sha1_file: release delta_stack on error in unpack_entry()
[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 #include "mru.h"
27 #include "list.h"
28 #include "mergesort.h"
29 #include "quote.h"
30
31 #define SZ_FMT PRIuMAX
32 static inline uintmax_t sz_fmt(size_t s) { return s; }
33
34 const unsigned char null_sha1[20];
35 const struct object_id null_oid;
36 const struct object_id empty_tree_oid = {
37 EMPTY_TREE_SHA1_BIN_LITERAL
38 };
39 const struct object_id empty_blob_oid = {
40 EMPTY_BLOB_SHA1_BIN_LITERAL
41 };
42
43 /*
44 * This is meant to hold a *small* number of objects that you would
45 * want read_sha1_file() to be able to return, but yet you do not want
46 * to write them into the object store (e.g. a browse-only
47 * application).
48 */
49 static struct cached_object {
50 unsigned char sha1[20];
51 enum object_type type;
52 void *buf;
53 unsigned long size;
54 } *cached_objects;
55 static int cached_object_nr, cached_object_alloc;
56
57 static struct cached_object empty_tree = {
58 EMPTY_TREE_SHA1_BIN_LITERAL,
59 OBJ_TREE,
60 "",
61 0
62 };
63
64 static struct cached_object *find_cached_object(const unsigned char *sha1)
65 {
66 int i;
67 struct cached_object *co = cached_objects;
68
69 for (i = 0; i < cached_object_nr; i++, co++) {
70 if (!hashcmp(co->sha1, sha1))
71 return co;
72 }
73 if (!hashcmp(sha1, empty_tree.sha1))
74 return &empty_tree;
75 return NULL;
76 }
77
78 int mkdir_in_gitdir(const char *path)
79 {
80 if (mkdir(path, 0777)) {
81 int saved_errno = errno;
82 struct stat st;
83 struct strbuf sb = STRBUF_INIT;
84
85 if (errno != EEXIST)
86 return -1;
87 /*
88 * Are we looking at a path in a symlinked worktree
89 * whose original repository does not yet have it?
90 * e.g. .git/rr-cache pointing at its original
91 * repository in which the user hasn't performed any
92 * conflict resolution yet?
93 */
94 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
95 strbuf_readlink(&sb, path, st.st_size) ||
96 !is_absolute_path(sb.buf) ||
97 mkdir(sb.buf, 0777)) {
98 strbuf_release(&sb);
99 errno = saved_errno;
100 return -1;
101 }
102 strbuf_release(&sb);
103 }
104 return adjust_shared_perm(path);
105 }
106
107 enum scld_error safe_create_leading_directories(char *path)
108 {
109 char *next_component = path + offset_1st_component(path);
110 enum scld_error ret = SCLD_OK;
111
112 while (ret == SCLD_OK && next_component) {
113 struct stat st;
114 char *slash = next_component, slash_character;
115
116 while (*slash && !is_dir_sep(*slash))
117 slash++;
118
119 if (!*slash)
120 break;
121
122 next_component = slash + 1;
123 while (is_dir_sep(*next_component))
124 next_component++;
125 if (!*next_component)
126 break;
127
128 slash_character = *slash;
129 *slash = '\0';
130 if (!stat(path, &st)) {
131 /* path exists */
132 if (!S_ISDIR(st.st_mode))
133 ret = SCLD_EXISTS;
134 } else if (mkdir(path, 0777)) {
135 if (errno == EEXIST &&
136 !stat(path, &st) && S_ISDIR(st.st_mode))
137 ; /* somebody created it since we checked */
138 else if (errno == ENOENT)
139 /*
140 * Either mkdir() failed because
141 * somebody just pruned the containing
142 * directory, or stat() failed because
143 * the file that was in our way was
144 * just removed. Either way, inform
145 * the caller that it might be worth
146 * trying again:
147 */
148 ret = SCLD_VANISHED;
149 else
150 ret = SCLD_FAILED;
151 } else if (adjust_shared_perm(path)) {
152 ret = SCLD_PERMS;
153 }
154 *slash = slash_character;
155 }
156 return ret;
157 }
158
159 enum scld_error safe_create_leading_directories_const(const char *path)
160 {
161 /* path points to cache entries, so xstrdup before messing with it */
162 char *buf = xstrdup(path);
163 enum scld_error result = safe_create_leading_directories(buf);
164 free(buf);
165 return result;
166 }
167
168 static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
169 {
170 int i;
171 for (i = 0; i < 20; i++) {
172 static char hex[] = "0123456789abcdef";
173 unsigned int val = sha1[i];
174 strbuf_addch(buf, hex[val >> 4]);
175 strbuf_addch(buf, hex[val & 0xf]);
176 if (!i)
177 strbuf_addch(buf, '/');
178 }
179 }
180
181 const char *sha1_file_name(const unsigned char *sha1)
182 {
183 static struct strbuf buf = STRBUF_INIT;
184
185 strbuf_reset(&buf);
186 strbuf_addf(&buf, "%s/", get_object_directory());
187
188 fill_sha1_path(&buf, sha1);
189 return buf.buf;
190 }
191
192 struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
193 {
194 strbuf_setlen(&alt->scratch, alt->base_len);
195 return &alt->scratch;
196 }
197
198 static const char *alt_sha1_path(struct alternate_object_database *alt,
199 const unsigned char *sha1)
200 {
201 struct strbuf *buf = alt_scratch_buf(alt);
202 fill_sha1_path(buf, sha1);
203 return buf->buf;
204 }
205
206 char *odb_pack_name(struct strbuf *buf,
207 const unsigned char *sha1,
208 const char *ext)
209 {
210 strbuf_reset(buf);
211 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
212 sha1_to_hex(sha1), ext);
213 return buf->buf;
214 }
215
216 char *sha1_pack_name(const unsigned char *sha1)
217 {
218 static struct strbuf buf = STRBUF_INIT;
219 return odb_pack_name(&buf, sha1, "pack");
220 }
221
222 char *sha1_pack_index_name(const unsigned char *sha1)
223 {
224 static struct strbuf buf = STRBUF_INIT;
225 return odb_pack_name(&buf, sha1, "idx");
226 }
227
228 struct alternate_object_database *alt_odb_list;
229 static struct alternate_object_database **alt_odb_tail;
230
231 /*
232 * Return non-zero iff the path is usable as an alternate object database.
233 */
234 static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
235 {
236 struct alternate_object_database *alt;
237
238 /* Detect cases where alternate disappeared */
239 if (!is_directory(path->buf)) {
240 error("object directory %s does not exist; "
241 "check .git/objects/info/alternates.",
242 path->buf);
243 return 0;
244 }
245
246 /*
247 * Prevent the common mistake of listing the same
248 * thing twice, or object directory itself.
249 */
250 for (alt = alt_odb_list; alt; alt = alt->next) {
251 if (!fspathcmp(path->buf, alt->path))
252 return 0;
253 }
254 if (!fspathcmp(path->buf, normalized_objdir))
255 return 0;
256
257 return 1;
258 }
259
260 /*
261 * Prepare alternate object database registry.
262 *
263 * The variable alt_odb_list points at the list of struct
264 * alternate_object_database. The elements on this list come from
265 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
266 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
267 * whose contents is similar to that environment variable but can be
268 * LF separated. Its base points at a statically allocated buffer that
269 * contains "/the/directory/corresponding/to/.git/objects/...", while
270 * its name points just after the slash at the end of ".git/objects/"
271 * in the example above, and has enough space to hold 40-byte hex
272 * SHA1, an extra slash for the first level indirection, and the
273 * terminating NUL.
274 */
275 static int link_alt_odb_entry(const char *entry, const char *relative_base,
276 int depth, const char *normalized_objdir)
277 {
278 struct alternate_object_database *ent;
279 struct strbuf pathbuf = STRBUF_INIT;
280
281 if (!is_absolute_path(entry) && relative_base) {
282 strbuf_realpath(&pathbuf, relative_base, 1);
283 strbuf_addch(&pathbuf, '/');
284 }
285 strbuf_addstr(&pathbuf, entry);
286
287 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
288 error("unable to normalize alternate object path: %s",
289 pathbuf.buf);
290 strbuf_release(&pathbuf);
291 return -1;
292 }
293
294 /*
295 * The trailing slash after the directory name is given by
296 * this function at the end. Remove duplicates.
297 */
298 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
299 strbuf_setlen(&pathbuf, pathbuf.len - 1);
300
301 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
302 strbuf_release(&pathbuf);
303 return -1;
304 }
305
306 ent = alloc_alt_odb(pathbuf.buf);
307
308 /* add the alternate entry */
309 *alt_odb_tail = ent;
310 alt_odb_tail = &(ent->next);
311 ent->next = NULL;
312
313 /* recursively add alternates */
314 read_info_alternates(pathbuf.buf, depth + 1);
315
316 strbuf_release(&pathbuf);
317 return 0;
318 }
319
320 static const char *parse_alt_odb_entry(const char *string,
321 int sep,
322 struct strbuf *out)
323 {
324 const char *end;
325
326 strbuf_reset(out);
327
328 if (*string == '#') {
329 /* comment; consume up to next separator */
330 end = strchrnul(string, sep);
331 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
332 /*
333 * quoted path; unquote_c_style has copied the
334 * data for us and set "end". Broken quoting (e.g.,
335 * an entry that doesn't end with a quote) falls
336 * back to the unquoted case below.
337 */
338 } else {
339 /* normal, unquoted path */
340 end = strchrnul(string, sep);
341 strbuf_add(out, string, end - string);
342 }
343
344 if (*end)
345 end++;
346 return end;
347 }
348
349 static void link_alt_odb_entries(const char *alt, int len, int sep,
350 const char *relative_base, int depth)
351 {
352 struct strbuf objdirbuf = STRBUF_INIT;
353 struct strbuf entry = STRBUF_INIT;
354
355 if (depth > 5) {
356 error("%s: ignoring alternate object stores, nesting too deep.",
357 relative_base);
358 return;
359 }
360
361 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
362 if (strbuf_normalize_path(&objdirbuf) < 0)
363 die("unable to normalize object directory: %s",
364 objdirbuf.buf);
365
366 while (*alt) {
367 alt = parse_alt_odb_entry(alt, sep, &entry);
368 if (!entry.len)
369 continue;
370 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
371 }
372 strbuf_release(&entry);
373 strbuf_release(&objdirbuf);
374 }
375
376 void read_info_alternates(const char * relative_base, int depth)
377 {
378 char *map;
379 size_t mapsz;
380 struct stat st;
381 char *path;
382 int fd;
383
384 path = xstrfmt("%s/info/alternates", relative_base);
385 fd = git_open(path);
386 free(path);
387 if (fd < 0)
388 return;
389 if (fstat(fd, &st) || (st.st_size == 0)) {
390 close(fd);
391 return;
392 }
393 mapsz = xsize_t(st.st_size);
394 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
395 close(fd);
396
397 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
398
399 munmap(map, mapsz);
400 }
401
402 struct alternate_object_database *alloc_alt_odb(const char *dir)
403 {
404 struct alternate_object_database *ent;
405
406 FLEX_ALLOC_STR(ent, path, dir);
407 strbuf_init(&ent->scratch, 0);
408 strbuf_addf(&ent->scratch, "%s/", dir);
409 ent->base_len = ent->scratch.len;
410
411 return ent;
412 }
413
414 void add_to_alternates_file(const char *reference)
415 {
416 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
417 char *alts = git_pathdup("objects/info/alternates");
418 FILE *in, *out;
419
420 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
421 out = fdopen_lock_file(lock, "w");
422 if (!out)
423 die_errno("unable to fdopen alternates lockfile");
424
425 in = fopen(alts, "r");
426 if (in) {
427 struct strbuf line = STRBUF_INIT;
428 int found = 0;
429
430 while (strbuf_getline(&line, in) != EOF) {
431 if (!strcmp(reference, line.buf)) {
432 found = 1;
433 break;
434 }
435 fprintf_or_die(out, "%s\n", line.buf);
436 }
437
438 strbuf_release(&line);
439 fclose(in);
440
441 if (found) {
442 rollback_lock_file(lock);
443 lock = NULL;
444 }
445 }
446 else if (errno != ENOENT)
447 die_errno("unable to read alternates file");
448
449 if (lock) {
450 fprintf_or_die(out, "%s\n", reference);
451 if (commit_lock_file(lock))
452 die_errno("unable to move new alternates file into place");
453 if (alt_odb_tail)
454 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
455 }
456 free(alts);
457 }
458
459 void add_to_alternates_memory(const char *reference)
460 {
461 /*
462 * Make sure alternates are initialized, or else our entry may be
463 * overwritten when they are.
464 */
465 prepare_alt_odb();
466
467 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
468 }
469
470 /*
471 * Compute the exact path an alternate is at and returns it. In case of
472 * error NULL is returned and the human readable error is added to `err`
473 * `path` may be relative and should point to $GITDIR.
474 * `err` must not be null.
