link_alt_odb_entries: make empty input a noop
[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 void read_info_alternates(const char * relative_base, int depth);
351 static int link_alt_odb_entry(const char *entry, const char *relative_base,
352 int depth, const char *normalized_objdir)
353 {
354 struct alternate_object_database *ent;
355 struct strbuf pathbuf = STRBUF_INIT;
356
357 if (!is_absolute_path(entry) && relative_base) {
358 strbuf_realpath(&pathbuf, relative_base, 1);
359 strbuf_addch(&pathbuf, '/');
360 }
361 strbuf_addstr(&pathbuf, entry);
362
363 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
364 error("unable to normalize alternate object path: %s",
365 pathbuf.buf);
366 strbuf_release(&pathbuf);
367 return -1;
368 }
369
370 /*
371 * The trailing slash after the directory name is given by
372 * this function at the end. Remove duplicates.
373 */
374 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
375 strbuf_setlen(&pathbuf, pathbuf.len - 1);
376
377 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
378 strbuf_release(&pathbuf);
379 return -1;
380 }
381
382 ent = alloc_alt_odb(pathbuf.buf);
383
384 /* add the alternate entry */
385 *alt_odb_tail = ent;
386 alt_odb_tail = &(ent->next);
387 ent->next = NULL;
388
389 /* recursively add alternates */
390 read_info_alternates(pathbuf.buf, depth + 1);
391
392 strbuf_release(&pathbuf);
393 return 0;
394 }
395
396 static const char *parse_alt_odb_entry(const char *string,
397 int sep,
398 struct strbuf *out)
399 {
400 const char *end;
401
402 strbuf_reset(out);
403
404 if (*string == '#') {
405 /* comment; consume up to next separator */
406 end = strchrnul(string, sep);
407 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
408 /*
409 * quoted path; unquote_c_style has copied the
410 * data for us and set "end". Broken quoting (e.g.,
411 * an entry that doesn't end with a quote) falls
412 * back to the unquoted case below.
413 */
414 } else {
415 /* normal, unquoted path */
416 end = strchrnul(string, sep);
417 strbuf_add(out, string, end - string);
418 }
419
420 if (*end)
421 end++;
422 return end;
423 }
424
425 static void link_alt_odb_entries(const char *alt, int sep,
426 const char *relative_base, int depth)
427 {
428 struct strbuf objdirbuf = STRBUF_INIT;
429 struct strbuf entry = STRBUF_INIT;
430
431 if (!alt || !*alt)
432 return;
433
434 if (depth > 5) {
435 error("%s: ignoring alternate object stores, nesting too deep.",
436 relative_base);
437 return;
438 }
439
440 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
441 if (strbuf_normalize_path(&objdirbuf) < 0)
442 die("unable to normalize object directory: %s",
443 objdirbuf.buf);
444
445 while (*alt) {
446 alt = parse_alt_odb_entry(alt, sep, &entry);
447 if (!entry.len)
448 continue;
449 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
450 }
451 strbuf_release(&entry);
452 strbuf_release(&objdirbuf);
453 }
454
455 static void read_info_alternates(const char * relative_base, int depth)
456 {
457 char *path;
458 struct strbuf buf = STRBUF_INIT;
459
460 path = xstrfmt("%s/info/alternates", relative_base);
461 if (strbuf_read_file(&buf, path, 1024) < 0) {
462 warn_on_fopen_errors(path);
463 free(path);
464 return;
465 }
466
467 link_alt_odb_entries(buf.buf, '\n', relative_base, depth);
468 strbuf_release(&buf);
469 free(path);
470 }
471
472 struct alternate_object_database *alloc_alt_odb(const char *dir)
473 {
474 struct alternate_object_database *ent;
475
476 FLEX_ALLOC_STR(ent, path, dir);
477 strbuf_init(&ent->scratch, 0);
478 strbuf_addf(&ent->scratch, "%s/", dir);
479 ent->base_len = ent->scratch.len;
480
481 return ent;
482 }
483
484 void add_to_alternates_file(const char *reference)
485 {
486 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
487 char *alts = git_pathdup("objects/info/alternates");
488 FILE *in, *out;
489
490 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
491 out = fdopen_lock_file(lock, "w");
492 if (!out)
493 die_errno("unable to fdopen alternates lockfile");
494
495 in = fopen(alts, "r");
496 if (in) {
497 struct strbuf line = STRBUF_INIT;
498 int found = 0;
499
500 while (strbuf_getline(&line, in) != EOF) {
501 if (!strcmp(reference, line.buf)) {
502 found = 1;
503 break;
504 }
505 fprintf_or_die(out, "%s\n", line.buf);
506 }
507
508 strbuf_release(&line);
509 fclose(in);
510
511 if (found) {
512 rollback_lock_file(lock);
513 lock = NULL;
514 }
515 }
516 else if (errno != ENOENT)
517 die_errno("unable to read alternates file");
518
519 if (lock) {
520 fprintf_or_die(out, "%s\n", reference);
521 if (commit_lock_file(lock))
522 die_errno("unable to move new alternates file into place");
523 if (alt_odb_tail)
524 link_alt_odb_entries(reference, '\n', NULL, 0);
525 }
526 free(alts);
527 }
528
529 void add_to_alternates_memory(const char *reference)
530 {
531 /*
532 * Make sure alternates are initialized, or else our entry may be
533 * overwritten when they are.
534 */
535 prepare_alt_odb();
536
537 link_alt_odb_entries(reference, '\n', NULL, 0);
538 }
539
540 /*
541 * Compute the exact path an alternate is at and returns it. In case of
542 * error NULL is returned and the human readable error is added to `err`
543 * `path` may be relative and should point to $GITDIR.
544 * `err` must not be null.
545 */
546 char *compute_alternate_path(const char *path, struct strbuf *err)
547 {
548 char *ref_git = NULL;
549 const char *repo, *ref_git_s;
550 int seen_error = 0;
551
552 ref_git_s = real_path_if_valid(path);
553 if (!ref_git_s) {
554 seen_error = 1;
555 strbuf_addf(err, _("path '%s' does not exist"), path);
556 goto out;
557 } else
558 /*
559 * Beware: read_gitfile(), real_path() and mkpath()
560 * return static buffer
561 */
562 ref_git = xstrdup(ref_git_s);
563
564 repo = read_gitfile(ref_git);
565 if (!repo)
566 repo = read_gitfile(mkpath("%s/.git", ref_git));
567 if (repo) {
568 free(ref_git);
569 ref_git = xstrdup(repo);
570 }
571
572 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
573 char *ref_git_git = mkpathdup("%s/.git", ref_git);
574 free(ref_git);
575 ref_git = ref_git_git;
576 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
577 struct strbuf sb = STRBUF_INIT;
578 seen_error = 1;
579 if (get_common_dir(&sb, ref_git)) {
580 strbuf_addf(err,
581 _("reference repository '%s' as a linked "
582 "checkout is not supported yet."),
583 path);
584 goto out;
585 }
586
587 strbuf_addf(err, _("reference repository '%s' is not a "
588 "local repository."), path);
589 goto out;
590 }
591
592 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
593 strbuf_addf(err, _("reference repository '%s' is shallow"),
594 path);
595 seen_error = 1;
596 goto out;
597 }
598
599 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
600 strbuf_addf(err,
601 _("reference repository '%s' is grafted"),
602 path);
603 seen_error = 1;
604 goto out;
605 }
606
607 out:
608 if (seen_error) {
609 FREE_AND_NULL(ref_git);
610 }
611
612 return ref_git;
613 }
614
615 int foreach_alt_odb(alt_odb_fn fn, void *cb)
616 {
617 struct alternate_object_database *ent;
618 int r = 0;
619
620 prepare_alt_odb();
621 for (ent = alt_odb_list; ent; ent = ent->next) {
622 r = fn(ent, cb);
623 if (r)
624 break;
625 }
626 return r;
627 }
628
629 void prepare_alt_odb(void)
630 {
631 const char *alt;
632
633 if (alt_odb_tail)
634 return;
635
636 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
637
638 alt_odb_tail = &alt_odb_list;
639 link_alt_odb_entries(alt, PATH_SEP, NULL, 0);
640
641 read_info_alternates(get_object_directory(), 0);
642 }
643
644 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
645 static int freshen_file(const char *fn)
646 {
647 struct utimbuf t;
648 t.actime = t.modtime = time(NULL);
649 return !utime(fn, &t);
650 }
651
652 /*
653 * All of the check_and_freshen functions return 1 if the file exists and was
654 * freshened (if freshening was requested), 0 otherwise. If they return
655 * 0, you should not assume that it is safe to skip a write of the object (it
656 * either does not exist on disk, or has a stale mtime and may be subject to
657 * pruning).
658 */
659 int check_and_freshen_file(const char *fn, int freshen)
660 {
661 if (access(fn, F_OK))
662 return 0;
663 if (freshen && !freshen_file(fn))
664 return 0;
665 return 1;
666 }
667
668 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
669 {
670 return check_and_freshen_file(sha1_file_name(sha1), freshen);
671 }
672
673 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
674 {
675 struct alternate_object_database *alt;
676 prepare_alt_odb();
677 for (alt = alt_odb_list; alt; alt = alt->next) {
678 const char *path = alt_sha1_path(alt, sha1);
679 if (check_and_freshen_file(path, freshen))
680 return 1;
681 }
682 return 0;
683 }
684
685 static int check_and_freshen(const unsigned char *sha1, int freshen)
686 {
687 return check_and_freshen_local(sha1, freshen) ||
688 check_and_freshen_nonlocal(sha1, freshen);
689 }
690
691 int has_loose_object_nonlocal(const unsigned char *sha1)
692 {
693 return check_and_freshen_nonlocal(sha1, 0);
694 }
695
696 static int has_loose_object(const unsigned char *sha1)
697 {
698 return check_and_freshen(sha1, 0);
699 }
700
701 static unsigned int pack_used_ctr;
702 static unsigned int pack_mmap_calls;
703 static unsigned int peak_pack_open_windows;
704 static unsigned int pack_open_windows;
705 static unsigned int pack_open_fds;
706 static unsigned int pack_max_fds;
707 static size_t peak_pack_mapped;
708 static size_t pack_mapped;
709 struct packed_git *packed_git;
710
711 static struct mru packed_git_mru_storage;
712 struct mru *packed_git_mru = &packed_git_mru_storage;
713
714 void pack_report(void)
715 {
716 fprintf(stderr,
717 "pack_report: getpagesize() = %10" SZ_FMT "\n"
718 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
719 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
720 sz_fmt(getpagesize()),
721 sz_fmt(packed_git_window_size),
722 sz_fmt(packed_git_limit));
723 fprintf(stderr,
724 "pack_report: pack_used_ctr = %10u\n"
725 "pack_report: pack_mmap_calls = %10u\n"
726 "pack_report: pack_open_windows = %10u / %10u\n"
727 "pack_report: pack_mapped = "
728 "%10" SZ_FMT " / %10" SZ_FMT "\n",
729 pack_used_ctr,
730 pack_mmap_calls,
731 pack_open_windows, peak_pack_open_windows,
732 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
733 }
734
735 /*
736 * Open and mmap the index file at path, perform a couple of
737 * consistency checks, then record its information to p. Return 0 on
738 * success.
739 */
740 static int check_packed_git_idx(const char *path, struct packed_git *p)
741 {
742 void *idx_map;
743 struct pack_idx_header *hdr;
744 size_t idx_size;
745 uint32_t version, nr, i, *index;
746 int fd = git_open(path);
747 struct stat st;
748
749 if (fd < 0)
750 return -1;
751 if (fstat(fd, &st)) {
752 close(fd);
753 return -1;
754 }
755 idx_size = xsize_t(st.st_size);
756 if (idx_size < 4 * 256 + 20 + 20) {
757 close(fd);
758 return error("index file %s is too small", path);
759 }
760 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
761 close(fd);
762
763 hdr = idx_map;
764 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
765 version = ntohl(hdr->idx_version);
766 if (version < 2 || version > 2) {
767 munmap(idx_map, idx_size);
768 return error("index file %s is version %"PRIu32
769 " and is not supported by this binary"
770 " (try upgrading GIT to a newer version)",
771 path, version);
772 }
773 } else
774 version = 1;
775
776 nr = 0;
777 index = idx_map;
778 if (version > 1)
779 index += 2; /* skip index header */
780 for (i = 0; i < 256; i++) {
781 uint32_t n = ntohl(index[i]);
782 if (n < nr) {
783 munmap(idx_map, idx_size);
784 return error("non-monotonic index %s", path);
785 }
786 nr = n;
787 }
788
789 if (version == 1) {
790 /*
791 * Total size:
792 * - 256 index entries 4 bytes each
793 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
794 * - 20-byte SHA1 of the packfile
795 * - 20-byte SHA1 file checksum
796 */
797 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
798 munmap(idx_map, idx_size);
799 return error("wrong index v1 file size in %s", path);
800 }
801 } else if (version == 2) {
802 /*
803 * Minimum size:
804 * - 8 bytes of header
805 * - 256 index entries 4 bytes each
806 * - 20-byte sha1 entry * nr
807 * - 4-byte crc entry * nr
808 * - 4-byte offset entry * nr
809 * - 20-byte SHA1 of the packfile
810 * - 20-byte SHA1 file checksum
811 * And after the 4-byte offset table might be a
812 * variable sized table containing 8-byte entries
813 * for offsets larger than 2^31.
