c26fdcbd8840ede1865210a09ee6c6f85794985b
[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 struct strbuf *buf,
212 const char *which)
213 {
214 strbuf_reset(buf);
215 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
216 sha1_to_hex(sha1), which);
217 return buf->buf;
218 }
219
220 char *sha1_pack_name(const unsigned char *sha1)
221 {
222 static struct strbuf buf = STRBUF_INIT;
223 return sha1_get_pack_name(sha1, &buf, "pack");
224 }
225
226 char *sha1_pack_index_name(const unsigned char *sha1)
227 {
228 static struct strbuf buf = STRBUF_INIT;
229 return sha1_get_pack_name(sha1, &buf, "idx");
230 }
231
232 struct alternate_object_database *alt_odb_list;
233 static struct alternate_object_database **alt_odb_tail;
234
235 /*
236 * Prepare alternate object database registry.
237 *
238 * The variable alt_odb_list points at the list of struct
239 * alternate_object_database. The elements on this list come from
240 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
241 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
242 * whose contents is similar to that environment variable but can be
243 * LF separated. Its base points at a statically allocated buffer that
244 * contains "/the/directory/corresponding/to/.git/objects/...", while
245 * its name points just after the slash at the end of ".git/objects/"
246 * in the example above, and has enough space to hold 40-byte hex
247 * SHA1, an extra slash for the first level indirection, and the
248 * terminating NUL.
249 */
250 static int link_alt_odb_entry(const char *entry, const char *relative_base,
251 int depth, const char *normalized_objdir)
252 {
253 struct alternate_object_database *ent;
254 struct alternate_object_database *alt;
255 int pfxlen, entlen;
256 struct strbuf pathbuf = STRBUF_INIT;
257
258 if (!is_absolute_path(entry) && relative_base) {
259 strbuf_addstr(&pathbuf, real_path(relative_base));
260 strbuf_addch(&pathbuf, '/');
261 }
262 strbuf_addstr(&pathbuf, entry);
263
264 normalize_path_copy(pathbuf.buf, pathbuf.buf);
265
266 pfxlen = strlen(pathbuf.buf);
267
268 /*
269 * The trailing slash after the directory name is given by
270 * this function at the end. Remove duplicates.
271 */
272 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
273 pfxlen -= 1;
274
275 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
276 ent = xmalloc(sizeof(*ent) + entlen);
277 memcpy(ent->base, pathbuf.buf, pfxlen);
278 strbuf_release(&pathbuf);
279
280 ent->name = ent->base + pfxlen + 1;
281 ent->base[pfxlen + 3] = '/';
282 ent->base[pfxlen] = ent->base[entlen-1] = 0;
283
284 /* Detect cases where alternate disappeared */
285 if (!is_directory(ent->base)) {
286 error("object directory %s does not exist; "
287 "check .git/objects/info/alternates.",
288 ent->base);
289 free(ent);
290 return -1;
291 }
292
293 /* Prevent the common mistake of listing the same
294 * thing twice, or object directory itself.
295 */
296 for (alt = alt_odb_list; alt; alt = alt->next) {
297 if (pfxlen == alt->name - alt->base - 1 &&
298 !memcmp(ent->base, alt->base, pfxlen)) {
299 free(ent);
300 return -1;
301 }
302 }
303 if (!strcmp_icase(ent->base, normalized_objdir)) {
304 free(ent);
305 return -1;
306 }
307
308 /* add the alternate entry */
309 *alt_odb_tail = ent;
310 alt_odb_tail = &(ent->next);
311 ent->next = NULL;
312
313 /* recursively add alternates */
314 read_info_alternates(ent->base, depth + 1);
315
316 ent->base[pfxlen] = '/';
317
318 return 0;
319 }
320
321 static void link_alt_odb_entries(const char *alt, int len, int sep,
322 const char *relative_base, int depth)
323 {
324 struct string_list entries = STRING_LIST_INIT_NODUP;
325 char *alt_copy;
326 int i;
327 struct strbuf objdirbuf = STRBUF_INIT;
328
329 if (depth > 5) {
330 error("%s: ignoring alternate object stores, nesting too deep.",
331 relative_base);
332 return;
333 }
334
335 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
336 normalize_path_copy(objdirbuf.buf, objdirbuf.buf);
337
338 alt_copy = xmemdupz(alt, len);
339 string_list_split_in_place(&entries, alt_copy, sep, -1);
340 for (i = 0; i < entries.nr; i++) {
341 const char *entry = entries.items[i].string;
342 if (entry[0] == '\0' || entry[0] == '#')
343 continue;
344 if (!is_absolute_path(entry) && depth) {
345 error("%s: ignoring relative alternate object store %s",
346 relative_base, entry);
347 } else {
348 link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
349 }
350 }
351 string_list_clear(&entries, 0);
352 free(alt_copy);
353 strbuf_release(&objdirbuf);
354 }
355
356 void read_info_alternates(const char * relative_base, int depth)
357 {
358 char *map;
359 size_t mapsz;
360 struct stat st;
361 char *path;
362 int fd;
363
364 path = xstrfmt("%s/info/alternates", relative_base);
365 fd = git_open_noatime(path);
366 free(path);
367 if (fd < 0)
368 return;
369 if (fstat(fd, &st) || (st.st_size == 0)) {
370 close(fd);
371 return;
372 }
373 mapsz = xsize_t(st.st_size);
374 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
375 close(fd);
376
377 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
378
379 munmap(map, mapsz);
380 }
381
382 void add_to_alternates_file(const char *reference)
383 {
384 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
385 char *alts = git_pathdup("objects/info/alternates");
386 FILE *in, *out;
387
388 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
389 out = fdopen_lock_file(lock, "w");
390 if (!out)
391 die_errno("unable to fdopen alternates lockfile");
392
393 in = fopen(alts, "r");
394 if (in) {
395 struct strbuf line = STRBUF_INIT;
396 int found = 0;
397
398 while (strbuf_getline(&line, in, '\n') != EOF) {
399 if (!strcmp(reference, line.buf)) {
400 found = 1;
401 break;
402 }
403 fprintf_or_die(out, "%s\n", line.buf);
404 }
405
406 strbuf_release(&line);
407 fclose(in);
408
409 if (found) {
410 rollback_lock_file(lock);
411 lock = NULL;
412 }
413 }
414 else if (errno != ENOENT)
415 die_errno("unable to read alternates file");
416
417 if (lock) {
418 fprintf_or_die(out, "%s\n", reference);
419 if (commit_lock_file(lock))
420 die_errno("unable to move new alternates file into place");
421 if (alt_odb_tail)
422 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
423 }
424 free(alts);
425 }
426
427 int foreach_alt_odb(alt_odb_fn fn, void *cb)
428 {
429 struct alternate_object_database *ent;
430 int r = 0;
431
432 prepare_alt_odb();
433 for (ent = alt_odb_list; ent; ent = ent->next) {
434 r = fn(ent, cb);
435 if (r)
436 break;
437 }
438 return r;
439 }
440
441 void prepare_alt_odb(void)
442 {
443 const char *alt;
444
445 if (alt_odb_tail)
446 return;
447
448 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
449 if (!alt) alt = "";
450
451 alt_odb_tail = &alt_odb_list;
452 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
453
454 read_info_alternates(get_object_directory(), 0);
455 }
456
457 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
458 static int freshen_file(const char *fn)
459 {
460 struct utimbuf t;
461 t.actime = t.modtime = time(NULL);
462 return !utime(fn, &t);
463 }
464
465 /*
466 * All of the check_and_freshen functions return 1 if the file exists and was
467 * freshened (if freshening was requested), 0 otherwise. If they return
468 * 0, you should not assume that it is safe to skip a write of the object (it
469 * either does not exist on disk, or has a stale mtime and may be subject to
470 * pruning).
471 */
472 static int check_and_freshen_file(const char *fn, int freshen)
473 {
474 if (access(fn, F_OK))
475 return 0;
476 if (freshen && !freshen_file(fn))
477 return 0;
478 return 1;
479 }
480
481 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
482 {
483 return check_and_freshen_file(sha1_file_name(sha1), freshen);
484 }
485
486 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
487 {
488 struct alternate_object_database *alt;
489 prepare_alt_odb();
490 for (alt = alt_odb_list; alt; alt = alt->next) {
491 fill_sha1_path(alt->name, sha1);
492 if (check_and_freshen_file(alt->base, freshen))
493 return 1;
494 }
495 return 0;
496 }
497
498 static int check_and_freshen(const unsigned char *sha1, int freshen)
499 {
500 return check_and_freshen_local(sha1, freshen) ||
501 check_and_freshen_nonlocal(sha1, freshen);
502 }
503
504 int has_loose_object_nonlocal(const unsigned char *sha1)
505 {
506 return check_and_freshen_nonlocal(sha1, 0);
507 }
508
509 static int has_loose_object(const unsigned char *sha1)
510 {
511 return check_and_freshen(sha1, 0);
512 }
513
514 static unsigned int pack_used_ctr;
515 static unsigned int pack_mmap_calls;
516 static unsigned int peak_pack_open_windows;
517 static unsigned int pack_open_windows;
518 static unsigned int pack_open_fds;
519 static unsigned int pack_max_fds;
520 static size_t peak_pack_mapped;
521 static size_t pack_mapped;
522 struct packed_git *packed_git;
523
524 void pack_report(void)
525 {
526 fprintf(stderr,
527 "pack_report: getpagesize() = %10" SZ_FMT "\n"
528 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
529 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
530 sz_fmt(getpagesize()),
531 sz_fmt(packed_git_window_size),
532 sz_fmt(packed_git_limit));
533 fprintf(stderr,
534 "pack_report: pack_used_ctr = %10u\n"
535 "pack_report: pack_mmap_calls = %10u\n"
536 "pack_report: pack_open_windows = %10u / %10u\n"
537 "pack_report: pack_mapped = "
538 "%10" SZ_FMT " / %10" SZ_FMT "\n",
539 pack_used_ctr,
540 pack_mmap_calls,
541 pack_open_windows, peak_pack_open_windows,
542 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
543 }
544
545 /*
546 * Open and mmap the index file at path, perform a couple of
547 * consistency checks, then record its information to p. Return 0 on
548 * success.
549 */
550 static int check_packed_git_idx(const char *path, struct packed_git *p)
551 {
552 void *idx_map;
553 struct pack_idx_header *hdr;
554 size_t idx_size;
555 uint32_t version, nr, i, *index;
556 int fd = git_open_noatime(path);
557 struct stat st;
558
559 if (fd < 0)
560 return -1;
561 if (fstat(fd, &st)) {
562 close(fd);
563 return -1;
564 }
565 idx_size = xsize_t(st.st_size);
566 if (idx_size < 4 * 256 + 20 + 20) {
567 close(fd);
568 return error("index file %s is too small", path);
569 }
570 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
571 close(fd);
572
573 hdr = idx_map;
574 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
575 version = ntohl(hdr->idx_version);
576 if (version < 2 || version > 2) {
577 munmap(idx_map, idx_size);
578 return error("index file %s is version %"PRIu32
579 " and is not supported by this binary"
580 " (try upgrading GIT to a newer version)",
581 path, version);
582 }
583 } else
584 version = 1;
585
586 nr = 0;
587 index = idx_map;
588 if (version > 1)
589 index += 2; /* skip index header */
590 for (i = 0; i < 256; i++) {
591 uint32_t n = ntohl(index[i]);
592 if (n < nr) {
593 munmap(idx_map, idx_size);
594 return error("non-monotonic index %s", path);
595 }
596 nr = n;
597 }
598
599 if (version == 1) {
600 /*
601 * Total size:
602 * - 256 index entries 4 bytes each
603 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
604 * - 20-byte SHA1 of the packfile
605 * - 20-byte SHA1 file checksum
606 */
607 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
608 munmap(idx_map, idx_size);
609 return error("wrong index v1 file size in %s", path);
610 }
611 } else if (version == 2) {
612 /*
613 * Minimum size:
614 * - 8 bytes of header
615 * - 256 index entries 4 bytes each
616 * - 20-byte sha1 entry * nr
617 * - 4-byte crc entry * nr
618 * - 4-byte offset entry * nr
619 * - 20-byte SHA1 of the packfile
620 * - 20-byte SHA1 file checksum
621 * And after the 4-byte offset table might be a
622 * variable sized table containing 8-byte entries
623 * for offsets larger than 2^31.
624 */
625 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
626 unsigned long max_size = min_size;
627 if (nr)
628 max_size += (nr - 1)*8;
629 if (idx_size < min_size || idx_size > max_size) {
630 munmap(idx_map, idx_size);
631 return error("wrong index v2 file size in %s", path);
632 }
633 if (idx_size != min_size &&
634 /*
635 * make sure we can deal with large pack offsets.
