unpack_sha1_header(): detect malformed object header
[git/git.git] / sha1_file.c
1 /*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
9 #include "cache.h"
10 #include "string-list.h"
11 #include "lockfile.h"
12 #include "delta.h"
13 #include "pack.h"
14 #include "blob.h"
15 #include "commit.h"
16 #include "run-command.h"
17 #include "tag.h"
18 #include "tree.h"
19 #include "tree-walk.h"
20 #include "refs.h"
21 #include "pack-revindex.h"
22 #include "sha1-lookup.h"
23 #include "bulk-checkin.h"
24 #include "streaming.h"
25 #include "dir.h"
26
27 #ifndef O_NOATIME
28 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
29 #define O_NOATIME 01000000
30 #else
31 #define O_NOATIME 0
32 #endif
33 #endif
34
35 #define SZ_FMT PRIuMAX
36 static inline uintmax_t sz_fmt(size_t s) { return s; }
37
38 const unsigned char null_sha1[20];
39
40 /*
41 * This is meant to hold a *small* number of objects that you would
42 * want read_sha1_file() to be able to return, but yet you do not want
43 * to write them into the object store (e.g. a browse-only
44 * application).
45 */
46 static struct cached_object {
47 unsigned char sha1[20];
48 enum object_type type;
49 void *buf;
50 unsigned long size;
51 } *cached_objects;
52 static int cached_object_nr, cached_object_alloc;
53
54 static struct cached_object empty_tree = {
55 EMPTY_TREE_SHA1_BIN_LITERAL,
56 OBJ_TREE,
57 "",
58 0
59 };
60
61 /*
62 * A pointer to the last packed_git in which an object was found.
63 * When an object is sought, we look in this packfile first, because
64 * objects that are looked up at similar times are often in the same
65 * packfile as one another.
66 */
67 static struct packed_git *last_found_pack;
68
69 static struct cached_object *find_cached_object(const unsigned char *sha1)
70 {
71 int i;
72 struct cached_object *co = cached_objects;
73
74 for (i = 0; i < cached_object_nr; i++, co++) {
75 if (!hashcmp(co->sha1, sha1))
76 return co;
77 }
78 if (!hashcmp(sha1, empty_tree.sha1))
79 return &empty_tree;
80 return NULL;
81 }
82
83 int mkdir_in_gitdir(const char *path)
84 {
85 if (mkdir(path, 0777)) {
86 int saved_errno = errno;
87 struct stat st;
88 struct strbuf sb = STRBUF_INIT;
89
90 if (errno != EEXIST)
91 return -1;
92 /*
93 * Are we looking at a path in a symlinked worktree
94 * whose original repository does not yet have it?
95 * e.g. .git/rr-cache pointing at its original
96 * repository in which the user hasn't performed any
97 * conflict resolution yet?
98 */
99 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
100 strbuf_readlink(&sb, path, st.st_size) ||
101 !is_absolute_path(sb.buf) ||
102 mkdir(sb.buf, 0777)) {
103 strbuf_release(&sb);
104 errno = saved_errno;
105 return -1;
106 }
107 strbuf_release(&sb);
108 }
109 return adjust_shared_perm(path);
110 }
111
112 enum scld_error safe_create_leading_directories(char *path)
113 {
114 char *next_component = path + offset_1st_component(path);
115 enum scld_error ret = SCLD_OK;
116
117 while (ret == SCLD_OK && next_component) {
118 struct stat st;
119 char *slash = next_component, slash_character;
120
121 while (*slash && !is_dir_sep(*slash))
122 slash++;
123
124 if (!*slash)
125 break;
126
127 next_component = slash + 1;
128 while (is_dir_sep(*next_component))
129 next_component++;
130 if (!*next_component)
131 break;
132
133 slash_character = *slash;
134 *slash = '\0';
135 if (!stat(path, &st)) {
136 /* path exists */
137 if (!S_ISDIR(st.st_mode))
138 ret = SCLD_EXISTS;
139 } else if (mkdir(path, 0777)) {
140 if (errno == EEXIST &&
141 !stat(path, &st) && S_ISDIR(st.st_mode))
142 ; /* somebody created it since we checked */
143 else if (errno == ENOENT)
144 /*
145 * Either mkdir() failed because
146 * somebody just pruned the containing
147 * directory, or stat() failed because
148 * the file that was in our way was
149 * just removed. Either way, inform
150 * the caller that it might be worth
151 * trying again:
152 */
153 ret = SCLD_VANISHED;
154 else
155 ret = SCLD_FAILED;
156 } else if (adjust_shared_perm(path)) {
157 ret = SCLD_PERMS;
158 }
159 *slash = slash_character;
160 }
161 return ret;
162 }
163
164 enum scld_error safe_create_leading_directories_const(const char *path)
165 {
166 /* path points to cache entries, so xstrdup before messing with it */
167 char *buf = xstrdup(path);
168 enum scld_error result = safe_create_leading_directories(buf);
169 free(buf);
170 return result;
171 }
172
173 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
174 {
175 int i;
176 for (i = 0; i < 20; i++) {
177 static char hex[] = "0123456789abcdef";
178 unsigned int val = sha1[i];
179 char *pos = pathbuf + i*2 + (i > 0);
180 *pos++ = hex[val >> 4];
181 *pos = hex[val & 0xf];
182 }
183 }
184
185 const char *sha1_file_name(const unsigned char *sha1)
186 {
187 static char buf[PATH_MAX];
188 const char *objdir;
189 int len;
190
191 objdir = get_object_directory();
192 len = strlen(objdir);
193
194 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
195 if (len + 43 > PATH_MAX)
196 die("insanely long object directory %s", objdir);
197 memcpy(buf, objdir, len);
198 buf[len] = '/';
199 buf[len+3] = '/';
200 buf[len+42] = '\0';
201 fill_sha1_path(buf + len + 1, sha1);
202 return buf;
203 }
204
205 /*
206 * Return the name of the pack or index file with the specified sha1
207 * in its filename. *base and *name are scratch space that must be
208 * provided by the caller. which should be "pack" or "idx".
209 */
210 static char *sha1_get_pack_name(const unsigned char *sha1,
211 char **name, char **base, const char *which)
212 {
213 static const char hex[] = "0123456789abcdef";
214 char *buf;
215 int i;
216
217 if (!*base) {
218 const char *sha1_file_directory = get_object_directory();
219 int len = strlen(sha1_file_directory);
220 *base = xmalloc(len + 60);
221 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
222 sha1_file_directory, which);
223 *name = *base + len + 11;
224 }
225
226 buf = *name;
227
228 for (i = 0; i < 20; i++) {
229 unsigned int val = *sha1++;
230 *buf++ = hex[val >> 4];
231 *buf++ = hex[val & 0xf];
232 }
233
234 return *base;
235 }
236
237 char *sha1_pack_name(const unsigned char *sha1)
238 {
239 static char *name, *base;
240
241 return sha1_get_pack_name(sha1, &name, &base, "pack");
242 }
243
244 char *sha1_pack_index_name(const unsigned char *sha1)
245 {
246 static char *name, *base;
247
248 return sha1_get_pack_name(sha1, &name, &base, "idx");
249 }
250
251 struct alternate_object_database *alt_odb_list;
252 static struct alternate_object_database **alt_odb_tail;
253
254 /*
255 * Prepare alternate object database registry.
256 *
257 * The variable alt_odb_list points at the list of struct
258 * alternate_object_database. The elements on this list come from
259 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
260 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
261 * whose contents is similar to that environment variable but can be
262 * LF separated. Its base points at a statically allocated buffer that
263 * contains "/the/directory/corresponding/to/.git/objects/...", while
264 * its name points just after the slash at the end of ".git/objects/"
265 * in the example above, and has enough space to hold 40-byte hex
266 * SHA1, an extra slash for the first level indirection, and the
267 * terminating NUL.
268 */
269 static int link_alt_odb_entry(const char *entry, const char *relative_base,
270 int depth, const char *normalized_objdir)
271 {
272 struct alternate_object_database *ent;
273 struct alternate_object_database *alt;
274 int pfxlen, entlen;
275 struct strbuf pathbuf = STRBUF_INIT;
276
277 if (!is_absolute_path(entry) && relative_base) {
278 strbuf_addstr(&pathbuf, real_path(relative_base));
279 strbuf_addch(&pathbuf, '/');
280 }
281 strbuf_addstr(&pathbuf, entry);
282
283 normalize_path_copy(pathbuf.buf, pathbuf.buf);
284
285 pfxlen = strlen(pathbuf.buf);
286
287 /*
288 * The trailing slash after the directory name is given by
289 * this function at the end. Remove duplicates.
290 */
291 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
292 pfxlen -= 1;
293
294 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
295 ent = xmalloc(sizeof(*ent) + entlen);
296 memcpy(ent->base, pathbuf.buf, pfxlen);
297 strbuf_release(&pathbuf);
298
299 ent->name = ent->base + pfxlen + 1;
300 ent->base[pfxlen + 3] = '/';
301 ent->base[pfxlen] = ent->base[entlen-1] = 0;
302
303 /* Detect cases where alternate disappeared */
304 if (!is_directory(ent->base)) {
305 error("object directory %s does not exist; "
306 "check .git/objects/info/alternates.",
307 ent->base);
308 free(ent);
309 return -1;
310 }
311
312 /* Prevent the common mistake of listing the same
313 * thing twice, or object directory itself.
314 */
315 for (alt = alt_odb_list; alt; alt = alt->next) {
316 if (pfxlen == alt->name - alt->base - 1 &&
317 !memcmp(ent->base, alt->base, pfxlen)) {
318 free(ent);
319 return -1;
320 }
321 }
322 if (!strcmp_icase(ent->base, normalized_objdir)) {
323 free(ent);
324 return -1;
325 }
326
327 /* add the alternate entry */
328 *alt_odb_tail = ent;
329 alt_odb_tail = &(ent->next);
330 ent->next = NULL;
331
332 /* recursively add alternates */
333 read_info_alternates(ent->base, depth + 1);
334
335 ent->base[pfxlen] = '/';
336
337 return 0;
338 }
339
340 static void link_alt_odb_entries(const char *alt, int len, int sep,
341 const char *relative_base, int depth)
342 {
343 struct string_list entries = STRING_LIST_INIT_NODUP;
344 char *alt_copy;
345 int i;
346 struct strbuf objdirbuf = STRBUF_INIT;
347
348 if (depth > 5) {
349 error("%s: ignoring alternate object stores, nesting too deep.",
350 relative_base);
351 return;
352 }
353
354 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
355 normalize_path_copy(objdirbuf.buf, objdirbuf.buf);
356
357 alt_copy = xmemdupz(alt, len);
358 string_list_split_in_place(&entries, alt_copy, sep, -1);
359 for (i = 0; i < entries.nr; i++) {
360 const char *entry = entries.items[i].string;
361 if (entry[0] == '\0' || entry[0] == '#')
362 continue;
363 if (!is_absolute_path(entry) && depth) {
364 error("%s: ignoring relative alternate object store %s",
365 relative_base, entry);
366 } else {
367 link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
368 }
369 }
370 string_list_clear(&entries, 0);
371 free(alt_copy);
372 strbuf_release(&objdirbuf);
373 }
374
375 void read_info_alternates(const char * relative_base, int depth)
376 {
377 char *map;
378 size_t mapsz;
379 struct stat st;
380 char *path;
381 int fd;
382
383 path = xstrfmt("%s/info/alternates", relative_base);
384 fd = git_open_noatime(path);
385 free(path);
386 if (fd < 0)
387 return;
388 if (fstat(fd, &st) || (st.st_size == 0)) {
389 close(fd);
390 return;
391 }
392 mapsz = xsize_t(st.st_size);
393 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
394 close(fd);
395
396 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
397
398 munmap(map, mapsz);
399 }
400
401 void add_to_alternates_file(const char *reference)
402 {
403 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
404 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
405 const char *alt = mkpath("%s\n", reference);
406 write_or_die(fd, alt, strlen(alt));
407 if (commit_lock_file(lock))
408 die("could not close alternates file");
409 if (alt_odb_tail)
410 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
411 }
412
413 int foreach_alt_odb(alt_odb_fn fn, void *cb)
414 {
415 struct alternate_object_database *ent;
416 int r = 0;
417
418 prepare_alt_odb();
419 for (ent = alt_odb_list; ent; ent = ent->next) {
420 r = fn(ent, cb);
421 if (r)
422 break;
423 }
424 return r;
425 }
426
427 void prepare_alt_odb(void)
428 {
429 const char *alt;
430
431 if (alt_odb_tail)
432 return;
433
434 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
435 if (!alt) alt = "";
436
437 alt_odb_tail = &alt_odb_list;
438 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
439
440 read_info_alternates(get_object_directory(), 0);
441 }
442
443 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
444 static int freshen_file(const char *fn)
445 {
446 struct utimbuf t;
447 t.actime = t.modtime = time(NULL);
448 return !utime(fn, &t);
449 }
450
451 /*
452 * All of the check_and_freshen functions return 1 if the file exists and was
453 * freshened (if freshening was requested), 0 otherwise. If they return
454 * 0, you should not assume that it is safe to skip a write of the object (it
455 * either does not exist on disk, or has a stale mtime and may be subject to
456 * pruning).
