]> git.ipfire.org Git - thirdparty/git.git/blob - object-file.c
object-file: free(*contents) only in read_loose_object() caller
[thirdparty/git.git] / object-file.c
1 /*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This handles basic git 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 "hash-lookup.h"
24 #include "bulk-checkin.h"
25 #include "repository.h"
26 #include "replace-object.h"
27 #include "streaming.h"
28 #include "dir.h"
29 #include "list.h"
30 #include "mergesort.h"
31 #include "quote.h"
32 #include "packfile.h"
33 #include "object-store.h"
34 #include "promisor-remote.h"
35
36 /* The maximum size for an object header. */
37 #define MAX_HEADER_LEN 32
38
39
40 #define EMPTY_TREE_SHA1_BIN_LITERAL \
41 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60" \
42 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04"
43 #define EMPTY_TREE_SHA256_BIN_LITERAL \
44 "\x6e\xf1\x9b\x41\x22\x5c\x53\x69\xf1\xc1" \
45 "\x04\xd4\x5d\x8d\x85\xef\xa9\xb0\x57\xb5" \
46 "\x3b\x14\xb4\xb9\xb9\x39\xdd\x74\xde\xcc" \
47 "\x53\x21"
48
49 #define EMPTY_BLOB_SHA1_BIN_LITERAL \
50 "\xe6\x9d\xe2\x9b\xb2\xd1\xd6\x43\x4b\x8b" \
51 "\x29\xae\x77\x5a\xd8\xc2\xe4\x8c\x53\x91"
52 #define EMPTY_BLOB_SHA256_BIN_LITERAL \
53 "\x47\x3a\x0f\x4c\x3b\xe8\xa9\x36\x81\xa2" \
54 "\x67\xe3\xb1\xe9\xa7\xdc\xda\x11\x85\x43" \
55 "\x6f\xe1\x41\xf7\x74\x91\x20\xa3\x03\x72" \
56 "\x18\x13"
57
58 static const struct object_id empty_tree_oid = {
59 .hash = EMPTY_TREE_SHA1_BIN_LITERAL,
60 .algo = GIT_HASH_SHA1,
61 };
62 static const struct object_id empty_blob_oid = {
63 .hash = EMPTY_BLOB_SHA1_BIN_LITERAL,
64 .algo = GIT_HASH_SHA1,
65 };
66 static const struct object_id null_oid_sha1 = {
67 .hash = {0},
68 .algo = GIT_HASH_SHA1,
69 };
70 static const struct object_id empty_tree_oid_sha256 = {
71 .hash = EMPTY_TREE_SHA256_BIN_LITERAL,
72 .algo = GIT_HASH_SHA256,
73 };
74 static const struct object_id empty_blob_oid_sha256 = {
75 .hash = EMPTY_BLOB_SHA256_BIN_LITERAL,
76 .algo = GIT_HASH_SHA256,
77 };
78 static const struct object_id null_oid_sha256 = {
79 .hash = {0},
80 .algo = GIT_HASH_SHA256,
81 };
82
83 static void git_hash_sha1_init(git_hash_ctx *ctx)
84 {
85 git_SHA1_Init(&ctx->sha1);
86 }
87
88 static void git_hash_sha1_clone(git_hash_ctx *dst, const git_hash_ctx *src)
89 {
90 git_SHA1_Clone(&dst->sha1, &src->sha1);
91 }
92
93 static void git_hash_sha1_update(git_hash_ctx *ctx, const void *data, size_t len)
94 {
95 git_SHA1_Update(&ctx->sha1, data, len);
96 }
97
98 static void git_hash_sha1_final(unsigned char *hash, git_hash_ctx *ctx)
99 {
100 git_SHA1_Final(hash, &ctx->sha1);
101 }
102
103 static void git_hash_sha1_final_oid(struct object_id *oid, git_hash_ctx *ctx)
104 {
105 git_SHA1_Final(oid->hash, &ctx->sha1);
106 memset(oid->hash + GIT_SHA1_RAWSZ, 0, GIT_MAX_RAWSZ - GIT_SHA1_RAWSZ);
107 oid->algo = GIT_HASH_SHA1;
108 }
109
110
111 static void git_hash_sha256_init(git_hash_ctx *ctx)
112 {
113 git_SHA256_Init(&ctx->sha256);
114 }
115
116 static void git_hash_sha256_clone(git_hash_ctx *dst, const git_hash_ctx *src)
117 {
118 git_SHA256_Clone(&dst->sha256, &src->sha256);
119 }
120
121 static void git_hash_sha256_update(git_hash_ctx *ctx, const void *data, size_t len)
122 {
123 git_SHA256_Update(&ctx->sha256, data, len);
124 }
125
126 static void git_hash_sha256_final(unsigned char *hash, git_hash_ctx *ctx)
127 {
128 git_SHA256_Final(hash, &ctx->sha256);
129 }
130
131 static void git_hash_sha256_final_oid(struct object_id *oid, git_hash_ctx *ctx)
132 {
133 git_SHA256_Final(oid->hash, &ctx->sha256);
134 /*
135 * This currently does nothing, so the compiler should optimize it out,
136 * but keep it in case we extend the hash size again.
137 */
138 memset(oid->hash + GIT_SHA256_RAWSZ, 0, GIT_MAX_RAWSZ - GIT_SHA256_RAWSZ);
139 oid->algo = GIT_HASH_SHA256;
140 }
141
142 static void git_hash_unknown_init(git_hash_ctx *ctx)
143 {
144 BUG("trying to init unknown hash");
145 }
146
147 static void git_hash_unknown_clone(git_hash_ctx *dst, const git_hash_ctx *src)
148 {
149 BUG("trying to clone unknown hash");
150 }
151
152 static void git_hash_unknown_update(git_hash_ctx *ctx, const void *data, size_t len)
153 {
154 BUG("trying to update unknown hash");
155 }
156
157 static void git_hash_unknown_final(unsigned char *hash, git_hash_ctx *ctx)
158 {
159 BUG("trying to finalize unknown hash");
160 }
161
162 static void git_hash_unknown_final_oid(struct object_id *oid, git_hash_ctx *ctx)
163 {
164 BUG("trying to finalize unknown hash");
165 }
166
167
168 const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
169 {
170 NULL,
171 0x00000000,
172 0,
173 0,
174 0,
175 git_hash_unknown_init,
176 git_hash_unknown_clone,
177 git_hash_unknown_update,
178 git_hash_unknown_final,
179 git_hash_unknown_final_oid,
180 NULL,
181 NULL,
182 NULL,
183 },
184 {
185 "sha1",
186 /* "sha1", big-endian */
187 0x73686131,
188 GIT_SHA1_RAWSZ,
189 GIT_SHA1_HEXSZ,
190 GIT_SHA1_BLKSZ,
191 git_hash_sha1_init,
192 git_hash_sha1_clone,
193 git_hash_sha1_update,
194 git_hash_sha1_final,
195 git_hash_sha1_final_oid,
196 &empty_tree_oid,
197 &empty_blob_oid,
198 &null_oid_sha1,
199 },
200 {
201 "sha256",
202 /* "s256", big-endian */
203 0x73323536,
204 GIT_SHA256_RAWSZ,
205 GIT_SHA256_HEXSZ,
206 GIT_SHA256_BLKSZ,
207 git_hash_sha256_init,
208 git_hash_sha256_clone,
209 git_hash_sha256_update,
210 git_hash_sha256_final,
211 git_hash_sha256_final_oid,
212 &empty_tree_oid_sha256,
213 &empty_blob_oid_sha256,
214 &null_oid_sha256,
215 }
216 };
217
218 const struct object_id *null_oid(void)
219 {
220 return the_hash_algo->null_oid;
221 }
222
223 const char *empty_tree_oid_hex(void)
224 {
225 static char buf[GIT_MAX_HEXSZ + 1];
226 return oid_to_hex_r(buf, the_hash_algo->empty_tree);
227 }
228
229 const char *empty_blob_oid_hex(void)
230 {
231 static char buf[GIT_MAX_HEXSZ + 1];
232 return oid_to_hex_r(buf, the_hash_algo->empty_blob);
233 }
234
235 int hash_algo_by_name(const char *name)
236 {
237 int i;
238 if (!name)
239 return GIT_HASH_UNKNOWN;
240 for (i = 1; i < GIT_HASH_NALGOS; i++)
241 if (!strcmp(name, hash_algos[i].name))
242 return i;
243 return GIT_HASH_UNKNOWN;
244 }
245
246 int hash_algo_by_id(uint32_t format_id)
247 {
248 int i;
249 for (i = 1; i < GIT_HASH_NALGOS; i++)
250 if (format_id == hash_algos[i].format_id)
251 return i;
252 return GIT_HASH_UNKNOWN;
253 }
254
255 int hash_algo_by_length(int len)
256 {
257 int i;
258 for (i = 1; i < GIT_HASH_NALGOS; i++)
259 if (len == hash_algos[i].rawsz)
260 return i;
261 return GIT_HASH_UNKNOWN;
262 }
263
264 /*
265 * This is meant to hold a *small* number of objects that you would
266 * want read_object_file() to be able to return, but yet you do not want
267 * to write them into the object store (e.g. a browse-only
268 * application).
269 */
270 static struct cached_object {
271 struct object_id oid;
272 enum object_type type;
273 void *buf;
274 unsigned long size;
275 } *cached_objects;
276 static int cached_object_nr, cached_object_alloc;
277
278 static struct cached_object empty_tree = {
279 { EMPTY_TREE_SHA1_BIN_LITERAL },
280 OBJ_TREE,
281 "",
282 0
283 };
284
285 static struct cached_object *find_cached_object(const struct object_id *oid)
286 {
287 int i;
288 struct cached_object *co = cached_objects;
289
290 for (i = 0; i < cached_object_nr; i++, co++) {
291 if (oideq(&co->oid, oid))
292 return co;
293 }
294 if (oideq(oid, the_hash_algo->empty_tree))
295 return &empty_tree;
296 return NULL;
297 }
298
299
300 static int get_conv_flags(unsigned flags)
301 {
302 if (flags & HASH_RENORMALIZE)
303 return CONV_EOL_RENORMALIZE;
304 else if (flags & HASH_WRITE_OBJECT)
305 return global_conv_flags_eol | CONV_WRITE_OBJECT;
306 else
307 return 0;
308 }
309
310
311 int mkdir_in_gitdir(const char *path)
312 {
313 if (mkdir(path, 0777)) {
314 int saved_errno = errno;
315 struct stat st;
316 struct strbuf sb = STRBUF_INIT;
317
318 if (errno != EEXIST)
319 return -1;
320 /*
321 * Are we looking at a path in a symlinked worktree
322 * whose original repository does not yet have it?
