Merge branch 'dl/cocci-everywhere'
[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 "repository.h"
26 #include "replace-object.h"
27 #include "streaming.h"
28 #include "dir.h"
29 #include "list.h"
30 #include "mergesort.h"
31 #include "quote.h"
32 #include "packfile.h"
33 #include "object-store.h"
34 #include "promisor-remote.h"
35
36 /* The maximum size for an object header. */
37 #define MAX_HEADER_LEN 32
38
39
40 #define EMPTY_TREE_SHA1_BIN_LITERAL \
41 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60" \
42 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04"
43 #define EMPTY_TREE_SHA256_BIN_LITERAL \
44 "\x6e\xf1\x9b\x41\x22\x5c\x53\x69\xf1\xc1" \
45 "\x04\xd4\x5d\x8d\x85\xef\xa9\xb0\x57\xb5" \
46 "\x3b\x14\xb4\xb9\xb9\x39\xdd\x74\xde\xcc" \
47 "\x53\x21"
48
49 #define EMPTY_BLOB_SHA1_BIN_LITERAL \
50 "\xe6\x9d\xe2\x9b\xb2\xd1\xd6\x43\x4b\x8b" \
51 "\x29\xae\x77\x5a\xd8\xc2\xe4\x8c\x53\x91"
52 #define EMPTY_BLOB_SHA256_BIN_LITERAL \
53 "\x47\x3a\x0f\x4c\x3b\xe8\xa9\x36\x81\xa2" \
54 "\x67\xe3\xb1\xe9\xa7\xdc\xda\x11\x85\x43" \
55 "\x6f\xe1\x41\xf7\x74\x91\x20\xa3\x03\x72" \
56 "\x18\x13"
57
58 const unsigned char null_sha1[GIT_MAX_RAWSZ];
59 const struct object_id null_oid;
60 static const struct object_id empty_tree_oid = {
61 EMPTY_TREE_SHA1_BIN_LITERAL
62 };
63 static const struct object_id empty_blob_oid = {
64 EMPTY_BLOB_SHA1_BIN_LITERAL
65 };
66 static const struct object_id empty_tree_oid_sha256 = {
67 EMPTY_TREE_SHA256_BIN_LITERAL
68 };
69 static const struct object_id empty_blob_oid_sha256 = {
70 EMPTY_BLOB_SHA256_BIN_LITERAL
71 };
72
73 static void git_hash_sha1_init(git_hash_ctx *ctx)
74 {
75 git_SHA1_Init(&ctx->sha1);
76 }
77
78 static void git_hash_sha1_update(git_hash_ctx *ctx, const void *data, size_t len)
79 {
80 git_SHA1_Update(&ctx->sha1, data, len);
81 }
82
83 static void git_hash_sha1_final(unsigned char *hash, git_hash_ctx *ctx)
84 {
85 git_SHA1_Final(hash, &ctx->sha1);
86 }
87
88
89 static void git_hash_sha256_init(git_hash_ctx *ctx)
90 {
91 git_SHA256_Init(&ctx->sha256);
92 }
93
94 static void git_hash_sha256_update(git_hash_ctx *ctx, const void *data, size_t len)
95 {
96 git_SHA256_Update(&ctx->sha256, data, len);
97 }
98
99 static void git_hash_sha256_final(unsigned char *hash, git_hash_ctx *ctx)
100 {
101 git_SHA256_Final(hash, &ctx->sha256);
102 }
103
104 static void git_hash_unknown_init(git_hash_ctx *ctx)
105 {
106 BUG("trying to init unknown hash");
107 }
108
109 static void git_hash_unknown_update(git_hash_ctx *ctx, const void *data, size_t len)
110 {
111 BUG("trying to update unknown hash");
112 }
113
114 static void git_hash_unknown_final(unsigned char *hash, git_hash_ctx *ctx)
115 {
116 BUG("trying to finalize unknown hash");
117 }
118
119 const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
120 {
121 NULL,
122 0x00000000,
123 0,
124 0,
125 0,
126 git_hash_unknown_init,
127 git_hash_unknown_update,
128 git_hash_unknown_final,
129 NULL,
130 NULL,
131 },
132 {
133 "sha1",
134 /* "sha1", big-endian */
135 0x73686131,
136 GIT_SHA1_RAWSZ,
137 GIT_SHA1_HEXSZ,
138 GIT_SHA1_BLKSZ,
139 git_hash_sha1_init,
140 git_hash_sha1_update,
141 git_hash_sha1_final,
142 &empty_tree_oid,
143 &empty_blob_oid,
144 },
145 {
146 "sha256",
147 /* "s256", big-endian */
148 0x73323536,
149 GIT_SHA256_RAWSZ,
150 GIT_SHA256_HEXSZ,
151 GIT_SHA256_BLKSZ,
152 git_hash_sha256_init,
153 git_hash_sha256_update,
154 git_hash_sha256_final,
155 &empty_tree_oid_sha256,
156 &empty_blob_oid_sha256,
157 }
158 };
159
160 const char *empty_tree_oid_hex(void)
161 {
162 static char buf[GIT_MAX_HEXSZ + 1];
163 return oid_to_hex_r(buf, the_hash_algo->empty_tree);
164 }
165
166 const char *empty_blob_oid_hex(void)
167 {
168 static char buf[GIT_MAX_HEXSZ + 1];
169 return oid_to_hex_r(buf, the_hash_algo->empty_blob);
170 }
171
172 int hash_algo_by_name(const char *name)
173 {
174 int i;
175 if (!name)
176 return GIT_HASH_UNKNOWN;
177 for (i = 1; i < GIT_HASH_NALGOS; i++)
178 if (!strcmp(name, hash_algos[i].name))
179 return i;
180 return GIT_HASH_UNKNOWN;
181 }
182
183 int hash_algo_by_id(uint32_t format_id)
184 {
185 int i;
186 for (i = 1; i < GIT_HASH_NALGOS; i++)
187 if (format_id == hash_algos[i].format_id)
188 return i;
189 return GIT_HASH_UNKNOWN;
190 }
191
192 int hash_algo_by_length(int len)
193 {
194 int i;
195 for (i = 1; i < GIT_HASH_NALGOS; i++)
196 if (len == hash_algos[i].rawsz)
197 return i;
198 return GIT_HASH_UNKNOWN;
199 }
200
201 /*
202 * This is meant to hold a *small* number of objects that you would
203 * want read_object_file() to be able to return, but yet you do not want
204 * to write them into the object store (e.g. a browse-only
205 * application).
206 */
207 static struct cached_object {
208 struct object_id oid;
209 enum object_type type;
210 void *buf;
211 unsigned long size;
212 } *cached_objects;
213 static int cached_object_nr, cached_object_alloc;
214
215 static struct cached_object empty_tree = {
216 { EMPTY_TREE_SHA1_BIN_LITERAL },
217 OBJ_TREE,
218 "",
219 0
220 };
221
222 static struct cached_object *find_cached_object(const struct object_id *oid)
223 {
224 int i;
225 struct cached_object *co = cached_objects;
226
227 for (i = 0; i < cached_object_nr; i++, co++) {
228 if (oideq(&co->oid, oid))
229 return co;
230 }
231 if (oideq(oid, the_hash_algo->empty_tree))
232 return &empty_tree;
233 return NULL;
234 }
235
236
237 static int get_conv_flags(unsigned flags)
238 {
239 if (flags & HASH_RENORMALIZE)
240 return CONV_EOL_RENORMALIZE;
241 else if (flags & HASH_WRITE_OBJECT)
242 return global_conv_flags_eol | CONV_WRITE_OBJECT;
243 else
244 return 0;
245 }
246
247
248 int mkdir_in_gitdir(const char *path)
249 {
250 if (mkdir(path, 0777)) {
251 int saved_errno = errno;
252 struct stat st;
253 struct strbuf sb = STRBUF_INIT;
254
255 if (errno != EEXIST)
256 return -1;
257 /*
258 * Are we looking at a path in a symlinked worktree
259 * whose original repository does not yet have it?
260 * e.g. .git/rr-cache pointing at its original
261 * repository in which the user hasn't performed any
262 * conflict resolution yet?
263 */
264 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
265 strbuf_readlink(&sb, path, st.st_size) ||
266 !is_absolute_path(sb.buf) ||
267 mkdir(sb.buf, 0777)) {
268 strbuf_release(&sb);
269 errno = saved_errno;
270 return -1;
271 }
272 strbuf_release(&sb);
273 }
274 return adjust_shared_perm(path);
275 }
276
277 enum scld_error safe_create_leading_directories(char *path)
278 {
279 char *next_component = path + offset_1st_component(path);
280 enum scld_error ret = SCLD_OK;
281
282 while (ret == SCLD_OK && next_component) {
283 struct stat st;
284 char *slash = next_component, slash_character;
285
286 while (*slash && !is_dir_sep(*slash))
287 slash++;
288
289 if (!*slash)
290 break;
291
292 next_component = slash + 1;
293 while (is_dir_sep(*next_component))
294 next_component++;
295 if (!*next_component)
296 break;
297
298 slash_character = *slash;
299 *slash = '\0';
300 if (!stat(path, &st)) {
301 /* path exists */
302 if (!S_ISDIR(st.st_mode)) {
303 errno = ENOTDIR;
304 ret = SCLD_EXISTS;
305 }
306 } else if (mkdir(path, 0777)) {
307 if (errno == EEXIST &&
308 !stat(path, &st) && S_ISDIR(st.st_mode))
309 ; /* somebody created it since we checked */
310 else if (errno == ENOENT)
311 /*
312 * Either mkdir() failed because
313 * somebody just pruned the containing
314 * directory, or stat() failed because
315 * the file that was in our way was
316 * just removed. Either way, inform
317 * the caller that it might be worth
318 * trying again:
319 */
320 ret = SCLD_VANISHED;
321 else
322 ret = SCLD_FAILED;
323 } else if (adjust_shared_perm(path)) {
324 ret = SCLD_PERMS;
325 }
326 *slash = slash_character;
327 }
328 return ret;
329 }
330
331 enum scld_error safe_create_leading_directories_const(const char *path)
332 {
333 int save_errno;
334 /* path points to cache entries, so xstrdup before messing with it */
335 char *buf = xstrdup(path);
336 enum scld_error result = safe_create_leading_directories(buf);
337
338 save_errno = errno;
339 free(buf);
340 errno = save_errno;
341 return result;
342 }
343
344 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
345 {
346 /*
347 * The number of times we will try to remove empty directories
348 * in the way of path. This is only 1 because if another
349 * process is racily creating directories that conflict with
350 * us, we don't want to fight against them.
