]> git.ipfire.org Git - thirdparty/git.git/blob - object-file.c
l10n: bg.po: Updated Bulgarian translation (5195t)
[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 #include "submodule.h"
36
37 /* The maximum size for an object header. */
38 #define MAX_HEADER_LEN 32
39
40
41 #define EMPTY_TREE_SHA1_BIN_LITERAL \
42 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60" \
43 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04"
44 #define EMPTY_TREE_SHA256_BIN_LITERAL \
45 "\x6e\xf1\x9b\x41\x22\x5c\x53\x69\xf1\xc1" \
46 "\x04\xd4\x5d\x8d\x85\xef\xa9\xb0\x57\xb5" \
47 "\x3b\x14\xb4\xb9\xb9\x39\xdd\x74\xde\xcc" \
48 "\x53\x21"
49
50 #define EMPTY_BLOB_SHA1_BIN_LITERAL \
51 "\xe6\x9d\xe2\x9b\xb2\xd1\xd6\x43\x4b\x8b" \
52 "\x29\xae\x77\x5a\xd8\xc2\xe4\x8c\x53\x91"
53 #define EMPTY_BLOB_SHA256_BIN_LITERAL \
54 "\x47\x3a\x0f\x4c\x3b\xe8\xa9\x36\x81\xa2" \
55 "\x67\xe3\xb1\xe9\xa7\xdc\xda\x11\x85\x43" \
56 "\x6f\xe1\x41\xf7\x74\x91\x20\xa3\x03\x72" \
57 "\x18\x13"
58
59 static const struct object_id empty_tree_oid = {
60 .hash = EMPTY_TREE_SHA1_BIN_LITERAL,
61 .algo = GIT_HASH_SHA1,
62 };
63 static const struct object_id empty_blob_oid = {
64 .hash = EMPTY_BLOB_SHA1_BIN_LITERAL,
65 .algo = GIT_HASH_SHA1,
66 };
67 static const struct object_id null_oid_sha1 = {
68 .hash = {0},
69 .algo = GIT_HASH_SHA1,
70 };
71 static const struct object_id empty_tree_oid_sha256 = {
72 .hash = EMPTY_TREE_SHA256_BIN_LITERAL,
73 .algo = GIT_HASH_SHA256,
74 };
75 static const struct object_id empty_blob_oid_sha256 = {
76 .hash = EMPTY_BLOB_SHA256_BIN_LITERAL,
77 .algo = GIT_HASH_SHA256,
78 };
79 static const struct object_id null_oid_sha256 = {
80 .hash = {0},
81 .algo = GIT_HASH_SHA256,
82 };
83
84 static void git_hash_sha1_init(git_hash_ctx *ctx)
85 {
86 git_SHA1_Init(&ctx->sha1);
87 }
88
89 static void git_hash_sha1_clone(git_hash_ctx *dst, const git_hash_ctx *src)
90 {
91 git_SHA1_Clone(&dst->sha1, &src->sha1);
92 }
93
94 static void git_hash_sha1_update(git_hash_ctx *ctx, const void *data, size_t len)
95 {
96 git_SHA1_Update(&ctx->sha1, data, len);
97 }
98
99 static void git_hash_sha1_final(unsigned char *hash, git_hash_ctx *ctx)
100 {
101 git_SHA1_Final(hash, &ctx->sha1);
102 }
103
104 static void git_hash_sha1_final_oid(struct object_id *oid, git_hash_ctx *ctx)
105 {
106 git_SHA1_Final(oid->hash, &ctx->sha1);
107 memset(oid->hash + GIT_SHA1_RAWSZ, 0, GIT_MAX_RAWSZ - GIT_SHA1_RAWSZ);
108 oid->algo = GIT_HASH_SHA1;
109 }
110
111
112 static void git_hash_sha256_init(git_hash_ctx *ctx)
113 {
114 git_SHA256_Init(&ctx->sha256);
115 }
116
117 static void git_hash_sha256_clone(git_hash_ctx *dst, const git_hash_ctx *src)
118 {
119 git_SHA256_Clone(&dst->sha256, &src->sha256);
120 }
121
122 static void git_hash_sha256_update(git_hash_ctx *ctx, const void *data, size_t len)
123 {
124 git_SHA256_Update(&ctx->sha256, data, len);
125 }
126
127 static void git_hash_sha256_final(unsigned char *hash, git_hash_ctx *ctx)
128 {
129 git_SHA256_Final(hash, &ctx->sha256);
130 }
131
132 static void git_hash_sha256_final_oid(struct object_id *oid, git_hash_ctx *ctx)
133 {
134 git_SHA256_Final(oid->hash, &ctx->sha256);
135 /*
136 * This currently does nothing, so the compiler should optimize it out,
137 * but keep it in case we extend the hash size again.
138 */
139 memset(oid->hash + GIT_SHA256_RAWSZ, 0, GIT_MAX_RAWSZ - GIT_SHA256_RAWSZ);
140 oid->algo = GIT_HASH_SHA256;
141 }
142
143 static void git_hash_unknown_init(git_hash_ctx *ctx)
144 {
145 BUG("trying to init unknown hash");
146 }
147
148 static void git_hash_unknown_clone(git_hash_ctx *dst, const git_hash_ctx *src)
149 {
150 BUG("trying to clone unknown hash");
151 }
152
153 static void git_hash_unknown_update(git_hash_ctx *ctx, const void *data, size_t len)
154 {
155 BUG("trying to update unknown hash");
156 }
157
158 static void git_hash_unknown_final(unsigned char *hash, git_hash_ctx *ctx)
159 {
160 BUG("trying to finalize unknown hash");
161 }
162
163 static void git_hash_unknown_final_oid(struct object_id *oid, git_hash_ctx *ctx)
164 {
165 BUG("trying to finalize unknown hash");
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 GIT_SHA1_FORMAT_ID,
187 GIT_SHA1_RAWSZ,
188 GIT_SHA1_HEXSZ,
189 GIT_SHA1_BLKSZ,
190 git_hash_sha1_init,
191 git_hash_sha1_clone,
192 git_hash_sha1_update,
193 git_hash_sha1_final,
194 git_hash_sha1_final_oid,
195 &empty_tree_oid,
196 &empty_blob_oid,
197 &null_oid_sha1,
198 },
199 {
200 "sha256",
201 GIT_SHA256_FORMAT_ID,
202 GIT_SHA256_RAWSZ,
203 GIT_SHA256_HEXSZ,
204 GIT_SHA256_BLKSZ,
205 git_hash_sha256_init,
206 git_hash_sha256_clone,
207 git_hash_sha256_update,
208 git_hash_sha256_final,
209 git_hash_sha256_final_oid,
210 &empty_tree_oid_sha256,
211 &empty_blob_oid_sha256,
212 &null_oid_sha256,
213 }
214 };
215
216 const struct object_id *null_oid(void)
217 {
218 return the_hash_algo->null_oid;
219 }
220
221 const char *empty_tree_oid_hex(void)
222 {
223 static char buf[GIT_MAX_HEXSZ + 1];
224 return oid_to_hex_r(buf, the_hash_algo->empty_tree);
225 }
226
227 const char *empty_blob_oid_hex(void)
228 {
229 static char buf[GIT_MAX_HEXSZ + 1];
230 return oid_to_hex_r(buf, the_hash_algo->empty_blob);
231 }
232
233 int hash_algo_by_name(const char *name)
234 {
235 int i;
236 if (!name)
237 return GIT_HASH_UNKNOWN;
238 for (i = 1; i < GIT_HASH_NALGOS; i++)
239 if (!strcmp(name, hash_algos[i].name))
240 return i;
241 return GIT_HASH_UNKNOWN;
242 }
243
244 int hash_algo_by_id(uint32_t format_id)
245 {
246 int i;
247 for (i = 1; i < GIT_HASH_NALGOS; i++)
248 if (format_id == hash_algos[i].format_id)
249 return i;
250 return GIT_HASH_UNKNOWN;
251 }
252
253 int hash_algo_by_length(int len)
254 {
255 int i;
256 for (i = 1; i < GIT_HASH_NALGOS; i++)
257 if (len == hash_algos[i].rawsz)
258 return i;
259 return GIT_HASH_UNKNOWN;
260 }
261
262 /*
263 * This is meant to hold a *small* number of objects that you would
264 * want read_object_file() to be able to return, but yet you do not want
265 * to write them into the object store (e.g. a browse-only
266 * application).
267 */
268 static struct cached_object {
269 struct object_id oid;
270 enum object_type type;
271 void *buf;
272 unsigned long size;
273 } *cached_objects;
274 static int cached_object_nr, cached_object_alloc;
275
276 static struct cached_object empty_tree = {
277 { EMPTY_TREE_SHA1_BIN_LITERAL },
278 OBJ_TREE,
279 "",
280 0
281 };
282
283 static struct cached_object *find_cached_object(const struct object_id *oid)
284 {
285 int i;
286 struct cached_object *co = cached_objects;
287
288 for (i = 0; i < cached_object_nr; i++, co++) {
289 if (oideq(&co->oid, oid))
290 return co;
291 }
292 if (oideq(oid, the_hash_algo->empty_tree))
293 return &empty_tree;
294 return NULL;
295 }
296
297
298 static int get_conv_flags(unsigned flags)
299 {
300 if (flags & HASH_RENORMALIZE)
301 return CONV_EOL_RENORMALIZE;
302 else if (flags & HASH_WRITE_OBJECT)
303 return global_conv_flags_eol | CONV_WRITE_OBJECT;
304 else
305 return 0;
306 }
307
308
309 int mkdir_in_gitdir(const char *path)
310 {
311 if (mkdir(path, 0777)) {
312 int saved_errno = errno;
313 struct stat st;
314 struct strbuf sb = STRBUF_INIT;
315
316 if (errno != EEXIST)
317 return -1;
318 /*
319 * Are we looking at a path in a symlinked worktree
320 * whose original repository does not yet have it?
321 * e.g. .git/rr-cache pointing at its original
322 * repository in which the user hasn't performed any
323 * conflict resolution yet?
