user-manual.txt: render ASCII art correctly under Asciidoctor
[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 "fetch-object.h"
34 #include "object-store.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) {
956 if (!length)
957 return NULL;
958 release_pack_memory(length);
959 ret = mmap(start, length, prot, flags, fd, offset);
960 }
961 return ret;
962 }
963
964 void *xmmap(void *start, size_t length,
965 int prot, int flags, int fd, off_t offset)
966 {
967 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
968 if (ret == MAP_FAILED)
969 die_errno(_("mmap failed"));
970 return ret;
971 }
972
973 /*
974 * With an in-core object data in "map", rehash it to make sure the
975 * object name actually matches "oid" to detect object corruption.
976 * With "map" == NULL, try reading the object named with "oid" using
977 * the streaming interface and rehash it to do the same.
978 */
979 int check_object_signature(const struct object_id *oid, void *map,
980 unsigned long size, const char *type)
981 {
982 struct object_id real_oid;
983 enum object_type obj_type;
984 struct git_istream *st;
985 git_hash_ctx c;
986 char hdr[MAX_HEADER_LEN];
987 int hdrlen;
988
989 if (map) {
990 hash_object_file(map, size, type, &real_oid);
991 return !oideq(oid, &real_oid) ? -1 : 0;
992 }
993
994 st = open_istream(oid, &obj_type, &size, NULL);
995 if (!st)
996 return -1;
997
998 /* Generate the header */
999 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(obj_type), (uintmax_t)size) + 1;
1000
1001 /* Sha1.. */
1002 the_hash_algo->init_fn(&c);
1003 the_hash_algo->update_fn(&c, hdr, hdrlen);
1004 for (;;) {
1005 char buf[1024 * 16];
1006 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1007
1008 if (readlen < 0) {
1009 close_istream(st);
1010 return -1;
1011 }
1012 if (!readlen)
1013 break;
1014 the_hash_algo->update_fn(&c, buf, readlen);
1015 }
1016 the_hash_algo->final_fn(real_oid.hash, &c);
1017 close_istream(st);
1018 return !oideq(oid, &real_oid) ? -1 : 0;
1019 }
1020
1021 int git_open_cloexec(const char *name, int flags)
1022 {
1023 int fd;
1024 static int o_cloexec = O_CLOEXEC;
1025
1026 fd = open(name, flags | o_cloexec);
1027 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1028 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1029 o_cloexec &= ~O_CLOEXEC;
1030 fd = open(name, flags | o_cloexec);
1031 }
1032
1033 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1034 {
1035 static int fd_cloexec = FD_CLOEXEC;
1036
1037 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1038 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1039 int flags = fcntl(fd, F_GETFD);
1040 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1041 fd_cloexec = 0;
1042 }
1043 }
1044 #endif
1045 return fd;
1046 }
1047
1048 /*
1049 * Find "oid" as a loose object in the local repository or in an alternate.
1050 * Returns 0 on success, negative on failure.
1051 *
1052 * The "path" out-parameter will give the path of the object we found (if any).
1053 * Note that it may point to static storage and is only valid until another
1054 * call to stat_loose_object().
1055 */
1056 static int stat_loose_object(struct repository *r, const struct object_id *oid,
1057 struct stat *st, const char **path)
1058 {
1059 struct object_directory *odb;
1060 static struct strbuf buf = STRBUF_INIT;
1061
1062 prepare_alt_odb(r);
1063 for (odb = r->objects->odb; odb; odb = odb->next) {
1064 *path = odb_loose_path(odb, &buf, oid);
1065 if (!lstat(*path, st))
1066 return 0;
1067 }
1068
1069 return -1;
1070 }
1071
1072 /*
1073 * Like stat_loose_object(), but actually open the object and return the
1074 * descriptor. See the caveats on the "path" parameter above.
1075 */
1076 static int open_loose_object(struct repository *r,
1077 const struct object_id *oid, const char **path)
1078 {
1079 int fd;
1080 struct object_directory *odb;
1081 int most_interesting_errno = ENOENT;
1082 static struct strbuf buf = STRBUF_INIT;
1083
1084 prepare_alt_odb(r);
1085 for (odb = r->objects->odb; odb; odb = odb->next) {
1086 *path = odb_loose_path(odb, &buf, oid);
1087 fd = git_open(*path);
1088 if (fd >= 0)
1089 return fd;
1090
1091 if (most_interesting_errno == ENOENT)
1092 most_interesting_errno = errno;
1093 }
1094 errno = most_interesting_errno;
1095 return -1;
1096 }
1097
1098 static int quick_has_loose(struct repository *r,
1099 const struct object_id *oid)
1100 {
1101 struct object_directory *odb;
1102
1103 prepare_alt_odb(r);
1104 for (odb = r->objects->odb; odb; odb = odb->next) {
1105 if (oid_array_lookup(odb_loose_cache(odb, oid), oid) >= 0)
1106 return 1;
1107 }
1108 return 0;
1109 }
1110
1111 /*
1112 * Map the loose object at "path" if it is not NULL, or the path found by
1113 * searching for a loose object named "oid".