475 */
476 char *compute_alternate_path(const char *path, struct strbuf *err)
477 {
478 char *ref_git = NULL;
479 const char *repo, *ref_git_s;
480 int seen_error = 0;
481
482 ref_git_s = real_path_if_valid(path);
483 if (!ref_git_s) {
484 seen_error = 1;
485 strbuf_addf(err, _("path '%s' does not exist"), path);
486 goto out;
487 } else
488 /*
489 * Beware: read_gitfile(), real_path() and mkpath()
490 * return static buffer
491 */
492 ref_git = xstrdup(ref_git_s);
493
494 repo = read_gitfile(ref_git);
495 if (!repo)
496 repo = read_gitfile(mkpath("%s/.git", ref_git));
497 if (repo) {
498 free(ref_git);
499 ref_git = xstrdup(repo);
500 }
501
502 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
503 char *ref_git_git = mkpathdup("%s/.git", ref_git);
504 free(ref_git);
505 ref_git = ref_git_git;
506 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
507 struct strbuf sb = STRBUF_INIT;
508 seen_error = 1;
509 if (get_common_dir(&sb, ref_git)) {
510 strbuf_addf(err,
511 _("reference repository '%s' as a linked "
512 "checkout is not supported yet."),
513 path);
514 goto out;
515 }
516
517 strbuf_addf(err, _("reference repository '%s' is not a "
518 "local repository."), path);
519 goto out;
520 }
521
522 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
523 strbuf_addf(err, _("reference repository '%s' is shallow"),
524 path);
525 seen_error = 1;
526 goto out;
527 }
528
529 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
530 strbuf_addf(err,
531 _("reference repository '%s' is grafted"),
532 path);
533 seen_error = 1;
534 goto out;
535 }
536
537 out:
538 if (seen_error) {
539 free(ref_git);
540 ref_git = NULL;
541 }
542
543 return ref_git;
544 }
545
546 int foreach_alt_odb(alt_odb_fn fn, void *cb)
547 {
548 struct alternate_object_database *ent;
549 int r = 0;
550
551 prepare_alt_odb();
552 for (ent = alt_odb_list; ent; ent = ent->next) {
553 r = fn(ent, cb);
554 if (r)
555 break;
556 }
557 return r;
558 }
559
560 void prepare_alt_odb(void)
561 {
562 const char *alt;
563
564 if (alt_odb_tail)
565 return;
566
567 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
568 if (!alt) alt = "";
569
570 alt_odb_tail = &alt_odb_list;
571 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
572
573 read_info_alternates(get_object_directory(), 0);
574 }
575
576 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
577 static int freshen_file(const char *fn)
578 {
579 struct utimbuf t;
580 t.actime = t.modtime = time(NULL);
581 return !utime(fn, &t);
582 }
583
584 /*
585 * All of the check_and_freshen functions return 1 if the file exists and was
586 * freshened (if freshening was requested), 0 otherwise. If they return
587 * 0, you should not assume that it is safe to skip a write of the object (it
588 * either does not exist on disk, or has a stale mtime and may be subject to
589 * pruning).
590 */
591 static int check_and_freshen_file(const char *fn, int freshen)
592 {
593 if (access(fn, F_OK))
594 return 0;
595 if (freshen && !freshen_file(fn))
596 return 0;
597 return 1;
598 }
599
600 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
601 {
602 return check_and_freshen_file(sha1_file_name(sha1), freshen);
603 }
604
605 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
606 {
607 struct alternate_object_database *alt;
608 prepare_alt_odb();
609 for (alt = alt_odb_list; alt; alt = alt->next) {
610 const char *path = alt_sha1_path(alt, sha1);
611 if (check_and_freshen_file(path, freshen))
612 return 1;
613 }
614 return 0;
615 }
616
617 static int check_and_freshen(const unsigned char *sha1, int freshen)
618 {
619 return check_and_freshen_local(sha1, freshen) ||
620 check_and_freshen_nonlocal(sha1, freshen);
621 }
622
623 int has_loose_object_nonlocal(const unsigned char *sha1)
624 {
625 return check_and_freshen_nonlocal(sha1, 0);
626 }
627
628 static int has_loose_object(const unsigned char *sha1)
629 {
630 return check_and_freshen(sha1, 0);
631 }
632
633 static unsigned int pack_used_ctr;
634 static unsigned int pack_mmap_calls;
635 static unsigned int peak_pack_open_windows;
636 static unsigned int pack_open_windows;
637 static unsigned int pack_open_fds;
638 static unsigned int pack_max_fds;
639 static size_t peak_pack_mapped;
640 static size_t pack_mapped;
641 struct packed_git *packed_git;
642
643 static struct mru packed_git_mru_storage;
644 struct mru *packed_git_mru = &packed_git_mru_storage;
645
646 void pack_report(void)
647 {
648 fprintf(stderr,
649 "pack_report: getpagesize() = %10" SZ_FMT "\n"
650 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
651 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
652 sz_fmt(getpagesize()),
653 sz_fmt(packed_git_window_size),
654 sz_fmt(packed_git_limit));
655 fprintf(stderr,
656 "pack_report: pack_used_ctr = %10u\n"
657 "pack_report: pack_mmap_calls = %10u\n"
658 "pack_report: pack_open_windows = %10u / %10u\n"
659 "pack_report: pack_mapped = "
660 "%10" SZ_FMT " / %10" SZ_FMT "\n",
661 pack_used_ctr,
662 pack_mmap_calls,
663 pack_open_windows, peak_pack_open_windows,
664 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
665 }
666
667 /*
668 * Open and mmap the index file at path, perform a couple of
669 * consistency checks, then record its information to p. Return 0 on
670 * success.
671 */
672 static int check_packed_git_idx(const char *path, struct packed_git *p)
673 {
674 void *idx_map;
675 struct pack_idx_header *hdr;
676 size_t idx_size;
677 uint32_t version, nr, i, *index;
678 int fd = git_open(path);
679 struct stat st;
680
681 if (fd < 0)
682 return -1;
683 if (fstat(fd, &st)) {
684 close(fd);
685 return -1;
686 }
687 idx_size = xsize_t(st.st_size);
688 if (idx_size < 4 * 256 + 20 + 20) {
689 close(fd);
690 return error("index file %s is too small", path);
691 }
692 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
693 close(fd);
694
695 hdr = idx_map;
696 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
697 version = ntohl(hdr->idx_version);
698 if (version < 2 || version > 2) {
699 munmap(idx_map, idx_size);
700 return error("index file %s is version %"PRIu32
701 " and is not supported by this binary"
702 " (try upgrading GIT to a newer version)",
703 path, version);
704 }
705 } else
706 version = 1;
707
708 nr = 0;
709 index = idx_map;
710 if (version > 1)
711 index += 2; /* skip index header */
712 for (i = 0; i < 256; i++) {
713 uint32_t n = ntohl(index[i]);
714 if (n < nr) {
715 munmap(idx_map, idx_size);
716 return error("non-monotonic index %s", path);
717 }
718 nr = n;
719 }
720
721 if (version == 1) {
722 /*
723 * Total size:
724 * - 256 index entries 4 bytes each
725 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
726 * - 20-byte SHA1 of the packfile
727 * - 20-byte SHA1 file checksum
728 */
729 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
730 munmap(idx_map, idx_size);
731 return error("wrong index v1 file size in %s", path);
732 }
733 } else if (version == 2) {
734 /*
735 * Minimum size:
736 * - 8 bytes of header
737 * - 256 index entries 4 bytes each
738 * - 20-byte sha1 entry * nr
739 * - 4-byte crc entry * nr
740 * - 4-byte offset entry * nr
741 * - 20-byte SHA1 of the packfile
742 * - 20-byte SHA1 file checksum
743 * And after the 4-byte offset table might be a
744 * variable sized table containing 8-byte entries
745 * for offsets larger than 2^31.
746 */
747 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
748 unsigned long max_size = min_size;
749 if (nr)
750 max_size += (nr - 1)*8;
751 if (idx_size < min_size || idx_size > max_size) {
752 munmap(idx_map, idx_size);
753 return error("wrong index v2 file size in %s", path);
754 }
755 if (idx_size != min_size &&
756 /*
757 * make sure we can deal with large pack offsets.
758 * 31-bit signed offset won't be enough, neither
759 * 32-bit unsigned one will be.
760 */
761 (sizeof(off_t) <= 4)) {
762 munmap(idx_map, idx_size);
763 return error("pack too large for current definition of off_t in %s", path);
764 }
765 }
766
767 p->index_version = version;
768 p->index_data = idx_map;
769 p->index_size = idx_size;
770 p->num_objects = nr;
771 return 0;
772 }
773
774 int open_pack_index(struct packed_git *p)
775 {
776 char *idx_name;
777 size_t len;
778 int ret;
779
780 if (p->index_data)
781 return 0;
782
783 if (!strip_suffix(p->pack_name, ".pack", &len))
784 die("BUG: pack_name does not end in .pack");
785 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
786 ret = check_packed_git_idx(idx_name, p);
787 free(idx_name);
788 return ret;
789 }
790
791 static void scan_windows(struct packed_git *p,
792 struct packed_git **lru_p,
793 struct pack_window **lru_w,
794 struct pack_window **lru_l)
795 {
796 struct pack_window *w, *w_l;
797
798 for (w_l = NULL, w = p->windows; w; w = w->next) {
799 if (!w->inuse_cnt) {
800 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
801 *lru_p = p;
802 *lru_w = w;
803 *lru_l = w_l;
804 }
805 }
806 w_l = w;
807 }
808 }
809
810 static int unuse_one_window(struct packed_git *current)
811 {
812 struct packed_git *p, *lru_p = NULL;
813 struct pack_window *lru_w = NULL, *lru_l = NULL;
814
815 if (current)
816 scan_windows(current, &lru_p, &lru_w, &lru_l);
817 for (p = packed_git; p; p = p->next)
818 scan_windows(p, &lru_p, &lru_w, &lru_l);
819 if (lru_p) {
820 munmap(lru_w->base, lru_w->len);
821 pack_mapped -= lru_w->len;
822 if (lru_l)
823 lru_l->next = lru_w->next;
824 else
825 lru_p->windows = lru_w->next;
826 free(lru_w);
827 pack_open_windows--;
828 return 1;
829 }
830 return 0;
831 }
832
833 void release_pack_memory(size_t need)
834 {
835 size_t cur = pack_mapped;
836 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
837 ; /* nothing */
838 }
839
840 static void mmap_limit_check(size_t length)
841 {
842 static size_t limit = 0;
843 if (!limit) {
844 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
845 if (!limit)
846 limit = SIZE_MAX;
847 }
848 if (length > limit)
849 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
850 (uintmax_t)length, (uintmax_t)limit);
851 }
852
853 void *xmmap_gently(void *start, size_t length,
854 int prot, int flags, int fd, off_t offset)
855 {
856 void *ret;
857
858 mmap_limit_check(length);
859 ret = mmap(start, length, prot, flags, fd, offset);
860 if (ret == MAP_FAILED) {
861 if (!length)
862 return NULL;
863 release_pack_memory(length);
864 ret = mmap(start, length, prot, flags, fd, offset);
865 }
866 return ret;
867 }
868
869 void *xmmap(void *start, size_t length,
870 int prot, int flags, int fd, off_t offset)
871 {
872 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
873 if (ret == MAP_FAILED)
874 die_errno("mmap failed");
875 return ret;
876 }
877
878 void close_pack_windows(struct packed_git *p)
879 {
880 while (p->windows) {
881 struct pack_window *w = p->windows;
882
883 if (w->inuse_cnt)
884 die("pack '%s' still has open windows to it",
885 p->pack_name);
886 munmap(w->base, w->len);
887 pack_mapped -= w->len;
888 pack_open_windows--;
889 p->windows = w->next;
890 free(w);
891 }
892 }
893
894 static int close_pack_fd(struct packed_git *p)
895 {
896 if (p->pack_fd < 0)
897 return 0;
898
899 close(p->pack_fd);
900 pack_open_fds--;
901 p->pack_fd = -1;
902
903 return 1;
904 }
905
906 static void close_pack(struct packed_git *p)
907 {
908 close_pack_windows(p);
909 close_pack_fd(p);
910 close_pack_index(p);
911 }
912
913 void close_all_packs(void)
914 {
915 struct packed_git *p;
916
917 for (p = packed_git; p; p = p->next)
918 if (p->do_not_close)
919 die("BUG: want to close pack marked 'do-not-close'");
920 else
921 close_pack(p);
922 }
923
924
925 /*
926 * The LRU pack is the one with the oldest MRU window, preferring packs
927 * with no used windows, or the oldest mtime if it has no windows allocated.
928 */
929 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
930 {
931 struct pack_window *w, *this_mru_w;
932 int has_windows_inuse = 0;
933
934 /*
935 * Reject this pack if it has windows and the previously selected
936 * one does not. If this pack does not have windows, reject
937 * it if the pack file is newer than the previously selected one.
938 */
939 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
940 return;
941
942 for (w = this_mru_w = p->windows; w; w = w->next) {
943 /*
944 * Reject this pack if any of its windows are in use,
945 * but the previously selected pack did not have any
946 * inuse windows. Otherwise, record that this pack
947 * has windows in use.
948 */
949 if (w->inuse_cnt) {
950 if (*accept_windows_inuse)
951 has_windows_inuse = 1;
952 else
953 return;
954 }
955
956 if (w->last_used > this_mru_w->last_used)
957 this_mru_w = w;
958
959 /*
960 * Reject this pack if it has windows that have been
961 * used more recently than the previously selected pack.
962 * If the previously selected pack had windows inuse and
963 * we have not encountered a window in this pack that is
964 * inuse, skip this check since we prefer a pack with no
965 * inuse windows to one that has inuse windows.
966 */
967 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
968 this_mru_w->last_used > (*mru_w)->last_used)
969 return;
970 }
971
972 /*
973 * Select this pack.
974 */
975 *mru_w = this_mru_w;
976 *lru_p = p;
977 *accept_windows_inuse = has_windows_inuse;
978 }
979
980 static int close_one_pack(void)
981 {
982 struct packed_git *p, *lru_p = NULL;
983 struct pack_window *mru_w = NULL;
984 int accept_windows_inuse = 1;
985
986 for (p = packed_git; p; p = p->next) {
987 if (p->pack_fd == -1)
988 continue;
989 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
990 }
991
992 if (lru_p)
993 return close_pack_fd(lru_p);
994
995 return 0;
996 }
997
998 void unuse_pack(struct pack_window **w_cursor)
999 {
1000 struct pack_window *w = *w_cursor;
1001 if (w) {
1002 w->inuse_cnt--;
1003 *w_cursor = NULL;
1004 }
1005 }
1006
1007 void close_pack_index(struct packed_git *p)
1008 {
1009 if (p->index_data) {
1010 munmap((void *)p->index_data, p->index_size);
1011 p->index_data = NULL;
1012 }
1013 }
1014
1015 static unsigned int get_max_fd_limit(void)
1016 {
1017 #ifdef RLIMIT_NOFILE
1018 {
1019 struct rlimit lim;
1020
1021 if (!getrlimit(RLIMIT_NOFILE, &lim))
1022 return lim.rlim_cur;
1023 }
1024 #endif
1025
1026 #ifdef _SC_OPEN_MAX
1027 {
1028 long open_max = sysconf(_SC_OPEN_MAX);
1029 if (0 < open_max)
1030 return open_max;
1031 /*
1032 * Otherwise, we got -1 for one of the two
1033 * reasons:
1034 *
1035 * (1) sysconf() did not understand _SC_OPEN_MAX
1036 * and signaled an error with -1; or
1037 * (2) sysconf() said there is no limit.