324 */
325 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
326 strbuf_readlink(&sb, path, st.st_size) ||
327 !is_absolute_path(sb.buf) ||
328 mkdir(sb.buf, 0777)) {
329 strbuf_release(&sb);
330 errno = saved_errno;
331 return -1;
332 }
333 strbuf_release(&sb);
334 }
335 return adjust_shared_perm(path);
336 }
337
338 static enum scld_error safe_create_leading_directories_1(char *path, int share)
339 {
340 char *next_component = path + offset_1st_component(path);
341 enum scld_error ret = SCLD_OK;
342
343 while (ret == SCLD_OK && next_component) {
344 struct stat st;
345 char *slash = next_component, slash_character;
346
347 while (*slash && !is_dir_sep(*slash))
348 slash++;
349
350 if (!*slash)
351 break;
352
353 next_component = slash + 1;
354 while (is_dir_sep(*next_component))
355 next_component++;
356 if (!*next_component)
357 break;
358
359 slash_character = *slash;
360 *slash = '\0';
361 if (!stat(path, &st)) {
362 /* path exists */
363 if (!S_ISDIR(st.st_mode)) {
364 errno = ENOTDIR;
365 ret = SCLD_EXISTS;
366 }
367 } else if (mkdir(path, 0777)) {
368 if (errno == EEXIST &&
369 !stat(path, &st) && S_ISDIR(st.st_mode))
370 ; /* somebody created it since we checked */
371 else if (errno == ENOENT)
372 /*
373 * Either mkdir() failed because
374 * somebody just pruned the containing
375 * directory, or stat() failed because
376 * the file that was in our way was
377 * just removed. Either way, inform
378 * the caller that it might be worth
379 * trying again:
380 */
381 ret = SCLD_VANISHED;
382 else
383 ret = SCLD_FAILED;
384 } else if (share && adjust_shared_perm(path)) {
385 ret = SCLD_PERMS;
386 }
387 *slash = slash_character;
388 }
389 return ret;
390 }
391
392 enum scld_error safe_create_leading_directories(char *path)
393 {
394 return safe_create_leading_directories_1(path, 1);
395 }
396
397 enum scld_error safe_create_leading_directories_no_share(char *path)
398 {
399 return safe_create_leading_directories_1(path, 0);
400 }
401
402 enum scld_error safe_create_leading_directories_const(const char *path)
403 {
404 int save_errno;
405 /* path points to cache entries, so xstrdup before messing with it */
406 char *buf = xstrdup(path);
407 enum scld_error result = safe_create_leading_directories(buf);
408
409 save_errno = errno;
410 free(buf);
411 errno = save_errno;
412 return result;
413 }
414
415 static void fill_loose_path(struct strbuf *buf, const struct object_id *oid)
416 {
417 int i;
418 for (i = 0; i < the_hash_algo->rawsz; i++) {
419 static char hex[] = "0123456789abcdef";
420 unsigned int val = oid->hash[i];
421 strbuf_addch(buf, hex[val >> 4]);
422 strbuf_addch(buf, hex[val & 0xf]);
423 if (!i)
424 strbuf_addch(buf, '/');
425 }
426 }
427
428 static const char *odb_loose_path(struct object_directory *odb,
429 struct strbuf *buf,
430 const struct object_id *oid)
431 {
432 strbuf_reset(buf);
433 strbuf_addstr(buf, odb->path);
434 strbuf_addch(buf, '/');
435 fill_loose_path(buf, oid);
436 return buf->buf;
437 }
438
439 const char *loose_object_path(struct repository *r, struct strbuf *buf,
440 const struct object_id *oid)
441 {
442 return odb_loose_path(r->objects->odb, buf, oid);
443 }
444
445 /*
446 * Return non-zero iff the path is usable as an alternate object database.
447 */
448 static int alt_odb_usable(struct raw_object_store *o,
449 struct strbuf *path,
450 const char *normalized_objdir, khiter_t *pos)
451 {
452 int r;
453
454 /* Detect cases where alternate disappeared */
455 if (!is_directory(path->buf)) {
456 error(_("object directory %s does not exist; "
457 "check .git/objects/info/alternates"),
458 path->buf);
459 return 0;
460 }
461
462 /*
463 * Prevent the common mistake of listing the same
464 * thing twice, or object directory itself.
465 */
466 if (!o->odb_by_path) {
467 khiter_t p;
468
469 o->odb_by_path = kh_init_odb_path_map();
470 assert(!o->odb->next);
471 p = kh_put_odb_path_map(o->odb_by_path, o->odb->path, &r);
472 assert(r == 1); /* never used */
473 kh_value(o->odb_by_path, p) = o->odb;
474 }
475 if (fspatheq(path->buf, normalized_objdir))
476 return 0;
477 *pos = kh_put_odb_path_map(o->odb_by_path, path->buf, &r);
478 /* r: 0 = exists, 1 = never used, 2 = deleted */
479 return r == 0 ? 0 : 1;
480 }
481
482 /*
483 * Prepare alternate object database registry.
484 *
485 * The variable alt_odb_list points at the list of struct
486 * object_directory. The elements on this list come from
487 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
488 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
489 * whose contents is similar to that environment variable but can be
490 * LF separated. Its base points at a statically allocated buffer that
491 * contains "/the/directory/corresponding/to/.git/objects/...", while
492 * its name points just after the slash at the end of ".git/objects/"
493 * in the example above, and has enough space to hold all hex characters
494 * of the object ID, an extra slash for the first level indirection, and
495 * the terminating NUL.
496 */
497 static void read_info_alternates(struct repository *r,
498 const char *relative_base,
499 int depth);
500 static int link_alt_odb_entry(struct repository *r, const struct strbuf *entry,
501 const char *relative_base, int depth, const char *normalized_objdir)
502 {
503 struct object_directory *ent;
504 struct strbuf pathbuf = STRBUF_INIT;
505 khiter_t pos;
506
507 if (!is_absolute_path(entry->buf) && relative_base) {
508 strbuf_realpath(&pathbuf, relative_base, 1);
509 strbuf_addch(&pathbuf, '/');
510 }
511 strbuf_addbuf(&pathbuf, entry);
512
513 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
514 error(_("unable to normalize alternate object path: %s"),
515 pathbuf.buf);
516 strbuf_release(&pathbuf);
517 return -1;
518 }
519
520 /*
521 * The trailing slash after the directory name is given by
522 * this function at the end. Remove duplicates.
523 */
524 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
525 strbuf_setlen(&pathbuf, pathbuf.len - 1);
526
527 if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir, &pos)) {
528 strbuf_release(&pathbuf);
529 return -1;
530 }
531
532 CALLOC_ARRAY(ent, 1);
533 /* pathbuf.buf is already in r->objects->odb_by_path */
534 ent->path = strbuf_detach(&pathbuf, NULL);
535
536 /* add the alternate entry */
537 *r->objects->odb_tail = ent;
538 r->objects->odb_tail = &(ent->next);
539 ent->next = NULL;
540 assert(r->objects->odb_by_path);
541 kh_value(r->objects->odb_by_path, pos) = ent;
542
543 /* recursively add alternates */
544 read_info_alternates(r, ent->path, depth + 1);
545
546 return 0;
547 }
548
549 static const char *parse_alt_odb_entry(const char *string,
550 int sep,
551 struct strbuf *out)
552 {
553 const char *end;
554
555 strbuf_reset(out);
556
557 if (*string == '#') {
558 /* comment; consume up to next separator */
559 end = strchrnul(string, sep);
560 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
561 /*
562 * quoted path; unquote_c_style has copied the
563 * data for us and set "end". Broken quoting (e.g.,
564 * an entry that doesn't end with a quote) falls
565 * back to the unquoted case below.
566 */
567 } else {
568 /* normal, unquoted path */
569 end = strchrnul(string, sep);
570 strbuf_add(out, string, end - string);
571 }
572
573 if (*end)
574 end++;
575 return end;
576 }
577
578 static void link_alt_odb_entries(struct repository *r, const char *alt,
579 int sep, const char *relative_base, int depth)
580 {
581 struct strbuf objdirbuf = STRBUF_INIT;
582 struct strbuf entry = STRBUF_INIT;
583
584 if (!alt || !*alt)
585 return;
586
587 if (depth > 5) {
588 error(_("%s: ignoring alternate object stores, nesting too deep"),
589 relative_base);
590 return;
591 }
592
593 strbuf_add_absolute_path(&objdirbuf, r->objects->odb->path);
594 if (strbuf_normalize_path(&objdirbuf) < 0)
595 die(_("unable to normalize object directory: %s"),
596 objdirbuf.buf);
597
598 while (*alt) {
599 alt = parse_alt_odb_entry(alt, sep, &entry);
600 if (!entry.len)
601 continue;
602 link_alt_odb_entry(r, &entry,
603 relative_base, depth, objdirbuf.buf);
604 }
605 strbuf_release(&entry);
606 strbuf_release(&objdirbuf);
607 }
608
609 static void read_info_alternates(struct repository *r,
610 const char *relative_base,
611 int depth)
612 {
613 char *path;
614 struct strbuf buf = STRBUF_INIT;
615
616 path = xstrfmt("%s/info/alternates", relative_base);
617 if (strbuf_read_file(&buf, path, 1024) < 0) {
618 warn_on_fopen_errors(path);
619 free(path);
620 return;
621 }
622
623 link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
624 strbuf_release(&buf);
625 free(path);
626 }
627
628 void add_to_alternates_file(const char *reference)
629 {
630 struct lock_file lock = LOCK_INIT;
631 char *alts = git_pathdup("objects/info/alternates");
632 FILE *in, *out;
633 int found = 0;
634
635 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
636 out = fdopen_lock_file(&lock, "w");
637 if (!out)
638 die_errno(_("unable to fdopen alternates lockfile"));
639
640 in = fopen(alts, "r");
641 if (in) {
642 struct strbuf line = STRBUF_INIT;
643
644 while (strbuf_getline(&line, in) != EOF) {
645 if (!strcmp(reference, line.buf)) {
646 found = 1;
647 break;
648 }
649 fprintf_or_die(out, "%s\n", line.buf);
650 }
651
652 strbuf_release(&line);
653 fclose(in);
654 }
655 else if (errno != ENOENT)
656 die_errno(_("unable to read alternates file"));
657
658 if (found) {
659 rollback_lock_file(&lock);
660 } else {
661 fprintf_or_die(out, "%s\n", reference);
662 if (commit_lock_file(&lock))
663 die_errno(_("unable to move new alternates file into place"));
664 if (the_repository->objects->loaded_alternates)
665 link_alt_odb_entries(the_repository, reference,
666 '\n', NULL, 0);
667 }
668 free(alts);
669 }
670
671 void add_to_alternates_memory(const char *reference)
672 {
673 /*
674 * Make sure alternates are initialized, or else our entry may be
675 * overwritten when they are.