1114 */
1115 static void *map_loose_object_1(struct repository *r, const char *path,
1116 const struct object_id *oid, unsigned long *size)
1117 {
1118 void *map;
1119 int fd;
1120
1121 if (path)
1122 fd = git_open(path);
1123 else
1124 fd = open_loose_object(r, oid, &path);
1125 map = NULL;
1126 if (fd >= 0) {
1127 struct stat st;
1128
1129 if (!fstat(fd, &st)) {
1130 *size = xsize_t(st.st_size);
1131 if (!*size) {
1132 /* mmap() is forbidden on empty files */
1133 error(_("object file %s is empty"), path);
1134 close(fd);
1135 return NULL;
1136 }
1137 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1138 }
1139 close(fd);
1140 }
1141 return map;
1142 }
1143
1144 void *map_loose_object(struct repository *r,
1145 const struct object_id *oid,
1146 unsigned long *size)
1147 {
1148 return map_loose_object_1(r, NULL, oid, size);
1149 }
1150
1151 static int unpack_loose_short_header(git_zstream *stream,
1152 unsigned char *map, unsigned long mapsize,
1153 void *buffer, unsigned long bufsiz)
1154 {
1155 /* Get the data stream */
1156 memset(stream, 0, sizeof(*stream));
1157 stream->next_in = map;
1158 stream->avail_in = mapsize;
1159 stream->next_out = buffer;
1160 stream->avail_out = bufsiz;
1161
1162 git_inflate_init(stream);
1163 return git_inflate(stream, 0);
1164 }
1165
1166 int unpack_loose_header(git_zstream *stream,
1167 unsigned char *map, unsigned long mapsize,
1168 void *buffer, unsigned long bufsiz)
1169 {
1170 int status = unpack_loose_short_header(stream, map, mapsize,
1171 buffer, bufsiz);
1172
1173 if (status < Z_OK)
1174 return status;
1175
1176 /* Make sure we have the terminating NUL */
1177 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1178 return -1;
1179 return 0;
1180 }
1181
1182 static int unpack_loose_header_to_strbuf(git_zstream *stream, unsigned char *map,
1183 unsigned long mapsize, void *buffer,
1184 unsigned long bufsiz, struct strbuf *header)
1185 {
1186 int status;
1187
1188 status = unpack_loose_short_header(stream, map, mapsize, buffer, bufsiz);
1189 if (status < Z_OK)
1190 return -1;
1191
1192 /*
1193 * Check if entire header is unpacked in the first iteration.
1194 */
1195 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1196 return 0;
1197
1198 /*
1199 * buffer[0..bufsiz] was not large enough. Copy the partial
1200 * result out to header, and then append the result of further
1201 * reading the stream.
1202 */
1203 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1204 stream->next_out = buffer;
1205 stream->avail_out = bufsiz;
1206
1207 do {
1208 status = git_inflate(stream, 0);
1209 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1210 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1211 return 0;
1212 stream->next_out = buffer;
1213 stream->avail_out = bufsiz;
1214 } while (status != Z_STREAM_END);
1215 return -1;
1216 }
1217
1218 static void *unpack_loose_rest(git_zstream *stream,
1219 void *buffer, unsigned long size,
1220 const struct object_id *oid)
1221 {
1222 int bytes = strlen(buffer) + 1;
1223 unsigned char *buf = xmallocz(size);
1224 unsigned long n;
1225 int status = Z_OK;
1226
1227 n = stream->total_out - bytes;
1228 if (n > size)
1229 n = size;
1230 memcpy(buf, (char *) buffer + bytes, n);
1231 bytes = n;
1232 if (bytes <= size) {
1233 /*
1234 * The above condition must be (bytes <= size), not
1235 * (bytes < size). In other words, even though we
1236 * expect no more output and set avail_out to zero,
1237 * the input zlib stream may have bytes that express
1238 * "this concludes the stream", and we *do* want to
1239 * eat that input.
1240 *
1241 * Otherwise we would not be able to test that we
1242 * consumed all the input to reach the expected size;
1243 * we also want to check that zlib tells us that all
1244 * went well with status == Z_STREAM_END at the end.
1245 */
1246 stream->next_out = buf + bytes;
1247 stream->avail_out = size - bytes;
1248 while (status == Z_OK)
1249 status = git_inflate(stream, Z_FINISH);
1250 }
1251 if (status == Z_STREAM_END && !stream->avail_in) {
1252 git_inflate_end(stream);
1253 return buf;
1254 }
1255
1256 if (status < 0)
1257 error(_("corrupt loose object '%s'"), oid_to_hex(oid));
1258 else if (stream->avail_in)
1259 error(_("garbage at end of loose object '%s'"),
1260 oid_to_hex(oid));
1261 free(buf);
1262 return NULL;
1263 }
1264
1265 /*
1266 * We used to just use "sscanf()", but that's actually way
1267 * too permissive for what we want to check. So do an anal
1268 * object header parse by hand.
1269 */
1270 static int parse_loose_header_extended(const char *hdr, struct object_info *oi,
1271 unsigned int flags)
1272 {
1273 const char *type_buf = hdr;
1274 unsigned long size;
1275 int type, type_len = 0;
1276
1277 /*
1278 * The type can be of any size but is followed by
1279 * a space.
1280 */
1281 for (;;) {
1282 char c = *hdr++;
1283 if (!c)
1284 return -1;
1285 if (c == ' ')
1286 break;
1287 type_len++;
1288 }
1289
1290 type = type_from_string_gently(type_buf, type_len, 1);
1291 if (oi->type_name)
1292 strbuf_add(oi->type_name, type_buf, type_len);
1293 /*
1294 * Set type to 0 if its an unknown object and
1295 * we're obtaining the type using '--allow-unknown-type'
1296 * option.
1297 */
1298 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1299 type = 0;
1300 else if (type < 0)
1301 die(_("invalid object type"));
1302 if (oi->typep)
1303 *oi->typep = type;
1304
1305 /*
1306 * The length must follow immediately, and be in canonical
1307 * decimal format (ie "010" is not valid).
1308 */
1309 size = *hdr++ - '0';
1310 if (size > 9)
1311 return -1;
1312 if (size) {
1313 for (;;) {
1314 unsigned long c = *hdr - '0';
1315 if (c > 9)
1316 break;
1317 hdr++;
1318 size = size * 10 + c;
1319 }
1320 }
1321
1322 if (oi->sizep)
1323 *oi->sizep = size;
1324
1325 /*
1326 * The length must be followed by a zero byte
1327 */
1328 return *hdr ? -1 : type;
1329 }
1330
1331 int parse_loose_header(const char *hdr, unsigned long *sizep)
1332 {
1333 struct object_info oi = OBJECT_INFO_INIT;
1334
1335 oi.sizep = sizep;
1336 return parse_loose_header_extended(hdr, &oi, 0);
1337 }
1338
1339 static int loose_object_info(struct repository *r,
1340 const struct object_id *oid,
1341 struct object_info *oi, int flags)
1342 {
1343 int status = 0;
1344 unsigned long mapsize;
1345 void *map;
1346 git_zstream stream;
1347 char hdr[MAX_HEADER_LEN];
1348 struct strbuf hdrbuf = STRBUF_INIT;
1349 unsigned long size_scratch;
1350
1351 if (oi->delta_base_sha1)
1352 hashclr(oi->delta_base_sha1);
1353
1354 /*
1355 * If we don't care about type or size, then we don't
1356 * need to look inside the object at all. Note that we
1357 * do not optimize out the stat call, even if the
1358 * caller doesn't care about the disk-size, since our
1359 * return value implicitly indicates whether the
1360 * object even exists.