323 * e.g. .git/rr-cache pointing at its original
324 * repository in which the user hasn't performed any
325 * conflict resolution yet?
326 */
327 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
328 strbuf_readlink(&sb, path, st.st_size) ||
329 !is_absolute_path(sb.buf) ||
330 mkdir(sb.buf, 0777)) {
331 strbuf_release(&sb);
332 errno = saved_errno;
333 return -1;
334 }
335 strbuf_release(&sb);
336 }
337 return adjust_shared_perm(path);
338 }
339
340 static enum scld_error safe_create_leading_directories_1(char *path, int share)
341 {
342 char *next_component = path + offset_1st_component(path);
343 enum scld_error ret = SCLD_OK;
344
345 while (ret == SCLD_OK && next_component) {
346 struct stat st;
347 char *slash = next_component, slash_character;
348
349 while (*slash && !is_dir_sep(*slash))
350 slash++;
351
352 if (!*slash)
353 break;
354
355 next_component = slash + 1;
356 while (is_dir_sep(*next_component))
357 next_component++;
358 if (!*next_component)
359 break;
360
361 slash_character = *slash;
362 *slash = '\0';
363 if (!stat(path, &st)) {
364 /* path exists */
365 if (!S_ISDIR(st.st_mode)) {
366 errno = ENOTDIR;
367 ret = SCLD_EXISTS;
368 }
369 } else if (mkdir(path, 0777)) {
370 if (errno == EEXIST &&
371 !stat(path, &st) && S_ISDIR(st.st_mode))
372 ; /* somebody created it since we checked */
373 else if (errno == ENOENT)
374 /*
375 * Either mkdir() failed because
376 * somebody just pruned the containing
377 * directory, or stat() failed because
378 * the file that was in our way was
379 * just removed. Either way, inform
380 * the caller that it might be worth
381 * trying again:
382 */
383 ret = SCLD_VANISHED;
384 else
385 ret = SCLD_FAILED;
386 } else if (share && adjust_shared_perm(path)) {
387 ret = SCLD_PERMS;
388 }
389 *slash = slash_character;
390 }
391 return ret;
392 }
393
394 enum scld_error safe_create_leading_directories(char *path)
395 {
396 return safe_create_leading_directories_1(path, 1);
397 }
398
399 enum scld_error safe_create_leading_directories_no_share(char *path)
400 {
401 return safe_create_leading_directories_1(path, 0);
402 }
403
404 enum scld_error safe_create_leading_directories_const(const char *path)
405 {
406 int save_errno;
407 /* path points to cache entries, so xstrdup before messing with it */
408 char *buf = xstrdup(path);
409 enum scld_error result = safe_create_leading_directories(buf);
410
411 save_errno = errno;
412 free(buf);
413 errno = save_errno;
414 return result;
415 }
416
417 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
418 {
419 /*
420 * The number of times we will try to remove empty directories
421 * in the way of path. This is only 1 because if another
422 * process is racily creating directories that conflict with
423 * us, we don't want to fight against them.
424 */
425 int remove_directories_remaining = 1;
426
427 /*
428 * The number of times that we will try to create the
429 * directories containing path. We are willing to attempt this
430 * more than once, because another process could be trying to
431 * clean up empty directories at the same time as we are
432 * trying to create them.
433 */
434 int create_directories_remaining = 3;
435
436 /* A scratch copy of path, filled lazily if we need it: */
437 struct strbuf path_copy = STRBUF_INIT;
438
439 int ret, save_errno;
440
441 /* Sanity check: */
442 assert(*path);
443
444 retry_fn:
445 ret = fn(path, cb);
446 save_errno = errno;
447 if (!ret)
448 goto out;
449
450 if (errno == EISDIR && remove_directories_remaining-- > 0) {
451 /*
452 * A directory is in the way. Maybe it is empty; try
453 * to remove it:
454 */
455 if (!path_copy.len)
456 strbuf_addstr(&path_copy, path);
457
458 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
459 goto retry_fn;
460 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
461 /*
462 * Maybe the containing directory didn't exist, or
463 * maybe it was just deleted by a process that is
464 * racing with us to clean up empty directories. Try
465 * to create it:
466 */
467 enum scld_error scld_result;
468
469 if (!path_copy.len)
470 strbuf_addstr(&path_copy, path);
471
472 do {
473 scld_result = safe_create_leading_directories(path_copy.buf);
474 if (scld_result == SCLD_OK)
475 goto retry_fn;
476 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
477 }
478
479 out:
480 strbuf_release(&path_copy);
481 errno = save_errno;
482 return ret;
483 }
484
485 static void fill_loose_path(struct strbuf *buf, const struct object_id *oid)
486 {
487 int i;
488 for (i = 0; i < the_hash_algo->rawsz; i++) {
489 static char hex[] = "0123456789abcdef";
490 unsigned int val = oid->hash[i];
491 strbuf_addch(buf, hex[val >> 4]);
492 strbuf_addch(buf, hex[val & 0xf]);
493 if (!i)
494 strbuf_addch(buf, '/');
495 }
496 }
497
498 static const char *odb_loose_path(struct object_directory *odb,
499 struct strbuf *buf,
500 const struct object_id *oid)
501 {
502 strbuf_reset(buf);
503 strbuf_addstr(buf, odb->path);
504 strbuf_addch(buf, '/');
505 fill_loose_path(buf, oid);
506 return buf->buf;
507 }
508
509 const char *loose_object_path(struct repository *r, struct strbuf *buf,
510 const struct object_id *oid)
511 {
512 return odb_loose_path(r->objects->odb, buf, oid);
513 }
514
515 /*
516 * Return non-zero iff the path is usable as an alternate object database.
517 */
518 static int alt_odb_usable(struct raw_object_store *o,
519 struct strbuf *path,
520 const char *normalized_objdir)
521 {
522 struct object_directory *odb;
523
524 /* Detect cases where alternate disappeared */
525 if (!is_directory(path->buf)) {
526 error(_("object directory %s does not exist; "
527 "check .git/objects/info/alternates"),
528 path->buf);
529 return 0;
530 }
531
532 /*
533 * Prevent the common mistake of listing the same
534 * thing twice, or object directory itself.
535 */
536 for (odb = o->odb; odb; odb = odb->next) {
537 if (!fspathcmp(path->buf, odb->path))
538 return 0;
539 }
540 if (!fspathcmp(path->buf, normalized_objdir))
541 return 0;
542
543 return 1;
544 }
545
546 /*
547 * Prepare alternate object database registry.
548 *
549 * The variable alt_odb_list points at the list of struct
550 * object_directory. The elements on this list come from
551 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
552 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
553 * whose contents is similar to that environment variable but can be
554 * LF separated. Its base points at a statically allocated buffer that
555 * contains "/the/directory/corresponding/to/.git/objects/...", while
556 * its name points just after the slash at the end of ".git/objects/"
557 * in the example above, and has enough space to hold all hex characters
558 * of the object ID, an extra slash for the first level indirection, and
559 * the terminating NUL.
560 */
561 static void read_info_alternates(struct repository *r,
562 const char *relative_base,
563 int depth);
564 static int link_alt_odb_entry(struct repository *r, const char *entry,
565 const char *relative_base, int depth, const char *normalized_objdir)
566 {
567 struct object_directory *ent;
568 struct strbuf pathbuf = STRBUF_INIT;
569
570 if (!is_absolute_path(entry) && relative_base) {
571 strbuf_realpath(&pathbuf, relative_base, 1);
572 strbuf_addch(&pathbuf, '/');
573 }
574 strbuf_addstr(&pathbuf, entry);
575
576 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
577 error(_("unable to normalize alternate object path: %s"),
578 pathbuf.buf);
579 strbuf_release(&pathbuf);
580 return -1;
581 }
582
583 /*
584 * The trailing slash after the directory name is given by
585 * this function at the end. Remove duplicates.
586 */
587 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
588 strbuf_setlen(&pathbuf, pathbuf.len - 1);
589
590 if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir)) {
591 strbuf_release(&pathbuf);
592 return -1;
593 }
594
595 CALLOC_ARRAY(ent, 1);
596 ent->path = xstrdup(pathbuf.buf);
597
598 /* add the alternate entry */
599 *r->objects->odb_tail = ent;
600 r->objects->odb_tail = &(ent->next);
601 ent->next = NULL;
602
603 /* recursively add alternates */
604 read_info_alternates(r, pathbuf.buf, depth + 1);
605
606 strbuf_release(&pathbuf);
607 return 0;
608 }
609
610 static const char *parse_alt_odb_entry(const char *string,
611 int sep,
612 struct strbuf *out)
613 {
614 const char *end;
615
616 strbuf_reset(out);
617
618 if (*string == '#') {
619 /* comment; consume up to next separator */
620 end = strchrnul(string, sep);
621 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
622 /*
623 * quoted path; unquote_c_style has copied the
624 * data for us and set "end". Broken quoting (e.g.,
625 * an entry that doesn't end with a quote) falls
626 * back to the unquoted case below.