676 */
677 prepare_alt_odb(the_repository);
678
679 link_alt_odb_entries(the_repository, reference,
680 '\n', NULL, 0);
681 }
682
683 struct object_directory *set_temporary_primary_odb(const char *dir, int will_destroy)
684 {
685 struct object_directory *new_odb;
686
687 /*
688 * Make sure alternates are initialized, or else our entry may be
689 * overwritten when they are.
690 */
691 prepare_alt_odb(the_repository);
692
693 /*
694 * Make a new primary odb and link the old primary ODB in as an
695 * alternate
696 */
697 new_odb = xcalloc(1, sizeof(*new_odb));
698 new_odb->path = xstrdup(dir);
699
700 /*
701 * Disable ref updates while a temporary odb is active, since
702 * the objects in the database may roll back.
703 */
704 new_odb->disable_ref_updates = 1;
705 new_odb->will_destroy = will_destroy;
706 new_odb->next = the_repository->objects->odb;
707 the_repository->objects->odb = new_odb;
708 return new_odb->next;
709 }
710
711 void restore_primary_odb(struct object_directory *restore_odb, const char *old_path)
712 {
713 struct object_directory *cur_odb = the_repository->objects->odb;
714
715 if (strcmp(old_path, cur_odb->path))
716 BUG("expected %s as primary object store; found %s",
717 old_path, cur_odb->path);
718
719 if (cur_odb->next != restore_odb)
720 BUG("we expect the old primary object store to be the first alternate");
721
722 the_repository->objects->odb = restore_odb;
723 free_object_directory(cur_odb);
724 }
725
726 /*
727 * Compute the exact path an alternate is at and returns it. In case of
728 * error NULL is returned and the human readable error is added to `err`
729 * `path` may be relative and should point to $GIT_DIR.
730 * `err` must not be null.
731 */
732 char *compute_alternate_path(const char *path, struct strbuf *err)
733 {
734 char *ref_git = NULL;
735 const char *repo;
736 int seen_error = 0;
737
738 ref_git = real_pathdup(path, 0);
739 if (!ref_git) {
740 seen_error = 1;
741 strbuf_addf(err, _("path '%s' does not exist"), path);
742 goto out;
743 }
744
745 repo = read_gitfile(ref_git);
746 if (!repo)
747 repo = read_gitfile(mkpath("%s/.git", ref_git));
748 if (repo) {
749 free(ref_git);
750 ref_git = xstrdup(repo);
751 }
752
753 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
754 char *ref_git_git = mkpathdup("%s/.git", ref_git);
755 free(ref_git);
756 ref_git = ref_git_git;
757 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
758 struct strbuf sb = STRBUF_INIT;
759 seen_error = 1;
760 if (get_common_dir(&sb, ref_git)) {
761 strbuf_addf(err,
762 _("reference repository '%s' as a linked "
763 "checkout is not supported yet."),
764 path);
765 goto out;
766 }
767
768 strbuf_addf(err, _("reference repository '%s' is not a "
769 "local repository."), path);
770 goto out;
771 }
772
773 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
774 strbuf_addf(err, _("reference repository '%s' is shallow"),
775 path);
776 seen_error = 1;
777 goto out;
778 }
779
780 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
781 strbuf_addf(err,
782 _("reference repository '%s' is grafted"),
783 path);
784 seen_error = 1;
785 goto out;
786 }
787
788 out:
789 if (seen_error) {
790 FREE_AND_NULL(ref_git);
791 }
792
793 return ref_git;
794 }
795
796 struct object_directory *find_odb(struct repository *r, const char *obj_dir)
797 {
798 struct object_directory *odb;
799 char *obj_dir_real = real_pathdup(obj_dir, 1);
800 struct strbuf odb_path_real = STRBUF_INIT;
801
802 prepare_alt_odb(r);
803 for (odb = r->objects->odb; odb; odb = odb->next) {
804 strbuf_realpath(&odb_path_real, odb->path, 1);
805 if (!strcmp(obj_dir_real, odb_path_real.buf))
806 break;
807 }
808
809 free(obj_dir_real);
810 strbuf_release(&odb_path_real);
811
812 if (!odb)
813 die(_("could not find object directory matching %s"), obj_dir);
814 return odb;
815 }
816
817 static void fill_alternate_refs_command(struct child_process *cmd,
818 const char *repo_path)
819 {
820 const char *value;
821
822 if (!git_config_get_value("core.alternateRefsCommand", &value)) {
823 cmd->use_shell = 1;
824
825 strvec_push(&cmd->args, value);
826 strvec_push(&cmd->args, repo_path);
827 } else {
828 cmd->git_cmd = 1;
829
830 strvec_pushf(&cmd->args, "--git-dir=%s", repo_path);
831 strvec_push(&cmd->args, "for-each-ref");
832 strvec_push(&cmd->args, "--format=%(objectname)");
833
834 if (!git_config_get_value("core.alternateRefsPrefixes", &value)) {
835 strvec_push(&cmd->args, "--");
836 strvec_split(&cmd->args, value);
837 }
838 }
839
840 strvec_pushv(&cmd->env_array, (const char **)local_repo_env);
841 cmd->out = -1;
842 }
843
844 static void read_alternate_refs(const char *path,
845 alternate_ref_fn *cb,
846 void *data)
847 {
848 struct child_process cmd = CHILD_PROCESS_INIT;
849 struct strbuf line = STRBUF_INIT;
850 FILE *fh;
851
852 fill_alternate_refs_command(&cmd, path);
853
854 if (start_command(&cmd))
855 return;
856
857 fh = xfdopen(cmd.out, "r");
858 while (strbuf_getline_lf(&line, fh) != EOF) {
859 struct object_id oid;
860 const char *p;
861
862 if (parse_oid_hex(line.buf, &oid, &p) || *p) {
863 warning(_("invalid line while parsing alternate refs: %s"),
864 line.buf);
865 break;
866 }
867
868 cb(&oid, data);
869 }
870
871 fclose(fh);
872 finish_command(&cmd);
873 strbuf_release(&line);
874 }
875
876 struct alternate_refs_data {
877 alternate_ref_fn *fn;
878 void *data;
879 };
880
881 static int refs_from_alternate_cb(struct object_directory *e,
882 void *data)
883 {
884 struct strbuf path = STRBUF_INIT;
885 size_t base_len;
886 struct alternate_refs_data *cb = data;
887
888 if (!strbuf_realpath(&path, e->path, 0))
889 goto out;
890 if (!strbuf_strip_suffix(&path, "/objects"))
891 goto out;
892 base_len = path.len;
893
894 /* Is this a git repository with refs? */
895 strbuf_addstr(&path, "/refs");
896 if (!is_directory(path.buf))
897 goto out;
898 strbuf_setlen(&path, base_len);
899
900 read_alternate_refs(path.buf, cb->fn, cb->data);
901
902 out:
903 strbuf_release(&path);
904 return 0;
905 }
906
907 void for_each_alternate_ref(alternate_ref_fn fn, void *data)
908 {
909 struct alternate_refs_data cb;
910 cb.fn = fn;
911 cb.data = data;
912 foreach_alt_odb(refs_from_alternate_cb, &cb);
913 }
914
915 int foreach_alt_odb(alt_odb_fn fn, void *cb)
916 {
917 struct object_directory *ent;
918 int r = 0;
919
920 prepare_alt_odb(the_repository);
921 for (ent = the_repository->objects->odb->next; ent; ent = ent->next) {
922 r = fn(ent, cb);
923 if (r)
924 break;
925 }
926 return r;
927 }
928
929 void prepare_alt_odb(struct repository *r)
930 {
931 if (r->objects->loaded_alternates)
932 return;
933
934 link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
935
936 read_info_alternates(r, r->objects->odb->path, 0);
937 r->objects->loaded_alternates = 1;
938 }
939
940 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
941 static int freshen_file(const char *fn)
942 {
943 return !utime(fn, NULL);
944 }
945
946 /*
947 * All of the check_and_freshen functions return 1 if the file exists and was
948 * freshened (if freshening was requested), 0 otherwise. If they return
949 * 0, you should not assume that it is safe to skip a write of the object (it
950 * either does not exist on disk, or has a stale mtime and may be subject to
951 * pruning).
952 */
953 int check_and_freshen_file(const char *fn, int freshen)
954 {
955 if (access(fn, F_OK))
956 return 0;
957 if (freshen && !freshen_file(fn))
958 return 0;
959 return 1;
960 }
961
962 static int check_and_freshen_odb(struct object_directory *odb,
963 const struct object_id *oid,
964 int freshen)
965 {
966 static struct strbuf path = STRBUF_INIT;
967 odb_loose_path(odb, &path, oid);
968 return check_and_freshen_file(path.buf, freshen);
969 }
970
971 static int check_and_freshen_local(const struct object_id *oid, int freshen)
972 {
973 return check_and_freshen_odb(the_repository->objects->odb, oid, freshen);
974 }
975
976 static int check_and_freshen_nonlocal(const struct object_id *oid, int freshen)
977 {
978 struct object_directory *odb;
979
980 prepare_alt_odb(the_repository);
981 for (odb = the_repository->objects->odb->next; odb; odb = odb->next) {
982 if (check_and_freshen_odb(odb, oid, freshen))
983 return 1;
984 }
985 return 0;
986 }
987
988 static int check_and_freshen(const struct object_id *oid, int freshen)
989 {
990 return check_and_freshen_local(oid, freshen) ||
991 check_and_freshen_nonlocal(oid, freshen);
992 }
993
994 int has_loose_object_nonlocal(const struct object_id *oid)
995 {
996 return check_and_freshen_nonlocal(oid, 0);
997 }
998
999 static int has_loose_object(const struct object_id *oid)
1000 {
1001 return check_and_freshen(oid, 0);
1002 }
1003
1004 static void mmap_limit_check(size_t length)
1005 {
1006 static size_t limit = 0;
1007 if (!limit) {
1008 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
1009 if (!limit)
1010 limit = SIZE_MAX;
1011 }
1012 if (length > limit)
1013 die(_("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX),
1014 (uintmax_t)length, (uintmax_t)limit);
1015 }
1016
1017 void *xmmap_gently(void *start, size_t length,
1018 int prot, int flags, int fd, off_t offset)
1019 {
1020 void *ret;
1021
1022 mmap_limit_check(length);
1023 ret = mmap(start, length, prot, flags, fd, offset);
1024 if (ret == MAP_FAILED && !length)
1025 ret = NULL;
1026 return ret;
1027 }
1028
1029 const char *mmap_os_err(void)
1030 {
1031 static const char blank[] = "";
1032 #if defined(__linux__)
1033 if (errno == ENOMEM) {
1034 /* this continues an existing error message: */
1035 static const char enomem[] =
1036 ", check sys.vm.max_map_count and/or RLIMIT_DATA";
1037 return enomem;
1038 }
1039 #endif /* OS-specific bits */
1040 return blank;
1041 }
1042
1043 void *xmmap(void *start, size_t length,
1044 int prot, int flags, int fd, off_t offset)
1045 {
1046 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
1047 if (ret == MAP_FAILED)
1048 die_errno(_("mmap failed%s"), mmap_os_err());
1049 return ret;
1050 }
1051
1052 /*
1053 * With an in-core object data in "map", rehash it to make sure the
1054 * object name actually matches "oid" to detect object corruption.