814 */
815 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
816 unsigned long max_size = min_size;
817 if (nr)
818 max_size += (nr - 1)*8;
819 if (idx_size < min_size || idx_size > max_size) {
820 munmap(idx_map, idx_size);
821 return error("wrong index v2 file size in %s", path);
822 }
823 if (idx_size != min_size &&
824 /*
825 * make sure we can deal with large pack offsets.
826 * 31-bit signed offset won't be enough, neither
827 * 32-bit unsigned one will be.
828 */
829 (sizeof(off_t) <= 4)) {
830 munmap(idx_map, idx_size);
831 return error("pack too large for current definition of off_t in %s", path);
832 }
833 }
834
835 p->index_version = version;
836 p->index_data = idx_map;
837 p->index_size = idx_size;
838 p->num_objects = nr;
839 return 0;
840 }
841
842 int open_pack_index(struct packed_git *p)
843 {
844 char *idx_name;
845 size_t len;
846 int ret;
847
848 if (p->index_data)
849 return 0;
850
851 if (!strip_suffix(p->pack_name, ".pack", &len))
852 die("BUG: pack_name does not end in .pack");
853 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
854 ret = check_packed_git_idx(idx_name, p);
855 free(idx_name);
856 return ret;
857 }
858
859 static void scan_windows(struct packed_git *p,
860 struct packed_git **lru_p,
861 struct pack_window **lru_w,
862 struct pack_window **lru_l)
863 {
864 struct pack_window *w, *w_l;
865
866 for (w_l = NULL, w = p->windows; w; w = w->next) {
867 if (!w->inuse_cnt) {
868 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
869 *lru_p = p;
870 *lru_w = w;
871 *lru_l = w_l;
872 }
873 }
874 w_l = w;
875 }
876 }
877
878 static int unuse_one_window(struct packed_git *current)
879 {
880 struct packed_git *p, *lru_p = NULL;
881 struct pack_window *lru_w = NULL, *lru_l = NULL;
882
883 if (current)
884 scan_windows(current, &lru_p, &lru_w, &lru_l);
885 for (p = packed_git; p; p = p->next)
886 scan_windows(p, &lru_p, &lru_w, &lru_l);
887 if (lru_p) {
888 munmap(lru_w->base, lru_w->len);
889 pack_mapped -= lru_w->len;
890 if (lru_l)
891 lru_l->next = lru_w->next;
892 else
893 lru_p->windows = lru_w->next;
894 free(lru_w);
895 pack_open_windows--;
896 return 1;
897 }
898 return 0;
899 }
900
901 void release_pack_memory(size_t need)
902 {
903 size_t cur = pack_mapped;
904 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
905 ; /* nothing */
906 }
907
908 static void mmap_limit_check(size_t length)
909 {
910 static size_t limit = 0;
911 if (!limit) {
912 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
913 if (!limit)
914 limit = SIZE_MAX;
915 }
916 if (length > limit)
917 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
918 (uintmax_t)length, (uintmax_t)limit);
919 }
920
921 void *xmmap_gently(void *start, size_t length,
922 int prot, int flags, int fd, off_t offset)
923 {
924 void *ret;
925
926 mmap_limit_check(length);
927 ret = mmap(start, length, prot, flags, fd, offset);
928 if (ret == MAP_FAILED) {
929 if (!length)
930 return NULL;
931 release_pack_memory(length);
932 ret = mmap(start, length, prot, flags, fd, offset);
933 }
934 return ret;
935 }
936
937 void *xmmap(void *start, size_t length,
938 int prot, int flags, int fd, off_t offset)
939 {
940 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
941 if (ret == MAP_FAILED)
942 die_errno("mmap failed");
943 return ret;
944 }
945
946 void close_pack_windows(struct packed_git *p)
947 {
948 while (p->windows) {
949 struct pack_window *w = p->windows;
950
951 if (w->inuse_cnt)
952 die("pack '%s' still has open windows to it",
953 p->pack_name);
954 munmap(w->base, w->len);
955 pack_mapped -= w->len;
956 pack_open_windows--;
957 p->windows = w->next;
958 free(w);
959 }
960 }
961
962 static int close_pack_fd(struct packed_git *p)
963 {
964 if (p->pack_fd < 0)
965 return 0;
966
967 close(p->pack_fd);
968 pack_open_fds--;
969 p->pack_fd = -1;
970
971 return 1;
972 }
973
974 static void close_pack(struct packed_git *p)
975 {
976 close_pack_windows(p);
977 close_pack_fd(p);
978 close_pack_index(p);
979 }
980
981 void close_all_packs(void)
982 {
983 struct packed_git *p;
984
985 for (p = packed_git; p; p = p->next)
986 if (p->do_not_close)
987 die("BUG: want to close pack marked 'do-not-close'");
988 else
989 close_pack(p);
990 }
991
992
993 /*
994 * The LRU pack is the one with the oldest MRU window, preferring packs
995 * with no used windows, or the oldest mtime if it has no windows allocated.
996 */
997 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
998 {
999 struct pack_window *w, *this_mru_w;
1000 int has_windows_inuse = 0;
1001
1002 /*
1003 * Reject this pack if it has windows and the previously selected
1004 * one does not. If this pack does not have windows, reject
1005 * it if the pack file is newer than the previously selected one.
1006 */
1007 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
1008 return;
1009
1010 for (w = this_mru_w = p->windows; w; w = w->next) {
1011 /*
1012 * Reject this pack if any of its windows are in use,
1013 * but the previously selected pack did not have any
1014 * inuse windows. Otherwise, record that this pack
1015 * has windows in use.
1016 */
1017 if (w->inuse_cnt) {
1018 if (*accept_windows_inuse)
1019 has_windows_inuse = 1;
1020 else
1021 return;
1022 }
1023
1024 if (w->last_used > this_mru_w->last_used)
1025 this_mru_w = w;
1026
1027 /*
1028 * Reject this pack if it has windows that have been
1029 * used more recently than the previously selected pack.
1030 * If the previously selected pack had windows inuse and
1031 * we have not encountered a window in this pack that is
1032 * inuse, skip this check since we prefer a pack with no
1033 * inuse windows to one that has inuse windows.
1034 */
1035 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
1036 this_mru_w->last_used > (*mru_w)->last_used)
1037 return;
1038 }
1039
1040 /*
1041 * Select this pack.
1042 */
1043 *mru_w = this_mru_w;
1044 *lru_p = p;
1045 *accept_windows_inuse = has_windows_inuse;
1046 }
1047
1048 static int close_one_pack(void)
1049 {
1050 struct packed_git *p, *lru_p = NULL;
1051 struct pack_window *mru_w = NULL;
1052 int accept_windows_inuse = 1;
1053
1054 for (p = packed_git; p; p = p->next) {
1055 if (p->pack_fd == -1)
1056 continue;
1057 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
1058 }
1059
1060 if (lru_p)
1061 return close_pack_fd(lru_p);
1062
1063 return 0;
1064 }
1065
1066 void unuse_pack(struct pack_window **w_cursor)
1067 {
1068 struct pack_window *w = *w_cursor;
1069 if (w) {
1070 w->inuse_cnt--;
1071 *w_cursor = NULL;
1072 }
1073 }
1074
1075 void close_pack_index(struct packed_git *p)
1076 {
1077 if (p->index_data) {
1078 munmap((void *)p->index_data, p->index_size);
1079 p->index_data = NULL;
1080 }
1081 }
1082
1083 static unsigned int get_max_fd_limit(void)
1084 {
1085 #ifdef RLIMIT_NOFILE
1086 {
1087 struct rlimit lim;
1088
1089 if (!getrlimit(RLIMIT_NOFILE, &lim))
1090 return lim.rlim_cur;
1091 }
1092 #endif
1093
1094 #ifdef _SC_OPEN_MAX
1095 {
1096 long open_max = sysconf(_SC_OPEN_MAX);
1097 if (0 < open_max)
1098 return open_max;
1099 /*
1100 * Otherwise, we got -1 for one of the two
1101 * reasons:
1102 *
1103 * (1) sysconf() did not understand _SC_OPEN_MAX
1104 * and signaled an error with -1; or
1105 * (2) sysconf() said there is no limit.
1106 *
1107 * We _could_ clear errno before calling sysconf() to
1108 * tell these two cases apart and return a huge number
1109 * in the latter case to let the caller cap it to a
1110 * value that is not so selfish, but letting the
1111 * fallback OPEN_MAX codepath take care of these cases
1112 * is a lot simpler.
1113 */
1114 }
1115 #endif
1116
1117 #ifdef OPEN_MAX
1118 return OPEN_MAX;
1119 #else
1120 return 1; /* see the caller ;-) */
1121 #endif
1122 }
1123
1124 /*
1125 * Do not call this directly as this leaks p->pack_fd on error return;
1126 * call open_packed_git() instead.
1127 */
1128 static int open_packed_git_1(struct packed_git *p)
1129 {
1130 struct stat st;
1131 struct pack_header hdr;
1132 unsigned char sha1[20];
1133 unsigned char *idx_sha1;
1134 long fd_flag;
1135
1136 if (!p->index_data && open_pack_index(p))
1137 return error("packfile %s index unavailable", p->pack_name);
1138
1139 if (!pack_max_fds) {
1140 unsigned int max_fds = get_max_fd_limit();
1141
1142 /* Save 3 for stdin/stdout/stderr, 22 for work */
1143 if (25 < max_fds)
1144 pack_max_fds = max_fds - 25;
1145 else
1146 pack_max_fds = 1;
1147 }
1148
1149 while (pack_max_fds <= pack_open_fds && close_one_pack())
1150 ; /* nothing */
1151
1152 p->pack_fd = git_open(p->pack_name);
1153 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
1154 return -1;
1155 pack_open_fds++;
1156
1157 /* If we created the struct before we had the pack we lack size. */
1158 if (!p->pack_size) {
1159 if (!S_ISREG(st.st_mode))
1160 return error("packfile %s not a regular file", p->pack_name);
1161 p->pack_size = st.st_size;
1162 } else if (p->pack_size != st.st_size)
1163 return error("packfile %s size changed", p->pack_name);
1164
1165 /* We leave these file descriptors open with sliding mmap;
1166 * there is no point keeping them open across exec(), though.
1167 */
1168 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
1169 if (fd_flag < 0)
1170 return error("cannot determine file descriptor flags");
1171 fd_flag |= FD_CLOEXEC;
1172 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
1173 return error("cannot set FD_CLOEXEC");
1174
1175 /* Verify we recognize this pack file format. */
1176 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
1177 return error("file %s is far too short to be a packfile", p->pack_name);
1178 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
1179 return error("file %s is not a GIT packfile", p->pack_name);
1180 if (!pack_version_ok(hdr.hdr_version))
1181 return error("packfile %s is version %"PRIu32" and not"
1182 " supported (try upgrading GIT to a newer version)",
1183 p->pack_name, ntohl(hdr.hdr_version));
1184
1185 /* Verify the pack matches its index. */
1186 if (p->num_objects != ntohl(hdr.hdr_entries))
1187 return error("packfile %s claims to have %"PRIu32" objects"
1188 " while index indicates %"PRIu32" objects",
1189 p->pack_name, ntohl(hdr.hdr_entries),
1190 p->num_objects);
1191 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
1192 return error("end of packfile %s is unavailable", p->pack_name);
1193 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
1194 return error("packfile %s signature is unavailable", p->pack_name);
1195 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
1196 if (hashcmp(sha1, idx_sha1))
1197 return error("packfile %s does not match index", p->pack_name);
1198 return 0;
1199 }
1200
1201 static int open_packed_git(struct packed_git *p)
1202 {
1203 if (!open_packed_git_1(p))
1204 return 0;
1205 close_pack_fd(p);
1206 return -1;
1207 }
1208
1209 static int in_window(struct pack_window *win, off_t offset)
1210 {
1211 /* We must promise at least 20 bytes (one hash) after the
1212 * offset is available from this window, otherwise the offset
1213 * is not actually in this window and a different window (which
1214 * has that one hash excess) must be used. This is to support
1215 * the object header and delta base parsing routines below.
1216 */
1217 off_t win_off = win->offset;
1218 return win_off <= offset
1219 && (offset + 20) <= (win_off + win->len);
1220 }
1221
1222 unsigned char *use_pack(struct packed_git *p,
1223 struct pack_window **w_cursor,
1224 off_t offset,
1225 unsigned long *left)
1226 {
1227 struct pack_window *win = *w_cursor;
1228
1229 /* Since packfiles end in a hash of their content and it's
1230 * pointless to ask for an offset into the middle of that
1231 * hash, and the in_window function above wouldn't match
1232 * don't allow an offset too close to the end of the file.