351 */
352 int remove_directories_remaining = 1;
353
354 /*
355 * The number of times that we will try to create the
356 * directories containing path. We are willing to attempt this
357 * more than once, because another process could be trying to
358 * clean up empty directories at the same time as we are
359 * trying to create them.
360 */
361 int create_directories_remaining = 3;
362
363 /* A scratch copy of path, filled lazily if we need it: */
364 struct strbuf path_copy = STRBUF_INIT;
365
366 int ret, save_errno;
367
368 /* Sanity check: */
369 assert(*path);
370
371 retry_fn:
372 ret = fn(path, cb);
373 save_errno = errno;
374 if (!ret)
375 goto out;
376
377 if (errno == EISDIR && remove_directories_remaining-- > 0) {
378 /*
379 * A directory is in the way. Maybe it is empty; try
380 * to remove it:
381 */
382 if (!path_copy.len)
383 strbuf_addstr(&path_copy, path);
384
385 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
386 goto retry_fn;
387 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
388 /*
389 * Maybe the containing directory didn't exist, or
390 * maybe it was just deleted by a process that is
391 * racing with us to clean up empty directories. Try
392 * to create it:
393 */
394 enum scld_error scld_result;
395
396 if (!path_copy.len)
397 strbuf_addstr(&path_copy, path);
398
399 do {
400 scld_result = safe_create_leading_directories(path_copy.buf);
401 if (scld_result == SCLD_OK)
402 goto retry_fn;
403 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
404 }
405
406 out:
407 strbuf_release(&path_copy);
408 errno = save_errno;
409 return ret;
410 }
411
412 static void fill_loose_path(struct strbuf *buf, const struct object_id *oid)
413 {
414 int i;
415 for (i = 0; i < the_hash_algo->rawsz; i++) {
416 static char hex[] = "0123456789abcdef";
417 unsigned int val = oid->hash[i];
418 strbuf_addch(buf, hex[val >> 4]);
419 strbuf_addch(buf, hex[val & 0xf]);
420 if (!i)
421 strbuf_addch(buf, '/');
422 }
423 }
424
425 static const char *odb_loose_path(struct object_directory *odb,
426 struct strbuf *buf,
427 const struct object_id *oid)
428 {
429 strbuf_reset(buf);
430 strbuf_addstr(buf, odb->path);
431 strbuf_addch(buf, '/');
432 fill_loose_path(buf, oid);
433 return buf->buf;
434 }
435
436 const char *loose_object_path(struct repository *r, struct strbuf *buf,
437 const struct object_id *oid)
438 {
439 return odb_loose_path(r->objects->odb, buf, oid);
440 }
441
442 /*
443 * Return non-zero iff the path is usable as an alternate object database.
444 */
445 static int alt_odb_usable(struct raw_object_store *o,
446 struct strbuf *path,
447 const char *normalized_objdir)
448 {
449 struct object_directory *odb;
450
451 /* Detect cases where alternate disappeared */
452 if (!is_directory(path->buf)) {
453 error(_("object directory %s does not exist; "
454 "check .git/objects/info/alternates"),
455 path->buf);
456 return 0;
457 }
458
459 /*
460 * Prevent the common mistake of listing the same
461 * thing twice, or object directory itself.
462 */
463 for (odb = o->odb; odb; odb = odb->next) {
464 if (!fspathcmp(path->buf, odb->path))
465 return 0;
466 }
467 if (!fspathcmp(path->buf, normalized_objdir))
468 return 0;
469
470 return 1;
471 }
472
473 /*
474 * Prepare alternate object database registry.
475 *
476 * The variable alt_odb_list points at the list of struct
477 * object_directory. The elements on this list come from
478 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
479 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
480 * whose contents is similar to that environment variable but can be
481 * LF separated. Its base points at a statically allocated buffer that
482 * contains "/the/directory/corresponding/to/.git/objects/...", while
483 * its name points just after the slash at the end of ".git/objects/"
484 * in the example above, and has enough space to hold 40-byte hex
485 * SHA1, an extra slash for the first level indirection, and the
486 * terminating NUL.
487 */
488 static void read_info_alternates(struct repository *r,
489 const char *relative_base,
490 int depth);
491 static int link_alt_odb_entry(struct repository *r, const char *entry,
492 const char *relative_base, int depth, const char *normalized_objdir)
493 {
494 struct object_directory *ent;
495 struct strbuf pathbuf = STRBUF_INIT;
496
497 if (!is_absolute_path(entry) && relative_base) {
498 strbuf_realpath(&pathbuf, relative_base, 1);
499 strbuf_addch(&pathbuf, '/');
500 }
501 strbuf_addstr(&pathbuf, entry);
502
503 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
504 error(_("unable to normalize alternate object path: %s"),
505 pathbuf.buf);
506 strbuf_release(&pathbuf);
507 return -1;
508 }
509
510 /*
511 * The trailing slash after the directory name is given by
512 * this function at the end. Remove duplicates.
513 */
514 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
515 strbuf_setlen(&pathbuf, pathbuf.len - 1);
516
517 if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir)) {
518 strbuf_release(&pathbuf);
519 return -1;
520 }
521
522 ent = xcalloc(1, sizeof(*ent));
523 ent->path = xstrdup(pathbuf.buf);
524
525 /* add the alternate entry */
526 *r->objects->odb_tail = ent;
527 r->objects->odb_tail = &(ent->next);
528 ent->next = NULL;
529
530 /* recursively add alternates */
531 read_info_alternates(r, pathbuf.buf, depth + 1);
532
533 strbuf_release(&pathbuf);
534 return 0;
535 }
536
537 static const char *parse_alt_odb_entry(const char *string,
538 int sep,
539 struct strbuf *out)
540 {
541 const char *end;
542
543 strbuf_reset(out);
544
545 if (*string == '#') {
546 /* comment; consume up to next separator */
547 end = strchrnul(string, sep);
548 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
549 /*
550 * quoted path; unquote_c_style has copied the
551 * data for us and set "end". Broken quoting (e.g.,
552 * an entry that doesn't end with a quote) falls
553 * back to the unquoted case below.
554 */
555 } else {
556 /* normal, unquoted path */
557 end = strchrnul(string, sep);
558 strbuf_add(out, string, end - string);
559 }
560
561 if (*end)
562 end++;
563 return end;
564 }
565
566 static void link_alt_odb_entries(struct repository *r, const char *alt,
567 int sep, const char *relative_base, int depth)
568 {
569 struct strbuf objdirbuf = STRBUF_INIT;
570 struct strbuf entry = STRBUF_INIT;
571
572 if (!alt || !*alt)
573 return;
574
575 if (depth > 5) {
576 error(_("%s: ignoring alternate object stores, nesting too deep"),
577 relative_base);
578 return;
579 }
580
581 strbuf_add_absolute_path(&objdirbuf, r->objects->odb->path);
582 if (strbuf_normalize_path(&objdirbuf) < 0)
583 die(_("unable to normalize object directory: %s"),
584 objdirbuf.buf);
585
586 while (*alt) {
587 alt = parse_alt_odb_entry(alt, sep, &entry);
588 if (!entry.len)
589 continue;
590 link_alt_odb_entry(r, entry.buf,
591 relative_base, depth, objdirbuf.buf);
592 }
593 strbuf_release(&entry);
594 strbuf_release(&objdirbuf);
595 }
596
597 static void read_info_alternates(struct repository *r,
598 const char *relative_base,
599 int depth)
600 {
601 char *path;
602 struct strbuf buf = STRBUF_INIT;
603
604 path = xstrfmt("%s/info/alternates", relative_base);
605 if (strbuf_read_file(&buf, path, 1024) < 0) {
606 warn_on_fopen_errors(path);
607 free(path);
608 return;
609 }
610
611 link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
612 strbuf_release(&buf);
613 free(path);
614 }
615
616 void add_to_alternates_file(const char *reference)
617 {
618 struct lock_file lock = LOCK_INIT;
619 char *alts = git_pathdup("objects/info/alternates");
620 FILE *in, *out;
621 int found = 0;
622
623 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
624 out = fdopen_lock_file(&lock, "w");
625 if (!out)
626 die_errno(_("unable to fdopen alternates lockfile"));
627
628 in = fopen(alts, "r");
629 if (in) {
630 struct strbuf line = STRBUF_INIT;
631
632 while (strbuf_getline(&line, in) != EOF) {
633 if (!strcmp(reference, line.buf)) {
634 found = 1;
635 break;
636 }
637 fprintf_or_die(out, "%s\n", line.buf);
638 }
639
640 strbuf_release(&line);
641 fclose(in);
642 }
643 else if (errno != ENOENT)
644 die_errno(_("unable to read alternates file"));
645
646 if (found) {
647 rollback_lock_file(&lock);
648 } else {
649 fprintf_or_die(out, "%s\n", reference);
650 if (commit_lock_file(&lock))
651 die_errno(_("unable to move new alternates file into place"));
652 if (the_repository->objects->loaded_alternates)
653 link_alt_odb_entries(the_repository, reference,
654 '\n', NULL, 0);
655 }
656 free(alts);
657 }
658
659 void add_to_alternates_memory(const char *reference)
660 {
661 /*
662 * Make sure alternates are initialized, or else our entry may be
663 * overwritten when they are.
