]> git.ipfire.org Git - thirdparty/git.git/blob - object-file.c
refs: print error message in debug output
[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 /*
684 * Compute the exact path an alternate is at and returns it. In case of
685 * error NULL is returned and the human readable error is added to `err`
686 * `path` may be relative and should point to $GIT_DIR.
687 * `err` must not be null.
688 */
689 char *compute_alternate_path(const char *path, struct strbuf *err)
690 {
691 char *ref_git = NULL;
692 const char *repo;
693 int seen_error = 0;
694
695 ref_git = real_pathdup(path, 0);
696 if (!ref_git) {
697 seen_error = 1;
698 strbuf_addf(err, _("path '%s' does not exist"), path);
699 goto out;
700 }
701
702 repo = read_gitfile(ref_git);
703 if (!repo)
704 repo = read_gitfile(mkpath("%s/.git", ref_git));
705 if (repo) {
706 free(ref_git);
707 ref_git = xstrdup(repo);
708 }
709
710 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
711 char *ref_git_git = mkpathdup("%s/.git", ref_git);
712 free(ref_git);
713 ref_git = ref_git_git;
714 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
715 struct strbuf sb = STRBUF_INIT;
716 seen_error = 1;
717 if (get_common_dir(&sb, ref_git)) {
718 strbuf_addf(err,
719 _("reference repository '%s' as a linked "
720 "checkout is not supported yet."),
721 path);
722 goto out;
723 }
724
725 strbuf_addf(err, _("reference repository '%s' is not a "
726 "local repository."), path);
727 goto out;
728 }
729
730 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
731 strbuf_addf(err, _("reference repository '%s' is shallow"),
732 path);
733 seen_error = 1;
734 goto out;
735 }
736
737 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
738 strbuf_addf(err,
739 _("reference repository '%s' is grafted"),
740 path);
741 seen_error = 1;
742 goto out;
743 }
744
745 out:
746 if (seen_error) {
747 FREE_AND_NULL(ref_git);
748 }
749
750 return ref_git;
751 }
752
753 struct object_directory *find_odb(struct repository *r, const char *obj_dir)
754 {
755 struct object_directory *odb;
756 char *obj_dir_real = real_pathdup(obj_dir, 1);
757 struct strbuf odb_path_real = STRBUF_INIT;
758
759 prepare_alt_odb(r);
760 for (odb = r->objects->odb; odb; odb = odb->next) {
761 strbuf_realpath(&odb_path_real, odb->path, 1);
762 if (!strcmp(obj_dir_real, odb_path_real.buf))
763 break;
764 }
765
766 free(obj_dir_real);
767 strbuf_release(&odb_path_real);
768
769 if (!odb)
770 die(_("could not find object directory matching %s"), obj_dir);
771 return odb;
772 }
773
774 static void fill_alternate_refs_command(struct child_process *cmd,
775 const char *repo_path)
776 {
777 const char *value;
778
779 if (!git_config_get_value("core.alternateRefsCommand", &value)) {
780 cmd->use_shell = 1;
781
782 strvec_push(&cmd->args, value);
783 strvec_push(&cmd->args, repo_path);
784 } else {
785 cmd->git_cmd = 1;
786
787 strvec_pushf(&cmd->args, "--git-dir=%s", repo_path);
788 strvec_push(&cmd->args, "for-each-ref");
789 strvec_push(&cmd->args, "--format=%(objectname)");
790
791 if (!git_config_get_value("core.alternateRefsPrefixes", &value)) {
792 strvec_push(&cmd->args, "--");
793 strvec_split(&cmd->args, value);
794 }
795 }
796
797 strvec_pushv(&cmd->env_array, (const char **)local_repo_env);
798 cmd->out = -1;
799 }
800
801 static void read_alternate_refs(const char *path,
802 alternate_ref_fn *cb,
803 void *data)
804 {
805 struct child_process cmd = CHILD_PROCESS_INIT;
806 struct strbuf line = STRBUF_INIT;
807 FILE *fh;
808
809 fill_alternate_refs_command(&cmd, path);
810
811 if (start_command(&cmd))
812 return;
813
814 fh = xfdopen(cmd.out, "r");
815 while (strbuf_getline_lf(&line, fh) != EOF) {
816 struct object_id oid;
817 const char *p;
818
819 if (parse_oid_hex(line.buf, &oid, &p) || *p) {
820 warning(_("invalid line while parsing alternate refs: %s"),
821 line.buf);
822 break;
823 }
824
825 cb(&oid, data);
826 }
827
828 fclose(fh);
829 finish_command(&cmd);
830 strbuf_release(&line);
831 }
832
833 struct alternate_refs_data {
834 alternate_ref_fn *fn;
835 void *data;
836 };
837
838 static int refs_from_alternate_cb(struct object_directory *e,
839 void *data)
840 {
841 struct strbuf path = STRBUF_INIT;
842 size_t base_len;
843 struct alternate_refs_data *cb = data;
844
845 if (!strbuf_realpath(&path, e->path, 0))
846 goto out;
847 if (!strbuf_strip_suffix(&path, "/objects"))
848 goto out;
