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
58const unsigned char null_sha1[GIT_MAX_RAWSZ];
59const struct object_id null_oid;
60static const struct object_id empty_tree_oid = {
61 EMPTY_TREE_SHA1_BIN_LITERAL
62};
63static const struct object_id empty_blob_oid = {
64 EMPTY_BLOB_SHA1_BIN_LITERAL
65};
66static const struct object_id empty_tree_oid_sha256 = {
67 EMPTY_TREE_SHA256_BIN_LITERAL
68};
69static const struct object_id empty_blob_oid_sha256 = {
70 EMPTY_BLOB_SHA256_BIN_LITERAL
71};
72
73static void git_hash_sha1_init(git_hash_ctx *ctx)
74{
75 git_SHA1_Init(&ctx->sha1);
76}
77
78static 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
83static void git_hash_sha1_final(unsigned char *hash, git_hash_ctx *ctx)
84{
85 git_SHA1_Final(hash, &ctx->sha1);
86}
87
88
89static void git_hash_sha256_init(git_hash_ctx *ctx)
90{
91 git_SHA256_Init(&ctx->sha256);
92}
93
94static 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
99static void git_hash_sha256_final(unsigned char *hash, git_hash_ctx *ctx)
100{
101 git_SHA256_Final(hash, &ctx->sha256);
102}
103
104static void git_hash_unknown_init(git_hash_ctx *ctx)
105{
106 BUG("trying to init unknown hash");
107}
108
109static 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
114static void git_hash_unknown_final(unsigned char *hash, git_hash_ctx *ctx)
115{
116 BUG("trying to finalize unknown hash");
117}
118
119const 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
160const 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
166const 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
172int 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
183int 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
192int 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 */
207static struct cached_object {
208 struct object_id oid;
209 enum object_type type;
210 void *buf;
211 unsigned long size;
212} *cached_objects;
213static int cached_object_nr, cached_object_alloc;
214
215static struct cached_object empty_tree = {
216 { EMPTY_TREE_SHA1_BIN_LITERAL },
217 OBJ_TREE,
218 "",
219 0
220};
221
222static 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
237static 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
248int 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
277enum 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
331enum 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
344int 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
371retry_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
406out:
407 strbuf_release(&path_copy);
408 errno = save_errno;
409 return ret;
410}
411
412static 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
425static 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
436const 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 */
445static 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 */
488static void read_info_alternates(struct repository *r,
489 const char *relative_base,
490 int depth);
491static 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
537static 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
566static 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
597static 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
616void 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
659void 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 */
677char *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
738out:
739 if (seen_error) {
740 FREE_AND_NULL(ref_git);
741 }
742
743 return ref_git;
744}
745
746static 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
773static 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}
803
804struct alternate_refs_data {
805 alternate_ref_fn *fn;
806 void *data;
807};
808
809static int refs_from_alternate_cb(struct object_directory *e,
810 void *data)
811{
812 struct strbuf path = STRBUF_INIT;
813 size_t base_len;
814 struct alternate_refs_data *cb = data;
815
816 if (!strbuf_realpath(&path, e->path, 0))
817 goto out;
818 if (!strbuf_strip_suffix(&path, "/objects"))
819 goto out;
820 base_len = path.len;
821
822 /* Is this a git repository with refs? */
823 strbuf_addstr(&path, "/refs");
824 if (!is_directory(path.buf))
825 goto out;
826 strbuf_setlen(&path, base_len);
827
828 read_alternate_refs(path.buf, cb->fn, cb->data);
829
830out:
831 strbuf_release(&path);
832 return 0;
833}
834
835void for_each_alternate_ref(alternate_ref_fn fn, void *data)
836{
837 struct alternate_refs_data cb;
838 cb.fn = fn;
839 cb.data = data;
840 foreach_alt_odb(refs_from_alternate_cb, &cb);
841}
842
843int foreach_alt_odb(alt_odb_fn fn, void *cb)
844{
845 struct object_directory *ent;
846 int r = 0;
847
848 prepare_alt_odb(the_repository);
849 for (ent = the_repository->objects->odb->next; ent; ent = ent->next) {
850 r = fn(ent, cb);
851 if (r)
852 break;
853 }
854 return r;
855}
856
857void prepare_alt_odb(struct repository *r)
858{
859 if (r->objects->loaded_alternates)
860 return;
861
862 link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
863
864 read_info_alternates(r, r->objects->odb->path, 0);
865 r->objects->loaded_alternates = 1;
866}
867
868/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
869static int freshen_file(const char *fn)
870{
871 struct utimbuf t;
872 t.actime = t.modtime = time(NULL);
873 return !utime(fn, &t);
874}
875
876/*
877 * All of the check_and_freshen functions return 1 if the file exists and was
878 * freshened (if freshening was requested), 0 otherwise. If they return
879 * 0, you should not assume that it is safe to skip a write of the object (it
880 * either does not exist on disk, or has a stale mtime and may be subject to
881 * pruning).
882 */
883int check_and_freshen_file(const char *fn, int freshen)
884{
885 if (access(fn, F_OK))
886 return 0;
887 if (freshen && !freshen_file(fn))
888 return 0;
889 return 1;
890}
891
892static int check_and_freshen_odb(struct object_directory *odb,
893 const struct object_id *oid,
894 int freshen)
895{
896 static struct strbuf path = STRBUF_INIT;
897 odb_loose_path(odb, &path, oid);
898 return check_and_freshen_file(path.buf, freshen);
899}
900
901static int check_and_freshen_local(const struct object_id *oid, int freshen)
902{
903 return check_and_freshen_odb(the_repository->objects->odb, oid, freshen);
904}
905
906static int check_and_freshen_nonlocal(const struct object_id *oid, int freshen)
907{
908 struct object_directory *odb;
909
910 prepare_alt_odb(the_repository);
911 for (odb = the_repository->objects->odb->next; odb; odb = odb->next) {
912 if (check_and_freshen_odb(odb, oid, freshen))
913 return 1;
914 }
915 return 0;
916}
917
918static int check_and_freshen(const struct object_id *oid, int freshen)
919{
920 return check_and_freshen_local(oid, freshen) ||
921 check_and_freshen_nonlocal(oid, freshen);
922}
923
924int has_loose_object_nonlocal(const struct object_id *oid)
925{
926 return check_and_freshen_nonlocal(oid, 0);
927}
928
929static int has_loose_object(const struct object_id *oid)
930{
931 return check_and_freshen(oid, 0);
932}
933
934static void mmap_limit_check(size_t length)
935{
936 static size_t limit = 0;
937 if (!limit) {
938 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
939 if (!limit)
940 limit = SIZE_MAX;
941 }
942 if (length > limit)
943 die(_("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX),
944 (uintmax_t)length, (uintmax_t)limit);
945}
946
947void *xmmap_gently(void *start, size_t length,
948 int prot, int flags, int fd, off_t offset)
949{
950 void *ret;
951
952 mmap_limit_check(length);
953 ret = mmap(start, length, prot, flags, fd, offset);
954 if (ret == MAP_FAILED) {
955 if (!length)
956 return NULL;
957 release_pack_memory(length);
958 ret = mmap(start, length, prot, flags, fd, offset);
959 }
960 return ret;
961}
962
963void *xmmap(void *start, size_t length,
964 int prot, int flags, int fd, off_t offset)
965{
966 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
967 if (ret == MAP_FAILED)
968 die_errno(_("mmap failed"));
969 return ret;
970}
971
972/*
973 * With an in-core object data in "map", rehash it to make sure the
974 * object name actually matches "oid" to detect object corruption.
