Merge branch 'jt/sha1-file-cleanup'
[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 "config.h"
11 #include "string-list.h"
12 #include "lockfile.h"
13 #include "delta.h"
14 #include "pack.h"
15 #include "blob.h"
16 #include "commit.h"
17 #include "run-command.h"
18 #include "tag.h"
19 #include "tree.h"
20 #include "tree-walk.h"
21 #include "refs.h"
22 #include "pack-revindex.h"
23 #include "sha1-lookup.h"
24 #include "bulk-checkin.h"
25 #include "streaming.h"
26 #include "dir.h"
27 #include "mru.h"
28 #include "list.h"
29 #include "mergesort.h"
30 #include "quote.h"
31
32 #define SZ_FMT PRIuMAX
33 static inline uintmax_t sz_fmt(size_t s) { return s; }
34
35 const unsigned char null_sha1[20];
36 const struct object_id null_oid;
37 const struct object_id empty_tree_oid = {
38 EMPTY_TREE_SHA1_BIN_LITERAL
39 };
40 const struct object_id empty_blob_oid = {
41 EMPTY_BLOB_SHA1_BIN_LITERAL
42 };
43
44 /*
45 * This is meant to hold a *small* number of objects that you would
46 * want read_sha1_file() to be able to return, but yet you do not want
47 * to write them into the object store (e.g. a browse-only
48 * application).
49 */
50 static struct cached_object {
51 unsigned char sha1[20];
52 enum object_type type;
53 void *buf;
54 unsigned long size;
55 } *cached_objects;
56 static int cached_object_nr, cached_object_alloc;
57
58 static struct cached_object empty_tree = {
59 EMPTY_TREE_SHA1_BIN_LITERAL,
60 OBJ_TREE,
61 "",
62 0
63 };
64
65 static struct cached_object *find_cached_object(const unsigned char *sha1)
66 {
67 int i;
68 struct cached_object *co = cached_objects;
69
70 for (i = 0; i < cached_object_nr; i++, co++) {
71 if (!hashcmp(co->sha1, sha1))
72 return co;
73 }
74 if (!hashcmp(sha1, empty_tree.sha1))
75 return &empty_tree;
76 return NULL;
77 }
78
79 int mkdir_in_gitdir(const char *path)
80 {
81 if (mkdir(path, 0777)) {
82 int saved_errno = errno;
83 struct stat st;
84 struct strbuf sb = STRBUF_INIT;
85
86 if (errno != EEXIST)
87 return -1;
88 /*
89 * Are we looking at a path in a symlinked worktree
90 * whose original repository does not yet have it?
91 * e.g. .git/rr-cache pointing at its original
92 * repository in which the user hasn't performed any
93 * conflict resolution yet?
94 */
95 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
96 strbuf_readlink(&sb, path, st.st_size) ||
97 !is_absolute_path(sb.buf) ||
98 mkdir(sb.buf, 0777)) {
99 strbuf_release(&sb);
100 errno = saved_errno;
101 return -1;
102 }
103 strbuf_release(&sb);
104 }
105 return adjust_shared_perm(path);
106 }
107
108 enum scld_error safe_create_leading_directories(char *path)
109 {
110 char *next_component = path + offset_1st_component(path);
111 enum scld_error ret = SCLD_OK;
112
113 while (ret == SCLD_OK && next_component) {
114 struct stat st;
115 char *slash = next_component, slash_character;
116
117 while (*slash && !is_dir_sep(*slash))
118 slash++;
119
120 if (!*slash)
121 break;
122
123 next_component = slash + 1;
124 while (is_dir_sep(*next_component))
125 next_component++;
126 if (!*next_component)
127 break;
128
129 slash_character = *slash;
130 *slash = '\0';
131 if (!stat(path, &st)) {
132 /* path exists */
133 if (!S_ISDIR(st.st_mode)) {
134 errno = ENOTDIR;
135 ret = SCLD_EXISTS;
136 }
137 } else if (mkdir(path, 0777)) {
138 if (errno == EEXIST &&
139 !stat(path, &st) && S_ISDIR(st.st_mode))
140 ; /* somebody created it since we checked */
141 else if (errno == ENOENT)
142 /*
143 * Either mkdir() failed because
144 * somebody just pruned the containing
145 * directory, or stat() failed because
146 * the file that was in our way was
147 * just removed. Either way, inform
148 * the caller that it might be worth
149 * trying again:
150 */
151 ret = SCLD_VANISHED;
152 else
153 ret = SCLD_FAILED;
154 } else if (adjust_shared_perm(path)) {
155 ret = SCLD_PERMS;
156 }
157 *slash = slash_character;
158 }
159 return ret;
160 }
161
162 enum scld_error safe_create_leading_directories_const(const char *path)
163 {
164 int save_errno;
165 /* path points to cache entries, so xstrdup before messing with it */
166 char *buf = xstrdup(path);
167 enum scld_error result = safe_create_leading_directories(buf);
168
169 save_errno = errno;
170 free(buf);
171 errno = save_errno;
172 return result;
173 }
174
175 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
176 {
177 /*
178 * The number of times we will try to remove empty directories
179 * in the way of path. This is only 1 because if another
180 * process is racily creating directories that conflict with
181 * us, we don't want to fight against them.
182 */
183 int remove_directories_remaining = 1;
184
185 /*
186 * The number of times that we will try to create the
187 * directories containing path. We are willing to attempt this
188 * more than once, because another process could be trying to
189 * clean up empty directories at the same time as we are
190 * trying to create them.
191 */
192 int create_directories_remaining = 3;
193
194 /* A scratch copy of path, filled lazily if we need it: */
195 struct strbuf path_copy = STRBUF_INIT;
196
197 int ret, save_errno;
198
199 /* Sanity check: */
200 assert(*path);
201
202 retry_fn:
203 ret = fn(path, cb);
204 save_errno = errno;
205 if (!ret)
206 goto out;
207
208 if (errno == EISDIR && remove_directories_remaining-- > 0) {
209 /*
210 * A directory is in the way. Maybe it is empty; try
211 * to remove it:
212 */
213 if (!path_copy.len)
214 strbuf_addstr(&path_copy, path);
215
216 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
217 goto retry_fn;
218 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
219 /*
220 * Maybe the containing directory didn't exist, or
221 * maybe it was just deleted by a process that is
222 * racing with us to clean up empty directories. Try
223 * to create it:
224 */
225 enum scld_error scld_result;
226
227 if (!path_copy.len)
228 strbuf_addstr(&path_copy, path);
229
230 do {
231 scld_result = safe_create_leading_directories(path_copy.buf);
232 if (scld_result == SCLD_OK)
233 goto retry_fn;
234 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
235 }
236
237 out:
238 strbuf_release(&path_copy);
239 errno = save_errno;
240 return ret;
241 }
242
243 static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
244 {
245 int i;
246 for (i = 0; i < 20; i++) {
247 static char hex[] = "0123456789abcdef";
248 unsigned int val = sha1[i];
249 strbuf_addch(buf, hex[val >> 4]);
250 strbuf_addch(buf, hex[val & 0xf]);
251 if (!i)
252 strbuf_addch(buf, '/');
253 }
254 }
255
256 const char *sha1_file_name(const unsigned char *sha1)
257 {
258 static struct strbuf buf = STRBUF_INIT;
259
260 strbuf_reset(&buf);
261 strbuf_addf(&buf, "%s/", get_object_directory());
262
263 fill_sha1_path(&buf, sha1);
264 return buf.buf;
265 }
266
267 struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
268 {
269 strbuf_setlen(&alt->scratch, alt->base_len);
270 return &alt->scratch;
271 }
272
273 static const char *alt_sha1_path(struct alternate_object_database *alt,
274 const unsigned char *sha1)
275 {
276 struct strbuf *buf = alt_scratch_buf(alt);
277 fill_sha1_path(buf, sha1);
278 return buf->buf;
279 }
280
281 char *odb_pack_name(struct strbuf *buf,
282 const unsigned char *sha1,
283 const char *ext)
284 {
285 strbuf_reset(buf);
286 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
287 sha1_to_hex(sha1), ext);
288 return buf->buf;
289 }
290
291 char *sha1_pack_name(const unsigned char *sha1)
292 {
293 static struct strbuf buf = STRBUF_INIT;
294 return odb_pack_name(&buf, sha1, "pack");
295 }
296
297 char *sha1_pack_index_name(const unsigned char *sha1)
298 {
299 static struct strbuf buf = STRBUF_INIT;
300 return odb_pack_name(&buf, sha1, "idx");
301 }
302
303 struct alternate_object_database *alt_odb_list;
304 static struct alternate_object_database **alt_odb_tail;
305
306 /*
307 * Return non-zero iff the path is usable as an alternate object database.
308 */
309 static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
310 {
311 struct alternate_object_database *alt;
312
313 /* Detect cases where alternate disappeared */
314 if (!is_directory(path->buf)) {
315 error("object directory %s does not exist; "
316 "check .git/objects/info/alternates.",
317 path->buf);
318 return 0;
319 }
320
321 /*
322 * Prevent the common mistake of listing the same
323 * thing twice, or object directory itself.
324 */
325 for (alt = alt_odb_list; alt; alt = alt->next) {
326 if (!fspathcmp(path->buf, alt->path))
327 return 0;
328 }
329 if (!fspathcmp(path->buf, normalized_objdir))
330 return 0;
331
332 return 1;
333 }
334
335 /*
336 * Prepare alternate object database registry.
337 *
338 * The variable alt_odb_list points at the list of struct
339 * alternate_object_database. The elements on this list come from
340 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
341 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
342 * whose contents is similar to that environment variable but can be
343 * LF separated. Its base points at a statically allocated buffer that
344 * contains "/the/directory/corresponding/to/.git/objects/...", while
345 * its name points just after the slash at the end of ".git/objects/"
346 * in the example above, and has enough space to hold 40-byte hex
347 * SHA1, an extra slash for the first level indirection, and the
348 * terminating NUL.
349 */
350 static int link_alt_odb_entry(const char *entry, const char *relative_base,
351 int depth, const char *normalized_objdir)
352 {
353 struct alternate_object_database *ent;
354 struct strbuf pathbuf = STRBUF_INIT;
355
356 if (!is_absolute_path(entry) && relative_base) {
357 strbuf_realpath(&pathbuf, relative_base, 1);
358 strbuf_addch(&pathbuf, '/');
359 }
360 strbuf_addstr(&pathbuf, entry);
361
362 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
363 error("unable to normalize alternate object path: %s",
364 pathbuf.buf);
365 strbuf_release(&pathbuf);
366 return -1;
367 }
368
369 /*
370 * The trailing slash after the directory name is given by
371 * this function at the end. Remove duplicates.
372 */
373 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
374 strbuf_setlen(&pathbuf, pathbuf.len - 1);
375
376 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
377 strbuf_release(&pathbuf);
378 return -1;
379 }
380
381 ent = alloc_alt_odb(pathbuf.buf);
382
383 /* add the alternate entry */
384 *alt_odb_tail = ent;
385 alt_odb_tail = &(ent->next);
386 ent->next = NULL;
387
388 /* recursively add alternates */
389 read_info_alternates(pathbuf.buf, depth + 1);
390
391 strbuf_release(&pathbuf);
392 return 0;
393 }
394
395 static const char *parse_alt_odb_entry(const char *string,
396 int sep,
397 struct strbuf *out)
398 {
399 const char *end;
400
401 strbuf_reset(out);
402
403 if (*string == '#') {
404 /* comment; consume up to next separator */
405 end = strchrnul(string, sep);
406 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
407 /*
408 * quoted path; unquote_c_style has copied the
409 * data for us and set "end". Broken quoting (e.g.,
410 * an entry that doesn't end with a quote) falls
411 * back to the unquoted case below.
412 */
413 } else {
414 /* normal, unquoted path */
415 end = strchrnul(string, sep);
416 strbuf_add(out, string, end - string);
417 }
418
419 if (*end)
420 end++;
421 return end;
422 }
423
424 static void link_alt_odb_entries(const char *alt, int len, int sep,
425 const char *relative_base, int depth)
426 {
427 struct strbuf objdirbuf = STRBUF_INIT;
428 struct strbuf entry = STRBUF_INIT;
429
430 if (depth > 5) {
431 error("%s: ignoring alternate object stores, nesting too deep.",
432 relative_base);
433 return;
434 }
435
436 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
437 if (strbuf_normalize_path(&objdirbuf) < 0)
438 die("unable to normalize object directory: %s",
439 objdirbuf.buf);
440
441 while (*alt) {
442 alt = parse_alt_odb_entry(alt, sep, &entry);
443 if (!entry.len)
444 continue;
445 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
446 }
447 strbuf_release(&entry);
448 strbuf_release(&objdirbuf);
449 }
450
451 void read_info_alternates(const char * relative_base, int depth)
452 {
453 char *map;
454 size_t mapsz;
455 struct stat st;
456 char *path;
457 int fd;
458
459 path = xstrfmt("%s/info/alternates", relative_base);
460 fd = git_open(path);
461 free(path);
462 if (fd < 0)
463 return;
464 if (fstat(fd, &st) || (st.st_size == 0)) {
465 close(fd);
466 return;
467 }
468 mapsz = xsize_t(st.st_size);
469 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
470 close(fd);
471
472 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
473
474 munmap(map, mapsz);
475 }
476
477 struct alternate_object_database *alloc_alt_odb(const char *dir)
478 {
479 struct alternate_object_database *ent;
480
481 FLEX_ALLOC_STR(ent, path, dir);
482 strbuf_init(&ent->scratch, 0);
483 strbuf_addf(&ent->scratch, "%s/", dir);
484 ent->base_len = ent->scratch.len;
485
486 return ent;
487 }
488
489 void add_to_alternates_file(const char *reference)
490 {
491 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
492 char *alts = git_pathdup("objects/info/alternates");
493 FILE *in, *out;
494
495 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
496 out = fdopen_lock_file(lock, "w");
497 if (!out)
498 die_errno("unable to fdopen alternates lockfile");
499
500 in = fopen(alts, "r");
501 if (in) {
502 struct strbuf line = STRBUF_INIT;
503 int found = 0;
504
505 while (strbuf_getline(&line, in) != EOF) {
506 if (!strcmp(reference, line.buf)) {
507 found = 1;
508 break;
509 }
510 fprintf_or_die(out, "%s\n", line.buf);
511 }
512
513 strbuf_release(&line);
514 fclose(in);
515
516 if (found) {
517 rollback_lock_file(lock);
518 lock = NULL;
519 }
520 }
521 else if (errno != ENOENT)
522 die_errno("unable to read alternates file");
523
524 if (lock) {
525 fprintf_or_die(out, "%s\n", reference);
526 if (commit_lock_file(lock))
527 die_errno("unable to move new alternates file into place");
528 if (alt_odb_tail)
529 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
530 }
531 free(alts);
532 }
533
534 void add_to_alternates_memory(const char *reference)
535 {
536 /*
537 * Make sure alternates are initialized, or else our entry may be
538 * overwritten when they are.