1361 */
1362 if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1363 const char *path;
1364 struct stat st;
1365 if (!oi->disk_sizep && (flags & OBJECT_INFO_QUICK))
1366 return quick_has_loose(r, oid) ? 0 : -1;
1367 if (stat_loose_object(r, oid, &st, &path) < 0)
1368 return -1;
1369 if (oi->disk_sizep)
1370 *oi->disk_sizep = st.st_size;
1371 return 0;
1372 }
1373
1374 map = map_loose_object(r, oid, &mapsize);
1375 if (!map)
1376 return -1;
1377
1378 if (!oi->sizep)
1379 oi->sizep = &size_scratch;
1380
1381 if (oi->disk_sizep)
1382 *oi->disk_sizep = mapsize;
1383 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1384 if (unpack_loose_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1385 status = error(_("unable to unpack %s header with --allow-unknown-type"),
1386 oid_to_hex(oid));
1387 } else if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1388 status = error(_("unable to unpack %s header"),
1389 oid_to_hex(oid));
1390 if (status < 0)
1391 ; /* Do nothing */
1392 else if (hdrbuf.len) {
1393 if ((status = parse_loose_header_extended(hdrbuf.buf, oi, flags)) < 0)
1394 status = error(_("unable to parse %s header with --allow-unknown-type"),
1395 oid_to_hex(oid));
1396 } else if ((status = parse_loose_header_extended(hdr, oi, flags)) < 0)
1397 status = error(_("unable to parse %s header"), oid_to_hex(oid));
1398
1399 if (status >= 0 && oi->contentp) {
1400 *oi->contentp = unpack_loose_rest(&stream, hdr,
1401 *oi->sizep, oid);
1402 if (!*oi->contentp) {
1403 git_inflate_end(&stream);
1404 status = -1;
1405 }
1406 } else
1407 git_inflate_end(&stream);
1408
1409 munmap(map, mapsize);
1410 if (status && oi->typep)
1411 *oi->typep = status;
1412 if (oi->sizep == &size_scratch)
1413 oi->sizep = NULL;
1414 strbuf_release(&hdrbuf);
1415 oi->whence = OI_LOOSE;
1416 return (status < 0) ? status : 0;
1417 }
1418
1419 int fetch_if_missing = 1;
1420
1421 int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1422 struct object_info *oi, unsigned flags)
1423 {
1424 static struct object_info blank_oi = OBJECT_INFO_INIT;
1425 struct pack_entry e;
1426 int rtype;
1427 const struct object_id *real = oid;
1428 int already_retried = 0;
1429
1430 if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1431 real = lookup_replace_object(r, oid);
1432
1433 if (is_null_oid(real))
1434 return -1;
1435
1436 if (!oi)
1437 oi = &blank_oi;
1438
1439 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1440 struct cached_object *co = find_cached_object(real);
1441 if (co) {
1442 if (oi->typep)
1443 *(oi->typep) = co->type;
1444 if (oi->sizep)
1445 *(oi->sizep) = co->size;
1446 if (oi->disk_sizep)
1447 *(oi->disk_sizep) = 0;
1448 if (oi->delta_base_sha1)
1449 hashclr(oi->delta_base_sha1);
1450 if (oi->type_name)
1451 strbuf_addstr(oi->type_name, type_name(co->type));
1452 if (oi->contentp)
1453 *oi->contentp = xmemdupz(co->buf, co->size);
1454 oi->whence = OI_CACHED;
1455 return 0;
1456 }
1457 }
1458
1459 while (1) {
1460 if (find_pack_entry(r, real, &e))
1461 break;
1462
1463 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1464 return -1;
1465
1466 /* Most likely it's a loose object. */
1467 if (!loose_object_info(r, real, oi, flags))
1468 return 0;
1469
1470 /* Not a loose object; someone else may have just packed it. */
1471 if (!(flags & OBJECT_INFO_QUICK)) {
1472 reprepare_packed_git(r);
1473 if (find_pack_entry(r, real, &e))
1474 break;
1475 }
1476
1477 /* Check if it is a missing object */
1478 if (fetch_if_missing && repository_format_partial_clone &&
1479 !already_retried && r == the_repository &&
1480 !(flags & OBJECT_INFO_SKIP_FETCH_OBJECT)) {
1481 /*
1482 * TODO Investigate having fetch_object() return
1483 * TODO error/success and stopping the music here.
1484 * TODO Pass a repository struct through fetch_object,
1485 * such that arbitrary repositories work.
1486 */
1487 fetch_objects(repository_format_partial_clone, real, 1);
1488 already_retried = 1;
1489 continue;
1490 }
1491
1492 return -1;
1493 }
1494
1495 if (oi == &blank_oi)
1496 /*
1497 * We know that the caller doesn't actually need the
1498 * information below, so return early.
1499 */
1500 return 0;
1501 rtype = packed_object_info(r, e.p, e.offset, oi);
1502 if (rtype < 0) {
1503 mark_bad_packed_object(e.p, real->hash);
1504 return oid_object_info_extended(r, real, oi, 0);
1505 } else if (oi->whence == OI_PACKED) {
1506 oi->u.packed.offset = e.offset;
1507 oi->u.packed.pack = e.p;
1508 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1509 rtype == OBJ_OFS_DELTA);
1510 }
1511
1512 return 0;
1513 }
1514
1515 /* returns enum object_type or negative */
1516 int oid_object_info(struct repository *r,
1517 const struct object_id *oid,
1518 unsigned long *sizep)
1519 {
1520 enum object_type type;
1521 struct object_info oi = OBJECT_INFO_INIT;
1522
1523 oi.typep = &type;
1524 oi.sizep = sizep;
1525 if (oid_object_info_extended(r, oid, &oi,
1526 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1527 return -1;
1528 return type;
1529 }
1530
1531 static void *read_object(struct repository *r,
1532 const struct object_id *oid, enum object_type *type,
1533 unsigned long *size)
1534 {
1535 struct object_info oi = OBJECT_INFO_INIT;
1536 void *content;
1537 oi.typep = type;
1538 oi.sizep = size;
1539 oi.contentp = &content;
1540
1541 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1542 return NULL;
1543 return content;
1544 }
1545
1546 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1547 struct object_id *oid)
1548 {
1549 struct cached_object *co;
1550
1551 hash_object_file(buf, len, type_name(type), oid);
1552 if (has_object_file(oid) || find_cached_object(oid))
1553 return 0;
1554 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1555 co = &cached_objects[cached_object_nr++];
1556 co->size = len;
1557 co->type = type;
1558 co->buf = xmalloc(len);
1559 memcpy(co->buf, buf, len);
1560 oidcpy(&co->oid, oid);
1561 return 0;
1562 }
1563
1564 /*
1565 * This function dies on corrupt objects; the callers who want to
1566 * deal with them should arrange to call read_object() and give error
1567 * messages themselves.