664 */
665 prepare_alt_odb(the_repository);
666
667 link_alt_odb_entries(the_repository, reference,
668 '\n', NULL, 0);
669 }
670
671 /*
672 * Compute the exact path an alternate is at and returns it. In case of
673 * error NULL is returned and the human readable error is added to `err`
674 * `path` may be relative and should point to $GIT_DIR.
675 * `err` must not be null.
676 */
677 char *compute_alternate_path(const char *path, struct strbuf *err)
678 {
679 char *ref_git = NULL;
680 const char *repo, *ref_git_s;
681 int seen_error = 0;
682
683 ref_git_s = real_path_if_valid(path);
684 if (!ref_git_s) {
685 seen_error = 1;
686 strbuf_addf(err, _("path '%s' does not exist"), path);
687 goto out;
688 } else
689 /*
690 * Beware: read_gitfile(), real_path() and mkpath()
691 * return static buffer
692 */
693 ref_git = xstrdup(ref_git_s);
694
695 repo = read_gitfile(ref_git);
696 if (!repo)
697 repo = read_gitfile(mkpath("%s/.git", ref_git));
698 if (repo) {
699 free(ref_git);
700 ref_git = xstrdup(repo);
701 }
702
703 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
704 char *ref_git_git = mkpathdup("%s/.git", ref_git);
705 free(ref_git);
706 ref_git = ref_git_git;
707 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
708 struct strbuf sb = STRBUF_INIT;
709 seen_error = 1;
710 if (get_common_dir(&sb, ref_git)) {
711 strbuf_addf(err,
712 _("reference repository '%s' as a linked "
713 "checkout is not supported yet."),
714 path);
715 goto out;
716 }
717
718 strbuf_addf(err, _("reference repository '%s' is not a "
719 "local repository."), path);
720 goto out;
721 }
722
723 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
724 strbuf_addf(err, _("reference repository '%s' is shallow"),
725 path);
726 seen_error = 1;
727 goto out;
728 }
729
730 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
731 strbuf_addf(err,
732 _("reference repository '%s' is grafted"),
733 path);
734 seen_error = 1;
735 goto out;
736 }
737
738 out:
739 if (seen_error) {
740 FREE_AND_NULL(ref_git);
741 }
742
743 return ref_git;
744 }
745
746 static void fill_alternate_refs_command(struct child_process *cmd,
747 const char *repo_path)
748 {
749 const char *value;
750
751 if (!git_config_get_value("core.alternateRefsCommand", &value)) {
752 cmd->use_shell = 1;
753
754 argv_array_push(&cmd->args, value);
755 argv_array_push(&cmd->args, repo_path);
756 } else {
757 cmd->git_cmd = 1;
758
759 argv_array_pushf(&cmd->args, "--git-dir=%s", repo_path);
760 argv_array_push(&cmd->args, "for-each-ref");
761 argv_array_push(&cmd->args, "--format=%(objectname)");
762
763 if (!git_config_get_value("core.alternateRefsPrefixes", &value)) {
764 argv_array_push(&cmd->args, "--");
765 argv_array_split(&cmd->args, value);
766 }
767 }
768
769 cmd->env = local_repo_env;
770 cmd->out = -1;
771 }
772
773 static void read_alternate_refs(const char *path,
774 alternate_ref_fn *cb,
775 void *data)
776 {
777 struct child_process cmd = CHILD_PROCESS_INIT;
778 struct strbuf line = STRBUF_INIT;
779 FILE *fh;
780
781 fill_alternate_refs_command(&cmd, path);
782
783 if (start_command(&cmd))
784 return;
785
786 fh = xfdopen(cmd.out, "r");
787 while (strbuf_getline_lf(&line, fh) != EOF) {
788 struct object_id oid;
789 const char *p;
790
791 if (parse_oid_hex(line.buf, &oid, &p) || *p) {
792 warning(_("invalid line while parsing alternate refs: %s"),
793 line.buf);
794 break;
795 }
796
797 cb(&oid, data);
798 }
799
800 fclose(fh);
801 finish_command(&cmd);
802 strbuf_release(&line);
803 }
804
805 struct alternate_refs_data {
806 alternate_ref_fn *fn;
807 void *data;
808 };
809
810 static int refs_from_alternate_cb(struct object_directory *e,
811 void *data)
812 {
813 struct strbuf path = STRBUF_INIT;
814 size_t base_len;
815 struct alternate_refs_data *cb = data;
816
817 if (!strbuf_realpath(&path, e->path, 0))
818 goto out;
819 if (!strbuf_strip_suffix(&path, "/objects"))
820 goto out;
821 base_len = path.len;
822
823 /* Is this a git repository with refs? */
824 strbuf_addstr(&path, "/refs");
825 if (!is_directory(path.buf))
826 goto out;
827 strbuf_setlen(&path, base_len);
828
829 read_alternate_refs(path.buf, cb->fn, cb->data);
830
831 out:
832 strbuf_release(&path);
833 return 0;
834 }
835
836 void for_each_alternate_ref(alternate_ref_fn fn, void *data)
837 {
838 struct alternate_refs_data cb;
839 cb.fn = fn;
840 cb.data = data;
841 foreach_alt_odb(refs_from_alternate_cb, &cb);
842 }
843
844 int foreach_alt_odb(alt_odb_fn fn, void *cb)
845 {
846 struct object_directory *ent;
847 int r = 0;
848
849 prepare_alt_odb(the_repository);
850 for (ent = the_repository->objects->odb->next; ent; ent = ent->next) {
851 r = fn(ent, cb);
852 if (r)
853 break;
854 }
855 return r;
856 }
857
858 void prepare_alt_odb(struct repository *r)
859 {
860 if (r->objects->loaded_alternates)
861 return;
862
863 link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
864
865 read_info_alternates(r, r->objects->odb->path, 0);
866 r->objects->loaded_alternates = 1;
867 }
868
869 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
870 static int freshen_file(const char *fn)
871 {
872 struct utimbuf t;
873 t.actime = t.modtime = time(NULL);
874 return !utime(fn, &t);
875 }
876
877 /*
878 * All of the check_and_freshen functions return 1 if the file exists and was
879 * freshened (if freshening was requested), 0 otherwise. If they return
880 * 0, you should not assume that it is safe to skip a write of the object (it
881 * either does not exist on disk, or has a stale mtime and may be subject to
882 * pruning).
883 */
884 int check_and_freshen_file(const char *fn, int freshen)
885 {
886 if (access(fn, F_OK))
887 return 0;
888 if (freshen && !freshen_file(fn))
889 return 0;
890 return 1;
891 }
892
893 static int check_and_freshen_odb(struct object_directory *odb,
894 const struct object_id *oid,
895 int freshen)
896 {
897 static struct strbuf path = STRBUF_INIT;
898 odb_loose_path(odb, &path, oid);
899 return check_and_freshen_file(path.buf, freshen);
900 }
901
902 static int check_and_freshen_local(const struct object_id *oid, int freshen)
903 {
904 return check_and_freshen_odb(the_repository->objects->odb, oid, freshen);
905 }
906
907 static int check_and_freshen_nonlocal(const struct object_id *oid, int freshen)
908 {
909 struct object_directory *odb;
910
911 prepare_alt_odb(the_repository);
912 for (odb = the_repository->objects->odb->next; odb; odb = odb->next) {
913 if (check_and_freshen_odb(odb, oid, freshen))
914 return 1;
915 }
916 return 0;
917 }
918
919 static int check_and_freshen(const struct object_id *oid, int freshen)
920 {
921 return check_and_freshen_local(oid, freshen) ||
922 check_and_freshen_nonlocal(oid, freshen);
923 }
924
925 int has_loose_object_nonlocal(const struct object_id *oid)
926 {
927 return check_and_freshen_nonlocal(oid, 0);
928 }
929
930 static int has_loose_object(const struct object_id *oid)
931 {
932 return check_and_freshen(oid, 0);
933 }
934
935 static void mmap_limit_check(size_t length)
936 {
937 static size_t limit = 0;
938 if (!limit) {
939 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
940 if (!limit)
941 limit = SIZE_MAX;
942 }
943 if (length > limit)
944 die(_("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX),
945 (uintmax_t)length, (uintmax_t)limit);
946 }
947
948 void *xmmap_gently(void *start, size_t length,
949 int prot, int flags, int fd, off_t offset)
950 {
951 void *ret;
952
953 mmap_limit_check(length);
954 ret = mmap(start, length, prot, flags, fd, offset);
955 if (ret == MAP_FAILED && !length)
956 ret = NULL;
957 return ret;
958 }
959
960 void *xmmap(void *start, size_t length,
961 int prot, int flags, int fd, off_t offset)
962 {
963 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
964 if (ret == MAP_FAILED)
965 die_errno(_("mmap failed"));
966 return ret;
967 }
968
969 /*
970 * With an in-core object data in "map", rehash it to make sure the
971 * object name actually matches "oid" to detect object corruption.
972 * With "map" == NULL, try reading the object named with "oid" using
973 * the streaming interface and rehash it to do the same.