1055 * With "map" == NULL, try reading the object named with "oid" using
1056 * the streaming interface and rehash it to do the same.
1057 */
1058 int check_object_signature(struct repository *r, const struct object_id *oid,
1059 void *map, unsigned long size, const char *type,
1060 struct object_id *real_oidp)
1061 {
1062 struct object_id tmp;
1063 struct object_id *real_oid = real_oidp ? real_oidp : &tmp;
1064 enum object_type obj_type;
1065 struct git_istream *st;
1066 git_hash_ctx c;
1067 char hdr[MAX_HEADER_LEN];
1068 int hdrlen;
1069
1070 if (map) {
1071 hash_object_file(r->hash_algo, map, size, type, real_oid);
1072 return !oideq(oid, real_oid) ? -1 : 0;
1073 }
1074
1075 st = open_istream(r, oid, &obj_type, &size, NULL);
1076 if (!st)
1077 return -1;
1078
1079 /* Generate the header */
1080 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(obj_type), (uintmax_t)size) + 1;
1081
1082 /* Sha1.. */
1083 r->hash_algo->init_fn(&c);
1084 r->hash_algo->update_fn(&c, hdr, hdrlen);
1085 for (;;) {
1086 char buf[1024 * 16];
1087 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1088
1089 if (readlen < 0) {
1090 close_istream(st);
1091 return -1;
1092 }
1093 if (!readlen)
1094 break;
1095 r->hash_algo->update_fn(&c, buf, readlen);
1096 }
1097 r->hash_algo->final_oid_fn(real_oid, &c);
1098 close_istream(st);
1099 return !oideq(oid, real_oid) ? -1 : 0;
1100 }
1101
1102 int git_open_cloexec(const char *name, int flags)
1103 {
1104 int fd;
1105 static int o_cloexec = O_CLOEXEC;
1106
1107 fd = open(name, flags | o_cloexec);
1108 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1109 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1110 o_cloexec &= ~O_CLOEXEC;
1111 fd = open(name, flags | o_cloexec);
1112 }
1113
1114 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1115 {
1116 static int fd_cloexec = FD_CLOEXEC;
1117
1118 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1119 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1120 int flags = fcntl(fd, F_GETFD);
1121 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1122 fd_cloexec = 0;
1123 }
1124 }
1125 #endif
1126 return fd;
1127 }
1128
1129 /*
1130 * Find "oid" as a loose object in the local repository or in an alternate.
1131 * Returns 0 on success, negative on failure.
1132 *
1133 * The "path" out-parameter will give the path of the object we found (if any).
1134 * Note that it may point to static storage and is only valid until another
1135 * call to stat_loose_object().
1136 */
1137 static int stat_loose_object(struct repository *r, const struct object_id *oid,
1138 struct stat *st, const char **path)
1139 {
1140 struct object_directory *odb;
1141 static struct strbuf buf = STRBUF_INIT;
1142
1143 prepare_alt_odb(r);
1144 for (odb = r->objects->odb; odb; odb = odb->next) {
1145 *path = odb_loose_path(odb, &buf, oid);
1146 if (!lstat(*path, st))
1147 return 0;
1148 }
1149
1150 return -1;
1151 }
1152
1153 /*
1154 * Like stat_loose_object(), but actually open the object and return the
1155 * descriptor. See the caveats on the "path" parameter above.
1156 */
1157 static int open_loose_object(struct repository *r,
1158 const struct object_id *oid, const char **path)
1159 {
1160 int fd;
1161 struct object_directory *odb;
1162 int most_interesting_errno = ENOENT;
1163 static struct strbuf buf = STRBUF_INIT;
1164
1165 prepare_alt_odb(r);
1166 for (odb = r->objects->odb; odb; odb = odb->next) {
1167 *path = odb_loose_path(odb, &buf, oid);
1168 fd = git_open(*path);
1169 if (fd >= 0)
1170 return fd;
1171
1172 if (most_interesting_errno == ENOENT)
1173 most_interesting_errno = errno;
1174 }
1175 errno = most_interesting_errno;
1176 return -1;
1177 }
1178
1179 static int quick_has_loose(struct repository *r,
1180 const struct object_id *oid)
1181 {
1182 struct object_directory *odb;
1183
1184 prepare_alt_odb(r);
1185 for (odb = r->objects->odb; odb; odb = odb->next) {
1186 if (oidtree_contains(odb_loose_cache(odb, oid), oid))
1187 return 1;
1188 }
1189 return 0;
1190 }
1191
1192 /*
1193 * Map the loose object at "path" if it is not NULL, or the path found by
1194 * searching for a loose object named "oid".
1195 */
1196 static void *map_loose_object_1(struct repository *r, const char *path,
1197 const struct object_id *oid, unsigned long *size)
1198 {
1199 void *map;
1200 int fd;
1201
1202 if (path)
1203 fd = git_open(path);
1204 else
1205 fd = open_loose_object(r, oid, &path);
1206 map = NULL;
1207 if (fd >= 0) {
1208 struct stat st;
1209
1210 if (!fstat(fd, &st)) {
1211 *size = xsize_t(st.st_size);
1212 if (!*size) {
1213 /* mmap() is forbidden on empty files */
1214 error(_("object file %s is empty"), path);
1215 close(fd);
1216 return NULL;
1217 }
1218 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1219 }
1220 close(fd);
1221 }
1222 return map;
1223 }
1224
1225 void *map_loose_object(struct repository *r,
1226 const struct object_id *oid,
1227 unsigned long *size)
1228 {
1229 return map_loose_object_1(r, NULL, oid, size);
1230 }
1231
1232 enum unpack_loose_header_result unpack_loose_header(git_zstream *stream,
1233 unsigned char *map,
1234 unsigned long mapsize,
1235 void *buffer,
1236 unsigned long bufsiz,
1237 struct strbuf *header)
1238 {
1239 int status;
1240
1241 /* Get the data stream */
1242 memset(stream, 0, sizeof(*stream));
1243 stream->next_in = map;
1244 stream->avail_in = mapsize;
1245 stream->next_out = buffer;
1246 stream->avail_out = bufsiz;
1247
1248 git_inflate_init(stream);
1249 obj_read_unlock();
1250 status = git_inflate(stream, 0);
1251 obj_read_lock();
1252 if (status < Z_OK)
1253 return ULHR_BAD;
1254
1255 /*
1256 * Check if entire header is unpacked in the first iteration.
1257 */
1258 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1259 return ULHR_OK;
1260
1261 /*
1262 * We have a header longer than MAX_HEADER_LEN. The "header"
1263 * here is only non-NULL when we run "cat-file
1264 * --allow-unknown-type".
1265 */
1266 if (!header)
1267 return ULHR_TOO_LONG;
1268
1269 /*
1270 * buffer[0..bufsiz] was not large enough. Copy the partial
1271 * result out to header, and then append the result of further
1272 * reading the stream.
1273 */
1274 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1275 stream->next_out = buffer;
1276 stream->avail_out = bufsiz;
1277
1278 do {
1279 obj_read_unlock();
1280 status = git_inflate(stream, 0);
1281 obj_read_lock();
1282 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1283 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1284 return 0;
1285 stream->next_out = buffer;
1286 stream->avail_out = bufsiz;
1287 } while (status != Z_STREAM_END);
1288 return ULHR_TOO_LONG;
1289 }
1290
1291 static void *unpack_loose_rest(git_zstream *stream,
1292 void *buffer, unsigned long size,
1293 const struct object_id *oid)
1294 {
1295 int bytes = strlen(buffer) + 1;
1296 unsigned char *buf = xmallocz(size);
1297 unsigned long n;
1298 int status = Z_OK;
1299
1300 n = stream->total_out - bytes;
1301 if (n > size)
1302 n = size;
1303 memcpy(buf, (char *) buffer + bytes, n);
1304 bytes = n;
1305 if (bytes <= size) {
1306 /*
1307 * The above condition must be (bytes <= size), not
1308 * (bytes < size). In other words, even though we
1309 * expect no more output and set avail_out to zero,
1310 * the input zlib stream may have bytes that express
1311 * "this concludes the stream", and we *do* want to
1312 * eat that input.
1313 *
1314 * Otherwise we would not be able to test that we
1315 * consumed all the input to reach the expected size;
1316 * we also want to check that zlib tells us that all
1317 * went well with status == Z_STREAM_END at the end.
1318 */
1319 stream->next_out = buf + bytes;
1320 stream->avail_out = size - bytes;
1321 while (status == Z_OK) {
1322 obj_read_unlock();
1323 status = git_inflate(stream, Z_FINISH);
1324 obj_read_lock();
1325 }
1326 }
1327 if (status == Z_STREAM_END && !stream->avail_in) {
1328 git_inflate_end(stream);
1329 return buf;
1330 }
1331
1332 if (status < 0)
1333 error(_("corrupt loose object '%s'"), oid_to_hex(oid));
1334 else if (stream->avail_in)
1335 error(_("garbage at end of loose object '%s'"),
1336 oid_to_hex(oid));
1337 free(buf);
1338 return NULL;
1339 }
1340
1341 /*
1342 * We used to just use "sscanf()", but that's actually way
1343 * too permissive for what we want to check. So do an anal
1344 * object header parse by hand.
1345 */
1346 int parse_loose_header(const char *hdr, struct object_info *oi)
1347 {
1348 const char *type_buf = hdr;
1349 size_t size;
1350 int type, type_len = 0;
1351
1352 /*
1353 * The type can be of any size but is followed by
1354 * a space.