627 */
628 } else {
629 /* normal, unquoted path */
630 end = strchrnul(string, sep);
631 strbuf_add(out, string, end - string);
632 }
633
634 if (*end)
635 end++;
636 return end;
637 }
638
639 static void link_alt_odb_entries(struct repository *r, const char *alt,
640 int sep, const char *relative_base, int depth)
641 {
642 struct strbuf objdirbuf = STRBUF_INIT;
643 struct strbuf entry = STRBUF_INIT;
644
645 if (!alt || !*alt)
646 return;
647
648 if (depth > 5) {
649 error(_("%s: ignoring alternate object stores, nesting too deep"),
650 relative_base);
651 return;
652 }
653
654 strbuf_add_absolute_path(&objdirbuf, r->objects->odb->path);
655 if (strbuf_normalize_path(&objdirbuf) < 0)
656 die(_("unable to normalize object directory: %s"),
657 objdirbuf.buf);
658
659 while (*alt) {
660 alt = parse_alt_odb_entry(alt, sep, &entry);
661 if (!entry.len)
662 continue;
663 link_alt_odb_entry(r, entry.buf,
664 relative_base, depth, objdirbuf.buf);
665 }
666 strbuf_release(&entry);
667 strbuf_release(&objdirbuf);
668 }
669
670 static void read_info_alternates(struct repository *r,
671 const char *relative_base,
672 int depth)
673 {
674 char *path;
675 struct strbuf buf = STRBUF_INIT;
676
677 path = xstrfmt("%s/info/alternates", relative_base);
678 if (strbuf_read_file(&buf, path, 1024) < 0) {
679 warn_on_fopen_errors(path);
680 free(path);
681 return;
682 }
683
684 link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
685 strbuf_release(&buf);
686 free(path);
687 }
688
689 void add_to_alternates_file(const char *reference)
690 {
691 struct lock_file lock = LOCK_INIT;
692 char *alts = git_pathdup("objects/info/alternates");
693 FILE *in, *out;
694 int found = 0;
695
696 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
697 out = fdopen_lock_file(&lock, "w");
698 if (!out)
699 die_errno(_("unable to fdopen alternates lockfile"));
700
701 in = fopen(alts, "r");
702 if (in) {
703 struct strbuf line = STRBUF_INIT;
704
705 while (strbuf_getline(&line, in) != EOF) {
706 if (!strcmp(reference, line.buf)) {
707 found = 1;
708 break;
709 }
710 fprintf_or_die(out, "%s\n", line.buf);
711 }
712
713 strbuf_release(&line);
714 fclose(in);
715 }
716 else if (errno != ENOENT)
717 die_errno(_("unable to read alternates file"));
718
719 if (found) {
720 rollback_lock_file(&lock);
721 } else {
722 fprintf_or_die(out, "%s\n", reference);
723 if (commit_lock_file(&lock))
724 die_errno(_("unable to move new alternates file into place"));
725 if (the_repository->objects->loaded_alternates)
726 link_alt_odb_entries(the_repository, reference,
727 '\n', NULL, 0);
728 }
729 free(alts);
730 }
731
732 void add_to_alternates_memory(const char *reference)
733 {
734 /*
735 * Make sure alternates are initialized, or else our entry may be
736 * overwritten when they are.
737 */
738 prepare_alt_odb(the_repository);
739
740 link_alt_odb_entries(the_repository, reference,
741 '\n', NULL, 0);
742 }
743
744 /*
745 * Compute the exact path an alternate is at and returns it. In case of
746 * error NULL is returned and the human readable error is added to `err`
747 * `path` may be relative and should point to $GIT_DIR.
748 * `err` must not be null.
749 */
750 char *compute_alternate_path(const char *path, struct strbuf *err)
751 {
752 char *ref_git = NULL;
753 const char *repo;
754 int seen_error = 0;
755
756 ref_git = real_pathdup(path, 0);
757 if (!ref_git) {
758 seen_error = 1;
759 strbuf_addf(err, _("path '%s' does not exist"), path);
760 goto out;
761 }
762
763 repo = read_gitfile(ref_git);
764 if (!repo)
765 repo = read_gitfile(mkpath("%s/.git", ref_git));
766 if (repo) {
767 free(ref_git);
768 ref_git = xstrdup(repo);
769 }
770
771 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
772 char *ref_git_git = mkpathdup("%s/.git", ref_git);
773 free(ref_git);
774 ref_git = ref_git_git;
775 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
776 struct strbuf sb = STRBUF_INIT;
777 seen_error = 1;
778 if (get_common_dir(&sb, ref_git)) {
779 strbuf_addf(err,
780 _("reference repository '%s' as a linked "
781 "checkout is not supported yet."),
782 path);
783 goto out;
784 }
785
786 strbuf_addf(err, _("reference repository '%s' is not a "
787 "local repository."), path);
788 goto out;
789 }
790
791 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
792 strbuf_addf(err, _("reference repository '%s' is shallow"),
793 path);
794 seen_error = 1;
795 goto out;
796 }
797
798 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
799 strbuf_addf(err,
800 _("reference repository '%s' is grafted"),
801 path);
802 seen_error = 1;
803 goto out;
804 }
805
806 out:
807 if (seen_error) {
808 FREE_AND_NULL(ref_git);
809 }
810
811 return ref_git;
812 }
813
814 static void fill_alternate_refs_command(struct child_process *cmd,
815 const char *repo_path)
816 {
817 const char *value;
818
819 if (!git_config_get_value("core.alternateRefsCommand", &value)) {
820 cmd->use_shell = 1;
821
822 strvec_push(&cmd->args, value);
823 strvec_push(&cmd->args, repo_path);
824 } else {
825 cmd->git_cmd = 1;
826
827 strvec_pushf(&cmd->args, "--git-dir=%s", repo_path);
828 strvec_push(&cmd->args, "for-each-ref");
829 strvec_push(&cmd->args, "--format=%(objectname)");
830
831 if (!git_config_get_value("core.alternateRefsPrefixes", &value)) {
832 strvec_push(&cmd->args, "--");
833 strvec_split(&cmd->args, value);
834 }
835 }
836
837 cmd->env = local_repo_env;
838 cmd->out = -1;
839 }
840
841 static void read_alternate_refs(const char *path,
842 alternate_ref_fn *cb,
843 void *data)
844 {
845 struct child_process cmd = CHILD_PROCESS_INIT;
846 struct strbuf line = STRBUF_INIT;
847 FILE *fh;
848
849 fill_alternate_refs_command(&cmd, path);
850
851 if (start_command(&cmd))
852 return;
853
854 fh = xfdopen(cmd.out, "r");
855 while (strbuf_getline_lf(&line, fh) != EOF) {
856 struct object_id oid;
857 const char *p;
858
859 if (parse_oid_hex(line.buf, &oid, &p) || *p) {
860 warning(_("invalid line while parsing alternate refs: %s"),
861 line.buf);
862 break;
863 }
864
865 cb(&oid, data);
866 }
867
868 fclose(fh);
869 finish_command(&cmd);
870 strbuf_release(&line);
871 }
872
873 struct alternate_refs_data {
874 alternate_ref_fn *fn;
875 void *data;
876 };
877
878 static int refs_from_alternate_cb(struct object_directory *e,
879 void *data)
880 {
881 struct strbuf path = STRBUF_INIT;
882 size_t base_len;
883 struct alternate_refs_data *cb = data;
884
885 if (!strbuf_realpath(&path, e->path, 0))
886 goto out;
887 if (!strbuf_strip_suffix(&path, "/objects"))
888 goto out;
889 base_len = path.len;
890
891 /* Is this a git repository with refs? */
892 strbuf_addstr(&path, "/refs");
893 if (!is_directory(path.buf))
894 goto out;
895 strbuf_setlen(&path, base_len);
896
897 read_alternate_refs(path.buf, cb->fn, cb->data);
898
899 out:
900 strbuf_release(&path);
901 return 0;
902 }
903
904 void for_each_alternate_ref(alternate_ref_fn fn, void *data)
905 {
906 struct alternate_refs_data cb;
907 cb.fn = fn;
908 cb.data = data;
909 foreach_alt_odb(refs_from_alternate_cb, &cb);
910 }
911
912 int foreach_alt_odb(alt_odb_fn fn, void *cb)
913 {
914 struct object_directory *ent;
915 int r = 0;
916
917 prepare_alt_odb(the_repository);
918 for (ent = the_repository->objects->odb->next; ent; ent = ent->next) {
919 r = fn(ent, cb);
920 if (r)
921 break;
922 }
923 return r;
924 }
925
926 void prepare_alt_odb(struct repository *r)
927 {
928 if (r->objects->loaded_alternates)
929 return;
930
931 link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
932
933 read_info_alternates(r, r->objects->odb->path, 0);
934 r->objects->loaded_alternates = 1;
935 }
936
937 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
938 static int freshen_file(const char *fn)
939 {
940 return !utime(fn, NULL);
941 }
942
943 /*
944 * All of the check_and_freshen functions return 1 if the file exists and was
945 * freshened (if freshening was requested), 0 otherwise. If they return
946 * 0, you should not assume that it is safe to skip a write of the object (it
947 * either does not exist on disk, or has a stale mtime and may be subject to
948 * pruning).
949 */
950 int check_and_freshen_file(const char *fn, int freshen)
951 {
952 if (access(fn, F_OK))
953 return 0;
954 if (freshen && !freshen_file(fn))
955 return 0;
956 return 1;
957 }
958
959 static int check_and_freshen_odb(struct object_directory *odb,
960 const struct object_id *oid,
961 int freshen)
962 {
963 static struct strbuf path = STRBUF_INIT;
964 odb_loose_path(odb, &path, oid);
965 return check_and_freshen_file(path.buf, freshen);
966 }
967
968 static int check_and_freshen_local(const struct object_id *oid, int freshen)
969 {
970 return check_and_freshen_odb(the_repository->objects->odb, oid, freshen);
971 }
972
973 static int check_and_freshen_nonlocal(const struct object_id *oid, int freshen)
974 {
975 struct object_directory *odb;
976
977 prepare_alt_odb(the_repository);
978 for (odb = the_repository->objects->odb->next; odb; odb = odb->next) {
979 if (check_and_freshen_odb(odb, oid, freshen))
980 return 1;
981 }
982 return 0;
983 }
984
985 static int check_and_freshen(const struct object_id *oid, int freshen)
986 {
987 return check_and_freshen_local(oid, freshen) ||
988 check_and_freshen_nonlocal(oid, freshen);
989 }
990
991 int has_loose_object_nonlocal(const struct object_id *oid)
992 {
993 return check_and_freshen_nonlocal(oid, 0);
994 }
995
996 static int has_loose_object(const struct object_id *oid)
997 {
998 return check_and_freshen(oid, 0);
999 }
1000
1001 static void mmap_limit_check(size_t length)
1002 {
1003 static size_t limit = 0;
1004 if (!limit) {
1005 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
1006 if (!limit)
1007 limit = SIZE_MAX;
1008 }
1009 if (length > limit)
1010 die(_("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX),
1011 (uintmax_t)length, (uintmax_t)limit);
1012 }
1013
1014 void *xmmap_gently(void *start, size_t length,
1015 int prot, int flags, int fd, off_t offset)
1016 {
1017 void *ret;
1018
1019 mmap_limit_check(length);
1020 ret = mmap(start, length, prot, flags, fd, offset);
1021 if (ret == MAP_FAILED && !length)
1022 ret = NULL;
1023 return ret;
1024 }
1025
1026 void *xmmap(void *start, size_t length,
1027 int prot, int flags, int fd, off_t offset)
1028 {
1029 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
1030 if (ret == MAP_FAILED)
1031 die_errno(_("mmap failed"));
1032 return ret;
1033 }
1034
1035 /*
1036 * With an in-core object data in "map", rehash it to make sure the
1037 * object name actually matches "oid" to detect object corruption.