457 */
458 static int check_and_freshen_file(const char *fn, int freshen)
459 {
460 if (access(fn, F_OK))
461 return 0;
462 if (freshen && !freshen_file(fn))
463 return 0;
464 return 1;
465 }
466
467 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
468 {
469 return check_and_freshen_file(sha1_file_name(sha1), freshen);
470 }
471
472 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
473 {
474 struct alternate_object_database *alt;
475 prepare_alt_odb();
476 for (alt = alt_odb_list; alt; alt = alt->next) {
477 fill_sha1_path(alt->name, sha1);
478 if (check_and_freshen_file(alt->base, freshen))
479 return 1;
480 }
481 return 0;
482 }
483
484 static int check_and_freshen(const unsigned char *sha1, int freshen)
485 {
486 return check_and_freshen_local(sha1, freshen) ||
487 check_and_freshen_nonlocal(sha1, freshen);
488 }
489
490 int has_loose_object_nonlocal(const unsigned char *sha1)
491 {
492 return check_and_freshen_nonlocal(sha1, 0);
493 }
494
495 static int has_loose_object(const unsigned char *sha1)
496 {
497 return check_and_freshen(sha1, 0);
498 }
499
500 static unsigned int pack_used_ctr;
501 static unsigned int pack_mmap_calls;
502 static unsigned int peak_pack_open_windows;
503 static unsigned int pack_open_windows;
504 static unsigned int pack_open_fds;
505 static unsigned int pack_max_fds;
506 static size_t peak_pack_mapped;
507 static size_t pack_mapped;
508 struct packed_git *packed_git;
509
510 void pack_report(void)
511 {
512 fprintf(stderr,
513 "pack_report: getpagesize() = %10" SZ_FMT "\n"
514 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
515 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
516 sz_fmt(getpagesize()),
517 sz_fmt(packed_git_window_size),
518 sz_fmt(packed_git_limit));
519 fprintf(stderr,
520 "pack_report: pack_used_ctr = %10u\n"
521 "pack_report: pack_mmap_calls = %10u\n"
522 "pack_report: pack_open_windows = %10u / %10u\n"
523 "pack_report: pack_mapped = "
524 "%10" SZ_FMT " / %10" SZ_FMT "\n",
525 pack_used_ctr,
526 pack_mmap_calls,
527 pack_open_windows, peak_pack_open_windows,
528 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
529 }
530
531 /*
532 * Open and mmap the index file at path, perform a couple of
533 * consistency checks, then record its information to p. Return 0 on
534 * success.
535 */
536 static int check_packed_git_idx(const char *path, struct packed_git *p)
537 {
538 void *idx_map;
539 struct pack_idx_header *hdr;
540 size_t idx_size;
541 uint32_t version, nr, i, *index;
542 int fd = git_open_noatime(path);
543 struct stat st;
544
545 if (fd < 0)
546 return -1;
547 if (fstat(fd, &st)) {
548 close(fd);
549 return -1;
550 }
551 idx_size = xsize_t(st.st_size);
552 if (idx_size < 4 * 256 + 20 + 20) {
553 close(fd);
554 return error("index file %s is too small", path);
555 }
556 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
557 close(fd);
558
559 hdr = idx_map;
560 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
561 version = ntohl(hdr->idx_version);
562 if (version < 2 || version > 2) {
563 munmap(idx_map, idx_size);
564 return error("index file %s is version %"PRIu32
565 " and is not supported by this binary"
566 " (try upgrading GIT to a newer version)",
567 path, version);
568 }
569 } else
570 version = 1;
571
572 nr = 0;
573 index = idx_map;
574 if (version > 1)
575 index += 2; /* skip index header */
576 for (i = 0; i < 256; i++) {
577 uint32_t n = ntohl(index[i]);
578 if (n < nr) {
579 munmap(idx_map, idx_size);
580 return error("non-monotonic index %s", path);
581 }
582 nr = n;
583 }
584
585 if (version == 1) {
586 /*
587 * Total size:
588 * - 256 index entries 4 bytes each
589 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
590 * - 20-byte SHA1 of the packfile
591 * - 20-byte SHA1 file checksum
592 */
593 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
594 munmap(idx_map, idx_size);
595 return error("wrong index v1 file size in %s", path);
596 }
597 } else if (version == 2) {
598 /*
599 * Minimum size:
600 * - 8 bytes of header
601 * - 256 index entries 4 bytes each
602 * - 20-byte sha1 entry * nr
603 * - 4-byte crc entry * nr
604 * - 4-byte offset entry * nr
605 * - 20-byte SHA1 of the packfile
606 * - 20-byte SHA1 file checksum
607 * And after the 4-byte offset table might be a
608 * variable sized table containing 8-byte entries
609 * for offsets larger than 2^31.
610 */
611 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
612 unsigned long max_size = min_size;
613 if (nr)
614 max_size += (nr - 1)*8;
615 if (idx_size < min_size || idx_size > max_size) {
616 munmap(idx_map, idx_size);
617 return error("wrong index v2 file size in %s", path);
618 }
619 if (idx_size != min_size &&
620 /*
621 * make sure we can deal with large pack offsets.
622 * 31-bit signed offset won't be enough, neither
623 * 32-bit unsigned one will be.
624 */
625 (sizeof(off_t) <= 4)) {
626 munmap(idx_map, idx_size);
627 return error("pack too large for current definition of off_t in %s", path);
628 }
629 }
630
631 p->index_version = version;
632 p->index_data = idx_map;
633 p->index_size = idx_size;
634 p->num_objects = nr;
635 return 0;
636 }
637
638 int open_pack_index(struct packed_git *p)
639 {
640 char *idx_name;
641 int ret;
642
643 if (p->index_data)
644 return 0;
645
646 idx_name = xstrdup(p->pack_name);
647 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
648 ret = check_packed_git_idx(idx_name, p);
649 free(idx_name);
650 return ret;
651 }
652
653 static void scan_windows(struct packed_git *p,
654 struct packed_git **lru_p,
655 struct pack_window **lru_w,
656 struct pack_window **lru_l)
657 {
658 struct pack_window *w, *w_l;
659
660 for (w_l = NULL, w = p->windows; w; w = w->next) {
661 if (!w->inuse_cnt) {
662 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
663 *lru_p = p;
664 *lru_w = w;
665 *lru_l = w_l;
666 }
667 }
668 w_l = w;
669 }
670 }
671
672 static int unuse_one_window(struct packed_git *current)
673 {
674 struct packed_git *p, *lru_p = NULL;
675 struct pack_window *lru_w = NULL, *lru_l = NULL;
676
677 if (current)
678 scan_windows(current, &lru_p, &lru_w, &lru_l);
679 for (p = packed_git; p; p = p->next)
680 scan_windows(p, &lru_p, &lru_w, &lru_l);
681 if (lru_p) {
682 munmap(lru_w->base, lru_w->len);
683 pack_mapped -= lru_w->len;
684 if (lru_l)
685 lru_l->next = lru_w->next;
686 else
687 lru_p->windows = lru_w->next;
688 free(lru_w);
689 pack_open_windows--;
690 return 1;
691 }
692 return 0;
693 }
694
695 void release_pack_memory(size_t need)
696 {
697 size_t cur = pack_mapped;
698 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
699 ; /* nothing */
700 }
701
702 static void mmap_limit_check(size_t length)
703 {
704 static size_t limit = 0;
705 if (!limit) {
706 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
707 if (!limit)
708 limit = SIZE_MAX;
709 }
710 if (length > limit)
711 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
712 (uintmax_t)length, (uintmax_t)limit);
713 }
714
715 void *xmmap_gently(void *start, size_t length,
716 int prot, int flags, int fd, off_t offset)
717 {
718 void *ret;
719
720 mmap_limit_check(length);
721 ret = mmap(start, length, prot, flags, fd, offset);
722 if (ret == MAP_FAILED) {
723 if (!length)
724 return NULL;
725 release_pack_memory(length);
726 ret = mmap(start, length, prot, flags, fd, offset);
727 }
728 return ret;
729 }
730
731 void *xmmap(void *start, size_t length,
732 int prot, int flags, int fd, off_t offset)
733 {
734 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
735 if (ret == MAP_FAILED)
736 die_errno("mmap failed");
737 return ret;
738 }
739
740 void close_pack_windows(struct packed_git *p)
741 {
742 while (p->windows) {
743 struct pack_window *w = p->windows;
744
745 if (w->inuse_cnt)
746 die("pack '%s' still has open windows to it",
747 p->pack_name);
748 munmap(w->base, w->len);
749 pack_mapped -= w->len;
750 pack_open_windows--;
751 p->windows = w->next;
752 free(w);
753 }
754 }
755
756 /*
757 * The LRU pack is the one with the oldest MRU window, preferring packs
758 * with no used windows, or the oldest mtime if it has no windows allocated.
759 */
760 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
761 {
762 struct pack_window *w, *this_mru_w;
763 int has_windows_inuse = 0;
764
765 /*
766 * Reject this pack if it has windows and the previously selected
767 * one does not. If this pack does not have windows, reject
768 * it if the pack file is newer than the previously selected one.
769 */
770 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
771 return;
772
773 for (w = this_mru_w = p->windows; w; w = w->next) {
774 /*
775 * Reject this pack if any of its windows are in use,
776 * but the previously selected pack did not have any
777 * inuse windows. Otherwise, record that this pack
778 * has windows in use.
779 */
780 if (w->inuse_cnt) {
781 if (*accept_windows_inuse)
782 has_windows_inuse = 1;
783 else
784 return;
785 }
786
787 if (w->last_used > this_mru_w->last_used)
788 this_mru_w = w;
789
790 /*
791 * Reject this pack if it has windows that have been
792 * used more recently than the previously selected pack.
793 * If the previously selected pack had windows inuse and
794 * we have not encountered a window in this pack that is
795 * inuse, skip this check since we prefer a pack with no
796 * inuse windows to one that has inuse windows.
797 */
798 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
799 this_mru_w->last_used > (*mru_w)->last_used)
800 return;
801 }
802
803 /*
804 * Select this pack.
805 */
806 *mru_w = this_mru_w;
807 *lru_p = p;
808 *accept_windows_inuse = has_windows_inuse;
809 }
810
811 static int close_one_pack(void)
812 {
813 struct packed_git *p, *lru_p = NULL;
814 struct pack_window *mru_w = NULL;
815 int accept_windows_inuse = 1;
816
817 for (p = packed_git; p; p = p->next) {
818 if (p->pack_fd == -1)
819 continue;
820 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
821 }
822
823 if (lru_p) {
824 close(lru_p->pack_fd);
825 pack_open_fds--;
826 lru_p->pack_fd = -1;
827 return 1;
828 }
829
830 return 0;
831 }
832
833 void unuse_pack(struct pack_window **w_cursor)
834 {
835 struct pack_window *w = *w_cursor;
836 if (w) {
837 w->inuse_cnt--;
838 *w_cursor = NULL;
839 }
840 }
841
842 void close_pack_index(struct packed_git *p)
843 {
844 if (p->index_data) {
845 munmap((void *)p->index_data, p->index_size);
846 p->index_data = NULL;
847 }
848 }
849
850 /*
851 * This is used by git-repack in case a newly created pack happens to
852 * contain the same set of objects as an existing one. In that case
853 * the resulting file might be different even if its name would be the
854 * same. It is best to close any reference to the old pack before it is
855 * replaced on disk. Of course no index pointers or windows for given pack
856 * must subsist at this point. If ever objects from this pack are requested
857 * again, the new version of the pack will be reinitialized through
858 * reprepare_packed_git().
859 */
860 void free_pack_by_name(const char *pack_name)
861 {
862 struct packed_git *p, **pp = &packed_git;
863
864 while (*pp) {
865 p = *pp;
866 if (strcmp(pack_name, p->pack_name) == 0) {
867 clear_delta_base_cache();
868 close_pack_windows(p);
869 if (p->pack_fd != -1) {
870 close(p->pack_fd);
871 pack_open_fds--;
872 }
873 close_pack_index(p);
874 free(p->bad_object_sha1);
875 *pp = p->next;
876 if (last_found_pack == p)
877 last_found_pack = NULL;
878 free(p);
879 return;
880 }
881 pp = &p->next;
882 }
883 }
884
885 static unsigned int get_max_fd_limit(void)
886 {
887 #ifdef RLIMIT_NOFILE
888 {
889 struct rlimit lim;
890
891 if (!getrlimit(RLIMIT_NOFILE, &lim))
892 return lim.rlim_cur;
893 }
894 #endif
895
896 #ifdef _SC_OPEN_MAX
897 {
898 long open_max = sysconf(_SC_OPEN_MAX);
899 if (0 < open_max)
900 return open_max;
901 /*
902 * Otherwise, we got -1 for one of the two
903 * reasons:
904 *
905 * (1) sysconf() did not understand _SC_OPEN_MAX
906 * and signaled an error with -1; or
907 * (2) sysconf() said there is no limit.
908 *
909 * We _could_ clear errno before calling sysconf() to
910 * tell these two cases apart and return a huge number
911 * in the latter case to let the caller cap it to a
912 * value that is not so selfish, but letting the
913 * fallback OPEN_MAX codepath take care of these cases
914 * is a lot simpler.
915 */
916 }
917 #endif
918
919 #ifdef OPEN_MAX
920 return OPEN_MAX;
921 #else
922 return 1; /* see the caller ;-) */
923 #endif
924 }
925
926 /*
927 * Do not call this directly as this leaks p->pack_fd on error return;
928 * call open_packed_git() instead.