1355 */
1356 for (;;) {
1357 char c = *hdr++;
1358 if (!c)
1359 return -1;
1360 if (c == ' ')
1361 break;
1362 type_len++;
1363 }
1364
1365 type = type_from_string_gently(type_buf, type_len, 1);
1366 if (oi->type_name)
1367 strbuf_add(oi->type_name, type_buf, type_len);
1368 if (oi->typep)
1369 *oi->typep = type;
1370
1371 /*
1372 * The length must follow immediately, and be in canonical
1373 * decimal format (ie "010" is not valid).
1374 */
1375 size = *hdr++ - '0';
1376 if (size > 9)
1377 return -1;
1378 if (size) {
1379 for (;;) {
1380 unsigned long c = *hdr - '0';
1381 if (c > 9)
1382 break;
1383 hdr++;
1384 size = st_add(st_mult(size, 10), c);
1385 }
1386 }
1387
1388 if (oi->sizep)
1389 *oi->sizep = cast_size_t_to_ulong(size);
1390
1391 /*
1392 * The length must be followed by a zero byte
1393 */
1394 if (*hdr)
1395 return -1;
1396
1397 /*
1398 * The format is valid, but the type may still be bogus. The
1399 * Caller needs to check its oi->typep.
1400 */
1401 return 0;
1402 }
1403
1404 static int loose_object_info(struct repository *r,
1405 const struct object_id *oid,
1406 struct object_info *oi, int flags)
1407 {
1408 int status = 0;
1409 unsigned long mapsize;
1410 void *map;
1411 git_zstream stream;
1412 char hdr[MAX_HEADER_LEN];
1413 struct strbuf hdrbuf = STRBUF_INIT;
1414 unsigned long size_scratch;
1415 enum object_type type_scratch;
1416 int allow_unknown = flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE;
1417
1418 if (oi->delta_base_oid)
1419 oidclr(oi->delta_base_oid);
1420
1421 /*
1422 * If we don't care about type or size, then we don't
1423 * need to look inside the object at all. Note that we
1424 * do not optimize out the stat call, even if the
1425 * caller doesn't care about the disk-size, since our
1426 * return value implicitly indicates whether the
1427 * object even exists.
1428 */
1429 if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1430 const char *path;
1431 struct stat st;
1432 if (!oi->disk_sizep && (flags & OBJECT_INFO_QUICK))
1433 return quick_has_loose(r, oid) ? 0 : -1;
1434 if (stat_loose_object(r, oid, &st, &path) < 0)
1435 return -1;
1436 if (oi->disk_sizep)
1437 *oi->disk_sizep = st.st_size;
1438 return 0;
1439 }
1440
1441 map = map_loose_object(r, oid, &mapsize);
1442 if (!map)
1443 return -1;
1444
1445 if (!oi->sizep)
1446 oi->sizep = &size_scratch;
1447 if (!oi->typep)
1448 oi->typep = &type_scratch;
1449
1450 if (oi->disk_sizep)
1451 *oi->disk_sizep = mapsize;
1452
1453 switch (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr),
1454 allow_unknown ? &hdrbuf : NULL)) {
1455 case ULHR_OK:
1456 if (parse_loose_header(hdrbuf.len ? hdrbuf.buf : hdr, oi) < 0)
1457 status = error(_("unable to parse %s header"), oid_to_hex(oid));
1458 else if (!allow_unknown && *oi->typep < 0)
1459 die(_("invalid object type"));
1460
1461 if (!oi->contentp)
1462 break;
1463 *oi->contentp = unpack_loose_rest(&stream, hdr, *oi->sizep, oid);
1464 if (*oi->contentp)
1465 goto cleanup;
1466
1467 status = -1;
1468 break;
1469 case ULHR_BAD:
1470 status = error(_("unable to unpack %s header"),
1471 oid_to_hex(oid));
1472 break;
1473 case ULHR_TOO_LONG:
1474 status = error(_("header for %s too long, exceeds %d bytes"),
1475 oid_to_hex(oid), MAX_HEADER_LEN);
1476 break;
1477 }
1478
1479 git_inflate_end(&stream);
1480 cleanup:
1481 munmap(map, mapsize);
1482 if (oi->sizep == &size_scratch)
1483 oi->sizep = NULL;
1484 strbuf_release(&hdrbuf);
1485 if (oi->typep == &type_scratch)
1486 oi->typep = NULL;
1487 oi->whence = OI_LOOSE;
1488 return status;
1489 }
1490
1491 int obj_read_use_lock = 0;
1492 pthread_mutex_t obj_read_mutex;
1493
1494 void enable_obj_read_lock(void)
1495 {
1496 if (obj_read_use_lock)
1497 return;
1498
1499 obj_read_use_lock = 1;
1500 init_recursive_mutex(&obj_read_mutex);
1501 }
1502
1503 void disable_obj_read_lock(void)
1504 {
1505 if (!obj_read_use_lock)
1506 return;
1507
1508 obj_read_use_lock = 0;
1509 pthread_mutex_destroy(&obj_read_mutex);
1510 }
1511
1512 int fetch_if_missing = 1;
1513
1514 static int do_oid_object_info_extended(struct repository *r,
1515 const struct object_id *oid,
1516 struct object_info *oi, unsigned flags)
1517 {
1518 static struct object_info blank_oi = OBJECT_INFO_INIT;
1519 struct cached_object *co;
1520 struct pack_entry e;
1521 int rtype;
1522 const struct object_id *real = oid;
1523 int already_retried = 0;
1524
1525
1526 if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1527 real = lookup_replace_object(r, oid);
1528
1529 if (is_null_oid(real))
1530 return -1;
1531
1532 if (!oi)
1533 oi = &blank_oi;
1534
1535 co = find_cached_object(real);
1536 if (co) {
1537 if (oi->typep)
1538 *(oi->typep) = co->type;
1539 if (oi->sizep)
1540 *(oi->sizep) = co->size;
1541 if (oi->disk_sizep)
1542 *(oi->disk_sizep) = 0;
1543 if (oi->delta_base_oid)
1544 oidclr(oi->delta_base_oid);
1545 if (oi->type_name)
1546 strbuf_addstr(oi->type_name, type_name(co->type));
1547 if (oi->contentp)
1548 *oi->contentp = xmemdupz(co->buf, co->size);
1549 oi->whence = OI_CACHED;
1550 return 0;
1551 }
1552
1553 while (1) {
1554 if (find_pack_entry(r, real, &e))
1555 break;
1556
1557 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1558 return -1;
1559
1560 /* Most likely it's a loose object. */
1561 if (!loose_object_info(r, real, oi, flags))
1562 return 0;
1563
1564 /* Not a loose object; someone else may have just packed it. */
1565 if (!(flags & OBJECT_INFO_QUICK)) {
1566 reprepare_packed_git(r);
1567 if (find_pack_entry(r, real, &e))
1568 break;
1569 }
1570
1571 /*
1572 * If r is the_repository, this might be an attempt at
1573 * accessing a submodule object as if it were in the_repository
1574 * (having called add_submodule_odb() on that submodule's ODB).
1575 * If any such ODBs exist, register them and try again.
1576 */
1577 if (r == the_repository &&
1578 register_all_submodule_odb_as_alternates())
1579 /* We added some alternates; retry */
1580 continue;
1581
1582 /* Check if it is a missing object */
1583 if (fetch_if_missing && repo_has_promisor_remote(r) &&
1584 !already_retried &&
1585 !(flags & OBJECT_INFO_SKIP_FETCH_OBJECT)) {
1586 /*
1587 * TODO Investigate checking promisor_remote_get_direct()
1588 * TODO return value and stopping on error here.
1589 */
1590 promisor_remote_get_direct(r, real, 1);
1591 already_retried = 1;
1592 continue;
1593 }
1594
1595 return -1;
1596 }
1597
1598 if (oi == &blank_oi)
1599 /*
1600 * We know that the caller doesn't actually need the
1601 * information below, so return early.
1602 */
1603 return 0;
1604 rtype = packed_object_info(r, e.p, e.offset, oi);
1605 if (rtype < 0) {
1606 mark_bad_packed_object(e.p, real);
1607 return do_oid_object_info_extended(r, real, oi, 0);
1608 } else if (oi->whence == OI_PACKED) {
1609 oi->u.packed.offset = e.offset;
1610 oi->u.packed.pack = e.p;
1611 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1612 rtype == OBJ_OFS_DELTA);
1613 }
1614
1615 return 0;
1616 }
1617
1618 int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1619 struct object_info *oi, unsigned flags)
1620 {
1621 int ret;
1622 obj_read_lock();
1623 ret = do_oid_object_info_extended(r, oid, oi, flags);
1624 obj_read_unlock();
1625 return ret;
1626 }
1627
1628
1629 /* returns enum object_type or negative */
1630 int oid_object_info(struct repository *r,
1631 const struct object_id *oid,
1632 unsigned long *sizep)
1633 {
1634 enum object_type type;
1635 struct object_info oi = OBJECT_INFO_INIT;
1636
1637 oi.typep = &type;
1638 oi.sizep = sizep;
1639 if (oid_object_info_extended(r, oid, &oi,
1640 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1641 return -1;
1642 return type;
1643 }
1644
1645 static void *read_object(struct repository *r,
1646 const struct object_id *oid, enum object_type *type,
1647 unsigned long *size)
1648 {
1649 struct object_info oi = OBJECT_INFO_INIT;
1650 void *content;
1651 oi.typep = type;
1652 oi.sizep = size;
1653 oi.contentp = &content;
1654
1655 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1656 return NULL;
1657 return content;
1658 }
1659
1660 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1661 struct object_id *oid)
1662 {
1663 struct cached_object *co;
1664
1665 hash_object_file(the_hash_algo, buf, len, type_name(type), oid);
1666 if (has_object_file_with_flags(oid, OBJECT_INFO_QUICK | OBJECT_INFO_SKIP_FETCH_OBJECT) ||
1667 find_cached_object(oid))
1668 return 0;
1669 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1670 co = &cached_objects[cached_object_nr++];
1671 co->size = len;
1672 co->type = type;
1673 co->buf = xmalloc(len);
1674 memcpy(co->buf, buf, len);
1675 oidcpy(&co->oid, oid);
1676 return 0;
1677 }
1678
1679 /*
1680 * This function dies on corrupt objects; the callers who want to
1681 * deal with them should arrange to call read_object() and give error
1682 * messages themselves.