539 */
540 prepare_alt_odb();
541
542 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
543 }
544
545 /*
546 * Compute the exact path an alternate is at and returns it. In case of
547 * error NULL is returned and the human readable error is added to `err`
548 * `path` may be relative and should point to $GITDIR.
549 * `err` must not be null.
550 */
551 char *compute_alternate_path(const char *path, struct strbuf *err)
552 {
553 char *ref_git = NULL;
554 const char *repo, *ref_git_s;
555 int seen_error = 0;
556
557 ref_git_s = real_path_if_valid(path);
558 if (!ref_git_s) {
559 seen_error = 1;
560 strbuf_addf(err, _("path '%s' does not exist"), path);
561 goto out;
562 } else
563 /*
564 * Beware: read_gitfile(), real_path() and mkpath()
565 * return static buffer
566 */
567 ref_git = xstrdup(ref_git_s);
568
569 repo = read_gitfile(ref_git);
570 if (!repo)
571 repo = read_gitfile(mkpath("%s/.git", ref_git));
572 if (repo) {
573 free(ref_git);
574 ref_git = xstrdup(repo);
575 }
576
577 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
578 char *ref_git_git = mkpathdup("%s/.git", ref_git);
579 free(ref_git);
580 ref_git = ref_git_git;
581 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
582 struct strbuf sb = STRBUF_INIT;
583 seen_error = 1;
584 if (get_common_dir(&sb, ref_git)) {
585 strbuf_addf(err,
586 _("reference repository '%s' as a linked "
587 "checkout is not supported yet."),
588 path);
589 goto out;
590 }
591
592 strbuf_addf(err, _("reference repository '%s' is not a "
593 "local repository."), path);
594 goto out;
595 }
596
597 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
598 strbuf_addf(err, _("reference repository '%s' is shallow"),
599 path);
600 seen_error = 1;
601 goto out;
602 }
603
604 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
605 strbuf_addf(err,
606 _("reference repository '%s' is grafted"),
607 path);
608 seen_error = 1;
609 goto out;
610 }
611
612 out:
613 if (seen_error) {
614 FREE_AND_NULL(ref_git);
615 }
616
617 return ref_git;
618 }
619
620 int foreach_alt_odb(alt_odb_fn fn, void *cb)
621 {
622 struct alternate_object_database *ent;
623 int r = 0;
624
625 prepare_alt_odb();
626 for (ent = alt_odb_list; ent; ent = ent->next) {
627 r = fn(ent, cb);
628 if (r)
629 break;
630 }
631 return r;
632 }
633
634 void prepare_alt_odb(void)
635 {
636 const char *alt;
637
638 if (alt_odb_tail)
639 return;
640
641 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
642 if (!alt) alt = "";
643
644 alt_odb_tail = &alt_odb_list;
645 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
646
647 read_info_alternates(get_object_directory(), 0);
648 }
649
650 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
651 static int freshen_file(const char *fn)
652 {
653 struct utimbuf t;
654 t.actime = t.modtime = time(NULL);
655 return !utime(fn, &t);
656 }
657
658 /*
659 * All of the check_and_freshen functions return 1 if the file exists and was
660 * freshened (if freshening was requested), 0 otherwise. If they return
661 * 0, you should not assume that it is safe to skip a write of the object (it
662 * either does not exist on disk, or has a stale mtime and may be subject to
663 * pruning).
664 */
665 int check_and_freshen_file(const char *fn, int freshen)
666 {
667 if (access(fn, F_OK))
668 return 0;
669 if (freshen && !freshen_file(fn))
670 return 0;
671 return 1;
672 }
673
674 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
675 {
676 return check_and_freshen_file(sha1_file_name(sha1), freshen);
677 }
678
679 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
680 {
681 struct alternate_object_database *alt;
682 prepare_alt_odb();
683 for (alt = alt_odb_list; alt; alt = alt->next) {
684 const char *path = alt_sha1_path(alt, sha1);
685 if (check_and_freshen_file(path, freshen))
686 return 1;
687 }
688 return 0;
689 }
690
691 static int check_and_freshen(const unsigned char *sha1, int freshen)
692 {
693 return check_and_freshen_local(sha1, freshen) ||
694 check_and_freshen_nonlocal(sha1, freshen);
695 }
696
697 int has_loose_object_nonlocal(const unsigned char *sha1)
698 {
699 return check_and_freshen_nonlocal(sha1, 0);
700 }
701
702 static int has_loose_object(const unsigned char *sha1)
703 {
704 return check_and_freshen(sha1, 0);
705 }
706
707 static unsigned int pack_used_ctr;
708 static unsigned int pack_mmap_calls;
709 static unsigned int peak_pack_open_windows;
710 static unsigned int pack_open_windows;
711 static unsigned int pack_open_fds;
712 static unsigned int pack_max_fds;
713 static size_t peak_pack_mapped;
714 static size_t pack_mapped;
715 struct packed_git *packed_git;
716
717 static struct mru packed_git_mru_storage;
718 struct mru *packed_git_mru = &packed_git_mru_storage;
719
720 void pack_report(void)
721 {
722 fprintf(stderr,
723 "pack_report: getpagesize() = %10" SZ_FMT "\n"
724 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
725 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
726 sz_fmt(getpagesize()),
727 sz_fmt(packed_git_window_size),
728 sz_fmt(packed_git_limit));
729 fprintf(stderr,
730 "pack_report: pack_used_ctr = %10u\n"
731 "pack_report: pack_mmap_calls = %10u\n"
732 "pack_report: pack_open_windows = %10u / %10u\n"
733 "pack_report: pack_mapped = "
734 "%10" SZ_FMT " / %10" SZ_FMT "\n",
735 pack_used_ctr,
736 pack_mmap_calls,
737 pack_open_windows, peak_pack_open_windows,
738 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
739 }
740
741 /*
742 * Open and mmap the index file at path, perform a couple of
743 * consistency checks, then record its information to p. Return 0 on
744 * success.
745 */
746 static int check_packed_git_idx(const char *path, struct packed_git *p)
747 {
748 void *idx_map;
749 struct pack_idx_header *hdr;
750 size_t idx_size;
751 uint32_t version, nr, i, *index;
752 int fd = git_open(path);
753 struct stat st;
754
755 if (fd < 0)
756 return -1;
757 if (fstat(fd, &st)) {
758 close(fd);
759 return -1;
760 }
761 idx_size = xsize_t(st.st_size);
762 if (idx_size < 4 * 256 + 20 + 20) {
763 close(fd);
764 return error("index file %s is too small", path);
765 }
766 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
767 close(fd);
768
769 hdr = idx_map;
770 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
771 version = ntohl(hdr->idx_version);
772 if (version < 2 || version > 2) {
773 munmap(idx_map, idx_size);
774 return error("index file %s is version %"PRIu32
775 " and is not supported by this binary"
776 " (try upgrading GIT to a newer version)",
777 path, version);
778 }
779 } else
780 version = 1;
781
782 nr = 0;
783 index = idx_map;
784 if (version > 1)
785 index += 2; /* skip index header */
786 for (i = 0; i < 256; i++) {
787 uint32_t n = ntohl(index[i]);
788 if (n < nr) {
789 munmap(idx_map, idx_size);
790 return error("non-monotonic index %s", path);
791 }
792 nr = n;
793 }
794
795 if (version == 1) {
796 /*
797 * Total size:
798 * - 256 index entries 4 bytes each
799 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
800 * - 20-byte SHA1 of the packfile
801 * - 20-byte SHA1 file checksum
802 */
803 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
804 munmap(idx_map, idx_size);
805 return error("wrong index v1 file size in %s", path);
806 }
807 } else if (version == 2) {
808 /*
809 * Minimum size:
810 * - 8 bytes of header
811 * - 256 index entries 4 bytes each
812 * - 20-byte sha1 entry * nr
813 * - 4-byte crc entry * nr
814 * - 4-byte offset entry * nr
815 * - 20-byte SHA1 of the packfile
816 * - 20-byte SHA1 file checksum
817 * And after the 4-byte offset table might be a
818 * variable sized table containing 8-byte entries
819 * for offsets larger than 2^31.
820 */
821 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
822 unsigned long max_size = min_size;
823 if (nr)
824 max_size += (nr - 1)*8;
825 if (idx_size < min_size || idx_size > max_size) {
826 munmap(idx_map, idx_size);
827 return error("wrong index v2 file size in %s", path);
828 }
829 if (idx_size != min_size &&
830 /*
831 * make sure we can deal with large pack offsets.
832 * 31-bit signed offset won't be enough, neither
833 * 32-bit unsigned one will be.
834 */
835 (sizeof(off_t) <= 4)) {
836 munmap(idx_map, idx_size);
837 return error("pack too large for current definition of off_t in %s", path);
838 }
839 }
840
841 p->index_version = version;
842 p->index_data = idx_map;
843 p->index_size = idx_size;
844 p->num_objects = nr;
845 return 0;
846 }
847
848 int open_pack_index(struct packed_git *p)
849 {
850 char *idx_name;
851 size_t len;
852 int ret;
853
854 if (p->index_data)
855 return 0;
856
857 if (!strip_suffix(p->pack_name, ".pack", &len))
858 die("BUG: pack_name does not end in .pack");
859 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
860 ret = check_packed_git_idx(idx_name, p);
861 free(idx_name);
862 return ret;
863 }
864
865 static void scan_windows(struct packed_git *p,
866 struct packed_git **lru_p,
867 struct pack_window **lru_w,
868 struct pack_window **lru_l)
869 {
870 struct pack_window *w, *w_l;
871
872 for (w_l = NULL, w = p->windows; w; w = w->next) {
873 if (!w->inuse_cnt) {
874 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
875 *lru_p = p;
876 *lru_w = w;
877 *lru_l = w_l;
878 }
879 }
880 w_l = w;
881 }
882 }
883
884 static int unuse_one_window(struct packed_git *current)
885 {
886 struct packed_git *p, *lru_p = NULL;
887 struct pack_window *lru_w = NULL, *lru_l = NULL;
888
889 if (current)
890 scan_windows(current, &lru_p, &lru_w, &lru_l);
891 for (p = packed_git; p; p = p->next)
892 scan_windows(p, &lru_p, &lru_w, &lru_l);
893 if (lru_p) {
894 munmap(lru_w->base, lru_w->len);
895 pack_mapped -= lru_w->len;
896 if (lru_l)
897 lru_l->next = lru_w->next;
898 else
899 lru_p->windows = lru_w->next;
900 free(lru_w);
901 pack_open_windows--;
902 return 1;
903 }
904 return 0;
905 }
906
907 void release_pack_memory(size_t need)
908 {
909 size_t cur = pack_mapped;
910 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
911 ; /* nothing */
912 }
913
914 static void mmap_limit_check(size_t length)
915 {
916 static size_t limit = 0;
917 if (!limit) {
918 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
919 if (!limit)
920 limit = SIZE_MAX;
921 }
922 if (length > limit)
923 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
924 (uintmax_t)length, (uintmax_t)limit);
925 }
926
927 void *xmmap_gently(void *start, size_t length,
928 int prot, int flags, int fd, off_t offset)
929 {
930 void *ret;
931
932 mmap_limit_check(length);
933 ret = mmap(start, length, prot, flags, fd, offset);
934 if (ret == MAP_FAILED) {
935 if (!length)
936 return NULL;
937 release_pack_memory(length);
938 ret = mmap(start, length, prot, flags, fd, offset);
939 }
940 return ret;
941 }
942
943 void *xmmap(void *start, size_t length,
944 int prot, int flags, int fd, off_t offset)
945 {
946 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
947 if (ret == MAP_FAILED)
948 die_errno("mmap failed");
949 return ret;
950 }
951
952 void close_pack_windows(struct packed_git *p)
953 {
954 while (p->windows) {
955 struct pack_window *w = p->windows;
956
957 if (w->inuse_cnt)
958 die("pack '%s' still has open windows to it",
959 p->pack_name);
960 munmap(w->base, w->len);
961 pack_mapped -= w->len;
962 pack_open_windows--;
963 p->windows = w->next;
964 free(w);
965 }
966 }
967
968 static int close_pack_fd(struct packed_git *p)
969 {
970 if (p->pack_fd < 0)
971 return 0;
972
973 close(p->pack_fd);
974 pack_open_fds--;
975 p->pack_fd = -1;
976
977 return 1;
978 }
979
980 static void close_pack(struct packed_git *p)
981 {
982 close_pack_windows(p);
983 close_pack_fd(p);
984 close_pack_index(p);
985 }
986
987 void close_all_packs(void)
988 {
989 struct packed_git *p;
990
991 for (p = packed_git; p; p = p->next)
992 if (p->do_not_close)
993 die("BUG: want to close pack marked 'do-not-close'");
994 else
995 close_pack(p);
996 }
997
998
999 /*
1000 * The LRU pack is the one with the oldest MRU window, preferring packs
1001 * with no used windows, or the oldest mtime if it has no windows allocated.
1002 */
1003 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
1004 {
1005 struct pack_window *w, *this_mru_w;
1006 int has_windows_inuse = 0;
1007
1008 /*
1009 * Reject this pack if it has windows and the previously selected
1010 * one does not. If this pack does not have windows, reject
1011 * it if the pack file is newer than the previously selected one.
1012 */
1013 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
1014 return;
1015
1016 for (w = this_mru_w = p->windows; w; w = w->next) {
1017 /*
1018 * Reject this pack if any of its windows are in use,
1019 * but the previously selected pack did not have any
1020 * inuse windows. Otherwise, record that this pack
1021 * has windows in use.
1022 */
1023 if (w->inuse_cnt) {
1024 if (*accept_windows_inuse)
1025 has_windows_inuse = 1;
1026 else
1027 return;
1028 }
1029
1030 if (w->last_used > this_mru_w->last_used)
1031 this_mru_w = w;
1032
1033 /*
1034 * Reject this pack if it has windows that have been
1035 * used more recently than the previously selected pack.