1038 * With "map" == NULL, try reading the object named with "oid" using
1039 * the streaming interface and rehash it to do the same.
1040 */
1041 int check_object_signature(struct repository *r, const struct object_id *oid,
1042 void *map, unsigned long size, const char *type,
1043 struct object_id *real_oidp)
1044 {
1045 struct object_id tmp;
1046 struct object_id *real_oid = real_oidp ? real_oidp : &tmp;
1047 enum object_type obj_type;
1048 struct git_istream *st;
1049 git_hash_ctx c;
1050 char hdr[MAX_HEADER_LEN];
1051 int hdrlen;
1052
1053 if (map) {
1054 hash_object_file(r->hash_algo, map, size, type, real_oid);
1055 return !oideq(oid, real_oid) ? -1 : 0;
1056 }
1057
1058 st = open_istream(r, oid, &obj_type, &size, NULL);
1059 if (!st)
1060 return -1;
1061
1062 /* Generate the header */
1063 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(obj_type), (uintmax_t)size) + 1;
1064
1065 /* Sha1.. */
1066 r->hash_algo->init_fn(&c);
1067 r->hash_algo->update_fn(&c, hdr, hdrlen);
1068 for (;;) {
1069 char buf[1024 * 16];
1070 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1071
1072 if (readlen < 0) {
1073 close_istream(st);
1074 return -1;
1075 }
1076 if (!readlen)
1077 break;
1078 r->hash_algo->update_fn(&c, buf, readlen);
1079 }
1080 r->hash_algo->final_oid_fn(real_oid, &c);
1081 close_istream(st);
1082 return !oideq(oid, real_oid) ? -1 : 0;
1083 }
1084
1085 int git_open_cloexec(const char *name, int flags)
1086 {
1087 int fd;
1088 static int o_cloexec = O_CLOEXEC;
1089
1090 fd = open(name, flags | o_cloexec);
1091 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1092 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1093 o_cloexec &= ~O_CLOEXEC;
1094 fd = open(name, flags | o_cloexec);
1095 }
1096
1097 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1098 {
1099 static int fd_cloexec = FD_CLOEXEC;
1100
1101 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1102 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1103 int flags = fcntl(fd, F_GETFD);
1104 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1105 fd_cloexec = 0;
1106 }
1107 }
1108 #endif
1109 return fd;
1110 }
1111
1112 /*
1113 * Find "oid" as a loose object in the local repository or in an alternate.
1114 * Returns 0 on success, negative on failure.
1115 *
1116 * The "path" out-parameter will give the path of the object we found (if any).
1117 * Note that it may point to static storage and is only valid until another
1118 * call to stat_loose_object().
1119 */
1120 static int stat_loose_object(struct repository *r, const struct object_id *oid,
1121 struct stat *st, const char **path)
1122 {
1123 struct object_directory *odb;
1124 static struct strbuf buf = STRBUF_INIT;
1125
1126 prepare_alt_odb(r);
1127 for (odb = r->objects->odb; odb; odb = odb->next) {
1128 *path = odb_loose_path(odb, &buf, oid);
1129 if (!lstat(*path, st))
1130 return 0;
1131 }
1132
1133 return -1;
1134 }
1135
1136 /*
1137 * Like stat_loose_object(), but actually open the object and return the
1138 * descriptor. See the caveats on the "path" parameter above.
1139 */
1140 static int open_loose_object(struct repository *r,
1141 const struct object_id *oid, const char **path)
1142 {
1143 int fd;
1144 struct object_directory *odb;
1145 int most_interesting_errno = ENOENT;
1146 static struct strbuf buf = STRBUF_INIT;
1147
1148 prepare_alt_odb(r);
1149 for (odb = r->objects->odb; odb; odb = odb->next) {
1150 *path = odb_loose_path(odb, &buf, oid);
1151 fd = git_open(*path);
1152 if (fd >= 0)
1153 return fd;
1154
1155 if (most_interesting_errno == ENOENT)
1156 most_interesting_errno = errno;
1157 }
1158 errno = most_interesting_errno;
1159 return -1;
1160 }
1161
1162 static int quick_has_loose(struct repository *r,
1163 const struct object_id *oid)
1164 {
1165 struct object_directory *odb;
1166
1167 prepare_alt_odb(r);
1168 for (odb = r->objects->odb; odb; odb = odb->next) {
1169 if (oid_array_lookup(odb_loose_cache(odb, oid), oid) >= 0)
1170 return 1;
1171 }
1172 return 0;
1173 }
1174
1175 /*
1176 * Map the loose object at "path" if it is not NULL, or the path found by
1177 * searching for a loose object named "oid".
1178 */
1179 static void *map_loose_object_1(struct repository *r, const char *path,
1180 const struct object_id *oid, unsigned long *size)
1181 {
1182 void *map;
1183 int fd;
1184
1185 if (path)
1186 fd = git_open(path);
1187 else
1188 fd = open_loose_object(r, oid, &path);
1189 map = NULL;
1190 if (fd >= 0) {
1191 struct stat st;
1192
1193 if (!fstat(fd, &st)) {
1194 *size = xsize_t(st.st_size);
1195 if (!*size) {
1196 /* mmap() is forbidden on empty files */
1197 error(_("object file %s is empty"), path);
1198 close(fd);
1199 return NULL;
1200 }
1201 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1202 }
1203 close(fd);
1204 }
1205 return map;
1206 }
1207
1208 void *map_loose_object(struct repository *r,
1209 const struct object_id *oid,
1210 unsigned long *size)
1211 {
1212 return map_loose_object_1(r, NULL, oid, size);
1213 }
1214
1215 enum unpack_loose_header_result unpack_loose_header(git_zstream *stream,
1216 unsigned char *map,
1217 unsigned long mapsize,
1218 void *buffer,
1219 unsigned long bufsiz,
1220 struct strbuf *header)
1221 {
1222 int status;
1223
1224 /* Get the data stream */
1225 memset(stream, 0, sizeof(*stream));
1226 stream->next_in = map;
1227 stream->avail_in = mapsize;
1228 stream->next_out = buffer;
1229 stream->avail_out = bufsiz;
1230
1231 git_inflate_init(stream);
1232 obj_read_unlock();
1233 status = git_inflate(stream, 0);
1234 obj_read_lock();
1235 if (status < Z_OK)
1236 return ULHR_BAD;
1237
1238 /*
1239 * Check if entire header is unpacked in the first iteration.
1240 */
1241 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1242 return ULHR_OK;
1243
1244 /*
1245 * We have a header longer than MAX_HEADER_LEN. The "header"
1246 * here is only non-NULL when we run "cat-file
1247 * --allow-unknown-type".
1248 */
1249 if (!header)
1250 return ULHR_TOO_LONG;
1251
1252 /*
1253 * buffer[0..bufsiz] was not large enough. Copy the partial
1254 * result out to header, and then append the result of further
1255 * reading the stream.
1256 */
1257 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1258 stream->next_out = buffer;
1259 stream->avail_out = bufsiz;
1260
1261 do {
1262 obj_read_unlock();
1263 status = git_inflate(stream, 0);
1264 obj_read_lock();
1265 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1266 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1267 return 0;
1268 stream->next_out = buffer;
1269 stream->avail_out = bufsiz;
1270 } while (status != Z_STREAM_END);
1271 return ULHR_TOO_LONG;
1272 }
1273
1274 static void *unpack_loose_rest(git_zstream *stream,
1275 void *buffer, unsigned long size,
1276 const struct object_id *oid)
1277 {
1278 int bytes = strlen(buffer) + 1;
1279 unsigned char *buf = xmallocz(size);
1280 unsigned long n;
1281 int status = Z_OK;
1282
1283 n = stream->total_out - bytes;
1284 if (n > size)
1285 n = size;
1286 memcpy(buf, (char *) buffer + bytes, n);
1287 bytes = n;
1288 if (bytes <= size) {
1289 /*
1290 * The above condition must be (bytes <= size), not
1291 * (bytes < size). In other words, even though we
1292 * expect no more output and set avail_out to zero,
1293 * the input zlib stream may have bytes that express
1294 * "this concludes the stream", and we *do* want to
1295 * eat that input.
1296 *
1297 * Otherwise we would not be able to test that we
1298 * consumed all the input to reach the expected size;
1299 * we also want to check that zlib tells us that all
1300 * went well with status == Z_STREAM_END at the end.
1301 */
1302 stream->next_out = buf + bytes;
1303 stream->avail_out = size - bytes;
1304 while (status == Z_OK) {
1305 obj_read_unlock();
1306 status = git_inflate(stream, Z_FINISH);
1307 obj_read_lock();
1308 }
1309 }
1310 if (status == Z_STREAM_END && !stream->avail_in) {
1311 git_inflate_end(stream);
1312 return buf;
1313 }
1314
1315 if (status < 0)
1316 error(_("corrupt loose object '%s'"), oid_to_hex(oid));
1317 else if (stream->avail_in)
1318 error(_("garbage at end of loose object '%s'"),
1319 oid_to_hex(oid));
1320 free(buf);
1321 return NULL;
1322 }
1323
1324 /*
1325 * We used to just use "sscanf()", but that's actually way
1326 * too permissive for what we want to check. So do an anal
1327 * object header parse by hand.
1328 */
1329 int parse_loose_header(const char *hdr, struct object_info *oi)
1330 {
1331 const char *type_buf = hdr;
1332 unsigned long size;
1333 int type, type_len = 0;
1334
1335 /*
1336 * The type can be of any size but is followed by
1337 * a space.