974 */
975 int check_object_signature(const struct object_id *oid, void *map,
976 unsigned long size, const char *type)
977 {
978 struct object_id real_oid;
979 enum object_type obj_type;
980 struct git_istream *st;
981 git_hash_ctx c;
982 char hdr[MAX_HEADER_LEN];
983 int hdrlen;
984
985 if (map) {
986 hash_object_file(map, size, type, &real_oid);
987 return !oideq(oid, &real_oid) ? -1 : 0;
988 }
989
990 st = open_istream(oid, &obj_type, &size, NULL);
991 if (!st)
992 return -1;
993
994 /* Generate the header */
995 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(obj_type), (uintmax_t)size) + 1;
996
997 /* Sha1.. */
998 the_hash_algo->init_fn(&c);
999 the_hash_algo->update_fn(&c, hdr, hdrlen);
1000 for (;;) {
1001 char buf[1024 * 16];
1002 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1003
1004 if (readlen < 0) {
1005 close_istream(st);
1006 return -1;
1007 }
1008 if (!readlen)
1009 break;
1010 the_hash_algo->update_fn(&c, buf, readlen);
1011 }
1012 the_hash_algo->final_fn(real_oid.hash, &c);
1013 close_istream(st);
1014 return !oideq(oid, &real_oid) ? -1 : 0;
1015 }
1016
1017 int git_open_cloexec(const char *name, int flags)
1018 {
1019 int fd;
1020 static int o_cloexec = O_CLOEXEC;
1021
1022 fd = open(name, flags | o_cloexec);
1023 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1024 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1025 o_cloexec &= ~O_CLOEXEC;
1026 fd = open(name, flags | o_cloexec);
1027 }
1028
1029 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1030 {
1031 static int fd_cloexec = FD_CLOEXEC;
1032
1033 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1034 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1035 int flags = fcntl(fd, F_GETFD);
1036 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1037 fd_cloexec = 0;
1038 }
1039 }
1040 #endif
1041 return fd;
1042 }
1043
1044 /*
1045 * Find "oid" as a loose object in the local repository or in an alternate.
1046 * Returns 0 on success, negative on failure.
1047 *
1048 * The "path" out-parameter will give the path of the object we found (if any).
1049 * Note that it may point to static storage and is only valid until another
1050 * call to stat_loose_object().
1051 */
1052 static int stat_loose_object(struct repository *r, const struct object_id *oid,
1053 struct stat *st, const char **path)
1054 {
1055 struct object_directory *odb;
1056 static struct strbuf buf = STRBUF_INIT;
1057
1058 prepare_alt_odb(r);
1059 for (odb = r->objects->odb; odb; odb = odb->next) {
1060 *path = odb_loose_path(odb, &buf, oid);
1061 if (!lstat(*path, st))
1062 return 0;
1063 }
1064
1065 return -1;
1066 }
1067
1068 /*
1069 * Like stat_loose_object(), but actually open the object and return the
1070 * descriptor. See the caveats on the "path" parameter above.
1071 */
1072 static int open_loose_object(struct repository *r,
1073 const struct object_id *oid, const char **path)
1074 {
1075 int fd;
1076 struct object_directory *odb;
1077 int most_interesting_errno = ENOENT;
1078 static struct strbuf buf = STRBUF_INIT;
1079
1080 prepare_alt_odb(r);
1081 for (odb = r->objects->odb; odb; odb = odb->next) {
1082 *path = odb_loose_path(odb, &buf, oid);
1083 fd = git_open(*path);
1084 if (fd >= 0)
1085 return fd;
1086
1087 if (most_interesting_errno == ENOENT)
1088 most_interesting_errno = errno;
1089 }
1090 errno = most_interesting_errno;
1091 return -1;
1092 }
1093
1094 static int quick_has_loose(struct repository *r,
1095 const struct object_id *oid)
1096 {
1097 struct object_directory *odb;
1098
1099 prepare_alt_odb(r);
1100 for (odb = r->objects->odb; odb; odb = odb->next) {
1101 if (oid_array_lookup(odb_loose_cache(odb, oid), oid) >= 0)
1102 return 1;
1103 }
1104 return 0;
1105 }
1106
1107 /*
1108 * Map the loose object at "path" if it is not NULL, or the path found by
1109 * searching for a loose object named "oid".
1110 */
1111 static void *map_loose_object_1(struct repository *r, const char *path,
1112 const struct object_id *oid, unsigned long *size)
1113 {
1114 void *map;
1115 int fd;
1116
1117 if (path)
1118 fd = git_open(path);
1119 else
1120 fd = open_loose_object(r, oid, &path);
1121 map = NULL;
1122 if (fd >= 0) {
1123 struct stat st;
1124
1125 if (!fstat(fd, &st)) {
1126 *size = xsize_t(st.st_size);
1127 if (!*size) {
1128 /* mmap() is forbidden on empty files */
1129 error(_("object file %s is empty"), path);
1130 close(fd);
1131 return NULL;
1132 }
1133 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1134 }
1135 close(fd);
1136 }
1137 return map;
1138 }
1139
1140 void *map_loose_object(struct repository *r,
1141 const struct object_id *oid,
1142 unsigned long *size)
1143 {
1144 return map_loose_object_1(r, NULL, oid, size);
1145 }
1146
1147 static int unpack_loose_short_header(git_zstream *stream,
1148 unsigned char *map, unsigned long mapsize,
1149 void *buffer, unsigned long bufsiz)
1150 {
1151 /* Get the data stream */
1152 memset(stream, 0, sizeof(*stream));
1153 stream->next_in = map;
1154 stream->avail_in = mapsize;
1155 stream->next_out = buffer;
1156 stream->avail_out = bufsiz;
1157
1158 git_inflate_init(stream);
1159 return git_inflate(stream, 0);
1160 }
1161
1162 int unpack_loose_header(git_zstream *stream,
1163 unsigned char *map, unsigned long mapsize,
1164 void *buffer, unsigned long bufsiz)
1165 {
1166 int status = unpack_loose_short_header(stream, map, mapsize,
1167 buffer, bufsiz);
1168
1169 if (status < Z_OK)
1170 return status;
1171
1172 /* Make sure we have the terminating NUL */
1173 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1174 return -1;
1175 return 0;
1176 }
1177
1178 static int unpack_loose_header_to_strbuf(git_zstream *stream, unsigned char *map,
1179 unsigned long mapsize, void *buffer,
1180 unsigned long bufsiz, struct strbuf *header)
1181 {
1182 int status;
1183
1184 status = unpack_loose_short_header(stream, map, mapsize, buffer, bufsiz);
1185 if (status < Z_OK)
1186 return -1;
1187
1188 /*
1189 * Check if entire header is unpacked in the first iteration.
1190 */
1191 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1192 return 0;
1193
1194 /*
1195 * buffer[0..bufsiz] was not large enough. Copy the partial
1196 * result out to header, and then append the result of further
1197 * reading the stream.
1198 */
1199 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1200 stream->next_out = buffer;
1201 stream->avail_out = bufsiz;
1202
1203 do {
1204 status = git_inflate(stream, 0);
1205 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1206 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1207 return 0;
1208 stream->next_out = buffer;
1209 stream->avail_out = bufsiz;
1210 } while (status != Z_STREAM_END);
1211 return -1;
1212 }
1213
1214 static void *unpack_loose_rest(git_zstream *stream,
1215 void *buffer, unsigned long size,
1216 const struct object_id *oid)
1217 {
1218 int bytes = strlen(buffer) + 1;
1219 unsigned char *buf = xmallocz(size);
1220 unsigned long n;
1221 int status = Z_OK;
1222
1223 n = stream->total_out - bytes;
1224 if (n > size)
1225 n = size;
1226 memcpy(buf, (char *) buffer + bytes, n);
1227 bytes = n;
1228 if (bytes <= size) {
1229 /*
1230 * The above condition must be (bytes <= size), not
1231 * (bytes < size). In other words, even though we
1232 * expect no more output and set avail_out to zero,
1233 * the input zlib stream may have bytes that express
1234 * "this concludes the stream", and we *do* want to
1235 * eat that input.
1236 *
1237 * Otherwise we would not be able to test that we
1238 * consumed all the input to reach the expected size;
1239 * we also want to check that zlib tells us that all
1240 * went well with status == Z_STREAM_END at the end.
1241 */
1242 stream->next_out = buf + bytes;
1243 stream->avail_out = size - bytes;
1244 while (status == Z_OK)
1245 status = git_inflate(stream, Z_FINISH);
1246 }
1247 if (status == Z_STREAM_END && !stream->avail_in) {
1248 git_inflate_end(stream);
1249 return buf;
1250 }
1251
1252 if (status < 0)
1253 error(_("corrupt loose object '%s'"), oid_to_hex(oid));
1254 else if (stream->avail_in)
1255 error(_("garbage at end of loose object '%s'"),
1256 oid_to_hex(oid));
1257 free(buf);
1258 return NULL;
1259 }
1260
1261 /*
1262 * We used to just use "sscanf()", but that's actually way
1263 * too permissive for what we want to check. So do an anal
1264 * object header parse by hand.
1265 */
1266 static int parse_loose_header_extended(const char *hdr, struct object_info *oi,
1267 unsigned int flags)
1268 {
1269 const char *type_buf = hdr;
1270 unsigned long size;
1271 int type, type_len = 0;
1272
1273 /*
1274 * The type can be of any size but is followed by
1275 * a space.
1276 */
1277 for (;;) {
1278 char c = *hdr++;
1279 if (!c)
1280 return -1;
1281 if (c == ' ')
1282 break;
1283 type_len++;
1284 }
1285
1286 type = type_from_string_gently(type_buf, type_len, 1);
1287 if (oi->type_name)
1288 strbuf_add(oi->type_name, type_buf, type_len);
1289 /*
1290 * Set type to 0 if its an unknown object and
1291 * we're obtaining the type using '--allow-unknown-type'
1292 * option.