1568 */
1569 void *read_object_file_extended(struct repository *r,
1570 const struct object_id *oid,
1571 enum object_type *type,
1572 unsigned long *size,
1573 int lookup_replace)
1574 {
1575 void *data;
1576 const struct packed_git *p;
1577 const char *path;
1578 struct stat st;
1579 const struct object_id *repl = lookup_replace ?
1580 lookup_replace_object(r, oid) : oid;
1581
1582 errno = 0;
1583 data = read_object(r, repl, type, size);
1584 if (data)
1585 return data;
1586
1587 if (errno && errno != ENOENT)
1588 die_errno(_("failed to read object %s"), oid_to_hex(oid));
1589
1590 /* die if we replaced an object with one that does not exist */
1591 if (repl != oid)
1592 die(_("replacement %s not found for %s"),
1593 oid_to_hex(repl), oid_to_hex(oid));
1594
1595 if (!stat_loose_object(r, repl, &st, &path))
1596 die(_("loose object %s (stored in %s) is corrupt"),
1597 oid_to_hex(repl), path);
1598
1599 if ((p = has_packed_and_bad(r, repl->hash)) != NULL)
1600 die(_("packed object %s (stored in %s) is corrupt"),
1601 oid_to_hex(repl), p->pack_name);
1602
1603 return NULL;
1604 }
1605
1606 void *read_object_with_reference(struct repository *r,
1607 const struct object_id *oid,
1608 const char *required_type_name,
1609 unsigned long *size,
1610 struct object_id *actual_oid_return)
1611 {
1612 enum object_type type, required_type;
1613 void *buffer;
1614 unsigned long isize;
1615 struct object_id actual_oid;
1616
1617 required_type = type_from_string(required_type_name);
1618 oidcpy(&actual_oid, oid);
1619 while (1) {
1620 int ref_length = -1;
1621 const char *ref_type = NULL;
1622
1623 buffer = repo_read_object_file(r, &actual_oid, &type, &isize);
1624 if (!buffer)
1625 return NULL;
1626 if (type == required_type) {
1627 *size = isize;
1628 if (actual_oid_return)
1629 oidcpy(actual_oid_return, &actual_oid);
1630 return buffer;
1631 }
1632 /* Handle references */
1633 else if (type == OBJ_COMMIT)
1634 ref_type = "tree ";
1635 else if (type == OBJ_TAG)
1636 ref_type = "object ";
1637 else {
1638 free(buffer);
1639 return NULL;
1640 }
1641 ref_length = strlen(ref_type);
1642
1643 if (ref_length + the_hash_algo->hexsz > isize ||
1644 memcmp(buffer, ref_type, ref_length) ||
1645 get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1646 free(buffer);
1647 return NULL;
1648 }
1649 free(buffer);
1650 /* Now we have the ID of the referred-to object in
1651 * actual_oid. Check again. */
1652 }
1653 }
1654
1655 static void write_object_file_prepare(const void *buf, unsigned long len,
1656 const char *type, struct object_id *oid,
1657 char *hdr, int *hdrlen)
1658 {
1659 git_hash_ctx c;
1660
1661 /* Generate the header */
1662 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %"PRIuMAX , type, (uintmax_t)len)+1;
1663
1664 /* Sha1.. */
1665 the_hash_algo->init_fn(&c);
1666 the_hash_algo->update_fn(&c, hdr, *hdrlen);
1667 the_hash_algo->update_fn(&c, buf, len);
1668 the_hash_algo->final_fn(oid->hash, &c);
1669 }
1670
1671 /*
1672 * Move the just written object into its final resting place.
1673 */
1674 int finalize_object_file(const char *tmpfile, const char *filename)
1675 {
1676 int ret = 0;
1677
1678 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1679 goto try_rename;
1680 else if (link(tmpfile, filename))
1681 ret = errno;
1682
1683 /*
1684 * Coda hack - coda doesn't like cross-directory links,
1685 * so we fall back to a rename, which will mean that it
1686 * won't be able to check collisions, but that's not a
1687 * big deal.
1688 *
1689 * The same holds for FAT formatted media.
1690 *
1691 * When this succeeds, we just return. We have nothing
1692 * left to unlink.
1693 */
1694 if (ret && ret != EEXIST) {
1695 try_rename:
1696 if (!rename(tmpfile, filename))
1697 goto out;
1698 ret = errno;
1699 }
1700 unlink_or_warn(tmpfile);
1701 if (ret) {
1702 if (ret != EEXIST) {
1703 return error_errno(_("unable to write file %s"), filename);
1704 }
1705 /* FIXME!!! Collision check here ? */
1706 }
1707
1708 out:
1709 if (adjust_shared_perm(filename))
1710 return error(_("unable to set permission to '%s'"), filename);
1711 return 0;
1712 }
1713
1714 static int write_buffer(int fd, const void *buf, size_t len)
1715 {
1716 if (write_in_full(fd, buf, len) < 0)
1717 return error_errno(_("file write error"));
1718 return 0;
1719 }
1720
1721 int hash_object_file(const void *buf, unsigned long len, const char *type,
1722 struct object_id *oid)
1723 {
1724 char hdr[MAX_HEADER_LEN];
1725 int hdrlen = sizeof(hdr);
1726 write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1727 return 0;
1728 }
1729
1730 /* Finalize a file on disk, and close it. */
1731 static void close_loose_object(int fd)
1732 {
1733 if (fsync_object_files)
1734 fsync_or_die(fd, "loose object file");
1735 if (close(fd) != 0)
1736 die_errno(_("error when closing loose object file"));
1737 }
1738
1739 /* Size of directory component, including the ending '/' */
1740 static inline int directory_size(const char *filename)
1741 {
1742 const char *s = strrchr(filename, '/');
1743 if (!s)
1744 return 0;
1745 return s - filename + 1;
1746 }
1747
1748 /*
1749 * This creates a temporary file in the same directory as the final
1750 * 'filename'
1751 *
1752 * We want to avoid cross-directory filename renames, because those
1753 * can have problems on various filesystems (FAT, NFS, Coda).