929 */
930 static int open_packed_git_1(struct packed_git *p)
931 {
932 struct stat st;
933 struct pack_header hdr;
934 unsigned char sha1[20];
935 unsigned char *idx_sha1;
936 long fd_flag;
937
938 if (!p->index_data && open_pack_index(p))
939 return error("packfile %s index unavailable", p->pack_name);
940
941 if (!pack_max_fds) {
942 unsigned int max_fds = get_max_fd_limit();
943
944 /* Save 3 for stdin/stdout/stderr, 22 for work */
945 if (25 < max_fds)
946 pack_max_fds = max_fds - 25;
947 else
948 pack_max_fds = 1;
949 }
950
951 while (pack_max_fds <= pack_open_fds && close_one_pack())
952 ; /* nothing */
953
954 p->pack_fd = git_open_noatime(p->pack_name);
955 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
956 return -1;
957 pack_open_fds++;
958
959 /* If we created the struct before we had the pack we lack size. */
960 if (!p->pack_size) {
961 if (!S_ISREG(st.st_mode))
962 return error("packfile %s not a regular file", p->pack_name);
963 p->pack_size = st.st_size;
964 } else if (p->pack_size != st.st_size)
965 return error("packfile %s size changed", p->pack_name);
966
967 /* We leave these file descriptors open with sliding mmap;
968 * there is no point keeping them open across exec(), though.
969 */
970 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
971 if (fd_flag < 0)
972 return error("cannot determine file descriptor flags");
973 fd_flag |= FD_CLOEXEC;
974 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
975 return error("cannot set FD_CLOEXEC");
976
977 /* Verify we recognize this pack file format. */
978 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
979 return error("file %s is far too short to be a packfile", p->pack_name);
980 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
981 return error("file %s is not a GIT packfile", p->pack_name);
982 if (!pack_version_ok(hdr.hdr_version))
983 return error("packfile %s is version %"PRIu32" and not"
984 " supported (try upgrading GIT to a newer version)",
985 p->pack_name, ntohl(hdr.hdr_version));
986
987 /* Verify the pack matches its index. */
988 if (p->num_objects != ntohl(hdr.hdr_entries))
989 return error("packfile %s claims to have %"PRIu32" objects"
990 " while index indicates %"PRIu32" objects",
991 p->pack_name, ntohl(hdr.hdr_entries),
992 p->num_objects);
993 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
994 return error("end of packfile %s is unavailable", p->pack_name);
995 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
996 return error("packfile %s signature is unavailable", p->pack_name);
997 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
998 if (hashcmp(sha1, idx_sha1))
999 return error("packfile %s does not match index", p->pack_name);
1000 return 0;
1001 }
1002
1003 static int open_packed_git(struct packed_git *p)
1004 {
1005 if (!open_packed_git_1(p))
1006 return 0;
1007 if (p->pack_fd != -1) {
1008 close(p->pack_fd);
1009 pack_open_fds--;
1010 p->pack_fd = -1;
1011 }
1012 return -1;
1013 }
1014
1015 static int in_window(struct pack_window *win, off_t offset)
1016 {
1017 /* We must promise at least 20 bytes (one hash) after the
1018 * offset is available from this window, otherwise the offset
1019 * is not actually in this window and a different window (which
1020 * has that one hash excess) must be used. This is to support
1021 * the object header and delta base parsing routines below.
1022 */
1023 off_t win_off = win->offset;
1024 return win_off <= offset
1025 && (offset + 20) <= (win_off + win->len);
1026 }
1027
1028 unsigned char *use_pack(struct packed_git *p,
1029 struct pack_window **w_cursor,
1030 off_t offset,
1031 unsigned long *left)
1032 {
1033 struct pack_window *win = *w_cursor;
1034
1035 /* Since packfiles end in a hash of their content and it's
1036 * pointless to ask for an offset into the middle of that
1037 * hash, and the in_window function above wouldn't match
1038 * don't allow an offset too close to the end of the file.
1039 */
1040 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1041 die("packfile %s cannot be accessed", p->pack_name);
1042 if (offset > (p->pack_size - 20))
1043 die("offset beyond end of packfile (truncated pack?)");
1044
1045 if (!win || !in_window(win, offset)) {
1046 if (win)
1047 win->inuse_cnt--;
1048 for (win = p->windows; win; win = win->next) {
1049 if (in_window(win, offset))
1050 break;
1051 }
1052 if (!win) {
1053 size_t window_align = packed_git_window_size / 2;
1054 off_t len;
1055
1056 if (p->pack_fd == -1 && open_packed_git(p))
1057 die("packfile %s cannot be accessed", p->pack_name);
1058
1059 win = xcalloc(1, sizeof(*win));
1060 win->offset = (offset / window_align) * window_align;
1061 len = p->pack_size - win->offset;
1062 if (len > packed_git_window_size)
1063 len = packed_git_window_size;
1064 win->len = (size_t)len;
1065 pack_mapped += win->len;
1066 while (packed_git_limit < pack_mapped
1067 && unuse_one_window(p))
1068 ; /* nothing */
1069 win->base = xmmap(NULL, win->len,
1070 PROT_READ, MAP_PRIVATE,
1071 p->pack_fd, win->offset);
1072 if (win->base == MAP_FAILED)
1073 die("packfile %s cannot be mapped: %s",
1074 p->pack_name,
1075 strerror(errno));
1076 if (!win->offset && win->len == p->pack_size
1077 && !p->do_not_close) {
1078 close(p->pack_fd);
1079 pack_open_fds--;
1080 p->pack_fd = -1;
1081 }
1082 pack_mmap_calls++;
1083 pack_open_windows++;
1084 if (pack_mapped > peak_pack_mapped)
1085 peak_pack_mapped = pack_mapped;
1086 if (pack_open_windows > peak_pack_open_windows)
1087 peak_pack_open_windows = pack_open_windows;
1088 win->next = p->windows;
1089 p->windows = win;
1090 }
1091 }
1092 if (win != *w_cursor) {
1093 win->last_used = pack_used_ctr++;
1094 win->inuse_cnt++;
1095 *w_cursor = win;
1096 }
1097 offset -= win->offset;
1098 if (left)
1099 *left = win->len - xsize_t(offset);
1100 return win->base + offset;
1101 }
1102
1103 static struct packed_git *alloc_packed_git(int extra)
1104 {
1105 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1106 memset(p, 0, sizeof(*p));
1107 p->pack_fd = -1;
1108 return p;
1109 }
1110
1111 static void try_to_free_pack_memory(size_t size)
1112 {
1113 release_pack_memory(size);
1114 }
1115
1116 struct packed_git *add_packed_git(const char *path, int path_len, int local)
1117 {
1118 static int have_set_try_to_free_routine;
1119 struct stat st;
1120 struct packed_git *p = alloc_packed_git(path_len + 2);
1121
1122 if (!have_set_try_to_free_routine) {
1123 have_set_try_to_free_routine = 1;
1124 set_try_to_free_routine(try_to_free_pack_memory);
1125 }
1126
1127 /*
1128 * Make sure a corresponding .pack file exists and that
1129 * the index looks sane.
1130 */
1131 path_len -= strlen(".idx");
1132 if (path_len < 1) {
1133 free(p);
1134 return NULL;
1135 }
1136 memcpy(p->pack_name, path, path_len);
1137
1138 strcpy(p->pack_name + path_len, ".keep");
1139 if (!access(p->pack_name, F_OK))
1140 p->pack_keep = 1;
1141
1142 strcpy(p->pack_name + path_len, ".pack");
1143 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1144 free(p);
1145 return NULL;
1146 }
1147
1148 /* ok, it looks sane as far as we can check without
1149 * actually mapping the pack file.
1150 */
1151 p->pack_size = st.st_size;
1152 p->pack_local = local;
1153 p->mtime = st.st_mtime;
1154 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1155 hashclr(p->sha1);
1156 return p;
1157 }
1158
1159 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1160 {
1161 const char *path = sha1_pack_name(sha1);
1162 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
1163
1164 strcpy(p->pack_name, path);
1165 hashcpy(p->sha1, sha1);
1166 if (check_packed_git_idx(idx_path, p)) {
1167 free(p);
1168 return NULL;
1169 }
1170
1171 return p;
1172 }
1173
1174 void install_packed_git(struct packed_git *pack)
1175 {
1176 if (pack->pack_fd != -1)
1177 pack_open_fds++;
1178
1179 pack->next = packed_git;
1180 packed_git = pack;
1181 }
1182
1183 void (*report_garbage)(const char *desc, const char *path);
1184
1185 static void report_helper(const struct string_list *list,
1186 int seen_bits, int first, int last)
1187 {
1188 const char *msg;
1189 switch (seen_bits) {
1190 case 0:
1191 msg = "no corresponding .idx or .pack";
1192 break;
1193 case 1:
1194 msg = "no corresponding .idx";
1195 break;
1196 case 2:
1197 msg = "no corresponding .pack";
1198 break;
1199 default:
1200 return;
1201 }
1202 for (; first < last; first++)
1203 report_garbage(msg, list->items[first].string);
1204 }
1205
1206 static void report_pack_garbage(struct string_list *list)
1207 {
1208 int i, baselen = -1, first = 0, seen_bits = 0;
1209
1210 if (!report_garbage)
1211 return;
1212
1213 string_list_sort(list);
1214
1215 for (i = 0; i < list->nr; i++) {
1216 const char *path = list->items[i].string;
1217 if (baselen != -1 &&
1218 strncmp(path, list->items[first].string, baselen)) {
1219 report_helper(list, seen_bits, first, i);
1220 baselen = -1;
1221 seen_bits = 0;
1222 }
1223 if (baselen == -1) {
1224 const char *dot = strrchr(path, '.');
1225 if (!dot) {
1226 report_garbage("garbage found", path);
1227 continue;
1228 }
1229 baselen = dot - path + 1;
1230 first = i;
1231 }
1232 if (!strcmp(path + baselen, "pack"))
1233 seen_bits |= 1;
1234 else if (!strcmp(path + baselen, "idx"))
1235 seen_bits |= 2;
1236 }
1237 report_helper(list, seen_bits, first, list->nr);
1238 }
1239
1240 static void prepare_packed_git_one(char *objdir, int local)
1241 {
1242 struct strbuf path = STRBUF_INIT;
1243 size_t dirnamelen;
1244 DIR *dir;
1245 struct dirent *de;
1246 struct string_list garbage = STRING_LIST_INIT_DUP;
1247
1248 strbuf_addstr(&path, objdir);
1249 strbuf_addstr(&path, "/pack");
1250 dir = opendir(path.buf);
1251 if (!dir) {
1252 if (errno != ENOENT)
1253 error("unable to open object pack directory: %s: %s",
1254 path.buf, strerror(errno));
1255 strbuf_release(&path);
1256 return;
1257 }
1258 strbuf_addch(&path, '/');
1259 dirnamelen = path.len;
1260 while ((de = readdir(dir)) != NULL) {
1261 struct packed_git *p;
1262 size_t base_len;
1263
1264 if (is_dot_or_dotdot(de->d_name))
1265 continue;
1266
1267 strbuf_setlen(&path, dirnamelen);
1268 strbuf_addstr(&path, de->d_name);
1269
1270 base_len = path.len;
1271 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1272 /* Don't reopen a pack we already have. */
1273 for (p = packed_git; p; p = p->next) {
1274 size_t len;
1275 if (strip_suffix(p->pack_name, ".pack", &len) &&
1276 len == base_len &&
1277 !memcmp(p->pack_name, path.buf, len))
1278 break;
1279 }
1280 if (p == NULL &&
1281 /*
1282 * See if it really is a valid .idx file with
1283 * corresponding .pack file that we can map.
1284 */
1285 (p = add_packed_git(path.buf, path.len, local)) != NULL)
1286 install_packed_git(p);
1287 }
1288
1289 if (!report_garbage)
1290 continue;
1291
1292 if (ends_with(de->d_name, ".idx") ||
1293 ends_with(de->d_name, ".pack") ||
1294 ends_with(de->d_name, ".bitmap") ||
1295 ends_with(de->d_name, ".keep"))
1296 string_list_append(&garbage, path.buf);
1297 else
1298 report_garbage("garbage found", path.buf);
1299 }
1300 closedir(dir);
1301 report_pack_garbage(&garbage);
1302 string_list_clear(&garbage, 0);
1303 strbuf_release(&path);
1304 }
1305
1306 static int sort_pack(const void *a_, const void *b_)
1307 {
1308 struct packed_git *a = *((struct packed_git **)a_);
1309 struct packed_git *b = *((struct packed_git **)b_);
1310 int st;
1311
1312 /*
1313 * Local packs tend to contain objects specific to our
1314 * variant of the project than remote ones. In addition,
1315 * remote ones could be on a network mounted filesystem.
1316 * Favor local ones for these reasons.
1317 */
1318 st = a->pack_local - b->pack_local;
1319 if (st)
1320 return -st;
1321
1322 /*
1323 * Younger packs tend to contain more recent objects,
1324 * and more recent objects tend to get accessed more
1325 * often.