1293 */
1294 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1295 type = 0;
1296 else if (type < 0)
1297 die(_("invalid object type"));
1298 if (oi->typep)
1299 *oi->typep = type;
1300
1301 /*
1302 * The length must follow immediately, and be in canonical
1303 * decimal format (ie "010" is not valid).
1304 */
1305 size = *hdr++ - '0';
1306 if (size > 9)
1307 return -1;
1308 if (size) {
1309 for (;;) {
1310 unsigned long c = *hdr - '0';
1311 if (c > 9)
1312 break;
1313 hdr++;
1314 size = size * 10 + c;
1315 }
1316 }
1317
1318 if (oi->sizep)
1319 *oi->sizep = size;
1320
1321 /*
1322 * The length must be followed by a zero byte
1323 */
1324 return *hdr ? -1 : type;
1325 }
1326
1327 int parse_loose_header(const char *hdr, unsigned long *sizep)
1328 {
1329 struct object_info oi = OBJECT_INFO_INIT;
1330
1331 oi.sizep = sizep;
1332 return parse_loose_header_extended(hdr, &oi, 0);
1333 }
1334
1335 static int loose_object_info(struct repository *r,
1336 const struct object_id *oid,
1337 struct object_info *oi, int flags)
1338 {
1339 int status = 0;
1340 unsigned long mapsize;
1341 void *map;
1342 git_zstream stream;
1343 char hdr[MAX_HEADER_LEN];
1344 struct strbuf hdrbuf = STRBUF_INIT;
1345 unsigned long size_scratch;
1346
1347 if (oi->delta_base_sha1)
1348 hashclr(oi->delta_base_sha1);
1349
1350 /*
1351 * If we don't care about type or size, then we don't
1352 * need to look inside the object at all. Note that we
1353 * do not optimize out the stat call, even if the
1354 * caller doesn't care about the disk-size, since our
1355 * return value implicitly indicates whether the
1356 * object even exists.
1357 */
1358 if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1359 const char *path;
1360 struct stat st;
1361 if (!oi->disk_sizep && (flags & OBJECT_INFO_QUICK))
1362 return quick_has_loose(r, oid) ? 0 : -1;
1363 if (stat_loose_object(r, oid, &st, &path) < 0)
1364 return -1;
1365 if (oi->disk_sizep)
1366 *oi->disk_sizep = st.st_size;
1367 return 0;
1368 }
1369
1370 map = map_loose_object(r, oid, &mapsize);
1371 if (!map)
1372 return -1;
1373
1374 if (!oi->sizep)
1375 oi->sizep = &size_scratch;
1376
1377 if (oi->disk_sizep)
1378 *oi->disk_sizep = mapsize;
1379 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1380 if (unpack_loose_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1381 status = error(_("unable to unpack %s header with --allow-unknown-type"),
1382 oid_to_hex(oid));
1383 } else if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1384 status = error(_("unable to unpack %s header"),
1385 oid_to_hex(oid));
1386 if (status < 0)
1387 ; /* Do nothing */
1388 else if (hdrbuf.len) {
1389 if ((status = parse_loose_header_extended(hdrbuf.buf, oi, flags)) < 0)
1390 status = error(_("unable to parse %s header with --allow-unknown-type"),
1391 oid_to_hex(oid));
1392 } else if ((status = parse_loose_header_extended(hdr, oi, flags)) < 0)
1393 status = error(_("unable to parse %s header"), oid_to_hex(oid));
1394
1395 if (status >= 0 && oi->contentp) {
1396 *oi->contentp = unpack_loose_rest(&stream, hdr,
1397 *oi->sizep, oid);
1398 if (!*oi->contentp) {
1399 git_inflate_end(&stream);
1400 status = -1;
1401 }
1402 } else
1403 git_inflate_end(&stream);
1404
1405 munmap(map, mapsize);
1406 if (status && oi->typep)
1407 *oi->typep = status;
1408 if (oi->sizep == &size_scratch)
1409 oi->sizep = NULL;
1410 strbuf_release(&hdrbuf);
1411 oi->whence = OI_LOOSE;
1412 return (status < 0) ? status : 0;
1413 }
1414
1415 int fetch_if_missing = 1;
1416
1417 int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1418 struct object_info *oi, unsigned flags)
1419 {
1420 static struct object_info blank_oi = OBJECT_INFO_INIT;
1421 struct pack_entry e;
1422 int rtype;
1423 const struct object_id *real = oid;
1424 int already_retried = 0;
1425
1426 if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1427 real = lookup_replace_object(r, oid);
1428
1429 if (is_null_oid(real))
1430 return -1;
1431
1432 if (!oi)
1433 oi = &blank_oi;
1434
1435 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1436 struct cached_object *co = find_cached_object(real);
1437 if (co) {
1438 if (oi->typep)
1439 *(oi->typep) = co->type;
1440 if (oi->sizep)
1441 *(oi->sizep) = co->size;
1442 if (oi->disk_sizep)
1443 *(oi->disk_sizep) = 0;
1444 if (oi->delta_base_sha1)
1445 hashclr(oi->delta_base_sha1);
1446 if (oi->type_name)
1447 strbuf_addstr(oi->type_name, type_name(co->type));
1448 if (oi->contentp)
1449 *oi->contentp = xmemdupz(co->buf, co->size);
1450 oi->whence = OI_CACHED;
1451 return 0;
1452 }
1453 }
1454
1455 while (1) {
1456 if (find_pack_entry(r, real, &e))
1457 break;
1458
1459 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1460 return -1;
1461
1462 /* Most likely it's a loose object. */
1463 if (!loose_object_info(r, real, oi, flags))
1464 return 0;
1465
1466 /* Not a loose object; someone else may have just packed it. */
1467 if (!(flags & OBJECT_INFO_QUICK)) {
1468 reprepare_packed_git(r);
1469 if (find_pack_entry(r, real, &e))
1470 break;
1471 }
1472
1473 /* Check if it is a missing object */
1474 if (fetch_if_missing && has_promisor_remote() &&
1475 !already_retried && r == the_repository &&
1476 !(flags & OBJECT_INFO_SKIP_FETCH_OBJECT)) {
1477 /*
1478 * TODO Investigate checking promisor_remote_get_direct()
1479 * TODO return value and stopping on error here.
1480 * TODO Pass a repository struct through
1481 * promisor_remote_get_direct(), such that arbitrary
1482 * repositories work.
1483 */
1484 promisor_remote_get_direct(r, real, 1);
1485 already_retried = 1;
1486 continue;
1487 }
1488
1489 return -1;
1490 }
1491
1492 if (oi == &blank_oi)
1493 /*
1494 * We know that the caller doesn't actually need the
1495 * information below, so return early.
1496 */
1497 return 0;
1498 rtype = packed_object_info(r, e.p, e.offset, oi);
1499 if (rtype < 0) {
1500 mark_bad_packed_object(e.p, real->hash);
1501 return oid_object_info_extended(r, real, oi, 0);
1502 } else if (oi->whence == OI_PACKED) {
1503 oi->u.packed.offset = e.offset;
1504 oi->u.packed.pack = e.p;
1505 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1506 rtype == OBJ_OFS_DELTA);
1507 }
1508
1509 return 0;
1510 }
1511
1512 /* returns enum object_type or negative */
1513 int oid_object_info(struct repository *r,
1514 const struct object_id *oid,
1515 unsigned long *sizep)
1516 {
1517 enum object_type type;
1518 struct object_info oi = OBJECT_INFO_INIT;
1519
1520 oi.typep = &type;
1521 oi.sizep = sizep;
1522 if (oid_object_info_extended(r, oid, &oi,
1523 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1524 return -1;
1525 return type;
1526 }
1527
1528 static void *read_object(struct repository *r,
1529 const struct object_id *oid, enum object_type *type,
1530 unsigned long *size)
1531 {
1532 struct object_info oi = OBJECT_INFO_INIT;
1533 void *content;
1534 oi.typep = type;
1535 oi.sizep = size;
1536 oi.contentp = &content;
1537
1538 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1539 return NULL;
1540 return content;
1541 }
1542
1543 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1544 struct object_id *oid)
1545 {
1546 struct cached_object *co;
1547
1548 hash_object_file(buf, len, type_name(type), oid);
1549 if (has_object_file(oid) || find_cached_object(oid))
1550 return 0;
1551 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1552 co = &cached_objects[cached_object_nr++];
1553 co->size = len;
1554 co->type = type;
1555 co->buf = xmalloc(len);
1556 memcpy(co->buf, buf, len);
1557 oidcpy(&co->oid, oid);
1558 return 0;
1559 }
1560
1561 /*
1562 * This function dies on corrupt objects; the callers who want to
1563 * deal with them should arrange to call read_object() and give error
1564 * messages themselves.
1565 */
1566 void *read_object_file_extended(struct repository *r,
1567 const struct object_id *oid,
1568 enum object_type *type,
1569 unsigned long *size,
1570 int lookup_replace)
1571 {
1572 void *data;
1573 const struct packed_git *p;
1574 const char *path;
1575 struct stat st;
1576 const struct object_id *repl = lookup_replace ?