975 * With "map" == NULL, try reading the object named with "oid" using
976 * the streaming interface and rehash it to do the same.
977 */
978int check_object_signature(const struct object_id *oid, void *map,
979 unsigned long size, const char *type)
980{
981 struct object_id real_oid;
982 enum object_type obj_type;
983 struct git_istream *st;
984 git_hash_ctx c;
985 char hdr[MAX_HEADER_LEN];
986 int hdrlen;
987
988 if (map) {
989 hash_object_file(map, size, type, &real_oid);
990 return !oideq(oid, &real_oid) ? -1 : 0;
991 }
992
993 st = open_istream(oid, &obj_type, &size, NULL);
994 if (!st)
995 return -1;
996
997 /* Generate the header */
998 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(obj_type), (uintmax_t)size) + 1;
999
1000 /* Sha1.. */
1001 the_hash_algo->init_fn(&c);
1002 the_hash_algo->update_fn(&c, hdr, hdrlen);
1003 for (;;) {
1004 char buf[1024 * 16];
1005 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1006
1007 if (readlen < 0) {
1008 close_istream(st);
1009 return -1;
1010 }
1011 if (!readlen)
1012 break;
1013 the_hash_algo->update_fn(&c, buf, readlen);
1014 }
1015 the_hash_algo->final_fn(real_oid.hash, &c);
1016 close_istream(st);
1017 return !oideq(oid, &real_oid) ? -1 : 0;
1018}
1019
1020int git_open_cloexec(const char *name, int flags)
1021{
1022 int fd;
1023 static int o_cloexec = O_CLOEXEC;
1024
1025 fd = open(name, flags | o_cloexec);
1026 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1027 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1028 o_cloexec &= ~O_CLOEXEC;
1029 fd = open(name, flags | o_cloexec);
1030 }
1031
1032#if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1033 {
1034 static int fd_cloexec = FD_CLOEXEC;
1035
1036 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1037 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1038 int flags = fcntl(fd, F_GETFD);
1039 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1040 fd_cloexec = 0;
1041 }
1042 }
1043#endif
1044 return fd;
1045}
1046
1047/*
1048 * Find "oid" as a loose object in the local repository or in an alternate.
1049 * Returns 0 on success, negative on failure.
1050 *
1051 * The "path" out-parameter will give the path of the object we found (if any).
1052 * Note that it may point to static storage and is only valid until another
1053 * call to stat_loose_object().
1054 */
1055static int stat_loose_object(struct repository *r, const struct object_id *oid,
1056 struct stat *st, const char **path)
1057{
1058 struct object_directory *odb;
1059 static struct strbuf buf = STRBUF_INIT;
1060
1061 prepare_alt_odb(r);
1062 for (odb = r->objects->odb; odb; odb = odb->next) {
1063 *path = odb_loose_path(odb, &buf, oid);
1064 if (!lstat(*path, st))
1065 return 0;
1066 }
1067
1068 return -1;
1069}
1070
1071/*
1072 * Like stat_loose_object(), but actually open the object and return the
1073 * descriptor. See the caveats on the "path" parameter above.
1074 */
1075static int open_loose_object(struct repository *r,
1076 const struct object_id *oid, const char **path)
1077{
1078 int fd;
1079 struct object_directory *odb;
1080 int most_interesting_errno = ENOENT;
1081 static struct strbuf buf = STRBUF_INIT;
1082
1083 prepare_alt_odb(r);
1084 for (odb = r->objects->odb; odb; odb = odb->next) {
1085 *path = odb_loose_path(odb, &buf, oid);
1086 fd = git_open(*path);
1087 if (fd >= 0)
1088 return fd;
1089
1090 if (most_interesting_errno == ENOENT)
1091 most_interesting_errno = errno;
1092 }
1093 errno = most_interesting_errno;
1094 return -1;
1095}
1096
1097static int quick_has_loose(struct repository *r,
1098 const struct object_id *oid)
1099{
1100 struct object_directory *odb;
1101
1102 prepare_alt_odb(r);
1103 for (odb = r->objects->odb; odb; odb = odb->next) {
1104 if (oid_array_lookup(odb_loose_cache(odb, oid), oid) >= 0)
1105 return 1;
1106 }
1107 return 0;
1108}
1109
1110/*
1111 * Map the loose object at "path" if it is not NULL, or the path found by
1112 * searching for a loose object named "oid".
1113 */
1114static void *map_loose_object_1(struct repository *r, const char *path,
1115 const struct object_id *oid, unsigned long *size)
1116{
1117 void *map;
1118 int fd;
1119
1120 if (path)
1121 fd = git_open(path);
1122 else
1123 fd = open_loose_object(r, oid, &path);
1124 map = NULL;
1125 if (fd >= 0) {
1126 struct stat st;
1127
1128 if (!fstat(fd, &st)) {
1129 *size = xsize_t(st.st_size);
1130 if (!*size) {
1131 /* mmap() is forbidden on empty files */
1132 error(_("object file %s is empty"), path);
1133 close(fd);
1134 return NULL;
1135 }
1136 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1137 }
1138 close(fd);
1139 }
1140 return map;
1141}
1142
1143void *map_loose_object(struct repository *r,
1144 const struct object_id *oid,
1145 unsigned long *size)
1146{
1147 return map_loose_object_1(r, NULL, oid, size);
1148}
1149
1150static int unpack_loose_short_header(git_zstream *stream,
1151 unsigned char *map, unsigned long mapsize,
1152 void *buffer, unsigned long bufsiz)
1153{
1154 /* Get the data stream */
1155 memset(stream, 0, sizeof(*stream));
1156 stream->next_in = map;
1157 stream->avail_in = mapsize;
1158 stream->next_out = buffer;
1159 stream->avail_out = bufsiz;
1160
1161 git_inflate_init(stream);
1162 return git_inflate(stream, 0);
1163}
1164
1165int unpack_loose_header(git_zstream *stream,
1166 unsigned char *map, unsigned long mapsize,
1167 void *buffer, unsigned long bufsiz)
1168{
1169 int status = unpack_loose_short_header(stream, map, mapsize,
1170 buffer, bufsiz);
1171
1172 if (status < Z_OK)
1173 return status;
1174
1175 /* Make sure we have the terminating NUL */
1176 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1177 return -1;
1178 return 0;
1179}
1180
1181static int unpack_loose_header_to_strbuf(git_zstream *stream, unsigned char *map,
1182 unsigned long mapsize, void *buffer,
1183 unsigned long bufsiz, struct strbuf *header)
1184{
1185 int status;
1186
1187 status = unpack_loose_short_header(stream, map, mapsize, buffer, bufsiz);
1188 if (status < Z_OK)
1189 return -1;
1190
1191 /*
1192 * Check if entire header is unpacked in the first iteration.