1038 *
1039 * We _could_ clear errno before calling sysconf() to
1040 * tell these two cases apart and return a huge number
1041 * in the latter case to let the caller cap it to a
1042 * value that is not so selfish, but letting the
1043 * fallback OPEN_MAX codepath take care of these cases
1044 * is a lot simpler.
1045 */
1046 }
1047 #endif
1048
1049 #ifdef OPEN_MAX
1050 return OPEN_MAX;
1051 #else
1052 return 1; /* see the caller ;-) */
1053 #endif
1054 }
1055
1056 /*
1057 * Do not call this directly as this leaks p->pack_fd on error return;
1058 * call open_packed_git() instead.
1059 */
1060 static int open_packed_git_1(struct packed_git *p)
1061 {
1062 struct stat st;
1063 struct pack_header hdr;
1064 unsigned char sha1[20];
1065 unsigned char *idx_sha1;
1066 long fd_flag;
1067
1068 if (!p->index_data && open_pack_index(p))
1069 return error("packfile %s index unavailable", p->pack_name);
1070
1071 if (!pack_max_fds) {
1072 unsigned int max_fds = get_max_fd_limit();
1073
1074 /* Save 3 for stdin/stdout/stderr, 22 for work */
1075 if (25 < max_fds)
1076 pack_max_fds = max_fds - 25;
1077 else
1078 pack_max_fds = 1;
1079 }
1080
1081 while (pack_max_fds <= pack_open_fds && close_one_pack())
1082 ; /* nothing */
1083
1084 p->pack_fd = git_open(p->pack_name);
1085 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
1086 return -1;
1087 pack_open_fds++;
1088
1089 /* If we created the struct before we had the pack we lack size. */
1090 if (!p->pack_size) {
1091 if (!S_ISREG(st.st_mode))
1092 return error("packfile %s not a regular file", p->pack_name);
1093 p->pack_size = st.st_size;
1094 } else if (p->pack_size != st.st_size)
1095 return error("packfile %s size changed", p->pack_name);
1096
1097 /* We leave these file descriptors open with sliding mmap;
1098 * there is no point keeping them open across exec(), though.
1099 */
1100 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
1101 if (fd_flag < 0)
1102 return error("cannot determine file descriptor flags");
1103 fd_flag |= FD_CLOEXEC;
1104 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
1105 return error("cannot set FD_CLOEXEC");
1106
1107 /* Verify we recognize this pack file format. */
1108 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
1109 return error("file %s is far too short to be a packfile", p->pack_name);
1110 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
1111 return error("file %s is not a GIT packfile", p->pack_name);
1112 if (!pack_version_ok(hdr.hdr_version))
1113 return error("packfile %s is version %"PRIu32" and not"
1114 " supported (try upgrading GIT to a newer version)",
1115 p->pack_name, ntohl(hdr.hdr_version));
1116
1117 /* Verify the pack matches its index. */
1118 if (p->num_objects != ntohl(hdr.hdr_entries))
1119 return error("packfile %s claims to have %"PRIu32" objects"
1120 " while index indicates %"PRIu32" objects",
1121 p->pack_name, ntohl(hdr.hdr_entries),
1122 p->num_objects);
1123 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
1124 return error("end of packfile %s is unavailable", p->pack_name);
1125 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
1126 return error("packfile %s signature is unavailable", p->pack_name);
1127 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
1128 if (hashcmp(sha1, idx_sha1))
1129 return error("packfile %s does not match index", p->pack_name);
1130 return 0;
1131 }
1132
1133 static int open_packed_git(struct packed_git *p)
1134 {
1135 if (!open_packed_git_1(p))
1136 return 0;
1137 close_pack_fd(p);
1138 return -1;
1139 }
1140
1141 static int in_window(struct pack_window *win, off_t offset)
1142 {
1143 /* We must promise at least 20 bytes (one hash) after the
1144 * offset is available from this window, otherwise the offset
1145 * is not actually in this window and a different window (which
1146 * has that one hash excess) must be used. This is to support
1147 * the object header and delta base parsing routines below.
1148 */
1149 off_t win_off = win->offset;
1150 return win_off <= offset
1151 && (offset + 20) <= (win_off + win->len);
1152 }
1153
1154 unsigned char *use_pack(struct packed_git *p,
1155 struct pack_window **w_cursor,
1156 off_t offset,
1157 unsigned long *left)
1158 {
1159 struct pack_window *win = *w_cursor;
1160
1161 /* Since packfiles end in a hash of their content and it's
1162 * pointless to ask for an offset into the middle of that
1163 * hash, and the in_window function above wouldn't match
1164 * don't allow an offset too close to the end of the file.
1165 */
1166 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1167 die("packfile %s cannot be accessed", p->pack_name);
1168 if (offset > (p->pack_size - 20))
1169 die("offset beyond end of packfile (truncated pack?)");
1170 if (offset < 0)
1171 die(_("offset before end of packfile (broken .idx?)"));
1172
1173 if (!win || !in_window(win, offset)) {
1174 if (win)
1175 win->inuse_cnt--;
1176 for (win = p->windows; win; win = win->next) {
1177 if (in_window(win, offset))
1178 break;
1179 }
1180 if (!win) {
1181 size_t window_align = packed_git_window_size / 2;
1182 off_t len;
1183
1184 if (p->pack_fd == -1 && open_packed_git(p))
1185 die("packfile %s cannot be accessed", p->pack_name);
1186
1187 win = xcalloc(1, sizeof(*win));
1188 win->offset = (offset / window_align) * window_align;
1189 len = p->pack_size - win->offset;
1190 if (len > packed_git_window_size)
1191 len = packed_git_window_size;
1192 win->len = (size_t)len;
1193 pack_mapped += win->len;
1194 while (packed_git_limit < pack_mapped
1195 && unuse_one_window(p))
1196 ; /* nothing */
1197 win->base = xmmap(NULL, win->len,
1198 PROT_READ, MAP_PRIVATE,
1199 p->pack_fd, win->offset);
1200 if (win->base == MAP_FAILED)
1201 die_errno("packfile %s cannot be mapped",
1202 p->pack_name);
1203 if (!win->offset && win->len == p->pack_size
1204 && !p->do_not_close)
1205 close_pack_fd(p);
1206 pack_mmap_calls++;
1207 pack_open_windows++;
1208 if (pack_mapped > peak_pack_mapped)
1209 peak_pack_mapped = pack_mapped;
1210 if (pack_open_windows > peak_pack_open_windows)
1211 peak_pack_open_windows = pack_open_windows;
1212 win->next = p->windows;
1213 p->windows = win;
1214 }
1215 }
1216 if (win != *w_cursor) {
1217 win->last_used = pack_used_ctr++;
1218 win->inuse_cnt++;
1219 *w_cursor = win;
1220 }
1221 offset -= win->offset;
1222 if (left)
1223 *left = win->len - xsize_t(offset);
1224 return win->base + offset;
1225 }
1226
1227 static struct packed_git *alloc_packed_git(int extra)
1228 {
1229 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
1230 memset(p, 0, sizeof(*p));
1231 p->pack_fd = -1;
1232 return p;
1233 }
1234
1235 static void try_to_free_pack_memory(size_t size)
1236 {
1237 release_pack_memory(size);
1238 }
1239
1240 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
1241 {
1242 static int have_set_try_to_free_routine;
1243 struct stat st;
1244 size_t alloc;
1245 struct packed_git *p;
1246
1247 if (!have_set_try_to_free_routine) {
1248 have_set_try_to_free_routine = 1;
1249 set_try_to_free_routine(try_to_free_pack_memory);
1250 }
1251
1252 /*
1253 * Make sure a corresponding .pack file exists and that
1254 * the index looks sane.
1255 */
1256 if (!strip_suffix_mem(path, &path_len, ".idx"))
1257 return NULL;
1258
1259 /*
1260 * ".pack" is long enough to hold any suffix we're adding (and
1261 * the use xsnprintf double-checks that)
1262 */
1263 alloc = st_add3(path_len, strlen(".pack"), 1);
1264 p = alloc_packed_git(alloc);
1265 memcpy(p->pack_name, path, path_len);
1266
1267 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
1268 if (!access(p->pack_name, F_OK))
1269 p->pack_keep = 1;
1270
1271 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
1272 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1273 free(p);
1274 return NULL;
1275 }
1276
1277 /* ok, it looks sane as far as we can check without
1278 * actually mapping the pack file.
1279 */
1280 p->pack_size = st.st_size;
1281 p->pack_local = local;
1282 p->mtime = st.st_mtime;
1283 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1284 hashclr(p->sha1);
1285 return p;
1286 }
1287
1288 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1289 {
1290 const char *path = sha1_pack_name(sha1);
1291 size_t alloc = st_add(strlen(path), 1);
1292 struct packed_git *p = alloc_packed_git(alloc);
1293
1294 memcpy(p->pack_name, path, alloc); /* includes NUL */
1295 hashcpy(p->sha1, sha1);
1296 if (check_packed_git_idx(idx_path, p)) {
1297 free(p);
1298 return NULL;
1299 }
1300
1301 return p;
1302 }
1303
1304 void install_packed_git(struct packed_git *pack)
1305 {
1306 if (pack->pack_fd != -1)
1307 pack_open_fds++;
1308
1309 pack->next = packed_git;
1310 packed_git = pack;
1311 }
1312
1313 void (*report_garbage)(unsigned seen_bits, const char *path);
1314
1315 static void report_helper(const struct string_list *list,
1316 int seen_bits, int first, int last)
1317 {
1318 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
1319 return;
1320
1321 for (; first < last; first++)
1322 report_garbage(seen_bits, list->items[first].string);
1323 }
1324
1325 static void report_pack_garbage(struct string_list *list)
1326 {
1327 int i, baselen = -1, first = 0, seen_bits = 0;
1328
1329 if (!report_garbage)
1330 return;
1331
1332 string_list_sort(list);
1333
1334 for (i = 0; i < list->nr; i++) {
1335 const char *path = list->items[i].string;
1336 if (baselen != -1 &&
1337 strncmp(path, list->items[first].string, baselen)) {
1338 report_helper(list, seen_bits, first, i);
1339 baselen = -1;
1340 seen_bits = 0;
1341 }
1342 if (baselen == -1) {
1343 const char *dot = strrchr(path, '.');
1344 if (!dot) {
1345 report_garbage(PACKDIR_FILE_GARBAGE, path);
1346 continue;
1347 }
1348 baselen = dot - path + 1;
1349 first = i;
1350 }
1351 if (!strcmp(path + baselen, "pack"))
1352 seen_bits |= 1;
1353 else if (!strcmp(path + baselen, "idx"))
1354 seen_bits |= 2;
1355 }
1356 report_helper(list, seen_bits, first, list->nr);
1357 }
1358
1359 static void prepare_packed_git_one(char *objdir, int local)
1360 {
1361 struct strbuf path = STRBUF_INIT;
1362 size_t dirnamelen;
1363 DIR *dir;
1364 struct dirent *de;
1365 struct string_list garbage = STRING_LIST_INIT_DUP;
1366
1367 strbuf_addstr(&path, objdir);
1368 strbuf_addstr(&path, "/pack");
1369 dir = opendir(path.buf);
1370 if (!dir) {
1371 if (errno != ENOENT)
1372 error_errno("unable to open object pack directory: %s",
1373 path.buf);
1374 strbuf_release(&path);
1375 return;
1376 }
1377 strbuf_addch(&path, '/');
1378 dirnamelen = path.len;
1379 while ((de = readdir(dir)) != NULL) {
1380 struct packed_git *p;
1381 size_t base_len;
1382
1383 if (is_dot_or_dotdot(de->d_name))
1384 continue;
1385
1386 strbuf_setlen(&path, dirnamelen);
1387 strbuf_addstr(&path, de->d_name);
1388
1389 base_len = path.len;
1390 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1391 /* Don't reopen a pack we already have. */
1392 for (p = packed_git; p; p = p->next) {
1393 size_t len;
1394 if (strip_suffix(p->pack_name, ".pack", &len) &&
1395 len == base_len &&
1396 !memcmp(p->pack_name, path.buf, len))
1397 break;
1398 }
1399 if (p == NULL &&
1400 /*
1401 * See if it really is a valid .idx file with
1402 * corresponding .pack file that we can map.
1403 */
1404 (p = add_packed_git(path.buf, path.len, local)) != NULL)
1405 install_packed_git(p);
1406 }
1407
1408 if (!report_garbage)
1409 continue;
1410
1411 if (ends_with(de->d_name, ".idx") ||
1412 ends_with(de->d_name, ".pack") ||
1413 ends_with(de->d_name, ".bitmap") ||
1414 ends_with(de->d_name, ".keep"))
1415 string_list_append(&garbage, path.buf);
1416 else
1417 report_garbage(PACKDIR_FILE_GARBAGE, path.buf);
1418 }
1419 closedir(dir);
1420 report_pack_garbage(&garbage);
1421 string_list_clear(&garbage, 0);
1422 strbuf_release(&path);
1423 }
1424
1425 static int approximate_object_count_valid;
1426
1427 /*
1428 * Give a fast, rough count of the number of objects in the repository. This
1429 * ignores loose objects completely. If you have a lot of them, then either
1430 * you should repack because your performance will be awful, or they are
1431 * all unreachable objects about to be pruned, in which case they're not really
1432 * interesting as a measure of repo size in the first place.
1433 */
1434 unsigned long approximate_object_count(void)
1435 {
1436 static unsigned long count;
1437 if (!approximate_object_count_valid) {
1438 struct packed_git *p;
1439
1440 prepare_packed_git();
1441 count = 0;
1442 for (p = packed_git; p; p = p->next) {
1443 if (open_pack_index(p))
1444 continue;
1445 count += p->num_objects;
1446 }
1447 }
1448 return count;
1449 }
1450
1451 static void *get_next_packed_git(const void *p)
1452 {
1453 return ((const struct packed_git *)p)->next;
1454 }
1455
1456 static void set_next_packed_git(void *p, void *next)
1457 {
1458 ((struct packed_git *)p)->next = next;
1459 }
1460
1461 static int sort_pack(const void *a_, const void *b_)
1462 {
1463 const struct packed_git *a = a_;
1464 const struct packed_git *b = b_;
1465 int st;
1466
1467 /*
1468 * Local packs tend to contain objects specific to our
1469 * variant of the project than remote ones. In addition,
1470 * remote ones could be on a network mounted filesystem.