849 base_len = path.len;
850
851 /* Is this a git repository with refs? */
852 strbuf_addstr(&path, "/refs");
853 if (!is_directory(path.buf))
854 goto out;
855 strbuf_setlen(&path, base_len);
856
857 read_alternate_refs(path.buf, cb->fn, cb->data);
858
859 out:
860 strbuf_release(&path);
861 return 0;
862 }
863
864 void for_each_alternate_ref(alternate_ref_fn fn, void *data)
865 {
866 struct alternate_refs_data cb;
867 cb.fn = fn;
868 cb.data = data;
869 foreach_alt_odb(refs_from_alternate_cb, &cb);
870 }
871
872 int foreach_alt_odb(alt_odb_fn fn, void *cb)
873 {
874 struct object_directory *ent;
875 int r = 0;
876
877 prepare_alt_odb(the_repository);
878 for (ent = the_repository->objects->odb->next; ent; ent = ent->next) {
879 r = fn(ent, cb);
880 if (r)
881 break;
882 }
883 return r;
884 }
885
886 void prepare_alt_odb(struct repository *r)
887 {
888 if (r->objects->loaded_alternates)
889 return;
890
891 link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
892
893 read_info_alternates(r, r->objects->odb->path, 0);
894 r->objects->loaded_alternates = 1;
895 }
896
897 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
898 static int freshen_file(const char *fn)
899 {
900 return !utime(fn, NULL);
901 }
902
903 /*
904 * All of the check_and_freshen functions return 1 if the file exists and was
905 * freshened (if freshening was requested), 0 otherwise. If they return
906 * 0, you should not assume that it is safe to skip a write of the object (it
907 * either does not exist on disk, or has a stale mtime and may be subject to
908 * pruning).
909 */
910 int check_and_freshen_file(const char *fn, int freshen)
911 {
912 if (access(fn, F_OK))
913 return 0;
914 if (freshen && !freshen_file(fn))
915 return 0;
916 return 1;
917 }
918
919 static int check_and_freshen_odb(struct object_directory *odb,
920 const struct object_id *oid,
921 int freshen)
922 {
923 static struct strbuf path = STRBUF_INIT;
924 odb_loose_path(odb, &path, oid);
925 return check_and_freshen_file(path.buf, freshen);
926 }
927
928 static int check_and_freshen_local(const struct object_id *oid, int freshen)
929 {
930 return check_and_freshen_odb(the_repository->objects->odb, oid, freshen);
931 }
932
933 static int check_and_freshen_nonlocal(const struct object_id *oid, int freshen)
934 {
935 struct object_directory *odb;
936
937 prepare_alt_odb(the_repository);
938 for (odb = the_repository->objects->odb->next; odb; odb = odb->next) {
939 if (check_and_freshen_odb(odb, oid, freshen))
940 return 1;
941 }
942 return 0;
943 }
944
945 static int check_and_freshen(const struct object_id *oid, int freshen)
946 {
947 return check_and_freshen_local(oid, freshen) ||
948 check_and_freshen_nonlocal(oid, freshen);
949 }
950
951 int has_loose_object_nonlocal(const struct object_id *oid)
952 {
953 return check_and_freshen_nonlocal(oid, 0);
954 }
955
956 static int has_loose_object(const struct object_id *oid)
957 {
958 return check_and_freshen(oid, 0);
959 }
960
961 static void mmap_limit_check(size_t length)
962 {
963 static size_t limit = 0;
964 if (!limit) {
965 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
966 if (!limit)
967 limit = SIZE_MAX;
968 }
969 if (length > limit)
970 die(_("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX),
971 (uintmax_t)length, (uintmax_t)limit);
972 }
973
974 void *xmmap_gently(void *start, size_t length,
975 int prot, int flags, int fd, off_t offset)
976 {
977 void *ret;
978
979 mmap_limit_check(length);
980 ret = mmap(start, length, prot, flags, fd, offset);
981 if (ret == MAP_FAILED && !length)
982 ret = NULL;
983 return ret;
984 }
985
986 const char *mmap_os_err(void)
987 {
988 static const char blank[] = "";
989 #if defined(__linux__)
990 if (errno == ENOMEM) {
991 /* this continues an existing error message: */
992 static const char enomem[] =
993 ", check sys.vm.max_map_count and/or RLIMIT_DATA";
994 return enomem;
995 }
996 #endif /* OS-specific bits */
997 return blank;
998 }
999
1000 void *xmmap(void *start, size_t length,
1001 int prot, int flags, int fd, off_t offset)
1002 {
1003 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
1004 if (ret == MAP_FAILED)
1005 die_errno(_("mmap failed%s"), mmap_os_err());
1006 return ret;
1007 }
1008
1009 /*
1010 * With an in-core object data in "map", rehash it to make sure the
1011 * object name actually matches "oid" to detect object corruption.