636 * 31-bit signed offset won't be enough, neither
637 * 32-bit unsigned one will be.
638 */
639 (sizeof(off_t) <= 4)) {
640 munmap(idx_map, idx_size);
641 return error("pack too large for current definition of off_t in %s", path);
642 }
643 }
644
645 p->index_version = version;
646 p->index_data = idx_map;
647 p->index_size = idx_size;
648 p->num_objects = nr;
649 return 0;
650 }
651
652 int open_pack_index(struct packed_git *p)
653 {
654 char *idx_name;
655 size_t len;
656 int ret;
657
658 if (p->index_data)
659 return 0;
660
661 if (!strip_suffix(p->pack_name, ".pack", &len))
662 die("BUG: pack_name does not end in .pack");
663 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
664 ret = check_packed_git_idx(idx_name, p);
665 free(idx_name);
666 return ret;
667 }
668
669 static void scan_windows(struct packed_git *p,
670 struct packed_git **lru_p,
671 struct pack_window **lru_w,
672 struct pack_window **lru_l)
673 {
674 struct pack_window *w, *w_l;
675
676 for (w_l = NULL, w = p->windows; w; w = w->next) {
677 if (!w->inuse_cnt) {
678 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
679 *lru_p = p;
680 *lru_w = w;
681 *lru_l = w_l;
682 }
683 }
684 w_l = w;
685 }
686 }
687
688 static int unuse_one_window(struct packed_git *current)
689 {
690 struct packed_git *p, *lru_p = NULL;
691 struct pack_window *lru_w = NULL, *lru_l = NULL;
692
693 if (current)
694 scan_windows(current, &lru_p, &lru_w, &lru_l);
695 for (p = packed_git; p; p = p->next)
696 scan_windows(p, &lru_p, &lru_w, &lru_l);
697 if (lru_p) {
698 munmap(lru_w->base, lru_w->len);
699 pack_mapped -= lru_w->len;
700 if (lru_l)
701 lru_l->next = lru_w->next;
702 else
703 lru_p->windows = lru_w->next;
704 free(lru_w);
705 pack_open_windows--;
706 return 1;
707 }
708 return 0;
709 }
710
711 void release_pack_memory(size_t need)
712 {
713 size_t cur = pack_mapped;
714 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
715 ; /* nothing */
716 }
717
718 static void mmap_limit_check(size_t length)
719 {
720 static size_t limit = 0;
721 if (!limit) {
722 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
723 if (!limit)
724 limit = SIZE_MAX;
725 }
726 if (length > limit)
727 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
728 (uintmax_t)length, (uintmax_t)limit);
729 }
730
731 void *xmmap_gently(void *start, size_t length,
732 int prot, int flags, int fd, off_t offset)
733 {
734 void *ret;
735
736 mmap_limit_check(length);
737 ret = mmap(start, length, prot, flags, fd, offset);
738 if (ret == MAP_FAILED) {
739 if (!length)
740 return NULL;
741 release_pack_memory(length);
742 ret = mmap(start, length, prot, flags, fd, offset);
743 }
744 return ret;
745 }
746
747 void *xmmap(void *start, size_t length,
748 int prot, int flags, int fd, off_t offset)
749 {
750 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
751 if (ret == MAP_FAILED)
752 die_errno("mmap failed");
753 return ret;
754 }
755
756 void close_pack_windows(struct packed_git *p)
757 {
758 while (p->windows) {
759 struct pack_window *w = p->windows;
760
761 if (w->inuse_cnt)
762 die("pack '%s' still has open windows to it",
763 p->pack_name);
764 munmap(w->base, w->len);
765 pack_mapped -= w->len;
766 pack_open_windows--;
767 p->windows = w->next;
768 free(w);
769 }
770 }
771
772 /*
773 * The LRU pack is the one with the oldest MRU window, preferring packs
774 * with no used windows, or the oldest mtime if it has no windows allocated.
775 */
776 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
777 {
778 struct pack_window *w, *this_mru_w;
779 int has_windows_inuse = 0;
780
781 /*
782 * Reject this pack if it has windows and the previously selected
783 * one does not. If this pack does not have windows, reject
784 * it if the pack file is newer than the previously selected one.
785 */
786 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
787 return;
788
789 for (w = this_mru_w = p->windows; w; w = w->next) {
790 /*
791 * Reject this pack if any of its windows are in use,
792 * but the previously selected pack did not have any
793 * inuse windows. Otherwise, record that this pack
794 * has windows in use.
795 */
796 if (w->inuse_cnt) {
797 if (*accept_windows_inuse)
798 has_windows_inuse = 1;
799 else
800 return;
801 }
802
803 if (w->last_used > this_mru_w->last_used)
804 this_mru_w = w;
805
806 /*
807 * Reject this pack if it has windows that have been
808 * used more recently than the previously selected pack.
809 * If the previously selected pack had windows inuse and
810 * we have not encountered a window in this pack that is
811 * inuse, skip this check since we prefer a pack with no
812 * inuse windows to one that has inuse windows.
813 */
814 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
815 this_mru_w->last_used > (*mru_w)->last_used)
816 return;
817 }
818
819 /*
820 * Select this pack.
821 */
822 *mru_w = this_mru_w;
823 *lru_p = p;
824 *accept_windows_inuse = has_windows_inuse;
825 }
826
827 static int close_one_pack(void)
828 {
829 struct packed_git *p, *lru_p = NULL;
830 struct pack_window *mru_w = NULL;
831 int accept_windows_inuse = 1;
832
833 for (p = packed_git; p; p = p->next) {
834 if (p->pack_fd == -1)
835 continue;
836 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
837 }
838
839 if (lru_p) {
840 close(lru_p->pack_fd);
841 pack_open_fds--;
842 lru_p->pack_fd = -1;
843 return 1;
844 }
845
846 return 0;
847 }
848
849 void unuse_pack(struct pack_window **w_cursor)
850 {
851 struct pack_window *w = *w_cursor;
852 if (w) {
853 w->inuse_cnt--;
854 *w_cursor = NULL;
855 }
856 }
857
858 void close_pack_index(struct packed_git *p)
859 {
860 if (p->index_data) {
861 munmap((void *)p->index_data, p->index_size);
862 p->index_data = NULL;
863 }
864 }
865
866 /*
867 * This is used by git-repack in case a newly created pack happens to
868 * contain the same set of objects as an existing one. In that case
869 * the resulting file might be different even if its name would be the
870 * same. It is best to close any reference to the old pack before it is
871 * replaced on disk. Of course no index pointers or windows for given pack
872 * must subsist at this point. If ever objects from this pack are requested
873 * again, the new version of the pack will be reinitialized through
874 * reprepare_packed_git().
875 */
876 void free_pack_by_name(const char *pack_name)
877 {
878 struct packed_git *p, **pp = &packed_git;
879
880 while (*pp) {
881 p = *pp;
882 if (strcmp(pack_name, p->pack_name) == 0) {
883 clear_delta_base_cache();
884 close_pack_windows(p);
885 if (p->pack_fd != -1) {
886 close(p->pack_fd);
887 pack_open_fds--;
888 }
889 close_pack_index(p);
890 free(p->bad_object_sha1);
891 *pp = p->next;
892 if (last_found_pack == p)
893 last_found_pack = NULL;
894 free(p);
895 return;
896 }
897 pp = &p->next;
898 }
899 }
900
901 static unsigned int get_max_fd_limit(void)
902 {
903 #ifdef RLIMIT_NOFILE
904 {
905 struct rlimit lim;
906
907 if (!getrlimit(RLIMIT_NOFILE, &lim))
908 return lim.rlim_cur;
909 }
910 #endif
911
912 #ifdef _SC_OPEN_MAX
913 {
914 long open_max = sysconf(_SC_OPEN_MAX);
915 if (0 < open_max)
916 return open_max;
917 /*
918 * Otherwise, we got -1 for one of the two
919 * reasons:
920 *
921 * (1) sysconf() did not understand _SC_OPEN_MAX
922 * and signaled an error with -1; or
923 * (2) sysconf() said there is no limit.
924 *
925 * We _could_ clear errno before calling sysconf() to
926 * tell these two cases apart and return a huge number
927 * in the latter case to let the caller cap it to a
928 * value that is not so selfish, but letting the
929 * fallback OPEN_MAX codepath take care of these cases
930 * is a lot simpler.
931 */
932 }
933 #endif
934
935 #ifdef OPEN_MAX
936 return OPEN_MAX;
937 #else
938 return 1; /* see the caller ;-) */
939 #endif
940 }
941
942 /*
943 * Do not call this directly as this leaks p->pack_fd on error return;
944 * call open_packed_git() instead.
945 */
946 static int open_packed_git_1(struct packed_git *p)
947 {
948 struct stat st;
949 struct pack_header hdr;
950 unsigned char sha1[20];
951 unsigned char *idx_sha1;
952 long fd_flag;
953
954 if (!p->index_data && open_pack_index(p))
955 return error("packfile %s index unavailable", p->pack_name);
956
957 if (!pack_max_fds) {
958 unsigned int max_fds = get_max_fd_limit();
959
960 /* Save 3 for stdin/stdout/stderr, 22 for work */
961 if (25 < max_fds)
962 pack_max_fds = max_fds - 25;
963 else
964 pack_max_fds = 1;
965 }
966
967 while (pack_max_fds <= pack_open_fds && close_one_pack())
968 ; /* nothing */
969
970 p->pack_fd = git_open_noatime(p->pack_name);
971 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
972 return -1;
973 pack_open_fds++;
974
975 /* If we created the struct before we had the pack we lack size. */
976 if (!p->pack_size) {
977 if (!S_ISREG(st.st_mode))
978 return error("packfile %s not a regular file", p->pack_name);
979 p->pack_size = st.st_size;
980 } else if (p->pack_size != st.st_size)
981 return error("packfile %s size changed", p->pack_name);
982
983 /* We leave these file descriptors open with sliding mmap;
984 * there is no point keeping them open across exec(), though.
985 */
986 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
987 if (fd_flag < 0)
988 return error("cannot determine file descriptor flags");
989 fd_flag |= FD_CLOEXEC;
990 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
991 return error("cannot set FD_CLOEXEC");
992
993 /* Verify we recognize this pack file format. */
994 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
995 return error("file %s is far too short to be a packfile", p->pack_name);
996 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
997 return error("file %s is not a GIT packfile", p->pack_name);
998 if (!pack_version_ok(hdr.hdr_version))
999 return error("packfile %s is version %"PRIu32" and not"
1000 " supported (try upgrading GIT to a newer version)",
1001 p->pack_name, ntohl(hdr.hdr_version));
1002
1003 /* Verify the pack matches its index. */
1004 if (p->num_objects != ntohl(hdr.hdr_entries))
1005 return error("packfile %s claims to have %"PRIu32" objects"
1006 " while index indicates %"PRIu32" objects",
1007 p->pack_name, ntohl(hdr.hdr_entries),
1008 p->num_objects);
1009 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
1010 return error("end of packfile %s is unavailable", p->pack_name);
1011 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
1012 return error("packfile %s signature is unavailable", p->pack_name);
1013 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
1014 if (hashcmp(sha1, idx_sha1))
1015 return error("packfile %s does not match index", p->pack_name);
1016 return 0;
1017 }
1018
1019 static int open_packed_git(struct packed_git *p)
1020 {
1021 if (!open_packed_git_1(p))
1022 return 0;
1023 if (p->pack_fd != -1) {
1024 close(p->pack_fd);
1025 pack_open_fds--;
1026 p->pack_fd = -1;
1027 }
1028 return -1;
1029 }
1030
1031 static int in_window(struct pack_window *win, off_t offset)
1032 {
1033 /* We must promise at least 20 bytes (one hash) after the
1034 * offset is available from this window, otherwise the offset
1035 * is not actually in this window and a different window (which
1036 * has that one hash excess) must be used. This is to support
1037 * the object header and delta base parsing routines below.
1038 */
1039 off_t win_off = win->offset;
1040 return win_off <= offset
1041 && (offset + 20) <= (win_off + win->len);
1042 }
1043
1044 unsigned char *use_pack(struct packed_git *p,
1045 struct pack_window **w_cursor,
1046 off_t offset,
1047 unsigned long *left)
1048 {
1049 struct pack_window *win = *w_cursor;
1050
1051 /* Since packfiles end in a hash of their content and it's
1052 * pointless to ask for an offset into the middle of that
1053 * hash, and the in_window function above wouldn't match
1054 * don't allow an offset too close to the end of the file.
