int index_core(unsigned char *sha1, int fd, size_t size,
3338 enum object_type type, const char *path,
3339 unsigned flags)
3340 {
3341 int ret;
3342
3343 if (!size) {
3344 ret = index_mem(sha1, "", size, type, path, flags);
3345 } else if (size <= SMALL_FILE_SIZE) {
3346 char *buf = xmalloc(size);
3347 if (size == read_in_full(fd, buf, size))
3348 ret = index_mem(sha1, buf, size, type, path, flags);
3349 else
3350 ret = error("short read %s", strerror(errno));
3351 free(buf);
3352 } else {
3353 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3354 ret = index_mem(sha1, buf, size, type, path, flags);
3355 munmap(buf, size);
3356 }
3357 return ret;
3358 }
3359
3360 /*
3361 * This creates one packfile per large blob unless bulk-checkin
3362 * machinery is "plugged".
3363 *
3364 * This also bypasses the usual "convert-to-git" dance, and that is on
3365 * purpose. We could write a streaming version of the converting
3366 * functions and insert that before feeding the data to fast-import
3367 * (or equivalent in-core API described above). However, that is
3368 * somewhat complicated, as we do not know the size of the filter
3369 * result, which we need to know beforehand when writing a git object.
3370 * Since the primary motivation for trying to stream from the working
3371 * tree file and to avoid mmaping it in core is to deal with large
3372 * binary blobs, they generally do not want to get any conversion, and
3373 * callers should avoid this code path when filters are requested.
3374 */
3375 static int index_stream(unsigned char *sha1, int fd, size_t size,
3376 enum object_type type, const char *path,
3377 unsigned flags)
3378 {
3379 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3380 }
3381
3382 int index_fd(unsigned char *sha1, int fd, struct stat *st,
3383 enum object_type type, const char *path, unsigned flags)
3384 {
3385 int ret;
3386
3387 /*
3388 * Call xsize_t() only when needed to avoid potentially unnecessary
3389 * die() for large files.
3390 */
3391 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3392 ret = index_stream_convert_blob(sha1, fd, path, flags);
3393 else if (!S_ISREG(st->st_mode))
3394 ret = index_pipe(sha1, fd, type, path, flags);
3395 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3396 (path && would_convert_to_git(path)))
3397 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3398 flags);
3399 else
3400 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3401 flags);
3402 close(fd);
3403 return ret;
3404 }
3405
3406 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3407 {
3408 int fd;
3409 struct strbuf sb = STRBUF_INIT;
3410
3411 switch (st->st_mode & S_IFMT) {
3412 case S_IFREG:
3413 fd = open(path, O_RDONLY);
3414 if (fd < 0)
3415 return error("open(\"%s\"): %s", path,
3416 strerror(errno));
3417 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3418 return error("%s: failed to insert into database",
3419 path);
3420 break;
3421 case S_IFLNK:
3422 if (strbuf_readlink(&sb, path, st->st_size)) {
3423 char *errstr = strerror(errno);
3424 return error("readlink(\"%s\"): %s", path,
3425 errstr);
3426 }
3427 if (!(flags & HASH_WRITE_OBJECT))
3428 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3429 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3430 return error("%s: failed to insert into database",
3431 path);
3432 strbuf_release(&sb);
3433 break;
3434 case S_IFDIR:
3435 return resolve_gitlink_ref(path, "HEAD", sha1);
3436 default:
3437 return error("%s: unsupported file type", path);
3438 }
3439 return 0;
3440 }
3441
3442 int read_pack_header(int fd, struct pack_header *header)
3443 {
3444 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3445 /* "eof before pack header was fully read" */
3446 return PH_ERROR_EOF;
3447
3448 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3449 /* "protocol error (pack signature mismatch detected)" */
3450 return PH_ERROR_PACK_SIGNATURE;
3451 if (!pack_version_ok(header->hdr_version))
3452 /* "protocol error (pack version unsupported)" */
3453 return PH_ERROR_PROTOCOL;
3454 return 0;
3455 }
3456
3457 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3458 {
3459 enum object_type type = sha1_object_info(sha1, NULL);
3460 if (type < 0)