1471 * Favor local ones for these reasons.
1472 */
1473 st = a->pack_local - b->pack_local;
1474 if (st)
1475 return -st;
1476
1477 /*
1478 * Younger packs tend to contain more recent objects,
1479 * and more recent objects tend to get accessed more
1480 * often.
1481 */
1482 if (a->mtime < b->mtime)
1483 return 1;
1484 else if (a->mtime == b->mtime)
1485 return 0;
1486 return -1;
1487 }
1488
1489 static void rearrange_packed_git(void)
1490 {
1491 packed_git = llist_mergesort(packed_git, get_next_packed_git,
1492 set_next_packed_git, sort_pack);
1493 }
1494
1495 static void prepare_packed_git_mru(void)
1496 {
1497 struct packed_git *p;
1498
1499 mru_clear(packed_git_mru);
1500 for (p = packed_git; p; p = p->next)
1501 mru_append(packed_git_mru, p);
1502 }
1503
1504 static int prepare_packed_git_run_once = 0;
1505 void prepare_packed_git(void)
1506 {
1507 struct alternate_object_database *alt;
1508
1509 if (prepare_packed_git_run_once)
1510 return;
1511 prepare_packed_git_one(get_object_directory(), 1);
1512 prepare_alt_odb();
1513 for (alt = alt_odb_list; alt; alt = alt->next)
1514 prepare_packed_git_one(alt->path, 0);
1515 rearrange_packed_git();
1516 prepare_packed_git_mru();
1517 prepare_packed_git_run_once = 1;
1518 }
1519
1520 void reprepare_packed_git(void)
1521 {
1522 approximate_object_count_valid = 0;
1523 prepare_packed_git_run_once = 0;
1524 prepare_packed_git();
1525 }
1526
1527 static void mark_bad_packed_object(struct packed_git *p,
1528 const unsigned char *sha1)
1529 {
1530 unsigned i;
1531 for (i = 0; i < p->num_bad_objects; i++)
1532 if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
1533 return;
1534 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1535 st_mult(GIT_SHA1_RAWSZ,
1536 st_add(p->num_bad_objects, 1)));
1537 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
1538 p->num_bad_objects++;
1539 }
1540
1541 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1542 {
1543 struct packed_git *p;
1544 unsigned i;
1545
1546 for (p = packed_git; p; p = p->next)
1547 for (i = 0; i < p->num_bad_objects; i++)
1548 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1549 return p;
1550 return NULL;
1551 }
1552
1553 /*
1554 * With an in-core object data in "map", rehash it to make sure the
1555 * object name actually matches "sha1" to detect object corruption.
1556 * With "map" == NULL, try reading the object named with "sha1" using
1557 * the streaming interface and rehash it to do the same.
1558 */
1559 int check_sha1_signature(const unsigned char *sha1, void *map,
1560 unsigned long size, const char *type)
1561 {
1562 unsigned char real_sha1[20];
1563 enum object_type obj_type;
1564 struct git_istream *st;
1565 git_SHA_CTX c;
1566 char hdr[32];
1567 int hdrlen;
1568
1569 if (map) {
1570 hash_sha1_file(map, size, type, real_sha1);
1571 return hashcmp(sha1, real_sha1) ? -1 : 0;
1572 }
1573
1574 st = open_istream(sha1, &obj_type, &size, NULL);
1575 if (!st)
1576 return -1;
1577
1578 /* Generate the header */
1579 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
1580
1581 /* Sha1.. */
1582 git_SHA1_Init(&c);
1583 git_SHA1_Update(&c, hdr, hdrlen);
1584 for (;;) {
1585 char buf[1024 * 16];
1586 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1587
1588 if (readlen < 0) {
1589 close_istream(st);
1590 return -1;
1591 }
1592 if (!readlen)
1593 break;
1594 git_SHA1_Update(&c, buf, readlen);
1595 }
1596 git_SHA1_Final(real_sha1, &c);
1597 close_istream(st);
1598 return hashcmp(sha1, real_sha1) ? -1 : 0;
1599 }
1600
1601 int git_open_cloexec(const char *name, int flags)
1602 {
1603 int fd;
1604 static int o_cloexec = O_CLOEXEC;
1605
1606 fd = open(name, flags | o_cloexec);
1607 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1608 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1609 o_cloexec &= ~O_CLOEXEC;
1610 fd = open(name, flags | o_cloexec);
1611 }
1612
1613 #if defined(F_GETFL) && defined(F_SETFL) && defined(FD_CLOEXEC)
1614 {
1615 static int fd_cloexec = FD_CLOEXEC;
1616
1617 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1618 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1619 int flags = fcntl(fd, F_GETFL);
1620 if (fcntl(fd, F_SETFL, flags | fd_cloexec))
1621 fd_cloexec = 0;
1622 }
1623 }
1624 #endif
1625 return fd;
1626 }
1627
1628 /*
1629 * Find "sha1" as a loose object in the local repository or in an alternate.
1630 * Returns 0 on success, negative on failure.
1631 *
1632 * The "path" out-parameter will give the path of the object we found (if any).
1633 * Note that it may point to static storage and is only valid until another
1634 * call to sha1_file_name(), etc.
1635 */
1636 static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
1637 const char **path)
1638 {
1639 struct alternate_object_database *alt;
1640
1641 *path = sha1_file_name(sha1);
1642 if (!lstat(*path, st))
1643 return 0;
1644
1645 prepare_alt_odb();
1646 errno = ENOENT;
1647 for (alt = alt_odb_list; alt; alt = alt->next) {
1648 *path = alt_sha1_path(alt, sha1);
1649 if (!lstat(*path, st))
1650 return 0;
1651 }
1652
1653 return -1;
1654 }
1655
1656 /*
1657 * Like stat_sha1_file(), but actually open the object and return the
1658 * descriptor. See the caveats on the "path" parameter above.
1659 */
1660 static int open_sha1_file(const unsigned char *sha1, const char **path)
1661 {
1662 int fd;
1663 struct alternate_object_database *alt;
1664 int most_interesting_errno;
1665
1666 *path = sha1_file_name(sha1);
1667 fd = git_open(*path);
1668 if (fd >= 0)
1669 return fd;
1670 most_interesting_errno = errno;
1671
1672 prepare_alt_odb();
1673 for (alt = alt_odb_list; alt; alt = alt->next) {
1674 *path = alt_sha1_path(alt, sha1);
1675 fd = git_open(*path);
1676 if (fd >= 0)
1677 return fd;
1678 if (most_interesting_errno == ENOENT)
1679 most_interesting_errno = errno;
1680 }
1681 errno = most_interesting_errno;
1682 return -1;
1683 }
1684
1685 /*
1686 * Map the loose object at "path" if it is not NULL, or the path found by
1687 * searching for a loose object named "sha1".
1688 */
1689 static void *map_sha1_file_1(const char *path,
1690 const unsigned char *sha1,
1691 unsigned long *size)
1692 {
1693 void *map;
1694 int fd;
1695
1696 if (path)
1697 fd = git_open(path);
1698 else
1699 fd = open_sha1_file(sha1, &path);
1700 map = NULL;
1701 if (fd >= 0) {
1702 struct stat st;
1703
1704 if (!fstat(fd, &st)) {
1705 *size = xsize_t(st.st_size);
1706 if (!*size) {
1707 /* mmap() is forbidden on empty files */
1708 error("object file %s is empty", path);
1709 return NULL;
1710 }
1711 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1712 }
1713 close(fd);
1714 }
1715 return map;
1716 }
1717
1718 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1719 {
1720 return map_sha1_file_1(NULL, sha1, size);
1721 }
1722
1723 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1724 unsigned long len, enum object_type *type, unsigned long *sizep)
1725 {
1726 unsigned shift;
1727 unsigned long size, c;
1728 unsigned long used = 0;
1729
1730 c = buf[used++];
1731 *type = (c >> 4) & 7;
1732 size = c & 15;
1733 shift = 4;
1734 while (c & 0x80) {
1735 if (len <= used || bitsizeof(long) <= shift) {
1736 error("bad object header");
1737 size = used = 0;
1738 break;
1739 }
1740 c = buf[used++];
1741 size += (c & 0x7f) << shift;
1742 shift += 7;
1743 }
1744 *sizep = size;
1745 return used;
1746 }
1747
1748 static int unpack_sha1_short_header(git_zstream *stream,
1749 unsigned char *map, unsigned long mapsize,
1750 void *buffer, unsigned long bufsiz)
1751 {
1752 /* Get the data stream */
1753 memset(stream, 0, sizeof(*stream));
1754 stream->next_in = map;
1755 stream->avail_in = mapsize;
1756 stream->next_out = buffer;
1757 stream->avail_out = bufsiz;
1758
1759 git_inflate_init(stream);
1760 return git_inflate(stream, 0);
1761 }
1762
1763 int unpack_sha1_header(git_zstream *stream,
1764 unsigned char *map, unsigned long mapsize,
1765 void *buffer, unsigned long bufsiz)
1766 {
1767 int status = unpack_sha1_short_header(stream, map, mapsize,
1768 buffer, bufsiz);
1769
1770 if (status < Z_OK)
1771 return status;
1772
1773 /* Make sure we have the terminating NUL */
1774 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1775 return -1;
1776 return 0;
1777 }
1778
1779 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1780 unsigned long mapsize, void *buffer,
1781 unsigned long bufsiz, struct strbuf *header)
1782 {
1783 int status;
1784
1785 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1786 if (status < Z_OK)
1787 return -1;
1788
1789 /*
1790 * Check if entire header is unpacked in the first iteration.
1791 */
1792 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1793 return 0;
1794
1795 /*
1796 * buffer[0..bufsiz] was not large enough. Copy the partial
1797 * result out to header, and then append the result of further
1798 * reading the stream.
1799 */
1800 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1801 stream->next_out = buffer;
1802 stream->avail_out = bufsiz;
1803
1804 do {
1805 status = git_inflate(stream, 0);
1806 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1807 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1808 return 0;
1809 stream->next_out = buffer;
1810 stream->avail_out = bufsiz;
1811 } while (status != Z_STREAM_END);
1812 return -1;
1813 }
1814
1815 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1816 {
1817 int bytes = strlen(buffer) + 1;
1818 unsigned char *buf = xmallocz(size);
1819 unsigned long n;
1820 int status = Z_OK;
1821
1822 n = stream->total_out - bytes;
1823 if (n > size)
1824 n = size;
1825 memcpy(buf, (char *) buffer + bytes, n);
1826 bytes = n;
1827 if (bytes <= size) {
1828 /*
1829 * The above condition must be (bytes <= size), not
1830 * (bytes < size). In other words, even though we
1831 * expect no more output and set avail_out to zero,
1832 * the input zlib stream may have bytes that express
1833 * "this concludes the stream", and we *do* want to
1834 * eat that input.
1835 *
1836 * Otherwise we would not be able to test that we
1837 * consumed all the input to reach the expected size;
1838 * we also want to check that zlib tells us that all
1839 * went well with status == Z_STREAM_END at the end.
1840 */
1841 stream->next_out = buf + bytes;
1842 stream->avail_out = size - bytes;
1843 while (status == Z_OK)
1844 status = git_inflate(stream, Z_FINISH);
1845 }
1846 if (status == Z_STREAM_END && !stream->avail_in) {
1847 git_inflate_end(stream);
1848 return buf;
1849 }
1850
1851 if (status < 0)
1852 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1853 else if (stream->avail_in)
1854 error("garbage at end of loose object '%s'",
1855 sha1_to_hex(sha1));
1856 free(buf);
1857 return NULL;
1858 }
1859
1860 /*
1861 * We used to just use "sscanf()", but that's actually way
1862 * too permissive for what we want to check. So do an anal
1863 * object header parse by hand.
1864 */
1865 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1866 unsigned int flags)
1867 {
1868 const char *type_buf = hdr;
1869 unsigned long size;
1870 int type, type_len = 0;
1871
1872 /*
1873 * The type can be of any size but is followed by
1874 * a space.
1875 */
1876 for (;;) {
1877 char c = *hdr++;
1878 if (!c)
1879 return -1;
1880 if (c == ' ')
1881 break;
1882 type_len++;
1883 }
1884
1885 type = type_from_string_gently(type_buf, type_len, 1);
1886 if (oi->typename)
1887 strbuf_add(oi->typename, type_buf, type_len);
1888 /*
1889 * Set type to 0 if its an unknown object and
1890 * we're obtaining the type using '--allow-unknown-type'
1891 * option.
1892 */
1893 if ((flags & LOOKUP_UNKNOWN_OBJECT) && (type < 0))
1894 type = 0;
1895 else if (type < 0)
1896 die("invalid object type");
1897 if (oi->typep)
1898 *oi->typep = type;
1899
1900 /*
1901 * The length must follow immediately, and be in canonical
1902 * decimal format (ie "010" is not valid).
1903 */
1904 size = *hdr++ - '0';
1905 if (size > 9)
1906 return -1;
1907 if (size) {
1908 for (;;) {
1909 unsigned long c = *hdr - '0';
1910 if (c > 9)
1911 break;
1912 hdr++;
1913 size = size * 10 + c;
1914 }
1915 }
1916
1917 if (oi->sizep)
1918 *oi->sizep = size;
1919
1920 /*
1921 * The length must be followed by a zero byte
1922 */
1923 return *hdr ? -1 : type;
1924 }
1925
1926 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1927 {
1928 struct object_info oi = OBJECT_INFO_INIT;
1929
1930 oi.sizep = sizep;
1931 return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
1932 }
1933
1934 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1935 {
1936 int ret;
1937 git_zstream stream;
1938 char hdr[8192];
1939
1940 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1941 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1942 return NULL;
1943
1944 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1945 }
1946
1947 unsigned long get_size_from_delta(struct packed_git *p,
1948 struct pack_window **w_curs,
1949 off_t curpos)
1950 {
1951 const unsigned char *data;
1952 unsigned char delta_head[20], *in;
1953 git_zstream stream;
1954 int st;
1955
1956 memset(&stream, 0, sizeof(stream));
1957 stream.next_out = delta_head;
1958 stream.avail_out = sizeof(delta_head);
1959
1960 git_inflate_init(&stream);
1961 do {
1962 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1963 stream.next_in = in;
1964 st = git_inflate(&stream, Z_FINISH);
1965 curpos += stream.next_in - in;
1966 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1967 stream.total_out < sizeof(delta_head));
1968 git_inflate_end(&stream);
1969 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1970 error("delta data unpack-initial failed");
1971 return 0;
1972 }
1973
1974 /* Examine the initial part of the delta to figure out
1975 * the result size.