1055 */
1056 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1057 die("packfile %s cannot be accessed", p->pack_name);
1058 if (offset > (p->pack_size - 20))
1059 die("offset beyond end of packfile (truncated pack?)");
1060
1061 if (!win || !in_window(win, offset)) {
1062 if (win)
1063 win->inuse_cnt--;
1064 for (win = p->windows; win; win = win->next) {
1065 if (in_window(win, offset))
1066 break;
1067 }
1068 if (!win) {
1069 size_t window_align = packed_git_window_size / 2;
1070 off_t len;
1071
1072 if (p->pack_fd == -1 && open_packed_git(p))
1073 die("packfile %s cannot be accessed", p->pack_name);
1074
1075 win = xcalloc(1, sizeof(*win));
1076 win->offset = (offset / window_align) * window_align;
1077 len = p->pack_size - win->offset;
1078 if (len > packed_git_window_size)
1079 len = packed_git_window_size;
1080 win->len = (size_t)len;
1081 pack_mapped += win->len;
1082 while (packed_git_limit < pack_mapped
1083 && unuse_one_window(p))
1084 ; /* nothing */
1085 win->base = xmmap(NULL, win->len,
1086 PROT_READ, MAP_PRIVATE,
1087 p->pack_fd, win->offset);
1088 if (win->base == MAP_FAILED)
1089 die("packfile %s cannot be mapped: %s",
1090 p->pack_name,
1091 strerror(errno));
1092 if (!win->offset && win->len == p->pack_size
1093 && !p->do_not_close) {
1094 close(p->pack_fd);
1095 pack_open_fds--;
1096 p->pack_fd = -1;
1097 }
1098 pack_mmap_calls++;
1099 pack_open_windows++;
1100 if (pack_mapped > peak_pack_mapped)
1101 peak_pack_mapped = pack_mapped;
1102 if (pack_open_windows > peak_pack_open_windows)
1103 peak_pack_open_windows = pack_open_windows;
1104 win->next = p->windows;
1105 p->windows = win;
1106 }
1107 }
1108 if (win != *w_cursor) {
1109 win->last_used = pack_used_ctr++;
1110 win->inuse_cnt++;
1111 *w_cursor = win;
1112 }
1113 offset -= win->offset;
1114 if (left)
1115 *left = win->len - xsize_t(offset);
1116 return win->base + offset;
1117 }
1118
1119 static struct packed_git *alloc_packed_git(int extra)
1120 {
1121 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1122 memset(p, 0, sizeof(*p));
1123 p->pack_fd = -1;
1124 return p;
1125 }
1126
1127 static void try_to_free_pack_memory(size_t size)
1128 {
1129 release_pack_memory(size);
1130 }
1131
1132 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
1133 {
1134 static int have_set_try_to_free_routine;
1135 struct stat st;
1136 size_t alloc;
1137 struct packed_git *p;
1138
1139 if (!have_set_try_to_free_routine) {
1140 have_set_try_to_free_routine = 1;
1141 set_try_to_free_routine(try_to_free_pack_memory);
1142 }
1143
1144 /*
1145 * Make sure a corresponding .pack file exists and that
1146 * the index looks sane.
1147 */
1148 if (!strip_suffix_mem(path, &path_len, ".idx"))
1149 return NULL;
1150
1151 /*
1152 * ".pack" is long enough to hold any suffix we're adding (and
1153 * the use xsnprintf double-checks that)
1154 */
1155 alloc = path_len + strlen(".pack") + 1;
1156 p = alloc_packed_git(alloc);
1157 memcpy(p->pack_name, path, path_len);
1158
1159 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
1160 if (!access(p->pack_name, F_OK))
1161 p->pack_keep = 1;
1162
1163 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
1164 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1165 free(p);
1166 return NULL;
1167 }
1168
1169 /* ok, it looks sane as far as we can check without
1170 * actually mapping the pack file.
1171 */
1172 p->pack_size = st.st_size;
1173 p->pack_local = local;
1174 p->mtime = st.st_mtime;
1175 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1176 hashclr(p->sha1);
1177 return p;
1178 }
1179
1180 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1181 {
1182 const char *path = sha1_pack_name(sha1);
1183 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
1184
1185 strcpy(p->pack_name, path);
1186 hashcpy(p->sha1, sha1);
1187 if (check_packed_git_idx(idx_path, p)) {
1188 free(p);
1189 return NULL;
1190 }
1191
1192 return p;
1193 }
1194
1195 void install_packed_git(struct packed_git *pack)
1196 {
1197 if (pack->pack_fd != -1)
1198 pack_open_fds++;
1199
1200 pack->next = packed_git;
1201 packed_git = pack;
1202 }
1203
1204 void (*report_garbage)(const char *desc, const char *path);
1205
1206 static void report_helper(const struct string_list *list,
1207 int seen_bits, int first, int last)
1208 {
1209 const char *msg;
1210 switch (seen_bits) {
1211 case 0:
1212 msg = "no corresponding .idx or .pack";
1213 break;
1214 case 1:
1215 msg = "no corresponding .idx";
1216 break;
1217 case 2:
1218 msg = "no corresponding .pack";
1219 break;
1220 default:
1221 return;
1222 }
1223 for (; first < last; first++)
1224 report_garbage(msg, list->items[first].string);
1225 }
1226
1227 static void report_pack_garbage(struct string_list *list)
1228 {
1229 int i, baselen = -1, first = 0, seen_bits = 0;
1230
1231 if (!report_garbage)
1232 return;
1233
1234 string_list_sort(list);
1235
1236 for (i = 0; i < list->nr; i++) {
1237 const char *path = list->items[i].string;
1238 if (baselen != -1 &&
1239 strncmp(path, list->items[first].string, baselen)) {
1240 report_helper(list, seen_bits, first, i);
1241 baselen = -1;
1242 seen_bits = 0;
1243 }
1244 if (baselen == -1) {
1245 const char *dot = strrchr(path, '.');
1246 if (!dot) {
1247 report_garbage("garbage found", path);
1248 continue;
1249 }
1250 baselen = dot - path + 1;
1251 first = i;
1252 }
1253 if (!strcmp(path + baselen, "pack"))
1254 seen_bits |= 1;
1255 else if (!strcmp(path + baselen, "idx"))
1256 seen_bits |= 2;
1257 }
1258 report_helper(list, seen_bits, first, list->nr);
1259 }
1260
1261 static void prepare_packed_git_one(char *objdir, int local)
1262 {
1263 struct strbuf path = STRBUF_INIT;
1264 size_t dirnamelen;
1265 DIR *dir;
1266 struct dirent *de;
1267 struct string_list garbage = STRING_LIST_INIT_DUP;
1268
1269 strbuf_addstr(&path, objdir);
1270 strbuf_addstr(&path, "/pack");
1271 dir = opendir(path.buf);
1272 if (!dir) {
1273 if (errno != ENOENT)
1274 error("unable to open object pack directory: %s: %s",
1275 path.buf, strerror(errno));
1276 strbuf_release(&path);
1277 return;
1278 }
1279 strbuf_addch(&path, '/');
1280 dirnamelen = path.len;
1281 while ((de = readdir(dir)) != NULL) {
1282 struct packed_git *p;
1283 size_t base_len;
1284
1285 if (is_dot_or_dotdot(de->d_name))
1286 continue;
1287
1288 strbuf_setlen(&path, dirnamelen);
1289 strbuf_addstr(&path, de->d_name);
1290
1291 base_len = path.len;
1292 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1293 /* Don't reopen a pack we already have. */
1294 for (p = packed_git; p; p = p->next) {
1295 size_t len;
1296 if (strip_suffix(p->pack_name, ".pack", &len) &&
1297 len == base_len &&
1298 !memcmp(p->pack_name, path.buf, len))
1299 break;
1300 }
1301 if (p == NULL &&
1302 /*
1303 * See if it really is a valid .idx file with
1304 * corresponding .pack file that we can map.
1305 */
1306 (p = add_packed_git(path.buf, path.len, local)) != NULL)
1307 install_packed_git(p);
1308 }
1309
1310 if (!report_garbage)
1311 continue;
1312
1313 if (ends_with(de->d_name, ".idx") ||
1314 ends_with(de->d_name, ".pack") ||
1315 ends_with(de->d_name, ".bitmap") ||
1316 ends_with(de->d_name, ".keep"))
1317 string_list_append(&garbage, path.buf);
1318 else
1319 report_garbage("garbage found", path.buf);
1320 }
1321 closedir(dir);
1322 report_pack_garbage(&garbage);
1323 string_list_clear(&garbage, 0);
1324 strbuf_release(&path);
1325 }
1326
1327 static int sort_pack(const void *a_, const void *b_)
1328 {
1329 struct packed_git *a = *((struct packed_git **)a_);
1330 struct packed_git *b = *((struct packed_git **)b_);
1331 int st;
1332
1333 /*
1334 * Local packs tend to contain objects specific to our
1335 * variant of the project than remote ones. In addition,
1336 * remote ones could be on a network mounted filesystem.
1337 * Favor local ones for these reasons.
1338 */
1339 st = a->pack_local - b->pack_local;
1340 if (st)
1341 return -st;
1342
1343 /*
1344 * Younger packs tend to contain more recent objects,
1345 * and more recent objects tend to get accessed more
1346 * often.
1347 */
1348 if (a->mtime < b->mtime)
1349 return 1;
1350 else if (a->mtime == b->mtime)
1351 return 0;
1352 return -1;
1353 }
1354
1355 static void rearrange_packed_git(void)
1356 {
1357 struct packed_git **ary, *p;
1358 int i, n;
1359
1360 for (n = 0, p = packed_git; p; p = p->next)
1361 n++;
1362 if (n < 2)
1363 return;
1364
1365 /* prepare an array of packed_git for easier sorting */
1366 ary = xcalloc(n, sizeof(struct packed_git *));
1367 for (n = 0, p = packed_git; p; p = p->next)
1368 ary[n++] = p;
1369
1370 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1371
1372 /* link them back again */
1373 for (i = 0; i < n - 1; i++)
1374 ary[i]->next = ary[i + 1];
1375 ary[n - 1]->next = NULL;
1376 packed_git = ary[0];
1377
1378 free(ary);
1379 }
1380
1381 static int prepare_packed_git_run_once = 0;
1382 void prepare_packed_git(void)
1383 {
1384 struct alternate_object_database *alt;
1385
1386 if (prepare_packed_git_run_once)
1387 return;
1388 prepare_packed_git_one(get_object_directory(), 1);
1389 prepare_alt_odb();
1390 for (alt = alt_odb_list; alt; alt = alt->next) {
1391 alt->name[-1] = 0;
1392 prepare_packed_git_one(alt->base, 0);
1393 alt->name[-1] = '/';
1394 }
1395 rearrange_packed_git();
1396 prepare_packed_git_run_once = 1;
1397 }
1398
1399 void reprepare_packed_git(void)
1400 {
1401 prepare_packed_git_run_once = 0;
1402 prepare_packed_git();
1403 }
1404
1405 static void mark_bad_packed_object(struct packed_git *p,
1406 const unsigned char *sha1)
1407 {
1408 unsigned i;
1409 for (i = 0; i < p->num_bad_objects; i++)
1410 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1411 return;
1412 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1413 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1414 p->num_bad_objects++;
1415 }
1416
1417 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1418 {
1419 struct packed_git *p;
1420 unsigned i;
1421
1422 for (p = packed_git; p; p = p->next)
1423 for (i = 0; i < p->num_bad_objects; i++)
1424 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1425 return p;
1426 return NULL;
1427 }
1428
1429 /*
1430 * With an in-core object data in "map", rehash it to make sure the
1431 * object name actually matches "sha1" to detect object corruption.
1432 * With "map" == NULL, try reading the object named with "sha1" using
1433 * the streaming interface and rehash it to do the same.