1577 lookup_replace_object(r, oid) : oid;
1578
1579 errno = 0;
1580 data = read_object(r, repl, type, size);
1581 if (data)
1582 return data;
1583
1584 if (errno && errno != ENOENT)
1585 die_errno(_("failed to read object %s"), oid_to_hex(oid));
1586
1587 /* die if we replaced an object with one that does not exist */
1588 if (repl != oid)
1589 die(_("replacement %s not found for %s"),
1590 oid_to_hex(repl), oid_to_hex(oid));
1591
1592 if (!stat_loose_object(r, repl, &st, &path))
1593 die(_("loose object %s (stored in %s) is corrupt"),
1594 oid_to_hex(repl), path);
1595
1596 if ((p = has_packed_and_bad(r, repl->hash)) != NULL)
1597 die(_("packed object %s (stored in %s) is corrupt"),
1598 oid_to_hex(repl), p->pack_name);
1599
1600 return NULL;
1601 }
1602
1603 void *read_object_with_reference(struct repository *r,
1604 const struct object_id *oid,
1605 const char *required_type_name,
1606 unsigned long *size,
1607 struct object_id *actual_oid_return)
1608 {
1609 enum object_type type, required_type;
1610 void *buffer;
1611 unsigned long isize;
1612 struct object_id actual_oid;
1613
1614 required_type = type_from_string(required_type_name);
1615 oidcpy(&actual_oid, oid);
1616 while (1) {
1617 int ref_length = -1;
1618 const char *ref_type = NULL;
1619
1620 buffer = repo_read_object_file(r, &actual_oid, &type, &isize);
1621 if (!buffer)
1622 return NULL;
1623 if (type == required_type) {
1624 *size = isize;
1625 if (actual_oid_return)
1626 oidcpy(actual_oid_return, &actual_oid);
1627 return buffer;
1628 }
1629 /* Handle references */
1630 else if (type == OBJ_COMMIT)
1631 ref_type = "tree ";
1632 else if (type == OBJ_TAG)
1633 ref_type = "object ";
1634 else {
1635 free(buffer);
1636 return NULL;
1637 }
1638 ref_length = strlen(ref_type);
1639
1640 if (ref_length + the_hash_algo->hexsz > isize ||
1641 memcmp(buffer, ref_type, ref_length) ||
1642 get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1643 free(buffer);
1644 return NULL;
1645 }
1646 free(buffer);
1647 /* Now we have the ID of the referred-to object in
1648 * actual_oid. Check again. */
1649 }
1650 }
1651
1652 static void write_object_file_prepare(const void *buf, unsigned long len,
1653 const char *type, struct object_id *oid,
1654 char *hdr, int *hdrlen)
1655 {
1656 git_hash_ctx c;
1657
1658 /* Generate the header */
1659 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %"PRIuMAX , type, (uintmax_t)len)+1;
1660
1661 /* Sha1.. */
1662 the_hash_algo->init_fn(&c);
1663 the_hash_algo->update_fn(&c, hdr, *hdrlen);
1664 the_hash_algo->update_fn(&c, buf, len);
1665 the_hash_algo->final_fn(oid->hash, &c);
1666 }
1667
1668 /*
1669 * Move the just written object into its final resting place.
1670 */
1671 int finalize_object_file(const char *tmpfile, const char *filename)
1672 {
1673 int ret = 0;
1674
1675 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1676 goto try_rename;
1677 else if (link(tmpfile, filename))
1678 ret = errno;
1679
1680 /*
1681 * Coda hack - coda doesn't like cross-directory links,
1682 * so we fall back to a rename, which will mean that it
1683 * won't be able to check collisions, but that's not a
1684 * big deal.
1685 *
1686 * The same holds for FAT formatted media.
1687 *
1688 * When this succeeds, we just return. We have nothing
1689 * left to unlink.
1690 */
1691 if (ret && ret != EEXIST) {
1692 try_rename:
1693 if (!rename(tmpfile, filename))
1694 goto out;
1695 ret = errno;
1696 }
1697 unlink_or_warn(tmpfile);
1698 if (ret) {
1699 if (ret != EEXIST) {
1700 return error_errno(_("unable to write file %s"), filename);
1701 }
1702 /* FIXME!!! Collision check here ? */
1703 }
1704
1705 out:
1706 if (adjust_shared_perm(filename))
1707 return error(_("unable to set permission to '%s'"), filename);
1708 return 0;
1709 }
1710
1711 static int write_buffer(int fd, const void *buf, size_t len)
1712 {
1713 if (write_in_full(fd, buf, len) < 0)
1714 return error_errno(_("file write error"));
1715 return 0;
1716 }
1717
1718 int hash_object_file(const void *buf, unsigned long len, const char *type,
1719 struct object_id *oid)
1720 {
1721 char hdr[MAX_HEADER_LEN];
1722 int hdrlen = sizeof(hdr);
1723 write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1724 return 0;
1725 }
1726
1727 /* Finalize a file on disk, and close it. */
1728 static void close_loose_object(int fd)
1729 {
1730 if (fsync_object_files)
1731 fsync_or_die(fd, "loose object file");
1732 if (close(fd) != 0)
1733 die_errno(_("error when closing loose object file"));
1734 }
1735
1736 /* Size of directory component, including the ending '/' */
1737 static inline int directory_size(const char *filename)
1738 {
1739 const char *s = strrchr(filename, '/');
1740 if (!s)
1741 return 0;
1742 return s - filename + 1;
1743 }
1744
1745 /*
1746 * This creates a temporary file in the same directory as the final
1747 * 'filename'
1748 *
1749 * We want to avoid cross-directory filename renames, because those
1750 * can have problems on various filesystems (FAT, NFS, Coda).
1751 */
1752 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1753 {
1754 int fd, dirlen = directory_size(filename);
1755
1756 strbuf_reset(tmp);
1757 strbuf_add(tmp, filename, dirlen);
1758 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1759 fd = git_mkstemp_mode(tmp->buf, 0444);
1760 if (fd < 0 && dirlen && errno == ENOENT) {
1761 /*
1762 * Make sure the directory exists; note that the contents
1763 * of the buffer are undefined after mkstemp returns an
1764 * error, so we have to rewrite the whole buffer from
1765 * scratch.
1766 */
1767 strbuf_reset(tmp);
1768 strbuf_add(tmp, filename, dirlen - 1);
1769 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1770 return -1;
1771 if (adjust_shared_perm(tmp->buf))
1772 return -1;
1773
1774 /* Try again */
1775 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1776 fd = git_mkstemp_mode(tmp->buf, 0444);
1777 }
1778 return fd;
1779 }
1780
1781 static int write_loose_object(const struct object_id *oid, char *hdr,
1782 int hdrlen, const void *buf, unsigned long len,
1783 time_t mtime)
1784 {
1785 int fd, ret;
1786 unsigned char compressed[4096];
1787 git_zstream stream;
1788 git_hash_ctx c;
1789 struct object_id parano_oid;
1790 static struct strbuf tmp_file = STRBUF_INIT;
1791 static struct strbuf filename = STRBUF_INIT;
1792
1793 loose_object_path(the_repository, &filename, oid);
1794
1795 fd = create_tmpfile(&tmp_file, filename.buf);
1796 if (fd < 0) {
1797 if (errno == EACCES)
1798 return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
1799 else
1800 return error_errno(_("unable to create temporary file"));
1801 }
1802
1803 /* Set it up */
1804 git_deflate_init(&stream, zlib_compression_level);
1805 stream.next_out = compressed;
1806 stream.avail_out = sizeof(compressed);
1807 the_hash_algo->init_fn(&c);
1808
1809 /* First header.. */
1810 stream.next_in = (unsigned char *)hdr;
1811 stream.avail_in = hdrlen;
1812 while (git_deflate(&stream, 0) == Z_OK)
1813 ; /* nothing */
1814 the_hash_algo->update_fn(&c, hdr, hdrlen);
1815
1816 /* Then the data itself.. */
1817 stream.next_in = (void *)buf;
1818 stream.avail_in = len;
1819 do {
1820 unsigned char *in0 = stream.next_in;
1821 ret = git_deflate(&stream, Z_FINISH);
1822 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1823 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1824 die(_("unable to write loose object file"));
1825 stream.next_out = compressed;
1826 stream.avail_out = sizeof(compressed);
1827 } while (ret == Z_OK);
1828
1829 if (ret != Z_STREAM_END)
1830 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
1831 ret);
1832 ret = git_deflate_end_gently(&stream);
1833 if (ret != Z_OK)
1834 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
1835 ret);
1836 the_hash_algo->final_fn(parano_oid.hash, &c);
1837 if (!oideq(oid, &parano_oid))
1838 die(_("confused by unstable object source data for %s"),
1839 oid_to_hex(oid));
1840
1841 close_loose_object(fd);
1842
1843 if (mtime) {
1844 struct utimbuf utb;
1845 utb.actime = mtime;
1846 utb.modtime = mtime;
1847 if (utime(tmp_file.buf, &utb) < 0)
1848 warning_errno(_("failed utime() on %s"), tmp_file.buf);
1849 }
1850
1851 return finalize_object_file(tmp_file.buf, filename.buf);
1852 }
1853
1854 static int freshen_loose_object(const struct object_id *oid)
1855 {
1856 return check_and_freshen(oid, 1);
1857 }
1858
1859 static int freshen_packed_object(const struct object_id *oid)
1860 {
1861 struct pack_entry e;
1862 if (!find_pack_entry(the_repository, oid, &e))
1863 return 0;
1864 if (e.p->freshened)
1865 return 1;
1866 if (!freshen_file(e.p->pack_name))
1867 return 0;
1868 e.p->freshened = 1;
1869 return 1;
1870 }
1871
1872 int write_object_file(const void *buf, unsigned long len, const char *type,
1873 struct object_id *oid)
1874 {
1875 char hdr[MAX_HEADER_LEN];
1876 int hdrlen = sizeof(hdr);
1877
1878 /* Normally if we have it in the pack then we do not bother writing
1879 * it out into .git/objects/??/?{38} file.