1193 */
1194 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1195 return 0;
1196
1197 /*
1198 * buffer[0..bufsiz] was not large enough. Copy the partial
1199 * result out to header, and then append the result of further
1200 * reading the stream.
1201 */
1202 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1203 stream->next_out = buffer;
1204 stream->avail_out = bufsiz;
1205
1206 do {
1207 status = git_inflate(stream, 0);
1208 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1209 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1210 return 0;
1211 stream->next_out = buffer;
1212 stream->avail_out = bufsiz;
1213 } while (status != Z_STREAM_END);
1214 return -1;
1215}
1216
1217static void *unpack_loose_rest(git_zstream *stream,
1218 void *buffer, unsigned long size,
1219 const struct object_id *oid)
1220{
1221 int bytes = strlen(buffer) + 1;
1222 unsigned char *buf = xmallocz(size);
1223 unsigned long n;
1224 int status = Z_OK;
1225
1226 n = stream->total_out - bytes;
1227 if (n > size)
1228 n = size;
1229 memcpy(buf, (char *) buffer + bytes, n);
1230 bytes = n;
1231 if (bytes <= size) {
1232 /*
1233 * The above condition must be (bytes <= size), not
1234 * (bytes < size). In other words, even though we
1235 * expect no more output and set avail_out to zero,
1236 * the input zlib stream may have bytes that express
1237 * "this concludes the stream", and we *do* want to
1238 * eat that input.
1239 *
1240 * Otherwise we would not be able to test that we
1241 * consumed all the input to reach the expected size;
1242 * we also want to check that zlib tells us that all
1243 * went well with status == Z_STREAM_END at the end.
1244 */
1245 stream->next_out = buf + bytes;
1246 stream->avail_out = size - bytes;
1247 while (status == Z_OK)
1248 status = git_inflate(stream, Z_FINISH);
1249 }
1250 if (status == Z_STREAM_END && !stream->avail_in) {
1251 git_inflate_end(stream);
1252 return buf;
1253 }
1254
1255 if (status < 0)
1256 error(_("corrupt loose object '%s'"), oid_to_hex(oid));
1257 else if (stream->avail_in)
1258 error(_("garbage at end of loose object '%s'"),
1259 oid_to_hex(oid));
1260 free(buf);
1261 return NULL;
1262}
1263
1264/*
1265 * We used to just use "sscanf()", but that's actually way
1266 * too permissive for what we want to check. So do an anal
1267 * object header parse by hand.
1268 */
1269static int parse_loose_header_extended(const char *hdr, struct object_info *oi,
1270 unsigned int flags)
1271{
1272 const char *type_buf = hdr;
1273 unsigned long size;
1274 int type, type_len = 0;
1275
1276 /*
1277 * The type can be of any size but is followed by
1278 * a space.
1279 */
1280 for (;;) {
1281 char c = *hdr++;
1282 if (!c)
1283 return -1;
1284 if (c == ' ')
1285 break;
1286 type_len++;
1287 }
1288
1289 type = type_from_string_gently(type_buf, type_len, 1);
1290 if (oi->type_name)
1291 strbuf_add(oi->type_name, type_buf, type_len);
1292 /*
1293 * Set type to 0 if its an unknown object and
1294 * we're obtaining the type using '--allow-unknown-type'
1295 * option.
1296 */
1297 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1298 type = 0;
1299 else if (type < 0)
1300 die(_("invalid object type"));
1301 if (oi->typep)
1302 *oi->typep = type;
1303
1304 /*
1305 * The length must follow immediately, and be in canonical
1306 * decimal format (ie "010" is not valid).
1307 */
1308 size = *hdr++ - '0';
1309 if (size > 9)
1310 return -1;
1311 if (size) {
1312 for (;;) {
1313 unsigned long c = *hdr - '0';
1314 if (c > 9)
1315 break;
1316 hdr++;
1317 size = size * 10 + c;
1318 }
1319 }
1320
1321 if (oi->sizep)
1322 *oi->sizep = size;
1323
1324 /*
1325 * The length must be followed by a zero byte
1326 */
1327 return *hdr ? -1 : type;
1328}
1329
1330int parse_loose_header(const char *hdr, unsigned long *sizep)
1331{
1332 struct object_info oi = OBJECT_INFO_INIT;
1333
1334 oi.sizep = sizep;
1335 return parse_loose_header_extended(hdr, &oi, 0);
1336}
1337
1338static int loose_object_info(struct repository *r,
1339 const struct object_id *oid,
1340 struct object_info *oi, int flags)
1341{
1342 int status = 0;
1343 unsigned long mapsize;
1344 void *map;
1345 git_zstream stream;
1346 char hdr[MAX_HEADER_LEN];
1347 struct strbuf hdrbuf = STRBUF_INIT;
1348 unsigned long size_scratch;
1349
1350 if (oi->delta_base_sha1)
1351 hashclr(oi->delta_base_sha1);
1352
1353 /*
1354 * If we don't care about type or size, then we don't
1355 * need to look inside the object at all. Note that we
1356 * do not optimize out the stat call, even if the
1357 * caller doesn't care about the disk-size, since our
1358 * return value implicitly indicates whether the
1359 * object even exists.