1338 */
1339 for (;;) {
1340 char c = *hdr++;
1341 if (!c)
1342 return -1;
1343 if (c == ' ')
1344 break;
1345 type_len++;
1346 }
1347
1348 type = type_from_string_gently(type_buf, type_len, 1);
1349 if (oi->type_name)
1350 strbuf_add(oi->type_name, type_buf, type_len);
1351 if (oi->typep)
1352 *oi->typep = type;
1353
1354 /*
1355 * The length must follow immediately, and be in canonical
1356 * decimal format (ie "010" is not valid).
1357 */
1358 size = *hdr++ - '0';
1359 if (size > 9)
1360 return -1;
1361 if (size) {
1362 for (;;) {
1363 unsigned long c = *hdr - '0';
1364 if (c > 9)
1365 break;
1366 hdr++;
1367 size = size * 10 + c;
1368 }
1369 }
1370
1371 if (oi->sizep)
1372 *oi->sizep = size;
1373
1374 /*
1375 * The length must be followed by a zero byte
1376 */
1377 if (*hdr)
1378 return -1;
1379
1380 /*
1381 * The format is valid, but the type may still be bogus. The
1382 * Caller needs to check its oi->typep.
1383 */
1384 return 0;
1385 }
1386
1387 static int loose_object_info(struct repository *r,
1388 const struct object_id *oid,
1389 struct object_info *oi, int flags)
1390 {
1391 int status = 0;
1392 unsigned long mapsize;
1393 void *map;
1394 git_zstream stream;
1395 char hdr[MAX_HEADER_LEN];
1396 struct strbuf hdrbuf = STRBUF_INIT;
1397 unsigned long size_scratch;
1398 enum object_type type_scratch;
1399 int allow_unknown = flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE;
1400
1401 if (oi->delta_base_oid)
1402 oidclr(oi->delta_base_oid);
1403
1404 /*
1405 * If we don't care about type or size, then we don't
1406 * need to look inside the object at all. Note that we
1407 * do not optimize out the stat call, even if the
1408 * caller doesn't care about the disk-size, since our
1409 * return value implicitly indicates whether the
1410 * object even exists.
1411 */
1412 if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1413 const char *path;
1414 struct stat st;
1415 if (!oi->disk_sizep && (flags & OBJECT_INFO_QUICK))
1416 return quick_has_loose(r, oid) ? 0 : -1;
1417 if (stat_loose_object(r, oid, &st, &path) < 0)
1418 return -1;
1419 if (oi->disk_sizep)
1420 *oi->disk_sizep = st.st_size;
1421 return 0;
1422 }
1423
1424 map = map_loose_object(r, oid, &mapsize);
1425 if (!map)
1426 return -1;
1427
1428 if (!oi->sizep)
1429 oi->sizep = &size_scratch;
1430 if (!oi->typep)
1431 oi->typep = &type_scratch;
1432
1433 if (oi->disk_sizep)
1434 *oi->disk_sizep = mapsize;
1435
1436 switch (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr),
1437 allow_unknown ? &hdrbuf : NULL)) {
1438 case ULHR_OK:
1439 if (parse_loose_header(hdrbuf.len ? hdrbuf.buf : hdr, oi) < 0)
1440 status = error(_("unable to parse %s header"), oid_to_hex(oid));
1441 else if (!allow_unknown && *oi->typep < 0)
1442 die(_("invalid object type"));
1443
1444 if (!oi->contentp)
1445 break;
1446 *oi->contentp = unpack_loose_rest(&stream, hdr, *oi->sizep, oid);
1447 if (*oi->contentp)
1448 goto cleanup;
1449
1450 status = -1;
1451 break;
1452 case ULHR_BAD:
1453 status = error(_("unable to unpack %s header"),
1454 oid_to_hex(oid));
1455 break;
1456 case ULHR_TOO_LONG:
1457 status = error(_("header for %s too long, exceeds %d bytes"),
1458 oid_to_hex(oid), MAX_HEADER_LEN);
1459 break;
1460 }
1461
1462 git_inflate_end(&stream);
1463 cleanup:
1464 munmap(map, mapsize);
1465 if (oi->sizep == &size_scratch)
1466 oi->sizep = NULL;
1467 strbuf_release(&hdrbuf);
1468 if (oi->typep == &type_scratch)
1469 oi->typep = NULL;
1470 oi->whence = OI_LOOSE;
1471 return status;
1472 }
1473
1474 int obj_read_use_lock = 0;
1475 pthread_mutex_t obj_read_mutex;
1476
1477 void enable_obj_read_lock(void)
1478 {
1479 if (obj_read_use_lock)
1480 return;
1481
1482 obj_read_use_lock = 1;
1483 init_recursive_mutex(&obj_read_mutex);
1484 }
1485
1486 void disable_obj_read_lock(void)
1487 {
1488 if (!obj_read_use_lock)
1489 return;
1490
1491 obj_read_use_lock = 0;
1492 pthread_mutex_destroy(&obj_read_mutex);
1493 }
1494
1495 int fetch_if_missing = 1;
1496
1497 static int do_oid_object_info_extended(struct repository *r,
1498 const struct object_id *oid,
1499 struct object_info *oi, unsigned flags)
1500 {
1501 static struct object_info blank_oi = OBJECT_INFO_INIT;
1502 struct cached_object *co;
1503 struct pack_entry e;
1504 int rtype;
1505 const struct object_id *real = oid;
1506 int already_retried = 0;
1507
1508
1509 if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1510 real = lookup_replace_object(r, oid);
1511
1512 if (is_null_oid(real))
1513 return -1;
1514
1515 if (!oi)
1516 oi = &blank_oi;
1517
1518 co = find_cached_object(real);
1519 if (co) {
1520 if (oi->typep)
1521 *(oi->typep) = co->type;
1522 if (oi->sizep)
1523 *(oi->sizep) = co->size;
1524 if (oi->disk_sizep)
1525 *(oi->disk_sizep) = 0;
1526 if (oi->delta_base_oid)
1527 oidclr(oi->delta_base_oid);
1528 if (oi->type_name)
1529 strbuf_addstr(oi->type_name, type_name(co->type));
1530 if (oi->contentp)
1531 *oi->contentp = xmemdupz(co->buf, co->size);
1532 oi->whence = OI_CACHED;
1533 return 0;
1534 }
1535
1536 while (1) {
1537 if (find_pack_entry(r, real, &e))
1538 break;
1539
1540 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1541 return -1;
1542
1543 /* Most likely it's a loose object. */
1544 if (!loose_object_info(r, real, oi, flags))
1545 return 0;
1546
1547 /* Not a loose object; someone else may have just packed it. */
1548 if (!(flags & OBJECT_INFO_QUICK)) {
1549 reprepare_packed_git(r);
1550 if (find_pack_entry(r, real, &e))
1551 break;
1552 }
1553
1554 /* Check if it is a missing object */
1555 if (fetch_if_missing && has_promisor_remote() &&
1556 !already_retried && r == the_repository &&
1557 !(flags & OBJECT_INFO_SKIP_FETCH_OBJECT)) {
1558 /*
1559 * TODO Investigate checking promisor_remote_get_direct()
1560 * TODO return value and stopping on error here.
1561 * TODO Pass a repository struct through
1562 * promisor_remote_get_direct(), such that arbitrary
1563 * repositories work.
1564 */
1565 promisor_remote_get_direct(r, real, 1);
1566 already_retried = 1;
1567 continue;
1568 }
1569
1570 return -1;
1571 }
1572
1573 if (oi == &blank_oi)
1574 /*
1575 * We know that the caller doesn't actually need the
1576 * information below, so return early.
1577 */
1578 return 0;
1579 rtype = packed_object_info(r, e.p, e.offset, oi);
1580 if (rtype < 0) {
1581 mark_bad_packed_object(e.p, real->hash);
1582 return do_oid_object_info_extended(r, real, oi, 0);
1583 } else if (oi->whence == OI_PACKED) {
1584 oi->u.packed.offset = e.offset;
1585 oi->u.packed.pack = e.p;
1586 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1587 rtype == OBJ_OFS_DELTA);
1588 }
1589
1590 return 0;
1591 }
1592
1593 int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1594 struct object_info *oi, unsigned flags)
1595 {
1596 int ret;
1597 obj_read_lock();
1598 ret = do_oid_object_info_extended(r, oid, oi, flags);
1599 obj_read_unlock();
1600 return ret;
1601 }
1602
1603
1604 /* returns enum object_type or negative */
1605 int oid_object_info(struct repository *r,
1606 const struct object_id *oid,
1607 unsigned long *sizep)
1608 {
1609 enum object_type type;
1610 struct object_info oi = OBJECT_INFO_INIT;
1611
1612 oi.typep = &type;
1613 oi.sizep = sizep;
1614 if (oid_object_info_extended(r, oid, &oi,
1615 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1616 return -1;
1617 return type;
1618 }
1619
1620 static void *read_object(struct repository *r,
1621 const struct object_id *oid, enum object_type *type,
1622 unsigned long *size)
1623 {
1624 struct object_info oi = OBJECT_INFO_INIT;
1625 void *content;
1626 oi.typep = type;
1627 oi.sizep = size;
1628 oi.contentp = &content;
1629
1630 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1631 return NULL;
1632 return content;
1633 }
1634
1635 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1636 struct object_id *oid)
1637 {
1638 struct cached_object *co;
1639
1640 hash_object_file(the_hash_algo, buf, len, type_name(type), oid);
1641 if (has_object_file_with_flags(oid, OBJECT_INFO_QUICK | OBJECT_INFO_SKIP_FETCH_OBJECT) ||
1642 find_cached_object(oid))
1643 return 0;
1644 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1645 co = &cached_objects[cached_object_nr++];
1646 co->size = len;
1647 co->type = type;
1648 co->buf = xmalloc(len);
1649 memcpy(co->buf, buf, len);
1650 oidcpy(&co->oid, oid);
1651 return 0;
1652 }
1653
1654 /*
1655 * This function dies on corrupt objects; the callers who want to
1656 * deal with them should arrange to call read_object() and give error
1657 * messages themselves.