1233 */
1234 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1235 die("packfile %s cannot be accessed", p->pack_name);
1236 if (offset > (p->pack_size - 20))
1237 die("offset beyond end of packfile (truncated pack?)");
1238 if (offset < 0)
1239 die(_("offset before end of packfile (broken .idx?)"));
1240
1241 if (!win || !in_window(win, offset)) {
1242 if (win)
1243 win->inuse_cnt--;
1244 for (win = p->windows; win; win = win->next) {
1245 if (in_window(win, offset))
1246 break;
1247 }
1248 if (!win) {
1249 size_t window_align = packed_git_window_size / 2;
1250 off_t len;
1251
1252 if (p->pack_fd == -1 && open_packed_git(p))
1253 die("packfile %s cannot be accessed", p->pack_name);
1254
1255 win = xcalloc(1, sizeof(*win));
1256 win->offset = (offset / window_align) * window_align;
1257 len = p->pack_size - win->offset;
1258 if (len > packed_git_window_size)
1259 len = packed_git_window_size;
1260 win->len = (size_t)len;
1261 pack_mapped += win->len;
1262 while (packed_git_limit < pack_mapped
1263 && unuse_one_window(p))
1264 ; /* nothing */
1265 win->base = xmmap(NULL, win->len,
1266 PROT_READ, MAP_PRIVATE,
1267 p->pack_fd, win->offset);
1268 if (win->base == MAP_FAILED)
1269 die_errno("packfile %s cannot be mapped",
1270 p->pack_name);
1271 if (!win->offset && win->len == p->pack_size
1272 && !p->do_not_close)
1273 close_pack_fd(p);
1274 pack_mmap_calls++;
1275 pack_open_windows++;
1276 if (pack_mapped > peak_pack_mapped)
1277 peak_pack_mapped = pack_mapped;
1278 if (pack_open_windows > peak_pack_open_windows)
1279 peak_pack_open_windows = pack_open_windows;
1280 win->next = p->windows;
1281 p->windows = win;
1282 }
1283 }
1284 if (win != *w_cursor) {
1285 win->last_used = pack_used_ctr++;
1286 win->inuse_cnt++;
1287 *w_cursor = win;
1288 }
1289 offset -= win->offset;
1290 if (left)
1291 *left = win->len - xsize_t(offset);
1292 return win->base + offset;
1293 }
1294
1295 static struct packed_git *alloc_packed_git(int extra)
1296 {
1297 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
1298 memset(p, 0, sizeof(*p));
1299 p->pack_fd = -1;
1300 return p;
1301 }
1302
1303 static void try_to_free_pack_memory(size_t size)
1304 {
1305 release_pack_memory(size);
1306 }
1307
1308 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
1309 {
1310 static int have_set_try_to_free_routine;
1311 struct stat st;
1312 size_t alloc;
1313 struct packed_git *p;
1314
1315 if (!have_set_try_to_free_routine) {
1316 have_set_try_to_free_routine = 1;
1317 set_try_to_free_routine(try_to_free_pack_memory);
1318 }
1319
1320 /*
1321 * Make sure a corresponding .pack file exists and that
1322 * the index looks sane.
1323 */
1324 if (!strip_suffix_mem(path, &path_len, ".idx"))
1325 return NULL;
1326
1327 /*
1328 * ".pack" is long enough to hold any suffix we're adding (and
1329 * the use xsnprintf double-checks that)
1330 */
1331 alloc = st_add3(path_len, strlen(".pack"), 1);
1332 p = alloc_packed_git(alloc);
1333 memcpy(p->pack_name, path, path_len);
1334
1335 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
1336 if (!access(p->pack_name, F_OK))
1337 p->pack_keep = 1;
1338
1339 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
1340 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1341 free(p);
1342 return NULL;
1343 }
1344
1345 /* ok, it looks sane as far as we can check without
1346 * actually mapping the pack file.
1347 */
1348 p->pack_size = st.st_size;
1349 p->pack_local = local;
1350 p->mtime = st.st_mtime;
1351 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1352 hashclr(p->sha1);
1353 return p;
1354 }
1355
1356 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1357 {
1358 const char *path = sha1_pack_name(sha1);
1359 size_t alloc = st_add(strlen(path), 1);
1360 struct packed_git *p = alloc_packed_git(alloc);
1361
1362 memcpy(p->pack_name, path, alloc); /* includes NUL */
1363 hashcpy(p->sha1, sha1);
1364 if (check_packed_git_idx(idx_path, p)) {
1365 free(p);
1366 return NULL;
1367 }
1368
1369 return p;
1370 }
1371
1372 void install_packed_git(struct packed_git *pack)
1373 {
1374 if (pack->pack_fd != -1)
1375 pack_open_fds++;
1376
1377 pack->next = packed_git;
1378 packed_git = pack;
1379 }
1380
1381 void (*report_garbage)(unsigned seen_bits, const char *path);
1382
1383 static void report_helper(const struct string_list *list,
1384 int seen_bits, int first, int last)
1385 {
1386 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
1387 return;
1388
1389 for (; first < last; first++)
1390 report_garbage(seen_bits, list->items[first].string);
1391 }
1392
1393 static void report_pack_garbage(struct string_list *list)
1394 {
1395 int i, baselen = -1, first = 0, seen_bits = 0;
1396
1397 if (!report_garbage)
1398 return;
1399
1400 string_list_sort(list);
1401
1402 for (i = 0; i < list->nr; i++) {
1403 const char *path = list->items[i].string;
1404 if (baselen != -1 &&
1405 strncmp(path, list->items[first].string, baselen)) {
1406 report_helper(list, seen_bits, first, i);
1407 baselen = -1;
1408 seen_bits = 0;
1409 }
1410 if (baselen == -1) {
1411 const char *dot = strrchr(path, '.');
1412 if (!dot) {
1413 report_garbage(PACKDIR_FILE_GARBAGE, path);
1414 continue;
1415 }
1416 baselen = dot - path + 1;
1417 first = i;
1418 }
1419 if (!strcmp(path + baselen, "pack"))
1420 seen_bits |= 1;
1421 else if (!strcmp(path + baselen, "idx"))
1422 seen_bits |= 2;
1423 }
1424 report_helper(list, seen_bits, first, list->nr);
1425 }
1426
1427 static void prepare_packed_git_one(char *objdir, int local)
1428 {
1429 struct strbuf path = STRBUF_INIT;
1430 size_t dirnamelen;
1431 DIR *dir;
1432 struct dirent *de;
1433 struct string_list garbage = STRING_LIST_INIT_DUP;
1434
1435 strbuf_addstr(&path, objdir);
1436 strbuf_addstr(&path, "/pack");
1437 dir = opendir(path.buf);
1438 if (!dir) {
1439 if (errno != ENOENT)
1440 error_errno("unable to open object pack directory: %s",
1441 path.buf);
1442 strbuf_release(&path);
1443 return;
1444 }
1445 strbuf_addch(&path, '/');
1446 dirnamelen = path.len;
1447 while ((de = readdir(dir)) != NULL) {
1448 struct packed_git *p;
1449 size_t base_len;
1450
1451 if (is_dot_or_dotdot(de->d_name))
1452 continue;
1453
1454 strbuf_setlen(&path, dirnamelen);
1455 strbuf_addstr(&path, de->d_name);
1456
1457 base_len = path.len;
1458 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1459 /* Don't reopen a pack we already have. */
1460 for (p = packed_git; p; p = p->next) {
1461 size_t len;
1462 if (strip_suffix(p->pack_name, ".pack", &len) &&
1463 len == base_len &&
1464 !memcmp(p->pack_name, path.buf, len))
1465 break;
1466 }
1467 if (p == NULL &&
1468 /*
1469 * See if it really is a valid .idx file with
1470 * corresponding .pack file that we can map.
1471 */
1472 (p = add_packed_git(path.buf, path.len, local)) != NULL)
1473 install_packed_git(p);
1474 }
1475
1476 if (!report_garbage)
1477 continue;
1478
1479 if (ends_with(de->d_name, ".idx") ||
1480 ends_with(de->d_name, ".pack") ||
1481 ends_with(de->d_name, ".bitmap") ||
1482 ends_with(de->d_name, ".keep"))
1483 string_list_append(&garbage, path.buf);
1484 else
1485 report_garbage(PACKDIR_FILE_GARBAGE, path.buf);
1486 }
1487 closedir(dir);
1488 report_pack_garbage(&garbage);
1489 string_list_clear(&garbage, 0);
1490 strbuf_release(&path);
1491 }
1492
1493 static int approximate_object_count_valid;
1494
1495 /*
1496 * Give a fast, rough count of the number of objects in the repository. This
1497 * ignores loose objects completely. If you have a lot of them, then either
1498 * you should repack because your performance will be awful, or they are
1499 * all unreachable objects about to be pruned, in which case they're not really
1500 * interesting as a measure of repo size in the first place.
1501 */
1502 unsigned long approximate_object_count(void)
1503 {
1504 static unsigned long count;
1505 if (!approximate_object_count_valid) {
1506 struct packed_git *p;
1507
1508 prepare_packed_git();
1509 count = 0;
1510 for (p = packed_git; p; p = p->next) {
1511 if (open_pack_index(p))
1512 continue;
1513 count += p->num_objects;
1514 }
1515 }
1516 return count;
1517 }
1518
1519 static void *get_next_packed_git(const void *p)
1520 {
1521 return ((const struct packed_git *)p)->next;
1522 }
1523
1524 static void set_next_packed_git(void *p, void *next)
1525 {
1526 ((struct packed_git *)p)->next = next;
1527 }
1528
1529 static int sort_pack(const void *a_, const void *b_)
1530 {
1531 const struct packed_git *a = a_;
1532 const struct packed_git *b = b_;
1533 int st;
1534
1535 /*
1536 * Local packs tend to contain objects specific to our
1537 * variant of the project than remote ones. In addition,
1538 * remote ones could be on a network mounted filesystem.
1539 * Favor local ones for these reasons.
1540 */
1541 st = a->pack_local - b->pack_local;
1542 if (st)
1543 return -st;
1544
1545 /*
1546 * Younger packs tend to contain more recent objects,
1547 * and more recent objects tend to get accessed more
1548 * often.
1549 */
1550 if (a->mtime < b->mtime)
1551 return 1;
1552 else if (a->mtime == b->mtime)
1553 return 0;
1554 return -1;
1555 }
1556
1557 static void rearrange_packed_git(void)
1558 {
1559 packed_git = llist_mergesort(packed_git, get_next_packed_git,
1560 set_next_packed_git, sort_pack);
1561 }
1562
1563 static void prepare_packed_git_mru(void)
1564 {
1565 struct packed_git *p;
1566
1567 mru_clear(packed_git_mru);
1568 for (p = packed_git; p; p = p->next)
1569 mru_append(packed_git_mru, p);
1570 }
1571
1572 static int prepare_packed_git_run_once = 0;
1573 void prepare_packed_git(void)
1574 {
1575 struct alternate_object_database *alt;
1576
1577 if (prepare_packed_git_run_once)
1578 return;
1579 prepare_packed_git_one(get_object_directory(), 1);
1580 prepare_alt_odb();
1581 for (alt = alt_odb_list; alt; alt = alt->next)
1582 prepare_packed_git_one(alt->path, 0);
1583 rearrange_packed_git();
1584 prepare_packed_git_mru();
1585 prepare_packed_git_run_once = 1;
1586 }
1587
1588 void reprepare_packed_git(void)
1589 {
1590 approximate_object_count_valid = 0;
1591 prepare_packed_git_run_once = 0;
1592 prepare_packed_git();
1593 }
1594
1595 static void mark_bad_packed_object(struct packed_git *p,
1596 const unsigned char *sha1)
1597 {
1598 unsigned i;
1599 for (i = 0; i < p->num_bad_objects; i++)
1600 if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
1601 return;
1602 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1603 st_mult(GIT_MAX_RAWSZ,
1604 st_add(p->num_bad_objects, 1)));
1605 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
1606 p->num_bad_objects++;
1607 }
1608
1609 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1610 {
1611 struct packed_git *p;
1612 unsigned i;
1613
1614 for (p = packed_git; p; p = p->next)
1615 for (i = 0; i < p->num_bad_objects; i++)
1616 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1617 return p;
1618 return NULL;
1619 }
1620
1621 /*
1622 * With an in-core object data in "map", rehash it to make sure the
1623 * object name actually matches "sha1" to detect object corruption.
1624 * With "map" == NULL, try reading the object named with "sha1" using
1625 * the streaming interface and rehash it to do the same.
1626 */
1627 int check_sha1_signature(const unsigned char *sha1, void *map,
1628 unsigned long size, const char *type)
1629 {
1630 unsigned char real_sha1[20];
1631 enum object_type obj_type;
1632 struct git_istream *st;
1633 git_SHA_CTX c;
1634 char hdr[32];
1635 int hdrlen;
1636
1637 if (map) {
1638 hash_sha1_file(map, size, type, real_sha1);
1639 return hashcmp(sha1, real_sha1) ? -1 : 0;
1640 }
1641
1642 st = open_istream(sha1, &obj_type, &size, NULL);
1643 if (!st)
1644 return -1;
1645
1646 /* Generate the header */
1647 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
1648
1649 /* Sha1.. */
1650 git_SHA1_Init(&c);
1651 git_SHA1_Update(&c, hdr, hdrlen);
1652 for (;;) {
1653 char buf[1024 * 16];
1654 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1655
1656 if (readlen < 0) {
1657 close_istream(st);
1658 return -1;
1659 }
1660 if (!readlen)
1661 break;
1662 git_SHA1_Update(&c, buf, readlen);
1663 }
1664 git_SHA1_Final(real_sha1, &c);
1665 close_istream(st);
1666 return hashcmp(sha1, real_sha1) ? -1 : 0;
1667 }
1668
1669 int git_open_cloexec(const char *name, int flags)
1670 {
1671 int fd;
1672 static int o_cloexec = O_CLOEXEC;
1673
1674 fd = open(name, flags | o_cloexec);
1675 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1676 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1677 o_cloexec &= ~O_CLOEXEC;
1678 fd = open(name, flags | o_cloexec);
1679 }
1680
1681 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1682 {
1683 static int fd_cloexec = FD_CLOEXEC;
1684
1685 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1686 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1687 int flags = fcntl(fd, F_GETFD);
1688 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1689 fd_cloexec = 0;
1690 }
1691 }
1692 #endif
1693 return fd;
1694 }
1695
1696 /*
1697 * Find "sha1" as a loose object in the local repository or in an alternate.