1434 */
1435 int check_sha1_signature(const unsigned char *sha1, void *map,
1436 unsigned long size, const char *type)
1437 {
1438 unsigned char real_sha1[20];
1439 enum object_type obj_type;
1440 struct git_istream *st;
1441 git_SHA_CTX c;
1442 char hdr[32];
1443 int hdrlen;
1444
1445 if (map) {
1446 hash_sha1_file(map, size, type, real_sha1);
1447 return hashcmp(sha1, real_sha1) ? -1 : 0;
1448 }
1449
1450 st = open_istream(sha1, &obj_type, &size, NULL);
1451 if (!st)
1452 return -1;
1453
1454 /* Generate the header */
1455 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
1456
1457 /* Sha1.. */
1458 git_SHA1_Init(&c);
1459 git_SHA1_Update(&c, hdr, hdrlen);
1460 for (;;) {
1461 char buf[1024 * 16];
1462 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1463
1464 if (readlen < 0) {
1465 close_istream(st);
1466 return -1;
1467 }
1468 if (!readlen)
1469 break;
1470 git_SHA1_Update(&c, buf, readlen);
1471 }
1472 git_SHA1_Final(real_sha1, &c);
1473 close_istream(st);
1474 return hashcmp(sha1, real_sha1) ? -1 : 0;
1475 }
1476
1477 int git_open_noatime(const char *name)
1478 {
1479 static int sha1_file_open_flag = O_NOATIME;
1480
1481 for (;;) {
1482 int fd;
1483
1484 errno = 0;
1485 fd = open(name, O_RDONLY | sha1_file_open_flag);
1486 if (fd >= 0)
1487 return fd;
1488
1489 /* Might the failure be due to O_NOATIME? */
1490 if (errno != ENOENT && sha1_file_open_flag) {
1491 sha1_file_open_flag = 0;
1492 continue;
1493 }
1494
1495 return -1;
1496 }
1497 }
1498
1499 static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1500 {
1501 struct alternate_object_database *alt;
1502
1503 if (!lstat(sha1_file_name(sha1), st))
1504 return 0;
1505
1506 prepare_alt_odb();
1507 errno = ENOENT;
1508 for (alt = alt_odb_list; alt; alt = alt->next) {
1509 fill_sha1_path(alt->name, sha1);
1510 if (!lstat(alt->base, st))
1511 return 0;
1512 }
1513
1514 return -1;
1515 }
1516
1517 static int open_sha1_file(const unsigned char *sha1)
1518 {
1519 int fd;
1520 struct alternate_object_database *alt;
1521 int most_interesting_errno;
1522
1523 fd = git_open_noatime(sha1_file_name(sha1));
1524 if (fd >= 0)
1525 return fd;
1526 most_interesting_errno = errno;
1527
1528 prepare_alt_odb();
1529 for (alt = alt_odb_list; alt; alt = alt->next) {
1530 fill_sha1_path(alt->name, sha1);
1531 fd = git_open_noatime(alt->base);
1532 if (fd >= 0)
1533 return fd;
1534 if (most_interesting_errno == ENOENT)
1535 most_interesting_errno = errno;
1536 }
1537 errno = most_interesting_errno;
1538 return -1;
1539 }
1540
1541 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1542 {
1543 void *map;
1544 int fd;
1545
1546 fd = open_sha1_file(sha1);
1547 map = NULL;
1548 if (fd >= 0) {
1549 struct stat st;
1550
1551 if (!fstat(fd, &st)) {
1552 *size = xsize_t(st.st_size);
1553 if (!*size) {
1554 /* mmap() is forbidden on empty files */
1555 error("object file %s is empty", sha1_file_name(sha1));
1556 return NULL;
1557 }
1558 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1559 }
1560 close(fd);
1561 }
1562 return map;
1563 }
1564
1565 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1566 unsigned long len, enum object_type *type, unsigned long *sizep)
1567 {
1568 unsigned shift;
1569 unsigned long size, c;
1570 unsigned long used = 0;
1571
1572 c = buf[used++];
1573 *type = (c >> 4) & 7;
1574 size = c & 15;
1575 shift = 4;
1576 while (c & 0x80) {
1577 if (len <= used || bitsizeof(long) <= shift) {
1578 error("bad object header");
1579 size = used = 0;
1580 break;
1581 }
1582 c = buf[used++];
1583 size += (c & 0x7f) << shift;
1584 shift += 7;
1585 }
1586 *sizep = size;
1587 return used;
1588 }
1589
1590 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1591 {
1592 /* Get the data stream */
1593 memset(stream, 0, sizeof(*stream));
1594 stream->next_in = map;
1595 stream->avail_in = mapsize;
1596 stream->next_out = buffer;
1597 stream->avail_out = bufsiz;
1598
1599 git_inflate_init(stream);
1600 return git_inflate(stream, 0);
1601 }
1602
1603 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1604 unsigned long mapsize, void *buffer,
1605 unsigned long bufsiz, struct strbuf *header)
1606 {
1607 int status;
1608
1609 status = unpack_sha1_header(stream, map, mapsize, buffer, bufsiz);
1610
1611 /*
1612 * Check if entire header is unpacked in the first iteration.
1613 */
1614 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1615 return 0;
1616
1617 /*
1618 * buffer[0..bufsiz] was not large enough. Copy the partial
1619 * result out to header, and then append the result of further
1620 * reading the stream.
1621 */
1622 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1623 stream->next_out = buffer;
1624 stream->avail_out = bufsiz;
1625
1626 do {
1627 status = git_inflate(stream, 0);
1628 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1629 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1630 return 0;
1631 stream->next_out = buffer;
1632 stream->avail_out = bufsiz;
1633 } while (status != Z_STREAM_END);
1634 return -1;
1635 }
1636
1637 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1638 {
1639 int bytes = strlen(buffer) + 1;
1640 unsigned char *buf = xmallocz(size);
1641 unsigned long n;
1642 int status = Z_OK;
1643
1644 n = stream->total_out - bytes;
1645 if (n > size)
1646 n = size;
1647 memcpy(buf, (char *) buffer + bytes, n);
1648 bytes = n;
1649 if (bytes <= size) {
1650 /*
1651 * The above condition must be (bytes <= size), not
1652 * (bytes < size). In other words, even though we
1653 * expect no more output and set avail_out to zero,
1654 * the input zlib stream may have bytes that express
1655 * "this concludes the stream", and we *do* want to
1656 * eat that input.
1657 *
1658 * Otherwise we would not be able to test that we
1659 * consumed all the input to reach the expected size;
1660 * we also want to check that zlib tells us that all
1661 * went well with status == Z_STREAM_END at the end.
1662 */
1663 stream->next_out = buf + bytes;
1664 stream->avail_out = size - bytes;
1665 while (status == Z_OK)
1666 status = git_inflate(stream, Z_FINISH);
1667 }
1668 if (status == Z_STREAM_END && !stream->avail_in) {
1669 git_inflate_end(stream);
1670 return buf;
1671 }
1672
1673 if (status < 0)
1674 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1675 else if (stream->avail_in)
1676 error("garbage at end of loose object '%s'",
1677 sha1_to_hex(sha1));
1678 free(buf);
1679 return NULL;
1680 }
1681
1682 /*
1683 * We used to just use "sscanf()", but that's actually way
1684 * too permissive for what we want to check. So do an anal
1685 * object header parse by hand.
1686 */
1687 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1688 unsigned int flags)
1689 {
1690 const char *type_buf = hdr;
1691 unsigned long size;
1692 int type, type_len = 0;
1693
1694 /*
1695 * The type can be of any size but is followed by
1696 * a space.
1697 */
1698 for (;;) {
1699 char c = *hdr++;
1700 if (c == ' ')
1701 break;
1702 type_len++;
1703 }
1704
1705 type = type_from_string_gently(type_buf, type_len, 1);
1706 if (oi->typename)
1707 strbuf_add(oi->typename, type_buf, type_len);
1708 /*
1709 * Set type to 0 if its an unknown object and
1710 * we're obtaining the type using '--allow-unkown-type'
1711 * option.
1712 */
1713 if ((flags & LOOKUP_UNKNOWN_OBJECT) && (type < 0))
1714 type = 0;
1715 else if (type < 0)
1716 die("invalid object type");
1717 if (oi->typep)
1718 *oi->typep = type;
1719
1720 /*
1721 * The length must follow immediately, and be in canonical
1722 * decimal format (ie "010" is not valid).
1723 */
1724 size = *hdr++ - '0';
1725 if (size > 9)
1726 return -1;
1727 if (size) {
1728 for (;;) {
1729 unsigned long c = *hdr - '0';
1730 if (c > 9)
1731 break;
1732 hdr++;
1733 size = size * 10 + c;
1734 }
1735 }
1736
1737 if (oi->sizep)
1738 *oi->sizep = size;
1739
1740 /*
1741 * The length must be followed by a zero byte
1742 */
1743 return *hdr ? -1 : type;
1744 }
1745
1746 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1747 {
1748 struct object_info oi;
1749
1750 oi.sizep = sizep;
1751 oi.typename = NULL;
1752 oi.typep = NULL;
1753 return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
1754 }
1755
1756 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1757 {
1758 int ret;
1759 git_zstream stream;
1760 char hdr[8192];
1761
1762 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1763 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1764 return NULL;
1765
1766 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1767 }
1768
1769 unsigned long get_size_from_delta(struct packed_git *p,
1770 struct pack_window **w_curs,
1771 off_t curpos)
1772 {
1773 const unsigned char *data;
1774 unsigned char delta_head[20], *in;
1775 git_zstream stream;
1776 int st;
1777
1778 memset(&stream, 0, sizeof(stream));
1779 stream.next_out = delta_head;
1780 stream.avail_out = sizeof(delta_head);
1781
1782 git_inflate_init(&stream);
1783 do {
1784 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1785 stream.next_in = in;
1786 st = git_inflate(&stream, Z_FINISH);
1787 curpos += stream.next_in - in;
1788 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1789 stream.total_out < sizeof(delta_head));
1790 git_inflate_end(&stream);
1791 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1792 error("delta data unpack-initial failed");
1793 return 0;
1794 }
1795
1796 /* Examine the initial part of the delta to figure out
1797 * the result size.
1798 */
1799 data = delta_head;
1800
1801 /* ignore base size */
1802 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1803
1804 /* Read the result size */
1805 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1806 }
1807
1808 static off_t get_delta_base(struct packed_git *p,
1809 struct pack_window **w_curs,
1810 off_t *curpos,
1811 enum object_type type,
1812 off_t delta_obj_offset)
1813 {
1814 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1815 off_t base_offset;
1816
1817 /* use_pack() assured us we have [base_info, base_info + 20)
1818 * as a range that we can look at without walking off the
1819 * end of the mapped window. Its actually the hash size
1820 * that is assured. An OFS_DELTA longer than the hash size
1821 * is stupid, as then a REF_DELTA would be smaller to store.
1822 */
1823 if (type == OBJ_OFS_DELTA) {
1824 unsigned used = 0;
1825 unsigned char c = base_info[used++];
1826 base_offset = c & 127;
1827 while (c & 128) {
1828 base_offset += 1;
1829 if (!base_offset || MSB(base_offset, 7))
1830 return 0; /* overflow */
1831 c = base_info[used++];
1832 base_offset = (base_offset << 7) + (c & 127);
1833 }
1834 base_offset = delta_obj_offset - base_offset;
1835 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1836 return 0; /* out of bound */
1837 *curpos += used;
1838 } else if (type == OBJ_REF_DELTA) {
1839 /* The base entry _must_ be in the same pack */
1840 base_offset = find_pack_entry_one(base_info, p);
1841 *curpos += 20;
1842 } else
1843 die("I am totally screwed");
1844 return base_offset;
1845 }
1846
1847 /*
1848 * Like get_delta_base above, but we return the sha1 instead of the pack
1849 * offset. This means it is cheaper for REF deltas (we do not have to do
1850 * the final object lookup), but more expensive for OFS deltas (we
1851 * have to load the revidx to convert the offset back into a sha1).
1852 */
1853 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1854 struct pack_window **w_curs,
1855 off_t curpos,
1856 enum object_type type,
1857 off_t delta_obj_offset)
1858 {
1859 if (type == OBJ_REF_DELTA) {
1860 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1861 return base;
1862 } else if (type == OBJ_OFS_DELTA) {
1863 struct revindex_entry *revidx;
1864 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1865 type, delta_obj_offset);
1866
1867 if (!base_offset)
1868 return NULL;
1869
1870 revidx = find_pack_revindex(p, base_offset);
1871 if (!revidx)
1872 return NULL;
1873
1874 return nth_packed_object_sha1(p, revidx->nr);
1875 } else
1876 return NULL;
1877 }
1878
1879 int unpack_object_header(struct packed_git *p,
1880 struct pack_window **w_curs,
1881 off_t *curpos,
1882 unsigned long *sizep)
1883 {
1884 unsigned char *base;
1885 unsigned long left;
1886 unsigned long used;
1887 enum object_type type;
1888
1889 /* use_pack() assures us we have [base, base + 20) available
1890 * as a range that we can look at. (Its actually the hash
1891 * size that is assured.) With our object header encoding
1892 * the maximum deflated object size is 2^137, which is just
1893 * insane, so we know won't exceed what we have been given.