1360 */
1361 if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1362 const char *path;
1363 struct stat st;
1364 if (!oi->disk_sizep && (flags & OBJECT_INFO_QUICK))
1365 return quick_has_loose(r, oid) ? 0 : -1;
1366 if (stat_loose_object(r, oid, &st, &path) < 0)
1367 return -1;
1368 if (oi->disk_sizep)
1369 *oi->disk_sizep = st.st_size;
1370 return 0;
1371 }
1372
1373 map = map_loose_object(r, oid, &mapsize);
1374 if (!map)
1375 return -1;
1376
1377 if (!oi->sizep)
1378 oi->sizep = &size_scratch;
1379
1380 if (oi->disk_sizep)
1381 *oi->disk_sizep = mapsize;
1382 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1383 if (unpack_loose_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1384 status = error(_("unable to unpack %s header with --allow-unknown-type"),
1385 oid_to_hex(oid));
1386 } else if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1387 status = error(_("unable to unpack %s header"),
1388 oid_to_hex(oid));
1389 if (status < 0)
1390 ; /* Do nothing */
1391 else if (hdrbuf.len) {
1392 if ((status = parse_loose_header_extended(hdrbuf.buf, oi, flags)) < 0)
1393 status = error(_("unable to parse %s header with --allow-unknown-type"),
1394 oid_to_hex(oid));
1395 } else if ((status = parse_loose_header_extended(hdr, oi, flags)) < 0)
1396 status = error(_("unable to parse %s header"), oid_to_hex(oid));
1397
1398 if (status >= 0 && oi->contentp) {
1399 *oi->contentp = unpack_loose_rest(&stream, hdr,
1400 *oi->sizep, oid);
1401 if (!*oi->contentp) {
1402 git_inflate_end(&stream);
1403 status = -1;
1404 }
1405 } else
1406 git_inflate_end(&stream);
1407
1408 munmap(map, mapsize);
1409 if (status && oi->typep)
1410 *oi->typep = status;
1411 if (oi->sizep == &size_scratch)
1412 oi->sizep = NULL;
1413 strbuf_release(&hdrbuf);
1414 oi->whence = OI_LOOSE;
1415 return (status < 0) ? status : 0;
1416}
1417
1418int fetch_if_missing = 1;
1419
1420int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1421 struct object_info *oi, unsigned flags)
1422{
1423 static struct object_info blank_oi = OBJECT_INFO_INIT;
1424 struct pack_entry e;
1425 int rtype;
1426 const struct object_id *real = oid;
1427 int already_retried = 0;
1428
1429 if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1430 real = lookup_replace_object(r, oid);
1431
1432 if (is_null_oid(real))
1433 return -1;
1434
1435 if (!oi)
1436 oi = &blank_oi;
1437
1438 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1439 struct cached_object *co = find_cached_object(real);
1440 if (co) {
1441 if (oi->typep)
1442 *(oi->typep) = co->type;
1443 if (oi->sizep)
1444 *(oi->sizep) = co->size;
1445 if (oi->disk_sizep)
1446 *(oi->disk_sizep) = 0;
1447 if (oi->delta_base_sha1)
1448 hashclr(oi->delta_base_sha1);
1449 if (oi->type_name)
1450 strbuf_addstr(oi->type_name, type_name(co->type));
1451 if (oi->contentp)
1452 *oi->contentp = xmemdupz(co->buf, co->size);
1453 oi->whence = OI_CACHED;
1454 return 0;
1455 }
1456 }
1457
1458 while (1) {
1459 if (find_pack_entry(r, real, &e))
1460 break;
1461
1462 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1463 return -1;
1464
1465 /* Most likely it's a loose object. */
1466 if (!loose_object_info(r, real, oi, flags))
1467 return 0;
1468
1469 /* Not a loose object; someone else may have just packed it. */
1470 if (!(flags & OBJECT_INFO_QUICK)) {
1471 reprepare_packed_git(r);
1472 if (find_pack_entry(r, real, &e))
1473 break;
1474 }
1475
1476 /* Check if it is a missing object */
1477 if (fetch_if_missing && repository_format_partial_clone &&
1478 !already_retried && r == the_repository &&
1479 !(flags & OBJECT_INFO_SKIP_FETCH_OBJECT)) {
1480 /*
1481 * TODO Investigate having fetch_object() return
1482 * TODO error/success and stopping the music here.
1483 * TODO Pass a repository struct through fetch_object,
1484 * such that arbitrary repositories work.
1485 */
1486 fetch_objects(repository_format_partial_clone, real, 1);
1487 already_retried = 1;
1488 continue;
1489 }
1490
1491 return -1;
1492 }
1493
1494 if (oi == &blank_oi)
1495 /*
1496 * We know that the caller doesn't actually need the
1497 * information below, so return early.
1498 */
1499 return 0;
1500 rtype = packed_object_info(r, e.p, e.offset, oi);
1501 if (rtype < 0) {
1502 mark_bad_packed_object(e.p, real->hash);
1503 return oid_object_info_extended(r, real, oi, 0);
1504 } else if (oi->whence == OI_PACKED) {
1505 oi->u.packed.offset = e.offset;
1506 oi->u.packed.pack = e.p;
1507 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1508 rtype == OBJ_OFS_DELTA);
1509 }
1510
1511 return 0;
1512}
1513
1514/* returns enum object_type or negative */
1515int oid_object_info(struct repository *r,
1516 const struct object_id *oid,
1517 unsigned long *sizep)
1518{
1519 enum object_type type;
1520 struct object_info oi = OBJECT_INFO_INIT;
1521
1522 oi.typep = &type;
1523 oi.sizep = sizep;
1524 if (oid_object_info_extended(r, oid, &oi,
1525 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1526 return -1;
1527 return type;
1528}
1529
1530static void *read_object(struct repository *r,
1531 const struct object_id *oid, enum object_type *type,
1532 unsigned long *size)
1533{
1534 struct object_info oi = OBJECT_INFO_INIT;
1535 void *content;
1536 oi.typep = type;
1537 oi.sizep = size;
1538 oi.contentp = &content;
1539
1540 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1541 return NULL;
1542 return content;
1543}
1544
1545int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1546 struct object_id *oid)
1547{
1548 struct cached_object *co;
1549
1550 hash_object_file(buf, len, type_name(type), oid);
1551 if (has_object_file(oid) || find_cached_object(oid))
1552 return 0;
1553 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1554 co = &cached_objects[cached_object_nr++];
1555 co->size = len;
1556 co->type = type;
1557 co->buf = xmalloc(len);
1558 memcpy(co->buf, buf, len);
1559 oidcpy(&co->oid, oid);
1560 return 0;
1561}
1562
1563/*
1564 * This function dies on corrupt objects; the callers who want to
1565 * deal with them should arrange to call read_object() and give error
1566 * messages themselves.
1567 */
1568void *read_object_file_extended(struct repository *r,
1569 const struct object_id *oid,
1570 enum object_type *type,
1571 unsigned long *size,
1572 int lookup_replace)
1573{
1574 void *data;
1575 const struct packed_git *p;
1576 const char *path;
1577 struct stat st;
1578 const struct object_id *repl = lookup_replace ?