1698 * Returns 0 on success, negative on failure.
1699 *
1700 * The "path" out-parameter will give the path of the object we found (if any).
1701 * Note that it may point to static storage and is only valid until another
1702 * call to sha1_file_name(), etc.
1703 */
1704 static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
1705 const char **path)
1706 {
1707 struct alternate_object_database *alt;
1708
1709 *path = sha1_file_name(sha1);
1710 if (!lstat(*path, st))
1711 return 0;
1712
1713 prepare_alt_odb();
1714 errno = ENOENT;
1715 for (alt = alt_odb_list; alt; alt = alt->next) {
1716 *path = alt_sha1_path(alt, sha1);
1717 if (!lstat(*path, st))
1718 return 0;
1719 }
1720
1721 return -1;
1722 }
1723
1724 /*
1725 * Like stat_sha1_file(), but actually open the object and return the
1726 * descriptor. See the caveats on the "path" parameter above.
1727 */
1728 static int open_sha1_file(const unsigned char *sha1, const char **path)
1729 {
1730 int fd;
1731 struct alternate_object_database *alt;
1732 int most_interesting_errno;
1733
1734 *path = sha1_file_name(sha1);
1735 fd = git_open(*path);
1736 if (fd >= 0)
1737 return fd;
1738 most_interesting_errno = errno;
1739
1740 prepare_alt_odb();
1741 for (alt = alt_odb_list; alt; alt = alt->next) {
1742 *path = alt_sha1_path(alt, sha1);
1743 fd = git_open(*path);
1744 if (fd >= 0)
1745 return fd;
1746 if (most_interesting_errno == ENOENT)
1747 most_interesting_errno = errno;
1748 }
1749 errno = most_interesting_errno;
1750 return -1;
1751 }
1752
1753 /*
1754 * Map the loose object at "path" if it is not NULL, or the path found by
1755 * searching for a loose object named "sha1".
1756 */
1757 static void *map_sha1_file_1(const char *path,
1758 const unsigned char *sha1,
1759 unsigned long *size)
1760 {
1761 void *map;
1762 int fd;
1763
1764 if (path)
1765 fd = git_open(path);
1766 else
1767 fd = open_sha1_file(sha1, &path);
1768 map = NULL;
1769 if (fd >= 0) {
1770 struct stat st;
1771
1772 if (!fstat(fd, &st)) {
1773 *size = xsize_t(st.st_size);
1774 if (!*size) {
1775 /* mmap() is forbidden on empty files */
1776 error("object file %s is empty", path);
1777 return NULL;
1778 }
1779 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1780 }
1781 close(fd);
1782 }
1783 return map;
1784 }
1785
1786 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1787 {
1788 return map_sha1_file_1(NULL, sha1, size);
1789 }
1790
1791 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1792 unsigned long len, enum object_type *type, unsigned long *sizep)
1793 {
1794 unsigned shift;
1795 unsigned long size, c;
1796 unsigned long used = 0;
1797
1798 c = buf[used++];
1799 *type = (c >> 4) & 7;
1800 size = c & 15;
1801 shift = 4;
1802 while (c & 0x80) {
1803 if (len <= used || bitsizeof(long) <= shift) {
1804 error("bad object header");
1805 size = used = 0;
1806 break;
1807 }
1808 c = buf[used++];
1809 size += (c & 0x7f) << shift;
1810 shift += 7;
1811 }
1812 *sizep = size;
1813 return used;
1814 }
1815
1816 static int unpack_sha1_short_header(git_zstream *stream,
1817 unsigned char *map, unsigned long mapsize,
1818 void *buffer, unsigned long bufsiz)
1819 {
1820 /* Get the data stream */
1821 memset(stream, 0, sizeof(*stream));
1822 stream->next_in = map;
1823 stream->avail_in = mapsize;
1824 stream->next_out = buffer;
1825 stream->avail_out = bufsiz;
1826
1827 git_inflate_init(stream);
1828 return git_inflate(stream, 0);
1829 }
1830
1831 int unpack_sha1_header(git_zstream *stream,
1832 unsigned char *map, unsigned long mapsize,
1833 void *buffer, unsigned long bufsiz)
1834 {
1835 int status = unpack_sha1_short_header(stream, map, mapsize,
1836 buffer, bufsiz);
1837
1838 if (status < Z_OK)
1839 return status;
1840
1841 /* Make sure we have the terminating NUL */
1842 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1843 return -1;
1844 return 0;
1845 }
1846
1847 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1848 unsigned long mapsize, void *buffer,
1849 unsigned long bufsiz, struct strbuf *header)
1850 {
1851 int status;
1852
1853 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1854 if (status < Z_OK)
1855 return -1;
1856
1857 /*
1858 * Check if entire header is unpacked in the first iteration.
1859 */
1860 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1861 return 0;
1862
1863 /*
1864 * buffer[0..bufsiz] was not large enough. Copy the partial
1865 * result out to header, and then append the result of further
1866 * reading the stream.
1867 */
1868 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1869 stream->next_out = buffer;
1870 stream->avail_out = bufsiz;
1871
1872 do {
1873 status = git_inflate(stream, 0);
1874 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1875 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1876 return 0;
1877 stream->next_out = buffer;
1878 stream->avail_out = bufsiz;
1879 } while (status != Z_STREAM_END);
1880 return -1;
1881 }
1882
1883 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1884 {
1885 int bytes = strlen(buffer) + 1;
1886 unsigned char *buf = xmallocz(size);
1887 unsigned long n;
1888 int status = Z_OK;
1889
1890 n = stream->total_out - bytes;
1891 if (n > size)
1892 n = size;
1893 memcpy(buf, (char *) buffer + bytes, n);
1894 bytes = n;
1895 if (bytes <= size) {
1896 /*
1897 * The above condition must be (bytes <= size), not
1898 * (bytes < size). In other words, even though we
1899 * expect no more output and set avail_out to zero,
1900 * the input zlib stream may have bytes that express
1901 * "this concludes the stream", and we *do* want to
1902 * eat that input.
1903 *
1904 * Otherwise we would not be able to test that we
1905 * consumed all the input to reach the expected size;
1906 * we also want to check that zlib tells us that all
1907 * went well with status == Z_STREAM_END at the end.
1908 */
1909 stream->next_out = buf + bytes;
1910 stream->avail_out = size - bytes;
1911 while (status == Z_OK)
1912 status = git_inflate(stream, Z_FINISH);
1913 }
1914 if (status == Z_STREAM_END && !stream->avail_in) {
1915 git_inflate_end(stream);
1916 return buf;
1917 }
1918
1919 if (status < 0)
1920 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1921 else if (stream->avail_in)
1922 error("garbage at end of loose object '%s'",
1923 sha1_to_hex(sha1));
1924 free(buf);
1925 return NULL;
1926 }
1927
1928 /*
1929 * We used to just use "sscanf()", but that's actually way
1930 * too permissive for what we want to check. So do an anal
1931 * object header parse by hand.
1932 */
1933 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1934 unsigned int flags)
1935 {
1936 const char *type_buf = hdr;
1937 unsigned long size;
1938 int type, type_len = 0;
1939
1940 /*
1941 * The type can be of any size but is followed by
1942 * a space.
1943 */
1944 for (;;) {
1945 char c = *hdr++;
1946 if (!c)
1947 return -1;
1948 if (c == ' ')
1949 break;
1950 type_len++;
1951 }
1952
1953 type = type_from_string_gently(type_buf, type_len, 1);
1954 if (oi->typename)
1955 strbuf_add(oi->typename, type_buf, type_len);
1956 /*
1957 * Set type to 0 if its an unknown object and
1958 * we're obtaining the type using '--allow-unknown-type'
1959 * option.
1960 */
1961 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1962 type = 0;
1963 else if (type < 0)
1964 die("invalid object type");
1965 if (oi->typep)
1966 *oi->typep = type;
1967
1968 /*
1969 * The length must follow immediately, and be in canonical
1970 * decimal format (ie "010" is not valid).
1971 */
1972 size = *hdr++ - '0';
1973 if (size > 9)
1974 return -1;
1975 if (size) {
1976 for (;;) {
1977 unsigned long c = *hdr - '0';
1978 if (c > 9)
1979 break;
1980 hdr++;
1981 size = size * 10 + c;
1982 }
1983 }
1984
1985 if (oi->sizep)
1986 *oi->sizep = size;
1987
1988 /*
1989 * The length must be followed by a zero byte
1990 */
1991 return *hdr ? -1 : type;
1992 }
1993
1994 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1995 {
1996 struct object_info oi = OBJECT_INFO_INIT;
1997
1998 oi.sizep = sizep;
1999 return parse_sha1_header_extended(hdr, &oi, 0);
2000 }
2001
2002 unsigned long get_size_from_delta(struct packed_git *p,
2003 struct pack_window **w_curs,
2004 off_t curpos)
2005 {
2006 const unsigned char *data;
2007 unsigned char delta_head[20], *in;
2008 git_zstream stream;
2009 int st;
2010
2011 memset(&stream, 0, sizeof(stream));
2012 stream.next_out = delta_head;
2013 stream.avail_out = sizeof(delta_head);
2014
2015 git_inflate_init(&stream);
2016 do {
2017 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2018 stream.next_in = in;
2019 st = git_inflate(&stream, Z_FINISH);
2020 curpos += stream.next_in - in;
2021 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
2022 stream.total_out < sizeof(delta_head));
2023 git_inflate_end(&stream);
2024 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
2025 error("delta data unpack-initial failed");
2026 return 0;
2027 }
2028
2029 /* Examine the initial part of the delta to figure out
2030 * the result size.
2031 */
2032 data = delta_head;
2033
2034 /* ignore base size */
2035 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
2036
2037 /* Read the result size */
2038 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
2039 }
2040
2041 static off_t get_delta_base(struct packed_git *p,
2042 struct pack_window **w_curs,
2043 off_t *curpos,
2044 enum object_type type,
2045 off_t delta_obj_offset)
2046 {
2047 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
2048 off_t base_offset;
2049
2050 /* use_pack() assured us we have [base_info, base_info + 20)
2051 * as a range that we can look at without walking off the
2052 * end of the mapped window. Its actually the hash size
2053 * that is assured. An OFS_DELTA longer than the hash size
2054 * is stupid, as then a REF_DELTA would be smaller to store.
2055 */
2056 if (type == OBJ_OFS_DELTA) {
2057 unsigned used = 0;
2058 unsigned char c = base_info[used++];
2059 base_offset = c & 127;
2060 while (c & 128) {
2061 base_offset += 1;
2062 if (!base_offset || MSB(base_offset, 7))
2063 return 0; /* overflow */
2064 c = base_info[used++];
2065 base_offset = (base_offset << 7) + (c & 127);
2066 }
2067 base_offset = delta_obj_offset - base_offset;
2068 if (base_offset <= 0 || base_offset >= delta_obj_offset)
2069 return 0; /* out of bound */
2070 *curpos += used;
2071 } else if (type == OBJ_REF_DELTA) {
2072 /* The base entry _must_ be in the same pack */
2073 base_offset = find_pack_entry_one(base_info, p);
2074 *curpos += 20;
2075 } else
2076 die("I am totally screwed");
2077 return base_offset;
2078 }
2079
2080 /*
2081 * Like get_delta_base above, but we return the sha1 instead of the pack
2082 * offset. This means it is cheaper for REF deltas (we do not have to do
2083 * the final object lookup), but more expensive for OFS deltas (we
2084 * have to load the revidx to convert the offset back into a sha1).
2085 */
2086 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
2087 struct pack_window **w_curs,
2088 off_t curpos,
2089 enum object_type type,
2090 off_t delta_obj_offset)
2091 {
2092 if (type == OBJ_REF_DELTA) {
2093 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
2094 return base;
2095 } else if (type == OBJ_OFS_DELTA) {
2096 struct revindex_entry *revidx;
2097 off_t base_offset = get_delta_base(p, w_curs, &curpos,
2098 type, delta_obj_offset);
2099
2100 if (!base_offset)
2101 return NULL;
2102
2103 revidx = find_pack_revindex(p, base_offset);
2104 if (!revidx)
2105 return NULL;
2106
2107 return nth_packed_object_sha1(p, revidx->nr);
2108 } else
2109 return NULL;
2110 }
2111
2112 int unpack_object_header(struct packed_git *p,
2113 struct pack_window **w_curs,
2114 off_t *curpos,
2115 unsigned long *sizep)
2116 {
2117 unsigned char *base;
2118 unsigned long left;
2119 unsigned long used;
2120 enum object_type type;
2121
2122 /* use_pack() assures us we have [base, base + 20) available
2123 * as a range that we can look at. (Its actually the hash
2124 * size that is assured.) With our object header encoding
2125 * the maximum deflated object size is 2^137, which is just
2126 * insane, so we know won't exceed what we have been given.