1036 * If the previously selected pack had windows inuse and
1037 * we have not encountered a window in this pack that is
1038 * inuse, skip this check since we prefer a pack with no
1039 * inuse windows to one that has inuse windows.
1040 */
1041 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
1042 this_mru_w->last_used > (*mru_w)->last_used)
1043 return;
1044 }
1045
1046 /*
1047 * Select this pack.
1048 */
1049 *mru_w = this_mru_w;
1050 *lru_p = p;
1051 *accept_windows_inuse = has_windows_inuse;
1052 }
1053
1054 static int close_one_pack(void)
1055 {
1056 struct packed_git *p, *lru_p = NULL;
1057 struct pack_window *mru_w = NULL;
1058 int accept_windows_inuse = 1;
1059
1060 for (p = packed_git; p; p = p->next) {
1061 if (p->pack_fd == -1)
1062 continue;
1063 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
1064 }
1065
1066 if (lru_p)
1067 return close_pack_fd(lru_p);
1068
1069 return 0;
1070 }
1071
1072 void unuse_pack(struct pack_window **w_cursor)
1073 {
1074 struct pack_window *w = *w_cursor;
1075 if (w) {
1076 w->inuse_cnt--;
1077 *w_cursor = NULL;
1078 }
1079 }
1080
1081 void close_pack_index(struct packed_git *p)
1082 {
1083 if (p->index_data) {
1084 munmap((void *)p->index_data, p->index_size);
1085 p->index_data = NULL;
1086 }
1087 }
1088
1089 static unsigned int get_max_fd_limit(void)
1090 {
1091 #ifdef RLIMIT_NOFILE
1092 {
1093 struct rlimit lim;
1094
1095 if (!getrlimit(RLIMIT_NOFILE, &lim))
1096 return lim.rlim_cur;
1097 }
1098 #endif
1099
1100 #ifdef _SC_OPEN_MAX
1101 {
1102 long open_max = sysconf(_SC_OPEN_MAX);
1103 if (0 < open_max)
1104 return open_max;
1105 /*
1106 * Otherwise, we got -1 for one of the two
1107 * reasons:
1108 *
1109 * (1) sysconf() did not understand _SC_OPEN_MAX
1110 * and signaled an error with -1; or
1111 * (2) sysconf() said there is no limit.
1112 *
1113 * We _could_ clear errno before calling sysconf() to
1114 * tell these two cases apart and return a huge number
1115 * in the latter case to let the caller cap it to a
1116 * value that is not so selfish, but letting the
1117 * fallback OPEN_MAX codepath take care of these cases
1118 * is a lot simpler.
1119 */
1120 }
1121 #endif
1122
1123 #ifdef OPEN_MAX
1124 return OPEN_MAX;
1125 #else
1126 return 1; /* see the caller ;-) */
1127 #endif
1128 }
1129
1130 /*
1131 * Do not call this directly as this leaks p->pack_fd on error return;
1132 * call open_packed_git() instead.
1133 */
1134 static int open_packed_git_1(struct packed_git *p)
1135 {
1136 struct stat st;
1137 struct pack_header hdr;
1138 unsigned char sha1[20];
1139 unsigned char *idx_sha1;
1140 long fd_flag;
1141
1142 if (!p->index_data && open_pack_index(p))
1143 return error("packfile %s index unavailable", p->pack_name);
1144
1145 if (!pack_max_fds) {
1146 unsigned int max_fds = get_max_fd_limit();
1147
1148 /* Save 3 for stdin/stdout/stderr, 22 for work */
1149 if (25 < max_fds)
1150 pack_max_fds = max_fds - 25;
1151 else
1152 pack_max_fds = 1;
1153 }
1154
1155 while (pack_max_fds <= pack_open_fds && close_one_pack())
1156 ; /* nothing */
1157
1158 p->pack_fd = git_open(p->pack_name);
1159 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
1160 return -1;
1161 pack_open_fds++;
1162
1163 /* If we created the struct before we had the pack we lack size. */
1164 if (!p->pack_size) {
1165 if (!S_ISREG(st.st_mode))
1166 return error("packfile %s not a regular file", p->pack_name);
1167 p->pack_size = st.st_size;
1168 } else if (p->pack_size != st.st_size)
1169 return error("packfile %s size changed", p->pack_name);
1170
1171 /* We leave these file descriptors open with sliding mmap;
1172 * there is no point keeping them open across exec(), though.
1173 */
1174 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
1175 if (fd_flag < 0)
1176 return error("cannot determine file descriptor flags");
1177 fd_flag |= FD_CLOEXEC;
1178 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
1179 return error("cannot set FD_CLOEXEC");
1180
1181 /* Verify we recognize this pack file format. */
1182 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
1183 return error("file %s is far too short to be a packfile", p->pack_name);
1184 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
1185 return error("file %s is not a GIT packfile", p->pack_name);
1186 if (!pack_version_ok(hdr.hdr_version))
1187 return error("packfile %s is version %"PRIu32" and not"
1188 " supported (try upgrading GIT to a newer version)",
1189 p->pack_name, ntohl(hdr.hdr_version));
1190
1191 /* Verify the pack matches its index. */
1192 if (p->num_objects != ntohl(hdr.hdr_entries))
1193 return error("packfile %s claims to have %"PRIu32" objects"
1194 " while index indicates %"PRIu32" objects",
1195 p->pack_name, ntohl(hdr.hdr_entries),
1196 p->num_objects);
1197 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
1198 return error("end of packfile %s is unavailable", p->pack_name);
1199 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
1200 return error("packfile %s signature is unavailable", p->pack_name);
1201 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
1202 if (hashcmp(sha1, idx_sha1))
1203 return error("packfile %s does not match index", p->pack_name);
1204 return 0;
1205 }
1206
1207 static int open_packed_git(struct packed_git *p)
1208 {
1209 if (!open_packed_git_1(p))
1210 return 0;
1211 close_pack_fd(p);
1212 return -1;
1213 }
1214
1215 static int in_window(struct pack_window *win, off_t offset)
1216 {
1217 /* We must promise at least 20 bytes (one hash) after the
1218 * offset is available from this window, otherwise the offset
1219 * is not actually in this window and a different window (which
1220 * has that one hash excess) must be used. This is to support
1221 * the object header and delta base parsing routines below.
1222 */
1223 off_t win_off = win->offset;
1224 return win_off <= offset
1225 && (offset + 20) <= (win_off + win->len);
1226 }
1227
1228 unsigned char *use_pack(struct packed_git *p,
1229 struct pack_window **w_cursor,
1230 off_t offset,
1231 unsigned long *left)
1232 {
1233 struct pack_window *win = *w_cursor;
1234
1235 /* Since packfiles end in a hash of their content and it's
1236 * pointless to ask for an offset into the middle of that
1237 * hash, and the in_window function above wouldn't match
1238 * don't allow an offset too close to the end of the file.
1239 */
1240 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1241 die("packfile %s cannot be accessed", p->pack_name);
1242 if (offset > (p->pack_size - 20))
1243 die("offset beyond end of packfile (truncated pack?)");
1244 if (offset < 0)
1245 die(_("offset before end of packfile (broken .idx?)"));
1246
1247 if (!win || !in_window(win, offset)) {
1248 if (win)
1249 win->inuse_cnt--;
1250 for (win = p->windows; win; win = win->next) {
1251 if (in_window(win, offset))
1252 break;
1253 }
1254 if (!win) {
1255 size_t window_align = packed_git_window_size / 2;
1256 off_t len;
1257
1258 if (p->pack_fd == -1 && open_packed_git(p))
1259 die("packfile %s cannot be accessed", p->pack_name);
1260
1261 win = xcalloc(1, sizeof(*win));
1262 win->offset = (offset / window_align) * window_align;
1263 len = p->pack_size - win->offset;
1264 if (len > packed_git_window_size)
1265 len = packed_git_window_size;
1266 win->len = (size_t)len;
1267 pack_mapped += win->len;
1268 while (packed_git_limit < pack_mapped
1269 && unuse_one_window(p))
1270 ; /* nothing */
1271 win->base = xmmap(NULL, win->len,
1272 PROT_READ, MAP_PRIVATE,
1273 p->pack_fd, win->offset);
1274 if (win->base == MAP_FAILED)
1275 die_errno("packfile %s cannot be mapped",
1276 p->pack_name);
1277 if (!win->offset && win->len == p->pack_size
1278 && !p->do_not_close)
1279 close_pack_fd(p);
1280 pack_mmap_calls++;
1281 pack_open_windows++;
1282 if (pack_mapped > peak_pack_mapped)
1283 peak_pack_mapped = pack_mapped;
1284 if (pack_open_windows > peak_pack_open_windows)
1285 peak_pack_open_windows = pack_open_windows;
1286 win->next = p->windows;
1287 p->windows = win;
1288 }
1289 }
1290 if (win != *w_cursor) {
1291 win->last_used = pack_used_ctr++;
1292 win->inuse_cnt++;
1293 *w_cursor = win;
1294 }
1295 offset -= win->offset;
1296 if (left)
1297 *left = win->len - xsize_t(offset);
1298 return win->base + offset;
1299 }
1300
1301 static struct packed_git *alloc_packed_git(int extra)
1302 {
1303 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
1304 memset(p, 0, sizeof(*p));
1305 p->pack_fd = -1;
1306 return p;
1307 }
1308
1309 static void try_to_free_pack_memory(size_t size)
1310 {
1311 release_pack_memory(size);
1312 }
1313
1314 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
1315 {
1316 static int have_set_try_to_free_routine;
1317 struct stat st;
1318 size_t alloc;
1319 struct packed_git *p;
1320
1321 if (!have_set_try_to_free_routine) {
1322 have_set_try_to_free_routine = 1;
1323 set_try_to_free_routine(try_to_free_pack_memory);
1324 }
1325
1326 /*
1327 * Make sure a corresponding .pack file exists and that
1328 * the index looks sane.
1329 */
1330 if (!strip_suffix_mem(path, &path_len, ".idx"))
1331 return NULL;
1332
1333 /*
1334 * ".pack" is long enough to hold any suffix we're adding (and
1335 * the use xsnprintf double-checks that)
1336 */
1337 alloc = st_add3(path_len, strlen(".pack"), 1);
1338 p = alloc_packed_git(alloc);
1339 memcpy(p->pack_name, path, path_len);
1340
1341 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
1342 if (!access(p->pack_name, F_OK))
1343 p->pack_keep = 1;
1344
1345 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
1346 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1347 free(p);
1348 return NULL;
1349 }
1350
1351 /* ok, it looks sane as far as we can check without
1352 * actually mapping the pack file.
1353 */
1354 p->pack_size = st.st_size;
1355 p->pack_local = local;
1356 p->mtime = st.st_mtime;
1357 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1358 hashclr(p->sha1);
1359 return p;
1360 }
1361
1362 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1363 {
1364 const char *path = sha1_pack_name(sha1);
1365 size_t alloc = st_add(strlen(path), 1);
1366 struct packed_git *p = alloc_packed_git(alloc);
1367
1368 memcpy(p->pack_name, path, alloc); /* includes NUL */
1369 hashcpy(p->sha1, sha1);
1370 if (check_packed_git_idx(idx_path, p)) {
1371 free(p);
1372 return NULL;
1373 }
1374
1375 return p;
1376 }
1377
1378 void install_packed_git(struct packed_git *pack)
1379 {
1380 if (pack->pack_fd != -1)
1381 pack_open_fds++;
1382
1383 pack->next = packed_git;
1384 packed_git = pack;
1385 }
1386
1387 void (*report_garbage)(unsigned seen_bits, const char *path);
1388
1389 static void report_helper(const struct string_list *list,
1390 int seen_bits, int first, int last)
1391 {
1392 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
1393 return;
1394
1395 for (; first < last; first++)
1396 report_garbage(seen_bits, list->items[first].string);
1397 }
1398
1399 static void report_pack_garbage(struct string_list *list)
1400 {
1401 int i, baselen = -1, first = 0, seen_bits = 0;
1402
1403 if (!report_garbage)
1404 return;
1405
1406 string_list_sort(list);
1407
1408 for (i = 0; i < list->nr; i++) {
1409 const char *path = list->items[i].string;
1410 if (baselen != -1 &&
1411 strncmp(path, list->items[first].string, baselen)) {
1412 report_helper(list, seen_bits, first, i);
1413 baselen = -1;
1414 seen_bits = 0;
1415 }
1416 if (baselen == -1) {
1417 const char *dot = strrchr(path, '.');
1418 if (!dot) {
1419 report_garbage(PACKDIR_FILE_GARBAGE, path);
1420 continue;
1421 }
1422 baselen = dot - path + 1;
1423 first = i;
1424 }
1425 if (!strcmp(path + baselen, "pack"))
1426 seen_bits |= 1;
1427 else if (!strcmp(path + baselen, "idx"))
1428 seen_bits |= 2;
1429 }
1430 report_helper(list, seen_bits, first, list->nr);
1431 }
1432
1433 static void prepare_packed_git_one(char *objdir, int local)
1434 {
1435 struct strbuf path = STRBUF_INIT;
1436 size_t dirnamelen;
1437 DIR *dir;
1438 struct dirent *de;
1439 struct string_list garbage = STRING_LIST_INIT_DUP;
1440
1441 strbuf_addstr(&path, objdir);
1442 strbuf_addstr(&path, "/pack");
1443 dir = opendir(path.buf);
1444 if (!dir) {
1445 if (errno != ENOENT)
1446 error_errno("unable to open object pack directory: %s",
1447 path.buf);
1448 strbuf_release(&path);
1449 return;
1450 }
1451 strbuf_addch(&path, '/');
1452 dirnamelen = path.len;
1453 while ((de = readdir(dir)) != NULL) {
1454 struct packed_git *p;
1455 size_t base_len;
1456
1457 if (is_dot_or_dotdot(de->d_name))
1458 continue;
1459
1460 strbuf_setlen(&path, dirnamelen);
1461 strbuf_addstr(&path, de->d_name);
1462
1463 base_len = path.len;
1464 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1465 /* Don't reopen a pack we already have. */
1466 for (p = packed_git; p; p = p->next) {
1467 size_t len;
1468 if (strip_suffix(p->pack_name, ".pack", &len) &&
1469 len == base_len &&
1470 !memcmp(p->pack_name, path.buf, len))
1471 break;
1472 }
1473 if (p == NULL &&
1474 /*
1475 * See if it really is a valid .idx file with
1476 * corresponding .pack file that we can map.