1754 */
1755 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1756 {
1757 int fd, dirlen = directory_size(filename);
1758
1759 strbuf_reset(tmp);
1760 strbuf_add(tmp, filename, dirlen);
1761 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1762 fd = git_mkstemp_mode(tmp->buf, 0444);
1763 if (fd < 0 && dirlen && errno == ENOENT) {
1764 /*
1765 * Make sure the directory exists; note that the contents
1766 * of the buffer are undefined after mkstemp returns an
1767 * error, so we have to rewrite the whole buffer from
1768 * scratch.
1769 */
1770 strbuf_reset(tmp);
1771 strbuf_add(tmp, filename, dirlen - 1);
1772 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1773 return -1;
1774 if (adjust_shared_perm(tmp->buf))
1775 return -1;
1776
1777 /* Try again */
1778 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1779 fd = git_mkstemp_mode(tmp->buf, 0444);
1780 }
1781 return fd;
1782 }
1783
1784 static int write_loose_object(const struct object_id *oid, char *hdr,
1785 int hdrlen, const void *buf, unsigned long len,
1786 time_t mtime)
1787 {
1788 int fd, ret;
1789 unsigned char compressed[4096];
1790 git_zstream stream;
1791 git_hash_ctx c;
1792 struct object_id parano_oid;
1793 static struct strbuf tmp_file = STRBUF_INIT;
1794 static struct strbuf filename = STRBUF_INIT;
1795
1796 loose_object_path(the_repository, &filename, oid);
1797
1798 fd = create_tmpfile(&tmp_file, filename.buf);
1799 if (fd < 0) {
1800 if (errno == EACCES)
1801 return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
1802 else
1803 return error_errno(_("unable to create temporary file"));
1804 }
1805
1806 /* Set it up */
1807 git_deflate_init(&stream, zlib_compression_level);
1808 stream.next_out = compressed;
1809 stream.avail_out = sizeof(compressed);
1810 the_hash_algo->init_fn(&c);
1811
1812 /* First header.. */
1813 stream.next_in = (unsigned char *)hdr;
1814 stream.avail_in = hdrlen;
1815 while (git_deflate(&stream, 0) == Z_OK)
1816 ; /* nothing */
1817 the_hash_algo->update_fn(&c, hdr, hdrlen);
1818
1819 /* Then the data itself.. */
1820 stream.next_in = (void *)buf;
1821 stream.avail_in = len;
1822 do {
1823 unsigned char *in0 = stream.next_in;
1824 ret = git_deflate(&stream, Z_FINISH);
1825 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1826 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1827 die(_("unable to write loose object file"));
1828 stream.next_out = compressed;
1829 stream.avail_out = sizeof(compressed);
1830 } while (ret == Z_OK);
1831
1832 if (ret != Z_STREAM_END)
1833 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
1834 ret);
1835 ret = git_deflate_end_gently(&stream);
1836 if (ret != Z_OK)
1837 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
1838 ret);
1839 the_hash_algo->final_fn(parano_oid.hash, &c);
1840 if (!oideq(oid, &parano_oid))
1841 die(_("confused by unstable object source data for %s"),
1842 oid_to_hex(oid));
1843
1844 close_loose_object(fd);
1845
1846 if (mtime) {
1847 struct utimbuf utb;
1848 utb.actime = mtime;
1849 utb.modtime = mtime;
1850 if (utime(tmp_file.buf, &utb) < 0)
1851 warning_errno(_("failed utime() on %s"), tmp_file.buf);
1852 }
1853
1854 return finalize_object_file(tmp_file.buf, filename.buf);
1855 }
1856
1857 static int freshen_loose_object(const struct object_id *oid)
1858 {
1859 return check_and_freshen(oid, 1);
1860 }
1861
1862 static int freshen_packed_object(const struct object_id *oid)
1863 {
1864 struct pack_entry e;
1865 if (!find_pack_entry(the_repository, oid, &e))
1866 return 0;
1867 if (e.p->freshened)
1868 return 1;
1869 if (!freshen_file(e.p->pack_name))
1870 return 0;
1871 e.p->freshened = 1;
1872 return 1;
1873 }
1874
1875 int write_object_file(const void *buf, unsigned long len, const char *type,
1876 struct object_id *oid)
1877 {
1878 char hdr[MAX_HEADER_LEN];
1879 int hdrlen = sizeof(hdr);
1880
1881 /* Normally if we have it in the pack then we do not bother writing
1882 * it out into .git/objects/??/?{38} file.