1894 */
1895 base = use_pack(p, w_curs, *curpos, &left);
1896 used = unpack_object_header_buffer(base, left, &type, sizep);
1897 if (!used) {
1898 type = OBJ_BAD;
1899 } else
1900 *curpos += used;
1901
1902 return type;
1903 }
1904
1905 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1906 {
1907 int type;
1908 struct revindex_entry *revidx;
1909 const unsigned char *sha1;
1910 revidx = find_pack_revindex(p, obj_offset);
1911 if (!revidx)
1912 return OBJ_BAD;
1913 sha1 = nth_packed_object_sha1(p, revidx->nr);
1914 mark_bad_packed_object(p, sha1);
1915 type = sha1_object_info(sha1, NULL);
1916 if (type <= OBJ_NONE)
1917 return OBJ_BAD;
1918 return type;
1919 }
1920
1921 #define POI_STACK_PREALLOC 64
1922
1923 static enum object_type packed_to_object_type(struct packed_git *p,
1924 off_t obj_offset,
1925 enum object_type type,
1926 struct pack_window **w_curs,
1927 off_t curpos)
1928 {
1929 off_t small_poi_stack[POI_STACK_PREALLOC];
1930 off_t *poi_stack = small_poi_stack;
1931 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1932
1933 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1934 off_t base_offset;
1935 unsigned long size;
1936 /* Push the object we're going to leave behind */
1937 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1938 poi_stack_alloc = alloc_nr(poi_stack_nr);
1939 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1940 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1941 } else {
1942 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1943 }
1944 poi_stack[poi_stack_nr++] = obj_offset;
1945 /* If parsing the base offset fails, just unwind */
1946 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1947 if (!base_offset)
1948 goto unwind;
1949 curpos = obj_offset = base_offset;
1950 type = unpack_object_header(p, w_curs, &curpos, &size);
1951 if (type <= OBJ_NONE) {
1952 /* If getting the base itself fails, we first
1953 * retry the base, otherwise unwind */
1954 type = retry_bad_packed_offset(p, base_offset);
1955 if (type > OBJ_NONE)
1956 goto out;
1957 goto unwind;
1958 }
1959 }
1960
1961 switch (type) {
1962 case OBJ_BAD:
1963 case OBJ_COMMIT:
1964 case OBJ_TREE:
1965 case OBJ_BLOB:
1966 case OBJ_TAG:
1967 break;
1968 default:
1969 error("unknown object type %i at offset %"PRIuMAX" in %s",
1970 type, (uintmax_t)obj_offset, p->pack_name);
1971 type = OBJ_BAD;
1972 }
1973
1974 out:
1975 if (poi_stack != small_poi_stack)
1976 free(poi_stack);
1977 return type;
1978
1979 unwind:
1980 while (poi_stack_nr) {
1981 obj_offset = poi_stack[--poi_stack_nr];
1982 type = retry_bad_packed_offset(p, obj_offset);
1983 if (type > OBJ_NONE)
1984 goto out;
1985 }
1986 type = OBJ_BAD;
1987 goto out;
1988 }
1989
1990 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1991 struct object_info *oi)
1992 {
1993 struct pack_window *w_curs = NULL;
1994 unsigned long size;
1995 off_t curpos = obj_offset;
1996 enum object_type type;
1997
1998 /*
1999 * We always get the representation type, but only convert it to
2000 * a "real" type later if the caller is interested.
2001 */
2002 type = unpack_object_header(p, &w_curs, &curpos, &size);
2003
2004 if (oi->sizep) {
2005 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2006 off_t tmp_pos = curpos;
2007 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
2008 type, obj_offset);
2009 if (!base_offset) {
2010 type = OBJ_BAD;
2011 goto out;
2012 }
2013 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2014 if (*oi->sizep == 0) {
2015 type = OBJ_BAD;
2016 goto out;
2017 }
2018 } else {
2019 *oi->sizep = size;
2020 }
2021 }
2022
2023 if (oi->disk_sizep) {
2024 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2025 *oi->disk_sizep = revidx[1].offset - obj_offset;
2026 }
2027
2028 if (oi->typep) {
2029 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
2030 if (*oi->typep < 0) {
2031 type = OBJ_BAD;
2032 goto out;
2033 }
2034 }
2035
2036 if (oi->delta_base_sha1) {
2037 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2038 const unsigned char *base;
2039
2040 base = get_delta_base_sha1(p, &w_curs, curpos,
2041 type, obj_offset);
2042 if (!base) {
2043 type = OBJ_BAD;
2044 goto out;
2045 }
2046
2047 hashcpy(oi->delta_base_sha1, base);
2048 } else
2049 hashclr(oi->delta_base_sha1);
2050 }
2051
2052 out:
2053 unuse_pack(&w_curs);
2054 return type;
2055 }
2056
2057 static void *unpack_compressed_entry(struct packed_git *p,
2058 struct pack_window **w_curs,
2059 off_t curpos,
2060 unsigned long size)
2061 {
2062 int st;
2063 git_zstream stream;
2064 unsigned char *buffer, *in;
2065
2066 buffer = xmallocz_gently(size);
2067 if (!buffer)
2068 return NULL;
2069 memset(&stream, 0, sizeof(stream));
2070 stream.next_out = buffer;
2071 stream.avail_out = size + 1;
2072
2073 git_inflate_init(&stream);
2074 do {
2075 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2076 stream.next_in = in;
2077 st = git_inflate(&stream, Z_FINISH);
2078 if (!stream.avail_out)
2079 break; /* the payload is larger than it should be */
2080 curpos += stream.next_in - in;
2081 } while (st == Z_OK || st == Z_BUF_ERROR);
2082 git_inflate_end(&stream);
2083 if ((st != Z_STREAM_END) || stream.total_out != size) {
2084 free(buffer);
2085 return NULL;
2086 }
2087
2088 return buffer;
2089 }
2090
2091 #define MAX_DELTA_CACHE (256)
2092
2093 static size_t delta_base_cached;
2094
2095 static struct delta_base_cache_lru_list {
2096 struct delta_base_cache_lru_list *prev;
2097 struct delta_base_cache_lru_list *next;
2098 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
2099
2100 static struct delta_base_cache_entry {
2101 struct delta_base_cache_lru_list lru;
2102 void *data;
2103 struct packed_git *p;
2104 off_t base_offset;
2105 unsigned long size;
2106 enum object_type type;
2107 } delta_base_cache[MAX_DELTA_CACHE];
2108
2109 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
2110 {
2111 unsigned long hash;
2112
2113 hash = (unsigned long)p + (unsigned long)base_offset;
2114 hash += (hash >> 8) + (hash >> 16);
2115 return hash % MAX_DELTA_CACHE;
2116 }
2117
2118 static struct delta_base_cache_entry *
2119 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2120 {
2121 unsigned long hash = pack_entry_hash(p, base_offset);
2122 return delta_base_cache + hash;
2123 }
2124
2125 static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
2126 struct packed_git *p, off_t base_offset)
2127 {
2128 return (ent->data && ent->p == p && ent->base_offset == base_offset);
2129 }
2130
2131 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2132 {
2133 struct delta_base_cache_entry *ent;
2134 ent = get_delta_base_cache_entry(p, base_offset);
2135 return eq_delta_base_cache_entry(ent, p, base_offset);
2136 }
2137
2138 static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2139 {
2140 ent->data = NULL;
2141 ent->lru.next->prev = ent->lru.prev;
2142 ent->lru.prev->next = ent->lru.next;
2143 delta_base_cached -= ent->size;
2144 }
2145
2146 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2147 unsigned long *base_size, enum object_type *type, int keep_cache)
2148 {
2149 struct delta_base_cache_entry *ent;
2150 void *ret;
2151
2152 ent = get_delta_base_cache_entry(p, base_offset);
2153
2154 if (!eq_delta_base_cache_entry(ent, p, base_offset))
2155 return unpack_entry(p, base_offset, type, base_size);
2156
2157 ret = ent->data;
2158
2159 if (!keep_cache)
2160 clear_delta_base_cache_entry(ent);
2161 else
2162 ret = xmemdupz(ent->data, ent->size);
2163 *type = ent->type;
2164 *base_size = ent->size;
2165 return ret;
2166 }
2167
2168 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2169 {
2170 if (ent->data) {
2171 free(ent->data);
2172 ent->data = NULL;
2173 ent->lru.next->prev = ent->lru.prev;
2174 ent->lru.prev->next = ent->lru.next;
2175 delta_base_cached -= ent->size;
2176 }
2177 }
2178
2179 void clear_delta_base_cache(void)
2180 {
2181 unsigned long p;
2182 for (p = 0; p < MAX_DELTA_CACHE; p++)
2183 release_delta_base_cache(&delta_base_cache[p]);
2184 }
2185
2186 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2187 void *base, unsigned long base_size, enum object_type type)
2188 {
2189 unsigned long hash = pack_entry_hash(p, base_offset);
2190 struct delta_base_cache_entry *ent = delta_base_cache + hash;
2191 struct delta_base_cache_lru_list *lru;
2192
2193 release_delta_base_cache(ent);
2194 delta_base_cached += base_size;
2195
2196 for (lru = delta_base_cache_lru.next;
2197 delta_base_cached > delta_base_cache_limit
2198 && lru != &delta_base_cache_lru;
2199 lru = lru->next) {
2200 struct delta_base_cache_entry *f = (void *)lru;
2201 if (f->type == OBJ_BLOB)
2202 release_delta_base_cache(f);
2203 }
2204 for (lru = delta_base_cache_lru.next;
2205 delta_base_cached > delta_base_cache_limit
2206 && lru != &delta_base_cache_lru;
2207 lru = lru->next) {
2208 struct delta_base_cache_entry *f = (void *)lru;
2209 release_delta_base_cache(f);
2210 }
2211
2212 ent->p = p;
2213 ent->base_offset = base_offset;
2214 ent->type = type;
2215 ent->data = base;
2216 ent->size = base_size;
2217 ent->lru.next = &delta_base_cache_lru;
2218 ent->lru.prev = delta_base_cache_lru.prev;
2219 delta_base_cache_lru.prev->next = &ent->lru;
2220 delta_base_cache_lru.prev = &ent->lru;
2221 }
2222
2223 static void *read_object(const unsigned char *sha1, enum object_type *type,
2224 unsigned long *size);
2225
2226 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2227 {
2228 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2229 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2230 p->pack_name, (uintmax_t)obj_offset);
2231 }
2232
2233 int do_check_packed_object_crc;
2234
2235 #define UNPACK_ENTRY_STACK_PREALLOC 64
2236 struct unpack_entry_stack_ent {
2237 off_t obj_offset;
2238 off_t curpos;
2239 unsigned long size;
2240 };
2241
2242 void *unpack_entry(struct packed_git *p, off_t obj_offset,
2243 enum object_type *final_type, unsigned long *final_size)
2244 {
2245 struct pack_window *w_curs = NULL;
2246 off_t curpos = obj_offset;
2247 void *data = NULL;
2248 unsigned long size;
2249 enum object_type type;
2250 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2251 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2252 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2253 int base_from_cache = 0;
2254
2255 write_pack_access_log(p, obj_offset);
2256
2257 /* PHASE 1: drill down to the innermost base object */
2258 for (;;) {
2259 off_t base_offset;
2260 int i;
2261 struct delta_base_cache_entry *ent;
2262
2263 ent = get_delta_base_cache_entry(p, curpos);
2264 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2265 type = ent->type;
2266 data = ent->data;
2267 size = ent->size;
2268 clear_delta_base_cache_entry(ent);
2269 base_from_cache = 1;
2270 break;
2271 }
2272
2273 if (do_check_packed_object_crc && p->index_version > 1) {
2274 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2275 unsigned long len = revidx[1].offset - obj_offset;
2276 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2277 const unsigned char *sha1 =
2278 nth_packed_object_sha1(p, revidx->nr);
2279 error("bad packed object CRC for %s",
2280 sha1_to_hex(sha1));
2281 mark_bad_packed_object(p, sha1);
2282 unuse_pack(&w_curs);
2283 return NULL;
2284 }
2285 }
2286
2287 type = unpack_object_header(p, &w_curs, &curpos, &size);
2288 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2289 break;
2290
2291 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2292 if (!base_offset) {
2293 error("failed to validate delta base reference "
2294 "at offset %"PRIuMAX" from %s",
2295 (uintmax_t)curpos, p->pack_name);
2296 /* bail to phase 2, in hopes of recovery */
2297 data = NULL;
2298 break;
2299 }
2300
2301 /* push object, proceed to base */
2302 if (delta_stack_nr >= delta_stack_alloc
2303 && delta_stack == small_delta_stack) {
2304 delta_stack_alloc = alloc_nr(delta_stack_nr);
2305 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2306 memcpy(delta_stack, small_delta_stack,
2307 sizeof(*delta_stack)*delta_stack_nr);
2308 } else {
2309 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2310 }
2311 i = delta_stack_nr++;
2312 delta_stack[i].obj_offset = obj_offset;
2313 delta_stack[i].curpos = curpos;
2314 delta_stack[i].size = size;
2315
2316 curpos = obj_offset = base_offset;
2317 }
2318
2319 /* PHASE 2: handle the base */
2320 switch (type) {
2321 case OBJ_OFS_DELTA:
2322 case OBJ_REF_DELTA:
2323 if (data)
2324 die("BUG in unpack_entry: left loop at a valid delta");
2325 break;
2326 case OBJ_COMMIT:
2327 case OBJ_TREE:
2328 case OBJ_BLOB:
2329 case OBJ_TAG:
2330 if (!base_from_cache)
2331 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2332 break;
2333 default:
2334 data = NULL;
2335 error("unknown object type %i at offset %"PRIuMAX" in %s",
2336 type, (uintmax_t)obj_offset, p->pack_name);
2337 }
2338
2339 /* PHASE 3: apply deltas in order */
2340
2341 /* invariants:
2342 * 'data' holds the base data, or NULL if there was corruption
2343 */
2344 while (delta_stack_nr) {
2345 void *delta_data;
2346 void *base = data;
2347 unsigned long delta_size, base_size = size;
2348 int i;
2349
2350 data = NULL;
2351
2352 if (base)
2353 add_delta_base_cache(p, obj_offset, base, base_size, type);
2354
2355 if (!base) {
2356 /*
2357 * We're probably in deep shit, but let's try to fetch
2358 * the required base anyway from another pack or loose.