2127 */
2128 base = use_pack(p, w_curs, *curpos, &left);
2129 used = unpack_object_header_buffer(base, left, &type, sizep);
2130 if (!used) {
2131 type = OBJ_BAD;
2132 } else
2133 *curpos += used;
2134
2135 return type;
2136 }
2137
2138 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
2139 {
2140 int type;
2141 struct revindex_entry *revidx;
2142 const unsigned char *sha1;
2143 revidx = find_pack_revindex(p, obj_offset);
2144 if (!revidx)
2145 return OBJ_BAD;
2146 sha1 = nth_packed_object_sha1(p, revidx->nr);
2147 mark_bad_packed_object(p, sha1);
2148 type = sha1_object_info(sha1, NULL);
2149 if (type <= OBJ_NONE)
2150 return OBJ_BAD;
2151 return type;
2152 }
2153
2154 #define POI_STACK_PREALLOC 64
2155
2156 static enum object_type packed_to_object_type(struct packed_git *p,
2157 off_t obj_offset,
2158 enum object_type type,
2159 struct pack_window **w_curs,
2160 off_t curpos)
2161 {
2162 off_t small_poi_stack[POI_STACK_PREALLOC];
2163 off_t *poi_stack = small_poi_stack;
2164 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
2165
2166 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2167 off_t base_offset;
2168 unsigned long size;
2169 /* Push the object we're going to leave behind */
2170 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
2171 poi_stack_alloc = alloc_nr(poi_stack_nr);
2172 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
2173 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
2174 } else {
2175 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
2176 }
2177 poi_stack[poi_stack_nr++] = obj_offset;
2178 /* If parsing the base offset fails, just unwind */
2179 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
2180 if (!base_offset)
2181 goto unwind;
2182 curpos = obj_offset = base_offset;
2183 type = unpack_object_header(p, w_curs, &curpos, &size);
2184 if (type <= OBJ_NONE) {
2185 /* If getting the base itself fails, we first
2186 * retry the base, otherwise unwind */
2187 type = retry_bad_packed_offset(p, base_offset);
2188 if (type > OBJ_NONE)
2189 goto out;
2190 goto unwind;
2191 }
2192 }
2193
2194 switch (type) {
2195 case OBJ_BAD:
2196 case OBJ_COMMIT:
2197 case OBJ_TREE:
2198 case OBJ_BLOB:
2199 case OBJ_TAG:
2200 break;
2201 default:
2202 error("unknown object type %i at offset %"PRIuMAX" in %s",
2203 type, (uintmax_t)obj_offset, p->pack_name);
2204 type = OBJ_BAD;
2205 }
2206
2207 out:
2208 if (poi_stack != small_poi_stack)
2209 free(poi_stack);
2210 return type;
2211
2212 unwind:
2213 while (poi_stack_nr) {
2214 obj_offset = poi_stack[--poi_stack_nr];
2215 type = retry_bad_packed_offset(p, obj_offset);
2216 if (type > OBJ_NONE)
2217 goto out;
2218 }
2219 type = OBJ_BAD;
2220 goto out;
2221 }
2222
2223 static struct hashmap delta_base_cache;
2224 static size_t delta_base_cached;
2225
2226 static LIST_HEAD(delta_base_cache_lru);
2227
2228 struct delta_base_cache_key {
2229 struct packed_git *p;
2230 off_t base_offset;
2231 };
2232
2233 struct delta_base_cache_entry {
2234 struct hashmap hash;
2235 struct delta_base_cache_key key;
2236 struct list_head lru;
2237 void *data;
2238 unsigned long size;
2239 enum object_type type;
2240 };
2241
2242 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
2243 {
2244 unsigned int hash;
2245
2246 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
2247 hash += (hash >> 8) + (hash >> 16);
2248 return hash;
2249 }
2250
2251 static struct delta_base_cache_entry *
2252 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2253 {
2254 struct hashmap_entry entry;
2255 struct delta_base_cache_key key;
2256
2257 if (!delta_base_cache.cmpfn)
2258 return NULL;
2259
2260 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
2261 key.p = p;
2262 key.base_offset = base_offset;
2263 return hashmap_get(&delta_base_cache, &entry, &key);
2264 }
2265
2266 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
2267 const struct delta_base_cache_key *b)
2268 {
2269 return a->p == b->p && a->base_offset == b->base_offset;
2270 }
2271
2272 static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
2273 const void *va, const void *vb,
2274 const void *vkey)
2275 {
2276 const struct delta_base_cache_entry *a = va, *b = vb;
2277 const struct delta_base_cache_key *key = vkey;
2278 if (key)
2279 return !delta_base_cache_key_eq(&a->key, key);
2280 else
2281 return !delta_base_cache_key_eq(&a->key, &b->key);
2282 }
2283
2284 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2285 {
2286 return !!get_delta_base_cache_entry(p, base_offset);
2287 }
2288
2289 /*
2290 * Remove the entry from the cache, but do _not_ free the associated
2291 * entry data. The caller takes ownership of the "data" buffer, and
2292 * should copy out any fields it wants before detaching.
2293 */
2294 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2295 {
2296 hashmap_remove(&delta_base_cache, ent, &ent->key);
2297 list_del(&ent->lru);
2298 delta_base_cached -= ent->size;
2299 free(ent);
2300 }
2301
2302 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2303 unsigned long *base_size, enum object_type *type)
2304 {
2305 struct delta_base_cache_entry *ent;
2306
2307 ent = get_delta_base_cache_entry(p, base_offset);
2308 if (!ent)
2309 return unpack_entry(p, base_offset, type, base_size);
2310
2311 if (type)
2312 *type = ent->type;
2313 if (base_size)
2314 *base_size = ent->size;
2315 return xmemdupz(ent->data, ent->size);
2316 }
2317
2318 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2319 {
2320 free(ent->data);
2321 detach_delta_base_cache_entry(ent);
2322 }
2323
2324 void clear_delta_base_cache(void)
2325 {
2326 struct list_head *lru, *tmp;
2327 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
2328 struct delta_base_cache_entry *entry =
2329 list_entry(lru, struct delta_base_cache_entry, lru);
2330 release_delta_base_cache(entry);
2331 }
2332 }
2333
2334 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2335 void *base, unsigned long base_size, enum object_type type)
2336 {
2337 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
2338 struct list_head *lru, *tmp;
2339
2340 delta_base_cached += base_size;
2341
2342 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
2343 struct delta_base_cache_entry *f =
2344 list_entry(lru, struct delta_base_cache_entry, lru);
2345 if (delta_base_cached <= delta_base_cache_limit)
2346 break;
2347 release_delta_base_cache(f);
2348 }
2349
2350 ent->key.p = p;
2351 ent->key.base_offset = base_offset;
2352 ent->type = type;
2353 ent->data = base;
2354 ent->size = base_size;
2355 list_add_tail(&ent->lru, &delta_base_cache_lru);
2356
2357 if (!delta_base_cache.cmpfn)
2358 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
2359 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
2360 hashmap_add(&delta_base_cache, ent);
2361 }
2362
2363 int packed_object_info(struct packed_git *p, off_t obj_offset,
2364 struct object_info *oi)
2365 {
2366 struct pack_window *w_curs = NULL;
2367 unsigned long size;
2368 off_t curpos = obj_offset;
2369 enum object_type type;
2370
2371 /*
2372 * We always get the representation type, but only convert it to
2373 * a "real" type later if the caller is interested.
2374 */
2375 if (oi->contentp) {
2376 *oi->contentp = cache_or_unpack_entry(p, obj_offset, oi->sizep,
2377 &type);
2378 if (!*oi->contentp)
2379 type = OBJ_BAD;
2380 } else {
2381 type = unpack_object_header(p, &w_curs, &curpos, &size);
2382 }
2383
2384 if (!oi->contentp && oi->sizep) {
2385 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2386 off_t tmp_pos = curpos;
2387 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
2388 type, obj_offset);
2389 if (!base_offset) {
2390 type = OBJ_BAD;
2391 goto out;
2392 }
2393 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2394 if (*oi->sizep == 0) {
2395 type = OBJ_BAD;
2396 goto out;
2397 }
2398 } else {
2399 *oi->sizep = size;
2400 }
2401 }
2402
2403 if (oi->disk_sizep) {
2404 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2405 *oi->disk_sizep = revidx[1].offset - obj_offset;
2406 }
2407
2408 if (oi->typep || oi->typename) {
2409 enum object_type ptot;
2410 ptot = packed_to_object_type(p, obj_offset, type, &w_curs,
2411 curpos);
2412 if (oi->typep)
2413 *oi->typep = ptot;
2414 if (oi->typename) {
2415 const char *tn = typename(ptot);
2416 if (tn)
2417 strbuf_addstr(oi->typename, tn);
2418 }
2419 if (ptot < 0) {
2420 type = OBJ_BAD;
2421 goto out;
2422 }
2423 }
2424
2425 if (oi->delta_base_sha1) {
2426 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2427 const unsigned char *base;
2428
2429 base = get_delta_base_sha1(p, &w_curs, curpos,
2430 type, obj_offset);
2431 if (!base) {
2432 type = OBJ_BAD;
2433 goto out;
2434 }
2435
2436 hashcpy(oi->delta_base_sha1, base);
2437 } else
2438 hashclr(oi->delta_base_sha1);
2439 }
2440
2441 out:
2442 unuse_pack(&w_curs);
2443 return type;
2444 }
2445
2446 static void *unpack_compressed_entry(struct packed_git *p,
2447 struct pack_window **w_curs,
2448 off_t curpos,
2449 unsigned long size)
2450 {
2451 int st;
2452 git_zstream stream;
2453 unsigned char *buffer, *in;
2454
2455 buffer = xmallocz_gently(size);
2456 if (!buffer)
2457 return NULL;
2458 memset(&stream, 0, sizeof(stream));
2459 stream.next_out = buffer;
2460 stream.avail_out = size + 1;
2461
2462 git_inflate_init(&stream);
2463 do {
2464 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2465 stream.next_in = in;
2466 st = git_inflate(&stream, Z_FINISH);
2467 if (!stream.avail_out)
2468 break; /* the payload is larger than it should be */
2469 curpos += stream.next_in - in;
2470 } while (st == Z_OK || st == Z_BUF_ERROR);
2471 git_inflate_end(&stream);
2472 if ((st != Z_STREAM_END) || stream.total_out != size) {
2473 free(buffer);
2474 return NULL;
2475 }
2476
2477 return buffer;
2478 }
2479
2480 static void *read_object(const unsigned char *sha1, enum object_type *type,
2481 unsigned long *size);
2482
2483 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2484 {
2485 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2486 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2487 p->pack_name, (uintmax_t)obj_offset);
2488 }
2489
2490 int do_check_packed_object_crc;
2491
2492 #define UNPACK_ENTRY_STACK_PREALLOC 64
2493 struct unpack_entry_stack_ent {
2494 off_t obj_offset;
2495 off_t curpos;
2496 unsigned long size;
2497 };
2498
2499 void *unpack_entry(struct packed_git *p, off_t obj_offset,
2500 enum object_type *final_type, unsigned long *final_size)
2501 {
2502 struct pack_window *w_curs = NULL;
2503 off_t curpos = obj_offset;
2504 void *data = NULL;
2505 unsigned long size;
2506 enum object_type type;
2507 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2508 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2509 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2510 int base_from_cache = 0;
2511
2512 write_pack_access_log(p, obj_offset);
2513
2514 /* PHASE 1: drill down to the innermost base object */
2515 for (;;) {
2516 off_t base_offset;
2517 int i;
2518 struct delta_base_cache_entry *ent;
2519
2520 ent = get_delta_base_cache_entry(p, curpos);
2521 if (ent) {
2522 type = ent->type;
2523 data = ent->data;
2524 size = ent->size;
2525 detach_delta_base_cache_entry(ent);
2526 base_from_cache = 1;
2527 break;
2528 }
2529
2530 if (do_check_packed_object_crc && p->index_version > 1) {
2531 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2532 off_t len = revidx[1].offset - obj_offset;
2533 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2534 const unsigned char *sha1 =
2535 nth_packed_object_sha1(p, revidx->nr);
2536 error("bad packed object CRC for %s",
2537 sha1_to_hex(sha1));
2538 mark_bad_packed_object(p, sha1);
2539 data = NULL;
2540 goto out;
2541 }
2542 }
2543
2544 type = unpack_object_header(p, &w_curs, &curpos, &size);
2545 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2546 break;
2547
2548 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2549 if (!base_offset) {
2550 error("failed to validate delta base reference "
2551 "at offset %"PRIuMAX" from %s",
2552 (uintmax_t)curpos, p->pack_name);
2553 /* bail to phase 2, in hopes of recovery */
2554 data = NULL;
2555 break;
2556 }
2557
2558 /* push object, proceed to base */
2559 if (delta_stack_nr >= delta_stack_alloc
2560 && delta_stack == small_delta_stack) {
2561 delta_stack_alloc = alloc_nr(delta_stack_nr);
2562 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
2563 memcpy(delta_stack, small_delta_stack,
2564 sizeof(*delta_stack)*delta_stack_nr);
2565 } else {
2566 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2567 }
2568 i = delta_stack_nr++;
2569 delta_stack[i].obj_offset = obj_offset;
2570 delta_stack[i].curpos = curpos;
2571 delta_stack[i].size = size;
2572
2573 curpos = obj_offset = base_offset;
2574 }
2575
2576 /* PHASE 2: handle the base */
2577 switch (type) {
2578 case OBJ_OFS_DELTA:
2579 case OBJ_REF_DELTA:
2580 if (data)
2581 die("BUG: unpack_entry: left loop at a valid delta");
2582 break;
2583 case OBJ_COMMIT:
2584 case OBJ_TREE:
2585 case OBJ_BLOB:
2586 case OBJ_TAG:
2587 if (!base_from_cache)
2588 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2589 break;
2590 default:
2591 data = NULL;
2592 error("unknown object type %i at offset %"PRIuMAX" in %s",
2593 type, (uintmax_t)obj_offset, p->pack_name);
2594 }
2595
2596 /* PHASE 3: apply deltas in order */
2597
2598 /* invariants:
2599 * 'data' holds the base data, or NULL if there was corruption
2600 */
2601 while (delta_stack_nr) {
2602 void *delta_data;
2603 void *base = data;
2604 void *external_base = NULL;
2605 unsigned long delta_size, base_size = size;
2606 int i;
2607
2608 data = NULL;
2609
2610 if (base)
2611 add_delta_base_cache(p, obj_offset, base, base_size, type);
2612
2613 if (!base) {
2614 /*
2615 * We're probably in deep shit, but let's try to fetch
2616 * the required base anyway from another pack or loose.