1883 */
1884 write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1885 if (freshen_packed_object(oid) || freshen_loose_object(oid))
1886 return 0;
1887 return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
1888 }
1889
1890 int hash_object_file_literally(const void *buf, unsigned long len,
1891 const char *type, struct object_id *oid,
1892 unsigned flags)
1893 {
1894 char *header;
1895 int hdrlen, status = 0;
1896
1897 /* type string, SP, %lu of the length plus NUL must fit this */
1898 hdrlen = strlen(type) + MAX_HEADER_LEN;
1899 header = xmalloc(hdrlen);
1900 write_object_file_prepare(buf, len, type, oid, header, &hdrlen);
1901
1902 if (!(flags & HASH_WRITE_OBJECT))
1903 goto cleanup;
1904 if (freshen_packed_object(oid) || freshen_loose_object(oid))
1905 goto cleanup;
1906 status = write_loose_object(oid, header, hdrlen, buf, len, 0);
1907
1908 cleanup:
1909 free(header);
1910 return status;
1911 }
1912
1913 int force_object_loose(const struct object_id *oid, time_t mtime)
1914 {
1915 void *buf;
1916 unsigned long len;
1917 enum object_type type;
1918 char hdr[MAX_HEADER_LEN];
1919 int hdrlen;
1920 int ret;
1921
1922 if (has_loose_object(oid))
1923 return 0;
1924 buf = read_object(the_repository, oid, &type, &len);
1925 if (!buf)
1926 return error(_("cannot read object for %s"), oid_to_hex(oid));
1927 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(type), (uintmax_t)len) + 1;
1928 ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
1929 free(buf);
1930
1931 return ret;
1932 }
1933
1934 int repo_has_object_file_with_flags(struct repository *r,
1935 const struct object_id *oid, int flags)
1936 {
1937 if (!startup_info->have_repository)
1938 return 0;
1939 return oid_object_info_extended(r, oid, NULL,
1940 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1941 }
1942
1943 int repo_has_object_file(struct repository *r,
1944 const struct object_id *oid)
1945 {
1946 return repo_has_object_file_with_flags(r, oid, 0);
1947 }
1948
1949 static void check_tree(const void *buf, size_t size)
1950 {
1951 struct tree_desc desc;
1952 struct name_entry entry;
1953
1954 init_tree_desc(&desc, buf, size);
1955 while (tree_entry(&desc, &entry))
1956 /* do nothing
1957 * tree_entry() will die() on malformed entries */
1958 ;
1959 }
1960
1961 static void check_commit(const void *buf, size_t size)
1962 {
1963 struct commit c;
1964 memset(&c, 0, sizeof(c));
1965 if (parse_commit_buffer(the_repository, &c, buf, size, 0))
1966 die(_("corrupt commit"));
1967 }
1968
1969 static void check_tag(const void *buf, size_t size)
1970 {
1971 struct tag t;
1972 memset(&t, 0, sizeof(t));
1973 if (parse_tag_buffer(the_repository, &t, buf, size))
1974 die(_("corrupt tag"));
1975 }
1976
1977 static int index_mem(struct index_state *istate,
1978 struct object_id *oid, void *buf, size_t size,
1979 enum object_type type,
1980 const char *path, unsigned flags)
1981 {
1982 int ret, re_allocated = 0;
1983 int write_object = flags & HASH_WRITE_OBJECT;
1984
1985 if (!type)
1986 type = OBJ_BLOB;
1987
1988 /*
1989 * Convert blobs to git internal format
1990 */
1991 if ((type == OBJ_BLOB) && path) {
1992 struct strbuf nbuf = STRBUF_INIT;
1993 if (convert_to_git(istate, path, buf, size, &nbuf,
1994 get_conv_flags(flags))) {
1995 buf = strbuf_detach(&nbuf, &size);
1996 re_allocated = 1;
1997 }
1998 }
1999 if (flags & HASH_FORMAT_CHECK) {
2000 if (type == OBJ_TREE)
2001 check_tree(buf, size);
2002 if (type == OBJ_COMMIT)
2003 check_commit(buf, size);
2004 if (type == OBJ_TAG)
2005 check_tag(buf, size);
2006 }
2007
2008 if (write_object)
2009 ret = write_object_file(buf, size, type_name(type), oid);
2010 else
2011 ret = hash_object_file(buf, size, type_name(type), oid);
2012 if (re_allocated)
2013 free(buf);
2014 return ret;
2015 }
2016
2017 static int index_stream_convert_blob(struct index_state *istate,
2018 struct object_id *oid,
2019 int fd,
2020 const char *path,
2021 unsigned flags)
2022 {
2023 int ret;
2024 const int write_object = flags & HASH_WRITE_OBJECT;
2025 struct strbuf sbuf = STRBUF_INIT;
2026
2027 assert(path);
2028 assert(would_convert_to_git_filter_fd(istate, path));
2029
2030 convert_to_git_filter_fd(istate, path, fd, &sbuf,
2031 get_conv_flags(flags));
2032
2033 if (write_object)
2034 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
2035 oid);
2036 else
2037 ret = hash_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
2038 oid);
2039 strbuf_release(&sbuf);
2040 return ret;
2041 }
2042
2043 static int index_pipe(struct index_state *istate, struct object_id *oid,
2044 int fd, enum object_type type,
2045 const char *path, unsigned flags)
2046 {
2047 struct strbuf sbuf = STRBUF_INIT;
2048 int ret;
2049
2050 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2051 ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags);
2052 else
2053 ret = -1;
2054 strbuf_release(&sbuf);
2055 return ret;
2056 }
2057
2058 #define SMALL_FILE_SIZE (32*1024)
2059
2060 static int index_core(struct index_state *istate,
2061 struct object_id *oid, int fd, size_t size,
2062 enum object_type type, const char *path,
2063 unsigned flags)
2064 {
2065 int ret;
2066
2067 if (!size) {
2068 ret = index_mem(istate, oid, "", size, type, path, flags);
2069 } else if (size <= SMALL_FILE_SIZE) {
2070 char *buf = xmalloc(size);
2071 ssize_t read_result = read_in_full(fd, buf, size);
2072 if (read_result < 0)
2073 ret = error_errno(_("read error while indexing %s"),
2074 path ? path : "<unknown>");
2075 else if (read_result != size)
2076 ret = error(_("short read while indexing %s"),
2077 path ? path : "<unknown>");
2078 else
2079 ret = index_mem(istate, oid, buf, size, type, path, flags);
2080 free(buf);
2081 } else {
2082 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2083 ret = index_mem(istate, oid, buf, size, type, path, flags);
2084 munmap(buf, size);
2085 }
2086 return ret;
2087 }
2088
2089 /*
2090 * This creates one packfile per large blob unless bulk-checkin
2091 * machinery is "plugged".
2092 *
2093 * This also bypasses the usual "convert-to-git" dance, and that is on
2094 * purpose. We could write a streaming version of the converting
2095 * functions and insert that before feeding the data to fast-import
2096 * (or equivalent in-core API described above). However, that is
2097 * somewhat complicated, as we do not know the size of the filter
2098 * result, which we need to know beforehand when writing a git object.