1976 */
1977 data = delta_head;
1978
1979 /* ignore base size */
1980 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1981
1982 /* Read the result size */
1983 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1984 }
1985
1986 static off_t get_delta_base(struct packed_git *p,
1987 struct pack_window **w_curs,
1988 off_t *curpos,
1989 enum object_type type,
1990 off_t delta_obj_offset)
1991 {
1992 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1993 off_t base_offset;
1994
1995 /* use_pack() assured us we have [base_info, base_info + 20)
1996 * as a range that we can look at without walking off the
1997 * end of the mapped window. Its actually the hash size
1998 * that is assured. An OFS_DELTA longer than the hash size
1999 * is stupid, as then a REF_DELTA would be smaller to store.
2000 */
2001 if (type == OBJ_OFS_DELTA) {
2002 unsigned used = 0;
2003 unsigned char c = base_info[used++];
2004 base_offset = c & 127;
2005 while (c & 128) {
2006 base_offset += 1;
2007 if (!base_offset || MSB(base_offset, 7))
2008 return 0; /* overflow */
2009 c = base_info[used++];
2010 base_offset = (base_offset << 7) + (c & 127);
2011 }
2012 base_offset = delta_obj_offset - base_offset;
2013 if (base_offset <= 0 || base_offset >= delta_obj_offset)
2014 return 0; /* out of bound */
2015 *curpos += used;
2016 } else if (type == OBJ_REF_DELTA) {
2017 /* The base entry _must_ be in the same pack */
2018 base_offset = find_pack_entry_one(base_info, p);
2019 *curpos += 20;
2020 } else
2021 die("I am totally screwed");
2022 return base_offset;
2023 }
2024
2025 /*
2026 * Like get_delta_base above, but we return the sha1 instead of the pack
2027 * offset. This means it is cheaper for REF deltas (we do not have to do
2028 * the final object lookup), but more expensive for OFS deltas (we
2029 * have to load the revidx to convert the offset back into a sha1).
2030 */
2031 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
2032 struct pack_window **w_curs,
2033 off_t curpos,
2034 enum object_type type,
2035 off_t delta_obj_offset)
2036 {
2037 if (type == OBJ_REF_DELTA) {
2038 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
2039 return base;
2040 } else if (type == OBJ_OFS_DELTA) {
2041 struct revindex_entry *revidx;
2042 off_t base_offset = get_delta_base(p, w_curs, &curpos,
2043 type, delta_obj_offset);
2044
2045 if (!base_offset)
2046 return NULL;
2047
2048 revidx = find_pack_revindex(p, base_offset);
2049 if (!revidx)
2050 return NULL;
2051
2052 return nth_packed_object_sha1(p, revidx->nr);
2053 } else
2054 return NULL;
2055 }
2056
2057 int unpack_object_header(struct packed_git *p,
2058 struct pack_window **w_curs,
2059 off_t *curpos,
2060 unsigned long *sizep)
2061 {
2062 unsigned char *base;
2063 unsigned long left;
2064 unsigned long used;
2065 enum object_type type;
2066
2067 /* use_pack() assures us we have [base, base + 20) available
2068 * as a range that we can look at. (Its actually the hash
2069 * size that is assured.) With our object header encoding
2070 * the maximum deflated object size is 2^137, which is just
2071 * insane, so we know won't exceed what we have been given.
2072 */
2073 base = use_pack(p, w_curs, *curpos, &left);
2074 used = unpack_object_header_buffer(base, left, &type, sizep);
2075 if (!used) {
2076 type = OBJ_BAD;
2077 } else
2078 *curpos += used;
2079
2080 return type;
2081 }
2082
2083 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
2084 {
2085 int type;
2086 struct revindex_entry *revidx;
2087 const unsigned char *sha1;
2088 revidx = find_pack_revindex(p, obj_offset);
2089 if (!revidx)
2090 return OBJ_BAD;
2091 sha1 = nth_packed_object_sha1(p, revidx->nr);
2092 mark_bad_packed_object(p, sha1);
2093 type = sha1_object_info(sha1, NULL);
2094 if (type <= OBJ_NONE)
2095 return OBJ_BAD;
2096 return type;
2097 }
2098
2099 #define POI_STACK_PREALLOC 64
2100
2101 static enum object_type packed_to_object_type(struct packed_git *p,
2102 off_t obj_offset,
2103 enum object_type type,
2104 struct pack_window **w_curs,
2105 off_t curpos)
2106 {
2107 off_t small_poi_stack[POI_STACK_PREALLOC];
2108 off_t *poi_stack = small_poi_stack;
2109 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
2110
2111 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2112 off_t base_offset;
2113 unsigned long size;
2114 /* Push the object we're going to leave behind */
2115 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
2116 poi_stack_alloc = alloc_nr(poi_stack_nr);
2117 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
2118 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
2119 } else {
2120 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
2121 }
2122 poi_stack[poi_stack_nr++] = obj_offset;
2123 /* If parsing the base offset fails, just unwind */
2124 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
2125 if (!base_offset)
2126 goto unwind;
2127 curpos = obj_offset = base_offset;
2128 type = unpack_object_header(p, w_curs, &curpos, &size);
2129 if (type <= OBJ_NONE) {
2130 /* If getting the base itself fails, we first
2131 * retry the base, otherwise unwind */
2132 type = retry_bad_packed_offset(p, base_offset);
2133 if (type > OBJ_NONE)
2134 goto out;
2135 goto unwind;
2136 }
2137 }
2138
2139 switch (type) {
2140 case OBJ_BAD:
2141 case OBJ_COMMIT:
2142 case OBJ_TREE:
2143 case OBJ_BLOB:
2144 case OBJ_TAG:
2145 break;
2146 default:
2147 error("unknown object type %i at offset %"PRIuMAX" in %s",
2148 type, (uintmax_t)obj_offset, p->pack_name);
2149 type = OBJ_BAD;
2150 }
2151
2152 out:
2153 if (poi_stack != small_poi_stack)
2154 free(poi_stack);
2155 return type;
2156
2157 unwind:
2158 while (poi_stack_nr) {
2159 obj_offset = poi_stack[--poi_stack_nr];
2160 type = retry_bad_packed_offset(p, obj_offset);
2161 if (type > OBJ_NONE)
2162 goto out;
2163 }
2164 type = OBJ_BAD;
2165 goto out;
2166 }
2167
2168 int packed_object_info(struct packed_git *p, off_t obj_offset,
2169 struct object_info *oi)
2170 {
2171 struct pack_window *w_curs = NULL;
2172 unsigned long size;
2173 off_t curpos = obj_offset;
2174 enum object_type type;
2175
2176 /*
2177 * We always get the representation type, but only convert it to
2178 * a "real" type later if the caller is interested.
2179 */
2180 type = unpack_object_header(p, &w_curs, &curpos, &size);
2181
2182 if (oi->sizep) {
2183 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2184 off_t tmp_pos = curpos;
2185 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
2186 type, obj_offset);
2187 if (!base_offset) {
2188 type = OBJ_BAD;
2189 goto out;
2190 }
2191 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2192 if (*oi->sizep == 0) {
2193 type = OBJ_BAD;
2194 goto out;
2195 }
2196 } else {
2197 *oi->sizep = size;
2198 }
2199 }
2200
2201 if (oi->disk_sizep) {
2202 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2203 *oi->disk_sizep = revidx[1].offset - obj_offset;
2204 }
2205
2206 if (oi->typep) {
2207 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
2208 if (*oi->typep < 0) {
2209 type = OBJ_BAD;
2210 goto out;
2211 }
2212 }
2213
2214 if (oi->delta_base_sha1) {
2215 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2216 const unsigned char *base;
2217
2218 base = get_delta_base_sha1(p, &w_curs, curpos,
2219 type, obj_offset);
2220 if (!base) {
2221 type = OBJ_BAD;
2222 goto out;
2223 }
2224
2225 hashcpy(oi->delta_base_sha1, base);
2226 } else
2227 hashclr(oi->delta_base_sha1);
2228 }
2229
2230 out:
2231 unuse_pack(&w_curs);
2232 return type;
2233 }
2234
2235 static void *unpack_compressed_entry(struct packed_git *p,
2236 struct pack_window **w_curs,
2237 off_t curpos,
2238 unsigned long size)
2239 {
2240 int st;
2241 git_zstream stream;
2242 unsigned char *buffer, *in;
2243
2244 buffer = xmallocz_gently(size);
2245 if (!buffer)
2246 return NULL;
2247 memset(&stream, 0, sizeof(stream));
2248 stream.next_out = buffer;
2249 stream.avail_out = size + 1;
2250
2251 git_inflate_init(&stream);
2252 do {
2253 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2254 stream.next_in = in;
2255 st = git_inflate(&stream, Z_FINISH);
2256 if (!stream.avail_out)
2257 break; /* the payload is larger than it should be */
2258 curpos += stream.next_in - in;
2259 } while (st == Z_OK || st == Z_BUF_ERROR);
2260 git_inflate_end(&stream);
2261 if ((st != Z_STREAM_END) || stream.total_out != size) {
2262 free(buffer);
2263 return NULL;
2264 }
2265
2266 return buffer;
2267 }
2268
2269 static struct hashmap delta_base_cache;
2270 static size_t delta_base_cached;
2271
2272 static LIST_HEAD(delta_base_cache_lru);
2273
2274 struct delta_base_cache_key {
2275 struct packed_git *p;
2276 off_t base_offset;
2277 };
2278
2279 struct delta_base_cache_entry {
2280 struct hashmap hash;
2281 struct delta_base_cache_key key;
2282 struct list_head lru;
2283 void *data;
2284 unsigned long size;
2285 enum object_type type;
2286 };
2287
2288 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
2289 {
2290 unsigned int hash;
2291
2292 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
2293 hash += (hash >> 8) + (hash >> 16);
2294 return hash;
2295 }
2296
2297 static struct delta_base_cache_entry *
2298 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2299 {
2300 struct hashmap_entry entry;
2301 struct delta_base_cache_key key;
2302
2303 if (!delta_base_cache.cmpfn)
2304 return NULL;
2305
2306 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
2307 key.p = p;
2308 key.base_offset = base_offset;
2309 return hashmap_get(&delta_base_cache, &entry, &key);
2310 }
2311
2312 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
2313 const struct delta_base_cache_key *b)
2314 {
2315 return a->p == b->p && a->base_offset == b->base_offset;
2316 }
2317
2318 static int delta_base_cache_hash_cmp(const void *va, const void *vb,
2319 const void *vkey)
2320 {
2321 const struct delta_base_cache_entry *a = va, *b = vb;
2322 const struct delta_base_cache_key *key = vkey;
2323 if (key)
2324 return !delta_base_cache_key_eq(&a->key, key);
2325 else
2326 return !delta_base_cache_key_eq(&a->key, &b->key);
2327 }
2328
2329 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2330 {
2331 return !!get_delta_base_cache_entry(p, base_offset);
2332 }
2333
2334 /*
2335 * Remove the entry from the cache, but do _not_ free the associated
2336 * entry data. The caller takes ownership of the "data" buffer, and
2337 * should copy out any fields it wants before detaching.