1012 * With "map" == NULL, try reading the object named with "oid" using
1013 * the streaming interface and rehash it to do the same.
1014 */
1015 int check_object_signature(struct repository *r, const struct object_id *oid,
1016 void *map, unsigned long size, const char *type,
1017 struct object_id *real_oidp)
1018 {
1019 struct object_id tmp;
1020 struct object_id *real_oid = real_oidp ? real_oidp : &tmp;
1021 enum object_type obj_type;
1022 struct git_istream *st;
1023 git_hash_ctx c;
1024 char hdr[MAX_HEADER_LEN];
1025 int hdrlen;
1026
1027 if (map) {
1028 hash_object_file(r->hash_algo, map, size, type, real_oid);
1029 return !oideq(oid, real_oid) ? -1 : 0;
1030 }
1031
1032 st = open_istream(r, oid, &obj_type, &size, NULL);
1033 if (!st)
1034 return -1;
1035
1036 /* Generate the header */
1037 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(obj_type), (uintmax_t)size) + 1;
1038
1039 /* Sha1.. */
1040 r->hash_algo->init_fn(&c);
1041 r->hash_algo->update_fn(&c, hdr, hdrlen);
1042 for (;;) {
1043 char buf[1024 * 16];
1044 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1045
1046 if (readlen < 0) {
1047 close_istream(st);
1048 return -1;
1049 }
1050 if (!readlen)
1051 break;
1052 r->hash_algo->update_fn(&c, buf, readlen);
1053 }
1054 r->hash_algo->final_oid_fn(real_oid, &c);
1055 close_istream(st);
1056 return !oideq(oid, real_oid) ? -1 : 0;
1057 }
1058
1059 int git_open_cloexec(const char *name, int flags)
1060 {
1061 int fd;
1062 static int o_cloexec = O_CLOEXEC;
1063
1064 fd = open(name, flags | o_cloexec);
1065 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1066 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1067 o_cloexec &= ~O_CLOEXEC;
1068 fd = open(name, flags | o_cloexec);
1069 }
1070
1071 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1072 {
1073 static int fd_cloexec = FD_CLOEXEC;
1074
1075 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1076 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1077 int flags = fcntl(fd, F_GETFD);
1078 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1079 fd_cloexec = 0;
1080 }
1081 }
1082 #endif
1083 return fd;
1084 }
1085
1086 /*
1087 * Find "oid" as a loose object in the local repository or in an alternate.
1088 * Returns 0 on success, negative on failure.
1089 *
1090 * The "path" out-parameter will give the path of the object we found (if any).
1091 * Note that it may point to static storage and is only valid until another
1092 * call to stat_loose_object().
1093 */
1094 static int stat_loose_object(struct repository *r, const struct object_id *oid,
1095 struct stat *st, const char **path)
1096 {
1097 struct object_directory *odb;
1098 static struct strbuf buf = STRBUF_INIT;
1099
1100 prepare_alt_odb(r);
1101 for (odb = r->objects->odb; odb; odb = odb->next) {
1102 *path = odb_loose_path(odb, &buf, oid);
1103 if (!lstat(*path, st))
1104 return 0;
1105 }
1106
1107 return -1;
1108 }
1109
1110 /*
1111 * Like stat_loose_object(), but actually open the object and return the
1112 * descriptor. See the caveats on the "path" parameter above.