2359 * This is costly but should happen only in the presence
2360 * of a corrupted pack, and is better than failing outright.
2361 */
2362 struct revindex_entry *revidx;
2363 const unsigned char *base_sha1;
2364 revidx = find_pack_revindex(p, obj_offset);
2365 if (revidx) {
2366 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2367 error("failed to read delta base object %s"
2368 " at offset %"PRIuMAX" from %s",
2369 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2370 p->pack_name);
2371 mark_bad_packed_object(p, base_sha1);
2372 base = read_object(base_sha1, &type, &base_size);
2373 }
2374 }
2375
2376 i = --delta_stack_nr;
2377 obj_offset = delta_stack[i].obj_offset;
2378 curpos = delta_stack[i].curpos;
2379 delta_size = delta_stack[i].size;
2380
2381 if (!base)
2382 continue;
2383
2384 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2385
2386 if (!delta_data) {
2387 error("failed to unpack compressed delta "
2388 "at offset %"PRIuMAX" from %s",
2389 (uintmax_t)curpos, p->pack_name);
2390 data = NULL;
2391 continue;
2392 }
2393
2394 data = patch_delta(base, base_size,
2395 delta_data, delta_size,
2396 &size);
2397
2398 /*
2399 * We could not apply the delta; warn the user, but keep going.
2400 * Our failure will be noticed either in the next iteration of
2401 * the loop, or if this is the final delta, in the caller when
2402 * we return NULL. Those code paths will take care of making
2403 * a more explicit warning and retrying with another copy of
2404 * the object.
2405 */
2406 if (!data)
2407 error("failed to apply delta");
2408
2409 free(delta_data);
2410 }
2411
2412 *final_type = type;
2413 *final_size = size;
2414
2415 unuse_pack(&w_curs);
2416
2417 if (delta_stack != small_delta_stack)
2418 free(delta_stack);
2419
2420 return data;
2421 }
2422
2423 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2424 uint32_t n)
2425 {
2426 const unsigned char *index = p->index_data;
2427 if (!index) {
2428 if (open_pack_index(p))
2429 return NULL;
2430 index = p->index_data;
2431 }
2432 if (n >= p->num_objects)
2433 return NULL;
2434 index += 4 * 256;
2435 if (p->index_version == 1) {
2436 return index + 24 * n + 4;
2437 } else {
2438 index += 8;
2439 return index + 20 * n;
2440 }
2441 }
2442
2443 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2444 {
2445 const unsigned char *index = p->index_data;
2446 index += 4 * 256;
2447 if (p->index_version == 1) {
2448 return ntohl(*((uint32_t *)(index + 24 * n)));
2449 } else {
2450 uint32_t off;
2451 index += 8 + p->num_objects * (20 + 4);
2452 off = ntohl(*((uint32_t *)(index + 4 * n)));
2453 if (!(off & 0x80000000))
2454 return off;
2455 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2456 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2457 ntohl(*((uint32_t *)(index + 4)));
2458 }
2459 }
2460
2461 off_t find_pack_entry_one(const unsigned char *sha1,
2462 struct packed_git *p)
2463 {
2464 const uint32_t *level1_ofs = p->index_data;
2465 const unsigned char *index = p->index_data;
2466 unsigned hi, lo, stride;
2467 static int use_lookup = -1;
2468 static int debug_lookup = -1;
2469
2470 if (debug_lookup < 0)
2471 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2472
2473 if (!index) {
2474 if (open_pack_index(p))
2475 return 0;
2476 level1_ofs = p->index_data;
2477 index = p->index_data;
2478 }
2479 if (p->index_version > 1) {
2480 level1_ofs += 2;
2481 index += 8;
2482 }
2483 index += 4 * 256;
2484 hi = ntohl(level1_ofs[*sha1]);
2485 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2486 if (p->index_version > 1) {
2487 stride = 20;
2488 } else {
2489 stride = 24;
2490 index += 4;
2491 }
2492
2493 if (debug_lookup)
2494 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2495 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2496
2497 if (use_lookup < 0)
2498 use_lookup = !!getenv("GIT_USE_LOOKUP");
2499 if (use_lookup) {
2500 int pos = sha1_entry_pos(index, stride, 0,
2501 lo, hi, p->num_objects, sha1);
2502 if (pos < 0)
2503 return 0;
2504 return nth_packed_object_offset(p, pos);
2505 }
2506
2507 do {
2508 unsigned mi = (lo + hi) / 2;
2509 int cmp = hashcmp(index + mi * stride, sha1);
2510
2511 if (debug_lookup)
2512 printf("lo %u hi %u rg %u mi %u\n",
2513 lo, hi, hi - lo, mi);
2514 if (!cmp)
2515 return nth_packed_object_offset(p, mi);
2516 if (cmp > 0)
2517 hi = mi;
2518 else
2519 lo = mi+1;
2520 } while (lo < hi);
2521 return 0;
2522 }
2523
2524 int is_pack_valid(struct packed_git *p)
2525 {
2526 /* An already open pack is known to be valid. */
2527 if (p->pack_fd != -1)
2528 return 1;
2529
2530 /* If the pack has one window completely covering the
2531 * file size, the pack is known to be valid even if
2532 * the descriptor is not currently open.
2533 */
2534 if (p->windows) {
2535 struct pack_window *w = p->windows;
2536
2537 if (!w->offset && w->len == p->pack_size)
2538 return 1;
2539 }
2540
2541 /* Force the pack to open to prove its valid. */
2542 return !open_packed_git(p);
2543 }
2544
2545 static int fill_pack_entry(const unsigned char *sha1,
2546 struct pack_entry *e,
2547 struct packed_git *p)
2548 {
2549 off_t offset;
2550
2551 if (p->num_bad_objects) {
2552 unsigned i;
2553 for (i = 0; i < p->num_bad_objects; i++)
2554 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2555 return 0;
2556 }
2557
2558 offset = find_pack_entry_one(sha1, p);
2559 if (!offset)
2560 return 0;
2561
2562 /*
2563 * We are about to tell the caller where they can locate the
2564 * requested object. We better make sure the packfile is
2565 * still here and can be accessed before supplying that
2566 * answer, as it may have been deleted since the index was
2567 * loaded!
2568 */
2569 if (!is_pack_valid(p))
2570 return 0;
2571 e->offset = offset;
2572 e->p = p;
2573 hashcpy(e->sha1, sha1);
2574 return 1;
2575 }
2576
2577 /*
2578 * Iff a pack file contains the object named by sha1, return true and
2579 * store its location to e.
2580 */
2581 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2582 {
2583 struct packed_git *p;
2584
2585 prepare_packed_git();
2586 if (!packed_git)
2587 return 0;
2588
2589 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2590 return 1;
2591
2592 for (p = packed_git; p; p = p->next) {
2593 if (p == last_found_pack)
2594 continue; /* we already checked this one */
2595
2596 if (fill_pack_entry(sha1, e, p)) {
2597 last_found_pack = p;
2598 return 1;
2599 }
2600 }
2601 return 0;
2602 }
2603
2604 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2605 struct packed_git *packs)
2606 {
2607 struct packed_git *p;
2608
2609 for (p = packs; p; p = p->next) {
2610 if (find_pack_entry_one(sha1, p))
2611 return p;
2612 }
2613 return NULL;
2614
2615 }
2616
2617 static int sha1_loose_object_info(const unsigned char *sha1,
2618 struct object_info *oi,
2619 int flags)
2620 {
2621 int status = 0;
2622 unsigned long mapsize;
2623 void *map;
2624 git_zstream stream;
2625 char hdr[32];
2626 struct strbuf hdrbuf = STRBUF_INIT;
2627
2628 if (oi->delta_base_sha1)
2629 hashclr(oi->delta_base_sha1);
2630
2631 /*
2632 * If we don't care about type or size, then we don't
2633 * need to look inside the object at all. Note that we
2634 * do not optimize out the stat call, even if the
2635 * caller doesn't care about the disk-size, since our
2636 * return value implicitly indicates whether the
2637 * object even exists.
2638 */
2639 if (!oi->typep && !oi->typename && !oi->sizep) {
2640 struct stat st;
2641 if (stat_sha1_file(sha1, &st) < 0)
2642 return -1;
2643 if (oi->disk_sizep)
2644 *oi->disk_sizep = st.st_size;
2645 return 0;
2646 }
2647
2648 map = map_sha1_file(sha1, &mapsize);
2649 if (!map)
2650 return -1;
2651 if (oi->disk_sizep)
2652 *oi->disk_sizep = mapsize;
2653 if ((flags & LOOKUP_UNKNOWN_OBJECT)) {
2654 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2655 status = error("unable to unpack %s header with --allow-unknown-type",
2656 sha1_to_hex(sha1));
2657 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2658 status = error("unable to unpack %s header",
2659 sha1_to_hex(sha1));
2660 if (status < 0)
2661 ; /* Do nothing */
2662 else if (hdrbuf.len) {
2663 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2664 status = error("unable to parse %s header with --allow-unknown-type",
2665 sha1_to_hex(sha1));
2666 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
2667 status = error("unable to parse %s header", sha1_to_hex(sha1));
2668 git_inflate_end(&stream);
2669 munmap(map, mapsize);
2670 if (status && oi->typep)
2671 *oi->typep = status;
2672 strbuf_release(&hdrbuf);
2673 return 0;
2674 }
2675
2676 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2677 {
2678 struct cached_object *co;
2679 struct pack_entry e;
2680 int rtype;
2681 enum object_type real_type;
2682 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2683
2684 co = find_cached_object(real);
2685 if (co) {
2686 if (oi->typep)
2687 *(oi->typep) = co->type;
2688 if (oi->sizep)
2689 *(oi->sizep) = co->size;
2690 if (oi->disk_sizep)
2691 *(oi->disk_sizep) = 0;
2692 if (oi->delta_base_sha1)
2693 hashclr(oi->delta_base_sha1);
2694 if (oi->typename)
2695 strbuf_addstr(oi->typename, typename(co->type));
2696 oi->whence = OI_CACHED;
2697 return 0;
2698 }
2699
2700 if (!find_pack_entry(real, &e)) {
2701 /* Most likely it's a loose object. */
2702 if (!sha1_loose_object_info(real, oi, flags)) {
2703 oi->whence = OI_LOOSE;
2704 return 0;
2705 }
2706
2707 /* Not a loose object; someone else may have just packed it. */
2708 reprepare_packed_git();
2709 if (!find_pack_entry(real, &e))
2710 return -1;
2711 }
2712
2713 /*
2714 * packed_object_info() does not follow the delta chain to
2715 * find out the real type, unless it is given oi->typep.
2716 */
2717 if (oi->typename && !oi->typep)
2718 oi->typep = &real_type;
2719
2720 rtype = packed_object_info(e.p, e.offset, oi);
2721 if (rtype < 0) {
2722 mark_bad_packed_object(e.p, real);
2723 if (oi->typep == &real_type)
2724 oi->typep = NULL;
2725 return sha1_object_info_extended(real, oi, 0);
2726 } else if (in_delta_base_cache(e.p, e.offset)) {
2727 oi->whence = OI_DBCACHED;
2728 } else {
2729 oi->whence = OI_PACKED;
2730 oi->u.packed.offset = e.offset;
2731 oi->u.packed.pack = e.p;
2732 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2733 rtype == OBJ_OFS_DELTA);
2734 }
2735 if (oi->typename)
2736 strbuf_addstr(oi->typename, typename(*oi->typep));
2737 if (oi->typep == &real_type)
2738 oi->typep = NULL;
2739
2740 return 0;
2741 }
2742
2743 /* returns enum object_type or negative */
2744 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2745 {
2746 enum object_type type;
2747 struct object_info oi = {NULL};
2748
2749 oi.typep = &type;
2750 oi.sizep = sizep;
2751 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2752 return -1;
2753 return type;
2754 }
2755
2756 static void *read_packed_sha1(const unsigned char *sha1,
2757 enum object_type *type, unsigned long *size)
2758 {
2759 struct pack_entry e;
2760 void *data;
2761
2762 if (!find_pack_entry(sha1, &e))
2763 return NULL;
2764 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2765 if (!data) {
2766 /*
2767 * We're probably in deep shit, but let's try to fetch
2768 * the required object anyway from another pack or loose.