3461 die("%s is not a valid object", sha1_to_hex(sha1));
3462 if (type != expect)
3463 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3464 typename(expect));
3465 }
3466
3467 static int for_each_file_in_obj_subdir(int subdir_nr,
3468 struct strbuf *path,
3469 each_loose_object_fn obj_cb,
3470 each_loose_cruft_fn cruft_cb,
3471 each_loose_subdir_fn subdir_cb,
3472 void *data)
3473 {
3474 size_t baselen = path->len;
3475 DIR *dir = opendir(path->buf);
3476 struct dirent *de;
3477 int r = 0;
3478
3479 if (!dir) {
3480 if (errno == ENOENT)
3481 return 0;
3482 return error("unable to open %s: %s", path->buf, strerror(errno));
3483 }
3484
3485 while ((de = readdir(dir))) {
3486 if (is_dot_or_dotdot(de->d_name))
3487 continue;
3488
3489 strbuf_setlen(path, baselen);
3490 strbuf_addf(path, "/%s", de->d_name);
3491
3492 if (strlen(de->d_name) == 38) {
3493 char hex[41];
3494 unsigned char sha1[20];
3495
3496 snprintf(hex, sizeof(hex), "%02x%s",
3497 subdir_nr, de->d_name);
3498 if (!get_sha1_hex(hex, sha1)) {
3499 if (obj_cb) {
3500 r = obj_cb(sha1, path->buf, data);
3501 if (r)
3502 break;
3503 }
3504 continue;
3505 }
3506 }
3507
3508 if (cruft_cb) {
3509 r = cruft_cb(de->d_name, path->buf, data);
3510 if (r)
3511 break;
3512 }
3513 }
3514 strbuf_setlen(path, baselen);
3515
3516 if (!r && subdir_cb)
3517 r = subdir_cb(subdir_nr, path->buf, data);
3518
3519 closedir(dir);
3520 return r;
3521 }
3522
3523 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3524 each_loose_object_fn obj_cb,
3525 each_loose_cruft_fn cruft_cb,
3526 each_loose_subdir_fn subdir_cb,
3527 void *data)
3528 {
3529 size_t baselen = path->len;
3530 int r = 0;
3531 int i;
3532
3533 for (i = 0; i < 256; i++) {
3534 strbuf_addf(path, "/%02x", i);
3535 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3536 subdir_cb, data);
3537 strbuf_setlen(path, baselen);
3538 if (r)
3539 break;
3540 }
3541
3542 return r;
3543 }
3544
3545 int for_each_loose_file_in_objdir(const char *path,
3546 each_loose_object_fn obj_cb,
3547 each_loose_cruft_fn cruft_cb,
3548 each_loose_subdir_fn subdir_cb,
3549 void *data)
3550 {
3551 struct strbuf buf = STRBUF_INIT;
3552 int r;
3553
3554 strbuf_addstr(&buf, path);
3555 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3556 subdir_cb, data);
3557 strbuf_release(&buf);
3558
3559 return r;
3560 }
3561
3562 struct loose_alt_odb_data {
3563 each_loose_object_fn *cb;
3564 void *data;
3565 };
3566
3567 static int loose_from_alt_odb(struct alternate_object_database *alt,
3568 void *vdata)
3569 {
3570 struct loose_alt_odb_data *data = vdata;
3571 struct strbuf buf = STRBUF_INIT;
3572 int r;
3573
3574 /* copy base not including trailing '/' */
3575 strbuf_add(&buf, alt->base, alt->name - alt->base - 1);
3576 r = for_each_loose_file_in_objdir_buf(&buf,
3577 data->cb, NULL, NULL,
3578 data->data);
3579 strbuf_release(&buf);
3580 return r;
3581 }
3582
3583 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3584 {
3585 struct loose_alt_odb_data alt;
3586 int r;
3587
3588 r = for_each_loose_file_in_objdir(get_object_directory(),
3589 cb, NULL, NULL, data);
3590 if (r)
3591 return r;
3592
3593 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3594 return 0;
3595
3596 alt.cb = cb;
3597 alt.data = data;
3598 return foreach_alt_odb(loose_from_alt_odb, &alt);
3599 }
3600
3601 static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3602 {
3603 uint32_t i;
3604 int r = 0;
3605
3606 for (i = 0; i < p->num_objects; i++) {
3607 const unsigned char *sha1 = nth_packed_object_sha1(p, i);
3608
3609 if (!sha1)
3610 return error("unable to get sha1 of object %u in %s",
3611 i, p->pack_name);
3612
3613 r = cb(sha1, p, i, data);
3614 if (r)
3615 break;
3616 }
3617 return r;
3618 }
3619
3620 int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3621 {
3622 struct packed_git *p;
3623 int r = 0;
3624 int pack_errors = 0;
3625
3626 prepare_packed_git();
3627 for (p = packed_git; p; p = p->next) {
3628 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3629 continue;
3630 if (open_pack_index(p)) {
3631 pack_errors = 1;
3632 continue;
3633 }
3634 r = for_each_object_in_pack(p, cb, data);
3635 if (r)
3636 break;
3637 }
3638 return r ? r : pack_errors;
3639 }