1113 */
1114 static int open_loose_object(struct repository *r,
1115 const struct object_id *oid, const char **path)
1116 {
1117 int fd;
1118 struct object_directory *odb;
1119 int most_interesting_errno = ENOENT;
1120 static struct strbuf buf = STRBUF_INIT;
1121
1122 prepare_alt_odb(r);
1123 for (odb = r->objects->odb; odb; odb = odb->next) {
1124 *path = odb_loose_path(odb, &buf, oid);
1125 fd = git_open(*path);
1126 if (fd >= 0)
1127 return fd;
1128
1129 if (most_interesting_errno == ENOENT)
1130 most_interesting_errno = errno;
1131 }
1132 errno = most_interesting_errno;
1133 return -1;
1134 }
1135
1136 static int quick_has_loose(struct repository *r,
1137 const struct object_id *oid)
1138 {
1139 struct object_directory *odb;
1140
1141 prepare_alt_odb(r);
1142 for (odb = r->objects->odb; odb; odb = odb->next) {
1143 if (oidtree_contains(odb_loose_cache(odb, oid), oid))
1144 return 1;
1145 }
1146 return 0;
1147 }
1148
1149 /*
1150 * Map the loose object at "path" if it is not NULL, or the path found by
1151 * searching for a loose object named "oid".
1152 */
1153 static void *map_loose_object_1(struct repository *r, const char *path,
1154 const struct object_id *oid, unsigned long *size)
1155 {
1156 void *map;
1157 int fd;
1158
1159 if (path)
1160 fd = git_open(path);
1161 else
1162 fd = open_loose_object(r, oid, &path);
1163 map = NULL;
1164 if (fd >= 0) {
1165 struct stat st;
1166
1167 if (!fstat(fd, &st)) {
1168 *size = xsize_t(st.st_size);
1169 if (!*size) {
1170 /* mmap() is forbidden on empty files */
1171 error(_("object file %s is empty"), path);
1172 close(fd);
1173 return NULL;
1174 }
1175 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1176 }
1177 close(fd);
1178 }
1179 return map;
1180 }
1181
1182 void *map_loose_object(struct repository *r,
1183 const struct object_id *oid,
1184 unsigned long *size)
1185 {
1186 return map_loose_object_1(r, NULL, oid, size);
1187 }
1188
1189 enum unpack_loose_header_result unpack_loose_header(git_zstream *stream,
1190 unsigned char *map,
1191 unsigned long mapsize,
1192 void *buffer,
1193 unsigned long bufsiz,
1194 struct strbuf *header)
1195 {
1196 int status;
1197
1198 /* Get the data stream */
1199 memset(stream, 0, sizeof(*stream));
1200 stream->next_in = map;
1201 stream->avail_in = mapsize;
1202 stream->next_out = buffer;
1203 stream->avail_out = bufsiz;
1204
1205 git_inflate_init(stream);
1206 obj_read_unlock();
1207 status = git_inflate(stream, 0);
1208 obj_read_lock();
1209 if (status < Z_OK)
1210 return ULHR_BAD;
1211
1212 /*
1213 * Check if entire header is unpacked in the first iteration.
1214 */
1215 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1216 return ULHR_OK;
1217
1218 /*
1219 * We have a header longer than MAX_HEADER_LEN. The "header"
1220 * here is only non-NULL when we run "cat-file
1221 * --allow-unknown-type".
1222 */
1223 if (!header)
1224 return ULHR_TOO_LONG;
1225
1226 /*
1227 * buffer[0..bufsiz] was not large enough. Copy the partial
1228 * result out to header, and then append the result of further
1229 * reading the stream.
1230 */
1231 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1232 stream->next_out = buffer;
1233 stream->avail_out = bufsiz;
1234
1235 do {
1236 obj_read_unlock();
1237 status = git_inflate(stream, 0);
1238 obj_read_lock();
1239 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1240 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1241 return 0;
1242 stream->next_out = buffer;
1243 stream->avail_out = bufsiz;
1244 } while (status != Z_STREAM_END);
1245 return ULHR_TOO_LONG;
1246 }
1247
1248 static void *unpack_loose_rest(git_zstream *stream,
1249 void *buffer, unsigned long size,
1250 const struct object_id *oid)
1251 {
1252 int bytes = strlen(buffer) + 1;
1253 unsigned char *buf = xmallocz(size);
1254 unsigned long n;
1255 int status = Z_OK;
1256
1257 n = stream->total_out - bytes;
1258 if (n > size)
1259 n = size;
1260 memcpy(buf, (char *) buffer + bytes, n);
1261 bytes = n;
1262 if (bytes <= size) {
1263 /*
1264 * The above condition must be (bytes <= size), not
1265 * (bytes < size). In other words, even though we
1266 * expect no more output and set avail_out to zero,
1267 * the input zlib stream may have bytes that express
1268 * "this concludes the stream", and we *do* want to
1269 * eat that input.