2617 * This is costly but should happen only in the presence
2618 * of a corrupted pack, and is better than failing outright.
2619 */
2620 struct revindex_entry *revidx;
2621 const unsigned char *base_sha1;
2622 revidx = find_pack_revindex(p, obj_offset);
2623 if (revidx) {
2624 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2625 error("failed to read delta base object %s"
2626 " at offset %"PRIuMAX" from %s",
2627 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2628 p->pack_name);
2629 mark_bad_packed_object(p, base_sha1);
2630 base = read_object(base_sha1, &type, &base_size);
2631 external_base = base;
2632 }
2633 }
2634
2635 i = --delta_stack_nr;
2636 obj_offset = delta_stack[i].obj_offset;
2637 curpos = delta_stack[i].curpos;
2638 delta_size = delta_stack[i].size;
2639
2640 if (!base)
2641 continue;
2642
2643 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2644
2645 if (!delta_data) {
2646 error("failed to unpack compressed delta "
2647 "at offset %"PRIuMAX" from %s",
2648 (uintmax_t)curpos, p->pack_name);
2649 data = NULL;
2650 free(external_base);
2651 continue;
2652 }
2653
2654 data = patch_delta(base, base_size,
2655 delta_data, delta_size,
2656 &size);
2657
2658 /*
2659 * We could not apply the delta; warn the user, but keep going.
2660 * Our failure will be noticed either in the next iteration of
2661 * the loop, or if this is the final delta, in the caller when
2662 * we return NULL. Those code paths will take care of making
2663 * a more explicit warning and retrying with another copy of
2664 * the object.
2665 */
2666 if (!data)
2667 error("failed to apply delta");
2668
2669 free(delta_data);
2670 free(external_base);
2671 }
2672
2673 if (final_type)
2674 *final_type = type;
2675 if (final_size)
2676 *final_size = size;
2677
2678 out:
2679 unuse_pack(&w_curs);
2680
2681 if (delta_stack != small_delta_stack)
2682 free(delta_stack);
2683
2684 return data;
2685 }
2686
2687 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2688 uint32_t n)
2689 {
2690 const unsigned char *index = p->index_data;
2691 if (!index) {
2692 if (open_pack_index(p))
2693 return NULL;
2694 index = p->index_data;
2695 }
2696 if (n >= p->num_objects)
2697 return NULL;
2698 index += 4 * 256;
2699 if (p->index_version == 1) {
2700 return index + 24 * n + 4;
2701 } else {
2702 index += 8;
2703 return index + 20 * n;
2704 }
2705 }
2706
2707 const struct object_id *nth_packed_object_oid(struct object_id *oid,
2708 struct packed_git *p,
2709 uint32_t n)
2710 {
2711 const unsigned char *hash = nth_packed_object_sha1(p, n);
2712 if (!hash)
2713 return NULL;
2714 hashcpy(oid->hash, hash);
2715 return oid;
2716 }
2717
2718 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
2719 {
2720 const unsigned char *ptr = vptr;
2721 const unsigned char *start = p->index_data;
2722 const unsigned char *end = start + p->index_size;
2723 if (ptr < start)
2724 die(_("offset before start of pack index for %s (corrupt index?)"),
2725 p->pack_name);
2726 /* No need to check for underflow; .idx files must be at least 8 bytes */
2727 if (ptr >= end - 8)
2728 die(_("offset beyond end of pack index for %s (truncated index?)"),
2729 p->pack_name);
2730 }
2731
2732 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2733 {
2734 const unsigned char *index = p->index_data;
2735 index += 4 * 256;
2736 if (p->index_version == 1) {
2737 return ntohl(*((uint32_t *)(index + 24 * n)));
2738 } else {
2739 uint32_t off;
2740 index += 8 + p->num_objects * (20 + 4);
2741 off = ntohl(*((uint32_t *)(index + 4 * n)));
2742 if (!(off & 0x80000000))
2743 return off;
2744 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2745 check_pack_index_ptr(p, index);
2746 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2747 ntohl(*((uint32_t *)(index + 4)));
2748 }
2749 }
2750
2751 off_t find_pack_entry_one(const unsigned char *sha1,
2752 struct packed_git *p)
2753 {
2754 const uint32_t *level1_ofs = p->index_data;
2755 const unsigned char *index = p->index_data;
2756 unsigned hi, lo, stride;
2757 static int use_lookup = -1;
2758 static int debug_lookup = -1;
2759
2760 if (debug_lookup < 0)
2761 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2762
2763 if (!index) {
2764 if (open_pack_index(p))
2765 return 0;
2766 level1_ofs = p->index_data;
2767 index = p->index_data;
2768 }
2769 if (p->index_version > 1) {
2770 level1_ofs += 2;
2771 index += 8;
2772 }
2773 index += 4 * 256;
2774 hi = ntohl(level1_ofs[*sha1]);
2775 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2776 if (p->index_version > 1) {
2777 stride = 20;
2778 } else {
2779 stride = 24;
2780 index += 4;
2781 }
2782
2783 if (debug_lookup)
2784 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2785 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2786
2787 if (use_lookup < 0)
2788 use_lookup = !!getenv("GIT_USE_LOOKUP");
2789 if (use_lookup) {
2790 int pos = sha1_entry_pos(index, stride, 0,
2791 lo, hi, p->num_objects, sha1);
2792 if (pos < 0)
2793 return 0;
2794 return nth_packed_object_offset(p, pos);
2795 }
2796
2797 while (lo < hi) {
2798 unsigned mi = (lo + hi) / 2;
2799 int cmp = hashcmp(index + mi * stride, sha1);
2800
2801 if (debug_lookup)
2802 printf("lo %u hi %u rg %u mi %u\n",
2803 lo, hi, hi - lo, mi);
2804 if (!cmp)
2805 return nth_packed_object_offset(p, mi);
2806 if (cmp > 0)
2807 hi = mi;
2808 else
2809 lo = mi+1;
2810 }
2811 return 0;
2812 }
2813
2814 int is_pack_valid(struct packed_git *p)
2815 {
2816 /* An already open pack is known to be valid. */
2817 if (p->pack_fd != -1)
2818 return 1;
2819
2820 /* If the pack has one window completely covering the
2821 * file size, the pack is known to be valid even if
2822 * the descriptor is not currently open.
2823 */
2824 if (p->windows) {
2825 struct pack_window *w = p->windows;
2826
2827 if (!w->offset && w->len == p->pack_size)
2828 return 1;
2829 }
2830
2831 /* Force the pack to open to prove its valid. */
2832 return !open_packed_git(p);
2833 }
2834
2835 static int fill_pack_entry(const unsigned char *sha1,
2836 struct pack_entry *e,
2837 struct packed_git *p)
2838 {
2839 off_t offset;
2840
2841 if (p->num_bad_objects) {
2842 unsigned i;
2843 for (i = 0; i < p->num_bad_objects; i++)
2844 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2845 return 0;
2846 }
2847
2848 offset = find_pack_entry_one(sha1, p);
2849 if (!offset)
2850 return 0;
2851
2852 /*
2853 * We are about to tell the caller where they can locate the
2854 * requested object. We better make sure the packfile is
2855 * still here and can be accessed before supplying that
2856 * answer, as it may have been deleted since the index was
2857 * loaded!
2858 */
2859 if (!is_pack_valid(p))
2860 return 0;
2861 e->offset = offset;
2862 e->p = p;
2863 hashcpy(e->sha1, sha1);
2864 return 1;
2865 }
2866
2867 /*
2868 * Iff a pack file contains the object named by sha1, return true and
2869 * store its location to e.
2870 */
2871 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2872 {
2873 struct mru_entry *p;
2874
2875 prepare_packed_git();
2876 if (!packed_git)
2877 return 0;
2878
2879 for (p = packed_git_mru->head; p; p = p->next) {
2880 if (fill_pack_entry(sha1, e, p->item)) {
2881 mru_mark(packed_git_mru, p);
2882 return 1;
2883 }
2884 }
2885 return 0;
2886 }
2887
2888 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2889 struct packed_git *packs)
2890 {
2891 struct packed_git *p;
2892
2893 for (p = packs; p; p = p->next) {
2894 if (find_pack_entry_one(sha1, p))
2895 return p;
2896 }
2897 return NULL;
2898
2899 }
2900
2901 static int sha1_loose_object_info(const unsigned char *sha1,
2902 struct object_info *oi,
2903 int flags)
2904 {
2905 int status = 0;
2906 unsigned long mapsize;
2907 void *map;
2908 git_zstream stream;
2909 char hdr[32];
2910 struct strbuf hdrbuf = STRBUF_INIT;
2911 unsigned long size_scratch;
2912
2913 if (oi->delta_base_sha1)
2914 hashclr(oi->delta_base_sha1);
2915
2916 /*
2917 * If we don't care about type or size, then we don't
2918 * need to look inside the object at all. Note that we
2919 * do not optimize out the stat call, even if the
2920 * caller doesn't care about the disk-size, since our
2921 * return value implicitly indicates whether the
2922 * object even exists.
2923 */
2924 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
2925 const char *path;
2926 struct stat st;
2927 if (stat_sha1_file(sha1, &st, &path) < 0)
2928 return -1;
2929 if (oi->disk_sizep)
2930 *oi->disk_sizep = st.st_size;
2931 return 0;
2932 }
2933
2934 map = map_sha1_file(sha1, &mapsize);
2935 if (!map)
2936 return -1;
2937
2938 if (!oi->sizep)
2939 oi->sizep = &size_scratch;
2940
2941 if (oi->disk_sizep)
2942 *oi->disk_sizep = mapsize;
2943 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
2944 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2945 status = error("unable to unpack %s header with --allow-unknown-type",
2946 sha1_to_hex(sha1));
2947 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2948 status = error("unable to unpack %s header",
2949 sha1_to_hex(sha1));
2950 if (status < 0)
2951 ; /* Do nothing */
2952 else if (hdrbuf.len) {
2953 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2954 status = error("unable to parse %s header with --allow-unknown-type",
2955 sha1_to_hex(sha1));
2956 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
2957 status = error("unable to parse %s header", sha1_to_hex(sha1));
2958
2959 if (status >= 0 && oi->contentp)
2960 *oi->contentp = unpack_sha1_rest(&stream, hdr,
2961 *oi->sizep, sha1);
2962 else
2963 git_inflate_end(&stream);
2964
2965 munmap(map, mapsize);
2966 if (status && oi->typep)
2967 *oi->typep = status;
2968 if (oi->sizep == &size_scratch)
2969 oi->sizep = NULL;
2970 strbuf_release(&hdrbuf);
2971 return (status < 0) ? status : 0;
2972 }
2973
2974 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2975 {
2976 static struct object_info blank_oi = OBJECT_INFO_INIT;
2977 struct pack_entry e;
2978 int rtype;
2979 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
2980 lookup_replace_object(sha1) :
2981 sha1;
2982
2983 if (!oi)
2984 oi = &blank_oi;
2985
2986 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
2987 struct cached_object *co = find_cached_object(real);
2988 if (co) {
2989 if (oi->typep)
2990 *(oi->typep) = co->type;
2991 if (oi->sizep)
2992 *(oi->sizep) = co->size;
2993 if (oi->disk_sizep)
2994 *(oi->disk_sizep) = 0;
2995 if (oi->delta_base_sha1)
2996 hashclr(oi->delta_base_sha1);
2997 if (oi->typename)
2998 strbuf_addstr(oi->typename, typename(co->type));
2999 if (oi->contentp)
3000 *oi->contentp = xmemdupz(co->buf, co->size);
3001 oi->whence = OI_CACHED;
3002 return 0;
3003 }
3004 }
3005
3006 if (!find_pack_entry(real, &e)) {
3007 /* Most likely it's a loose object. */
3008 if (!sha1_loose_object_info(real, oi, flags)) {
3009 oi->whence = OI_LOOSE;
3010 return 0;
3011 }
3012
3013 /* Not a loose object; someone else may have just packed it. */
3014 if (flags & OBJECT_INFO_QUICK) {
3015 return -1;
3016 } else {
3017 reprepare_packed_git();
3018 if (!find_pack_entry(real, &e))
3019 return -1;
3020 }
3021 }
3022
3023 if (oi == &blank_oi)
3024 /*
3025 * We know that the caller doesn't actually need the
3026 * information below, so return early.