1683 */
1684 void *read_object_file_extended(struct repository *r,
1685 const struct object_id *oid,
1686 enum object_type *type,
1687 unsigned long *size,
1688 int lookup_replace)
1689 {
1690 void *data;
1691 const struct packed_git *p;
1692 const char *path;
1693 struct stat st;
1694 const struct object_id *repl = lookup_replace ?
1695 lookup_replace_object(r, oid) : oid;
1696
1697 errno = 0;
1698 data = read_object(r, repl, type, size);
1699 if (data)
1700 return data;
1701
1702 obj_read_lock();
1703 if (errno && errno != ENOENT)
1704 die_errno(_("failed to read object %s"), oid_to_hex(oid));
1705
1706 /* die if we replaced an object with one that does not exist */
1707 if (repl != oid)
1708 die(_("replacement %s not found for %s"),
1709 oid_to_hex(repl), oid_to_hex(oid));
1710
1711 if (!stat_loose_object(r, repl, &st, &path))
1712 die(_("loose object %s (stored in %s) is corrupt"),
1713 oid_to_hex(repl), path);
1714
1715 if ((p = has_packed_and_bad(r, repl)) != NULL)
1716 die(_("packed object %s (stored in %s) is corrupt"),
1717 oid_to_hex(repl), p->pack_name);
1718 obj_read_unlock();
1719
1720 return NULL;
1721 }
1722
1723 void *read_object_with_reference(struct repository *r,
1724 const struct object_id *oid,
1725 const char *required_type_name,
1726 unsigned long *size,
1727 struct object_id *actual_oid_return)
1728 {
1729 enum object_type type, required_type;
1730 void *buffer;
1731 unsigned long isize;
1732 struct object_id actual_oid;
1733
1734 required_type = type_from_string(required_type_name);
1735 oidcpy(&actual_oid, oid);
1736 while (1) {
1737 int ref_length = -1;
1738 const char *ref_type = NULL;
1739
1740 buffer = repo_read_object_file(r, &actual_oid, &type, &isize);
1741 if (!buffer)
1742 return NULL;
1743 if (type == required_type) {
1744 *size = isize;
1745 if (actual_oid_return)
1746 oidcpy(actual_oid_return, &actual_oid);
1747 return buffer;
1748 }
1749 /* Handle references */
1750 else if (type == OBJ_COMMIT)
1751 ref_type = "tree ";
1752 else if (type == OBJ_TAG)
1753 ref_type = "object ";
1754 else {
1755 free(buffer);
1756 return NULL;
1757 }
1758 ref_length = strlen(ref_type);
1759
1760 if (ref_length + the_hash_algo->hexsz > isize ||
1761 memcmp(buffer, ref_type, ref_length) ||
1762 get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1763 free(buffer);
1764 return NULL;
1765 }
1766 free(buffer);
1767 /* Now we have the ID of the referred-to object in
1768 * actual_oid. Check again. */
1769 }
1770 }
1771
1772 static void write_object_file_prepare(const struct git_hash_algo *algo,
1773 const void *buf, unsigned long len,
1774 const char *type, struct object_id *oid,
1775 char *hdr, int *hdrlen)
1776 {
1777 git_hash_ctx c;
1778
1779 /* Generate the header */
1780 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %"PRIuMAX , type, (uintmax_t)len)+1;
1781
1782 /* Sha1.. */
1783 algo->init_fn(&c);
1784 algo->update_fn(&c, hdr, *hdrlen);
1785 algo->update_fn(&c, buf, len);
1786 algo->final_oid_fn(oid, &c);
1787 }
1788
1789 /*
1790 * Move the just written object into its final resting place.
1791 */
1792 int finalize_object_file(const char *tmpfile, const char *filename)
1793 {
1794 int ret = 0;
1795
1796 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1797 goto try_rename;
1798 else if (link(tmpfile, filename))
1799 ret = errno;
1800
1801 /*
1802 * Coda hack - coda doesn't like cross-directory links,
1803 * so we fall back to a rename, which will mean that it
1804 * won't be able to check collisions, but that's not a
1805 * big deal.
1806 *
1807 * The same holds for FAT formatted media.
1808 *
1809 * When this succeeds, we just return. We have nothing
1810 * left to unlink.
1811 */
1812 if (ret && ret != EEXIST) {
1813 try_rename:
1814 if (!rename(tmpfile, filename))
1815 goto out;
1816 ret = errno;
1817 }
1818 unlink_or_warn(tmpfile);
1819 if (ret) {
1820 if (ret != EEXIST) {
1821 return error_errno(_("unable to write file %s"), filename);
1822 }
1823 /* FIXME!!! Collision check here ? */
1824 }
1825
1826 out:
1827 if (adjust_shared_perm(filename))
1828 return error(_("unable to set permission to '%s'"), filename);
1829 return 0;
1830 }
1831
1832 static int write_buffer(int fd, const void *buf, size_t len)
1833 {
1834 if (write_in_full(fd, buf, len) < 0)
1835 return error_errno(_("file write error"));
1836 return 0;
1837 }
1838
1839 int hash_object_file(const struct git_hash_algo *algo, const void *buf,
1840 unsigned long len, const char *type,
1841 struct object_id *oid)
1842 {
1843 char hdr[MAX_HEADER_LEN];
1844 int hdrlen = sizeof(hdr);
1845 write_object_file_prepare(algo, buf, len, type, oid, hdr, &hdrlen);
1846 return 0;
1847 }
1848
1849 /* Finalize a file on disk, and close it. */
1850 static void close_loose_object(int fd)
1851 {
1852 if (!the_repository->objects->odb->will_destroy) {
1853 if (fsync_object_files)
1854 fsync_or_die(fd, "loose object file");
1855 }
1856
1857 if (close(fd) != 0)
1858 die_errno(_("error when closing loose object file"));
1859 }
1860
1861 /* Size of directory component, including the ending '/' */
1862 static inline int directory_size(const char *filename)
1863 {
1864 const char *s = strrchr(filename, '/');
1865 if (!s)
1866 return 0;
1867 return s - filename + 1;
1868 }
1869
1870 /*
1871 * This creates a temporary file in the same directory as the final
1872 * 'filename'
1873 *
1874 * We want to avoid cross-directory filename renames, because those
1875 * can have problems on various filesystems (FAT, NFS, Coda).
1876 */
1877 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1878 {
1879 int fd, dirlen = directory_size(filename);
1880
1881 strbuf_reset(tmp);
1882 strbuf_add(tmp, filename, dirlen);
1883 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1884 fd = git_mkstemp_mode(tmp->buf, 0444);
1885 if (fd < 0 && dirlen && errno == ENOENT) {
1886 /*
1887 * Make sure the directory exists; note that the contents
1888 * of the buffer are undefined after mkstemp returns an
1889 * error, so we have to rewrite the whole buffer from
1890 * scratch.
1891 */
1892 strbuf_reset(tmp);
1893 strbuf_add(tmp, filename, dirlen - 1);
1894 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1895 return -1;
1896 if (adjust_shared_perm(tmp->buf))
1897 return -1;
1898
1899 /* Try again */
1900 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1901 fd = git_mkstemp_mode(tmp->buf, 0444);
1902 }
1903 return fd;
1904 }
1905
1906 static int write_loose_object(const struct object_id *oid, char *hdr,
1907 int hdrlen, const void *buf, unsigned long len,
1908 time_t mtime, unsigned flags)
1909 {
1910 int fd, ret;
1911 unsigned char compressed[4096];
1912 git_zstream stream;
1913 git_hash_ctx c;
1914 struct object_id parano_oid;
1915 static struct strbuf tmp_file = STRBUF_INIT;
1916 static struct strbuf filename = STRBUF_INIT;
1917
1918 loose_object_path(the_repository, &filename, oid);
1919
1920 fd = create_tmpfile(&tmp_file, filename.buf);
1921 if (fd < 0) {
1922 if (flags & HASH_SILENT)
1923 return -1;
1924 else if (errno == EACCES)
1925 return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
1926 else
1927 return error_errno(_("unable to create temporary file"));
1928 }
1929
1930 /* Set it up */
1931 git_deflate_init(&stream, zlib_compression_level);
1932 stream.next_out = compressed;
1933 stream.avail_out = sizeof(compressed);
1934 the_hash_algo->init_fn(&c);
1935
1936 /* First header.. */
1937 stream.next_in = (unsigned char *)hdr;
1938 stream.avail_in = hdrlen;
1939 while (git_deflate(&stream, 0) == Z_OK)
1940 ; /* nothing */
1941 the_hash_algo->update_fn(&c, hdr, hdrlen);
1942
1943 /* Then the data itself.. */
1944 stream.next_in = (void *)buf;
1945 stream.avail_in = len;
1946 do {
1947 unsigned char *in0 = stream.next_in;
1948 ret = git_deflate(&stream, Z_FINISH);
1949 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1950 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1951 die(_("unable to write loose object file"));
1952 stream.next_out = compressed;
1953 stream.avail_out = sizeof(compressed);
1954 } while (ret == Z_OK);
1955
1956 if (ret != Z_STREAM_END)
1957 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
1958 ret);
1959 ret = git_deflate_end_gently(&stream);
1960 if (ret != Z_OK)
1961 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
1962 ret);
1963 the_hash_algo->final_oid_fn(&parano_oid, &c);
1964 if (!oideq(oid, &parano_oid))
1965 die(_("confused by unstable object source data for %s"),
1966 oid_to_hex(oid));
1967
1968 close_loose_object(fd);
1969
1970 if (mtime) {
1971 struct utimbuf utb;
1972 utb.actime = mtime;
1973 utb.modtime = mtime;
1974 if (utime(tmp_file.buf, &utb) < 0 &&
1975 !(flags & HASH_SILENT))
1976 warning_errno(_("failed utime() on %s"), tmp_file.buf);
1977 }
1978
1979 return finalize_object_file(tmp_file.buf, filename.buf);
1980 }
1981
1982 static int freshen_loose_object(const struct object_id *oid)
1983 {
1984 return check_and_freshen(oid, 1);
1985 }
1986
1987 static int freshen_packed_object(const struct object_id *oid)
1988 {
1989 struct pack_entry e;
1990 if (!find_pack_entry(the_repository, oid, &e))
1991 return 0;
1992 if (e.p->freshened)
1993 return 1;
1994 if (!freshen_file(e.p->pack_name))
1995 return 0;
1996 e.p->freshened = 1;
1997 return 1;
1998 }
1999
2000 int write_object_file_flags(const void *buf, unsigned long len,
2001 const char *type, struct object_id *oid,
2002 unsigned flags)
2003 {
2004 char hdr[MAX_HEADER_LEN];
2005 int hdrlen = sizeof(hdr);
2006
2007 /* Normally if we have it in the pack then we do not bother writing
2008 * it out into .git/objects/??/?{38} file.