2769 * This should happen only in the presence of a corrupted
2770 * pack, and is better than failing outright.
2771 */
2772 error("failed to read object %s at offset %"PRIuMAX" from %s",
2773 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2774 mark_bad_packed_object(e.p, sha1);
2775 data = read_object(sha1, type, size);
2776 }
2777 return data;
2778 }
2779
2780 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2781 unsigned char *sha1)
2782 {
2783 struct cached_object *co;
2784
2785 hash_sha1_file(buf, len, typename(type), sha1);
2786 if (has_sha1_file(sha1) || find_cached_object(sha1))
2787 return 0;
2788 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2789 co = &cached_objects[cached_object_nr++];
2790 co->size = len;
2791 co->type = type;
2792 co->buf = xmalloc(len);
2793 memcpy(co->buf, buf, len);
2794 hashcpy(co->sha1, sha1);
2795 return 0;
2796 }
2797
2798 static void *read_object(const unsigned char *sha1, enum object_type *type,
2799 unsigned long *size)
2800 {
2801 unsigned long mapsize;
2802 void *map, *buf;
2803 struct cached_object *co;
2804
2805 co = find_cached_object(sha1);
2806 if (co) {
2807 *type = co->type;
2808 *size = co->size;
2809 return xmemdupz(co->buf, co->size);
2810 }
2811
2812 buf = read_packed_sha1(sha1, type, size);
2813 if (buf)
2814 return buf;
2815 map = map_sha1_file(sha1, &mapsize);
2816 if (map) {
2817 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2818 munmap(map, mapsize);
2819 return buf;
2820 }
2821 reprepare_packed_git();
2822 return read_packed_sha1(sha1, type, size);
2823 }
2824
2825 /*
2826 * This function dies on corrupt objects; the callers who want to
2827 * deal with them should arrange to call read_object() and give error
2828 * messages themselves.
2829 */
2830 void *read_sha1_file_extended(const unsigned char *sha1,
2831 enum object_type *type,
2832 unsigned long *size,
2833 unsigned flag)
2834 {
2835 void *data;
2836 const struct packed_git *p;
2837 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
2838
2839 errno = 0;
2840 data = read_object(repl, type, size);
2841 if (data)
2842 return data;
2843
2844 if (errno && errno != ENOENT)
2845 die_errno("failed to read object %s", sha1_to_hex(sha1));
2846
2847 /* die if we replaced an object with one that does not exist */
2848 if (repl != sha1)
2849 die("replacement %s not found for %s",
2850 sha1_to_hex(repl), sha1_to_hex(sha1));
2851
2852 if (has_loose_object(repl)) {
2853 const char *path = sha1_file_name(sha1);
2854
2855 die("loose object %s (stored in %s) is corrupt",
2856 sha1_to_hex(repl), path);
2857 }
2858
2859 if ((p = has_packed_and_bad(repl)) != NULL)
2860 die("packed object %s (stored in %s) is corrupt",
2861 sha1_to_hex(repl), p->pack_name);
2862
2863 return NULL;
2864 }
2865
2866 void *read_object_with_reference(const unsigned char *sha1,
2867 const char *required_type_name,
2868 unsigned long *size,
2869 unsigned char *actual_sha1_return)
2870 {
2871 enum object_type type, required_type;
2872 void *buffer;
2873 unsigned long isize;
2874 unsigned char actual_sha1[20];
2875
2876 required_type = type_from_string(required_type_name);
2877 hashcpy(actual_sha1, sha1);
2878 while (1) {
2879 int ref_length = -1;
2880 const char *ref_type = NULL;
2881
2882 buffer = read_sha1_file(actual_sha1, &type, &isize);
2883 if (!buffer)
2884 return NULL;
2885 if (type == required_type) {
2886 *size = isize;
2887 if (actual_sha1_return)
2888 hashcpy(actual_sha1_return, actual_sha1);
2889 return buffer;
2890 }
2891 /* Handle references */
2892 else if (type == OBJ_COMMIT)
2893 ref_type = "tree ";
2894 else if (type == OBJ_TAG)
2895 ref_type = "object ";
2896 else {
2897 free(buffer);
2898 return NULL;
2899 }
2900 ref_length = strlen(ref_type);
2901
2902 if (ref_length + 40 > isize ||
2903 memcmp(buffer, ref_type, ref_length) ||
2904 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2905 free(buffer);
2906 return NULL;
2907 }
2908 free(buffer);
2909 /* Now we have the ID of the referred-to object in
2910 * actual_sha1. Check again. */
2911 }
2912 }
2913
2914 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2915 const char *type, unsigned char *sha1,
2916 char *hdr, int *hdrlen)
2917 {
2918 git_SHA_CTX c;
2919
2920 /* Generate the header */
2921 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
2922
2923 /* Sha1.. */
2924 git_SHA1_Init(&c);
2925 git_SHA1_Update(&c, hdr, *hdrlen);
2926 git_SHA1_Update(&c, buf, len);
2927 git_SHA1_Final(sha1, &c);
2928 }
2929
2930 /*
2931 * Move the just written object into its final resting place.
2932 */
2933 int finalize_object_file(const char *tmpfile, const char *filename)
2934 {
2935 int ret = 0;
2936
2937 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2938 goto try_rename;
2939 else if (link(tmpfile, filename))
2940 ret = errno;
2941
2942 /*
2943 * Coda hack - coda doesn't like cross-directory links,
2944 * so we fall back to a rename, which will mean that it
2945 * won't be able to check collisions, but that's not a
2946 * big deal.
2947 *
2948 * The same holds for FAT formatted media.
2949 *
2950 * When this succeeds, we just return. We have nothing
2951 * left to unlink.
2952 */
2953 if (ret && ret != EEXIST) {
2954 try_rename:
2955 if (!rename(tmpfile, filename))
2956 goto out;
2957 ret = errno;
2958 }
2959 unlink_or_warn(tmpfile);
2960 if (ret) {
2961 if (ret != EEXIST) {
2962 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2963 }
2964 /* FIXME!!! Collision check here ? */
2965 }
2966
2967 out:
2968 if (adjust_shared_perm(filename))
2969 return error("unable to set permission to '%s'", filename);
2970 return 0;
2971 }
2972
2973 static int write_buffer(int fd, const void *buf, size_t len)
2974 {
2975 if (write_in_full(fd, buf, len) < 0)
2976 return error("file write error (%s)", strerror(errno));
2977 return 0;
2978 }
2979
2980 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2981 unsigned char *sha1)
2982 {
2983 char hdr[32];
2984 int hdrlen = sizeof(hdr);
2985 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2986 return 0;
2987 }
2988
2989 /* Finalize a file on disk, and close it. */
2990 static void close_sha1_file(int fd)
2991 {
2992 if (fsync_object_files)
2993 fsync_or_die(fd, "sha1 file");
2994 if (close(fd) != 0)
2995 die_errno("error when closing sha1 file");
2996 }
2997
2998 /* Size of directory component, including the ending '/' */
2999 static inline int directory_size(const char *filename)
3000 {
3001 const char *s = strrchr(filename, '/');
3002 if (!s)
3003 return 0;
3004 return s - filename + 1;
3005 }
3006
3007 /*
3008 * This creates a temporary file in the same directory as the final
3009 * 'filename'
3010 *
3011 * We want to avoid cross-directory filename renames, because those
3012 * can have problems on various filesystems (FAT, NFS, Coda).
3013 */
3014 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
3015 {
3016 int fd, dirlen = directory_size(filename);
3017
3018 if (dirlen + 20 > bufsiz) {
3019 errno = ENAMETOOLONG;
3020 return -1;
3021 }
3022 memcpy(buffer, filename, dirlen);
3023 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
3024 fd = git_mkstemp_mode(buffer, 0444);
3025 if (fd < 0 && dirlen && errno == ENOENT) {
3026 /* Make sure the directory exists */
3027 memcpy(buffer, filename, dirlen);
3028 buffer[dirlen-1] = 0;
3029 if (mkdir(buffer, 0777) && errno != EEXIST)
3030 return -1;
3031 if (adjust_shared_perm(buffer))
3032 return -1;
3033
3034 /* Try again */
3035 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
3036 fd = git_mkstemp_mode(buffer, 0444);
3037 }
3038 return fd;
3039 }
3040
3041 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
3042 const void *buf, unsigned long len, time_t mtime)
3043 {
3044 int fd, ret;
3045 unsigned char compressed[4096];
3046 git_zstream stream;
3047 git_SHA_CTX c;
3048 unsigned char parano_sha1[20];
3049 static char tmp_file[PATH_MAX];
3050 const char *filename = sha1_file_name(sha1);
3051
3052 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
3053 if (fd < 0) {
3054 if (errno == EACCES)
3055 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3056 else
3057 return error("unable to create temporary file: %s", strerror(errno));
3058 }
3059
3060 /* Set it up */
3061 git_deflate_init(&stream, zlib_compression_level);
3062 stream.next_out = compressed;
3063 stream.avail_out = sizeof(compressed);
3064 git_SHA1_Init(&c);
3065
3066 /* First header.. */
3067 stream.next_in = (unsigned char *)hdr;
3068 stream.avail_in = hdrlen;
3069 while (git_deflate(&stream, 0) == Z_OK)
3070 ; /* nothing */
3071 git_SHA1_Update(&c, hdr, hdrlen);
3072
3073 /* Then the data itself.. */
3074 stream.next_in = (void *)buf;
3075 stream.avail_in = len;
3076 do {
3077 unsigned char *in0 = stream.next_in;
3078 ret = git_deflate(&stream, Z_FINISH);
3079 git_SHA1_Update(&c, in0, stream.next_in - in0);
3080 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3081 die("unable to write sha1 file");
3082 stream.next_out = compressed;
3083 stream.avail_out = sizeof(compressed);
3084 } while (ret == Z_OK);
3085
3086 if (ret != Z_STREAM_END)
3087 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
3088 ret = git_deflate_end_gently(&stream);
3089 if (ret != Z_OK)
3090 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
3091 git_SHA1_Final(parano_sha1, &c);
3092 if (hashcmp(sha1, parano_sha1) != 0)
3093 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
3094
3095 close_sha1_file(fd);
3096
3097 if (mtime) {
3098 struct utimbuf utb;
3099 utb.actime = mtime;
3100 utb.modtime = mtime;
3101 if (utime(tmp_file, &utb) < 0)
3102 warning("failed utime() on %s: %s",
3103 tmp_file, strerror(errno));
3104 }
3105
3106 return finalize_object_file(tmp_file, filename);
3107 }
3108
3109 static int freshen_loose_object(const unsigned char *sha1)
3110 {
3111 return check_and_freshen(sha1, 1);
3112 }
3113
3114 static int freshen_packed_object(const unsigned char *sha1)
3115 {
3116 struct pack_entry e;
3117 if (!find_pack_entry(sha1, &e))
3118 return 0;
3119 if (e.p->freshened)
3120 return 1;
3121 if (!freshen_file(e.p->pack_name))
3122 return 0;
3123 e.p->freshened = 1;
3124 return 1;
3125 }
3126
3127 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3128 {
3129 char hdr[32];
3130 int hdrlen = sizeof(hdr);
3131
3132 /* Normally if we have it in the pack then we do not bother writing
3133 * it out into .git/objects/??/?{38} file.