3027 */
3028 return 0;
3029
3030 rtype = packed_object_info(e.p, e.offset, oi);
3031 if (rtype < 0) {
3032 mark_bad_packed_object(e.p, real);
3033 return sha1_object_info_extended(real, oi, 0);
3034 } else if (in_delta_base_cache(e.p, e.offset)) {
3035 oi->whence = OI_DBCACHED;
3036 } else {
3037 oi->whence = OI_PACKED;
3038 oi->u.packed.offset = e.offset;
3039 oi->u.packed.pack = e.p;
3040 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
3041 rtype == OBJ_OFS_DELTA);
3042 }
3043
3044 return 0;
3045 }
3046
3047 /* returns enum object_type or negative */
3048 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
3049 {
3050 enum object_type type;
3051 struct object_info oi = OBJECT_INFO_INIT;
3052
3053 oi.typep = &type;
3054 oi.sizep = sizep;
3055 if (sha1_object_info_extended(sha1, &oi,
3056 OBJECT_INFO_LOOKUP_REPLACE) < 0)
3057 return -1;
3058 return type;
3059 }
3060
3061 static void *read_packed_sha1(const unsigned char *sha1,
3062 enum object_type *type, unsigned long *size)
3063 {
3064 struct pack_entry e;
3065 void *data;
3066
3067 if (!find_pack_entry(sha1, &e))
3068 return NULL;
3069 data = cache_or_unpack_entry(e.p, e.offset, size, type);
3070 if (!data) {
3071 /*
3072 * We're probably in deep shit, but let's try to fetch
3073 * the required object anyway from another pack or loose.
3074 * This should happen only in the presence of a corrupted
3075 * pack, and is better than failing outright.
3076 */
3077 error("failed to read object %s at offset %"PRIuMAX" from %s",
3078 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
3079 mark_bad_packed_object(e.p, sha1);
3080 data = read_object(sha1, type, size);
3081 }
3082 return data;
3083 }
3084
3085 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
3086 unsigned char *sha1)
3087 {
3088 struct cached_object *co;
3089
3090 hash_sha1_file(buf, len, typename(type), sha1);
3091 if (has_sha1_file(sha1) || find_cached_object(sha1))
3092 return 0;
3093 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
3094 co = &cached_objects[cached_object_nr++];
3095 co->size = len;
3096 co->type = type;
3097 co->buf = xmalloc(len);
3098 memcpy(co->buf, buf, len);
3099 hashcpy(co->sha1, sha1);
3100 return 0;
3101 }
3102
3103 static void *read_object(const unsigned char *sha1, enum object_type *type,
3104 unsigned long *size)
3105 {
3106 struct object_info oi = OBJECT_INFO_INIT;
3107 void *content;
3108 oi.typep = type;
3109 oi.sizep = size;
3110 oi.contentp = &content;
3111
3112 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
3113 return NULL;
3114 return content;
3115 }
3116
3117 /*
3118 * This function dies on corrupt objects; the callers who want to
3119 * deal with them should arrange to call read_object() and give error
3120 * messages themselves.
3121 */
3122 void *read_sha1_file_extended(const unsigned char *sha1,
3123 enum object_type *type,
3124 unsigned long *size,
3125 int lookup_replace)
3126 {
3127 void *data;
3128 const struct packed_git *p;
3129 const char *path;
3130 struct stat st;
3131 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
3132 : sha1;
3133
3134 errno = 0;
3135 data = read_object(repl, type, size);
3136 if (data)
3137 return data;
3138
3139 if (errno && errno != ENOENT)
3140 die_errno("failed to read object %s", sha1_to_hex(sha1));
3141
3142 /* die if we replaced an object with one that does not exist */
3143 if (repl != sha1)
3144 die("replacement %s not found for %s",
3145 sha1_to_hex(repl), sha1_to_hex(sha1));
3146
3147 if (!stat_sha1_file(repl, &st, &path))
3148 die("loose object %s (stored in %s) is corrupt",
3149 sha1_to_hex(repl), path);
3150
3151 if ((p = has_packed_and_bad(repl)) != NULL)
3152 die("packed object %s (stored in %s) is corrupt",
3153 sha1_to_hex(repl), p->pack_name);
3154
3155 return NULL;
3156 }
3157
3158 void *read_object_with_reference(const unsigned char *sha1,
3159 const char *required_type_name,
3160 unsigned long *size,
3161 unsigned char *actual_sha1_return)
3162 {
3163 enum object_type type, required_type;
3164 void *buffer;
3165 unsigned long isize;
3166 unsigned char actual_sha1[20];
3167
3168 required_type = type_from_string(required_type_name);
3169 hashcpy(actual_sha1, sha1);
3170 while (1) {
3171 int ref_length = -1;
3172 const char *ref_type = NULL;
3173
3174 buffer = read_sha1_file(actual_sha1, &type, &isize);
3175 if (!buffer)
3176 return NULL;
3177 if (type == required_type) {
3178 *size = isize;
3179 if (actual_sha1_return)
3180 hashcpy(actual_sha1_return, actual_sha1);
3181 return buffer;
3182 }
3183 /* Handle references */
3184 else if (type == OBJ_COMMIT)
3185 ref_type = "tree ";
3186 else if (type == OBJ_TAG)
3187 ref_type = "object ";
3188 else {
3189 free(buffer);
3190 return NULL;
3191 }
3192 ref_length = strlen(ref_type);
3193
3194 if (ref_length + 40 > isize ||
3195 memcmp(buffer, ref_type, ref_length) ||
3196 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
3197 free(buffer);
3198 return NULL;
3199 }
3200 free(buffer);
3201 /* Now we have the ID of the referred-to object in
3202 * actual_sha1. Check again. */
3203 }
3204 }
3205
3206 static void write_sha1_file_prepare(const void *buf, unsigned long len,
3207 const char *type, unsigned char *sha1,
3208 char *hdr, int *hdrlen)
3209 {
3210 git_SHA_CTX c;
3211
3212 /* Generate the header */
3213 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
3214
3215 /* Sha1.. */
3216 git_SHA1_Init(&c);
3217 git_SHA1_Update(&c, hdr, *hdrlen);
3218 git_SHA1_Update(&c, buf, len);
3219 git_SHA1_Final(sha1, &c);
3220 }
3221
3222 /*
3223 * Move the just written object into its final resting place.
3224 */
3225 int finalize_object_file(const char *tmpfile, const char *filename)
3226 {
3227 int ret = 0;
3228
3229 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
3230 goto try_rename;
3231 else if (link(tmpfile, filename))
3232 ret = errno;
3233
3234 /*
3235 * Coda hack - coda doesn't like cross-directory links,
3236 * so we fall back to a rename, which will mean that it
3237 * won't be able to check collisions, but that's not a
3238 * big deal.
3239 *
3240 * The same holds for FAT formatted media.
3241 *
3242 * When this succeeds, we just return. We have nothing
3243 * left to unlink.
3244 */
3245 if (ret && ret != EEXIST) {
3246 try_rename:
3247 if (!rename(tmpfile, filename))
3248 goto out;
3249 ret = errno;
3250 }
3251 unlink_or_warn(tmpfile);
3252 if (ret) {
3253 if (ret != EEXIST) {
3254 return error_errno("unable to write sha1 filename %s", filename);
3255 }
3256 /* FIXME!!! Collision check here ? */
3257 }
3258
3259 out:
3260 if (adjust_shared_perm(filename))
3261 return error("unable to set permission to '%s'", filename);
3262 return 0;
3263 }
3264
3265 static int write_buffer(int fd, const void *buf, size_t len)
3266 {
3267 if (write_in_full(fd, buf, len) < 0)
3268 return error_errno("file write error");
3269 return 0;
3270 }
3271
3272 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
3273 unsigned char *sha1)
3274 {
3275 char hdr[32];
3276 int hdrlen = sizeof(hdr);
3277 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3278 return 0;
3279 }
3280
3281 /* Finalize a file on disk, and close it. */
3282 static void close_sha1_file(int fd)
3283 {
3284 if (fsync_object_files)
3285 fsync_or_die(fd, "sha1 file");
3286 if (close(fd) != 0)
3287 die_errno("error when closing sha1 file");
3288 }
3289
3290 /* Size of directory component, including the ending '/' */
3291 static inline int directory_size(const char *filename)
3292 {
3293 const char *s = strrchr(filename, '/');
3294 if (!s)
3295 return 0;
3296 return s - filename + 1;
3297 }
3298
3299 /*
3300 * This creates a temporary file in the same directory as the final
3301 * 'filename'
3302 *
3303 * We want to avoid cross-directory filename renames, because those
3304 * can have problems on various filesystems (FAT, NFS, Coda).
3305 */
3306 static int create_tmpfile(struct strbuf *tmp, const char *filename)
3307 {
3308 int fd, dirlen = directory_size(filename);
3309
3310 strbuf_reset(tmp);
3311 strbuf_add(tmp, filename, dirlen);
3312 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
3313 fd = git_mkstemp_mode(tmp->buf, 0444);
3314 if (fd < 0 && dirlen && errno == ENOENT) {
3315 /*
3316 * Make sure the directory exists; note that the contents
3317 * of the buffer are undefined after mkstemp returns an
3318 * error, so we have to rewrite the whole buffer from
3319 * scratch.
3320 */
3321 strbuf_reset(tmp);
3322 strbuf_add(tmp, filename, dirlen - 1);
3323 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
3324 return -1;
3325 if (adjust_shared_perm(tmp->buf))
3326 return -1;
3327
3328 /* Try again */
3329 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
3330 fd = git_mkstemp_mode(tmp->buf, 0444);
3331 }
3332 return fd;
3333 }
3334
3335 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
3336 const void *buf, unsigned long len, time_t mtime)
3337 {
3338 int fd, ret;
3339 unsigned char compressed[4096];
3340 git_zstream stream;
3341 git_SHA_CTX c;
3342 unsigned char parano_sha1[20];
3343 static struct strbuf tmp_file = STRBUF_INIT;
3344 const char *filename = sha1_file_name(sha1);
3345
3346 fd = create_tmpfile(&tmp_file, filename);
3347 if (fd < 0) {
3348 if (errno == EACCES)
3349 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3350 else
3351 return error_errno("unable to create temporary file");
3352 }
3353
3354 /* Set it up */
3355 git_deflate_init(&stream, zlib_compression_level);
3356 stream.next_out = compressed;
3357 stream.avail_out = sizeof(compressed);
3358 git_SHA1_Init(&c);
3359
3360 /* First header.. */
3361 stream.next_in = (unsigned char *)hdr;
3362 stream.avail_in = hdrlen;
3363 while (git_deflate(&stream, 0) == Z_OK)
3364 ; /* nothing */
3365 git_SHA1_Update(&c, hdr, hdrlen);
3366
3367 /* Then the data itself.. */
3368 stream.next_in = (void *)buf;
3369 stream.avail_in = len;
3370 do {
3371 unsigned char *in0 = stream.next_in;
3372 ret = git_deflate(&stream, Z_FINISH);
3373 git_SHA1_Update(&c, in0, stream.next_in - in0);
3374 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3375 die("unable to write sha1 file");
3376 stream.next_out = compressed;
3377 stream.avail_out = sizeof(compressed);
3378 } while (ret == Z_OK);
3379
3380 if (ret != Z_STREAM_END)
3381 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
3382 ret = git_deflate_end_gently(&stream);
3383 if (ret != Z_OK)
3384 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
3385 git_SHA1_Final(parano_sha1, &c);
3386 if (hashcmp(sha1, parano_sha1) != 0)
3387 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
3388
3389 close_sha1_file(fd);
3390
3391 if (mtime) {
3392 struct utimbuf utb;
3393 utb.actime = mtime;
3394 utb.modtime = mtime;
3395 if (utime(tmp_file.buf, &utb) < 0)
3396 warning_errno("failed utime() on %s", tmp_file.buf);
3397 }
3398
3399 return finalize_object_file(tmp_file.buf, filename);
3400 }
3401
3402 static int freshen_loose_object(const unsigned char *sha1)
3403 {
3404 return check_and_freshen(sha1, 1);
3405 }
3406
3407 static int freshen_packed_object(const unsigned char *sha1)
3408 {
3409 struct pack_entry e;
3410 if (!find_pack_entry(sha1, &e))
3411 return 0;
3412 if (e.p->freshened)
3413 return 1;
3414 if (!freshen_file(e.p->pack_name))
3415 return 0;
3416 e.p->freshened = 1;
3417 return 1;
3418 }
3419
3420 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3421 {
3422 char hdr[32];
3423 int hdrlen = sizeof(hdr);
3424
3425 /* Normally if we have it in the pack then we do not bother writing
3426 * it out into .git/objects/??/?{38} file.
