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