1658 */
1659 void *read_object_file_extended(struct repository *r,
1660 const struct object_id *oid,
1661 enum object_type *type,
1662 unsigned long *size,
1663 int lookup_replace)
1664 {
1665 void *data;
1666 const struct packed_git *p;
1667 const char *path;
1668 struct stat st;
1669 const struct object_id *repl = lookup_replace ?
1670 lookup_replace_object(r, oid) : oid;
1671
1672 errno = 0;
1673 data = read_object(r, repl, type, size);
1674 if (data)
1675 return data;
1676
1677 obj_read_lock();
1678 if (errno && errno != ENOENT)
1679 die_errno(_("failed to read object %s"), oid_to_hex(oid));
1680
1681 /* die if we replaced an object with one that does not exist */
1682 if (repl != oid)
1683 die(_("replacement %s not found for %s"),
1684 oid_to_hex(repl), oid_to_hex(oid));
1685
1686 if (!stat_loose_object(r, repl, &st, &path))
1687 die(_("loose object %s (stored in %s) is corrupt"),
1688 oid_to_hex(repl), path);
1689
1690 if ((p = has_packed_and_bad(r, repl->hash)) != NULL)
1691 die(_("packed object %s (stored in %s) is corrupt"),
1692 oid_to_hex(repl), p->pack_name);
1693 obj_read_unlock();
1694
1695 return NULL;
1696 }
1697
1698 void *read_object_with_reference(struct repository *r,
1699 const struct object_id *oid,
1700 const char *required_type_name,
1701 unsigned long *size,
1702 struct object_id *actual_oid_return)
1703 {
1704 enum object_type type, required_type;
1705 void *buffer;
1706 unsigned long isize;
1707 struct object_id actual_oid;
1708
1709 required_type = type_from_string(required_type_name);
1710 oidcpy(&actual_oid, oid);
1711 while (1) {
1712 int ref_length = -1;
1713 const char *ref_type = NULL;
1714
1715 buffer = repo_read_object_file(r, &actual_oid, &type, &isize);
1716 if (!buffer)
1717 return NULL;
1718 if (type == required_type) {
1719 *size = isize;
1720 if (actual_oid_return)
1721 oidcpy(actual_oid_return, &actual_oid);
1722 return buffer;
1723 }
1724 /* Handle references */
1725 else if (type == OBJ_COMMIT)
1726 ref_type = "tree ";
1727 else if (type == OBJ_TAG)
1728 ref_type = "object ";
1729 else {
1730 free(buffer);
1731 return NULL;
1732 }
1733 ref_length = strlen(ref_type);
1734
1735 if (ref_length + the_hash_algo->hexsz > isize ||
1736 memcmp(buffer, ref_type, ref_length) ||
1737 get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1738 free(buffer);
1739 return NULL;
1740 }
1741 free(buffer);
1742 /* Now we have the ID of the referred-to object in
1743 * actual_oid. Check again. */
1744 }
1745 }
1746
1747 static void write_object_file_prepare(const struct git_hash_algo *algo,
1748 const void *buf, unsigned long len,
1749 const char *type, struct object_id *oid,
1750 char *hdr, int *hdrlen)
1751 {
1752 git_hash_ctx c;
1753
1754 /* Generate the header */
1755 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %"PRIuMAX , type, (uintmax_t)len)+1;
1756
1757 /* Sha1.. */
1758 algo->init_fn(&c);
1759 algo->update_fn(&c, hdr, *hdrlen);
1760 algo->update_fn(&c, buf, len);
1761 algo->final_oid_fn(oid, &c);
1762 }
1763
1764 /*
1765 * Move the just written object into its final resting place.
1766 */
1767 int finalize_object_file(const char *tmpfile, const char *filename)
1768 {
1769 int ret = 0;
1770
1771 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1772 goto try_rename;
1773 else if (link(tmpfile, filename))
1774 ret = errno;
1775
1776 /*
1777 * Coda hack - coda doesn't like cross-directory links,
1778 * so we fall back to a rename, which will mean that it
1779 * won't be able to check collisions, but that's not a
1780 * big deal.
1781 *
1782 * The same holds for FAT formatted media.
1783 *
1784 * When this succeeds, we just return. We have nothing
1785 * left to unlink.
1786 */
1787 if (ret && ret != EEXIST) {
1788 try_rename:
1789 if (!rename(tmpfile, filename))
1790 goto out;
1791 ret = errno;
1792 }
1793 unlink_or_warn(tmpfile);
1794 if (ret) {
1795 if (ret != EEXIST) {
1796 return error_errno(_("unable to write file %s"), filename);
1797 }
1798 /* FIXME!!! Collision check here ? */
1799 }
1800
1801 out:
1802 if (adjust_shared_perm(filename))
1803 return error(_("unable to set permission to '%s'"), filename);
1804 return 0;
1805 }
1806
1807 static int write_buffer(int fd, const void *buf, size_t len)
1808 {
1809 if (write_in_full(fd, buf, len) < 0)
1810 return error_errno(_("file write error"));
1811 return 0;
1812 }
1813
1814 int hash_object_file(const struct git_hash_algo *algo, const void *buf,
1815 unsigned long len, const char *type,
1816 struct object_id *oid)
1817 {
1818 char hdr[MAX_HEADER_LEN];
1819 int hdrlen = sizeof(hdr);
1820 write_object_file_prepare(algo, buf, len, type, oid, hdr, &hdrlen);
1821 return 0;
1822 }
1823
1824 /* Finalize a file on disk, and close it. */
1825 static void close_loose_object(int fd)
1826 {
1827 if (fsync_object_files)
1828 fsync_or_die(fd, "loose object file");
1829 if (close(fd) != 0)
1830 die_errno(_("error when closing loose object file"));
1831 }
1832
1833 /* Size of directory component, including the ending '/' */
1834 static inline int directory_size(const char *filename)
1835 {
1836 const char *s = strrchr(filename, '/');
1837 if (!s)
1838 return 0;
1839 return s - filename + 1;
1840 }
1841
1842 /*
1843 * This creates a temporary file in the same directory as the final
1844 * 'filename'
1845 *
1846 * We want to avoid cross-directory filename renames, because those
1847 * can have problems on various filesystems (FAT, NFS, Coda).
1848 */
1849 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1850 {
1851 int fd, dirlen = directory_size(filename);
1852
1853 strbuf_reset(tmp);
1854 strbuf_add(tmp, filename, dirlen);
1855 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1856 fd = git_mkstemp_mode(tmp->buf, 0444);
1857 if (fd < 0 && dirlen && errno == ENOENT) {
1858 /*
1859 * Make sure the directory exists; note that the contents
1860 * of the buffer are undefined after mkstemp returns an
1861 * error, so we have to rewrite the whole buffer from
1862 * scratch.
1863 */
1864 strbuf_reset(tmp);
1865 strbuf_add(tmp, filename, dirlen - 1);
1866 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1867 return -1;
1868 if (adjust_shared_perm(tmp->buf))
1869 return -1;
1870
1871 /* Try again */
1872 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1873 fd = git_mkstemp_mode(tmp->buf, 0444);
1874 }
1875 return fd;
1876 }
1877
1878 static int write_loose_object(const struct object_id *oid, char *hdr,
1879 int hdrlen, const void *buf, unsigned long len,
1880 time_t mtime)
1881 {
1882 int fd, ret;
1883 unsigned char compressed[4096];
1884 git_zstream stream;
1885 git_hash_ctx c;
1886 struct object_id parano_oid;
1887 static struct strbuf tmp_file = STRBUF_INIT;
1888 static struct strbuf filename = STRBUF_INIT;
1889
1890 loose_object_path(the_repository, &filename, oid);
1891
1892 fd = create_tmpfile(&tmp_file, filename.buf);
1893 if (fd < 0) {
1894 if (errno == EACCES)
1895 return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
1896 else
1897 return error_errno(_("unable to create temporary file"));
1898 }
1899
1900 /* Set it up */
1901 git_deflate_init(&stream, zlib_compression_level);
1902 stream.next_out = compressed;
1903 stream.avail_out = sizeof(compressed);
1904 the_hash_algo->init_fn(&c);
1905
1906 /* First header.. */
1907 stream.next_in = (unsigned char *)hdr;
1908 stream.avail_in = hdrlen;
1909 while (git_deflate(&stream, 0) == Z_OK)
1910 ; /* nothing */
1911 the_hash_algo->update_fn(&c, hdr, hdrlen);
1912
1913 /* Then the data itself.. */
1914 stream.next_in = (void *)buf;
1915 stream.avail_in = len;
1916 do {
1917 unsigned char *in0 = stream.next_in;
1918 ret = git_deflate(&stream, Z_FINISH);
1919 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1920 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1921 die(_("unable to write loose object file"));
1922 stream.next_out = compressed;
1923 stream.avail_out = sizeof(compressed);
1924 } while (ret == Z_OK);
1925
1926 if (ret != Z_STREAM_END)
1927 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
1928 ret);
1929 ret = git_deflate_end_gently(&stream);
1930 if (ret != Z_OK)
1931 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
1932 ret);
1933 the_hash_algo->final_oid_fn(&parano_oid, &c);
1934 if (!oideq(oid, &parano_oid))
1935 die(_("confused by unstable object source data for %s"),
1936 oid_to_hex(oid));
1937
1938 close_loose_object(fd);
1939
1940 if (mtime) {
1941 struct utimbuf utb;
1942 utb.actime = mtime;
1943 utb.modtime = mtime;
1944 if (utime(tmp_file.buf, &utb) < 0)
1945 warning_errno(_("failed utime() on %s"), tmp_file.buf);
1946 }
1947
1948 return finalize_object_file(tmp_file.buf, filename.buf);
1949 }
1950
1951 static int freshen_loose_object(const struct object_id *oid)
1952 {
1953 return check_and_freshen(oid, 1);
1954 }
1955
1956 static int freshen_packed_object(const struct object_id *oid)
1957 {
1958 struct pack_entry e;
1959 if (!find_pack_entry(the_repository, oid, &e))
1960 return 0;
1961 if (e.p->freshened)
1962 return 1;
1963 if (!freshen_file(e.p->pack_name))
1964 return 0;
1965 e.p->freshened = 1;
1966 return 1;
1967 }
1968
1969 int write_object_file(const void *buf, unsigned long len, const char *type,
1970 struct object_id *oid)
1971 {
1972 char hdr[MAX_HEADER_LEN];
1973 int hdrlen = sizeof(hdr);
1974
1975 /* Normally if we have it in the pack then we do not bother writing
1976 * it out into .git/objects/??/?{38} file.