2338 */
2339 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2340 {
2341 hashmap_remove(&delta_base_cache, ent, &ent->key);
2342 list_del(&ent->lru);
2343 delta_base_cached -= ent->size;
2344 free(ent);
2345 }
2346
2347 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2348 unsigned long *base_size, enum object_type *type)
2349 {
2350 struct delta_base_cache_entry *ent;
2351
2352 ent = get_delta_base_cache_entry(p, base_offset);
2353 if (!ent)
2354 return unpack_entry(p, base_offset, type, base_size);
2355
2356 *type = ent->type;
2357 *base_size = ent->size;
2358 return xmemdupz(ent->data, ent->size);
2359 }
2360
2361 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2362 {
2363 free(ent->data);
2364 detach_delta_base_cache_entry(ent);
2365 }
2366
2367 void clear_delta_base_cache(void)
2368 {
2369 struct list_head *lru, *tmp;
2370 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
2371 struct delta_base_cache_entry *entry =
2372 list_entry(lru, struct delta_base_cache_entry, lru);
2373 release_delta_base_cache(entry);
2374 }
2375 }
2376
2377 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2378 void *base, unsigned long base_size, enum object_type type)
2379 {
2380 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
2381 struct list_head *lru, *tmp;
2382
2383 delta_base_cached += base_size;
2384
2385 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
2386 struct delta_base_cache_entry *f =
2387 list_entry(lru, struct delta_base_cache_entry, lru);
2388 if (delta_base_cached <= delta_base_cache_limit)
2389 break;
2390 release_delta_base_cache(f);
2391 }
2392
2393 ent->key.p = p;
2394 ent->key.base_offset = base_offset;
2395 ent->type = type;
2396 ent->data = base;
2397 ent->size = base_size;
2398 list_add_tail(&ent->lru, &delta_base_cache_lru);
2399
2400 if (!delta_base_cache.cmpfn)
2401 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, 0);
2402 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
2403 hashmap_add(&delta_base_cache, ent);
2404 }
2405
2406 static void *read_object(const unsigned char *sha1, enum object_type *type,
2407 unsigned long *size);
2408
2409 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2410 {
2411 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2412 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2413 p->pack_name, (uintmax_t)obj_offset);
2414 }
2415
2416 int do_check_packed_object_crc;
2417
2418 #define UNPACK_ENTRY_STACK_PREALLOC 64
2419 struct unpack_entry_stack_ent {
2420 off_t obj_offset;
2421 off_t curpos;
2422 unsigned long size;
2423 };
2424
2425 void *unpack_entry(struct packed_git *p, off_t obj_offset,
2426 enum object_type *final_type, unsigned long *final_size)
2427 {
2428 struct pack_window *w_curs = NULL;
2429 off_t curpos = obj_offset;
2430 void *data = NULL;
2431 unsigned long size;
2432 enum object_type type;
2433 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2434 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2435 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2436 int base_from_cache = 0;
2437
2438 write_pack_access_log(p, obj_offset);
2439
2440 /* PHASE 1: drill down to the innermost base object */
2441 for (;;) {
2442 off_t base_offset;
2443 int i;
2444 struct delta_base_cache_entry *ent;
2445
2446 ent = get_delta_base_cache_entry(p, curpos);
2447 if (ent) {
2448 type = ent->type;
2449 data = ent->data;
2450 size = ent->size;
2451 detach_delta_base_cache_entry(ent);
2452 base_from_cache = 1;
2453 break;
2454 }
2455
2456 if (do_check_packed_object_crc && p->index_version > 1) {
2457 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2458 off_t len = revidx[1].offset - obj_offset;
2459 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2460 const unsigned char *sha1 =
2461 nth_packed_object_sha1(p, revidx->nr);
2462 error("bad packed object CRC for %s",
2463 sha1_to_hex(sha1));
2464 mark_bad_packed_object(p, sha1);
2465 data = NULL;
2466 goto out;
2467 }
2468 }
2469
2470 type = unpack_object_header(p, &w_curs, &curpos, &size);
2471 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2472 break;
2473
2474 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2475 if (!base_offset) {
2476 error("failed to validate delta base reference "
2477 "at offset %"PRIuMAX" from %s",
2478 (uintmax_t)curpos, p->pack_name);
2479 /* bail to phase 2, in hopes of recovery */
2480 data = NULL;
2481 break;
2482 }
2483
2484 /* push object, proceed to base */
2485 if (delta_stack_nr >= delta_stack_alloc
2486 && delta_stack == small_delta_stack) {
2487 delta_stack_alloc = alloc_nr(delta_stack_nr);
2488 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
2489 memcpy(delta_stack, small_delta_stack,
2490 sizeof(*delta_stack)*delta_stack_nr);
2491 } else {
2492 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2493 }
2494 i = delta_stack_nr++;
2495 delta_stack[i].obj_offset = obj_offset;
2496 delta_stack[i].curpos = curpos;
2497 delta_stack[i].size = size;
2498
2499 curpos = obj_offset = base_offset;
2500 }
2501
2502 /* PHASE 2: handle the base */
2503 switch (type) {
2504 case OBJ_OFS_DELTA:
2505 case OBJ_REF_DELTA:
2506 if (data)
2507 die("BUG: unpack_entry: left loop at a valid delta");
2508 break;
2509 case OBJ_COMMIT:
2510 case OBJ_TREE:
2511 case OBJ_BLOB:
2512 case OBJ_TAG:
2513 if (!base_from_cache)
2514 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2515 break;
2516 default:
2517 data = NULL;
2518 error("unknown object type %i at offset %"PRIuMAX" in %s",
2519 type, (uintmax_t)obj_offset, p->pack_name);
2520 }
2521
2522 /* PHASE 3: apply deltas in order */
2523
2524 /* invariants:
2525 * 'data' holds the base data, or NULL if there was corruption
2526 */
2527 while (delta_stack_nr) {
2528 void *delta_data;
2529 void *base = data;
2530 void *external_base = NULL;
2531 unsigned long delta_size, base_size = size;
2532 int i;
2533
2534 data = NULL;
2535
2536 if (base)
2537 add_delta_base_cache(p, obj_offset, base, base_size, type);
2538
2539 if (!base) {
2540 /*
2541 * We're probably in deep shit, but let's try to fetch
2542 * the required base anyway from another pack or loose.
2543 * This is costly but should happen only in the presence
2544 * of a corrupted pack, and is better than failing outright.
2545 */
2546 struct revindex_entry *revidx;
2547 const unsigned char *base_sha1;
2548 revidx = find_pack_revindex(p, obj_offset);
2549 if (revidx) {
2550 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2551 error("failed to read delta base object %s"
2552 " at offset %"PRIuMAX" from %s",
2553 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2554 p->pack_name);
2555 mark_bad_packed_object(p, base_sha1);
2556 base = read_object(base_sha1, &type, &base_size);
2557 external_base = base;
2558 }
2559 }
2560
2561 i = --delta_stack_nr;
2562 obj_offset = delta_stack[i].obj_offset;
2563 curpos = delta_stack[i].curpos;
2564 delta_size = delta_stack[i].size;
2565
2566 if (!base)
2567 continue;
2568
2569 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2570
2571 if (!delta_data) {
2572 error("failed to unpack compressed delta "
2573 "at offset %"PRIuMAX" from %s",
2574 (uintmax_t)curpos, p->pack_name);
2575 data = NULL;
2576 free(external_base);
2577 continue;
2578 }
2579
2580 data = patch_delta(base, base_size,
2581 delta_data, delta_size,
2582 &size);
2583
2584 /*
2585 * We could not apply the delta; warn the user, but keep going.
2586 * Our failure will be noticed either in the next iteration of
2587 * the loop, or if this is the final delta, in the caller when
2588 * we return NULL. Those code paths will take care of making
2589 * a more explicit warning and retrying with another copy of
2590 * the object.
2591 */
2592 if (!data)
2593 error("failed to apply delta");
2594
2595 free(delta_data);
2596 free(external_base);
2597 }
2598
2599 *final_type = type;
2600 *final_size = size;
2601
2602 out:
2603 unuse_pack(&w_curs);
2604
2605 if (delta_stack != small_delta_stack)
2606 free(delta_stack);
2607
2608 return data;
2609 }
2610
2611 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2612 uint32_t n)
2613 {
2614 const unsigned char *index = p->index_data;
2615 if (!index) {
2616 if (open_pack_index(p))
2617 return NULL;
2618 index = p->index_data;
2619 }
2620 if (n >= p->num_objects)
2621 return NULL;
2622 index += 4 * 256;
2623 if (p->index_version == 1) {
2624 return index + 24 * n + 4;
2625 } else {
2626 index += 8;
2627 return index + 20 * n;
2628 }
2629 }
2630
2631 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
2632 {
2633 const unsigned char *ptr = vptr;
2634 const unsigned char *start = p->index_data;
2635 const unsigned char *end = start + p->index_size;
2636 if (ptr < start)
2637 die(_("offset before start of pack index for %s (corrupt index?)"),
2638 p->pack_name);
2639 /* No need to check for underflow; .idx files must be at least 8 bytes */
2640 if (ptr >= end - 8)
2641 die(_("offset beyond end of pack index for %s (truncated index?)"),
2642 p->pack_name);
2643 }
2644
2645 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2646 {
2647 const unsigned char *index = p->index_data;
2648 index += 4 * 256;
2649 if (p->index_version == 1) {
2650 return ntohl(*((uint32_t *)(index + 24 * n)));
2651 } else {
2652 uint32_t off;
2653 index += 8 + p->num_objects * (20 + 4);
2654 off = ntohl(*((uint32_t *)(index + 4 * n)));
2655 if (!(off & 0x80000000))
2656 return off;
2657 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2658 check_pack_index_ptr(p, index);
2659 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2660 ntohl(*((uint32_t *)(index + 4)));
2661 }
2662 }
2663
2664 off_t find_pack_entry_one(const unsigned char *sha1,
2665 struct packed_git *p)
2666 {
2667 const uint32_t *level1_ofs = p->index_data;
2668 const unsigned char *index = p->index_data;
2669 unsigned hi, lo, stride;
2670 static int use_lookup = -1;
2671 static int debug_lookup = -1;
2672
2673 if (debug_lookup < 0)
2674 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2675
2676 if (!index) {
2677 if (open_pack_index(p))
2678 return 0;
2679 level1_ofs = p->index_data;
2680 index = p->index_data;
2681 }
2682 if (p->index_version > 1) {
2683 level1_ofs += 2;
2684 index += 8;
2685 }
2686 index += 4 * 256;
2687 hi = ntohl(level1_ofs[*sha1]);
2688 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2689 if (p->index_version > 1) {
2690 stride = 20;
2691 } else {
2692 stride = 24;
2693 index += 4;
2694 }
2695
2696 if (debug_lookup)
2697 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2698 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2699
2700 if (use_lookup < 0)
2701 use_lookup = !!getenv("GIT_USE_LOOKUP");
2702 if (use_lookup) {
2703 int pos = sha1_entry_pos(index, stride, 0,
2704 lo, hi, p->num_objects, sha1);
2705 if (pos < 0)
2706 return 0;
2707 return nth_packed_object_offset(p, pos);
2708 }
2709
2710 do {
2711 unsigned mi = (lo + hi) / 2;
2712 int cmp = hashcmp(index + mi * stride, sha1);
2713
2714 if (debug_lookup)
2715 printf("lo %u hi %u rg %u mi %u\n",
2716 lo, hi, hi - lo, mi);
2717 if (!cmp)
2718 return nth_packed_object_offset(p, mi);
2719 if (cmp > 0)
2720 hi = mi;
2721 else
2722 lo = mi+1;
2723 } while (lo < hi);
2724 return 0;
2725 }
2726
2727 int is_pack_valid(struct packed_git *p)
2728 {
2729 /* An already open pack is known to be valid. */
2730 if (p->pack_fd != -1)
2731 return 1;
2732
2733 /* If the pack has one window completely covering the
2734 * file size, the pack is known to be valid even if
2735 * the descriptor is not currently open.
2736 */
2737 if (p->windows) {
2738 struct pack_window *w = p->windows;
2739
2740 if (!w->offset && w->len == p->pack_size)
2741 return 1;
2742 }
2743
2744 /* Force the pack to open to prove its valid. */
2745 return !open_packed_git(p);
2746 }
2747
2748 static int fill_pack_entry(const unsigned char *sha1,
2749 struct pack_entry *e,
2750 struct packed_git *p)
2751 {
2752 off_t offset;
2753
2754 if (p->num_bad_objects) {
2755 unsigned i;
2756 for (i = 0; i < p->num_bad_objects; i++)
2757 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2758 return 0;
2759 }
2760
2761 offset = find_pack_entry_one(sha1, p);
2762 if (!offset)
2763 return 0;
2764
2765 /*
2766 * We are about to tell the caller where they can locate the
2767 * requested object. We better make sure the packfile is
2768 * still here and can be accessed before supplying that
2769 * answer, as it may have been deleted since the index was
2770 * loaded!
2771 */
2772 if (!is_pack_valid(p))
2773 return 0;
2774 e->offset = offset;
2775 e->p = p;
2776 hashcpy(e->sha1, sha1);
2777 return 1;
2778 }
2779
2780 /*
2781 * Iff a pack file contains the object named by sha1, return true and
2782 * store its location to e.
2783 */
2784 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2785 {
2786 struct mru_entry *p;
2787
2788 prepare_packed_git();
2789 if (!packed_git)
2790 return 0;
2791
2792 for (p = packed_git_mru->head; p; p = p->next) {
2793 if (fill_pack_entry(sha1, e, p->item)) {
2794 mru_mark(packed_git_mru, p);
2795 return 1;
2796 }
2797 }
2798 return 0;
2799 }
2800
2801 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2802 struct packed_git *packs)
2803 {
2804 struct packed_git *p;
2805
2806 for (p = packs; p; p = p->next) {
2807 if (find_pack_entry_one(sha1, p))
2808 return p;
2809 }
2810 return NULL;
2811
2812 }
2813
2814 static int sha1_loose_object_info(const unsigned char *sha1,
2815 struct object_info *oi,
2816 int flags)
2817 {
2818 int status = 0;
2819 unsigned long mapsize;
2820 void *map;
2821 git_zstream stream;
2822 char hdr[32];
2823 struct strbuf hdrbuf = STRBUF_INIT;
2824
2825 if (oi->delta_base_sha1)
2826 hashclr(oi->delta_base_sha1);
2827
2828 /*
2829 * If we don't care about type or size, then we don't
2830 * need to look inside the object at all. Note that we
2831 * do not optimize out the stat call, even if the
2832 * caller doesn't care about the disk-size, since our
2833 * return value implicitly indicates whether the
2834 * object even exists.
2835 */
2836 if (!oi->typep && !oi->typename && !oi->sizep) {
2837 const char *path;
2838 struct stat st;
2839 if (stat_sha1_file(sha1, &st, &path) < 0)
2840 return -1;
2841 if (oi->disk_sizep)
2842 *oi->disk_sizep = st.st_size;
2843 return 0;
2844 }
2845
2846 map = map_sha1_file(sha1, &mapsize);
2847 if (!map)
2848 return -1;
2849 if (oi->disk_sizep)
2850 *oi->disk_sizep = mapsize;
2851 if ((flags & LOOKUP_UNKNOWN_OBJECT)) {
2852 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2853 status = error("unable to unpack %s header with --allow-unknown-type",
2854 sha1_to_hex(sha1));
2855 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2856 status = error("unable to unpack %s header",
2857 sha1_to_hex(sha1));
2858 if (status < 0)
2859 ; /* Do nothing */
2860 else if (hdrbuf.len) {
2861 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2862 status = error("unable to parse %s header with --allow-unknown-type",
2863 sha1_to_hex(sha1));
2864 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
2865 status = error("unable to parse %s header", sha1_to_hex(sha1));
2866 git_inflate_end(&stream);
2867 munmap(map, mapsize);
2868 if (status && oi->typep)
2869 *oi->typep = status;
2870 strbuf_release(&hdrbuf);
2871 return 0;
2872 }
2873
2874 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2875 {
2876 struct cached_object *co;
2877 struct pack_entry e;
2878 int rtype;
2879 enum object_type real_type;
2880 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2881
2882 co = find_cached_object(real);
2883 if (co) {
2884 if (oi->typep)
2885 *(oi->typep) = co->type;
2886 if (oi->sizep)
2887 *(oi->sizep) = co->size;
2888 if (oi->disk_sizep)
2889 *(oi->disk_sizep) = 0;
2890 if (oi->delta_base_sha1)
2891 hashclr(oi->delta_base_sha1);
2892 if (oi->typename)
2893 strbuf_addstr(oi->typename, typename(co->type));
2894 oi->whence = OI_CACHED;
2895 return 0;
2896 }
2897
2898 if (!find_pack_entry(real, &e)) {
2899 /* Most likely it's a loose object. */
2900 if (!sha1_loose_object_info(real, oi, flags)) {
2901 oi->whence = OI_LOOSE;
2902 return 0;
2903 }
2904
2905 /* Not a loose object; someone else may have just packed it. */
2906 reprepare_packed_git();
2907 if (!find_pack_entry(real, &e))
2908 return -1;
2909 }
2910
2911 /*
2912 * packed_object_info() does not follow the delta chain to
2913 * find out the real type, unless it is given oi->typep.