1477 */
1478 (p = add_packed_git(path.buf, path.len, local)) != NULL)
1479 install_packed_git(p);
1480 }
1481
1482 if (!report_garbage)
1483 continue;
1484
1485 if (ends_with(de->d_name, ".idx") ||
1486 ends_with(de->d_name, ".pack") ||
1487 ends_with(de->d_name, ".bitmap") ||
1488 ends_with(de->d_name, ".keep"))
1489 string_list_append(&garbage, path.buf);
1490 else
1491 report_garbage(PACKDIR_FILE_GARBAGE, path.buf);
1492 }
1493 closedir(dir);
1494 report_pack_garbage(&garbage);
1495 string_list_clear(&garbage, 0);
1496 strbuf_release(&path);
1497 }
1498
1499 static int approximate_object_count_valid;
1500
1501 /*
1502 * Give a fast, rough count of the number of objects in the repository. This
1503 * ignores loose objects completely. If you have a lot of them, then either
1504 * you should repack because your performance will be awful, or they are
1505 * all unreachable objects about to be pruned, in which case they're not really
1506 * interesting as a measure of repo size in the first place.
1507 */
1508 unsigned long approximate_object_count(void)
1509 {
1510 static unsigned long count;
1511 if (!approximate_object_count_valid) {
1512 struct packed_git *p;
1513
1514 prepare_packed_git();
1515 count = 0;
1516 for (p = packed_git; p; p = p->next) {
1517 if (open_pack_index(p))
1518 continue;
1519 count += p->num_objects;
1520 }
1521 }
1522 return count;
1523 }
1524
1525 static void *get_next_packed_git(const void *p)
1526 {
1527 return ((const struct packed_git *)p)->next;
1528 }
1529
1530 static void set_next_packed_git(void *p, void *next)
1531 {
1532 ((struct packed_git *)p)->next = next;
1533 }
1534
1535 static int sort_pack(const void *a_, const void *b_)
1536 {
1537 const struct packed_git *a = a_;
1538 const struct packed_git *b = b_;
1539 int st;
1540
1541 /*
1542 * Local packs tend to contain objects specific to our
1543 * variant of the project than remote ones. In addition,
1544 * remote ones could be on a network mounted filesystem.
1545 * Favor local ones for these reasons.
1546 */
1547 st = a->pack_local - b->pack_local;
1548 if (st)
1549 return -st;
1550
1551 /*
1552 * Younger packs tend to contain more recent objects,
1553 * and more recent objects tend to get accessed more
1554 * often.
1555 */
1556 if (a->mtime < b->mtime)
1557 return 1;
1558 else if (a->mtime == b->mtime)
1559 return 0;
1560 return -1;
1561 }
1562
1563 static void rearrange_packed_git(void)
1564 {
1565 packed_git = llist_mergesort(packed_git, get_next_packed_git,
1566 set_next_packed_git, sort_pack);
1567 }
1568
1569 static void prepare_packed_git_mru(void)
1570 {
1571 struct packed_git *p;
1572
1573 mru_clear(packed_git_mru);
1574 for (p = packed_git; p; p = p->next)
1575 mru_append(packed_git_mru, p);
1576 }
1577
1578 static int prepare_packed_git_run_once = 0;
1579 void prepare_packed_git(void)
1580 {
1581 struct alternate_object_database *alt;
1582
1583 if (prepare_packed_git_run_once)
1584 return;
1585 prepare_packed_git_one(get_object_directory(), 1);
1586 prepare_alt_odb();
1587 for (alt = alt_odb_list; alt; alt = alt->next)
1588 prepare_packed_git_one(alt->path, 0);
1589 rearrange_packed_git();
1590 prepare_packed_git_mru();
1591 prepare_packed_git_run_once = 1;
1592 }
1593
1594 void reprepare_packed_git(void)
1595 {
1596 approximate_object_count_valid = 0;
1597 prepare_packed_git_run_once = 0;
1598 prepare_packed_git();
1599 }
1600
1601 static void mark_bad_packed_object(struct packed_git *p,
1602 const unsigned char *sha1)
1603 {
1604 unsigned i;
1605 for (i = 0; i < p->num_bad_objects; i++)
1606 if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
1607 return;
1608 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1609 st_mult(GIT_MAX_RAWSZ,
1610 st_add(p->num_bad_objects, 1)));
1611 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
1612 p->num_bad_objects++;
1613 }
1614
1615 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1616 {
1617 struct packed_git *p;
1618 unsigned i;
1619
1620 for (p = packed_git; p; p = p->next)
1621 for (i = 0; i < p->num_bad_objects; i++)
1622 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1623 return p;
1624 return NULL;
1625 }
1626
1627 /*
1628 * With an in-core object data in "map", rehash it to make sure the
1629 * object name actually matches "sha1" to detect object corruption.
1630 * With "map" == NULL, try reading the object named with "sha1" using
1631 * the streaming interface and rehash it to do the same.
1632 */
1633 int check_sha1_signature(const unsigned char *sha1, void *map,
1634 unsigned long size, const char *type)
1635 {
1636 unsigned char real_sha1[20];
1637 enum object_type obj_type;
1638 struct git_istream *st;
1639 git_SHA_CTX c;
1640 char hdr[32];
1641 int hdrlen;
1642
1643 if (map) {
1644 hash_sha1_file(map, size, type, real_sha1);
1645 return hashcmp(sha1, real_sha1) ? -1 : 0;
1646 }
1647
1648 st = open_istream(sha1, &obj_type, &size, NULL);
1649 if (!st)
1650 return -1;
1651
1652 /* Generate the header */
1653 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
1654
1655 /* Sha1.. */
1656 git_SHA1_Init(&c);
1657 git_SHA1_Update(&c, hdr, hdrlen);
1658 for (;;) {
1659 char buf[1024 * 16];
1660 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1661
1662 if (readlen < 0) {
1663 close_istream(st);
1664 return -1;
1665 }
1666 if (!readlen)
1667 break;
1668 git_SHA1_Update(&c, buf, readlen);
1669 }
1670 git_SHA1_Final(real_sha1, &c);
1671 close_istream(st);
1672 return hashcmp(sha1, real_sha1) ? -1 : 0;
1673 }
1674
1675 int git_open_cloexec(const char *name, int flags)
1676 {
1677 int fd;
1678 static int o_cloexec = O_CLOEXEC;
1679
1680 fd = open(name, flags | o_cloexec);
1681 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1682 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1683 o_cloexec &= ~O_CLOEXEC;
1684 fd = open(name, flags | o_cloexec);
1685 }
1686
1687 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1688 {
1689 static int fd_cloexec = FD_CLOEXEC;
1690
1691 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1692 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1693 int flags = fcntl(fd, F_GETFD);
1694 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1695 fd_cloexec = 0;
1696 }
1697 }
1698 #endif
1699 return fd;
1700 }
1701
1702 /*
1703 * Find "sha1" as a loose object in the local repository or in an alternate.
1704 * Returns 0 on success, negative on failure.
1705 *
1706 * The "path" out-parameter will give the path of the object we found (if any).
1707 * Note that it may point to static storage and is only valid until another
1708 * call to sha1_file_name(), etc.
1709 */
1710 static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
1711 const char **path)
1712 {
1713 struct alternate_object_database *alt;
1714
1715 *path = sha1_file_name(sha1);
1716 if (!lstat(*path, st))
1717 return 0;
1718
1719 prepare_alt_odb();
1720 errno = ENOENT;
1721 for (alt = alt_odb_list; alt; alt = alt->next) {
1722 *path = alt_sha1_path(alt, sha1);
1723 if (!lstat(*path, st))
1724 return 0;
1725 }
1726
1727 return -1;
1728 }
1729
1730 /*
1731 * Like stat_sha1_file(), but actually open the object and return the
1732 * descriptor. See the caveats on the "path" parameter above.
1733 */
1734 static int open_sha1_file(const unsigned char *sha1, const char **path)
1735 {
1736 int fd;
1737 struct alternate_object_database *alt;
1738 int most_interesting_errno;
1739
1740 *path = sha1_file_name(sha1);
1741 fd = git_open(*path);
1742 if (fd >= 0)
1743 return fd;
1744 most_interesting_errno = errno;
1745
1746 prepare_alt_odb();
1747 for (alt = alt_odb_list; alt; alt = alt->next) {
1748 *path = alt_sha1_path(alt, sha1);
1749 fd = git_open(*path);
1750 if (fd >= 0)
1751 return fd;
1752 if (most_interesting_errno == ENOENT)
1753 most_interesting_errno = errno;
1754 }
1755 errno = most_interesting_errno;
1756 return -1;
1757 }
1758
1759 /*
1760 * Map the loose object at "path" if it is not NULL, or the path found by
1761 * searching for a loose object named "sha1".
1762 */
1763 static void *map_sha1_file_1(const char *path,
1764 const unsigned char *sha1,
1765 unsigned long *size)
1766 {
1767 void *map;
1768 int fd;
1769
1770 if (path)
1771 fd = git_open(path);
1772 else
1773 fd = open_sha1_file(sha1, &path);
1774 map = NULL;
1775 if (fd >= 0) {
1776 struct stat st;
1777
1778 if (!fstat(fd, &st)) {
1779 *size = xsize_t(st.st_size);
1780 if (!*size) {
1781 /* mmap() is forbidden on empty files */
1782 error("object file %s is empty", path);
1783 return NULL;
1784 }
1785 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1786 }
1787 close(fd);
1788 }
1789 return map;
1790 }
1791
1792 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1793 {
1794 return map_sha1_file_1(NULL, sha1, size);
1795 }
1796
1797 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1798 unsigned long len, enum object_type *type, unsigned long *sizep)
1799 {
1800 unsigned shift;
1801 unsigned long size, c;
1802 unsigned long used = 0;
1803
1804 c = buf[used++];
1805 *type = (c >> 4) & 7;
1806 size = c & 15;
1807 shift = 4;
1808 while (c & 0x80) {
1809 if (len <= used || bitsizeof(long) <= shift) {
1810 error("bad object header");
1811 size = used = 0;
1812 break;
1813 }
1814 c = buf[used++];
1815 size += (c & 0x7f) << shift;
1816 shift += 7;
1817 }
1818 *sizep = size;
1819 return used;
1820 }
1821
1822 static int unpack_sha1_short_header(git_zstream *stream,
1823 unsigned char *map, unsigned long mapsize,
1824 void *buffer, unsigned long bufsiz)
1825 {
1826 /* Get the data stream */
1827 memset(stream, 0, sizeof(*stream));
1828 stream->next_in = map;
1829 stream->avail_in = mapsize;
1830 stream->next_out = buffer;
1831 stream->avail_out = bufsiz;
1832
1833 git_inflate_init(stream);
1834 return git_inflate(stream, 0);
1835 }
1836
1837 int unpack_sha1_header(git_zstream *stream,
1838 unsigned char *map, unsigned long mapsize,
1839 void *buffer, unsigned long bufsiz)
1840 {
1841 int status = unpack_sha1_short_header(stream, map, mapsize,
1842 buffer, bufsiz);
1843
1844 if (status < Z_OK)
1845 return status;
1846
1847 /* Make sure we have the terminating NUL */
1848 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1849 return -1;
1850 return 0;
1851 }
1852
1853 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1854 unsigned long mapsize, void *buffer,
1855 unsigned long bufsiz, struct strbuf *header)
1856 {
1857 int status;
1858
1859 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1860 if (status < Z_OK)
1861 return -1;
1862
1863 /*
1864 * Check if entire header is unpacked in the first iteration.
1865 */
1866 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1867 return 0;
1868
1869 /*
1870 * buffer[0..bufsiz] was not large enough. Copy the partial
1871 * result out to header, and then append the result of further
1872 * reading the stream.
1873 */
1874 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1875 stream->next_out = buffer;
1876 stream->avail_out = bufsiz;
1877
1878 do {
1879 status = git_inflate(stream, 0);
1880 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1881 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1882 return 0;
1883 stream->next_out = buffer;
1884 stream->avail_out = bufsiz;
1885 } while (status != Z_STREAM_END);
1886 return -1;
1887 }
1888
1889 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1890 {
1891 int bytes = strlen(buffer) + 1;
1892 unsigned char *buf = xmallocz(size);
1893 unsigned long n;
1894 int status = Z_OK;
1895
1896 n = stream->total_out - bytes;
1897 if (n > size)
1898 n = size;
1899 memcpy(buf, (char *) buffer + bytes, n);
1900 bytes = n;
1901 if (bytes <= size) {
1902 /*
1903 * The above condition must be (bytes <= size), not
1904 * (bytes < size). In other words, even though we
1905 * expect no more output and set avail_out to zero,
1906 * the input zlib stream may have bytes that express
1907 * "this concludes the stream", and we *do* want to
1908 * eat that input.
1909 *
1910 * Otherwise we would not be able to test that we
1911 * consumed all the input to reach the expected size;
1912 * we also want to check that zlib tells us that all
1913 * went well with status == Z_STREAM_END at the end.
1914 */
1915 stream->next_out = buf + bytes;
1916 stream->avail_out = size - bytes;
1917 while (status == Z_OK)
1918 status = git_inflate(stream, Z_FINISH);
1919 }
1920 if (status == Z_STREAM_END && !stream->avail_in) {
1921 git_inflate_end(stream);
1922 return buf;
1923 }
1924
1925 if (status < 0)
1926 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1927 else if (stream->avail_in)
1928 error("garbage at end of loose object '%s'",
1929 sha1_to_hex(sha1));
1930 free(buf);
1931 return NULL;
1932 }
1933
1934 /*
1935 * We used to just use "sscanf()", but that's actually way
1936 * too permissive for what we want to check. So do an anal
1937 * object header parse by hand.
1938 */
1939 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1940 unsigned int flags)
1941 {
1942 const char *type_buf = hdr;
1943 unsigned long size;
1944 int type, type_len = 0;
1945
1946 /*
1947 * The type can be of any size but is followed by
1948 * a space.