2009 */
2010 write_object_file_prepare(the_hash_algo, buf, len, type, oid, hdr,
2011 &hdrlen);
2012 if (freshen_packed_object(oid) || freshen_loose_object(oid))
2013 return 0;
2014 return write_loose_object(oid, hdr, hdrlen, buf, len, 0, flags);
2015 }
2016
2017 int hash_object_file_literally(const void *buf, unsigned long len,
2018 const char *type, struct object_id *oid,
2019 unsigned flags)
2020 {
2021 char *header;
2022 int hdrlen, status = 0;
2023
2024 /* type string, SP, %lu of the length plus NUL must fit this */
2025 hdrlen = strlen(type) + MAX_HEADER_LEN;
2026 header = xmalloc(hdrlen);
2027 write_object_file_prepare(the_hash_algo, buf, len, type, oid, header,
2028 &hdrlen);
2029
2030 if (!(flags & HASH_WRITE_OBJECT))
2031 goto cleanup;
2032 if (freshen_packed_object(oid) || freshen_loose_object(oid))
2033 goto cleanup;
2034 status = write_loose_object(oid, header, hdrlen, buf, len, 0, 0);
2035
2036 cleanup:
2037 free(header);
2038 return status;
2039 }
2040
2041 int force_object_loose(const struct object_id *oid, time_t mtime)
2042 {
2043 void *buf;
2044 unsigned long len;
2045 enum object_type type;
2046 char hdr[MAX_HEADER_LEN];
2047 int hdrlen;
2048 int ret;
2049
2050 if (has_loose_object(oid))
2051 return 0;
2052 buf = read_object(the_repository, oid, &type, &len);
2053 if (!buf)
2054 return error(_("cannot read object for %s"), oid_to_hex(oid));
2055 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(type), (uintmax_t)len) + 1;
2056 ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime, 0);
2057 free(buf);
2058
2059 return ret;
2060 }
2061
2062 int has_object(struct repository *r, const struct object_id *oid,
2063 unsigned flags)
2064 {
2065 int quick = !(flags & HAS_OBJECT_RECHECK_PACKED);
2066 unsigned object_info_flags = OBJECT_INFO_SKIP_FETCH_OBJECT |
2067 (quick ? OBJECT_INFO_QUICK : 0);
2068
2069 if (!startup_info->have_repository)
2070 return 0;
2071 return oid_object_info_extended(r, oid, NULL, object_info_flags) >= 0;
2072 }
2073
2074 int repo_has_object_file_with_flags(struct repository *r,
2075 const struct object_id *oid, int flags)
2076 {
2077 if (!startup_info->have_repository)
2078 return 0;
2079 return oid_object_info_extended(r, oid, NULL, flags) >= 0;
2080 }
2081
2082 int repo_has_object_file(struct repository *r,
2083 const struct object_id *oid)
2084 {
2085 return repo_has_object_file_with_flags(r, oid, 0);
2086 }
2087
2088 static void check_tree(const void *buf, size_t size)
2089 {
2090 struct tree_desc desc;
2091 struct name_entry entry;
2092
2093 init_tree_desc(&desc, buf, size);
2094 while (tree_entry(&desc, &entry))
2095 /* do nothing
2096 * tree_entry() will die() on malformed entries */
2097 ;
2098 }
2099
2100 static void check_commit(const void *buf, size_t size)
2101 {
2102 struct commit c;
2103 memset(&c, 0, sizeof(c));
2104 if (parse_commit_buffer(the_repository, &c, buf, size, 0))
2105 die(_("corrupt commit"));
2106 }
2107
2108 static void check_tag(const void *buf, size_t size)
2109 {
2110 struct tag t;
2111 memset(&t, 0, sizeof(t));
2112 if (parse_tag_buffer(the_repository, &t, buf, size))
2113 die(_("corrupt tag"));
2114 }
2115
2116 static int index_mem(struct index_state *istate,
2117 struct object_id *oid, void *buf, size_t size,
2118 enum object_type type,
2119 const char *path, unsigned flags)
2120 {
2121 int ret, re_allocated = 0;
2122 int write_object = flags & HASH_WRITE_OBJECT;
2123
2124 if (!type)
2125 type = OBJ_BLOB;
2126
2127 /*
2128 * Convert blobs to git internal format
2129 */
2130 if ((type == OBJ_BLOB) && path) {
2131 struct strbuf nbuf = STRBUF_INIT;
2132 if (convert_to_git(istate, path, buf, size, &nbuf,
2133 get_conv_flags(flags))) {
2134 buf = strbuf_detach(&nbuf, &size);
2135 re_allocated = 1;
2136 }
2137 }
2138 if (flags & HASH_FORMAT_CHECK) {
2139 if (type == OBJ_TREE)
2140 check_tree(buf, size);
2141 if (type == OBJ_COMMIT)
2142 check_commit(buf, size);
2143 if (type == OBJ_TAG)
2144 check_tag(buf, size);
2145 }
2146
2147 if (write_object)
2148 ret = write_object_file(buf, size, type_name(type), oid);
2149 else
2150 ret = hash_object_file(the_hash_algo, buf, size,
2151 type_name(type), oid);
2152 if (re_allocated)
2153 free(buf);
2154 return ret;
2155 }
2156
2157 static int index_stream_convert_blob(struct index_state *istate,
2158 struct object_id *oid,
2159 int fd,
2160 const char *path,
2161 unsigned flags)
2162 {
2163 int ret;
2164 const int write_object = flags & HASH_WRITE_OBJECT;
2165 struct strbuf sbuf = STRBUF_INIT;
2166
2167 assert(path);
2168 assert(would_convert_to_git_filter_fd(istate, path));
2169
2170 convert_to_git_filter_fd(istate, path, fd, &sbuf,
2171 get_conv_flags(flags));
2172
2173 if (write_object)
2174 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
2175 oid);
2176 else
2177 ret = hash_object_file(the_hash_algo, sbuf.buf, sbuf.len,
2178 type_name(OBJ_BLOB), oid);
2179 strbuf_release(&sbuf);
2180 return ret;
2181 }
2182
2183 static int index_pipe(struct index_state *istate, struct object_id *oid,
2184 int fd, enum object_type type,
2185 const char *path, unsigned flags)
2186 {
2187 struct strbuf sbuf = STRBUF_INIT;
2188 int ret;
2189
2190 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2191 ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags);
2192 else
2193 ret = -1;
2194 strbuf_release(&sbuf);
2195 return ret;
2196 }
2197
2198 #define SMALL_FILE_SIZE (32*1024)
2199
2200 static int index_core(struct index_state *istate,
2201 struct object_id *oid, int fd, size_t size,
2202 enum object_type type, const char *path,
2203 unsigned flags)
2204 {
2205 int ret;
2206
2207 if (!size) {
2208 ret = index_mem(istate, oid, "", size, type, path, flags);
2209 } else if (size <= SMALL_FILE_SIZE) {
2210 char *buf = xmalloc(size);
2211 ssize_t read_result = read_in_full(fd, buf, size);
2212 if (read_result < 0)
2213 ret = error_errno(_("read error while indexing %s"),
2214 path ? path : "<unknown>");
2215 else if (read_result != size)
2216 ret = error(_("short read while indexing %s"),
2217 path ? path : "<unknown>");
2218 else
2219 ret = index_mem(istate, oid, buf, size, type, path, flags);
2220 free(buf);
2221 } else {
2222 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2223 ret = index_mem(istate, oid, buf, size, type, path, flags);
2224 munmap(buf, size);
2225 }
2226 return ret;
2227 }
2228
2229 /*
2230 * This creates one packfile per large blob unless bulk-checkin
2231 * machinery is "plugged".
2232 *
2233 * This also bypasses the usual "convert-to-git" dance, and that is on
2234 * purpose. We could write a streaming version of the converting
2235 * functions and insert that before feeding the data to fast-import
2236 * (or equivalent in-core API described above). However, that is
2237 * somewhat complicated, as we do not know the size of the filter
2238 * result, which we need to know beforehand when writing a git object.
2239 * Since the primary motivation for trying to stream from the working
2240 * tree file and to avoid mmaping it in core is to deal with large
2241 * binary blobs, they generally do not want to get any conversion, and
2242 * callers should avoid this code path when filters are requested.
2243 */
2244 static int index_stream(struct object_id *oid, int fd, size_t size,
2245 enum object_type type, const char *path,
2246 unsigned flags)
2247 {
2248 return index_bulk_checkin(oid, fd, size, type, path, flags);
2249 }
2250
2251 int index_fd(struct index_state *istate, struct object_id *oid,
2252 int fd, struct stat *st,
2253 enum object_type type, const char *path, unsigned flags)
2254 {
2255 int ret;
2256
2257 /*
2258 * Call xsize_t() only when needed to avoid potentially unnecessary
2259 * die() for large files.