1326 */
1327 if (a->mtime < b->mtime)
1328 return 1;
1329 else if (a->mtime == b->mtime)
1330 return 0;
1331 return -1;
1332 }
1333
1334 static void rearrange_packed_git(void)
1335 {
1336 struct packed_git **ary, *p;
1337 int i, n;
1338
1339 for (n = 0, p = packed_git; p; p = p->next)
1340 n++;
1341 if (n < 2)
1342 return;
1343
1344 /* prepare an array of packed_git for easier sorting */
1345 ary = xcalloc(n, sizeof(struct packed_git *));
1346 for (n = 0, p = packed_git; p; p = p->next)
1347 ary[n++] = p;
1348
1349 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1350
1351 /* link them back again */
1352 for (i = 0; i < n - 1; i++)
1353 ary[i]->next = ary[i + 1];
1354 ary[n - 1]->next = NULL;
1355 packed_git = ary[0];
1356
1357 free(ary);
1358 }
1359
1360 static int prepare_packed_git_run_once = 0;
1361 void prepare_packed_git(void)
1362 {
1363 struct alternate_object_database *alt;
1364
1365 if (prepare_packed_git_run_once)
1366 return;
1367 prepare_packed_git_one(get_object_directory(), 1);
1368 prepare_alt_odb();
1369 for (alt = alt_odb_list; alt; alt = alt->next) {
1370 alt->name[-1] = 0;
1371 prepare_packed_git_one(alt->base, 0);
1372 alt->name[-1] = '/';
1373 }
1374 rearrange_packed_git();
1375 prepare_packed_git_run_once = 1;
1376 }
1377
1378 void reprepare_packed_git(void)
1379 {
1380 prepare_packed_git_run_once = 0;
1381 prepare_packed_git();
1382 }
1383
1384 static void mark_bad_packed_object(struct packed_git *p,
1385 const unsigned char *sha1)
1386 {
1387 unsigned i;
1388 for (i = 0; i < p->num_bad_objects; i++)
1389 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1390 return;
1391 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1392 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1393 p->num_bad_objects++;
1394 }
1395
1396 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1397 {
1398 struct packed_git *p;
1399 unsigned i;
1400
1401 for (p = packed_git; p; p = p->next)
1402 for (i = 0; i < p->num_bad_objects; i++)
1403 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1404 return p;
1405 return NULL;
1406 }
1407
1408 /*
1409 * With an in-core object data in "map", rehash it to make sure the
1410 * object name actually matches "sha1" to detect object corruption.
1411 * With "map" == NULL, try reading the object named with "sha1" using
1412 * the streaming interface and rehash it to do the same.
1413 */
1414 int check_sha1_signature(const unsigned char *sha1, void *map,
1415 unsigned long size, const char *type)
1416 {
1417 unsigned char real_sha1[20];
1418 enum object_type obj_type;
1419 struct git_istream *st;
1420 git_SHA_CTX c;
1421 char hdr[32];
1422 int hdrlen;
1423
1424 if (map) {
1425 hash_sha1_file(map, size, type, real_sha1);
1426 return hashcmp(sha1, real_sha1) ? -1 : 0;
1427 }
1428
1429 st = open_istream(sha1, &obj_type, &size, NULL);
1430 if (!st)
1431 return -1;
1432
1433 /* Generate the header */
1434 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1435
1436 /* Sha1.. */
1437 git_SHA1_Init(&c);
1438 git_SHA1_Update(&c, hdr, hdrlen);
1439 for (;;) {
1440 char buf[1024 * 16];
1441 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1442
1443 if (readlen < 0) {
1444 close_istream(st);
1445 return -1;
1446 }
1447 if (!readlen)
1448 break;
1449 git_SHA1_Update(&c, buf, readlen);
1450 }
1451 git_SHA1_Final(real_sha1, &c);
1452 close_istream(st);
1453 return hashcmp(sha1, real_sha1) ? -1 : 0;
1454 }
1455
1456 int git_open_noatime(const char *name)
1457 {
1458 static int sha1_file_open_flag = O_NOATIME;
1459
1460 for (;;) {
1461 int fd;
1462
1463 errno = 0;
1464 fd = open(name, O_RDONLY | sha1_file_open_flag);
1465 if (fd >= 0)
1466 return fd;
1467
1468 /* Might the failure be due to O_NOATIME? */
1469 if (errno != ENOENT && sha1_file_open_flag) {
1470 sha1_file_open_flag = 0;
1471 continue;
1472 }
1473
1474 return -1;
1475 }
1476 }
1477
1478 static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1479 {
1480 struct alternate_object_database *alt;
1481
1482 if (!lstat(sha1_file_name(sha1), st))
1483 return 0;
1484
1485 prepare_alt_odb();
1486 errno = ENOENT;
1487 for (alt = alt_odb_list; alt; alt = alt->next) {
1488 fill_sha1_path(alt->name, sha1);
1489 if (!lstat(alt->base, st))
1490 return 0;
1491 }
1492
1493 return -1;
1494 }
1495
1496 static int open_sha1_file(const unsigned char *sha1)
1497 {
1498 int fd;
1499 struct alternate_object_database *alt;
1500 int most_interesting_errno;
1501
1502 fd = git_open_noatime(sha1_file_name(sha1));
1503 if (fd >= 0)
1504 return fd;
1505 most_interesting_errno = errno;
1506
1507 prepare_alt_odb();
1508 for (alt = alt_odb_list; alt; alt = alt->next) {
1509 fill_sha1_path(alt->name, sha1);
1510 fd = git_open_noatime(alt->base);
1511 if (fd >= 0)
1512 return fd;
1513 if (most_interesting_errno == ENOENT)
1514 most_interesting_errno = errno;
1515 }
1516 errno = most_interesting_errno;
1517 return -1;
1518 }
1519
1520 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1521 {
1522 void *map;
1523 int fd;
1524
1525 fd = open_sha1_file(sha1);
1526 map = NULL;
1527 if (fd >= 0) {
1528 struct stat st;
1529
1530 if (!fstat(fd, &st)) {
1531 *size = xsize_t(st.st_size);
1532 if (!*size) {
1533 /* mmap() is forbidden on empty files */
1534 error("object file %s is empty", sha1_file_name(sha1));
1535 return NULL;
1536 }
1537 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1538 }
1539 close(fd);
1540 }
1541 return map;
1542 }
1543
1544 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1545 unsigned long len, enum object_type *type, unsigned long *sizep)
1546 {
1547 unsigned shift;
1548 unsigned long size, c;
1549 unsigned long used = 0;
1550
1551 c = buf[used++];
1552 *type = (c >> 4) & 7;
1553 size = c & 15;
1554 shift = 4;
1555 while (c & 0x80) {
1556 if (len <= used || bitsizeof(long) <= shift) {
1557 error("bad object header");
1558 size = used = 0;
1559 break;
1560 }
1561 c = buf[used++];
1562 size += (c & 0x7f) << shift;
1563 shift += 7;
1564 }
1565 *sizep = size;
1566 return used;
1567 }
1568
1569 static int unpack_sha1_short_header(git_zstream *stream,
1570 unsigned char *map, unsigned long mapsize,
1571 void *buffer, unsigned long bufsiz)
1572 {
1573 /* Get the data stream */
1574 memset(stream, 0, sizeof(*stream));
1575 stream->next_in = map;
1576 stream->avail_in = mapsize;
1577 stream->next_out = buffer;
1578 stream->avail_out = bufsiz;
1579
1580 git_inflate_init(stream);
1581 return git_inflate(stream, 0);
1582 }
1583
1584 int unpack_sha1_header(git_zstream *stream,
1585 unsigned char *map, unsigned long mapsize,
1586 void *buffer, unsigned long bufsiz)
1587 {
1588 int status = unpack_sha1_short_header(stream, map, mapsize,
1589 buffer, bufsiz);
1590
1591 if (status < Z_OK)
1592 return status;
1593
1594 /* Make sure we have the terminating NUL */
1595 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1596 return -1;
1597 return 0;
1598 }
1599
1600 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1601 unsigned long mapsize, void *buffer,
1602 unsigned long bufsiz, struct strbuf *header)
1603 {
1604 int status;
1605
1606 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1607 if (status < Z_OK)
1608 return -1;
1609
1610 /*
1611 * Check if entire header is unpacked in the first iteration.
1612 */
1613 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1614 return 0;
1615
1616 /*
1617 * buffer[0..bufsiz] was not large enough. Copy the partial
1618 * result out to header, and then append the result of further
1619 * reading the stream.
1620 */
1621 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1622 stream->next_out = buffer;
1623 stream->avail_out = bufsiz;
1624
1625 do {
1626 status = git_inflate(stream, 0);
1627 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1628 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1629 return 0;
1630 stream->next_out = buffer;
1631 stream->avail_out = bufsiz;
1632 } while (status != Z_STREAM_END);
1633 return -1;
1634 }
1635
1636 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1637 {
1638 int bytes = strlen(buffer) + 1;
1639 unsigned char *buf = xmallocz(size);
1640 unsigned long n;
1641 int status = Z_OK;
1642
1643 n = stream->total_out - bytes;
1644 if (n > size)
1645 n = size;
1646 memcpy(buf, (char *) buffer + bytes, n);
1647 bytes = n;
1648 if (bytes <= size) {
1649 /*
1650 * The above condition must be (bytes <= size), not
1651 * (bytes < size). In other words, even though we
1652 * expect no more output and set avail_out to zero,
1653 * the input zlib stream may have bytes that express
1654 * "this concludes the stream", and we *do* want to
1655 * eat that input.
1656 *
1657 * Otherwise we would not be able to test that we
1658 * consumed all the input to reach the expected size;
1659 * we also want to check that zlib tells us that all
1660 * went well with status == Z_STREAM_END at the end.
1661 */
1662 stream->next_out = buf + bytes;
1663 stream->avail_out = size - bytes;
1664 while (status == Z_OK)
1665 status = git_inflate(stream, Z_FINISH);
1666 }
1667 if (status == Z_STREAM_END && !stream->avail_in) {
1668 git_inflate_end(stream);
1669 return buf;
1670 }
1671
1672 if (status < 0)
1673 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1674 else if (stream->avail_in)
1675 error("garbage at end of loose object '%s'",
1676 sha1_to_hex(sha1));
1677 free(buf);
1678 return NULL;
1679 }
1680
1681 /*
1682 * We used to just use "sscanf()", but that's actually way
1683 * too permissive for what we want to check. So do an anal
1684 * object header parse by hand.
1685 */
1686 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1687 unsigned int flags)
1688 {
1689 const char *type_buf = hdr;
1690 unsigned long size;
1691 int type, type_len = 0;
1692
1693 /*
1694 * The type can be of any size but is followed by
1695 * a space.
1696 */
1697 for (;;) {
1698 char c = *hdr++;
1699 if (!c)
1700 return -1;
1701 if (c == ' ')
1702 break;
1703 type_len++;
1704 }
1705
1706 type = type_from_string_gently(type_buf, type_len, 1);
1707 if (oi->typename)
1708 strbuf_add(oi->typename, type_buf, type_len);
1709 /*
1710 * Set type to 0 if its an unknown object and
1711 * we're obtaining the type using '--allow-unkown-type'
1712 * option.
1713 */
1714 if ((flags & LOOKUP_UNKNOWN_OBJECT) && (type < 0))
1715 type = 0;
1716 else if (type < 0)
1717 die("invalid object type");
1718 if (oi->typep)
1719 *oi->typep = type;
1720
1721 /*
1722 * The length must follow immediately, and be in canonical
1723 * decimal format (ie "010" is not valid).
1724 */
1725 size = *hdr++ - '0';
1726 if (size > 9)
1727 return -1;
1728 if (size) {
1729 for (;;) {
1730 unsigned long c = *hdr - '0';
1731 if (c > 9)
1732 break;
1733 hdr++;
1734 size = size * 10 + c;
1735 }
1736 }
1737
1738 if (oi->sizep)
1739 *oi->sizep = size;
1740
1741 /*
1742 * The length must be followed by a zero byte
1743 */
1744 return *hdr ? -1 : type;
1745 }
1746
1747 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1748 {
1749 struct object_info oi;
1750
1751 oi.sizep = sizep;
1752 oi.typename = NULL;
1753 oi.typep = NULL;
1754 return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
1755 }
1756
1757 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1758 {
1759 int ret;
1760 git_zstream stream;
1761 char hdr[8192];
1762
1763 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1764 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1765 return NULL;
1766
1767 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1768 }
1769
1770 unsigned long get_size_from_delta(struct packed_git *p,
1771 struct pack_window **w_curs,
1772 off_t curpos)
1773 {
1774 const unsigned char *data;
1775 unsigned char delta_head[20], *in;
1776 git_zstream stream;
1777 int st;
1778
1779 memset(&stream, 0, sizeof(stream));
1780 stream.next_out = delta_head;
1781 stream.avail_out = sizeof(delta_head);
1782
1783 git_inflate_init(&stream);
1784 do {
1785 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1786 stream.next_in = in;
1787 st = git_inflate(&stream, Z_FINISH);
1788 curpos += stream.next_in - in;
1789 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1790 stream.total_out < sizeof(delta_head));
1791 git_inflate_end(&stream);
1792 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1793 error("delta data unpack-initial failed");
1794 return 0;
1795 }
1796
1797 /* Examine the initial part of the delta to figure out
1798 * the result size.
1799 */
1800 data = delta_head;
1801
1802 /* ignore base size */
1803 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1804
1805 /* Read the result size */
1806 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1807 }
1808
1809 static off_t get_delta_base(struct packed_git *p,
1810 struct pack_window **w_curs,
1811 off_t *curpos,
1812 enum object_type type,
1813 off_t delta_obj_offset)
1814 {
1815 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1816 off_t base_offset;
1817
1818 /* use_pack() assured us we have [base_info, base_info + 20)
1819 * as a range that we can look at without walking off the
1820 * end of the mapped window. Its actually the hash size
1821 * that is assured. An OFS_DELTA longer than the hash size
1822 * is stupid, as then a REF_DELTA would be smaller to store.