1880 */
1881 write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1882 if (freshen_packed_object(oid) || freshen_loose_object(oid))
1883 return 0;
1884 return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
1885 }
1886
1887 int hash_object_file_literally(const void *buf, unsigned long len,
1888 const char *type, struct object_id *oid,
1889 unsigned flags)
1890 {
1891 char *header;
1892 int hdrlen, status = 0;
1893
1894 /* type string, SP, %lu of the length plus NUL must fit this */
1895 hdrlen = strlen(type) + MAX_HEADER_LEN;
1896 header = xmalloc(hdrlen);
1897 write_object_file_prepare(buf, len, type, oid, header, &hdrlen);
1898
1899 if (!(flags & HASH_WRITE_OBJECT))
1900 goto cleanup;
1901 if (freshen_packed_object(oid) || freshen_loose_object(oid))
1902 goto cleanup;
1903 status = write_loose_object(oid, header, hdrlen, buf, len, 0);
1904
1905 cleanup:
1906 free(header);
1907 return status;
1908 }
1909
1910 int force_object_loose(const struct object_id *oid, time_t mtime)
1911 {
1912 void *buf;
1913 unsigned long len;
1914 enum object_type type;
1915 char hdr[MAX_HEADER_LEN];
1916 int hdrlen;
1917 int ret;
1918
1919 if (has_loose_object(oid))
1920 return 0;
1921 buf = read_object(the_repository, oid, &type, &len);
1922 if (!buf)
1923 return error(_("cannot read object for %s"), oid_to_hex(oid));
1924 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(type), (uintmax_t)len) + 1;
1925 ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
1926 free(buf);
1927
1928 return ret;
1929 }
1930
1931 int repo_has_object_file_with_flags(struct repository *r,
1932 const struct object_id *oid, int flags)
1933 {
1934 if (!startup_info->have_repository)
1935 return 0;
1936 return oid_object_info_extended(r, oid, NULL,
1937 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1938 }
1939
1940 int repo_has_object_file(struct repository *r,
1941 const struct object_id *oid)
1942 {
1943 return repo_has_object_file_with_flags(r, oid, 0);
1944 }
1945
1946 static void check_tree(const void *buf, size_t size)
1947 {
1948 struct tree_desc desc;
1949 struct name_entry entry;
1950
1951 init_tree_desc(&desc, buf, size);
1952 while (tree_entry(&desc, &entry))
1953 /* do nothing
1954 * tree_entry() will die() on malformed entries */
1955 ;
1956 }
1957
1958 static void check_commit(const void *buf, size_t size)
1959 {
1960 struct commit c;
1961 memset(&c, 0, sizeof(c));
1962 if (parse_commit_buffer(the_repository, &c, buf, size, 0))
1963 die(_("corrupt commit"));
1964 }
1965
1966 static void check_tag(const void *buf, size_t size)
1967 {
1968 struct tag t;
1969 memset(&t, 0, sizeof(t));
1970 if (parse_tag_buffer(the_repository, &t, buf, size))
1971 die(_("corrupt tag"));
1972 }
1973
1974 static int index_mem(struct index_state *istate,
1975 struct object_id *oid, void *buf, size_t size,
1976 enum object_type type,
1977 const char *path, unsigned flags)
1978 {
1979 int ret, re_allocated = 0;
1980 int write_object = flags & HASH_WRITE_OBJECT;
1981
1982 if (!type)
1983 type = OBJ_BLOB;
1984
1985 /*
1986 * Convert blobs to git internal format
1987 */
1988 if ((type == OBJ_BLOB) && path) {
1989 struct strbuf nbuf = STRBUF_INIT;
1990 if (convert_to_git(istate, path, buf, size, &nbuf,
1991 get_conv_flags(flags))) {
1992 buf = strbuf_detach(&nbuf, &size);
1993 re_allocated = 1;
1994 }
1995 }
1996 if (flags & HASH_FORMAT_CHECK) {
1997 if (type == OBJ_TREE)
1998 check_tree(buf, size);
1999 if (type == OBJ_COMMIT)
2000 check_commit(buf, size);
2001 if (type == OBJ_TAG)
2002 check_tag(buf, size);
2003 }
2004
2005 if (write_object)
2006 ret = write_object_file(buf, size, type_name(type), oid);
2007 else
2008 ret = hash_object_file(buf, size, type_name(type), oid);
2009 if (re_allocated)
2010 free(buf);
2011 return ret;
2012 }
2013
2014 static int index_stream_convert_blob(struct index_state *istate,
2015 struct object_id *oid,
2016 int fd,
2017 const char *path,
2018 unsigned flags)
2019 {
2020 int ret;
2021 const int write_object = flags & HASH_WRITE_OBJECT;
2022 struct strbuf sbuf = STRBUF_INIT;
2023
2024 assert(path);
2025 assert(would_convert_to_git_filter_fd(istate, path));
2026
2027 convert_to_git_filter_fd(istate, path, fd, &sbuf,
2028 get_conv_flags(flags));
2029
2030 if (write_object)
2031 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
2032 oid);
2033 else
2034 ret = hash_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
2035 oid);
2036 strbuf_release(&sbuf);
2037 return ret;
2038 }
2039
2040 static int index_pipe(struct index_state *istate, struct object_id *oid,
2041 int fd, enum object_type type,
2042 const char *path, unsigned flags)
2043 {
2044 struct strbuf sbuf = STRBUF_INIT;
2045 int ret;
2046
2047 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2048 ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags);
2049 else
2050 ret = -1;
2051 strbuf_release(&sbuf);
2052 return ret;
2053 }
2054
2055 #define SMALL_FILE_SIZE (32*1024)
2056
2057 static int index_core(struct index_state *istate,
2058 struct object_id *oid, int fd, size_t size,
2059 enum object_type type, const char *path,
2060 unsigned flags)
2061 {
2062 int ret;
2063
2064 if (!size) {
2065 ret = index_mem(istate, oid, "", size, type, path, flags);
2066 } else if (size <= SMALL_FILE_SIZE) {
2067 char *buf = xmalloc(size);
2068 ssize_t read_result = read_in_full(fd, buf, size);
2069 if (read_result < 0)
2070 ret = error_errno(_("read error while indexing %s"),
2071 path ? path : "<unknown>");
2072 else if (read_result != size)
2073 ret = error(_("short read while indexing %s"),
2074 path ? path : "<unknown>");
2075 else
2076 ret = index_mem(istate, oid, buf, size, type, path, flags);
2077 free(buf);
2078 } else {
2079 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2080 ret = index_mem(istate, oid, buf, size, type, path, flags);
2081 munmap(buf, size);
2082 }
2083 return ret;
2084 }
2085
2086 /*
2087 * This creates one packfile per large blob unless bulk-checkin
2088 * machinery is "plugged".
2089 *
2090 * This also bypasses the usual "convert-to-git" dance, and that is on
2091 * purpose. We could write a streaming version of the converting
2092 * functions and insert that before feeding the data to fast-import
2093 * (or equivalent in-core API described above). However, that is
2094 * somewhat complicated, as we do not know the size of the filter
2095 * result, which we need to know beforehand when writing a git object.
2096 * Since the primary motivation for trying to stream from the working
2097 * tree file and to avoid mmaping it in core is to deal with large
2098 * binary blobs, they generally do not want to get any conversion, and
2099 * callers should avoid this code path when filters are requested.
2100 */
2101 static int index_stream(struct object_id *oid, int fd, size_t size,
2102 enum object_type type, const char *path,
2103 unsigned flags)
2104 {
2105 return index_bulk_checkin(oid, fd, size, type, path, flags);
2106 }
2107
2108 int index_fd(struct index_state *istate, struct object_id *oid,
2109 int fd, struct stat *st,
2110 enum object_type type, const char *path, unsigned flags)
2111 {
2112 int ret;
2113
2114 /*
2115 * Call xsize_t() only when needed to avoid potentially unnecessary
2116 * die() for large files.