2099 * Since the primary motivation for trying to stream from the working
2100 * tree file and to avoid mmaping it in core is to deal with large
2101 * binary blobs, they generally do not want to get any conversion, and
2102 * callers should avoid this code path when filters are requested.
2103 */
2104 static int index_stream(struct object_id *oid, int fd, size_t size,
2105 enum object_type type, const char *path,
2106 unsigned flags)
2107 {
2108 return index_bulk_checkin(oid, fd, size, type, path, flags);
2109 }
2110
2111 int index_fd(struct index_state *istate, struct object_id *oid,
2112 int fd, struct stat *st,
2113 enum object_type type, const char *path, unsigned flags)
2114 {
2115 int ret;
2116
2117 /*
2118 * Call xsize_t() only when needed to avoid potentially unnecessary
2119 * die() for large files.
2120 */
2121 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, path))
2122 ret = index_stream_convert_blob(istate, oid, fd, path, flags);
2123 else if (!S_ISREG(st->st_mode))
2124 ret = index_pipe(istate, oid, fd, type, path, flags);
2125 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
2126 (path && would_convert_to_git(istate, path)))
2127 ret = index_core(istate, oid, fd, xsize_t(st->st_size),
2128 type, path, flags);
2129 else
2130 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
2131 flags);
2132 close(fd);
2133 return ret;
2134 }
2135
2136 int index_path(struct index_state *istate, struct object_id *oid,
2137 const char *path, struct stat *st, unsigned flags)
2138 {
2139 int fd;
2140 struct strbuf sb = STRBUF_INIT;
2141 int rc = 0;
2142
2143 switch (st->st_mode & S_IFMT) {
2144 case S_IFREG:
2145 fd = open(path, O_RDONLY);
2146 if (fd < 0)
2147 return error_errno("open(\"%s\")", path);
2148 if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0)
2149 return error(_("%s: failed to insert into database"),
2150 path);
2151 break;
2152 case S_IFLNK:
2153 if (strbuf_readlink(&sb, path, st->st_size))
2154 return error_errno("readlink(\"%s\")", path);
2155 if (!(flags & HASH_WRITE_OBJECT))
2156 hash_object_file(sb.buf, sb.len, blob_type, oid);
2157 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
2158 rc = error(_("%s: failed to insert into database"), path);
2159 strbuf_release(&sb);
2160 break;
2161 case S_IFDIR:
2162 return resolve_gitlink_ref(path, "HEAD", oid);
2163 default:
2164 return error(_("%s: unsupported file type"), path);
2165 }
2166 return rc;
2167 }
2168
2169 int read_pack_header(int fd, struct pack_header *header)
2170 {
2171 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
2172 /* "eof before pack header was fully read" */
2173 return PH_ERROR_EOF;
2174
2175 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2176 /* "protocol error (pack signature mismatch detected)" */
2177 return PH_ERROR_PACK_SIGNATURE;
2178 if (!pack_version_ok(header->hdr_version))
2179 /* "protocol error (pack version unsupported)" */
2180 return PH_ERROR_PROTOCOL;
2181 return 0;
2182 }
2183
2184 void assert_oid_type(const struct object_id *oid, enum object_type expect)
2185 {
2186 enum object_type type = oid_object_info(the_repository, oid, NULL);
2187 if (type < 0)
2188 die(_("%s is not a valid object"), oid_to_hex(oid));
2189 if (type != expect)
2190 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
2191 type_name(expect));
2192 }
2193
2194 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2195 struct strbuf *path,
2196 each_loose_object_fn obj_cb,
2197 each_loose_cruft_fn cruft_cb,
2198 each_loose_subdir_fn subdir_cb,
2199 void *data)
2200 {
2201 size_t origlen, baselen;
2202 DIR *dir;
2203 struct dirent *de;
2204 int r = 0;
2205 struct object_id oid;
2206
2207 if (subdir_nr > 0xff)
2208 BUG("invalid loose object subdirectory: %x", subdir_nr);
2209
2210 origlen = path->len;
2211 strbuf_complete(path, '/');
2212 strbuf_addf(path, "%02x", subdir_nr);
2213
2214 dir = opendir(path->buf);
2215 if (!dir) {
2216 if (errno != ENOENT)
2217 r = error_errno(_("unable to open %s"), path->buf);
2218 strbuf_setlen(path, origlen);
2219 return r;
2220 }
2221
2222 oid.hash[0] = subdir_nr;
2223 strbuf_addch(path, '/');
2224 baselen = path->len;
2225
2226 while ((de = readdir(dir))) {
2227 size_t namelen;
2228 if (is_dot_or_dotdot(de->d_name))
2229 continue;
2230
2231 namelen = strlen(de->d_name);
2232 strbuf_setlen(path, baselen);
2233 strbuf_add(path, de->d_name, namelen);
2234 if (namelen == the_hash_algo->hexsz - 2 &&
2235 !hex_to_bytes(oid.hash + 1, de->d_name,
2236 the_hash_algo->rawsz - 1)) {
2237 if (obj_cb) {
2238 r = obj_cb(&oid, path->buf, data);
2239 if (r)
2240 break;
2241 }
2242 continue;
2243 }
2244
2245 if (cruft_cb) {
2246 r = cruft_cb(de->d_name, path->buf, data);
2247 if (r)
2248 break;
2249 }
2250 }
2251 closedir(dir);
2252
2253 strbuf_setlen(path, baselen - 1);
2254 if (!r && subdir_cb)
2255 r = subdir_cb(subdir_nr, path->buf, data);
2256
2257 strbuf_setlen(path, origlen);
2258
2259 return r;
2260 }
2261
2262 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2263 each_loose_object_fn obj_cb,
2264 each_loose_cruft_fn cruft_cb,
2265 each_loose_subdir_fn subdir_cb,
2266 void *data)
2267 {
2268 int r = 0;
2269 int i;
2270
2271 for (i = 0; i < 256; i++) {
2272 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2273 subdir_cb, data);
2274 if (r)
2275 break;
2276 }
2277
2278 return r;
2279 }
2280