3427 */
3428 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3429 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3430 return 0;
3431 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3432 }
3433
3434 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3435 unsigned char *sha1, unsigned flags)
3436 {
3437 char *header;
3438 int hdrlen, status = 0;
3439
3440 /* type string, SP, %lu of the length plus NUL must fit this */
3441 hdrlen = strlen(type) + 32;
3442 header = xmalloc(hdrlen);
3443 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3444
3445 if (!(flags & HASH_WRITE_OBJECT))
3446 goto cleanup;
3447 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3448 goto cleanup;
3449 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3450
3451 cleanup:
3452 free(header);
3453 return status;
3454 }
3455
3456 int force_object_loose(const unsigned char *sha1, time_t mtime)
3457 {
3458 void *buf;
3459 unsigned long len;
3460 enum object_type type;
3461 char hdr[32];
3462 int hdrlen;
3463 int ret;
3464
3465 if (has_loose_object(sha1))
3466 return 0;
3467 buf = read_packed_sha1(sha1, &type, &len);
3468 if (!buf)
3469 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3470 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
3471 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3472 free(buf);
3473
3474 return ret;
3475 }
3476
3477 int has_pack_index(const unsigned char *sha1)
3478 {
3479 struct stat st;
3480 if (stat(sha1_pack_index_name(sha1), &st))
3481 return 0;
3482 return 1;
3483 }
3484
3485 int has_sha1_pack(const unsigned char *sha1)
3486 {
3487 struct pack_entry e;
3488 return find_pack_entry(sha1, &e);
3489 }
3490
3491 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3492 {
3493 if (!startup_info->have_repository)
3494 return 0;
3495 return sha1_object_info_extended(sha1, NULL,
3496 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
3497 }
3498
3499 int has_object_file(const struct object_id *oid)
3500 {
3501 return has_sha1_file(oid->hash);
3502 }
3503
3504 int has_object_file_with_flags(const struct object_id *oid, int flags)
3505 {
3506 return has_sha1_file_with_flags(oid->hash, flags);
3507 }
3508
3509 static void check_tree(const void *buf, size_t size)
3510 {
3511 struct tree_desc desc;
3512 struct name_entry entry;
3513
3514 init_tree_desc(&desc, buf, size);
3515 while (tree_entry(&desc, &entry))
3516 /* do nothing
3517 * tree_entry() will die() on malformed entries */
3518 ;
3519 }
3520
3521 static void check_commit(const void *buf, size_t size)
3522 {
3523 struct commit c;
3524 memset(&c, 0, sizeof(c));
3525 if (parse_commit_buffer(&c, buf, size))
3526 die("corrupt commit");
3527 }
3528
3529 static void check_tag(const void *buf, size_t size)
3530 {
3531 struct tag t;
3532 memset(&t, 0, sizeof(t));
3533 if (parse_tag_buffer(&t, buf, size))
3534 die("corrupt tag");
3535 }
3536
3537 static int index_mem(unsigned char *sha1, void *buf, size_t size,
3538 enum object_type type,
3539 const char *path, unsigned flags)
3540 {
3541 int ret, re_allocated = 0;
3542 int write_object = flags & HASH_WRITE_OBJECT;
3543
3544 if (!type)
3545 type = OBJ_BLOB;
3546
3547 /*
3548 * Convert blobs to git internal format
3549 */
3550 if ((type == OBJ_BLOB) && path) {
3551 struct strbuf nbuf = STRBUF_INIT;
3552 if (convert_to_git(&the_index, path, buf, size, &nbuf,
3553 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3554 buf = strbuf_detach(&nbuf, &size);
3555 re_allocated = 1;
3556 }
3557 }
3558 if (flags & HASH_FORMAT_CHECK) {
3559 if (type == OBJ_TREE)
3560 check_tree(buf, size);
3561 if (type == OBJ_COMMIT)
3562 check_commit(buf, size);
3563 if (type == OBJ_TAG)
3564 check_tag(buf, size);
3565 }
3566
3567 if (write_object)
3568 ret = write_sha1_file(buf, size, typename(type), sha1);
3569 else
3570 ret = hash_sha1_file(buf, size, typename(type), sha1);
3571 if (re_allocated)
3572 free(buf);
3573 return ret;
3574 }
3575
3576 static int index_stream_convert_blob(unsigned char *sha1, int fd,
3577 const char *path, unsigned flags)
3578 {
3579 int ret;
3580 const int write_object = flags & HASH_WRITE_OBJECT;
3581 struct strbuf sbuf = STRBUF_INIT;
3582
3583 assert(path);
3584 assert(would_convert_to_git_filter_fd(path));
3585
3586 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
3587 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3588
3589 if (write_object)
3590 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3591 sha1);
3592 else
3593 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3594 sha1);
3595 strbuf_release(&sbuf);
3596 return ret;
3597 }
3598
3599 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3600 const char *path, unsigned flags)
3601 {
3602 struct strbuf sbuf = STRBUF_INIT;
3603 int ret;
3604
3605 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3606 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3607 else
3608 ret = -1;
3609 strbuf_release(&sbuf);
3610 return ret;
3611 }
3612
3613 #define SMALL_FILE_SIZE (32*1024)
3614
3615 static int index_core(unsigned char *sha1, int fd, size_t size,
3616 enum object_type type, const char *path,
3617 unsigned flags)
3618 {
3619 int ret;
3620
3621 if (!size) {
3622 ret = index_mem(sha1, "", size, type, path, flags);
3623 } else if (size <= SMALL_FILE_SIZE) {
3624 char *buf = xmalloc(size);
3625 if (size == read_in_full(fd, buf, size))
3626 ret = index_mem(sha1, buf, size, type, path, flags);
3627 else
3628 ret = error_errno("short read");
3629 free(buf);
3630 } else {
3631 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3632 ret = index_mem(sha1, buf, size, type, path, flags);
3633 munmap(buf, size);
3634 }
3635 return ret;
3636 }
3637
3638 /*
3639 * This creates one packfile per large blob unless bulk-checkin
3640 * machinery is "plugged".
3641 *
3642 * This also bypasses the usual "convert-to-git" dance, and that is on
3643 * purpose. We could write a streaming version of the converting
3644 * functions and insert that before feeding the data to fast-import
3645 * (or equivalent in-core API described above). However, that is
3646 * somewhat complicated, as we do not know the size of the filter
3647 * result, which we need to know beforehand when writing a git object.
3648 * Since the primary motivation for trying to stream from the working
3649 * tree file and to avoid mmaping it in core is to deal with large
3650 * binary blobs, they generally do not want to get any conversion, and
3651 * callers should avoid this code path when filters are requested.
3652 */
3653 static int index_stream(unsigned char *sha1, int fd, size_t size,
3654 enum object_type type, const char *path,
3655 unsigned flags)
3656 {
3657 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3658 }
3659
3660 int index_fd(unsigned char *sha1, int fd, struct stat *st,
3661 enum object_type type, const char *path, unsigned flags)
3662 {
3663 int ret;
3664
3665 /*
3666 * Call xsize_t() only when needed to avoid potentially unnecessary
3667 * die() for large files.
3668 */
3669 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3670 ret = index_stream_convert_blob(sha1, fd, path, flags);
3671 else if (!S_ISREG(st->st_mode))
3672 ret = index_pipe(sha1, fd, type, path, flags);
3673 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3674 (path && would_convert_to_git(&the_index, path)))
3675 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3676 flags);
3677 else
3678 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3679 flags);
3680 close(fd);
3681 return ret;
3682 }
3683
3684 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3685 {
3686 int fd;
3687 struct strbuf sb = STRBUF_INIT;
3688
3689 switch (st->st_mode & S_IFMT) {
3690 case S_IFREG:
3691 fd = open(path, O_RDONLY);
3692 if (fd < 0)
3693 return error_errno("open(\"%s\")", path);
3694 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3695 return error("%s: failed to insert into database",
3696 path);
3697 break;
3698 case S_IFLNK:
3699 if (strbuf_readlink(&sb, path, st->st_size))
3700 return error_errno("readlink(\"%s\")", path);
3701 if (!(flags & HASH_WRITE_OBJECT))
3702 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3703 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3704 return error("%s: failed to insert into database",
3705 path);
3706 strbuf_release(&sb);
3707 break;
3708 case S_IFDIR:
3709 return resolve_gitlink_ref(path, "HEAD", sha1);
3710 default:
3711 return error("%s: unsupported file type", path);
3712 }
3713 return 0;
3714 }
3715
3716 int read_pack_header(int fd, struct pack_header *header)
3717 {
3718 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3719 /* "eof before pack header was fully read" */
3720 return PH_ERROR_EOF;
3721
3722 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3723 /* "protocol error (pack signature mismatch detected)" */
3724 return PH_ERROR_PACK_SIGNATURE;
3725 if (!pack_version_ok(header->hdr_version))
3726 /* "protocol error (pack version unsupported)" */
3727 return PH_ERROR_PROTOCOL;
3728 return 0;
3729 }
3730
3731 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3732 {
3733 enum object_type type = sha1_object_info(sha1, NULL);
3734 if (type < 0)
3735 die("%s is not a valid object", sha1_to_hex(sha1));
3736 if (type != expect)
3737 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3738 typename(expect));
3739 }
3740
3741 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
3742 struct strbuf *path,
3743 each_loose_object_fn obj_cb,
3744 each_loose_cruft_fn cruft_cb,
3745 each_loose_subdir_fn subdir_cb,
3746 void *data)
3747 {
3748 size_t origlen, baselen;
3749 DIR *dir;
3750 struct dirent *de;
3751 int r = 0;
3752
3753 if (subdir_nr > 0xff)
3754 BUG("invalid loose object subdirectory: %x", subdir_nr);
3755
3756 origlen = path->len;
3757 strbuf_complete(path, '/');
3758 strbuf_addf(path, "%02x", subdir_nr);
3759 baselen = path->len;
3760
3761 dir = opendir(path->buf);
3762 if (!dir) {
3763 if (errno != ENOENT)
3764 r = error_errno("unable to open %s", path->buf);
3765 strbuf_setlen(path, origlen);
3766 return r;
3767 }
3768
3769 while ((de = readdir(dir))) {
3770 if (is_dot_or_dotdot(de->d_name))
3771 continue;
3772
3773 strbuf_setlen(path, baselen);
3774 strbuf_addf(path, "/%s", de->d_name);
3775
3776 if (strlen(de->d_name) == GIT_SHA1_HEXSZ - 2) {
3777 char hex[GIT_MAX_HEXSZ+1];
3778 struct object_id oid;
3779
3780 xsnprintf(hex, sizeof(hex), "%02x%s",
3781 subdir_nr, de->d_name);
3782 if (!get_oid_hex(hex, &oid)) {
3783 if (obj_cb) {
3784 r = obj_cb(&oid, path->buf, data);
3785 if (r)
3786 break;
3787 }
3788 continue;
3789 }
3790 }
3791
3792 if (cruft_cb) {
3793 r = cruft_cb(de->d_name, path->buf, data);
3794 if (r)
3795 break;
3796 }
3797 }
3798 closedir(dir);
3799
3800 strbuf_setlen(path, baselen);
3801 if (!r && subdir_cb)
3802 r = subdir_cb(subdir_nr, path->buf, data);
3803
3804 strbuf_setlen(path, origlen);
3805
3806 return r;
3807 }
3808
3809 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3810 each_loose_object_fn obj_cb,
3811 each_loose_cruft_fn cruft_cb,
3812 each_loose_subdir_fn subdir_cb,
3813 void *data)
3814 {
3815 int r = 0;
3816 int i;
3817
3818 for (i = 0; i < 256; i++) {
3819 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3820 subdir_cb, data);
3821 if (r)
3822 break;
3823 }
3824
3825 return r;
3826 }
3827
3828 int for_each_loose_file_in_objdir(const char *path,
3829 each_loose_object_fn obj_cb,
3830 each_loose_cruft_fn cruft_cb,
3831 each_loose_subdir_fn subdir_cb,
3832 void *data)
3833 {
3834 struct strbuf buf = STRBUF_INIT;
3835 int r;
3836
3837 strbuf_addstr(&buf, path);
3838 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3839 subdir_cb, data);
3840 strbuf_release(&buf);
3841
3842 return r;
3843 }
3844
3845 struct loose_alt_odb_data {
3846 each_loose_object_fn *cb;
3847 void *data;
3848 };
3849
3850 static int loose_from_alt_odb(struct alternate_object_database *alt,
3851 void *vdata)
3852 {
3853 struct loose_alt_odb_data *data = vdata;
3854 struct strbuf buf = STRBUF_INIT;
3855 int r;
3856
3857 strbuf_addstr(&buf, alt->path);
3858 r = for_each_loose_file_in_objdir_buf(&buf,
3859 data->cb, NULL, NULL,
3860 data->data);
3861 strbuf_release(&buf);
3862 return r;
3863 }
3864
3865 int for_each_loose_object(each_loose_object_fn cb, void *