2260 */
2261 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, path))
2262 ret = index_stream_convert_blob(istate, oid, fd, path, flags);
2263 else if (!S_ISREG(st->st_mode))
2264 ret = index_pipe(istate, oid, fd, type, path, flags);
2265 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
2266 (path && would_convert_to_git(istate, path)))
2267 ret = index_core(istate, oid, fd, xsize_t(st->st_size),
2268 type, path, flags);
2269 else
2270 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
2271 flags);
2272 close(fd);
2273 return ret;
2274 }
2275
2276 int index_path(struct index_state *istate, struct object_id *oid,
2277 const char *path, struct stat *st, unsigned flags)
2278 {
2279 int fd;
2280 struct strbuf sb = STRBUF_INIT;
2281 int rc = 0;
2282
2283 switch (st->st_mode & S_IFMT) {
2284 case S_IFREG:
2285 fd = open(path, O_RDONLY);
2286 if (fd < 0)
2287 return error_errno("open(\"%s\")", path);
2288 if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0)
2289 return error(_("%s: failed to insert into database"),
2290 path);
2291 break;
2292 case S_IFLNK:
2293 if (strbuf_readlink(&sb, path, st->st_size))
2294 return error_errno("readlink(\"%s\")", path);
2295 if (!(flags & HASH_WRITE_OBJECT))
2296 hash_object_file(the_hash_algo, sb.buf, sb.len,
2297 blob_type, oid);
2298 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
2299 rc = error(_("%s: failed to insert into database"), path);
2300 strbuf_release(&sb);
2301 break;
2302 case S_IFDIR:
2303 return resolve_gitlink_ref(path, "HEAD", oid);
2304 default:
2305 return error(_("%s: unsupported file type"), path);
2306 }
2307 return rc;
2308 }
2309
2310 int read_pack_header(int fd, struct pack_header *header)
2311 {
2312 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
2313 /* "eof before pack header was fully read" */
2314 return PH_ERROR_EOF;
2315
2316 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2317 /* "protocol error (pack signature mismatch detected)" */
2318 return PH_ERROR_PACK_SIGNATURE;
2319 if (!pack_version_ok(header->hdr_version))
2320 /* "protocol error (pack version unsupported)" */
2321 return PH_ERROR_PROTOCOL;
2322 return 0;
2323 }
2324
2325 void assert_oid_type(const struct object_id *oid, enum object_type expect)
2326 {
2327 enum object_type type = oid_object_info(the_repository, oid, NULL);
2328 if (type < 0)
2329 die(_("%s is not a valid object"), oid_to_hex(oid));
2330 if (type != expect)
2331 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
2332 type_name(expect));
2333 }
2334
2335 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2336 struct strbuf *path,
2337 each_loose_object_fn obj_cb,
2338 each_loose_cruft_fn cruft_cb,
2339 each_loose_subdir_fn subdir_cb,
2340 void *data)
2341 {
2342 size_t origlen, baselen;
2343 DIR *dir;
2344 struct dirent *de;
2345 int r = 0;
2346 struct object_id oid;
2347
2348 if (subdir_nr > 0xff)
2349 BUG("invalid loose object subdirectory: %x", subdir_nr);
2350
2351 origlen = path->len;
2352 strbuf_complete(path, '/');
2353 strbuf_addf(path, "%02x", subdir_nr);
2354
2355 dir = opendir(path->buf);
2356 if (!dir) {
2357 if (errno != ENOENT)
2358 r = error_errno(_("unable to open %s"), path->buf);
2359 strbuf_setlen(path, origlen);
2360 return r;
2361 }
2362
2363 oid.hash[0] = subdir_nr;
2364 strbuf_addch(path, '/');
2365 baselen = path->len;
2366
2367 while ((de = readdir_skip_dot_and_dotdot(dir))) {
2368 size_t namelen;
2369
2370 namelen = strlen(de->d_name);
2371 strbuf_setlen(path, baselen);
2372 strbuf_add(path, de->d_name, namelen);
2373 if (namelen == the_hash_algo->hexsz - 2 &&
2374 !hex_to_bytes(oid.hash + 1, de->d_name,
2375 the_hash_algo->rawsz - 1)) {
2376 oid_set_algo(&oid, the_hash_algo);
2377 if (obj_cb) {
2378 r = obj_cb(&oid, path->buf, data);
2379 if (r)
2380 break;
2381 }
2382 continue;
2383 }
2384
2385 if (cruft_cb) {
2386 r = cruft_cb(de->d_name, path->buf, data);
2387 if (r)
2388 break;
2389 }
2390 }
2391 closedir(dir);
2392
2393 strbuf_setlen(path, baselen - 1);
2394 if (!r && subdir_cb)
2395 r = subdir_cb(subdir_nr, path->buf, data);
2396
2397 strbuf_setlen(path, origlen);
2398
2399 return r;
2400 }
2401
2402 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2403 each_loose_object_fn obj_cb,
2404 each_loose_cruft_fn cruft_cb,
2405 each_loose_subdir_fn subdir_cb,
2406 void *data)
2407 {
2408 int r = 0;
2409 int i;
2410
2411 for (i = 0; i < 256; i++) {
2412 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2413 subdir_cb, data);
2414 if (r)
2415 break;
2416 }
2417
2418 return r;
2419 }
2420
2421 int for_each_loose_file_in_objdir(const char *path,
2422 each_loose_object_fn obj_cb,
2423 each_loose_cruft_fn cruft_cb,
2424 each_loose_subdir_fn subdir_cb,
2425 void *data)
2426 {
2427 struct strbuf buf = STRBUF_INIT;
2428 int r;
2429
2430 strbuf_addstr(&buf, path);
2431 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2432 subdir_cb, data);
2433 strbuf_release(&buf);
2434
2435 return r;
2436 }
2437
2438 int for_each_loose_object(each_loose_object_fn cb, void *data,
2439 enum for_each_object_flags flags)
2440 {
2441 struct object_directory *odb;
2442
2443 prepare_alt_odb(the_repository);
2444 for (odb = the_repository->objects->odb; odb; odb = odb->next) {
2445 int r = for_each_loose_file_in_objdir(odb->path, cb, NULL,
2446 NULL, data);
2447 if (r)
2448 return r;
2449
2450 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2451 break;
2452 }
2453
2454 return 0;
2455 }
2456
2457 static int append_loose_object(const struct object_id *oid, const char *path,
2458 void *data)
2459 {
2460 oidtree_insert(data, oid);
2461 return 0;
2462 }
2463
2464 struct oidtree *odb_loose_cache(struct object_directory *odb,
2465 const struct object_id *oid)
2466 {
2467 int subdir_nr = oid->hash[0];
2468 struct strbuf buf = STRBUF_INIT;
2469 size_t word_bits = bitsizeof(odb->loose_objects_subdir_seen[0]);
2470 size_t word_index = subdir_nr / word_bits;
2471 size_t mask = (size_t)1u << (subdir_nr % word_bits);
2472 uint32_t *bitmap;
2473
2474 if (subdir_nr < 0 ||
2475 subdir_nr >= bitsizeof(odb->loose_objects_subdir_seen))
2476 BUG("subdir_nr out of range");
2477
2478 bitmap = &odb->loose_objects_subdir_seen[word_index];
2479 if (*bitmap & mask)
2480 return odb->loose_objects_cache;
2481 if (!odb->loose_objects_cache) {
2482 ALLOC_ARRAY(odb->loose_objects_cache, 1);
2483 oidtree_init(odb->loose_objects_cache);
2484 }
2485 strbuf_addstr(&buf, odb->path);
2486 for_each_file_in_obj_subdir(subdir_nr, &buf,
2487 append_loose_object,
2488 NULL, NULL,
2489 odb->loose_objects_cache);
2490 *bitmap |= mask;
2491 strbuf_release(&buf);
2492 return odb->loose_objects_cache;
2493 }
2494
2495 void odb_clear_loose_cache(struct object_directory *odb)
2496 {
2497 oidtree_clear(odb->loose_objects_cache);
2498 FREE_AND_NULL(odb->loose_objects_cache);
2499 memset(&odb->loose_objects_subdir_seen, 0,
2500 sizeof(odb->loose_objects_subdir_seen));
2501 }
2502
2503 static int check_stream_oid(git_zstream *stream,
2504 const char *hdr,
2505 unsigned long size,
2506 const char *path,
2507 const struct object_id *expected_oid)
2508 {
2509 git_hash_ctx c;
2510 struct object_id real_oid;
2511 unsigned char buf[4096];
2512 unsigned long total_read;
2513 int status = Z_OK;
2514
2515 the_hash_algo->init_fn(&c);
2516 the_hash_algo->update_fn(&c, hdr, stream->total_out);
2517
2518 /*
2519 * We already read some bytes into hdr, but the ones up to the NUL
2520 * do not count against the object's content size.
2521 */
2522 total_read = stream->total_out - strlen(hdr) - 1;
2523
2524 /*
2525 * This size comparison must be "<=" to read the final zlib packets;
2526 * see the comment in unpack_loose_rest for details.
2527 */
2528 while (total_read <= size &&
2529 (status == Z_OK ||
2530 (status == Z_BUF_ERROR && !stream->avail_out))) {
2531 stream->next_out = buf;
2532 stream->avail_out = sizeof(buf);
2533 if (size - total_read < stream->avail_out)
2534 stream->avail_out = size - total_read;
2535 status = git_inflate(stream, Z_FINISH);
2536 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2537 total_read += stream->next_out - buf;
2538 }
2539 git_inflate_end(stream);
2540
2541 if (status != Z_STREAM_END) {
2542 error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid));
2543 return -1;
2544 }
2545 if (stream->avail_in) {
2546 error(_("garbage at end of loose object '%s'"),
2547 oid_to_hex(expected_oid));
2548 return -1;
2549 }
2550
2551 the_hash_algo->final_oid_fn(&real_oid, &c);
2552 if (!oideq(expected_oid, &real_oid)) {
2553 error(_("hash mismatch for %s (expected %s)"), path,
2554 oid_to_hex(expected_oid));
2555 return -1;
2556 }
2557
2558 return 0;
2559 }
2560
2561 int read_loose_object(const char *path,
2562 const struct object_id *expected_oid,
2563 struct object_id *real_oid,
2564 void **contents,
2565 struct object_info *oi)
2566 {
2567 int ret = -1;
2568 void *map = NULL;
2569 unsigned long mapsize;
2570 git_zstream stream;
2571 char hdr[MAX_HEADER_LEN];
2572 unsigned long *size = oi->sizep;
2573
2574 map = map_loose_object_1(the_repository, path, NULL, &mapsize);
2575 if (!map) {
2576 error_errno(_("unable to mmap %s"), path);
2577 goto out;
2578 }
2579
2580 if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr),
2581 NULL) < 0) {
2582 error(_("unable to unpack header of %s"), path);
2583 goto out;
2584 }
2585
2586 if (parse_loose_header(hdr, oi) < 0) {
2587 error(_("unable to parse header of %s"), path);
2588 git_inflate_end(&stream);
2589 goto out;
2590 }
2591
2592 if (*oi->typep == OBJ_BLOB && *size > big_file_threshold) {
2593 if (check_stream_oid(&stream, hdr, *size, path, expected_oid) < 0)
2594 goto out;
2595 } else {
2596 *contents = unpack_loose_rest(&stream, hdr, *size, expected_oid);
2597 if (!*contents) {
2598 error(_("unable to unpack contents of %s"), path);
2599 git_inflate_end(&stream);
2600 goto out;
2601 }
2602 if (check_object_signature(the_repository, expected_oid,
2603 *contents, *size,
2604 oi->type_name->buf, real_oid))
2605 goto out;
2606 }
2607
2608 ret = 0; /* everything checks out */
2609
2610 out:
2611 if (map)
2612 munmap(map, mapsize);
2613 return ret;
2614 }