1823 */
1824 if (type == OBJ_OFS_DELTA) {
1825 unsigned used = 0;
1826 unsigned char c = base_info[used++];
1827 base_offset = c & 127;
1828 while (c & 128) {
1829 base_offset += 1;
1830 if (!base_offset || MSB(base_offset, 7))
1831 return 0; /* overflow */
1832 c = base_info[used++];
1833 base_offset = (base_offset << 7) + (c & 127);
1834 }
1835 base_offset = delta_obj_offset - base_offset;
1836 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1837 return 0; /* out of bound */
1838 *curpos += used;
1839 } else if (type == OBJ_REF_DELTA) {
1840 /* The base entry _must_ be in the same pack */
1841 base_offset = find_pack_entry_one(base_info, p);
1842 *curpos += 20;
1843 } else
1844 die("I am totally screwed");
1845 return base_offset;
1846 }
1847
1848 /*
1849 * Like get_delta_base above, but we return the sha1 instead of the pack
1850 * offset. This means it is cheaper for REF deltas (we do not have to do
1851 * the final object lookup), but more expensive for OFS deltas (we
1852 * have to load the revidx to convert the offset back into a sha1).
1853 */
1854 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1855 struct pack_window **w_curs,
1856 off_t curpos,
1857 enum object_type type,
1858 off_t delta_obj_offset)
1859 {
1860 if (type == OBJ_REF_DELTA) {
1861 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1862 return base;
1863 } else if (type == OBJ_OFS_DELTA) {
1864 struct revindex_entry *revidx;
1865 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1866 type, delta_obj_offset);
1867
1868 if (!base_offset)
1869 return NULL;
1870
1871 revidx = find_pack_revindex(p, base_offset);
1872 if (!revidx)
1873 return NULL;
1874
1875 return nth_packed_object_sha1(p, revidx->nr);
1876 } else
1877 return NULL;
1878 }
1879
1880 int unpack_object_header(struct packed_git *p,
1881 struct pack_window **w_curs,
1882 off_t *curpos,
1883 unsigned long *sizep)
1884 {
1885 unsigned char *base;
1886 unsigned long left;
1887 unsigned long used;
1888 enum object_type type;
1889
1890 /* use_pack() assures us we have [base, base + 20) available
1891 * as a range that we can look at. (Its actually the hash
1892 * size that is assured.) With our object header encoding
1893 * the maximum deflated object size is 2^137, which is just
1894 * insane, so we know won't exceed what we have been given.
1895 */
1896 base = use_pack(p, w_curs, *curpos, &left);
1897 used = unpack_object_header_buffer(base, left, &type, sizep);
1898 if (!used) {
1899 type = OBJ_BAD;
1900 } else
1901 *curpos += used;
1902
1903 return type;
1904 }
1905
1906 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1907 {
1908 int type;
1909 struct revindex_entry *revidx;
1910 const unsigned char *sha1;
1911 revidx = find_pack_revindex(p, obj_offset);
1912 if (!revidx)
1913 return OBJ_BAD;
1914 sha1 = nth_packed_object_sha1(p, revidx->nr);
1915 mark_bad_packed_object(p, sha1);
1916 type = sha1_object_info(sha1, NULL);
1917 if (type <= OBJ_NONE)
1918 return OBJ_BAD;
1919 return type;
1920 }
1921
1922 #define POI_STACK_PREALLOC 64
1923
1924 static enum object_type packed_to_object_type(struct packed_git *p,
1925 off_t obj_offset,
1926 enum object_type type,
1927 struct pack_window **w_curs,
1928 off_t curpos)
1929 {
1930 off_t small_poi_stack[POI_STACK_PREALLOC];
1931 off_t *poi_stack = small_poi_stack;
1932 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1933
1934 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1935 off_t base_offset;
1936 unsigned long size;
1937 /* Push the object we're going to leave behind */
1938 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1939 poi_stack_alloc = alloc_nr(poi_stack_nr);
1940 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1941 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1942 } else {
1943 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1944 }
1945 poi_stack[poi_stack_nr++] = obj_offset;
1946 /* If parsing the base offset fails, just unwind */
1947 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1948 if (!base_offset)
1949 goto unwind;
1950 curpos = obj_offset = base_offset;
1951 type = unpack_object_header(p, w_curs, &curpos, &size);
1952 if (type <= OBJ_NONE) {
1953 /* If getting the base itself fails, we first
1954 * retry the base, otherwise unwind */
1955 type = retry_bad_packed_offset(p, base_offset);
1956 if (type > OBJ_NONE)
1957 goto out;
1958 goto unwind;
1959 }
1960 }
1961
1962 switch (type) {
1963 case OBJ_BAD:
1964 case OBJ_COMMIT:
1965 case OBJ_TREE:
1966 case OBJ_BLOB:
1967 case OBJ_TAG:
1968 break;
1969 default:
1970 error("unknown object type %i at offset %"PRIuMAX" in %s",
1971 type, (uintmax_t)obj_offset, p->pack_name);
1972 type = OBJ_BAD;
1973 }
1974
1975 out:
1976 if (poi_stack != small_poi_stack)
1977 free(poi_stack);
1978 return type;
1979
1980 unwind:
1981 while (poi_stack_nr) {
1982 obj_offset = poi_stack[--poi_stack_nr];
1983 type = retry_bad_packed_offset(p, obj_offset);
1984 if (type > OBJ_NONE)
1985 goto out;
1986 }
1987 type = OBJ_BAD;
1988 goto out;
1989 }
1990
1991 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1992 struct object_info *oi)
1993 {
1994 struct pack_window *w_curs = NULL;
1995 unsigned long size;
1996 off_t curpos = obj_offset;
1997 enum object_type type;
1998
1999 /*
2000 * We always get the representation type, but only convert it to
2001 * a "real" type later if the caller is interested.
2002 */
2003 type = unpack_object_header(p, &w_curs, &curpos, &size);
2004
2005 if (oi->sizep) {
2006 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2007 off_t tmp_pos = curpos;
2008 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
2009 type, obj_offset);
2010 if (!base_offset) {
2011 type = OBJ_BAD;
2012 goto out;
2013 }
2014 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2015 if (*oi->sizep == 0) {
2016 type = OBJ_BAD;
2017 goto out;
2018 }
2019 } else {
2020 *oi->sizep = size;
2021 }
2022 }
2023
2024 if (oi->disk_sizep) {
2025 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2026 *oi->disk_sizep = revidx[1].offset - obj_offset;
2027 }
2028
2029 if (oi->typep) {
2030 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
2031 if (*oi->typep < 0) {
2032 type = OBJ_BAD;
2033 goto out;
2034 }
2035 }
2036
2037 if (oi->delta_base_sha1) {
2038 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2039 const unsigned char *base;
2040
2041 base = get_delta_base_sha1(p, &w_curs, curpos,
2042 type, obj_offset);
2043 if (!base) {
2044 type = OBJ_BAD;
2045 goto out;
2046 }
2047
2048 hashcpy(oi->delta_base_sha1, base);
2049 } else
2050 hashclr(oi->delta_base_sha1);
2051 }
2052
2053 out:
2054 unuse_pack(&w_curs);
2055 return type;
2056 }
2057
2058 static void *unpack_compressed_entry(struct packed_git *p,
2059 struct pack_window **w_curs,
2060 off_t curpos,
2061 unsigned long size)
2062 {
2063 int st;
2064 git_zstream stream;
2065 unsigned char *buffer, *in;
2066
2067 buffer = xmallocz_gently(size);
2068 if (!buffer)
2069 return NULL;
2070 memset(&stream, 0, sizeof(stream));
2071 stream.next_out = buffer;
2072 stream.avail_out = size + 1;
2073
2074 git_inflate_init(&stream);
2075 do {
2076 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2077 stream.next_in = in;
2078 st = git_inflate(&stream, Z_FINISH);
2079 if (!stream.avail_out)
2080 break; /* the payload is larger than it should be */
2081 curpos += stream.next_in - in;
2082 } while (st == Z_OK || st == Z_BUF_ERROR);
2083 git_inflate_end(&stream);
2084 if ((st != Z_STREAM_END) || stream.total_out != size) {
2085 free(buffer);
2086 return NULL;
2087 }
2088
2089 return buffer;
2090 }
2091
2092 #define MAX_DELTA_CACHE (256)
2093
2094 static size_t delta_base_cached;
2095
2096 static struct delta_base_cache_lru_list {
2097 struct delta_base_cache_lru_list *prev;
2098 struct delta_base_cache_lru_list *next;
2099 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
2100
2101 static struct delta_base_cache_entry {
2102 struct delta_base_cache_lru_list lru;
2103 void *data;
2104 struct packed_git *p;
2105 off_t base_offset;
2106 unsigned long size;
2107 enum object_type type;
2108 } delta_base_cache[MAX_DELTA_CACHE];
2109
2110 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
2111 {
2112 unsigned long hash;
2113
2114 hash = (unsigned long)p + (unsigned long)base_offset;
2115 hash += (hash >> 8) + (hash >> 16);
2116 return hash % MAX_DELTA_CACHE;
2117 }
2118
2119 static struct delta_base_cache_entry *
2120 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2121 {
2122 unsigned long hash = pack_entry_hash(p, base_offset);
2123 return delta_base_cache + hash;
2124 }
2125
2126 static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
2127 struct packed_git *p, off_t base_offset)
2128 {
2129 return (ent->data && ent->p == p && ent->base_offset == base_offset);
2130 }
2131
2132 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2133 {
2134 struct delta_base_cache_entry *ent;
2135 ent = get_delta_base_cache_entry(p, base_offset);
2136 return eq_delta_base_cache_entry(ent, p, base_offset);
2137 }
2138
2139 static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2140 {
2141 ent->data = NULL;
2142 ent->lru.next->prev = ent->lru.prev;
2143 ent->lru.prev->next = ent->lru.next;
2144 delta_base_cached -= ent->size;
2145 }
2146
2147 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2148 unsigned long *base_size, enum object_type *type, int keep_cache)
2149 {
2150 struct delta_base_cache_entry *ent;
2151 void *ret;
2152
2153 ent = get_delta_base_cache_entry(p, base_offset);
2154
2155 if (!eq_delta_base_cache_entry(ent, p, base_offset))
2156 return unpack_entry(p, base_offset, type, base_size);
2157
2158 ret = ent->data;
2159
2160 if (!keep_cache)
2161 clear_delta_base_cache_entry(ent);
2162 else
2163 ret = xmemdupz(ent->data, ent->size);
2164 *type = ent->type;
2165 *base_size = ent->size;
2166 return ret;
2167 }
2168
2169 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2170 {
2171 if (ent->data) {
2172 free(ent->data);
2173 ent->data = NULL;
2174 ent->lru.next->prev = ent->lru.prev;
2175 ent->lru.prev->next = ent->lru.next;
2176 delta_base_cached -= ent->size;
2177 }
2178 }
2179
2180 void clear_delta_base_cache(void)
2181 {
2182 unsigned long p;
2183 for (p = 0; p < MAX_DELTA_CACHE; p++)
2184 release_delta_base_cache(&delta_base_cache[p]);
2185 }
2186
2187 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2188 void *base, unsigned long base_size, enum object_type type)
2189 {
2190 unsigned long hash = pack_entry_hash(p, base_offset);
2191 struct delta_base_cache_entry *ent = delta_base_cache + hash;
2192 struct delta_base_cache_lru_list *lru;
2193
2194 release_delta_base_cache(ent);
2195 delta_base_cached += base_size;
2196
2197 for (lru = delta_base_cache_lru.next;
2198 delta_base_cached > delta_base_cache_limit
2199 && lru != &delta_base_cache_lru;
2200 lru = lru->next) {
2201 struct delta_base_cache_entry *f = (void *)lru;
2202 if (f->type == OBJ_BLOB)
2203 release_delta_base_cache(f);
2204 }
2205 for (lru = delta_base_cache_lru.next;
2206 delta_base_cached > delta_base_cache_limit
2207 && lru != &delta_base_cache_lru;
2208 lru = lru->next) {
2209 struct delta_base_cache_entry *f = (void *)lru;
2210 release_delta_base_cache(f);
2211 }
2212
2213 ent->p = p;
2214 ent->base_offset = base_offset;
2215 ent->type = type;
2216 ent->data = base;
2217 ent->size = base_size;
2218 ent->lru.next = &delta_base_cache_lru;
2219 ent->lru.prev = delta_base_cache_lru.prev;
2220 delta_base_cache_lru.prev->next = &ent->lru;
2221 delta_base_cache_lru.prev = &ent->lru;
2222 }
2223
2224 static void *read_object(const unsigned char *sha1, enum object_type *type,
2225 unsigned long *size);
2226
2227 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2228 {
2229 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2230 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2231 p->pack_name, (uintmax_t)obj_offset);
2232 }
2233
2234 int do_check_packed_object_crc;
2235
2236 #define UNPACK_ENTRY_STACK_PREALLOC 64
2237 struct unpack_entry_stack_ent {
2238 off_t obj_offset;
2239 off_t curpos;
2240 unsigned long size;
2241 };
2242
2243 void *unpack_entry(struct packed_git *p, off_t obj_offset,
2244 enum object_type *final_type, unsigned long *final_size)
2245 {
2246 struct pack_window *w_curs = NULL;
2247 off_t curpos = obj_offset;
2248 void *data = NULL;
2249 unsigned long size;
2250 enum object_type type;
2251 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2252 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2253 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2254 int base_from_cache = 0;
2255
2256 write_pack_access_log(p, obj_offset);
2257
2258 /* PHASE 1: drill down to the innermost base object */
2259 for (;;) {
2260 off_t base_offset;
2261 int i;
2262 struct delta_base_cache_entry *ent;
2263
2264 ent = get_delta_base_cache_entry(p, curpos);
2265 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2266 type = ent->type;
2267 data = ent->data;
2268 size = ent->size;
2269 clear_delta_base_cache_entry(ent);
2270 base_from_cache = 1;
2271 break;
2272 }
2273
2274 if (do_check_packed_object_crc && p->index_version > 1) {
2275 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2276 unsigned long len = revidx[1].offset - obj_offset;
2277 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2278 const unsigned char *sha1 =
2279 nth_packed_object_sha1(p, revidx->nr);
2280 error("bad packed object CRC for %s",
2281 sha1_to_hex(sha1));
2282 mark_bad_packed_object(p, sha1);
2283 unuse_pack(&w_curs);
2284 return NULL;
2285 }
2286 }
2287
2288 type = unpack_object_header(p, &w_curs, &curpos, &size);
2289 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2290 break;
2291
2292 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2293 if (!base_offset) {
2294 error("failed to validate delta base reference "
2295 "at offset %"PRIuMAX" from %s",
2296 (uintmax_t)curpos, p->pack_name);
2297 /* bail to phase 2, in hopes of recovery */
2298 data = NULL;
2299 break;
2300 }
2301
2302 /* push object, proceed to base */
2303 if (delta_stack_nr >= delta_stack_alloc
2304 && delta_stack == small_delta_stack) {
2305 delta_stack_alloc = alloc_nr(delta_stack_nr);
2306 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2307 memcpy(delta_stack, small_delta_stack,
2308 sizeof(*delta_stack)*delta_stack_nr);
2309 } else {
2310 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2311 }
2312 i = delta_stack_nr++;
2313 delta_stack[i].obj_offset = obj_offset;
2314 delta_stack[i].curpos = curpos;
2315 delta_stack[i].size = size;
2316
2317 curpos = obj_offset = base_offset;
2318 }
2319
2320 /* PHASE 2: handle the base */
2321 switch (type) {
2322 case OBJ_OFS_DELTA:
2323 case OBJ_REF_DELTA:
2324 if (data)
2325 die("BUG in unpack_entry: left loop at a valid delta");
2326 break;
2327 case OBJ_COMMIT:
2328 case OBJ_TREE:
2329 case OBJ_BLOB:
2330 case OBJ_TAG:
2331 if (!base_from_cache)
2332 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2333 break;
2334 default:
2335 data = NULL;
2336 error("unknown object type %i at offset %"PRIuMAX" in %s",
2337 type, (uintmax_t)obj_offset, p->pack_name);
2338 }
2339
2340 /* PHASE 3: apply deltas in order */
2341
2342 /* invariants:
2343 * 'data' holds the base data, or NULL if there was corruption
2344 */
2345 while (delta_stack_nr) {
2346 void *delta_data;
2347 void *base = data;
2348 unsigned long delta_size, base_size = size;
2349 int i;
2350
2351 data = NULL;
2352
2353 if (base)
2354 add_delta_base_cache(p, obj_offset, base, base_size, type);
2355
2356 if (!base) {
2357 /*
2358 * We're probably in deep shit, but let's try to fetch
2359 * the required base anyway from another pack or loose.