1949 */
1950 for (;;) {
1951 char c = *hdr++;
1952 if (!c)
1953 return -1;
1954 if (c == ' ')
1955 break;
1956 type_len++;
1957 }
1958
1959 type = type_from_string_gently(type_buf, type_len, 1);
1960 if (oi->typename)
1961 strbuf_add(oi->typename, type_buf, type_len);
1962 /*
1963 * Set type to 0 if its an unknown object and
1964 * we're obtaining the type using '--allow-unknown-type'
1965 * option.
1966 */
1967 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1968 type = 0;
1969 else if (type < 0)
1970 die("invalid object type");
1971 if (oi->typep)
1972 *oi->typep = type;
1973
1974 /*
1975 * The length must follow immediately, and be in canonical
1976 * decimal format (ie "010" is not valid).
1977 */
1978 size = *hdr++ - '0';
1979 if (size > 9)
1980 return -1;
1981 if (size) {
1982 for (;;) {
1983 unsigned long c = *hdr - '0';
1984 if (c > 9)
1985 break;
1986 hdr++;
1987 size = size * 10 + c;
1988 }
1989 }
1990
1991 if (oi->sizep)
1992 *oi->sizep = size;
1993
1994 /*
1995 * The length must be followed by a zero byte
1996 */
1997 return *hdr ? -1 : type;
1998 }
1999
2000 int parse_sha1_header(const char *hdr, unsigned long *sizep)
2001 {
2002 struct object_info oi = OBJECT_INFO_INIT;
2003
2004 oi.sizep = sizep;
2005 return parse_sha1_header_extended(hdr, &oi, 0);
2006 }
2007
2008 unsigned long get_size_from_delta(struct packed_git *p,
2009 struct pack_window **w_curs,
2010 off_t curpos)
2011 {
2012 const unsigned char *data;
2013 unsigned char delta_head[20], *in;
2014 git_zstream stream;
2015 int st;
2016
2017 memset(&stream, 0, sizeof(stream));
2018 stream.next_out = delta_head;
2019 stream.avail_out = sizeof(delta_head);
2020
2021 git_inflate_init(&stream);
2022 do {
2023 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2024 stream.next_in = in;
2025 st = git_inflate(&stream, Z_FINISH);
2026 curpos += stream.next_in - in;
2027 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
2028 stream.total_out < sizeof(delta_head));
2029 git_inflate_end(&stream);
2030 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
2031 error("delta data unpack-initial failed");
2032 return 0;
2033 }
2034
2035 /* Examine the initial part of the delta to figure out
2036 * the result size.
2037 */
2038 data = delta_head;
2039
2040 /* ignore base size */
2041 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
2042
2043 /* Read the result size */
2044 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
2045 }
2046
2047 static off_t get_delta_base(struct packed_git *p,
2048 struct pack_window **w_curs,
2049 off_t *curpos,
2050 enum object_type type,
2051 off_t delta_obj_offset)
2052 {
2053 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
2054 off_t base_offset;
2055
2056 /* use_pack() assured us we have [base_info, base_info + 20)
2057 * as a range that we can look at without walking off the
2058 * end of the mapped window. Its actually the hash size
2059 * that is assured. An OFS_DELTA longer than the hash size
2060 * is stupid, as then a REF_DELTA would be smaller to store.
2061 */
2062 if (type == OBJ_OFS_DELTA) {
2063 unsigned used = 0;
2064 unsigned char c = base_info[used++];
2065 base_offset = c & 127;
2066 while (c & 128) {
2067 base_offset += 1;
2068 if (!base_offset || MSB(base_offset, 7))
2069 return 0; /* overflow */
2070 c = base_info[used++];
2071 base_offset = (base_offset << 7) + (c & 127);
2072 }
2073 base_offset = delta_obj_offset - base_offset;
2074 if (base_offset <= 0 || base_offset >= delta_obj_offset)
2075 return 0; /* out of bound */
2076 *curpos += used;
2077 } else if (type == OBJ_REF_DELTA) {
2078 /* The base entry _must_ be in the same pack */
2079 base_offset = find_pack_entry_one(base_info, p);
2080 *curpos += 20;
2081 } else
2082 die("I am totally screwed");
2083 return base_offset;
2084 }
2085
2086 /*
2087 * Like get_delta_base above, but we return the sha1 instead of the pack
2088 * offset. This means it is cheaper for REF deltas (we do not have to do
2089 * the final object lookup), but more expensive for OFS deltas (we
2090 * have to load the revidx to convert the offset back into a sha1).
2091 */
2092 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
2093 struct pack_window **w_curs,
2094 off_t curpos,
2095 enum object_type type,
2096 off_t delta_obj_offset)
2097 {
2098 if (type == OBJ_REF_DELTA) {
2099 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
2100 return base;
2101 } else if (type == OBJ_OFS_DELTA) {
2102 struct revindex_entry *revidx;
2103 off_t base_offset = get_delta_base(p, w_curs, &curpos,
2104 type, delta_obj_offset);
2105
2106 if (!base_offset)
2107 return NULL;
2108
2109 revidx = find_pack_revindex(p, base_offset);
2110 if (!revidx)
2111 return NULL;
2112
2113 return nth_packed_object_sha1(p, revidx->nr);
2114 } else
2115 return NULL;
2116 }
2117
2118 int unpack_object_header(struct packed_git *p,
2119 struct pack_window **w_curs,
2120 off_t *curpos,
2121 unsigned long *sizep)
2122 {
2123 unsigned char *base;
2124 unsigned long left;
2125 unsigned long used;
2126 enum object_type type;
2127
2128 /* use_pack() assures us we have [base, base + 20) available
2129 * as a range that we can look at. (Its actually the hash
2130 * size that is assured.) With our object header encoding
2131 * the maximum deflated object size is 2^137, which is just
2132 * insane, so we know won't exceed what we have been given.
2133 */
2134 base = use_pack(p, w_curs, *curpos, &left);
2135 used = unpack_object_header_buffer(base, left, &type, sizep);
2136 if (!used) {
2137 type = OBJ_BAD;
2138 } else
2139 *curpos += used;
2140
2141 return type;
2142 }
2143
2144 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
2145 {
2146 int type;
2147 struct revindex_entry *revidx;
2148 const unsigned char *sha1;
2149 revidx = find_pack_revindex(p, obj_offset);
2150 if (!revidx)
2151 return OBJ_BAD;
2152 sha1 = nth_packed_object_sha1(p, revidx->nr);
2153 mark_bad_packed_object(p, sha1);
2154 type = sha1_object_info(sha1, NULL);
2155 if (type <= OBJ_NONE)
2156 return OBJ_BAD;
2157 return type;
2158 }
2159
2160 #define POI_STACK_PREALLOC 64
2161
2162 static enum object_type packed_to_object_type(struct packed_git *p,
2163 off_t obj_offset,
2164 enum object_type type,
2165 struct pack_window **w_curs,
2166 off_t curpos)
2167 {
2168 off_t small_poi_stack[POI_STACK_PREALLOC];
2169 off_t *poi_stack = small_poi_stack;
2170 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
2171
2172 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2173 off_t base_offset;
2174 unsigned long size;
2175 /* Push the object we're going to leave behind */
2176 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
2177 poi_stack_alloc = alloc_nr(poi_stack_nr);
2178 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
2179 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
2180 } else {
2181 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
2182 }
2183 poi_stack[poi_stack_nr++] = obj_offset;
2184 /* If parsing the base offset fails, just unwind */
2185 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
2186 if (!base_offset)
2187 goto unwind;
2188 curpos = obj_offset = base_offset;
2189 type = unpack_object_header(p, w_curs, &curpos, &size);
2190 if (type <= OBJ_NONE) {
2191 /* If getting the base itself fails, we first
2192 * retry the base, otherwise unwind */
2193 type = retry_bad_packed_offset(p, base_offset);
2194 if (type > OBJ_NONE)
2195 goto out;
2196 goto unwind;
2197 }
2198 }
2199
2200 switch (type) {
2201 case OBJ_BAD:
2202 case OBJ_COMMIT:
2203 case OBJ_TREE:
2204 case OBJ_BLOB:
2205 case OBJ_TAG:
2206 break;
2207 default:
2208 error("unknown object type %i at offset %"PRIuMAX" in %s",
2209 type, (uintmax_t)obj_offset, p->pack_name);
2210 type = OBJ_BAD;
2211 }
2212
2213 out:
2214 if (poi_stack != small_poi_stack)
2215 free(poi_stack);
2216 return type;
2217
2218 unwind:
2219 while (poi_stack_nr) {
2220 obj_offset = poi_stack[--poi_stack_nr];
2221 type = retry_bad_packed_offset(p, obj_offset);
2222 if (type > OBJ_NONE)
2223 goto out;
2224 }
2225 type = OBJ_BAD;
2226 goto out;
2227 }
2228
2229 static struct hashmap delta_base_cache;
2230 static size_t delta_base_cached;
2231
2232 static LIST_HEAD(delta_base_cache_lru);
2233
2234 struct delta_base_cache_key {
2235 struct packed_git *p;
2236 off_t base_offset;
2237 };
2238
2239 struct delta_base_cache_entry {
2240 struct hashmap hash;
2241 struct delta_base_cache_key key;
2242 struct list_head lru;
2243 void *data;
2244 unsigned long size;
2245 enum object_type type;
2246 };
2247
2248 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
2249 {
2250 unsigned int hash;
2251
2252 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
2253 hash += (hash >> 8) + (hash >> 16);
2254 return hash;
2255 }
2256
2257 static struct delta_base_cache_entry *
2258 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2259 {
2260 struct hashmap_entry entry;
2261 struct delta_base_cache_key key;
2262
2263 if (!delta_base_cache.cmpfn)
2264 return NULL;
2265
2266 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
2267 key.p = p;
2268 key.base_offset = base_offset;
2269 return hashmap_get(&delta_base_cache, &entry, &key);
2270 }
2271
2272 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
2273 const struct delta_base_cache_key *b)
2274 {
2275 return a->p == b->p && a->base_offset == b->base_offset;
2276 }
2277
2278 static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
2279 const void *va, const void *vb,
2280 const void *vkey)
2281 {
2282 const struct delta_base_cache_entry *a = va, *b = vb;
2283 const struct delta_base_cache_key *key = vkey;
2284 if (key)
2285 return !delta_base_cache_key_eq(&a->key, key);
2286 else
2287 return !delta_base_cache_key_eq(&a->key, &b->key);
2288 }
2289
2290 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2291 {
2292 return !!get_delta_base_cache_entry(p, base_offset);
2293 }
2294
2295 /*
2296 * Remove the entry from the cache, but do _not_ free the associated
2297 * entry data. The caller takes ownership of the "data" buffer, and
2298 * should copy out any fields it wants before detaching.
2299 */
2300 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2301 {
2302 hashmap_remove(&delta_base_cache, ent, &ent->key);
2303 list_del(&ent->lru);
2304 delta_base_cached -= ent->size;
2305 free(ent);
2306 }
2307
2308 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2309 unsigned long *base_size, enum object_type *type)
2310 {
2311 struct delta_base_cache_entry *ent;
2312
2313 ent = get_delta_base_cache_entry(p, base_offset);
2314 if (!ent)
2315 return unpack_entry(p, base_offset, type, base_size);
2316
2317 if (type)
2318 *type = ent->type;
2319 if (base_size)
2320 *base_size = ent->size;
2321 return xmemdupz(ent->data, ent->size);
2322 }
2323
2324 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2325 {
2326 free(ent->data);
2327 detach_delta_base_cache_entry(ent);
2328 }
2329
2330 void clear_delta_base_cache(void)
2331 {
2332 struct list_head *lru, *tmp;
2333 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
2334 struct delta_base_cache_entry *entry =
2335 list_entry(lru, struct delta_base_cache_entry, lru);
2336 release_delta_base_cache(entry);
2337 }
2338 }
2339
2340 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2341 void *base, unsigned long base_size, enum object_type type)
2342 {
2343 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
2344 struct list_head *lru, *tmp;
2345
2346 delta_base_cached += base_size;
2347
2348 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
2349 struct delta_base_cache_entry *f =
2350 list_entry(lru, struct delta_base_cache_entry, lru);
2351 if (delta_base_cached <= delta_base_cache_limit)
2352 break;
2353 release_delta_base_cache(f);
2354 }
2355
2356 ent->key.p = p;
2357 ent->key.base_offset = base_offset;
2358 ent->type = type;
2359 ent->data = base;
2360 ent->size = base_size;
2361 list_add_tail(&ent->lru, &delta_base_cache_lru);
2362
2363 if (!delta_base_cache.cmpfn)
2364 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
2365 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
2366 hashmap_add(&delta_base_cache, ent);
2367 }
2368
2369 int packed_object_info(struct packed_git *p, off_t obj_offset,
2370 struct object_info *oi)
2371 {
2372 struct pack_window *w_curs = NULL;
2373 unsigned long size;
2374 off_t curpos = obj_offset;
2375 enum object_type type;
2376
2377 /*
2378 * We always get the representation type, but only convert it to
2379 * a "real" type later if the caller is interested.