3134 */
3135 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3136 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3137 return 0;
3138 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3139 }
3140
3141 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3142 unsigned char *sha1, unsigned flags)
3143 {
3144 char *header;
3145 int hdrlen, status = 0;
3146
3147 /* type string, SP, %lu of the length plus NUL must fit this */
3148 hdrlen = strlen(type) + 32;
3149 header = xmalloc(hdrlen);
3150 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3151
3152 if (!(flags & HASH_WRITE_OBJECT))
3153 goto cleanup;
3154 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3155 goto cleanup;
3156 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3157
3158 cleanup:
3159 free(header);
3160 return status;
3161 }
3162
3163 int force_object_loose(const unsigned char *sha1, time_t mtime)
3164 {
3165 void *buf;
3166 unsigned long len;
3167 enum object_type type;
3168 char hdr[32];
3169 int hdrlen;
3170 int ret;
3171
3172 if (has_loose_object(sha1))
3173 return 0;
3174 buf = read_packed_sha1(sha1, &type, &len);
3175 if (!buf)
3176 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3177 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
3178 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3179 free(buf);
3180
3181 return ret;
3182 }
3183
3184 int has_pack_index(const unsigned char *sha1)
3185 {
3186 struct stat st;
3187 if (stat(sha1_pack_index_name(sha1), &st))
3188 return 0;
3189 return 1;
3190 }
3191
3192 int has_sha1_pack(const unsigned char *sha1)
3193 {
3194 struct pack_entry e;
3195 return find_pack_entry(sha1, &e);
3196 }
3197
3198 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3199 {
3200 struct pack_entry e;
3201
3202 if (find_pack_entry(sha1, &e))
3203 return 1;
3204 if (has_loose_object(sha1))
3205 return 1;
3206 if (flags & HAS_SHA1_QUICK)
3207 return 0;
3208 reprepare_packed_git();
3209 return find_pack_entry(sha1, &e);
3210 }
3211
3212 static void check_tree(const void *buf, size_t size)
3213 {
3214 struct tree_desc desc;
3215 struct name_entry entry;
3216
3217 init_tree_desc(&desc, buf, size);
3218 while (tree_entry(&desc, &entry))
3219 /* do nothing
3220 * tree_entry() will die() on malformed entries */
3221 ;
3222 }
3223
3224 static void check_commit(const void *buf, size_t size)
3225 {
3226 struct commit c;
3227 memset(&c, 0, sizeof(c));
3228 if (parse_commit_buffer(&c, buf, size))
3229 die("corrupt commit");
3230 }
3231
3232 static void check_tag(const void *buf, size_t size)
3233 {
3234 struct tag t;
3235 memset(&t, 0, sizeof(t));
3236 if (parse_tag_buffer(&t, buf, size))
3237 die("corrupt tag");
3238 }
3239
3240 static int index_mem(unsigned char *sha1, void *buf, size_t size,
3241 enum object_type type,
3242 const char *path, unsigned flags)
3243 {
3244 int ret, re_allocated = 0;
3245 int write_object = flags & HASH_WRITE_OBJECT;
3246
3247 if (!type)
3248 type = OBJ_BLOB;
3249
3250 /*
3251 * Convert blobs to git internal format
3252 */
3253 if ((type == OBJ_BLOB) && path) {
3254 struct strbuf nbuf = STRBUF_INIT;
3255 if (convert_to_git(path, buf, size, &nbuf,
3256 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3257 buf = strbuf_detach(&nbuf, &size);
3258 re_allocated = 1;
3259 }
3260 }
3261 if (flags & HASH_FORMAT_CHECK) {
3262 if (type == OBJ_TREE)
3263 check_tree(buf, size);
3264 if (type == OBJ_COMMIT)
3265 check_commit(buf, size);
3266 if (type == OBJ_TAG)
3267 check_tag(buf, size);
3268 }
3269
3270 if (write_object)
3271 ret = write_sha1_file(buf, size, typename(type), sha1);
3272 else
3273 ret = hash_sha1_file(buf, size, typename(type), sha1);
3274 if (re_allocated)
3275 free(buf);
3276 return ret;
3277 }
3278
3279 static int index_stream_convert_blob(unsigned char *sha1, int fd,
3280 const char *path, unsigned flags)
3281 {
3282 int ret;
3283 const int write_object = flags & HASH_WRITE_OBJECT;
3284 struct strbuf sbuf = STRBUF_INIT;
3285
3286 assert(path);
3287 assert(would_convert_to_git_filter_fd(path));
3288
3289 convert_to_git_filter_fd(path, fd, &sbuf,
3290 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3291
3292 if (write_object)
3293 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3294 sha1);
3295 else
3296 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3297 sha1);
3298 strbuf_release(&sbuf);
3299 return ret;
3300 }
3301
3302 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3303 const char *path, unsigned flags)
3304 {
3305 struct strbuf sbuf = STRBUF_INIT;
3306 int ret;
3307
3308 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3309 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3310 else
3311 ret = -1;
3312 strbuf_release(&sbuf);
3313 return ret;
3314 }
3315
3316 #define SMALL_FILE_SIZE (32*1024)
3317
3318 static int index_core(unsigned char *sha1, int fd, size_t size,
3319 enum object_type type, const char *path,
3320 unsigned flags)
3321 {
3322 int ret;
3323
3324 if (!size) {
3325 ret = index_mem(sha1, "", size, type, path, flags);
3326 } else if (size <= SMALL_FILE_SIZE) {
3327 char *buf = xmalloc(size);
3328 if (size == read_in_full(fd, buf, size))
3329 ret = index_mem(sha1, buf, size, type, path, flags);
3330 else
3331 ret = error("short read %s", strerror(errno));
3332 free(buf);
3333 } else {
3334 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3335 ret = index_mem(sha1, buf, size, type, path, flags);
3336 munmap(buf, size);
3337 }
3338 return ret;
3339 }
3340
3341 /*
3342 * This creates one packfile per large blob unless bulk-checkin
3343 * machinery is "plugged".
3344 *
3345 * This also bypasses the usual "convert-to-git" dance, and that is on
3346 * purpose. We could write a streaming version of the converting
3347 * functions and insert that before feeding the data to fast-import
3348 * (or equivalent in-core API described above). However, that is
3349 * somewhat complicated, as we do not know the size of the filter
3350 * result, which we need to know beforehand when writing a git object.
3351 * Since the primary motivation for trying to stream from the working
3352 * tree file and to avoid mmaping it in core is to deal with large
3353 * binary blobs, they generally do not want to get any conversion, and
3354 * callers should avoid this code path when filters are requested.
3355 */
3356 static int index_stream(unsigned char *sha1, int fd, size_t size,
3357 enum object_type type, const char *path,
3358 unsigned flags)
3359 {
3360 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3361 }
3362
3363 int index_fd(unsigned char *sha1, int fd, struct stat *st,
3364 enum object_type type, const char *path, unsigned flags)
3365 {
3366 int ret;
3367
3368 /*
3369 * Call xsize_t() only when needed to avoid potentially unnecessary
3370 * die() for large files.
3371 */
3372 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3373 ret = index_stream_convert_blob(sha1, fd, path, flags);
3374 else if (!S_ISREG(st->st_mode))
3375 ret = index_pipe(sha1, fd, type, path, flags);
3376 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3377 (path && would_convert_to_git(path)))
3378 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3379 flags);
3380 else
3381 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3382 flags);
3383 close(fd);
3384 return ret;
3385 }
3386
3387 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3388 {
3389 int fd;
3390 struct strbuf sb = STRBUF_INIT;
3391
3392 switch (st->st_mode & S_IFMT) {
3393 case S_IFREG:
3394 fd = open(path, O_RDONLY);
3395 if (fd < 0)
3396 return error("open(\"%s\"): %s", path,
3397 strerror(errno));
3398 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3399 return error("%s: failed to insert into database",
3400 path);
3401 break;
3402 case S_IFLNK:
3403 if (strbuf_readlink(&sb, path, st->st_size)) {
3404 char *errstr = strerror(errno);
3405 return error("readlink(\"%s\"): %s", path,
3406 errstr);
3407 }
3408 if (!(flags & HASH_WRITE_OBJECT))
3409 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3410 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3411 return error("%s: failed to insert into database",
3412 path);
3413 strbuf_release(&sb);
3414 break;
3415 case S_IFDIR:
3416 return resolve_gitlink_ref(path, "HEAD", sha1);
3417 default:
3418 return error("%s: unsupported file type", path);
3419 }
3420 return 0;
3421 }
3422
3423 int read_pack_header(int fd, struct pack_header *header)
3424 {
3425 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3426 /* "eof before pack header was fully read" */
3427 return PH_ERROR_EOF;
3428
3429 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3430 /* "protocol error (pack signature mismatch detected)" */
3431 return PH_ERROR_PACK_SIGNATURE;
3432 if (!pack_version_ok(header->hdr_version))
3433 /* "protocol error (pack version unsupported)" */
3434 return PH_ERROR_PROTOCOL;
3435 return 0;
3436 }
3437
3438 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3439 {
3440 enum object_type type = sha1_object_info(sha1, NULL);
3441 if (type < 0)
3442 die("%s is not a valid object", sha1_to_hex(sha1));
3443 if (type != expect)
3444 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3445 typename(expect));
3446 }
3447
3448 static int for_each_file_in_obj_subdir(int subdir_nr,
3449 struct strbuf *path,
3450 each_loose_object_fn obj_cb,
3451 each_loose_cruft_fn cruft_cb,
3452 each_loose_subdir_fn subdir_cb,
3453 void *data)
3454 {
3455 size_t baselen = path->len;
3456 DIR *dir = opendir(path->buf);
3457 struct dirent *de;
3458 int r = 0;
3459
3460 if (!dir) {
3461 if (errno == ENOENT)
3462 return 0;
3463 return error("unable to open %s: %s", path->buf, strerror(errno));
3464 }
3465
3466 while ((de = readdir(dir))) {
3467 if (is_dot_or_dotdot(de->d_name))
3468 continue;
3469
3470 strbuf_setlen(path, baselen);
3471 strbuf_addf(path, "/%s", de->d_name);
3472
3473 if (strlen(de->d_name) == 38) {
3474 char hex[41];
3475 unsigned char sha1[20];
3476
3477 snprintf(hex, sizeof(hex), "%02x%s",
3478 subdir_nr, de->d_name);
3479 if (!get_sha1_hex(hex, sha1)) {
3480 if (obj_cb) {
3481 r = obj_cb(sha1, path->buf, data);
3482 if (r)
3483 break;
3484 }
3485 continue;
3486 }
3487 }
3488
3489 if (cruft_cb) {
3490 r = cruft_cb(de->d_name, path->buf, data);
3491 if (r)
3492 break;
3493 }
3494 }
3495 strbuf_setlen(path, baselen);
3496
3497 if (!r && subdir_cb)
3498 r = subdir_cb(subdir_nr, path->buf, data);
3499
3500 closedir(dir);
3501 return r;
3502 }
3503
3504 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3505 each_loose_object_fn obj_cb,
3506 each_loose_cruft_fn cruft_cb,
3507 each_loose_subdir_fn subdir_cb,
3508 void *data)
3509 {
3510 size_t baselen = path->len;
3511 int r = 0;
3512 int i;
3513
3514 for (i = 0; i < 256; i++) {
3515 strbuf_addf(path, "/%02x", i);
3516 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3517 subdir_cb, data);
3518 strbuf_setlen(path, baselen);
3519 if (r)
3520 break;
3521 }
3522
3523 return r;
3524 }
3525
3526 int for_each_loose_file_in_objdir(const char *path,
3527 each_loose_object_fn obj_cb,
3528 each_loose_cruft_fn cruft_cb,
3529 each_loose_subdir_fn subdir_cb,
3530 void *data)
3531 {
3532 struct strbuf buf = STRBUF_INIT;
3533 int r;
3534
3535 strbuf_addstr(&buf, path);
3536 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3537 subdir_cb, data);
3538 strbuf_release(&buf);
3539
3540 return r;
3541 }
3542
3543 struct loose_alt_odb_data {
3544 each_loose_object_fn *cb;
3545 void *data;
3546 };
3547
3548 static int loose_from_alt_odb(struct alternate_object_database *alt,
3549 void *vdata)
3550 {
3551 struct loose_alt_odb_data *data = vdata;
3552 struct strbuf buf = STRBUF_INIT;
3553 int r;
3554
3555 /* copy base not including trailing '/' */
3556 strbuf_add(&buf, alt->base, alt->name - alt->base - 1);
3557 r = for_each_loose_file_in_objdir_buf(&buf,
3558 data->cb, NULL, NULL,
3559 data->data);
3560 strbuf_release(&buf);
3561 return r;
3562 }
3563
3564 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3565 {
3566 struct loose_alt_odb_data alt;
3567 int r;
3568
3569 r = for_each_loose_file_in_objdir(get_object_directory(),
3570 cb, NULL, NULL, data);
3571 if (r)
3572 return r;
3573
3574 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3575 return 0;
3576
3577 alt.cb = cb;
3578 alt.data = data;
3579 return foreach_alt_odb(loose_from_alt_odb, &alt);
3580 }
3581
3582 static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3583 {
3584 uint32_t i;
3585 int r = 0;
3586
3587 for (i = 0; i < p->num_objects; i++) {
3588 const unsigned char *sha1 = nth_packed_object_sha1(p, i);
3589
3590 if (!sha1)
3591 return error("unable to get sha1 of object %u in %s",
3592 i, p->pack_name);
3593
3594 r = cb(sha1, p, i, data);
3595 if (r)
3596 break;
3597 }
3598 return r;
3599 }
3600
3601 int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3602 {
3603 struct packed_git *p;
3604 int r = 0;
3605 int pack_errors = 0;
3606
3607 prepare_packed_git();
3608 for (p = packed_git; p; p = p->next) {
3609 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3610 continue;
3611 if (open_pack_index(p)) {
3612 pack_errors = 1;
3613 continue;
3614 }
3615 r = for_each_object_in_pack(p, cb, data);
3616 if (r)
3617 break;
3618 }
3619 return r ? r : pack_errors;
3620 }