2360 * This is costly but should happen only in the presence
2361 * of a corrupted pack, and is better than failing outright.
2362 */
2363 struct revindex_entry *revidx;
2364 const unsigned char *base_sha1;
2365 revidx = find_pack_revindex(p, obj_offset);
2366 if (revidx) {
2367 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2368 error("failed to read delta base object %s"
2369 " at offset %"PRIuMAX" from %s",
2370 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2371 p->pack_name);
2372 mark_bad_packed_object(p, base_sha1);
2373 base = read_object(base_sha1, &type, &base_size);
2374 }
2375 }
2376
2377 i = --delta_stack_nr;
2378 obj_offset = delta_stack[i].obj_offset;
2379 curpos = delta_stack[i].curpos;
2380 delta_size = delta_stack[i].size;
2381
2382 if (!base)
2383 continue;
2384
2385 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2386
2387 if (!delta_data) {
2388 error("failed to unpack compressed delta "
2389 "at offset %"PRIuMAX" from %s",
2390 (uintmax_t)curpos, p->pack_name);
2391 data = NULL;
2392 continue;
2393 }
2394
2395 data = patch_delta(base, base_size,
2396 delta_data, delta_size,
2397 &size);
2398
2399 /*
2400 * We could not apply the delta; warn the user, but keep going.
2401 * Our failure will be noticed either in the next iteration of
2402 * the loop, or if this is the final delta, in the caller when
2403 * we return NULL. Those code paths will take care of making
2404 * a more explicit warning and retrying with another copy of
2405 * the object.
2406 */
2407 if (!data)
2408 error("failed to apply delta");
2409
2410 free(delta_data);
2411 }
2412
2413 *final_type = type;
2414 *final_size = size;
2415
2416 unuse_pack(&w_curs);
2417
2418 if (delta_stack != small_delta_stack)
2419 free(delta_stack);
2420
2421 return data;
2422 }
2423
2424 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2425 uint32_t n)
2426 {
2427 const unsigned char *index = p->index_data;
2428 if (!index) {
2429 if (open_pack_index(p))
2430 return NULL;
2431 index = p->index_data;
2432 }
2433 if (n >= p->num_objects)
2434 return NULL;
2435 index += 4 * 256;
2436 if (p->index_version == 1) {
2437 return index + 24 * n + 4;
2438 } else {
2439 index += 8;
2440 return index + 20 * n;
2441 }
2442 }
2443
2444 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2445 {
2446 const unsigned char *index = p->index_data;
2447 index += 4 * 256;
2448 if (p->index_version == 1) {
2449 return ntohl(*((uint32_t *)(index + 24 * n)));
2450 } else {
2451 uint32_t off;
2452 index += 8 + p->num_objects * (20 + 4);
2453 off = ntohl(*((uint32_t *)(index + 4 * n)));
2454 if (!(off & 0x80000000))
2455 return off;
2456 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2457 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2458 ntohl(*((uint32_t *)(index + 4)));
2459 }
2460 }
2461
2462 off_t find_pack_entry_one(const unsigned char *sha1,
2463 struct packed_git *p)
2464 {
2465 const uint32_t *level1_ofs = p->index_data;
2466 const unsigned char *index = p->index_data;
2467 unsigned hi, lo, stride;
2468 static int use_lookup = -1;
2469 static int debug_lookup = -1;
2470
2471 if (debug_lookup < 0)
2472 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2473
2474 if (!index) {
2475 if (open_pack_index(p))
2476 return 0;
2477 level1_ofs = p->index_data;
2478 index = p->index_data;
2479 }
2480 if (p->index_version > 1) {
2481 level1_ofs += 2;
2482 index += 8;
2483 }
2484 index += 4 * 256;
2485 hi = ntohl(level1_ofs[*sha1]);
2486 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2487 if (p->index_version > 1) {
2488 stride = 20;
2489 } else {
2490 stride = 24;
2491 index += 4;
2492 }
2493
2494 if (debug_lookup)
2495 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2496 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2497
2498 if (use_lookup < 0)
2499 use_lookup = !!getenv("GIT_USE_LOOKUP");
2500 if (use_lookup) {
2501 int pos = sha1_entry_pos(index, stride, 0,
2502 lo, hi, p->num_objects, sha1);
2503 if (pos < 0)
2504 return 0;
2505 return nth_packed_object_offset(p, pos);
2506 }
2507
2508 do {
2509 unsigned mi = (lo + hi) / 2;
2510 int cmp = hashcmp(index + mi * stride, sha1);
2511
2512 if (debug_lookup)
2513 printf("lo %u hi %u rg %u mi %u\n",
2514 lo, hi, hi - lo, mi);
2515 if (!cmp)
2516 return nth_packed_object_offset(p, mi);
2517 if (cmp > 0)
2518 hi = mi;
2519 else
2520 lo = mi+1;
2521 } while (lo < hi);
2522 return 0;
2523 }
2524
2525 int is_pack_valid(struct packed_git *p)
2526 {
2527 /* An already open pack is known to be valid. */
2528 if (p->pack_fd != -1)
2529 return 1;
2530
2531 /* If the pack has one window completely covering the
2532 * file size, the pack is known to be valid even if
2533 * the descriptor is not currently open.
2534 */
2535 if (p->windows) {
2536 struct pack_window *w = p->windows;
2537
2538 if (!w->offset && w->len == p->pack_size)
2539 return 1;
2540 }
2541
2542 /* Force the pack to open to prove its valid. */
2543 return !open_packed_git(p);
2544 }
2545
2546 static int fill_pack_entry(const unsigned char *sha1,
2547 struct pack_entry *e,
2548 struct packed_git *p)
2549 {
2550 off_t offset;
2551
2552 if (p->num_bad_objects) {
2553 unsigned i;
2554 for (i = 0; i < p->num_bad_objects; i++)
2555 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2556 return 0;
2557 }
2558
2559 offset = find_pack_entry_one(sha1, p);
2560 if (!offset)
2561 return 0;
2562
2563 /*
2564 * We are about to tell the caller where they can locate the
2565 * requested object. We better make sure the packfile is
2566 * still here and can be accessed before supplying that
2567 * answer, as it may have been deleted since the index was
2568 * loaded!
2569 */
2570 if (!is_pack_valid(p))
2571 return 0;
2572 e->offset = offset;
2573 e->p = p;
2574 hashcpy(e->sha1, sha1);
2575 return 1;
2576 }
2577
2578 /*
2579 * Iff a pack file contains the object named by sha1, return true and
2580 * store its location to e.
2581 */
2582 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2583 {
2584 struct packed_git *p;
2585
2586 prepare_packed_git();
2587 if (!packed_git)
2588 return 0;
2589
2590 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2591 return 1;
2592
2593 for (p = packed_git; p; p = p->next) {
2594 if (p == last_found_pack)
2595 continue; /* we already checked this one */
2596
2597 if (fill_pack_entry(sha1, e, p)) {
2598 last_found_pack = p;
2599 return 1;
2600 }
2601 }
2602 return 0;
2603 }
2604
2605 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2606 struct packed_git *packs)
2607 {
2608 struct packed_git *p;
2609
2610 for (p = packs; p; p = p->next) {
2611 if (find_pack_entry_one(sha1, p))
2612 return p;
2613 }
2614 return NULL;
2615
2616 }
2617
2618 static int sha1_loose_object_info(const unsigned char *sha1,
2619 struct object_info *oi,
2620 int flags)
2621 {
2622 int status = 0;
2623 unsigned long mapsize;
2624 void *map;
2625 git_zstream stream;
2626 char hdr[32];
2627 struct strbuf hdrbuf = STRBUF_INIT;
2628
2629 if (oi->delta_base_sha1)
2630 hashclr(oi->delta_base_sha1);
2631
2632 /*
2633 * If we don't care about type or size, then we don't
2634 * need to look inside the object at all. Note that we
2635 * do not optimize out the stat call, even if the
2636 * caller doesn't care about the disk-size, since our
2637 * return value implicitly indicates whether the
2638 * object even exists.
2639 */
2640 if (!oi->typep && !oi->typename && !oi->sizep) {
2641 struct stat st;
2642 if (stat_sha1_file(sha1, &st) < 0)
2643 return -1;
2644 if (oi->disk_sizep)
2645 *oi->disk_sizep = st.st_size;
2646 return 0;
2647 }
2648
2649 map = map_sha1_file(sha1, &mapsize);
2650 if (!map)
2651 return -1;
2652 if (oi->disk_sizep)
2653 *oi->disk_sizep = mapsize;
2654 if ((flags & LOOKUP_UNKNOWN_OBJECT)) {
2655 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2656 status = error("unable to unpack %s header with --allow-unknown-type",
2657 sha1_to_hex(sha1));
2658 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2659 status = error("unable to unpack %s header",
2660 sha1_to_hex(sha1));
2661 if (status < 0)
2662 ; /* Do nothing */
2663 else if (hdrbuf.len) {
2664 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2665 status = error("unable to parse %s header with --allow-unknown-type",
2666 sha1_to_hex(sha1));
2667 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
2668 status = error("unable to parse %s header", sha1_to_hex(sha1));
2669 git_inflate_end(&stream);
2670 munmap(map, mapsize);
2671 if (status && oi->typep)
2672 *oi->typep = status;
2673 strbuf_release(&hdrbuf);
2674 return 0;
2675 }
2676
2677 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2678 {
2679 struct cached_object *co;
2680 struct pack_entry e;
2681 int rtype;
2682 enum object_type real_type;
2683 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2684
2685 co = find_cached_object(real);
2686 if (co) {
2687 if (oi->typep)
2688 *(oi->typep) = co->type;
2689 if (oi->sizep)
2690 *(oi->sizep) = co->size;
2691 if (oi->disk_sizep)
2692 *(oi->disk_sizep) = 0;
2693 if (oi->delta_base_sha1)
2694 hashclr(oi->delta_base_sha1);
2695 if (oi->typename)
2696 strbuf_addstr(oi->typename, typename(co->type));
2697 oi->whence = OI_CACHED;
2698 return 0;
2699 }
2700
2701 if (!find_pack_entry(real, &e)) {
2702 /* Most likely it's a loose object. */
2703 if (!sha1_loose_object_info(real, oi, flags)) {
2704 oi->whence = OI_LOOSE;
2705 return 0;
2706 }
2707
2708 /* Not a loose object; someone else may have just packed it. */
2709 reprepare_packed_git();
2710 if (!find_pack_entry(real, &e))
2711 return -1;
2712 }
2713
2714 /*
2715 * packed_object_info() does not follow the delta chain to
2716 * find out the real type, unless it is given oi->typep.