2380 */
2381 if (oi->contentp) {
2382 *oi->contentp = cache_or_unpack_entry(p, obj_offset, oi->sizep,
2383 &type);
2384 if (!*oi->contentp)
2385 type = OBJ_BAD;
2386 } else {
2387 type = unpack_object_header(p, &w_curs, &curpos, &size);
2388 }
2389
2390 if (!oi->contentp && oi->sizep) {
2391 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2392 off_t tmp_pos = curpos;
2393 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
2394 type, obj_offset);
2395 if (!base_offset) {
2396 type = OBJ_BAD;
2397 goto out;
2398 }
2399 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2400 if (*oi->sizep == 0) {
2401 type = OBJ_BAD;
2402 goto out;
2403 }
2404 } else {
2405 *oi->sizep = size;
2406 }
2407 }
2408
2409 if (oi->disk_sizep) {
2410 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2411 *oi->disk_sizep = revidx[1].offset - obj_offset;
2412 }
2413
2414 if (oi->typep || oi->typename) {
2415 enum object_type ptot;
2416 ptot = packed_to_object_type(p, obj_offset, type, &w_curs,
2417 curpos);
2418 if (oi->typep)
2419 *oi->typep = ptot;
2420 if (oi->typename) {
2421 const char *tn = typename(ptot);
2422 if (tn)
2423 strbuf_addstr(oi->typename, tn);
2424 }
2425 if (ptot < 0) {
2426 type = OBJ_BAD;
2427 goto out;
2428 }
2429 }
2430
2431 if (oi->delta_base_sha1) {
2432 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2433 const unsigned char *base;
2434
2435 base = get_delta_base_sha1(p, &w_curs, curpos,
2436 type, obj_offset);
2437 if (!base) {
2438 type = OBJ_BAD;
2439 goto out;
2440 }
2441
2442 hashcpy(oi->delta_base_sha1, base);
2443 } else
2444 hashclr(oi->delta_base_sha1);
2445 }
2446
2447 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
2448 OI_PACKED;
2449
2450 out:
2451 unuse_pack(&w_curs);
2452 return type;
2453 }
2454
2455 static void *unpack_compressed_entry(struct packed_git *p,
2456 struct pack_window **w_curs,
2457 off_t curpos,
2458 unsigned long size)
2459 {
2460 int st;
2461 git_zstream stream;
2462 unsigned char *buffer, *in;
2463
2464 buffer = xmallocz_gently(size);
2465 if (!buffer)
2466 return NULL;
2467 memset(&stream, 0, sizeof(stream));
2468 stream.next_out = buffer;
2469 stream.avail_out = size + 1;
2470
2471 git_inflate_init(&stream);
2472 do {
2473 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2474 stream.next_in = in;
2475 st = git_inflate(&stream, Z_FINISH);
2476 if (!stream.avail_out)
2477 break; /* the payload is larger than it should be */
2478 curpos += stream.next_in - in;
2479 } while (st == Z_OK || st == Z_BUF_ERROR);
2480 git_inflate_end(&stream);
2481 if ((st != Z_STREAM_END) || stream.total_out != size) {
2482 free(buffer);
2483 return NULL;
2484 }
2485
2486 return buffer;
2487 }
2488
2489 static void *read_object(const unsigned char *sha1, enum object_type *type,
2490 unsigned long *size);
2491
2492 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2493 {
2494 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2495 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2496 p->pack_name, (uintmax_t)obj_offset);
2497 }
2498
2499 int do_check_packed_object_crc;
2500
2501 #define UNPACK_ENTRY_STACK_PREALLOC 64
2502 struct unpack_entry_stack_ent {
2503 off_t obj_offset;
2504 off_t curpos;
2505 unsigned long size;
2506 };
2507
2508 void *unpack_entry(struct packed_git *p, off_t obj_offset,
2509 enum object_type *final_type, unsigned long *final_size)
2510 {
2511 struct pack_window *w_curs = NULL;
2512 off_t curpos = obj_offset;
2513 void *data = NULL;
2514 unsigned long size;
2515 enum object_type type;
2516 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2517 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2518 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2519 int base_from_cache = 0;
2520
2521 write_pack_access_log(p, obj_offset);
2522
2523 /* PHASE 1: drill down to the innermost base object */
2524 for (;;) {
2525 off_t base_offset;
2526 int i;
2527 struct delta_base_cache_entry *ent;
2528
2529 ent = get_delta_base_cache_entry(p, curpos);
2530 if (ent) {
2531 type = ent->type;
2532 data = ent->data;
2533 size = ent->size;
2534 detach_delta_base_cache_entry(ent);
2535 base_from_cache = 1;
2536 break;
2537 }
2538
2539 if (do_check_packed_object_crc && p->index_version > 1) {
2540 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2541 off_t len = revidx[1].offset - obj_offset;
2542 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2543 const unsigned char *sha1 =
2544 nth_packed_object_sha1(p, revidx->nr);
2545 error("bad packed object CRC for %s",
2546 sha1_to_hex(sha1));
2547 mark_bad_packed_object(p, sha1);
2548 data = NULL;
2549 goto out;
2550 }
2551 }
2552
2553 type = unpack_object_header(p, &w_curs, &curpos, &size);
2554 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2555 break;
2556
2557 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2558 if (!base_offset) {
2559 error("failed to validate delta base reference "
2560 "at offset %"PRIuMAX" from %s",
2561 (uintmax_t)curpos, p->pack_name);
2562 /* bail to phase 2, in hopes of recovery */
2563 data = NULL;
2564 break;
2565 }
2566
2567 /* push object, proceed to base */
2568 if (delta_stack_nr >= delta_stack_alloc
2569 && delta_stack == small_delta_stack) {
2570 delta_stack_alloc = alloc_nr(delta_stack_nr);
2571 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
2572 memcpy(delta_stack, small_delta_stack,
2573 sizeof(*delta_stack)*delta_stack_nr);
2574 } else {
2575 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2576 }
2577 i = delta_stack_nr++;
2578 delta_stack[i].obj_offset = obj_offset;
2579 delta_stack[i].curpos = curpos;
2580 delta_stack[i].size = size;
2581
2582 curpos = obj_offset = base_offset;
2583 }
2584
2585 /* PHASE 2: handle the base */
2586 switch (type) {
2587 case OBJ_OFS_DELTA:
2588 case OBJ_REF_DELTA:
2589 if (data)
2590 die("BUG: unpack_entry: left loop at a valid delta");
2591 break;
2592 case OBJ_COMMIT:
2593 case OBJ_TREE:
2594 case OBJ_BLOB:
2595 case OBJ_TAG:
2596 if (!base_from_cache)
2597 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2598 break;
2599 default:
2600 data = NULL;
2601 error("unknown object type %i at offset %"PRIuMAX" in %s",
2602 type, (uintmax_t)obj_offset, p->pack_name);
2603 }
2604
2605 /* PHASE 3: apply deltas in order */
2606
2607 /* invariants:
2608 * 'data' holds the base data, or NULL if there was corruption
2609 */
2610 while (delta_stack_nr) {
2611 void *delta_data;
2612 void *base = data;
2613 void *external_base = NULL;
2614 unsigned long delta_size, base_size = size;
2615 int i;
2616
2617 data = NULL;
2618
2619 if (base)
2620 add_delta_base_cache(p, obj_offset, base, base_size, type);
2621
2622 if (!base) {
2623 /*
2624 * We're probably in deep shit, but let's try to fetch
2625 * the required base anyway from another pack or loose.
2626 * This is costly but should happen only in the presence
2627 * of a corrupted pack, and is better than failing outright.
2628 */
2629 struct revindex_entry *revidx;
2630 const unsigned char *base_sha1;
2631 revidx = find_pack_revindex(p, obj_offset);
2632 if (revidx) {
2633 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2634 error("failed to read delta base object %s"
2635 " at offset %"PRIuMAX" from %s",
2636 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2637 p->pack_name);
2638 mark_bad_packed_object(p, base_sha1);
2639 base = read_object(base_sha1, &type, &base_size);
2640 external_base = base;
2641 }
2642 }
2643
2644 i = --delta_stack_nr;
2645 obj_offset = delta_stack[i].obj_offset;
2646 curpos = delta_stack[i].curpos;
2647 delta_size = delta_stack[i].size;
2648
2649 if (!base)
2650 continue;
2651
2652 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2653
2654 if (!delta_data) {
2655 error("failed to unpack compressed delta "
2656 "at offset %"PRIuMAX" from %s",
2657 (uintmax_t)curpos, p->pack_name);
2658 data = NULL;
2659 free(external_base);
2660 continue;
2661 }
2662
2663 data = patch_delta(base, base_size,
2664 delta_data, delta_size,
2665 &size);
2666
2667 /*
2668 * We could not apply the delta; warn the user, but keep going.
2669 * Our failure will be noticed either in the next iteration of
2670 * the loop, or if this is the final delta, in the caller when
2671 * we return NULL. Those code paths will take care of making
2672 * a more explicit warning and retrying with another copy of
2673 * the object.
2674 */
2675 if (!data)
2676 error("failed to apply delta");
2677
2678 free(delta_data);
2679 free(external_base);
2680 }
2681
2682 if (final_type)
2683 *final_type = type;
2684 if (final_size)
2685 *final_size = size;
2686
2687 out:
2688 unuse_pack(&w_curs);
2689
2690 if (delta_stack != small_delta_stack)
2691 free(delta_stack);
2692
2693 return data;
2694 }
2695
2696 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2697 uint32_t n)
2698 {
2699 const unsigned char *index = p->index_data;
2700 if (!index) {
2701 if (open_pack_index(p))
2702 return NULL;
2703 index = p->index_data;
2704 }
2705 if (n >= p->num_objects)
2706 return NULL;
2707 index += 4 * 256;
2708 if (p->index_version == 1) {
2709 return index + 24 * n + 4;
2710 } else {
2711 index += 8;
2712 return index + 20 * n;
2713 }
2714 }
2715
2716 const struct object_id *nth_packed_object_oid(struct object_id *oid,
2717 struct packed_git *p,
2718 uint32_t n)
2719 {
2720 const unsigned char *hash = nth_packed_object_sha1(p, n);
2721 if (!hash)
2722 return NULL;
2723 hashcpy(oid->hash, hash);
2724 return oid;
2725 }
2726
2727 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
2728 {
2729 const unsigned char *ptr = vptr;
2730 const unsigned char *start = p->index_data;
2731 const unsigned char *end = start + p->index_size;
2732 if (ptr < start)
2733 die(_("offset before start of pack index for %s (corrupt index?)"),
2734 p->pack_name);
2735 /* No need to check for underflow; .idx files must be at least 8 bytes */
2736 if (ptr >= end - 8)
2737 die(_("offset beyond end of pack index for %s (truncated index?)"),
2738 p->pack_name);
2739 }
2740
2741 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2742 {
2743 const unsigned char *index = p->index_data;
2744 index += 4 * 256;
2745 if (p->index_version == 1) {
2746 return ntohl(*((uint32_t *)(index + 24 * n)));
2747 } else {
2748 uint32_t off;
2749 index += 8 + p->num_objects * (20 + 4);
2750 off = ntohl(*((uint32_t *)(index + 4 * n)));
2751 if (!(off & 0x80000000))
2752 return off;
2753 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2754 check_pack_index_ptr(p, index);
2755 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2756 ntohl(*((uint32_t *)(index + 4)));
2757 }
2758 }
2759
2760 off_t find_pack_entry_one(const unsigned char *sha1,
2761 struct packed_git *p)
2762 {
2763 const uint32_t *level1_ofs = p->index_data;
2764 const unsigned char *index = p->index_data;
2765 unsigned hi, lo, stride;
2766 static int debug_lookup = -1;
2767
2768 if (debug_lookup < 0)
2769 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2770
2771 if (!index) {
2772 if (open_pack_index(p))
2773 return 0;
2774 level1_ofs = p->index_data;
2775 index = p->index_data;
2776 }
2777 if (p->index_version > 1) {
2778 level1_ofs += 2;
2779 index += 8;
2780 }
2781 index += 4 * 256;
2782 hi = ntohl(level1_ofs[*sha1]);
2783 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2784 if (p->index_version > 1) {
2785 stride = 20;
2786 } else {
2787 stride = 24;
2788 index += 4;
2789 }
2790
2791 if (debug_lookup)
2792 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2793 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2794
2795 while (lo < hi) {
2796 unsigned mi = (lo + hi) / 2;
2797 int cmp = hashcmp(index + mi * stride, sha1);
2798
2799 if (debug_lookup)
2800 printf("lo %u hi %u rg %u mi %u\n",
2801 lo, hi, hi - lo, mi);
2802 if (!cmp)
2803 return nth_packed_object_offset(p, mi);
2804 if (cmp > 0)
2805 hi = mi;
2806 else
2807 lo = mi+1;
2808 }
2809 return 0;
2810 }
2811
2812 int is_pack_valid(struct packed_git *p)
2813 {
2814 /* An already open pack is known to be valid. */
2815 if (p->pack_fd != -1)
2816 return 1;
2817
2818 /* If the pack has one window completely covering the
2819 * file size, the pack is known to be valid even if
2820 * the descriptor is not currently open.
2821 */
2822 if (p->windows) {
2823 struct pack_window *w = p->windows;
2824
2825 if (!w->offset && w->len == p->pack_size)
2826 return 1;
2827 }
2828
2829 /* Force the pack to open to prove its valid. */
2830 return !open_packed_git(p);
2831 }
2832
2833 static int fill_pack_entry(const unsigned char *sha1,
2834 struct pack_entry *e,
2835 struct packed_git *p)
2836 {
2837 off_t offset;
2838
2839 if (p->num_bad_objects) {
2840 unsigned i;
2841 for (i = 0; i < p->num_bad_objects; i++)
2842 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2843 return 0;
2844 }
2845
2846 offset = find_pack_entry_one(sha1, p);
2847 if (!offset)
2848 return 0;
2849
2850 /*
2851 * We are about to tell the caller where they can locate the
2852 * requested object. We better make sure the packfile is
2853 * still here and can be accessed before supplying that
2854 * answer, as it may have been deleted since the index was
2855 * loaded!
2856 */
2857 if (!is_pack_valid(p))
2858 return 0;
2859 e->offset = offset;
2860 e->p = p;
2861 hashcpy(e->sha1, sha1);
2862 return 1;
2863 }
2864
2865 /*
2866 * Iff a pack file contains the object named by sha1, return true and
2867 * store its location to e.
2868 */
2869 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2870 {
2871 struct mru_entry *p;
2872
2873 prepare_packed_git();
2874 if (!packed_git)
2875 return 0;
2876
2877 for (p = packed_git_mru->head; p; p = p->next) {
2878 if (fill_pack_entry(sha1, e, p->item)) {
2879 mru_mark(packed_git_mru, p);
2880 return 1;
2881 }
2882 }
2883 return 0;
2884 }
2885
2886 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2887 struct packed_git *packs)
2888 {
2889 struct packed_git *p;
2890
2891 for (p = packs; p; p = p->next) {
2892 if (find_pack_entry_one(sha1, p))
2893 return p;
2894 }
2895 return NULL;
2896
2897 }
2898
2899 static int sha1_loose_object_info(const unsigned char *sha1,
2900 struct object_info *oi,
2901 int flags)
2902 {
2903 int status = 0;
2904 unsigned long mapsize;
2905 void *map;
2906 git_zstream stream;
2907 char hdr[32];
2908 struct strbuf hdrbuf = STRBUF_INIT;
2909 unsigned long size_scratch;
2910
2911 if (oi->delta_base_sha1)
2912 hashclr(oi->delta_base_sha1);
2913
2914 /*
2915 * If we don't care about type or size, then we don't
2916 * need to look inside the object at all. Note that we
2917 * do not optimize out the stat call, even if the
2918 * caller doesn't care about the disk-size, since our
2919 * return value implicitly indicates whether the
2920 * object even exists.