1579 lookup_replace_object(r, oid) : oid;
1580
1581 errno = 0;
1582 data = read_object(r, repl, type, size);
1583 if (data)
1584 return data;
1585
1586 if (errno && errno != ENOENT)
1587 die_errno(_("failed to read object %s"), oid_to_hex(oid));
1588
1589 /* die if we replaced an object with one that does not exist */
1590 if (repl != oid)
1591 die(_("replacement %s not found for %s"),
1592 oid_to_hex(repl), oid_to_hex(oid));
1593
1594 if (!stat_loose_object(r, repl, &st, &path))
1595 die(_("loose object %s (stored in %s) is corrupt"),
1596 oid_to_hex(repl), path);
1597
1598 if ((p = has_packed_and_bad(r, repl->hash)) != NULL)
1599 die(_("packed object %s (stored in %s) is corrupt"),
1600 oid_to_hex(repl), p->pack_name);
1601
1602 return NULL;
1603}
1604
1605void *read_object_with_reference(struct repository *r,
1606 const struct object_id *oid,
1607 const char *required_type_name,
1608 unsigned long *size,
1609 struct object_id *actual_oid_return)
1610{
1611 enum object_type type, required_type;
1612 void *buffer;
1613 unsigned long isize;
1614 struct object_id actual_oid;
1615
1616 required_type = type_from_string(required_type_name);
1617 oidcpy(&actual_oid, oid);
1618 while (1) {
1619 int ref_length = -1;
1620 const char *ref_type = NULL;
1621
1622 buffer = repo_read_object_file(r, &actual_oid, &type, &isize);
1623 if (!buffer)
1624 return NULL;
1625 if (type == required_type) {
1626 *size = isize;
1627 if (actual_oid_return)
1628 oidcpy(actual_oid_return, &actual_oid);
1629 return buffer;
1630 }
1631 /* Handle references */
1632 else if (type == OBJ_COMMIT)
1633 ref_type = "tree ";
1634 else if (type == OBJ_TAG)
1635 ref_type = "object ";
1636 else {
1637 free(buffer);
1638 return NULL;
1639 }
1640 ref_length = strlen(ref_type);
1641
1642 if (ref_length + the_hash_algo->hexsz > isize ||
1643 memcmp(buffer, ref_type, ref_length) ||
1644 get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1645 free(buffer);
1646 return NULL;
1647 }
1648 free(buffer);
1649 /* Now we have the ID of the referred-to object in
1650 * actual_oid. Check again. */
1651 }
1652}
1653
1654static void write_object_file_prepare(const void *buf, unsigned long len,
1655 const char *type, struct object_id *oid,
1656 char *hdr, int *hdrlen)
1657{
1658 git_hash_ctx c;
1659
1660 /* Generate the header */
1661 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %"PRIuMAX , type, (uintmax_t)len)+1;
1662
1663 /* Sha1.. */
1664 the_hash_algo->init_fn(&c);
1665 the_hash_algo->update_fn(&c, hdr, *hdrlen);
1666 the_hash_algo->update_fn(&c, buf, len);
1667 the_hash_algo->final_fn(oid->hash, &c);
1668}
1669
1670/*
1671 * Move the just written object into its final resting place.
1672 */
1673int finalize_object_file(const char *tmpfile, const char *filename)
1674{
1675 int ret = 0;
1676
1677 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1678 goto try_rename;
1679 else if (link(tmpfile, filename))
1680 ret = errno;
1681
1682 /*
1683 * Coda hack - coda doesn't like cross-directory links,
1684 * so we fall back to a rename, which will mean that it
1685 * won't be able to check collisions, but that's not a
1686 * big deal.
1687 *
1688 * The same holds for FAT formatted media.
1689 *
1690 * When this succeeds, we just return. We have nothing
1691 * left to unlink.
1692 */
1693 if (ret && ret != EEXIST) {
1694 try_rename:
1695 if (!rename(tmpfile, filename))
1696 goto out;
1697 ret = errno;
1698 }
1699 unlink_or_warn(tmpfile);
1700 if (ret) {
1701 if (ret != EEXIST) {
1702 return error_errno(_("unable to write file %s"), filename);
1703 }
1704 /* FIXME!!! Collision check here ? */
1705 }
1706
1707out:
1708 if (adjust_shared_perm(filename))
1709 return error(_("unable to set permission to '%s'"), filename);
1710 return 0;
1711}
1712
1713static int write_buffer(int fd, const void *buf, size_t len)
1714{
1715 if (write_in_full(fd, buf, len) < 0)
1716 return error_errno(_("file write error"));
1717 return 0;
1718}
1719
1720int hash_object_file(const void *buf, unsigned long len, const char *type,
1721 struct object_id *oid)
1722{
1723 char hdr[MAX_HEADER_LEN];
1724 int hdrlen = sizeof(hdr);
1725 write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1726 return 0;
1727}
1728
1729/* Finalize a file on disk, and close it. */
1730static void close_loose_object(int fd)
1731{
1732 if (fsync_object_files)
1733 fsync_or_die(fd, "loose object file");
1734 if (close(fd) != 0)
1735 die_errno(_("error when closing loose object file"));
1736}
1737
1738/* Size of directory component, including the ending '/' */
1739static inline int directory_size(const char *filename)
1740{
1741 const char *s = strrchr(filename, '/');
1742 if (!s)
1743 return 0;
1744 return s - filename + 1;
1745}
1746
1747/*
1748 * This creates a temporary file in the same directory as the final
1749 * 'filename'
1750 *
1751 * We want to avoid cross-directory filename renames, because those
1752 * can have problems on various filesystems (FAT, NFS, Coda).
1753 */
1754static int create_tmpfile(struct strbuf *tmp, const char *filename)
1755{
1756 int fd, dirlen = directory_size(filename);
1757
1758 strbuf_reset(tmp);
1759 strbuf_add(tmp, filename, dirlen);
1760 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1761 fd = git_mkstemp_mode(tmp->buf, 0444);
1762 if (fd < 0 && dirlen && errno == ENOENT) {
1763 /*
1764 * Make sure the directory exists; note that the contents
1765 * of the buffer are undefined after mkstemp returns an
1766 * error, so we have to rewrite the whole buffer from
1767 * scratch.
1768 */
1769 strbuf_reset(tmp);
1770 strbuf_add(tmp, filename, dirlen - 1);
1771 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1772 return -1;
1773 if (adjust_shared_perm(tmp->buf))
1774 return -1;
1775
1776 /* Try again */
1777 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1778 fd = git_mkstemp_mode(tmp->buf, 0444);
1779 }
1780 return fd;
1781}
1782
1783static int write_loose_object(const struct object_id *oid, char *hdr,
1784 int hdrlen, const void *buf, unsigned long len,
1785 time_t mtime)
1786{
1787 int fd, ret;
1788 unsigned char compressed[4096];
1789 git_zstream stream;
1790 git_hash_ctx c;
1791 struct object_id parano_oid;
1792 static struct strbuf tmp_file = STRBUF_INIT;
1793 static struct strbuf filename = STRBUF_INIT;
1794
1795 loose_object_path(the_repository, &filename, oid);
1796
1797 fd = create_tmpfile(&tmp_file, filename.buf);
1798 if (fd < 0) {
1799 if (errno == EACCES)
1800 return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
1801 else
1802 return error_errno(_("unable to create temporary file"));
1803 }
1804
1805 /* Set it up */
1806 git_deflate_init(&stream, zlib_compression_level);
1807 stream.next_out = compressed;
1808 stream.avail_out = sizeof(compressed);
1809 the_hash_algo->init_fn(&c);
1810
1811 /* First header.. */
1812 stream.next_in = (unsigned char *)hdr;
1813 stream.avail_in = hdrlen;
1814 while (git_deflate(&stream, 0) == Z_OK)
1815 ; /* nothing */
1816 the_hash_algo->update_fn(&c, hdr, hdrlen);
1817
1818 /* Then the data itself.. */
1819 stream.next_in = (void *)buf;
1820 stream.avail_in = len;
1821 do {
1822 unsigned char *in0 = stream.next_in;
1823 ret = git_deflate(&stream, Z_FINISH);
1824 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1825 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1826 die(_("unable to write loose object file"));
1827 stream.next_out = compressed;
1828 stream.avail_out = sizeof(compressed);
1829 } while (ret == Z_OK);
1830
1831 if (ret != Z_STREAM_END)
1832 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
1833 ret);
1834 ret = git_deflate_end_gently(&stream);
1835 if (ret != Z_OK)
1836 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
1837 ret);
1838 the_hash_algo->final_fn(parano_oid.hash, &c);
1839 if (!oideq(oid, ¶no_oid))
1840 die(_("confused by unstable object source data for %s"),
1841 oid_to_hex(oid));
1842
1843 close_loose_object(fd);
1844
1845 if (mtime) {
1846 struct utimbuf utb;
1847 utb.actime = mtime;
1848 utb.modtime = mtime;
1849 if (utime(tmp_file.buf, &utb) < 0)
1850 warning_errno(_("failed utime() on %s"), tmp_file.buf);
1851 }
1852
1853 return finalize_object_file(tmp_file.buf, filename.buf);
1854}
1855
1856static int freshen_loose_object(const struct object_id *oid)
1857{
1858 return check_and_freshen(oid, 1);
1859}
1860
1861static int freshen_packed_object(const struct object_id *oid)
1862{
1863 struct pack_entry e;
1864 if (!find_pack_entry(the_repository, oid, &e))
1865 return 0;
1866 if (e.p->freshened)
1867 return 1;
1868 if (!freshen_file(e.p->pack_name))
1869 return 0;
1870 e.p->freshened = 1;
1871 return 1;
1872}
1873
1874int write_object_file(const void *buf, unsigned long len, const char *type,
1875 struct object_id *oid)
1876{
1877 char hdr[MAX_HEADER_LEN];
1878 int hdrlen = sizeof(hdr);
1879
1880 /* Normally if we have it in the pack then we do not bother writing
1881 * it out into .git/objects/??/?{38} file.