1977 */
1978 write_object_file_prepare(the_hash_algo, buf, len, type, oid, hdr,
1979 &hdrlen);
1980 if (freshen_packed_object(oid) || freshen_loose_object(oid))
1981 return 0;
1982 return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
1983 }
1984
1985 int hash_object_file_literally(const void *buf, unsigned long len,
1986 const char *type, struct object_id *oid,
1987 unsigned flags)
1988 {
1989 char *header;
1990 int hdrlen, status = 0;
1991
1992 /* type string, SP, %lu of the length plus NUL must fit this */
1993 hdrlen = strlen(type) + MAX_HEADER_LEN;
1994 header = xmalloc(hdrlen);
1995 write_object_file_prepare(the_hash_algo, buf, len, type, oid, header,
1996 &hdrlen);
1997
1998 if (!(flags & HASH_WRITE_OBJECT))
1999 goto cleanup;
2000 if (freshen_packed_object(oid) || freshen_loose_object(oid))
2001 goto cleanup;
2002 status = write_loose_object(oid, header, hdrlen, buf, len, 0);
2003
2004 cleanup:
2005 free(header);
2006 return status;
2007 }
2008
2009 int force_object_loose(const struct object_id *oid, time_t mtime)
2010 {
2011 void *buf;
2012 unsigned long len;
2013 enum object_type type;
2014 char hdr[MAX_HEADER_LEN];
2015 int hdrlen;
2016 int ret;
2017
2018 if (has_loose_object(oid))
2019 return 0;
2020 buf = read_object(the_repository, oid, &type, &len);
2021 if (!buf)
2022 return error(_("cannot read object for %s"), oid_to_hex(oid));
2023 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(type), (uintmax_t)len) + 1;
2024 ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
2025 free(buf);
2026
2027 return ret;
2028 }
2029
2030 int has_object(struct repository *r, const struct object_id *oid,
2031 unsigned flags)
2032 {
2033 int quick = !(flags & HAS_OBJECT_RECHECK_PACKED);
2034 unsigned object_info_flags = OBJECT_INFO_SKIP_FETCH_OBJECT |
2035 (quick ? OBJECT_INFO_QUICK : 0);
2036
2037 if (!startup_info->have_repository)
2038 return 0;
2039 return oid_object_info_extended(r, oid, NULL, object_info_flags) >= 0;
2040 }
2041
2042 int repo_has_object_file_with_flags(struct repository *r,
2043 const struct object_id *oid, int flags)
2044 {
2045 if (!startup_info->have_repository)
2046 return 0;
2047 return oid_object_info_extended(r, oid, NULL, flags) >= 0;
2048 }
2049
2050 int repo_has_object_file(struct repository *r,
2051 const struct object_id *oid)
2052 {
2053 return repo_has_object_file_with_flags(r, oid, 0);
2054 }
2055
2056 static void check_tree(const void *buf, size_t size)
2057 {
2058 struct tree_desc desc;
2059 struct name_entry entry;
2060
2061 init_tree_desc(&desc, buf, size);
2062 while (tree_entry(&desc, &entry))
2063 /* do nothing
2064 * tree_entry() will die() on malformed entries */
2065 ;
2066 }
2067
2068 static void check_commit(const void *buf, size_t size)
2069 {
2070 struct commit c;
2071 memset(&c, 0, sizeof(c));
2072 if (parse_commit_buffer(the_repository, &c, buf, size, 0))
2073 die(_("corrupt commit"));
2074 }
2075
2076 static void check_tag(const void *buf, size_t size)
2077 {
2078 struct tag t;
2079 memset(&t, 0, sizeof(t));
2080 if (parse_tag_buffer(the_repository, &t, buf, size))
2081 die(_("corrupt tag"));
2082 }
2083
2084 static int index_mem(struct index_state *istate,
2085 struct object_id *oid, void *buf, size_t size,
2086 enum object_type type,
2087 const char *path, unsigned flags)
2088 {
2089 int ret, re_allocated = 0;
2090 int write_object = flags & HASH_WRITE_OBJECT;
2091
2092 if (!type)
2093 type = OBJ_BLOB;
2094
2095 /*
2096 * Convert blobs to git internal format
2097 */
2098 if ((type == OBJ_BLOB) && path) {
2099 struct strbuf nbuf = STRBUF_INIT;
2100 if (convert_to_git(istate, path, buf, size, &nbuf,
2101 get_conv_flags(flags))) {
2102 buf = strbuf_detach(&nbuf, &size);
2103 re_allocated = 1;
2104 }
2105 }
2106 if (flags & HASH_FORMAT_CHECK) {
2107 if (type == OBJ_TREE)
2108 check_tree(buf, size);
2109 if (type == OBJ_COMMIT)
2110 check_commit(buf, size);
2111 if (type == OBJ_TAG)
2112 check_tag(buf, size);
2113 }
2114
2115 if (write_object)
2116 ret = write_object_file(buf, size, type_name(type), oid);
2117 else
2118 ret = hash_object_file(the_hash_algo, buf, size,
2119 type_name(type), oid);
2120 if (re_allocated)
2121 free(buf);
2122 return ret;
2123 }
2124
2125 static int index_stream_convert_blob(struct index_state *istate,
2126 struct object_id *oid,
2127 int fd,
2128 const char *path,
2129 unsigned flags)
2130 {
2131 int ret;
2132 const int write_object = flags & HASH_WRITE_OBJECT;
2133 struct strbuf sbuf = STRBUF_INIT;
2134
2135 assert(path);
2136 assert(would_convert_to_git_filter_fd(istate, path));
2137
2138 convert_to_git_filter_fd(istate, path, fd, &sbuf,
2139 get_conv_flags(flags));
2140
2141 if (write_object)
2142 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
2143 oid);
2144 else
2145 ret = hash_object_file(the_hash_algo, sbuf.buf, sbuf.len,
2146 type_name(OBJ_BLOB), oid);
2147 strbuf_release(&sbuf);
2148 return ret;
2149 }
2150
2151 static int index_pipe(struct index_state *istate, struct object_id *oid,
2152 int fd, enum object_type type,
2153 const char *path, unsigned flags)
2154 {
2155 struct strbuf sbuf = STRBUF_INIT;
2156 int ret;
2157
2158 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2159 ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags);
2160 else
2161 ret = -1;
2162 strbuf_release(&sbuf);
2163 return ret;
2164 }
2165
2166 #define SMALL_FILE_SIZE (32*1024)
2167
2168 static int index_core(struct index_state *istate,
2169 struct object_id *oid, int fd, size_t size,
2170 enum object_type type, const char *path,
2171 unsigned flags)
2172 {
2173 int ret;
2174
2175 if (!size) {
2176 ret = index_mem(istate, oid, "", size, type, path, flags);
2177 } else if (size <= SMALL_FILE_SIZE) {
2178 char *buf = xmalloc(size);
2179 ssize_t read_result = read_in_full(fd, buf, size);
2180 if (read_result < 0)
2181 ret = error_errno(_("read error while indexing %s"),
2182 path ? path : "<unknown>");
2183 else if (read_result != size)
2184 ret = error(_("short read while indexing %s"),
2185 path ? path : "<unknown>");
2186 else
2187 ret = index_mem(istate, oid, buf, size, type, path, flags);
2188 free(buf);
2189 } else {
2190 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2191 ret = index_mem(istate, oid, buf, size, type, path, flags);
2192 munmap(buf, size);
2193 }
2194 return ret;
2195 }
2196
2197 /*
2198 * This creates one packfile per large blob unless bulk-checkin
2199 * machinery is "plugged".
2200 *
2201 * This also bypasses the usual "convert-to-git" dance, and that is on
2202 * purpose. We could write a streaming version of the converting
2203 * functions and insert that before feeding the data to fast-import
2204 * (or equivalent in-core API described above). However, that is
2205 * somewhat complicated, as we do not know the size of the filter
2206 * result, which we need to know beforehand when writing a git object.
2207 * Since the primary motivation for trying to stream from the working
2208 * tree file and to avoid mmaping it in core is to deal with large
2209 * binary blobs, they generally do not want to get any conversion, and
2210 * callers should avoid this code path when filters are requested.
2211 */
2212 static int index_stream(struct object_id *oid, int fd, size_t size,
2213 enum object_type type, const char *path,
2214 unsigned flags)
2215 {
2216 return index_bulk_checkin(oid, fd, size, type, path, flags);
2217 }
2218
2219 int index_fd(struct index_state *istate, struct object_id *oid,
2220 int fd, struct stat *st,
2221 enum object_type type, const char *path, unsigned flags)
2222 {
2223 int ret;
2224
2225 /*
2226 * Call xsize_t() only when needed to avoid potentially unnecessary
2227 * die() for large files.