2921 */
2922 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
2923 const char *path;
2924 struct stat st;
2925 if (stat_sha1_file(sha1, &st, &path) < 0)
2926 return -1;
2927 if (oi->disk_sizep)
2928 *oi->disk_sizep = st.st_size;
2929 return 0;
2930 }
2931
2932 map = map_sha1_file(sha1, &mapsize);
2933 if (!map)
2934 return -1;
2935
2936 if (!oi->sizep)
2937 oi->sizep = &size_scratch;
2938
2939 if (oi->disk_sizep)
2940 *oi->disk_sizep = mapsize;
2941 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
2942 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2943 status = error("unable to unpack %s header with --allow-unknown-type",
2944 sha1_to_hex(sha1));
2945 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2946 status = error("unable to unpack %s header",
2947 sha1_to_hex(sha1));
2948 if (status < 0)
2949 ; /* Do nothing */
2950 else if (hdrbuf.len) {
2951 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2952 status = error("unable to parse %s header with --allow-unknown-type",
2953 sha1_to_hex(sha1));
2954 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
2955 status = error("unable to parse %s header", sha1_to_hex(sha1));
2956
2957 if (status >= 0 && oi->contentp)
2958 *oi->contentp = unpack_sha1_rest(&stream, hdr,
2959 *oi->sizep, sha1);
2960 else
2961 git_inflate_end(&stream);
2962
2963 munmap(map, mapsize);
2964 if (status && oi->typep)
2965 *oi->typep = status;
2966 if (oi->sizep == &size_scratch)
2967 oi->sizep = NULL;
2968 strbuf_release(&hdrbuf);
2969 oi->whence = OI_LOOSE;
2970 return (status < 0) ? status : 0;
2971 }
2972
2973 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2974 {
2975 static struct object_info blank_oi = OBJECT_INFO_INIT;
2976 struct pack_entry e;
2977 int rtype;
2978 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
2979 lookup_replace_object(sha1) :
2980 sha1;
2981
2982 if (!oi)
2983 oi = &blank_oi;
2984
2985 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
2986 struct cached_object *co = find_cached_object(real);
2987 if (co) {
2988 if (oi->typep)
2989 *(oi->typep) = co->type;
2990 if (oi->sizep)
2991 *(oi->sizep) = co->size;
2992 if (oi->disk_sizep)
2993 *(oi->disk_sizep) = 0;
2994 if (oi->delta_base_sha1)
2995 hashclr(oi->delta_base_sha1);
2996 if (oi->typename)
2997 strbuf_addstr(oi->typename, typename(co->type));
2998 if (oi->contentp)
2999 *oi->contentp = xmemdupz(co->buf, co->size);
3000 oi->whence = OI_CACHED;
3001 return 0;
3002 }
3003 }
3004
3005 if (!find_pack_entry(real, &e)) {
3006 /* Most likely it's a loose object. */
3007 if (!sha1_loose_object_info(real, oi, flags))
3008 return 0;
3009
3010 /* Not a loose object; someone else may have just packed it. */
3011 if (flags & OBJECT_INFO_QUICK) {
3012 return -1;
3013 } else {
3014 reprepare_packed_git();
3015 if (!find_pack_entry(real, &e))
3016 return -1;
3017 }
3018 }
3019
3020 if (oi == &blank_oi)
3021 /*
3022 * We know that the caller doesn't actually need the
3023 * information below, so return early.
3024 */
3025 return 0;
3026
3027 rtype = packed_object_info(e.p, e.offset, oi);
3028 if (rtype < 0) {
3029 mark_bad_packed_object(e.p, real);
3030 return sha1_object_info_extended(real, oi, 0);
3031 } else if (oi->whence == OI_PACKED) {
3032 oi->u.packed.offset = e.offset;
3033 oi->u.packed.pack = e.p;
3034 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
3035 rtype == OBJ_OFS_DELTA);
3036 }
3037
3038 return 0;
3039 }
3040
3041 /* returns enum object_type or negative */
3042 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
3043 {
3044 enum object_type type;
3045 struct object_info oi = OBJECT_INFO_INIT;
3046
3047 oi.typep = &type;
3048 oi.sizep = sizep;
3049 if (sha1_object_info_extended(sha1, &oi,
3050 OBJECT_INFO_LOOKUP_REPLACE) < 0)
3051 return -1;
3052 return type;
3053 }
3054
3055 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
3056 unsigned char *sha1)
3057 {
3058 struct cached_object *co;
3059
3060 hash_sha1_file(buf, len, typename(type), sha1);
3061 if (has_sha1_file(sha1) || find_cached_object(sha1))
3062 return 0;
3063 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
3064 co = &cached_objects[cached_object_nr++];
3065 co->size = len;
3066 co->type = type;
3067 co->buf = xmalloc(len);
3068 memcpy(co->buf, buf, len);
3069 hashcpy(co->sha1, sha1);
3070 return 0;
3071 }
3072
3073 static void *read_object(const unsigned char *sha1, enum object_type *type,
3074 unsigned long *size)
3075 {
3076 struct object_info oi = OBJECT_INFO_INIT;
3077 void *content;
3078 oi.typep = type;
3079 oi.sizep = size;
3080 oi.contentp = &content;
3081
3082 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
3083 return NULL;
3084 return content;
3085 }
3086
3087 /*
3088 * This function dies on corrupt objects; the callers who want to
3089 * deal with them should arrange to call read_object() and give error
3090 * messages themselves.
3091 */
3092 void *read_sha1_file_extended(const unsigned char *sha1,
3093 enum object_type *type,
3094 unsigned long *size,
3095 int lookup_replace)
3096 {
3097 void *data;
3098 const struct packed_git *p;
3099 const char *path;
3100 struct stat st;
3101 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
3102 : sha1;
3103
3104 errno = 0;
3105 data = read_object(repl, type, size);
3106 if (data)
3107 return data;
3108
3109 if (errno && errno != ENOENT)
3110 die_errno("failed to read object %s", sha1_to_hex(sha1));
3111
3112 /* die if we replaced an object with one that does not exist */
3113 if (repl != sha1)
3114 die("replacement %s not found for %s",
3115 sha1_to_hex(repl), sha1_to_hex(sha1));
3116
3117 if (!stat_sha1_file(repl, &st, &path))
3118 die("loose object %s (stored in %s) is corrupt",
3119 sha1_to_hex(repl), path);
3120
3121 if ((p = has_packed_and_bad(repl)) != NULL)
3122 die("packed object %s (stored in %s) is corrupt",
3123 sha1_to_hex(repl), p->pack_name);
3124
3125 return NULL;
3126 }
3127
3128 void *read_object_with_reference(const unsigned char *sha1,
3129 const char *required_type_name,
3130 unsigned long *size,
3131 unsigned char *actual_sha1_return)
3132 {
3133 enum object_type type, required_type;
3134 void *buffer;
3135 unsigned long isize;
3136 unsigned char actual_sha1[20];
3137
3138 required_type = type_from_string(required_type_name);
3139 hashcpy(actual_sha1, sha1);
3140 while (1) {
3141 int ref_length = -1;
3142 const char *ref_type = NULL;
3143
3144 buffer = read_sha1_file(actual_sha1, &type, &isize);
3145 if (!buffer)
3146 return NULL;
3147 if (type == required_type) {
3148 *size = isize;
3149 if (actual_sha1_return)
3150 hashcpy(actual_sha1_return, actual_sha1);
3151 return buffer;
3152 }
3153 /* Handle references */
3154 else if (type == OBJ_COMMIT)
3155 ref_type = "tree ";
3156 else if (type == OBJ_TAG)
3157 ref_type = "object ";
3158 else {
3159 free(buffer);
3160 return NULL;
3161 }
3162 ref_length = strlen(ref_type);
3163
3164 if (ref_length + 40 > isize ||
3165 memcmp(buffer, ref_type, ref_length) ||
3166 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
3167 free(buffer);
3168 return NULL;
3169 }
3170 free(buffer);
3171 /* Now we have the ID of the referred-to object in
3172 * actual_sha1. Check again. */
3173 }
3174 }
3175
3176 static void write_sha1_file_prepare(const void *buf, unsigned long len,
3177 const char *type, unsigned char *sha1,
3178 char *hdr, int *hdrlen)
3179 {
3180 git_SHA_CTX c;
3181
3182 /* Generate the header */
3183 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
3184
3185 /* Sha1.. */
3186 git_SHA1_Init(&c);
3187 git_SHA1_Update(&c, hdr, *hdrlen);
3188 git_SHA1_Update(&c, buf, len);
3189 git_SHA1_Final(sha1, &c);
3190 }
3191
3192 /*
3193 * Move the just written object into its final resting place.
3194 */
3195 int finalize_object_file(const char *tmpfile, const char *filename)
3196 {
3197 int ret = 0;
3198
3199 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
3200 goto try_rename;
3201 else if (link(tmpfile, filename))
3202 ret = errno;
3203
3204 /*
3205 * Coda hack - coda doesn't like cross-directory links,
3206 * so we fall back to a rename, which will mean that it
3207 * won't be able to check collisions, but that's not a
3208 * big deal.
3209 *
3210 * The same holds for FAT formatted media.
3211 *
3212 * When this succeeds, we just return. We have nothing
3213 * left to unlink.
3214 */
3215 if (ret && ret != EEXIST) {
3216 try_rename:
3217 if (!rename(tmpfile, filename))
3218 goto out;
3219 ret = errno;
3220 }
3221 unlink_or_warn(tmpfile);
3222 if (ret) {
3223 if (ret != EEXIST) {
3224 return error_errno("unable to write sha1 filename %s", filename);
3225 }
3226 /* FIXME!!! Collision check here ? */
3227 }
3228
3229 out:
3230 if (adjust_shared_perm(filename))
3231 return error("unable to set permission to '%s'", filename);
3232 return 0;
3233 }
3234
3235 static int write_buffer(int fd, const void *buf, size_t len)
3236 {
3237 if (write_in_full(fd, buf, len) < 0)
3238 return error_errno("file write error");
3239 return 0;
3240 }
3241
3242 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
3243 unsigned char *sha1)
3244 {
3245 char hdr[32];
3246 int hdrlen = sizeof(hdr);
3247 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3248 return 0;
3249 }
3250
3251 /* Finalize a file on disk, and close it. */
3252 static void close_sha1_file(int fd)
3253 {
3254 if (fsync_object_files)
3255 fsync_or_die(fd, "sha1 file");
3256 if (close(fd) != 0)
3257 die_errno("error when closing sha1 file");
3258 }
3259
3260 /* Size of directory component, including the ending '/' */
3261 static inline int directory_size(const char *filename)
3262 {
3263 const char *s = strrchr(filename, '/');
3264 if (!s)
3265 return 0;
3266 return s - filename + 1;
3267 }
3268
3269 /*
3270 * This creates a temporary file in the same directory as the final
3271 * 'filename'
3272 *
3273 * We want to avoid cross-directory filename renames, because those
3274 * can have problems on various filesystems (FAT, NFS, Coda).
3275 */
3276 static int create_tmpfile(struct strbuf *tmp, const char *filename)
3277 {
3278 int fd, dirlen = directory_size(filename);
3279
3280 strbuf_reset(tmp);
3281 strbuf_add(tmp, filename, dirlen);
3282 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
3283 fd = git_mkstemp_mode(tmp->buf, 0444);
3284 if (fd < 0 && dirlen && errno == ENOENT) {
3285 /*
3286 * Make sure the directory exists; note that the contents
3287 * of the buffer are undefined after mkstemp returns an
3288 * error, so we have to rewrite the whole buffer from
3289 * scratch.
3290 */
3291 strbuf_reset(tmp);
3292 strbuf_add(tmp, filename, dirlen - 1);
3293 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
3294 return -1;
3295 if (adjust_shared_perm(tmp->buf))
3296 return -1;
3297
3298 /* Try again */
3299 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
3300 fd = git_mkstemp_mode(tmp->buf, 0444);
3301 }
3302 return fd;
3303 }
3304
3305 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
3306 const void *buf, unsigned long len, time_t mtime)
3307 {
3308 int fd, ret;
3309 unsigned char compressed[4096];
3310 git_zstream stream;
3311 git_SHA_CTX c;
3312 unsigned char parano_sha1[20];
3313 static struct strbuf tmp_file = STRBUF_INIT;
3314 const char *filename = sha1_file_name(sha1);
3315
3316 fd = create_tmpfile(&tmp_file, filename);
3317 if (fd < 0) {
3318 if (errno == EACCES)
3319 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3320 else
3321 return error_errno("unable to create temporary file");
3322 }
3323
3324 /* Set it up */
3325 git_deflate_init(&stream, zlib_compression_level);
3326 stream.next_out = compressed;
3327 stream.avail_out = sizeof(compressed);
3328 git_SHA1_Init(&c);
3329
3330 /* First header.. */
3331 stream.next_in = (unsigned char *)hdr;
3332 stream.avail_in = hdrlen;
3333 while (git_deflate(&stream, 0) == Z_OK)
3334 ; /* nothing */
3335 git_SHA1_Update(&c, hdr, hdrlen);
3336
3337 /* Then the data itself.. */
3338 stream.next_in = (void *)buf;
3339 stream.avail_in = len;
3340 do {
3341 unsigned char *in0 = stream.next_in;
3342 ret = git_deflate(&stream, Z_FINISH);
3343 git_SHA1_Update(&c, in0, stream.next_in - in0);
3344 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3345 die("unable to write sha1 file");
3346 stream.next_out = compressed;
3347 stream.avail_out = sizeof(compressed);
3348 } while (ret == Z_OK);
3349
3350 if (ret != Z_STREAM_END)
3351 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
3352 ret = git_deflate_end_gently(&stream);
3353 if (ret != Z_OK)
3354 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
3355 git_SHA1_Final(parano_sha1, &c);
3356 if (hashcmp(sha1, parano_sha1) != 0)
3357 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
3358
3359 close_sha1_file(fd);
3360
3361 if (mtime) {
3362 struct utimbuf utb;
3363 utb.actime = mtime;
3364 utb.modtime = mtime;
3365 if (utime(tmp_file.buf, &utb) < 0)
3366 warning_errno("failed utime() on %s", tmp_file.buf);
3367 }
3368
3369 return finalize_object_file(tmp_file.buf, filename);
3370 }
3371
3372 static int freshen_loose_object(const unsigned char *sha1)
3373 {
3374 return check_and_freshen(sha1, 1);
3375 }
3376
3377 static int freshen_packed_object(const unsigned char *sha1)
3378 {
3379 struct pack_entry e;
3380 if (!find_pack_entry(sha1, &e))
3381 return 0;
3382 if (e.p->freshened)
3383 return 1;
3384 if (!freshen_file(e.p->pack_name))
3385 return 0;
3386 e.p->freshened = 1;
3387 return 1;
3388 }
3389
3390 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3391 {
3392 char hdr[32];
3393 int hdrlen = sizeof(hdr);
3394
3395 /* Normally if we have it in the pack then we do not bother writing
3396 * it out into .git/objects/??/?{38} file.