2914 */
2915 if (oi->typename && !oi->typep)
2916 oi->typep = &real_type;
2917
2918 rtype = packed_object_info(e.p, e.offset, oi);
2919 if (rtype < 0) {
2920 mark_bad_packed_object(e.p, real);
2921 if (oi->typep == &real_type)
2922 oi->typep = NULL;
2923 return sha1_object_info_extended(real, oi, 0);
2924 } else if (in_delta_base_cache(e.p, e.offset)) {
2925 oi->whence = OI_DBCACHED;
2926 } else {
2927 oi->whence = OI_PACKED;
2928 oi->u.packed.offset = e.offset;
2929 oi->u.packed.pack = e.p;
2930 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2931 rtype == OBJ_OFS_DELTA);
2932 }
2933 if (oi->typename)
2934 strbuf_addstr(oi->typename, typename(*oi->typep));
2935 if (oi->typep == &real_type)
2936 oi->typep = NULL;
2937
2938 return 0;
2939 }
2940
2941 /* returns enum object_type or negative */
2942 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2943 {
2944 enum object_type type;
2945 struct object_info oi = OBJECT_INFO_INIT;
2946
2947 oi.typep = &type;
2948 oi.sizep = sizep;
2949 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2950 return -1;
2951 return type;
2952 }
2953
2954 static void *read_packed_sha1(const unsigned char *sha1,
2955 enum object_type *type, unsigned long *size)
2956 {
2957 struct pack_entry e;
2958 void *data;
2959
2960 if (!find_pack_entry(sha1, &e))
2961 return NULL;
2962 data = cache_or_unpack_entry(e.p, e.offset, size, type);
2963 if (!data) {
2964 /*
2965 * We're probably in deep shit, but let's try to fetch
2966 * the required object anyway from another pack or loose.
2967 * This should happen only in the presence of a corrupted
2968 * pack, and is better than failing outright.
2969 */
2970 error("failed to read object %s at offset %"PRIuMAX" from %s",
2971 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2972 mark_bad_packed_object(e.p, sha1);
2973 data = read_object(sha1, type, size);
2974 }
2975 return data;
2976 }
2977
2978 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2979 unsigned char *sha1)
2980 {
2981 struct cached_object *co;
2982
2983 hash_sha1_file(buf, len, typename(type), sha1);
2984 if (has_sha1_file(sha1) || find_cached_object(sha1))
2985 return 0;
2986 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2987 co = &cached_objects[cached_object_nr++];
2988 co->size = len;
2989 co->type = type;
2990 co->buf = xmalloc(len);
2991 memcpy(co->buf, buf, len);
2992 hashcpy(co->sha1, sha1);
2993 return 0;
2994 }
2995
2996 static void *read_object(const unsigned char *sha1, enum object_type *type,
2997 unsigned long *size)
2998 {
2999 unsigned long mapsize;
3000 void *map, *buf;
3001 struct cached_object *co;
3002
3003 co = find_cached_object(sha1);
3004 if (co) {
3005 *type = co->type;
3006 *size = co->size;
3007 return xmemdupz(co->buf, co->size);
3008 }
3009
3010 buf = read_packed_sha1(sha1, type, size);
3011 if (buf)
3012 return buf;
3013 map = map_sha1_file(sha1, &mapsize);
3014 if (map) {
3015 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
3016 munmap(map, mapsize);
3017 return buf;
3018 }
3019 reprepare_packed_git();
3020 return read_packed_sha1(sha1, type, size);
3021 }
3022
3023 /*
3024 * This function dies on corrupt objects; the callers who want to
3025 * deal with them should arrange to call read_object() and give error
3026 * messages themselves.
3027 */
3028 void *read_sha1_file_extended(const unsigned char *sha1,
3029 enum object_type *type,
3030 unsigned long *size,
3031 unsigned flag)
3032 {
3033 void *data;
3034 const struct packed_git *p;
3035 const char *path;
3036 struct stat st;
3037 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
3038
3039 errno = 0;
3040 data = read_object(repl, type, size);
3041 if (data)
3042 return data;
3043
3044 if (errno && errno != ENOENT)
3045 die_errno("failed to read object %s", sha1_to_hex(sha1));
3046
3047 /* die if we replaced an object with one that does not exist */
3048 if (repl != sha1)
3049 die("replacement %s not found for %s",
3050 sha1_to_hex(repl), sha1_to_hex(sha1));
3051
3052 if (!stat_sha1_file(repl, &st, &path))
3053 die("loose object %s (stored in %s) is corrupt",
3054 sha1_to_hex(repl), path);
3055
3056 if ((p = has_packed_and_bad(repl)) != NULL)
3057 die("packed object %s (stored in %s) is corrupt",
3058 sha1_to_hex(repl), p->pack_name);
3059
3060 return NULL;
3061 }
3062
3063 void *read_object_with_reference(const unsigned char *sha1,
3064 const char *required_type_name,
3065 unsigned long *size,
3066 unsigned char *actual_sha1_return)
3067 {
3068 enum object_type type, required_type;
3069 void *buffer;
3070 unsigned long isize;
3071 unsigned char actual_sha1[20];
3072
3073 required_type = type_from_string(required_type_name);
3074 hashcpy(actual_sha1, sha1);
3075 while (1) {
3076 int ref_length = -1;
3077 const char *ref_type = NULL;
3078
3079 buffer = read_sha1_file(actual_sha1, &type, &isize);
3080 if (!buffer)
3081 return NULL;
3082 if (type == required_type) {
3083 *size = isize;
3084 if (actual_sha1_return)
3085 hashcpy(actual_sha1_return, actual_sha1);
3086 return buffer;
3087 }
3088 /* Handle references */
3089 else if (type == OBJ_COMMIT)
3090 ref_type = "tree ";
3091 else if (type == OBJ_TAG)
3092 ref_type = "object ";
3093 else {
3094 free(buffer);
3095 return NULL;
3096 }
3097 ref_length = strlen(ref_type);
3098
3099 if (ref_length + 40 > isize ||
3100 memcmp(buffer, ref_type, ref_length) ||
3101 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
3102 free(buffer);
3103 return NULL;
3104 }
3105 free(buffer);
3106 /* Now we have the ID of the referred-to object in
3107 * actual_sha1. Check again. */
3108 }
3109 }
3110
3111 static void write_sha1_file_prepare(const void *buf, unsigned long len,
3112 const char *type, unsigned char *sha1,
3113 char *hdr, int *hdrlen)
3114 {
3115 git_SHA_CTX c;
3116
3117 /* Generate the header */
3118 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
3119
3120 /* Sha1.. */
3121 git_SHA1_Init(&c);
3122 git_SHA1_Update(&c, hdr, *hdrlen);
3123 git_SHA1_Update(&c, buf, len);
3124 git_SHA1_Final(sha1, &c);
3125 }
3126
3127 /*
3128 * Move the just written object into its final resting place.
3129 */
3130 int finalize_object_file(const char *tmpfile, const char *filename)
3131 {
3132 int ret = 0;
3133
3134 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
3135 goto try_rename;
3136 else if (link(tmpfile, filename))
3137 ret = errno;
3138
3139 /*
3140 * Coda hack - coda doesn't like cross-directory links,
3141 * so we fall back to a rename, which will mean that it
3142 * won't be able to check collisions, but that's not a
3143 * big deal.
3144 *
3145 * The same holds for FAT formatted media.
3146 *
3147 * When this succeeds, we just return. We have nothing
3148 * left to unlink.
3149 */
3150 if (ret && ret != EEXIST) {
3151 try_rename:
3152 if (!rename(tmpfile, filename))
3153 goto out;
3154 ret = errno;
3155 }
3156 unlink_or_warn(tmpfile);
3157 if (ret) {
3158 if (ret != EEXIST) {
3159 return error_errno("unable to write sha1 filename %s", filename);
3160 }
3161 /* FIXME!!! Collision check here ? */
3162 }
3163
3164 out:
3165 if (adjust_shared_perm(filename))
3166 return error("unable to set permission to '%s'", filename);
3167 return 0;
3168 }
3169
3170 static int write_buffer(int fd, const void *buf, size_t len)
3171 {
3172 if (write_in_full(fd, buf, len) < 0)
3173 return error_errno("file write error");
3174 return 0;
3175 }
3176
3177 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
3178 unsigned char *sha1)
3179 {
3180 char hdr[32];
3181 int hdrlen = sizeof(hdr);
3182 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3183 return 0;
3184 }
3185
3186 /* Finalize a file on disk, and close it. */
3187 static void close_sha1_file(int fd)
3188 {
3189 if (fsync_object_files)
3190 fsync_or_die(fd, "sha1 file");
3191 if (close(fd) != 0)
3192 die_errno("error when closing sha1 file");
3193 }
3194
3195 /* Size of directory component, including the ending '/' */
3196 static inline int directory_size(const char *filename)
3197 {
3198 const char *s = strrchr(filename, '/');
3199 if (!s)
3200 return 0;
3201 return s - filename + 1;
3202 }
3203
3204 /*
3205 * This creates a temporary file in the same directory as the final
3206 * 'filename'
3207 *
3208 * We want to avoid cross-directory filename renames, because those
3209 * can have problems on various filesystems (FAT, NFS, Coda).
3210 */
3211 static int create_tmpfile(struct strbuf *tmp, const char *filename)
3212 {
3213 int fd, dirlen = directory_size(filename);
3214
3215 strbuf_reset(tmp);
3216 strbuf_add(tmp, filename, dirlen);
3217 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
3218 fd = git_mkstemp_mode(tmp->buf, 0444);
3219 if (fd < 0 && dirlen && errno == ENOENT) {
3220 /*
3221 * Make sure the directory exists; note that the contents
3222 * of the buffer are undefined after mkstemp returns an
3223 * error, so we have to rewrite the whole buffer from
3224 * scratch.
3225 */
3226 strbuf_reset(tmp);
3227 strbuf_add(tmp, filename, dirlen - 1);
3228 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
3229 return -1;
3230 if (adjust_shared_perm(tmp->buf))
3231 return -1;
3232
3233 /* Try again */
3234 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
3235 fd = git_mkstemp_mode(tmp->buf, 0444);
3236 }
3237 return fd;
3238 }
3239
3240 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
3241 const void *buf, unsigned long len, time_t mtime)
3242 {
3243 int fd, ret;
3244 unsigned char compressed[4096];
3245 git_zstream stream;
3246 git_SHA_CTX c;
3247 unsigned char parano_sha1[20];
3248 static struct strbuf tmp_file = STRBUF_INIT;
3249 const char *filename = sha1_file_name(sha1);
3250
3251 fd = create_tmpfile(&tmp_file, filename);
3252 if (fd < 0) {
3253 if (errno == EACCES)
3254 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3255 else
3256 return error_errno("unable to create temporary file");
3257 }
3258
3259 /* Set it up */
3260 git_deflate_init(&stream, zlib_compression_level);
3261 stream.next_out = compressed;
3262 stream.avail_out = sizeof(compressed);
3263 git_SHA1_Init(&c);
3264
3265 /* First header.. */
3266 stream.next_in = (unsigned char *)hdr;
3267 stream.avail_in = hdrlen;
3268 while (git_deflate(&stream, 0) == Z_OK)
3269 ; /* nothing */
3270 git_SHA1_Update(&c, hdr, hdrlen);
3271
3272 /* Then the data itself.. */
3273 stream.next_in = (void *)buf;
3274 stream.avail_in = len;
3275 do {
3276 unsigned char *in0 = stream.next_in;
3277 ret = git_deflate(&stream, Z_FINISH);
3278 git_SHA1_Update(&c, in0, stream.next_in - in0);
3279 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3280 die("unable to write sha1 file");
3281 stream.next_out = compressed;
3282 stream.avail_out = sizeof(compressed);
3283 } while (ret == Z_OK);
3284
3285 if (ret != Z_STREAM_END)
3286 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
3287 ret = git_deflate_end_gently(&stream);
3288 if (ret != Z_OK)
3289 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
3290 git_SHA1_Final(parano_sha1, &c);
3291 if (hashcmp(sha1, parano_sha1) != 0)
3292 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
3293
3294 close_sha1_file(fd);
3295
3296 if (mtime) {
3297 struct utimbuf utb;
3298 utb.actime = mtime;
3299 utb.modtime = mtime;
3300 if (utime(tmp_file.buf, &utb) < 0)
3301 warning_errno("failed utime() on %s", tmp_file.buf);
3302 }
3303
3304 return finalize_object_file(tmp_file.buf, filename);
3305 }
3306
3307 static int freshen_loose_object(const unsigned char *sha1)
3308 {
3309 return check_and_freshen(sha1, 1);
3310 }
3311
3312 static int freshen_packed_object(const unsigned char *sha1)
3313 {
3314 struct pack_entry e;
3315 if (!find_pack_entry(sha1, &e))
3316 return 0;
3317 if (e.p->freshened)
3318 return 1;
3319 if (!freshen_file(e.p->pack_name))
3320 return 0;
3321 e.p->freshened = 1;
3322 return 1;
3323 }
3324
3325 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3326 {
3327 char hdr[32];
3328 int hdrlen = sizeof(hdr);
3329
3330 /* Normally if we have it in the pack then we do not bother writing
3331 * it out into .git/objects/??/?{38} file.