1882 */
1883 write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1884 if (freshen_packed_object(oid) || freshen_loose_object(oid))
1885 return 0;
1886 return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
1887}
1888
1889int hash_object_file_literally(const void *buf, unsigned long len,
1890 const char *type, struct object_id *oid,
1891 unsigned flags)
1892{
1893 char *header;
1894 int hdrlen, status = 0;
1895
1896 /* type string, SP, %lu of the length plus NUL must fit this */
1897 hdrlen = strlen(type) + MAX_HEADER_LEN;
1898 header = xmalloc(hdrlen);
1899 write_object_file_prepare(buf, len, type, oid, header, &hdrlen);
1900
1901 if (!(flags & HASH_WRITE_OBJECT))
1902 goto cleanup;
1903 if (freshen_packed_object(oid) || freshen_loose_object(oid))
1904 goto cleanup;
1905 status = write_loose_object(oid, header, hdrlen, buf, len, 0);
1906
1907cleanup:
1908 free(header);
1909 return status;
1910}
1911
1912int force_object_loose(const struct object_id *oid, time_t mtime)
1913{
1914 void *buf;
1915 unsigned long len;
1916 enum object_type type;
1917 char hdr[MAX_HEADER_LEN];
1918 int hdrlen;
1919 int ret;
1920
1921 if (has_loose_object(oid))
1922 return 0;
1923 buf = read_object(the_repository, oid, &type, &len);
1924 if (!buf)
1925 return error(_("cannot read object for %s"), oid_to_hex(oid));
1926 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(type), (uintmax_t)len) + 1;
1927 ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
1928 free(buf);
1929
1930 return ret;
1931}
1932
1933int repo_has_object_file_with_flags(struct repository *r,
1934 const struct object_id *oid, int flags)
1935{
1936 if (!startup_info->have_repository)
1937 return 0;
1938 return oid_object_info_extended(r, oid, NULL,
1939 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1940}
1941
1942int repo_has_object_file(struct repository *r,
1943 const struct object_id *oid)
1944{
1945 return repo_has_object_file_with_flags(r, oid, 0);
1946}
1947
1948static void check_tree(const void *buf, size_t size)
1949{
1950 struct tree_desc desc;
1951 struct name_entry entry;
1952
1953 init_tree_desc(&desc, buf, size);
1954 while (tree_entry(&desc, &entry))
1955 /* do nothing
1956 * tree_entry() will die() on malformed entries */
1957 ;
1958}
1959
1960static void check_commit(const void *buf, size_t size)
1961{
1962 struct commit c;
1963 memset(&c, 0, sizeof(c));
1964 if (parse_commit_buffer(the_repository, &c, buf, size, 0))
1965 die(_("corrupt commit"));
1966}
1967
1968static void check_tag(const void *buf, size_t size)
1969{
1970 struct tag t;
1971 memset(&t, 0, sizeof(t));
1972 if (parse_tag_buffer(the_repository, &t, buf, size))
1973 die(_("corrupt tag"));
1974}
1975
1976static int index_mem(struct index_state *istate,
1977 struct object_id *oid, void *buf, size_t size,
1978 enum object_type type,
1979 const char *path, unsigned flags)
1980{
1981 int ret, re_allocated = 0;
1982 int write_object = flags & HASH_WRITE_OBJECT;
1983
1984 if (!type)
1985 type = OBJ_BLOB;
1986
1987 /*
1988 * Convert blobs to git internal format
1989 */
1990 if ((type == OBJ_BLOB) && path) {
1991 struct strbuf nbuf = STRBUF_INIT;
1992 if (convert_to_git(istate, path, buf, size, &nbuf,
1993 get_conv_flags(flags))) {
1994 buf = strbuf_detach(&nbuf, &size);
1995 re_allocated = 1;
1996 }
1997 }
1998 if (flags & HASH_FORMAT_CHECK) {
1999 if (type == OBJ_TREE)
2000 check_tree(buf, size);
2001 if (type == OBJ_COMMIT)
2002 check_commit(buf, size);
2003 if (type == OBJ_TAG)
2004 check_tag(buf, size);
2005 }
2006
2007 if (write_object)
2008 ret = write_object_file(buf, size, type_name(type), oid);
2009 else
2010 ret = hash_object_file(buf, size, type_name(type), oid);
2011 if (re_allocated)
2012 free(buf);
2013 return ret;
2014}
2015
2016static int index_stream_convert_blob(struct index_state *istate,
2017 struct object_id *oid,
2018 int fd,
2019 const char *path,
2020 unsigned flags)
2021{
2022 int ret;
2023 const int write_object = flags & HASH_WRITE_OBJECT;
2024 struct strbuf sbuf = STRBUF_INIT;
2025
2026 assert(path);
2027 assert(would_convert_to_git_filter_fd(istate, path));
2028
2029 convert_to_git_filter_fd(istate, path, fd, &sbuf,
2030 get_conv_flags(flags));
2031
2032 if (write_object)
2033 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
2034 oid);
2035 else
2036 ret = hash_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
2037 oid);
2038 strbuf_release(&sbuf);
2039 return ret;
2040}
2041
2042static int index_pipe(struct index_state *istate, struct object_id *oid,
2043 int fd, enum object_type type,
2044 const char *path, unsigned flags)
2045{
2046 struct strbuf sbuf = STRBUF_INIT;
2047 int ret;
2048
2049 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2050 ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags);
2051 else
2052 ret = -1;
2053 strbuf_release(&sbuf);
2054 return ret;
2055}
2056
2057#define SMALL_FILE_SIZE (32*1024)
2058
2059static int index_core(struct index_state *istate,
2060 struct object_id *oid, int fd, size_t size,
2061 enum object_type type, const char *path,
2062 unsigned flags)
2063{
2064 int ret;
2065
2066 if (!size) {
2067 ret = index_mem(istate, oid, "", size, type, path, flags);
2068 } else if (size <= SMALL_FILE_SIZE) {
2069 char *buf = xmalloc(size);
2070 ssize_t read_result = read_in_full(fd, buf, size);
2071 if (read_result < 0)
2072 ret = error_errno(_("read error while indexing %s"),
2073 path ? path : "<unknown>");
2074 else if (read_result != size)
2075 ret = error(_("short read while indexing %s"),
2076 path ? path : "<unknown>");
2077 else
2078 ret = index_mem(istate, oid, buf, size, type, path, flags);
2079 free(buf);
2080 } else {
2081 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2082 ret = index_mem(istate, oid, buf, size, type, path, flags);
2083 munmap(buf, size);
2084 }
2085 return ret;
2086}
2087
2088/*
2089 * This creates one packfile per large blob unless bulk-checkin
2090 * machinery is "plugged".