2228 */
2229 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, path))
2230 ret = index_stream_convert_blob(istate, oid, fd, path, flags);
2231 else if (!S_ISREG(st->st_mode))
2232 ret = index_pipe(istate, oid, fd, type, path, flags);
2233 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
2234 (path && would_convert_to_git(istate, path)))
2235 ret = index_core(istate, oid, fd, xsize_t(st->st_size),
2236 type, path, flags);
2237 else
2238 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
2239 flags);
2240 close(fd);
2241 return ret;
2242 }
2243
2244 int index_path(struct index_state *istate, struct object_id *oid,
2245 const char *path, struct stat *st, unsigned flags)
2246 {
2247 int fd;
2248 struct strbuf sb = STRBUF_INIT;
2249 int rc = 0;
2250
2251 switch (st->st_mode & S_IFMT) {
2252 case S_IFREG:
2253 fd = open(path, O_RDONLY);
2254 if (fd < 0)
2255 return error_errno("open(\"%s\")", path);
2256 if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0)
2257 return error(_("%s: failed to insert into database"),
2258 path);
2259 break;
2260 case S_IFLNK:
2261 if (strbuf_readlink(&sb, path, st->st_size))
2262 return error_errno("readlink(\"%s\")", path);
2263 if (!(flags & HASH_WRITE_OBJECT))
2264 hash_object_file(the_hash_algo, sb.buf, sb.len,
2265 blob_type, oid);
2266 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
2267 rc = error(_("%s: failed to insert into database"), path);
2268 strbuf_release(&sb);
2269 break;
2270 case S_IFDIR:
2271 return resolve_gitlink_ref(path, "HEAD", oid);
2272 default:
2273 return error(_("%s: unsupported file type"), path);
2274 }
2275 return rc;
2276 }
2277
2278 int read_pack_header(int fd, struct pack_header *header)
2279 {
2280 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
2281 /* "eof before pack header was fully read" */
2282 return PH_ERROR_EOF;
2283
2284 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2285 /* "protocol error (pack signature mismatch detected)" */
2286 return PH_ERROR_PACK_SIGNATURE;
2287 if (!pack_version_ok(header->hdr_version))
2288 /* "protocol error (pack version unsupported)" */
2289 return PH_ERROR_PROTOCOL;
2290 return 0;
2291 }
2292
2293 void assert_oid_type(const struct object_id *oid, enum object_type expect)
2294 {
2295 enum object_type type = oid_object_info(the_repository, oid, NULL);
2296 if (type < 0)
2297 die(_("%s is not a valid object"), oid_to_hex(oid));
2298 if (type != expect)
2299 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
2300 type_name(expect));
2301 }
2302
2303 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2304 struct strbuf *path,
2305 each_loose_object_fn obj_cb,
2306 each_loose_cruft_fn cruft_cb,
2307 each_loose_subdir_fn subdir_cb,
2308 void *data)
2309 {
2310 size_t origlen, baselen;
2311 DIR *dir;
2312 struct dirent *de;
2313 int r = 0;
2314 struct object_id oid;
2315
2316 if (subdir_nr > 0xff)
2317 BUG("invalid loose object subdirectory: %x", subdir_nr);
2318
2319 origlen = path->len;
2320 strbuf_complete(path, '/');
2321 strbuf_addf(path, "%02x", subdir_nr);
2322
2323 dir = opendir(path->buf);
2324 if (!dir) {
2325 if (errno != ENOENT)
2326 r = error_errno(_("unable to open %s"), path->buf);
2327 strbuf_setlen(path, origlen);
2328 return r;
2329 }
2330
2331 oid.hash[0] = subdir_nr;
2332 strbuf_addch(path, '/');
2333 baselen = path->len;
2334
2335 while ((de = readdir_skip_dot_and_dotdot(dir))) {
2336 size_t namelen;
2337
2338 namelen = strlen(de->d_name);
2339 strbuf_setlen(path, baselen);
2340 strbuf_add(path, de->d_name, namelen);
2341 if (namelen == the_hash_algo->hexsz - 2 &&
2342 !hex_to_bytes(oid.hash + 1, de->d_name,
2343 the_hash_algo->rawsz - 1)) {
2344 oid_set_algo(&oid, the_hash_algo);
2345 if (obj_cb) {
2346 r = obj_cb(&oid, path->buf, data);
2347 if (r)
2348 break;
2349 }
2350 continue;
2351 }
2352
2353 if (cruft_cb) {
2354 r = cruft_cb(de->d_name, path->buf, data);
2355 if (r)
2356 break;
2357 }
2358 }
2359 closedir(dir);
2360
2361 strbuf_setlen(path, baselen - 1);
2362 if (!r && subdir_cb)
2363 r = subdir_cb(subdir_nr, path->buf, data);
2364
2365 strbuf_setlen(path, origlen);
2366
2367 return r;
2368 }
2369
2370 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2371 each_loose_object_fn obj_cb,
2372 each_loose_cruft_fn cruft_cb,
2373 each_loose_subdir_fn subdir_cb,
2374 void *data)
2375 {
2376 int r = 0;
2377 int i;
2378
2379 for (i = 0; i < 256; i++) {
2380 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2381 subdir_cb, data);
2382 if (r)
2383 break;
2384 }
2385
2386 return r;
2387 }
2388
2389 int for_each_loose_file_in_objdir(const char *path,
2390 each_loose_object_fn obj_cb,
2391 each_loose_cruft_fn cruft_cb,
2392 each_loose_subdir_fn subdir_cb,
2393 void *data)
2394 {
2395 struct strbuf buf = STRBUF_INIT;
2396 int r;
2397
2398 strbuf_addstr(&buf, path);
2399 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2400 subdir_cb, data);
2401 strbuf_release(&buf);
2402
2403 return r;
2404 }
2405
2406 int for_each_loose_object(each_loose_object_fn cb, void *data,
2407 enum for_each_object_flags flags)
2408 {
2409 struct object_directory *odb;
2410
2411 prepare_alt_odb(the_repository);
2412 for (odb = the_repository->objects->odb; odb; odb = odb->next) {
2413 int r = for_each_loose_file_in_objdir(odb->path, cb, NULL,
2414 NULL, data);
2415 if (r)
2416 return r;
2417
2418 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2419 break;
2420 }
2421
2422 return 0;
2423 }
2424
2425 static int append_loose_object(const struct object_id *oid, const char *path,
2426 void *data)
2427 {
2428 oid_array_append(data, oid);
2429 return 0;
2430 }
2431
2432 struct oid_array *odb_loose_cache(struct object_directory *odb,
2433 const struct object_id *oid)
2434 {
2435 int subdir_nr = oid->hash[0];
2436 struct strbuf buf = STRBUF_INIT;
2437
2438 if (subdir_nr < 0 ||
2439 subdir_nr >= ARRAY_SIZE(odb->loose_objects_subdir_seen))
2440 BUG("subdir_nr out of range");
2441
2442 if (odb->loose_objects_subdir_seen[subdir_nr])
2443 return &odb->loose_objects_cache[subdir_nr];
2444
2445 strbuf_addstr(&buf, odb->path);
2446 for_each_file_in_obj_subdir(subdir_nr, &buf,
2447 append_loose_object,
2448 NULL, NULL,
2449 &odb->loose_objects_cache[subdir_nr]);
2450 odb->loose_objects_subdir_seen[subdir_nr] = 1;
2451 strbuf_release(&buf);
2452 return &odb->loose_objects_cache[subdir_nr];
2453 }
2454
2455 void odb_clear_loose_cache(struct object_directory *odb)
2456 {
2457 int i;
2458
2459 for (i = 0; i < ARRAY_SIZE(odb->loose_objects_cache); i++)
2460 oid_array_clear(&odb->loose_objects_cache[i]);
2461 memset(&odb->loose_objects_subdir_seen, 0,
2462 sizeof(odb->loose_objects_subdir_seen));
2463 }
2464
2465 static int check_stream_oid(git_zstream *stream,
2466 const char *hdr,
2467 unsigned long size,
2468 const char *path,
2469 const struct object_id *expected_oid)
2470 {
2471 git_hash_ctx c;
2472 struct object_id real_oid;
2473 unsigned char buf[4096];
2474 unsigned long total_read;
2475 int status = Z_OK;
2476
2477 the_hash_algo->init_fn(&c);
2478 the_hash_algo->update_fn(&c, hdr, stream->total_out);
2479
2480 /*
2481 * We already read some bytes into hdr, but the ones up to the NUL
2482 * do not count against the object's content size.
2483 */
2484 total_read = stream->total_out - strlen(hdr) - 1;
2485
2486 /*
2487 * This size comparison must be "<=" to read the final zlib packets;
2488 * see the comment in unpack_loose_rest for details.
2489 */
2490 while (total_read <= size &&
2491 (status == Z_OK ||
2492 (status == Z_BUF_ERROR && !stream->avail_out))) {
2493 stream->next_out = buf;
2494 stream->avail_out = sizeof(buf);
2495 if (size - total_read < stream->avail_out)
2496 stream->avail_out = size - total_read;
2497 status = git_inflate(stream, Z_FINISH);
2498 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2499 total_read += stream->next_out - buf;
2500 }
2501 git_inflate_end(stream);
2502
2503 if (status != Z_STREAM_END) {
2504 error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid));
2505 return -1;
2506 }
2507 if (stream->avail_in) {
2508 error(_("garbage at end of loose object '%s'"),
2509 oid_to_hex(expected_oid));
2510 return -1;
2511 }
2512
2513 the_hash_algo->final_oid_fn(&real_oid, &c);
2514 if (!oideq(expected_oid, &real_oid)) {
2515 error(_("hash mismatch for %s (expected %s)"), path,
2516 oid_to_hex(expected_oid));
2517 return -1;
2518 }
2519
2520 return 0;
2521 }
2522
2523 int read_loose_object(const char *path,
2524 const struct object_id *expected_oid,
2525 struct object_id *real_oid,
2526 void **contents,
2527 struct object_info *oi)
2528 {
2529 int ret = -1;
2530 void *map = NULL;
2531 unsigned long mapsize;
2532 git_zstream stream;
2533 char hdr[MAX_HEADER_LEN];
2534 unsigned long *size = oi->sizep;
2535
2536 map = map_loose_object_1(the_repository, path, NULL, &mapsize);
2537 if (!map) {
2538 error_errno(_("unable to mmap %s"), path);
2539 goto out;
2540 }
2541
2542 if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr),
2543 NULL) < 0) {
2544 error(_("unable to unpack header of %s"), path);
2545 goto out;
2546 }
2547
2548 if (parse_loose_header(hdr, oi) < 0) {
2549 error(_("unable to parse header of %s"), path);
2550 git_inflate_end(&stream);
2551 goto out;
2552 }
2553
2554 if (*oi->typep == OBJ_BLOB && *size > big_file_threshold) {
2555 if (check_stream_oid(&stream, hdr, *size, path, expected_oid) < 0)
2556 goto out;
2557 } else {
2558 *contents = unpack_loose_rest(&stream, hdr, *size, expected_oid);
2559 if (!*contents) {
2560 error(_("unable to unpack contents of %s"), path);
2561 git_inflate_end(&stream);
2562 goto out;
2563 }
2564 if (check_object_signature(the_repository, expected_oid,
2565 *contents, *size,
2566 oi->type_name->buf, real_oid))
2567 goto out;
2568 }
2569
2570 ret = 0; /* everything checks out */
2571
2572 out:
2573 if (map)
2574 munmap(map, mapsize);
2575 return ret;
2576 }