2717 */
2718 if (oi->typename && !oi->typep)
2719 oi->typep = &real_type;
2720
2721 rtype = packed_object_info(e.p, e.offset, oi);
2722 if (rtype < 0) {
2723 mark_bad_packed_object(e.p, real);
2724 if (oi->typep == &real_type)
2725 oi->typep = NULL;
2726 return sha1_object_info_extended(real, oi, 0);
2727 } else if (in_delta_base_cache(e.p, e.offset)) {
2728 oi->whence = OI_DBCACHED;
2729 } else {
2730 oi->whence = OI_PACKED;
2731 oi->u.packed.offset = e.offset;
2732 oi->u.packed.pack = e.p;
2733 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2734 rtype == OBJ_OFS_DELTA);
2735 }
2736 if (oi->typename)
2737 strbuf_addstr(oi->typename, typename(*oi->typep));
2738 if (oi->typep == &real_type)
2739 oi->typep = NULL;
2740
2741 return 0;
2742 }
2743
2744 /* returns enum object_type or negative */
2745 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2746 {
2747 enum object_type type;
2748 struct object_info oi = {NULL};
2749
2750 oi.typep = &type;
2751 oi.sizep = sizep;
2752 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2753 return -1;
2754 return type;
2755 }
2756
2757 static void *read_packed_sha1(const unsigned char *sha1,
2758 enum object_type *type, unsigned long *size)
2759 {
2760 struct pack_entry e;
2761 void *data;
2762
2763 if (!find_pack_entry(sha1, &e))
2764 return NULL;
2765 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2766 if (!data) {
2767 /*
2768 * We're probably in deep shit, but let's try to fetch
2769 * the required object anyway from another pack or loose.
2770 * This should happen only in the presence of a corrupted
2771 * pack, and is better than failing outright.
2772 */
2773 error("failed to read object %s at offset %"PRIuMAX" from %s",
2774 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2775 mark_bad_packed_object(e.p, sha1);
2776 data = read_object(sha1, type, size);
2777 }
2778 return data;
2779 }
2780
2781 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2782 unsigned char *sha1)
2783 {
2784 struct cached_object *co;
2785
2786 hash_sha1_file(buf, len, typename(type), sha1);
2787 if (has_sha1_file(sha1) || find_cached_object(sha1))
2788 return 0;
2789 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2790 co = &cached_objects[cached_object_nr++];
2791 co->size = len;
2792 co->type = type;
2793 co->buf = xmalloc(len);
2794 memcpy(co->buf, buf, len);
2795 hashcpy(co->sha1, sha1);
2796 return 0;
2797 }
2798
2799 static void *read_object(const unsigned char *sha1, enum object_type *type,
2800 unsigned long *size)
2801 {
2802 unsigned long mapsize;
2803 void *map, *buf;
2804 struct cached_object *co;
2805
2806 co = find_cached_object(sha1);
2807 if (co) {
2808 *type = co->type;
2809 *size = co->size;
2810 return xmemdupz(co->buf, co->size);
2811 }
2812
2813 buf = read_packed_sha1(sha1, type, size);
2814 if (buf)
2815 return buf;
2816 map = map_sha1_file(sha1, &mapsize);
2817 if (map) {
2818 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2819 munmap(map, mapsize);
2820 return buf;
2821 }
2822 reprepare_packed_git();
2823 return read_packed_sha1(sha1, type, size);
2824 }
2825
2826 /*
2827 * This function dies on corrupt objects; the callers who want to
2828 * deal with them should arrange to call read_object() and give error
2829 * messages themselves.
2830 */
2831 void *read_sha1_file_extended(const unsigned char *sha1,
2832 enum object_type *type,
2833 unsigned long *size,
2834 unsigned flag)
2835 {
2836 void *data;
2837 const struct packed_git *p;
2838 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
2839
2840 errno = 0;
2841 data = read_object(repl, type, size);
2842 if (data)
2843 return data;
2844
2845 if (errno && errno != ENOENT)
2846 die_errno("failed to read object %s", sha1_to_hex(sha1));
2847
2848 /* die if we replaced an object with one that does not exist */
2849 if (repl != sha1)
2850 die("replacement %s not found for %s",
2851 sha1_to_hex(repl), sha1_to_hex(sha1));
2852
2853 if (has_loose_object(repl)) {
2854 const char *path = sha1_file_name(sha1);
2855
2856 die("loose object %s (stored in %s) is corrupt",
2857 sha1_to_hex(repl), path);
2858 }
2859
2860 if ((p = has_packed_and_bad(repl)) != NULL)
2861 die("packed object %s (stored in %s) is corrupt",
2862 sha1_to_hex(repl), p->pack_name);
2863
2864 return NULL;
2865 }
2866
2867 void *read_object_with_reference(const unsigned char *sha1,
2868 const char *required_type_name,
2869 unsigned long *size,
2870 unsigned char *actual_sha1_return)
2871 {
2872 enum object_type type, required_type;
2873 void *buffer;
2874 unsigned long isize;
2875 unsigned char actual_sha1[20];
2876
2877 required_type = type_from_string(required_type_name);
2878 hashcpy(actual_sha1, sha1);
2879 while (1) {
2880 int ref_length = -1;
2881 const char *ref_type = NULL;
2882
2883 buffer = read_sha1_file(actual_sha1, &type, &isize);
2884 if (!buffer)
2885 return NULL;
2886 if (type == required_type) {
2887 *size = isize;
2888 if (actual_sha1_return)
2889 hashcpy(actual_sha1_return, actual_sha1);
2890 return buffer;
2891 }
2892 /* Handle references */
2893 else if (type == OBJ_COMMIT)
2894 ref_type = "tree ";
2895 else if (type == OBJ_TAG)
2896 ref_type = "object ";
2897 else {
2898 free(buffer);
2899 return NULL;
2900 }
2901 ref_length = strlen(ref_type);
2902
2903 if (ref_length + 40 > isize ||
2904 memcmp(buffer, ref_type, ref_length) ||
2905 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2906 free(buffer);
2907 return NULL;
2908 }
2909 free(buffer);
2910 /* Now we have the ID of the referred-to object in
2911 * actual_sha1. Check again. */
2912 }
2913 }
2914
2915 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2916 const char *type, unsigned char *sha1,
2917 char *hdr, int *hdrlen)
2918 {
2919 git_SHA_CTX c;
2920
2921 /* Generate the header */
2922 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2923
2924 /* Sha1.. */
2925 git_SHA1_Init(&c);
2926 git_SHA1_Update(&c, hdr, *hdrlen);
2927 git_SHA1_Update(&c, buf, len);
2928 git_SHA1_Final(sha1, &c);
2929 }
2930
2931 /*
2932 * Move the just written object into its final resting place.
2933 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2934 * "moving" is only a part of what it does, when no patch between
2935 * master to pu changes the call sites of this function.
2936 */
2937 int move_temp_to_file(const char *tmpfile, const char *filename)
2938 {
2939 int ret = 0;
2940
2941 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2942 goto try_rename;
2943 else if (link(tmpfile, filename))
2944 ret = errno;
2945
2946 /*
2947 * Coda hack - coda doesn't like cross-directory links,
2948 * so we fall back to a rename, which will mean that it
2949 * won't be able to check collisions, but that's not a
2950 * big deal.
2951 *
2952 * The same holds for FAT formatted media.
2953 *
2954 * When this succeeds, we just return. We have nothing
2955 * left to unlink.
2956 */
2957 if (ret && ret != EEXIST) {
2958 try_rename:
2959 if (!rename(tmpfile, filename))
2960 goto out;
2961 ret = errno;
2962 }
2963 unlink_or_warn(tmpfile);
2964 if (ret) {
2965 if (ret != EEXIST) {
2966 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2967 }
2968 /* FIXME!!! Collision check here ? */
2969 }
2970
2971 out:
2972 if (adjust_shared_perm(filename))
2973 return error("unable to set permission to '%s'", filename);
2974 return 0;
2975 }
2976
2977 static int write_buffer(int fd, const void *buf, size_t len)
2978 {
2979 if (write_in_full(fd, buf, len) < 0)
2980 return error("file write error (%s)", strerror(errno));
2981 return 0;
2982 }
2983
2984 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2985 unsigned char *sha1)
2986 {
2987 char hdr[32];
2988 int hdrlen;
2989 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2990 return 0;
2991 }
2992
2993 /* Finalize a file on disk, and close it. */
2994 static void close_sha1_file(int fd)
2995 {
2996 if (fsync_object_files)
2997 fsync_or_die(fd, "sha1 file");
2998 if (close(fd) != 0)
2999 die_errno("error when closing sha1 file");
3000 }
3001
3002 /* Size of directory component, including the ending '/' */
3003 static inline int directory_size(const char *filename)
3004 {
3005 const char *s = strrchr(filename, '/');
3006 if (!s)
3007 return 0;
3008 return s - filename + 1;
3009 }
3010
3011 /*
3012 * This creates a temporary file in the same directory as the final
3013 * 'filename'
3014 *
3015 * We want to avoid cross-directory filename renames, because those
3016 * can have problems on various filesystems (FAT, NFS, Coda).
3017 */
3018 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
3019 {
3020 int fd, dirlen = directory_size(filename);
3021
3022 if (dirlen + 20 > bufsiz) {
3023 errno = ENAMETOOLONG;
3024 return -1;
3025 }
3026 memcpy(buffer, filename, dirlen);
3027 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
3028 fd = git_mkstemp_mode(buffer, 0444);
3029 if (fd < 0 && dirlen && errno == ENOENT) {
3030 /* Make sure the directory exists */
3031 memcpy(buffer, filename, dirlen);
3032 buffer[dirlen-1] = 0;
3033 if (mkdir(buffer, 0777) && errno != EEXIST)
3034 return -1;
3035 if (adjust_shared_perm(buffer))
3036 return -1;
3037
3038 /* Try again */
3039 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
3040 fd = git_mkstemp_mode(buffer, 0444);
3041 }
3042 return fd;
3043 }
3044
3045 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
3046 const void *buf, unsigned long len, time_t mtime)
3047 {
3048 int fd, ret;
3049 unsigned char compressed[4096];
3050 git_zstream stream;
3051 git_SHA_CTX c;
3052 unsigned char parano_sha1[20];
3053 static char tmp_file[PATH_MAX];
3054 const char *filename = sha1_file_name(sha1);
3055
3056 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
3057 if (fd < 0) {
3058 if (errno == EACCES)
3059 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3060 else
3061 return error("unable to create temporary file: %s", strerror(errno));
3062 }
3063
3064 /* Set it up */
3065 git_deflate_init(&stream, zlib_compression_level);
3066 stream.next_out = compressed;
3067 stream.avail_out = sizeof(compressed);
3068 git_SHA1_Init(&c);
3069
3070 /* First header.. */
3071 stream.next_in = (unsigned char *)hdr;
3072 stream.avail_in = hdrlen;
3073 while (git_deflate(&stream, 0) == Z_OK)
3074 ; /* nothing */
3075 git_SHA1_Update(&c, hdr, hdrlen);
3076
3077 /* Then the data itself.. */
3078 stream.next_in = (void *)buf;
3079 stream.avail_in = len;
3080 do {
3081 unsigned char *in0 = stream.next_in;
3082 ret = git_deflate(&stream, Z_FINISH);
3083 git_SHA1_Update(&c, in0, stream.next_in - in0);
3084 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3085 die("unable to write sha1 file");
3086 stream.next_out = compressed;
3087 stream.avail_out = sizeof(compressed);
3088 } while (ret == Z_OK);
3089
3090 if (ret != Z_STREAM_END)
3091 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
3092 ret = git_deflate_end_gently(&stream);
3093 if (ret != Z_OK)
3094 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
3095 git_SHA1_Final(parano_sha1, &c);
3096 if (hashcmp(sha1, parano_sha1) != 0)
3097 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
3098
3099 close_sha1_file(fd);
3100
3101 if (mtime) {
3102 struct utimbuf utb;
3103 utb.actime = mtime;
3104 utb.modtime = mtime;
3105 if (utime(tmp_file, &utb) < 0)
3106 warning("failed utime() on %s: %s",
3107 tmp_file, strerror(errno));
3108 }
3109
3110 return move_temp_to_file(tmp_file, filename);
3111 }
3112
3113 static int freshen_loose_object(const unsigned char *sha1)
3114 {
3115 return check_and_freshen(sha1, 1);
3116 }
3117
3118 static int freshen_packed_object(const unsigned char *sha1)
3119 {
3120 struct pack_entry e;
3121 if (!find_pack_entry(sha1, &e))
3122 return 0;
3123 if (e.p->freshened)
3124 return 1;
3125 if (!freshen_file(e.p->pack_name))
3126 return 0;
3127 e.p->freshened = 1;
3128 return 1;
3129 }
3130
3131 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3132 {
3133 char hdr[32];
3134 int hdrlen;
3135
3136 /* Normally if we have it in the pack then we do not bother writing
3137 * it out into .git/objects/??/?{38} file.