2091 *
2092 * This also bypasses the usual "convert-to-git" dance, and that is on
2093 * purpose. We could write a streaming version of the converting
2094 * functions and insert that before feeding the data to fast-import
2095 * (or equivalent in-core API described above). However, that is
2096 * somewhat complicated, as we do not know the size of the filter
2097 * result, which we need to know beforehand when writing a git object.
2098 * Since the primary motivation for trying to stream from the working
2099 * tree file and to avoid mmaping it in core is to deal with large
2100 * binary blobs, they generally do not want to get any conversion, and
2101 * callers should avoid this code path when filters are requested.
2102 */
2103static int index_stream(struct object_id *oid, int fd, size_t size,
2104 enum object_type type, const char *path,
2105 unsigned flags)
2106{
2107 return index_bulk_checkin(oid, fd, size, type, path, flags);
2108}
2109
2110int index_fd(struct index_state *istate, struct object_id *oid,
2111 int fd, struct stat *st,
2112 enum object_type type, const char *path, unsigned flags)
2113{
2114 int ret;
2115
2116 /*
2117 * Call xsize_t() only when needed to avoid potentially unnecessary
2118 * die() for large files.
2119 */
2120 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, path))
2121 ret = index_stream_convert_blob(istate, oid, fd, path, flags);
2122 else if (!S_ISREG(st->st_mode))
2123 ret = index_pipe(istate, oid, fd, type, path, flags);
2124 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
2125 (path && would_convert_to_git(istate, path)))
2126 ret = index_core(istate, oid, fd, xsize_t(st->st_size),
2127 type, path, flags);
2128 else
2129 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
2130 flags);
2131 close(fd);
2132 return ret;
2133}
2134
2135int index_path(struct index_state *istate, struct object_id *oid,
2136 const char *path, struct stat *st, unsigned flags)
2137{
2138 int fd;
2139 struct strbuf sb = STRBUF_INIT;
2140 int rc = 0;
2141
2142 switch (st->st_mode & S_IFMT) {
2143 case S_IFREG:
2144 fd = open(path, O_RDONLY);
2145 if (fd < 0)
2146 return error_errno("open(\"%s\")", path);
2147 if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0)
2148 return error(_("%s: failed to insert into database"),
2149 path);
2150 break;
2151 case S_IFLNK:
2152 if (strbuf_readlink(&sb, path, st->st_size))
2153 return error_errno("readlink(\"%s\")", path);
2154 if (!(flags & HASH_WRITE_OBJECT))
2155 hash_object_file(sb.buf, sb.len, blob_type, oid);
2156 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
2157 rc = error(_("%s: failed to insert into database"), path);
2158 strbuf_release(&sb);
2159 break;
2160 case S_IFDIR:
2161 return resolve_gitlink_ref(path, "HEAD", oid);
2162 default:
2163 return error(_("%s: unsupported file type"), path);
2164 }
2165 return rc;
2166}
2167
2168int read_pack_header(int fd, struct pack_header *header)
2169{
2170 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
2171 /* "eof before pack header was fully read" */
2172 return PH_ERROR_EOF;
2173
2174 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2175 /* "protocol error (pack signature mismatch detected)" */
2176 return PH_ERROR_PACK_SIGNATURE;
2177 if (!pack_version_ok(header->hdr_version))
2178 /* "protocol error (pack version unsupported)" */
2179 return PH_ERROR_PROTOCOL;
2180 return 0;
2181}
2182
2183void assert_oid_type(const struct object_id *oid, enum object_type expect)
2184{
2185 enum object_type type = oid_object_info(the_repository, oid, NULL);
2186 if (type < 0)
2187 die(_("%s is not a valid object"), oid_to_hex(oid));
2188 if (type != expect)
2189 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
2190 type_name(expect));
2191}
2192
2193int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2194 struct strbuf *path,
2195 each_loose_object_fn obj_cb,
2196 each_loose_cruft_fn cruft_cb,
2197 each_loose_subdir_fn subdir_cb,
2198 void *data)
2199{
2200 size_t origlen, baselen;
2201 DIR *dir;
2202 struct dirent *de;
2203 int r = 0;
2204 struct object_id oid;
2205
2206 if (subdir_nr > 0xff)
2207 BUG("invalid loose object subdirectory: %x", subdir_nr);
2208
2209 origlen = path->len;
2210 strbuf_complete(path, '/');
2211 strbuf_addf(path, "%02x", subdir_nr);
2212
2213 dir = opendir(path->buf);
2214 if (!dir) {
2215 if (errno != ENOENT)
2216 r = error_errno(_("unable to open %s"), path->buf);
2217 strbuf_setlen(path, origlen);
2218 return r;
2219 }
2220
2221 oid.hash[0] = subdir_nr;
2222 strbuf_addch(path, '/');
2223 baselen = path->len;
2224
2225 while ((de = readdir(dir))) {
2226 size_t namelen;
2227 if (is_dot_or_dotdot(de->d_name))
2228 continue;
2229
2230 namelen = strlen(de->d_name);
2231 strbuf_setlen(path, baselen);
2232 strbuf_add(path, de->d_name, namelen);
2233 if (namelen == the_hash_algo->hexsz - 2 &&
2234 !hex_to_bytes(oid.hash + 1, de->d_name,
2235 the_hash_algo->rawsz - 1)) {
2236 if (obj_cb) {
2237 r = obj_cb(&oid, path->buf, data);
2238 if (r)
2239 break;
2240 }
2241 continue;
2242 }
2243
2244 if (cruft_cb) {
2245 r = cruft_cb(de->d_name, path->buf, data);
2246 if (r)
2247 break;
2248 }
2249 }
2250 closedir(dir);
2251
2252 strbuf_setlen(path, baselen - 1);
2253 if (!r && subdir_cb)
2254 r = subdir_cb(subdir_nr, path->buf, data);
2255
2256 strbuf_setlen(path, origlen);
2257
2258 return r;
2259}
2260
2261int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2262 each_loose_object_fn obj_cb,
2263 each_loose_cruft_fn cruft_cb,
2264 each_loose_subdir_fn subdir_cb,
2265 void *data)
2266{
2267 int r = 0;