2281 int for_each_loose_file_in_objdir(const char *path,
2282 each_loose_object_fn obj_cb,
2283 each_loose_cruft_fn cruft_cb,
2284 each_loose_subdir_fn subdir_cb,
2285 void *data)
2286 {
2287 struct strbuf buf = STRBUF_INIT;
2288 int r;
2289
2290 strbuf_addstr(&buf, path);
2291 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2292 subdir_cb, data);
2293 strbuf_release(&buf);
2294
2295 return r;
2296 }
2297
2298 int for_each_loose_object(each_loose_object_fn cb, void *data,
2299 enum for_each_object_flags flags)
2300 {
2301 struct object_directory *odb;
2302
2303 prepare_alt_odb(the_repository);
2304 for (odb = the_repository->objects->odb; odb; odb = odb->next) {
2305 int r = for_each_loose_file_in_objdir(odb->path, cb, NULL,
2306 NULL, data);
2307 if (r)
2308 return r;
2309
2310 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2311 break;
2312 }
2313
2314 return 0;
2315 }
2316
2317 static int append_loose_object(const struct object_id *oid, const char *path,
2318 void *data)
2319 {
2320 oid_array_append(data, oid);
2321 return 0;
2322 }
2323
2324 struct oid_array *odb_loose_cache(struct object_directory *odb,
2325 const struct object_id *oid)
2326 {
2327 int subdir_nr = oid->hash[0];
2328 struct strbuf buf = STRBUF_INIT;
2329
2330 if (subdir_nr < 0 ||
2331 subdir_nr >= ARRAY_SIZE(odb->loose_objects_subdir_seen))
2332 BUG("subdir_nr out of range");
2333
2334 if (odb->loose_objects_subdir_seen[subdir_nr])
2335 return &odb->loose_objects_cache[subdir_nr];
2336
2337 strbuf_addstr(&buf, odb->path);
2338 for_each_file_in_obj_subdir(subdir_nr, &buf,
2339 append_loose_object,
2340 NULL, NULL,
2341 &odb->loose_objects_cache[subdir_nr]);
2342 odb->loose_objects_subdir_seen[subdir_nr] = 1;
2343 strbuf_release(&buf);
2344 return &odb->loose_objects_cache[subdir_nr];
2345 }
2346
2347 void odb_clear_loose_cache(struct object_directory *odb)
2348 {
2349 int i;
2350
2351 for (i = 0; i < ARRAY_SIZE(odb->loose_objects_cache); i++)
2352 oid_array_clear(&odb->loose_objects_cache[i]);
2353 memset(&odb->loose_objects_subdir_seen, 0,
2354 sizeof(odb->loose_objects_subdir_seen));
2355 }
2356
2357 static int check_stream_oid(git_zstream *stream,
2358 const char *hdr,
2359 unsigned long size,
2360 const char *path,
2361 const struct object_id *expected_oid)
2362 {
2363 git_hash_ctx c;
2364 struct object_id real_oid;
2365 unsigned char buf[4096];
2366 unsigned long total_read;
2367 int status = Z_OK;
2368
2369 the_hash_algo->init_fn(&c);
2370 the_hash_algo->update_fn(&c, hdr, stream->total_out);
2371
2372 /*
2373 * We already read some bytes into hdr, but the ones up to the NUL
2374 * do not count against the object's content size.
2375 */
2376 total_read = stream->total_out - strlen(hdr) - 1;
2377
2378 /*
2379 * This size comparison must be "<=" to read the final zlib packets;
2380 * see the comment in unpack_loose_rest for details.
2381 */
2382 while (total_read <= size &&
2383 (status == Z_OK ||
2384 (status == Z_BUF_ERROR && !stream->avail_out))) {
2385 stream->next_out = buf;
2386 stream->avail_out = sizeof(buf);
2387 if (size - total_read < stream->avail_out)
2388 stream->avail_out = size - total_read;
2389 status = git_inflate(stream, Z_FINISH);
2390 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2391 total_read += stream->next_out - buf;
2392 }
2393 git_inflate_end(stream);
2394
2395 if (status != Z_STREAM_END) {
2396 error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid));
2397 return -1;
2398 }
2399 if (stream->avail_in) {
2400 error(_("garbage at end of loose object '%s'"),
2401 oid_to_hex(expected_oid));
2402 return -1;
2403 }
2404
2405 the_hash_algo->final_fn(real_oid.hash, &c);
2406 if (!oideq(expected_oid, &real_oid)) {
2407 error(_("hash mismatch for %s (expected %s)"), path,
2408 oid_to_hex(expected_oid));
2409 return -1;
2410 }
2411
2412 return 0;
2413 }
2414
2415 int read_loose_object(const char *path,
2416 const struct object_id *expected_oid,
2417 enum object_type *type,
2418 unsigned long *size,
2419 void **contents)
2420 {
2421 int ret = -1;
2422 void *map = NULL;
2423 unsigned long mapsize;
2424 git_zstream stream;
2425 char hdr[MAX_HEADER_LEN];
2426
2427 *contents = NULL;
2428
2429 map = map_loose_object_1(the_repository, path, NULL, &mapsize);
2430 if (!map) {
2431 error_errno(_("unable to mmap %s"), path);
2432 goto out;
2433 }
2434
2435 if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2436 error(_("unable to unpack header of %s"), path);
2437 goto out;
2438 }
2439
2440 *type = parse_loose_header(hdr, size);
2441 if (*type < 0) {
2442 error(_("unable to parse header of %s"), path);
2443 git_inflate_end(&stream);
2444 goto out;
2445 }
2446
2447 if (*type == OBJ_BLOB && *size > big_file_threshold) {
2448 if (check_stream_oid(&stream, hdr, *size, path, expected_oid) < 0)
2449 goto out;
2450 } else {
2451 *contents = unpack_loose_rest(&stream, hdr, *size, expected_oid);
2452 if (!*contents) {
2453 error(_("unable to unpack contents of %s"), path);
2454 git_inflate_end(&stream);
2455 goto out;
2456 }
2457 if (check_object_signature(expected_oid, *contents,
2458 *size, type_name(*type))) {
2459 error(_("hash mismatch for %s (expected %s)"), path,
2460 oid_to_hex(expected_oid));
2461 free(*contents);
2462 goto out;
2463 }
2464 }
2465
2466 ret = 0; /* everything checks out */
2467
2468 out:
2469 if (map)
2470 munmap(map, mapsize);
2471 return ret;
2472 }