3138 */
3139 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3140 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3141 return 0;
3142 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3143 }
3144
3145 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3146 unsigned char *sha1, unsigned flags)
3147 {
3148 char *header;
3149 int hdrlen, status = 0;
3150
3151 /* type string, SP, %lu of the length plus NUL must fit this */
3152 header = xmalloc(strlen(type) + 32);
3153 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3154
3155 if (!(flags & HASH_WRITE_OBJECT))
3156 goto cleanup;
3157 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3158 goto cleanup;
3159 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3160
3161 cleanup:
3162 free(header);
3163 return status;
3164 }
3165
3166 int force_object_loose(const unsigned char *sha1, time_t mtime)
3167 {
3168 void *buf;
3169 unsigned long len;
3170 enum object_type type;
3171 char hdr[32];
3172 int hdrlen;
3173 int ret;
3174
3175 if (has_loose_object(sha1))
3176 return 0;
3177 buf = read_packed_sha1(sha1, &type, &len);
3178 if (!buf)
3179 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3180 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
3181 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3182 free(buf);
3183
3184 return ret;
3185 }
3186
3187 int has_pack_index(const unsigned char *sha1)
3188 {
3189 struct stat st;
3190 if (stat(sha1_pack_index_name(sha1), &st))
3191 return 0;
3192 return 1;
3193 }
3194
3195 int has_sha1_pack(const unsigned char *sha1)
3196 {
3197 struct pack_entry e;
3198 return find_pack_entry(sha1, &e);
3199 }
3200
3201 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3202 {
3203 struct pack_entry e;
3204
3205 if (find_pack_entry(sha1, &e))
3206 return 1;
3207 if (has_loose_object(sha1))
3208 return 1;
3209 if (flags & HAS_SHA1_QUICK)
3210 return 0;
3211 reprepare_packed_git();
3212 return find_pack_entry(sha1, &e);
3213 }
3214
3215 static void check_tree(const void *buf, size_t size)
3216 {
3217 struct tree_desc desc;
3218 struct name_entry entry;
3219
3220 init_tree_desc(&desc, buf, size);
3221 while (tree_entry(&desc, &entry))
3222 /* do nothing
3223 * tree_entry() will die() on malformed entries */
3224 ;
3225 }
3226
3227 static void check_commit(const void *buf, size_t size)
3228 {
3229 struct commit c;
3230 memset(&c, 0, sizeof(c));
3231 if (parse_commit_buffer(&c, buf, size))
3232 die("corrupt commit");
3233 }
3234
3235 static void check_tag(const void *buf, size_t size)
3236 {
3237 struct tag t;
3238 memset(&t, 0, sizeof(t));
3239 if (parse_tag_buffer(&t, buf, size))
3240 die("corrupt tag");
3241 }
3242
3243 static int index_mem(unsigned char *sha1, void *buf, size_t size,
3244 enum object_type type,
3245 const char *path, unsigned flags)
3246 {
3247 int ret, re_allocated = 0;
3248 int write_object = flags & HASH_WRITE_OBJECT;
3249
3250 if (!type)
3251 type = OBJ_BLOB;
3252
3253 /*
3254 * Convert blobs to git internal format
3255 */
3256 if ((type == OBJ_BLOB) && path) {
3257 struct strbuf nbuf = STRBUF_INIT;
3258 if (convert_to_git(path, buf, size, &nbuf,
3259 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3260 buf = strbuf_detach(&nbuf, &size);
3261 re_allocated = 1;
3262 }
3263 }
3264 if (flags & HASH_FORMAT_CHECK) {
3265 if (type == OBJ_TREE)
3266 check_tree(buf, size);
3267 if (type == OBJ_COMMIT)
3268 check_commit(buf, size);
3269 if (type == OBJ_TAG)
3270 check_tag(buf, size);
3271 }
3272
3273 if (write_object)
3274 ret = write_sha1_file(buf, size, typename(type), sha1);
3275 else
3276 ret = hash_sha1_file(buf, size, typename(type), sha1);
3277 if (re_allocated)
3278 free(buf);
3279 return ret;
3280 }
3281
3282 static int index_stream_convert_blob(unsigned char *sha1, int fd,
3283 const char *path, unsigned flags)
3284 {
3285 int ret;
3286 const int write_object = flags & HASH_WRITE_OBJECT;
3287 struct strbuf sbuf = STRBUF_INIT;
3288
3289 assert(path);
3290 assert(would_convert_to_git_filter_fd(path));
3291
3292 convert_to_git_filter_fd(path, fd, &sbuf,
3293 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3294
3295 if (write_object)
3296 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3297 sha1);
3298 else
3299 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3300 sha1);
3301 strbuf_release(&sbuf);
3302 return ret;
3303 }
3304
3305 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3306 const char *path, unsigned flags)
3307 {
3308 struct strbuf sbuf = STRBUF_INIT;
3309 int ret;
3310
3311 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3312 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3313 else
3314 ret = -1;
3315 strbuf_release(&sbuf);
3316 return ret;
3317 }
3318
3319 #define SMALL_FILE_SIZE (32*1024)
3320
3321 static int index_core(unsigned char *sha1, int fd, size_t size,
3322 enum object_type type, const char *path,
3323 unsigned flags)
3324 {
3325 int ret;
3326
3327 if (!size) {
3328 ret = index_mem(sha1, "", size, type, path, flags);
3329 } else if (size <= SMALL_FILE_SIZE) {
3330 char *buf = xmalloc(size);
3331 if (size == read_in_full(fd, buf, size))
3332 ret = index_mem(sha1, buf, size, type, path, flags);
3333 else
3334 ret = error("short read %s", strerror(errno));
3335 free(buf);
3336 } else {
3337 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3338 ret = index_mem(sha1, buf, size, type, path, flags);
3339 munmap(buf, size);
3340 }
3341 return ret;
3342 }
3343
3344 /*
3345 * This creates one packfile per large blob unless bulk-checkin
3346 * machinery is "plugged".
3347 *
3348 * This also bypasses the usual "convert-to-git" dance, and that is on
3349 * purpose. We could write a streaming version of the converting
3350 * functions and insert that before feeding the data to fast-import
3351 * (or equivalent in-core API described above). However, that is
3352 * somewhat complicated, as we do not know the size of the filter
3353 * result, which we need to know beforehand when writing a git object.
3354 * Since the primary motivation for trying to stream from the working
3355 * tree file and to avoid mmaping it in core is to deal with large
3356 * binary blobs, they generally do not want to get any conversion, and
3357 * callers should avoid this code path when filters are requested.
3358 */
3359 static int index_stream(unsigned char *sha1, int fd, size_t size,
3360 enum object_type type, const char *path,
3361 unsigned flags)
3362 {
3363 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3364 }
3365
3366 int index_fd(unsigned char *sha1, int fd, struct stat *st,
3367 enum object_type type, const char *path, unsigned flags)
3368 {
3369 int ret;
3370
3371 /*
3372 * Call xsize_t() only when needed to avoid potentially unnecessary
3373 * die() for large files.
3374 */
3375 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3376 ret = index_stream_convert_blob(sha1, fd, path, flags);
3377 else if (!S_ISREG(st->st_mode))
3378 ret = index_pipe(sha1, fd, type, path, flags);
3379 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3380 (path && would_convert_to_git(path)))
3381 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3382 flags);
3383 else
3384 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3385 flags);
3386 close(fd);
3387 return ret;
3388 }
3389
3390 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3391 {
3392 int fd;
3393 struct strbuf sb = STRBUF_INIT;
3394
3395 switch (st->st_mode & S_IFMT) {
3396 case S_IFREG:
3397 fd = open(path, O_RDONLY);
3398 if (fd < 0)
3399 return error("open(\"%s\"): %s", path,
3400 strerror(errno));
3401 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3402 return error("%s: failed to insert into database",
3403 path);
3404 break;
3405 case S_IFLNK:
3406 if (strbuf_readlink(&sb, path, st->st_size)) {
3407 char *errstr = strerror(errno);
3408 return error("readlink(\"%s\"): %s", path,
3409 errstr);
3410 }
3411 if (!(flags & HASH_WRITE_OBJECT))
3412 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3413 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3414 return error("%s: failed to insert into database",
3415 path);
3416 strbuf_release(&sb);
3417 break;
3418 case S_IFDIR:
3419 return resolve_gitlink_ref(path, "HEAD", sha1);
3420 default:
3421 return error("%s: unsupported file type", path);
3422 }
3423 return 0;
3424 }
3425
3426 int read_pack_header(int fd, struct pack_header *header)
3427 {
3428 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3429 /* "eof before pack header was fully read" */
3430 return PH_ERROR_EOF;
3431
3432 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3433 /* "protocol error (pack signature mismatch detected)" */
3434 return PH_ERROR_PACK_SIGNATURE;
3435 if (!pack_version_ok(header->hdr_version))
3436 /* "protocol error (pack version unsupported)" */
3437 return PH_ERROR_PROTOCOL;
3438 return 0;
3439 }
3440
3441 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3442 {
3443 enum object_type type = sha1_object_info(sha1, NULL);
3444 if (type < 0)
3445 die("%s is not a valid object", sha1_to_hex(sha1));
3446 if (type != expect)
3447 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3448 typename(expect));
3449 }
3450
3451 static int for_each_file_in_obj_subdir(int subdir_nr,
3452 struct strbuf *path,
3453 each_loose_object_fn obj_cb,
3454 each_loose_cruft_fn cruft_cb,
3455 each_loose_subdir_fn subdir_cb,
3456 void *data)
3457 {
3458 size_t baselen = path->len;
3459 DIR *dir = opendir(path->buf);
3460 struct dirent *de;
3461 int r = 0;
3462
3463 if (!dir) {
3464 if (errno == ENOENT)
3465 return 0;
3466 return error("unable to open %s: %s", path->buf, strerror(errno));
3467 }
3468
3469 while ((de = readdir(dir))) {
3470 if (is_dot_or_dotdot(de->d_name))
3471 continue;
3472
3473 strbuf_setlen(path, baselen);
3474 strbuf_addf(path, "/%s", de->d_name);
3475
3476 if (strlen(de->d_name) == 38) {
3477 char hex[41];
3478 unsigned char sha1[20];
3479
3480 snprintf(hex, sizeof(hex), "%02x%s",
3481 subdir_nr, de->d_name);
3482 if (!get_sha1_hex(hex, sha1)) {
3483 if (obj_cb) {
3484 r = obj_cb(sha1, path->buf, data);
3485 if (r)
3486 break;
3487 }
3488 continue;
3489 }
3490 }
3491
3492 if (cruft_cb) {
3493 r = cruft_cb(de->d_name, path->buf, data);
3494 if (r)
3495 break;
3496 }
3497 }
3498 strbuf_setlen(path, baselen);
3499
3500 if (!r && subdir_cb)
3501 r = subdir_cb(subdir_nr, path->buf, data);
3502
3503 closedir(dir);
3504 return r;
3505 }
3506
3507 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3508 each_loose_object_fn obj_cb,
3509 each_loose_cruft_fn cruft_cb,
3510 each_loose_subdir_fn subdir_cb,
3511 void *data)
3512 {
3513 size_t baselen = path->len;
3514 int r = 0;
3515 int i;
3516
3517 for (i = 0; i < 256; i++) {
3518 strbuf_addf(path, "/%02x", i);
3519 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3520 subdir_cb, data);
3521 strbuf_setlen(path, baselen);
3522 if (r)
3523 break;
3524 }
3525
3526 return r;
3527 }
3528
3529 int for_each_loose_file_in_objdir(const char *path,
3530 each_loose_object_fn obj_cb,
3531 each_loose_cruft_fn cruft_cb,
3532 each_loose_subdir_fn subdir_cb,
3533 void *data)
3534 {
3535 struct strbuf buf = STRBUF_INIT;
3536 int r;
3537
3538 strbuf_addstr(&buf, path);
3539 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3540 subdir_cb, data);
3541 strbuf_release(&buf);
3542
3543 return r;
3544 }
3545
3546 struct loose_alt_odb_data {
3547 each_loose_object_fn *cb;
3548 void *data;
3549 };
3550
3551 static int loose_from_alt_odb(struct alternate_object_database *alt,
3552 void *vdata)
3553 {
3554 struct loose_alt_odb_data *data = vdata;
3555 struct strbuf buf = STRBUF_INIT;
3556 int r;
3557
3558 /* copy base not including trailing '/' */
3559 strbuf_add(&buf, alt->base, alt->name - alt->base - 1);
3560 r = for_each_loose_file_in_objdir_buf(&buf,
3561 data->cb, NULL, NULL,
3562 data->data);
3563 strbuf_release(&buf);
3564 return r;
3565 }
3566
3567 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3568 {
3569 struct loose_alt_odb_data alt;
3570 int r;
3571
3572 r = for_each_loose_file_in_objdir(get_object_directory(),
3573 cb, NULL, NULL, data);
3574 if (r)
3575 return r;
3576
3577 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3578 return 0;
3579
3580 alt.cb = cb;
3581 alt.data = data;
3582 return foreach_alt_odb(loose_from_alt_odb, &alt);
3583 }
3584
3585 static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3586 {
3587 uint32_t i;
3588 int r = 0;
3589
3590 for (i = 0; i < p->num_objects; i++) {
3591 const unsigned char *sha1 = nth_packed_object_sha1(p, i);
3592
3593 if (!sha1)
3594 return error("unable to get sha1 of object %u in %s",
3595 i, p->pack_name);
3596
3597 r = cb(sha1, p, i, data);
3598 if (r)
3599 break;
3600 }
3601 return r;
3602 }
3603
3604 int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3605 {
3606 struct packed_git *p;
3607 int r = 0;
3608 int pack_errors = 0;
3609
3610 prepare_packed_git();
3611 for (p = packed_git; p; p = p->next) {
3612 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3613 continue;
3614 if (open_pack_index(p)) {
3615 pack_errors = 1;
3616 continue;
3617 }
3618 r = for_each_object_in_pack(p, cb, data);
3619 if (r)
3620 break;
3621 }
3622 return r ? r : pack_errors;
3623 }