3397 */
3398 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3399 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3400 return 0;
3401 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3402 }
3403
3404 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3405 unsigned char *sha1, unsigned flags)
3406 {
3407 char *header;
3408 int hdrlen, status = 0;
3409
3410 /* type string, SP, %lu of the length plus NUL must fit this */
3411 hdrlen = strlen(type) + 32;
3412 header = xmalloc(hdrlen);
3413 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3414
3415 if (!(flags & HASH_WRITE_OBJECT))
3416 goto cleanup;
3417 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3418 goto cleanup;
3419 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3420
3421 cleanup:
3422 free(header);
3423 return status;
3424 }
3425
3426 int force_object_loose(const unsigned char *sha1, time_t mtime)
3427 {
3428 void *buf;
3429 unsigned long len;
3430 enum object_type type;
3431 char hdr[32];
3432 int hdrlen;
3433 int ret;
3434
3435 if (has_loose_object(sha1))
3436 return 0;
3437 buf = read_object(sha1, &type, &len);
3438 if (!buf)
3439 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3440 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
3441 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3442 free(buf);
3443
3444 return ret;
3445 }
3446
3447 int has_pack_index(const unsigned char *sha1)
3448 {
3449 struct stat st;
3450 if (stat(sha1_pack_index_name(sha1), &st))
3451 return 0;
3452 return 1;
3453 }
3454
3455 int has_sha1_pack(const unsigned char *sha1)
3456 {
3457 struct pack_entry e;
3458 return find_pack_entry(sha1, &e);
3459 }
3460
3461 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3462 {
3463 if (!startup_info->have_repository)
3464 return 0;
3465 return sha1_object_info_extended(sha1, NULL,
3466 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
3467 }
3468
3469 int has_object_file(const struct object_id *oid)
3470 {
3471 return has_sha1_file(oid->hash);
3472 }
3473
3474 int has_object_file_with_flags(const struct object_id *oid, int flags)
3475 {
3476 return has_sha1_file_with_flags(oid->hash, flags);
3477 }
3478
3479 static void check_tree(const void *buf, size_t size)
3480 {
3481 struct tree_desc desc;
3482 struct name_entry entry;
3483
3484 init_tree_desc(&desc, buf, size);
3485 while (tree_entry(&desc, &entry))
3486 /* do nothing
3487 * tree_entry() will die() on malformed entries */
3488 ;
3489 }
3490
3491 static void check_commit(const void *buf, size_t size)
3492 {
3493 struct commit c;
3494 memset(&c, 0, sizeof(c));
3495 if (parse_commit_buffer(&c, buf, size))
3496 die("corrupt commit");
3497 }
3498
3499 static void check_tag(const void *buf, size_t size)
3500 {
3501 struct tag t;
3502 memset(&t, 0, sizeof(t));
3503 if (parse_tag_buffer(&t, buf, size))
3504 die("corrupt tag");
3505 }
3506
3507 static int index_mem(unsigned char *sha1, void *buf, size_t size,
3508 enum object_type type,
3509 const char *path, unsigned flags)
3510 {
3511 int ret, re_allocated = 0;
3512 int write_object = flags & HASH_WRITE_OBJECT;
3513
3514 if (!type)
3515 type = OBJ_BLOB;
3516
3517 /*
3518 * Convert blobs to git internal format
3519 */
3520 if ((type == OBJ_BLOB) && path) {
3521 struct strbuf nbuf = STRBUF_INIT;
3522 if (convert_to_git(&the_index, path, buf, size, &nbuf,
3523 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3524 buf = strbuf_detach(&nbuf, &size);
3525 re_allocated = 1;
3526 }
3527 }
3528 if (flags & HASH_FORMAT_CHECK) {
3529 if (type == OBJ_TREE)
3530 check_tree(buf, size);
3531 if (type == OBJ_COMMIT)
3532 check_commit(buf, size);
3533 if (type == OBJ_TAG)
3534 check_tag(buf, size);
3535 }
3536
3537 if (write_object)
3538 ret = write_sha1_file(buf, size, typename(type), sha1);
3539 else
3540 ret = hash_sha1_file(buf, size, typename(type), sha1);
3541 if (re_allocated)
3542 free(buf);
3543 return ret;
3544 }
3545
3546 static int index_stream_convert_blob(unsigned char *sha1, int fd,
3547 const char *path, unsigned flags)
3548 {
3549 int ret;
3550 const int write_object = flags & HASH_WRITE_OBJECT;
3551 struct strbuf sbuf = STRBUF_INIT;
3552
3553 assert(path);
3554 assert(would_convert_to_git_filter_fd(path));
3555
3556 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
3557 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3558
3559 if (write_object)
3560 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3561 sha1);
3562 else
3563 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3564 sha1);
3565 strbuf_release(&sbuf);
3566 return ret;
3567 }
3568
3569 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3570 const char *path, unsigned flags)
3571 {
3572 struct strbuf sbuf = STRBUF_INIT;
3573 int ret;
3574
3575 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3576 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3577 else
3578 ret = -1;
3579 strbuf_release(&sbuf);
3580 return ret;
3581 }
3582
3583 #define SMALL_FILE_SIZE (32*1024)
3584
3585 static int index_core(unsigned char *sha1, int fd, size_t size,
3586 enum object_type type, const char *path,
3587 unsigned flags)
3588 {
3589 int ret;
3590
3591 if (!size) {
3592 ret = index_mem(sha1, "", size, type, path, flags);
3593 } else if (size <= SMALL_FILE_SIZE) {
3594 char *buf = xmalloc(size);
3595 if (size == read_in_full(fd, buf, size))
3596 ret = index_mem(sha1, buf, size, type, path, flags);
3597 else
3598 ret = error_errno("short read");
3599 free(buf);
3600 } else {
3601 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3602 ret = index_mem(sha1, buf, size, type, path, flags);
3603 munmap(buf, size);
3604 }
3605 return ret;
3606 }
3607
3608 /*
3609 * This creates one packfile per large blob unless bulk-checkin
3610 * machinery is "plugged".
3611 *
3612 * This also bypasses the usual "convert-to-git" dance, and that is on
3613 * purpose. We could write a streaming version of the converting
3614 * functions and insert that before feeding the data to fast-import
3615 * (or equivalent in-core API described above). However, that is
3616 * somewhat complicated, as we do not know the size of the filter
3617 * result, which we need to know beforehand when writing a git object.
3618 * Since the primary motivation for trying to stream from the working
3619 * tree file and to avoid mmaping it in core is to deal with large
3620 * binary blobs, they generally do not want to get any conversion, and
3621 * callers should avoid this code path when filters are requested.
3622 */
3623 static int index_stream(unsigned char *sha1, int fd, size_t size,
3624 enum object_type type, const char *path,
3625 unsigned flags)
3626 {
3627 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3628 }
3629
3630 int index_fd(unsigned char *sha1, int fd, struct stat *st,
3631 enum object_type type, const char *path, unsigned flags)
3632 {
3633 int ret;
3634
3635 /*
3636 * Call xsize_t() only when needed to avoid potentially unnecessary
3637 * die() for large files.
3638 */
3639 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3640 ret = index_stream_convert_blob(sha1, fd, path, flags);
3641 else if (!S_ISREG(st->st_mode))
3642 ret = index_pipe(sha1, fd, type, path, flags);
3643 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3644 (path && would_convert_to_git(&the_index, path)))
3645 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3646 flags);
3647 else
3648 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3649 flags);
3650 close(fd);
3651 return ret;
3652 }
3653
3654 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3655 {
3656 int fd;
3657 struct strbuf sb = STRBUF_INIT;
3658
3659 switch (st->st_mode & S_IFMT) {
3660 case S_IFREG:
3661 fd = open(path, O_RDONLY);
3662 if (fd < 0)
3663 return error_errno("open(\"%s\")", path);
3664 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3665 return error("%s: failed to insert into database",
3666 path);
3667 break;
3668 case S_IFLNK:
3669 if (strbuf_readlink(&sb, path, st->st_size))
3670 return error_errno("readlink(\"%s\")", path);
3671 if (!(flags & HASH_WRITE_OBJECT))
3672 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3673 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3674 return error("%s: failed to insert into database",
3675 path);
3676 strbuf_release(&sb);
3677 break;
3678 case S_IFDIR:
3679 return resolve_gitlink_ref(path, "HEAD", sha1);
3680 default:
3681 return error("%s: unsupported file type", path);
3682 }
3683 return 0;
3684 }
3685
3686 int read_pack_header(int fd, struct pack_header *header)
3687 {
3688 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3689 /* "eof before pack header was fully read" */
3690 return PH_ERROR_EOF;
3691
3692 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3693 /* "protocol error (pack signature mismatch detected)" */
3694 return PH_ERROR_PACK_SIGNATURE;
3695 if (!pack_version_ok(header->hdr_version))
3696 /* "protocol error (pack version unsupported)" */
3697 return PH_ERROR_PROTOCOL;
3698 return 0;
3699 }
3700
3701 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3702 {
3703 enum object_type type = sha1_object_info(sha1, NULL);
3704 if (type < 0)
3705 die("%s is not a valid object", sha1_to_hex(sha1));
3706 if (type != expect)
3707 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3708 typename(expect));
3709 }
3710
3711 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
3712 struct strbuf *path,
3713 each_loose_object_fn obj_cb,
3714 each_loose_cruft_fn cruft_cb,
3715 each_loose_subdir_fn subdir_cb,
3716 void *data)
3717 {
3718 size_t origlen, baselen;
3719 DIR *dir;
3720 struct dirent *de;
3721 int r = 0;
3722
3723 if (subdir_nr > 0xff)
3724 BUG("invalid loose object subdirectory: %x", subdir_nr);
3725
3726 origlen = path->len;
3727 strbuf_complete(path, '/');
3728 strbuf_addf(path, "%02x", subdir_nr);
3729 baselen = path->len;
3730
3731 dir = opendir(path->buf);
3732 if (!dir) {
3733 if (errno != ENOENT)
3734 r = error_errno("unable to open %s", path->buf);
3735 strbuf_setlen(path, origlen);
3736 return r;
3737 }
3738
3739 while ((de = readdir(dir))) {
3740 if (is_dot_or_dotdot(de->d_name))
3741 continue;
3742
3743 strbuf_setlen(path, baselen);
3744 strbuf_addf(path, "/%s", de->d_name);
3745
3746 if (strlen(de->d_name) == GIT_SHA1_HEXSZ - 2) {
3747 char hex[GIT_MAX_HEXSZ+1];
3748 struct object_id oid;
3749
3750 xsnprintf(hex, sizeof(hex), "%02x%s",
3751 subdir_nr, de->d_name);
3752 if (!get_oid_hex(hex, &oid)) {
3753 if (obj_cb) {
3754 r = obj_cb(&oid, path->buf, data);
3755 if (r)
3756 break;
3757 }
3758 continue;
3759 }
3760 }
3761
3762 if (cruft_cb) {
3763 r = cruft_cb(de->d_name, path->buf, data);
3764 if (r)
3765 break;
3766 }
3767 }
3768 closedir(dir);
3769
3770 strbuf_setlen(path, baselen);
3771 if (!r && subdir_cb)
3772 r = subdir_cb(subdir_nr, path->buf, data);
3773
3774 strbuf_setlen(path, origlen);
3775
3776 return r;
3777 }
3778
3779 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3780 each_loose_object_fn obj_cb,
3781 each_loose_cruft_fn cruft_cb,
3782 each_loose_subdir_fn subdir_cb,
3783 void *data)
3784 {
3785 int r = 0;
3786 int i;
3787
3788 for (i = 0; i < 256; i++) {
3789 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3790 subdir_cb, data);
3791 if (r)
3792 break;
3793 }
3794
3795 return r;
3796 }
3797
3798 int for_each_loose_file_in_objdir(const char *path,
3799 each_loose_object_fn obj_cb,
3800 each_loose_cruft_fn cruft_cb,
3801 each_loose_subdir_fn subdir_cb,
3802 void *data)
3803 {
3804 struct strbuf buf = STRBUF_INIT;
3805 int r;
3806
3807 strbuf_addstr(&buf, path);
3808 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3809 subdir_cb, data);
3810 strbuf_release(&buf);
3811
3812 return r;
3813 }
3814
3815 struct loose_alt_odb_data {
3816 each_loose_object_fn *cb;
3817 void *data;
3818 };
3819
3820 static int loose_from_alt_odb(struct alternate_object_database *alt,
3821 void *vdata)
3822 {
3823 struct loose_alt_odb_data *data = vdata;
3824 struct strbuf buf = STRBUF_INIT;
3825 int r;
3826
3827 strbuf_addstr(&buf, alt->path);
3828 r = for_each_loose_file_in_objdir_buf(&buf,
3829 data->cb, NULL, NULL,
3830 data->data);
3831 strbuf_release(&buf);
3832 return r;
3833 }
3834
3835 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3836 {
3837 struct loose_alt_odb_data alt;
3838 int r;
3839
3840 r = for_each_loose_file_in_objdir(get_object_directory(),
3841 cb, NULL, NULL, data);
3842 if (r)
3843 return r;
3844
3845 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3846 return 0;
3847
3848 alt.cb = cb;
3849 alt.data = data;
3850 return foreach_alt_odb(loose_from_alt_odb, &alt);
3851 }
3852
3853 static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3854 {
3855 uint32_t i;
3856 int r = 0;
3857
3858 for (i = 0; i < p->num_objects; i++) {
3859 struct object_id oid;
3860
3861 if (!nth_packed_object_oid(&oid, p, i))
3862 return error("unable to get sha1 of object %u in %s",
3863 i, p->pack_name);
3864
3865 r = cb(&oid, p, i, data);
3866 if (r)</