3332 */
3333 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3334 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3335 return 0;
3336 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3337 }
3338
3339 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3340 unsigned char *sha1, unsigned flags)
3341 {
3342 char *header;
3343 int hdrlen, status = 0;
3344
3345 /* type string, SP, %lu of the length plus NUL must fit this */
3346 hdrlen = strlen(type) + 32;
3347 header = xmalloc(hdrlen);
3348 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3349
3350 if (!(flags & HASH_WRITE_OBJECT))
3351 goto cleanup;
3352 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3353 goto cleanup;
3354 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3355
3356 cleanup:
3357 free(header);
3358 return status;
3359 }
3360
3361 int force_object_loose(const unsigned char *sha1, time_t mtime)
3362 {
3363 void *buf;
3364 unsigned long len;
3365 enum object_type type;
3366 char hdr[32];
3367 int hdrlen;
3368 int ret;
3369
3370 if (has_loose_object(sha1))
3371 return 0;
3372 buf = read_packed_sha1(sha1, &type, &len);
3373 if (!buf)
3374 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3375 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
3376 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3377 free(buf);
3378
3379 return ret;
3380 }
3381
3382 int has_pack_index(const unsigned char *sha1)
3383 {
3384 struct stat st;
3385 if (stat(sha1_pack_index_name(sha1), &st))
3386 return 0;
3387 return 1;
3388 }
3389
3390 int has_sha1_pack(const unsigned char *sha1)
3391 {
3392 struct pack_entry e;
3393 return find_pack_entry(sha1, &e);
3394 }
3395
3396 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3397 {
3398 struct pack_entry e;
3399
3400 if (find_pack_entry(sha1, &e))
3401 return 1;
3402 if (has_loose_object(sha1))
3403 return 1;
3404 if (flags & HAS_SHA1_QUICK)
3405 return 0;
3406 reprepare_packed_git();
3407 return find_pack_entry(sha1, &e);
3408 }
3409
3410 int has_object_file(const struct object_id *oid)
3411 {
3412 return has_sha1_file(oid->hash);
3413 }
3414
3415 int has_object_file_with_flags(const struct object_id *oid, int flags)
3416 {
3417 return has_sha1_file_with_flags(oid->hash, flags);
3418 }
3419
3420 static void check_tree(const void *buf, size_t size)
3421 {
3422 struct tree_desc desc;
3423 struct name_entry entry;
3424
3425 init_tree_desc(&desc, buf, size);
3426 while (tree_entry(&desc, &entry))
3427 /* do nothing
3428 * tree_entry() will die() on malformed entries */
3429 ;
3430 }
3431
3432 static void check_commit(const void *buf, size_t size)
3433 {
3434 struct commit c;
3435 memset(&c, 0, sizeof(c));
3436 if (parse_commit_buffer(&c, buf, size))
3437 die("corrupt commit");
3438 }
3439
3440 static void check_tag(const void *buf, size_t size)
3441 {
3442 struct tag t;
3443 memset(&t, 0, sizeof(t));
3444 if (parse_tag_buffer(&t, buf, size))
3445 die("corrupt tag");
3446 }
3447
3448 static int index_mem(unsigned char *sha1, void *buf, size_t size,
3449 enum object_type type,
3450 const char *path, unsigned flags)
3451 {
3452 int ret, re_allocated = 0;
3453 int write_object = flags & HASH_WRITE_OBJECT;
3454
3455 if (!type)
3456 type = OBJ_BLOB;
3457
3458 /*
3459 * Convert blobs to git internal format
3460 */
3461 if ((type == OBJ_BLOB) && path) {
3462 struct strbuf nbuf = STRBUF_INIT;
3463 if (convert_to_git(path, buf, size, &nbuf,
3464 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3465 buf = strbuf_detach(&nbuf, &size);
3466 re_allocated = 1;
3467 }
3468 }
3469 if (flags & HASH_FORMAT_CHECK) {
3470 if (type == OBJ_TREE)
3471 check_tree(buf, size);
3472 if (type == OBJ_COMMIT)
3473 check_commit(buf, size);
3474 if (type == OBJ_TAG)
3475 check_tag(buf, size);
3476 }
3477
3478 if (write_object)
3479 ret = write_sha1_file(buf, size, typename(type), sha1);
3480 else
3481 ret = hash_sha1_file(buf, size, typename(type), sha1);
3482 if (re_allocated)
3483 free(buf);
3484 return ret;
3485 }
3486
3487 static int index_stream_convert_blob(unsigned char *sha1, int fd,
3488 const char *path, unsigned flags)
3489 {
3490 int ret;
3491 const int write_object = flags & HASH_WRITE_OBJECT;
3492 struct strbuf sbuf = STRBUF_INIT;
3493
3494 assert(path);
3495 assert(would_convert_to_git_filter_fd(path));
3496
3497 convert_to_git_filter_fd(path, fd, &sbuf,
3498 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3499
3500 if (write_object)
3501 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3502 sha1);
3503 else
3504 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3505 sha1);
3506 strbuf_release(&sbuf);
3507 return ret;
3508 }
3509
3510 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3511 const char *path, unsigned flags)
3512 {
3513 struct strbuf sbuf = STRBUF_INIT;
3514 int ret;
3515
3516 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3517 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3518 else
3519 ret = -1;
3520 strbuf_release(&sbuf);
3521 return ret;
3522 }
3523
3524 #define SMALL_FILE_SIZE (32*1024)
3525
3526 static int index_core(unsigned char *sha1, int fd, size_t size,
3527 enum object_type type, const char *path,
3528 unsigned flags)
3529 {
3530 int ret;
3531
3532 if (!size) {
3533 ret = index_mem(sha1, "", size, type, path, flags);
3534 } else if (size <= SMALL_FILE_SIZE) {
3535 char *buf = xmalloc(size);
3536 if (size == read_in_full(fd, buf, size))
3537 ret = index_mem(sha1, buf, size, type, path, flags);
3538 else
3539 ret = error_errno("short read");
3540 free(buf);
3541 } else {
3542 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3543 ret = index_mem(sha1, buf, size, type, path, flags);
3544 munmap(buf, size);
3545 }
3546 return ret;
3547 }
3548
3549 /*
3550 * This creates one packfile per large blob unless bulk-checkin
3551 * machinery is "plugged".
3552 *
3553 * This also bypasses the usual "convert-to-git" dance, and that is on
3554 * purpose. We could write a streaming version of the converting
3555 * functions and insert that before feeding the data to fast-import
3556 * (or equivalent in-core API described above). However, that is
3557 * somewhat complicated, as we do not know the size of the filter
3558 * result, which we need to know beforehand when writing a git object.
3559 * Since the primary motivation for trying to stream from the working
3560 * tree file and to avoid mmaping it in core is to deal with large
3561 * binary blobs, they generally do not want to get any conversion, and
3562 * callers should avoid this code path when filters are requested.
3563 */
3564 static int index_stream(unsigned char *sha1, int fd, size_t size,
3565 enum object_type type, const char *path,
3566 unsigned flags)
3567 {
3568 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3569 }
3570
3571 int index_fd(unsigned char *sha1, int fd, struct stat *st,
3572 enum object_type type, const char *path, unsigned flags)
3573 {
3574 int ret;
3575
3576 /*
3577 * Call xsize_t() only when needed to avoid potentially unnecessary
3578 * die() for large files.
3579 */
3580 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3581 ret = index_stream_convert_blob(sha1, fd, path, flags);
3582 else if (!S_ISREG(st->st_mode))
3583 ret = index_pipe(sha1, fd, type, path, flags);
3584 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3585 (path && would_convert_to_git(path)))
3586 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3587 flags);
3588 else
3589 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3590 flags);
3591 close(fd);
3592 return ret;
3593 }
3594
3595 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3596 {
3597 int fd;
3598 struct strbuf sb = STRBUF_INIT;
3599
3600 switch (st->st_mode & S_IFMT) {
3601 case S_IFREG:
3602 fd = open(path, O_RDONLY);
3603 if (fd < 0)
3604 return error_errno("open(\"%s\")", path);
3605 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3606 return error("%s: failed to insert into database",
3607 path);
3608 break;
3609 case S_IFLNK:
3610 if (strbuf_readlink(&sb, path, st->st_size))
3611 return error_errno("readlink(\"%s\")", path);
3612 if (!(flags & HASH_WRITE_OBJECT))
3613 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3614 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3615 return error("%s: failed to insert into database",
3616 path);
3617 strbuf_release(&sb);
3618 break;
3619 case S_IFDIR:
3620 return resolve_gitlink_ref(path, "HEAD", sha1);
3621 default:
3622 return error("%s: unsupported file type", path);
3623 }
3624 return 0;
3625 }
3626
3627 int read_pack_header(int fd, struct pack_header *header)
3628 {
3629 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3630 /* "eof before pack header was fully read" */
3631 return PH_ERROR_EOF;
3632
3633 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3634 /* "protocol error (pack signature mismatch detected)" */
3635 return PH_ERROR_PACK_SIGNATURE;
3636 if (!pack_version_ok(header->hdr_version))
3637 /* "protocol error (pack version unsupported)" */
3638 return PH_ERROR_PROTOCOL;
3639 return 0;
3640 }
3641
3642 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3643 {
3644 enum object_type type = sha1_object_info(sha1, NULL);
3645 if (type < 0)
3646 die("%s is not a valid object", sha1_to_hex(sha1));
3647 if (type != expect)
3648 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3649 typename(expect));
3650 }
3651
3652 static int for_each_file_in_obj_subdir(int subdir_nr,
3653 struct strbuf *path,
3654 each_loose_object_fn obj_cb,
3655 each_loose_cruft_fn cruft_cb,
3656 each_loose_subdir_fn subdir_cb,
3657 void *data)
3658 {
3659 size_t baselen = path->len;
3660 DIR *dir = opendir(path->buf);
3661 struct dirent *de;
3662 int r = 0;
3663
3664 if (!dir) {
3665 if (errno == ENOENT)
3666 return 0;
3667 return error_errno("unable to open %s", path->buf);
3668 }
3669
3670 while ((de = readdir(dir))) {
3671 if (is_dot_or_dotdot(de->d_name))
3672 continue;
3673
3674 strbuf_setlen(path, baselen);
3675 strbuf_addf(path, "/%s", de->d_name);
3676
3677 if (strlen(de->d_name) == 38) {
3678 char hex[41];
3679 unsigned char sha1[20];
3680
3681 snprintf(hex, sizeof(hex), "%02x%s",
3682 subdir_nr, de->d_name);
3683 if (!get_sha1_hex(hex, sha1)) {
3684 if (obj_cb) {
3685 r = obj_cb(sha1, path->buf, data);
3686 if (r)
3687 break;
3688 }
3689 continue;
3690 }
3691 }
3692
3693 if (cruft_cb) {
3694 r = cruft_cb(de->d_name, path->buf, data);
3695 if (r)
3696 break;
3697 }
3698 }
3699 closedir(dir);
3700
3701 strbuf_setlen(path, baselen);
3702 if (!r && subdir_cb)
3703 r = subdir_cb(subdir_nr, path->buf, data);
3704
3705 return r;
3706 }
3707
3708 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3709 each_loose_object_fn obj_cb,
3710 each_loose_cruft_fn cruft_cb,
3711 each_loose_subdir_fn subdir_cb,
3712 void *data)
3713 {
3714 size_t baselen = path->len;
3715 int r = 0;
3716 int i;
3717
3718 for (i = 0; i < 256; i++) {
3719 strbuf_addf(path, "/%02x", i);
3720 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3721 subdir_cb, data);
3722 strbuf_setlen(path, baselen);
3723 if (r)
3724 break;
3725 }
3726
3727 return r;
3728 }
3729
3730 int for_each_loose_file_in_objdir(const char *path,
3731 each_loose_object_fn obj_cb,
3732 each_loose_cruft_fn cruft_cb,
3733 each_loose_subdir_fn subdir_cb,
3734 void *data)
3735 {
3736 struct strbuf buf = STRBUF_INIT;
3737 int r;
3738
3739 strbuf_addstr(&buf, path);
3740 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3741 subdir_cb, data);
3742 strbuf_release(&buf);
3743
3744 return r;
3745 }
3746
3747 struct loose_alt_odb_data {
3748 each_loose_object_fn *cb;
3749 void *data;
3750 };
3751
3752 static int loose_from_alt_odb(struct alternate_object_database *alt,
3753 void *vdata)
3754 {
3755 struct loose_alt_odb_data *data = vdata;
3756 struct strbuf buf = STRBUF_INIT;
3757 int r;
3758
3759 strbuf_addstr(&buf, alt->path);
3760 r = for_each_loose_file_in_objdir_buf(&buf,
3761 data->cb, NULL, NULL,
3762 data->data);
3763 strbuf_release(&buf);
3764 return r;
3765 }
3766
3767 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3768 {
3769 struct loose_alt_odb_data alt;
3770 int r;
3771
3772 r = for_each_loose_file_in_objdir(get_object_directory(),
3773 cb, NULL, NULL, data);
3774 if (r)
3775 return r;
3776
3777 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3778 return 0;
3779
3780 alt.cb = cb;
3781 alt.data = data;
3782 return foreach_alt_odb(loose_from_alt_odb, &alt);
3783 }
3784
3785 static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3786 {
3787 uint32_t i;
3788 int r = 0;
3789
3790 for (i = 0; i < p->num_objects; i++) {
3791 const unsigned char *sha1 = nth_packed_object_sha1(p, i);
3792
3793 if (!sha1)
3794 return error("unable to get sha1 of object %u in %s",
3795 i, p->pack_name);
3796
3797 r = cb(sha1, p, i, data);
3798 if (r)
3799 break;
3800 }
3801 return r;
3802 }
3803
3804 int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3805 {
3806 struct packed_git *p;
3807 int r = 0;
3808 int pack_errors = 0;
3809
3810 prepare_packed_git();
3811 for (p = packed_git; p; p = p->next) {
3812 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3813 continue;
3814 if (open_pack_index(p)) {
3815 pack_errors = 1;
3816 continue;
3817 }
3818 r = for_each_object_in_pack(p, cb, data);
3819 if (r)
3820 break;
3821 }
3822 return r ? r : pack_errors;
3823 }
3824
3825 static int check_stream_sha1(git_zstream *stream,
3826 const char *hdr,
3827 unsigned long size,
3828 const char *path,
3829 const unsigned char *expected_sha1)
3830 {
3831 git_SHA_CTX c;
3832 unsigned char real_sha1[GIT_SHA1_RAWSZ];
3833 unsigned char buf[4096];
3834 unsigned long total_read;
3835 int status = Z_OK;
3836
3837 git_SHA1_Init(&c);
3838 git_SHA1_Update(&c, hdr, stream->total_out);
3839
3840 /*
3841 * We already read some bytes into hdr, but the ones up to the NUL
3842 * do not count against the object's content size.
3843 */
3844 total_read = stream->total_out - strlen(hdr) - 1;
3845
3846 /*
3847 * This size comparison must be "<=" to read the final zlib packets;
3848 * see the comment in unpack_sha1_rest for details.
3849 */
3850 while (total_read <= size &&
3851 (status == Z_OK || status == Z_BUF_ERROR)) {
3852 stream->next_out = buf;
3853 stream->avail_out = sizeof(buf);
3854 if (size - total_read < stream->avail_out)
3855 stream->avail_out = size - total_read;
3856 status = git_inflate(stream, Z