2268 int i;
2269
2270 for (i = 0; i < 256; i++) {
2271 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2272 subdir_cb, data);
2273 if (r)
2274 break;
2275 }
2276
2277 return r;
2278}
2279
2280int for_each_loose_file_in_objdir(const char *path,
2281 each_loose_object_fn obj_cb,
2282 each_loose_cruft_fn cruft_cb,
2283 each_loose_subdir_fn subdir_cb,
2284 void *data)
2285{
2286 struct strbuf buf = STRBUF_INIT;
2287 int r;
2288
2289 strbuf_addstr(&buf, path);
2290 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2291 subdir_cb, data);
2292 strbuf_release(&buf);
2293
2294 return r;
2295}
2296
2297int for_each_loose_object(each_loose_object_fn cb, void *data,
2298 enum for_each_object_flags flags)
2299{
2300 struct object_directory *odb;
2301
2302 prepare_alt_odb(the_repository);
2303 for (odb = the_repository->objects->odb; odb; odb = odb->next) {
2304 int r = for_each_loose_file_in_objdir(odb->path, cb, NULL,
2305 NULL, data);
2306 if (r)
2307 return r;
2308
2309 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2310 break;
2311 }
2312
2313 return 0;
2314}
2315
2316static int append_loose_object(const struct object_id *oid, const char *path,
2317 void *data)
2318{
2319 oid_array_append(data, oid);
2320 return 0;
2321}
2322
2323struct oid_array *odb_loose_cache(struct object_directory *odb,
2324 const struct object_id *oid)
2325{
2326 int subdir_nr = oid->hash[0];
2327 struct strbuf buf = STRBUF_INIT;
2328
2329 if (subdir_nr < 0 ||
2330 subdir_nr >= ARRAY_SIZE(odb->loose_objects_subdir_seen))
2331 BUG("subdir_nr out of range");
2332
2333 if (odb->loose_objects_subdir_seen[subdir_nr])
2334 return &odb->loose_objects_cache[subdir_nr];
2335
2336 strbuf_addstr(&buf, odb->path);
2337 for_each_file_in_obj_subdir(subdir_nr, &buf,
2338 append_loose_object,
2339 NULL, NULL,
2340 &odb->loose_objects_cache[subdir_nr]);
2341 odb->loose_objects_subdir_seen[subdir_nr] = 1;
2342 strbuf_release(&buf);
2343 return &odb->loose_objects_cache[subdir_nr];
2344}
2345
2346void odb_clear_loose_cache(struct object_directory *odb)
2347{
2348 int i;
2349
2350 for (i = 0; i < ARRAY_SIZE(odb->loose_objects_cache); i++)
2351 oid_array_clear(&odb->loose_objects_cache[i]);
2352 memset(&odb->loose_objects_subdir_seen, 0,
2353 sizeof(odb->loose_objects_subdir_seen));
2354}
2355
2356static int check_stream_oid(git_zstream *stream,
2357 const char *hdr,
2358 unsigned long size,
2359 const char *path,
2360 const struct object_id *expected_oid)
2361{
2362 git_hash_ctx c;
2363 struct object_id real_oid;
2364 unsigned char buf[4096];
2365 unsigned long total_read;
2366 int status = Z_OK;
2367
2368 the_hash_algo->init_fn(&c);
2369 the_hash_algo->update_fn(&c, hdr, stream->total_out);
2370
2371 /*
2372 * We already read some bytes into hdr, but the ones up to the NUL
2373 * do not count against the object's content size.
2374 */
2375 total_read = stream->total_out - strlen(hdr) - 1;
2376
2377 /*
2378 * This size comparison must be "<=" to read the final zlib packets;
2379 * see the comment in unpack_loose_rest for details.
2380 */
2381 while (total_read <= size &&
2382 (status == Z_OK ||
2383 (status == Z_BUF_ERROR && !stream->avail_out))) {
2384 stream->next_out = buf;
2385 stream->avail_out = sizeof(buf);
2386 if (size - total_read < stream->avail_out)
2387 stream->avail_out = size - total_read;
2388 status = git_inflate(stream, Z_FINISH);
2389 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2390 total_read += stream->next_out - buf;
2391 }
2392 git_inflate_end(stream);
2393
2394 if (status != Z_STREAM_END) {
2395 error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid));
2396 return -1;
2397 }
2398 if (stream->avail_in) {
2399 error(_("garbage at end of loose object '%s'"),
2400 oid_to_hex(expected_oid));
2401 return -1;
2402 }
2403
2404 the_hash_algo->final_fn(real_oid.hash, &c);
2405 if (!oideq(expected_oid, &real_oid)) {
2406 error(_("hash mismatch for %s (expected %s)"), path,
2407 oid_to_hex(expected_oid));
2408 return -1;
2409 }
2410
2411 return 0;
2412}
2413
2414int read_loose_object(const char *path,
2415 const struct object_id *expected_oid,
2416 enum object_type *type,
2417 unsigned long *size,
2418 void **contents)
2419{
2420 int ret = -1;
2421 void *map = NULL;
2422 unsigned long mapsize;
2423 git_zstream stream;
2424 char hdr[MAX_HEADER_LEN];
2425
2426 *contents = NULL;
2427
2428 map = map_loose_object_1(the_repository, path, NULL, &mapsize);
2429 if (!map) {
2430 error_errno(_("unable to mmap %s"), path);
2431 goto out;
2432 }
2433
2434 if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2435 error(_("unable to unpack header of %s"), path);
2436 goto out;
2437 }
2438
2439 *type = parse_loose_header(hdr, size);
2440 if (*type < 0) {
2441 error(_("unable to parse header of %s"), path);
2442 git_inflate_end(&stream);
2443 goto out;
2444 }
2445
2446 if (*type == OBJ_BLOB && *size > big_file_threshold) {
2447 if (check_stream_oid(&stream, hdr, *size, path, expected_oid) < 0)
2448 goto out;
2449 } else {
2450 *contents = unpack_loose_rest(&stream, hdr, *size, expected_oid);
2451 if (!*contents) {
2452 error(_("unable to unpack contents of %s"), path);
2453 git_inflate_end(&stream);
2454 goto out;
2455 }
2456 if (check_object_signature(expected_oid, *contents,
2457 *size, type_name(*type))) {
2458 error(_("hash mismatch for %s (expected %s)"), path,
2459 oid_to_hex(expected_oid));
2460 free(*contents);
2461 goto out;
2462 }
2463 }
2464
2465 ret = 0; /* everything checks out */
2466
2467out:
2468 if (map)
2469 munmap(map, mapsize);
2470 return ret;
2471}