2117 */
2118 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, path))
2119 ret = index_stream_convert_blob(istate, oid, fd, path, flags);
2120 else if (!S_ISREG(st->st_mode))
2121 ret = index_pipe(istate, oid, fd, type, path, flags);
2122 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
2123 (path && would_convert_to_git(istate, path)))
2124 ret = index_core(istate, oid, fd, xsize_t(st->st_size),
2125 type, path, flags);
2126 else
2127 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
2128 flags);
2129 close(fd);
2130 return ret;
2131 }
2132
2133 int index_path(struct index_state *istate, struct object_id *oid,
2134 const char *path, struct stat *st, unsigned flags)
2135 {
2136 int fd;
2137 struct strbuf sb = STRBUF_INIT;
2138 int rc = 0;
2139
2140 switch (st->st_mode & S_IFMT) {
2141 case S_IFREG:
2142 fd = open(path, O_RDONLY);
2143 if (fd < 0)
2144 return error_errno("open(\"%s\")", path);
2145 if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0)
2146 return error(_("%s: failed to insert into database"),
2147 path);
2148 break;
2149 case S_IFLNK:
2150 if (strbuf_readlink(&sb, path, st->st_size))
2151 return error_errno("readlink(\"%s\")", path);
2152 if (!(flags & HASH_WRITE_OBJECT))
2153 hash_object_file(sb.buf, sb.len, blob_type, oid);
2154 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
2155 rc = error(_("%s: failed to insert into database"), path);
2156 strbuf_release(&sb);
2157 break;
2158 case S_IFDIR:
2159 return resolve_gitlink_ref(path, "HEAD", oid);
2160 default:
2161 return error(_("%s: unsupported file type"), path);
2162 }
2163 return rc;
2164 }
2165
2166 int read_pack_header(int fd, struct pack_header *header)
2167 {
2168 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
2169 /* "eof before pack header was fully read" */
2170 return PH_ERROR_EOF;
2171
2172 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2173 /* "protocol error (pack signature mismatch detected)" */
2174 return PH_ERROR_PACK_SIGNATURE;
2175 if (!pack_version_ok(header->hdr_version))
2176 /* "protocol error (pack version unsupported)" */
2177 return PH_ERROR_PROTOCOL;
2178 return 0;
2179 }
2180
2181 void assert_oid_type(const struct object_id *oid, enum object_type expect)
2182 {
2183 enum object_type type = oid_object_info(the_repository, oid, NULL);
2184 if (type < 0)
2185 die(_("%s is not a valid object"), oid_to_hex(oid));
2186 if (type != expect)
2187 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
2188 type_name(expect));
2189 }
2190
2191 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2192 struct strbuf *path,
2193 each_loose_object_fn obj_cb,
2194 each_loose_cruft_fn cruft_cb,
2195 each_loose_subdir_fn subdir_cb,
2196 void *data)
2197 {
2198 size_t origlen, baselen;
2199 DIR *dir;
2200 struct dirent *de;
2201 int r = 0;
2202 struct object_id oid;
2203
2204 if (subdir_nr > 0xff)
2205 BUG("invalid loose object subdirectory: %x", subdir_nr);
2206
2207 origlen = path->len;
2208 strbuf_complete(path, '/');
2209 strbuf_addf(path, "%02x", subdir_nr);
2210
2211 dir = opendir(path->buf);
2212 if (!dir) {
2213 if (errno != ENOENT)
2214 r = error_errno(_("unable to open %s"), path->buf);
2215 strbuf_setlen(path, origlen);
2216 return r;
2217 }
2218
2219 oid.hash[0] = subdir_nr;
2220 strbuf_addch(path, '/');
2221 baselen = path->len;
2222
2223 while ((de = readdir(dir))) {
2224 size_t namelen;
2225 if (is_dot_or_dotdot(de->d_name))
2226 continue;
2227
2228 namelen = strlen(de->d_name);
2229 strbuf_setlen(path, baselen);
2230 strbuf_add(path, de->d_name, namelen);
2231 if (namelen == the_hash_algo->hexsz - 2 &&
2232 !hex_to_bytes(oid.hash + 1, de->d_name,
2233 the_hash_algo->rawsz - 1)) {
2234 if (obj_cb) {
2235 r = obj_cb(&oid, path->buf, data);
2236 if (r)
2237 break;
2238 }
2239 continue;
2240 }
2241
2242 if (cruft_cb) {
2243 r = cruft_cb(de->d_name, path->buf, data);
2244 if (r)
2245 break;
2246 }
2247 }
2248 closedir(dir);
2249
2250 strbuf_setlen(path, baselen - 1);
2251 if (!r && subdir_cb)
2252 r = subdir_cb(subdir_nr, path->buf, data);
2253
2254 strbuf_setlen(path, origlen);
2255
2256 return r;
2257 }
2258
2259 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2260 each_loose_object_fn obj_cb,
2261 each_loose_cruft_fn cruft_cb,
2262 each_loose_subdir_fn subdir_cb,
2263 void *data)
2264 {
2265 int r = 0;
2266 int i;
2267
2268 for (i = 0; i < 256; i++) {
2269 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2270 subdir_cb, data);
2271 if (r)
2272 break;
2273 }
2274
2275 return r;
2276 }
2277
2278 int for_each_loose_file_in_objdir(const char *path,
2279 each_loose_object_fn obj_cb,
2280 each_loose_cruft_fn cruft_cb,
2281 each_loose_subdir_fn subdir_cb,
2282 void *data)
2283 {
2284 struct strbuf buf = STRBUF_INIT;
2285 int r;
2286
2287 strbuf_addstr(&buf, path);
2288 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2289 subdir_cb, data);
2290 strbuf_release(&buf);
2291
2292 return r;
2293 }
2294
2295 int for_each_loose_object(each_loose_object_fn cb, void *data,
2296 enum for_each_object_flags flags)
2297 {
2298 struct object_directory *odb;
2299
2300 prepare_alt_odb(the_repository);
2301 for (odb = the_repository->objects->odb; odb; odb = odb->next) {
2302 int r = for_each_loose_file_in_objdir(odb->path, cb, NULL,
2303 NULL, data);
2304 if (r)
2305 return r;
2306
2307 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2308 break;
2309 }
2310
2311 return 0;
2312 }
2313
2314 static int append_loose_object(const struct object_id *oid, const char *path,
2315 void *data)
2316 {
2317 oid_array_append(data, oid);
2318 return 0;
2319 }
2320
2321 struct oid_array *odb_loose_cache(struct object_directory *odb,
2322 const struct object_id *oid)
2323 {
2324 int subdir_nr = oid->hash[0];
2325 struct strbuf buf = STRBUF_INIT;
2326
2327 if (subdir_nr < 0 ||
2328 subdir_nr >= ARRAY_SIZE(odb->loose_objects_subdir_seen))
2329 BUG("subdir_nr out of range");
2330
2331 if (odb->loose_objects_subdir_seen[subdir_nr])
2332 return &odb->loose_objects_cache[subdir_nr];
2333
2334 strbuf_addstr(&buf, odb->path);
2335 for_each_file_in_obj_subdir(subdir_nr, &buf,
2336 append_loose_object,
2337 NULL, NULL,
2338 &odb->loose_objects_cache[subdir_nr]);
2339 odb->loose_objects_subdir_seen[subdir_nr] = 1;
2340 strbuf_release(&buf);
2341 return &odb->loose_objects_cache[subdir_nr];
2342 }
2343
2344 void odb_clear_loose_cache(struct object_directory *odb)
2345 {
2346 int i;
2347
2348 for (i = 0; i < ARRAY_SIZE(odb->loose_objects_cache); i++)
2349 oid_array_clear(&odb->loose_objects_cache[i]);
2350 memset(&odb->loose_objects_subdir_seen, 0,
2351 sizeof(odb->loose_objects_subdir_seen));
2352 }
2353
2354 static int check_stream_oid(git_zstream *stream,
2355 const char *hdr,
2356 unsigned long size,
2357 const char *path,
2358 const struct object_id *expected_oid)
2359 {
2360 git_hash_ctx c;
2361 struct object_id real_oid;
2362 unsigned char buf[4096];
2363 unsigned long total_read;
2364 int status = Z_OK;
2365
2366 the_hash_algo->init_fn(&c);
2367 the_hash_algo->update_fn(&c, hdr, stream->total_out);
2368
2369 /*
2370 * We already read some bytes into hdr, but the ones up to the NUL
2371 * do not count against the object's content size.
2372 */
2373 total_read = stream->total_out - strlen(hdr) - 1;
2374
2375 /*
2376 * This size comparison must be "<=" to read the final zlib packets;
2377 * see the comment in unpack_loose_rest for details.
2378 */
2379 while (total_read <= size &&
2380 (status == Z_OK ||
2381 (status == Z_BUF_ERROR && !stream->avail_out))) {
2382 stream->next_out = buf;
2383 stream->avail_out = sizeof(buf);
2384 if (size - total_read < stream->avail_out)
2385 stream->avail_out = size - total_read;
2386 status = git_inflate(stream, Z_FINISH);
2387 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2388 total_read += stream->next_out - buf;
2389 }
2390 git_inflate_end(stream);
2391
2392 if (status != Z_STREAM_END) {
2393 error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid));
2394 return -1;
2395 }
2396 if (stream->avail_in) {
2397 error(_("garbage at end of loose object '%s'"),
2398 oid_to_hex(expected_oid));
2399 return -1;
2400 }
2401
2402 the_hash_algo->final_fn(real_oid.hash, &c);
2403 if (!oideq(expected_oid, &real_oid)) {
2404 error(_("hash mismatch for %s (expected %s)"), path,
2405 oid_to_hex(expected_oid));
2406 return -1;
2407 }
2408
2409 return 0;
2410 }
2411
2412 int read_loose_object(const char *path,
2413 const struct object_id *expected_oid,
2414 enum object_type *type,
2415 unsigned long *size,
2416 void **contents)
2417 {
2418 int ret = -1;
2419 void *map = NULL;
2420 unsigned long mapsize;
2421 git_zstream stream;
2422 char hdr[MAX_HEADER_LEN];
2423
2424 *contents = NULL;
2425
2426 map = map_loose_object_1(the_repository, path, NULL, &mapsize);
2427 if (!map) {
2428 error_errno(_("unable to mmap %s"), path);
2429 goto out;
2430 }
2431
2432 if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2433 error(_("unable to unpack header of %s"), path);
2434 goto out;
2435 }
2436
2437 *type = parse_loose_header(hdr, size);
2438 if (*type < 0) {
2439 error(_("unable to parse header of %s"), path);
2440 git_inflate_end(&stream);
2441 goto out;
2442 }
2443
2444 if (*type == OBJ_BLOB && *size > big_file_threshold) {
2445 if (check_stream_oid(&stream, hdr, *size, path, expected_oid) < 0)
2446 goto out;
2447 } else {
2448 *contents = unpack_loose_rest(&stream, hdr, *size, expected_oid);
2449 if (!*contents) {
2450 error(_("unable to unpack contents of %s"), path);
2451 git_inflate_end(&stream);
2452 goto out;
2453 }
2454 if (check_object_signature(expected_oid, *contents,
2455 *size, type_name(*type))) {
2456 error(_("hash mismatch for %s (expected %s)"), path,
2457 oid_to_hex(expected_oid));
2458 free(*contents);
2459 goto out;
2460 }
2461 }
2462
2463 ret = 0; /* everything checks out */
2464
2465 out:
2466 if (map)
2467 munmap(map, mapsize);
2468 return ret;
2469 }