1270 *
1271 * Otherwise we would not be able to test that we
1272 * consumed all the input to reach the expected size;
1273 * we also want to check that zlib tells us that all
1274 * went well with status == Z_STREAM_END at the end.
1275 */
1276 stream->next_out = buf + bytes;
1277 stream->avail_out = size - bytes;
1278 while (status == Z_OK) {
1279 obj_read_unlock();
1280 status = git_inflate(stream, Z_FINISH);
1281 obj_read_lock();
1282 }
1283 }
1284 if (status == Z_STREAM_END && !stream->avail_in) {
1285 git_inflate_end(stream);
1286 return buf;
1287 }
1288
1289 if (status < 0)
1290 error(_("corrupt loose object '%s'"), oid_to_hex(oid));
1291 else if (stream->avail_in)
1292 error(_("garbage at end of loose object '%s'"),
1293 oid_to_hex(oid));
1294 free(buf);
1295 return NULL;
1296 }
1297
1298 /*
1299 * We used to just use "sscanf()", but that's actually way
1300 * too permissive for what we want to check. So do an anal
1301 * object header parse by hand.
1302 */
1303 int parse_loose_header(const char *hdr, struct object_info *oi)
1304 {
1305 const char *type_buf = hdr;
1306 size_t size;
1307 int type, type_len = 0;
1308
1309 /*
1310 * The type can be of any size but is followed by
1311 * a space.
1312 */
1313 for (;;) {
1314 char c = *hdr++;
1315 if (!c)
1316 return -1;
1317 if (c == ' ')
1318 break;
1319 type_len++;
1320 }
1321
1322 type = type_from_string_gently(type_buf, type_len, 1);
1323 if (oi->type_name)
1324 strbuf_add(oi->type_name, type_buf, type_len);
1325 if (oi->typep)
1326 *oi->typep = type;
1327
1328 /*
1329 * The length must follow immediately, and be in canonical
1330 * decimal format (ie "010" is not valid).
1331 */
1332 size = *hdr++ - '0';
1333 if (size > 9)
1334 return -1;
1335 if (size) {
1336 for (;;) {
1337 unsigned long c = *hdr - '0';
1338 if (c > 9)
1339 break;
1340 hdr++;
1341 size = st_add(st_mult(size, 10), c);
1342 }
1343 }
1344
1345 if (oi->sizep)
1346 *oi->sizep = cast_size_t_to_ulong(size);
1347
1348 /*
1349 * The length must be followed by a zero byte
1350 */
1351 if (*hdr)
1352 return -1;
1353
1354 /*
1355 * The format is valid, but the type may still be bogus. The
1356 * Caller needs to check its oi->typep.
1357 */
1358 return 0;
1359 }
1360
1361 static int loose_object_info(struct repository *r,
1362 const struct object_id *oid,
1363 struct object_info *oi, int flags)
1364 {
1365 int status = 0;
1366 unsigned long mapsize;
1367 void *map;
1368 git_zstream stream;
1369 char hdr[MAX_HEADER_LEN];
1370 struct strbuf hdrbuf = STRBUF_INIT;
1371 unsigned long size_scratch;
1372 enum object_type type_scratch;
1373 int allow_unknown = flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE;
1374
1375 if (oi->delta_base_oid)
1376 oidclr(oi->delta_base_oid);
1377
1378 /*
1379 * If we don't care about type or size, then we don't
1380 * need to look inside the object at all. Note that we
1381 * do not optimize out the stat call, even if the
1382 * caller doesn't care about the disk-size, since our
1383 * return value implicitly indicates whether the
1384 * object even exists.
1385 */
1386 if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1387 const char *path;
1388 struct stat st;
1389 if (!oi->disk_sizep && (flags & OBJECT_INFO_QUICK))
1390 return quick_has_loose(r, oid) ? 0 : -1;
1391 if (stat_loose_object(r, oid, &st, &path) < 0)
1392 return -1;
1393 if (oi->disk_sizep)
1394 *oi->disk_sizep = st.st_size;
1395 return 0;
1396 }
1397
1398 map = map_loose_object(r, oid, &mapsize);
1399 if (!map)
1400 return -1;
1401
1402 if (!oi->sizep)
1403 oi->sizep = &size_scratch;
1404 if (!oi->typep)
1405 oi->typep = &type_scratch;
1406
1407 if (oi->disk_sizep)
1408 *oi->disk_sizep = mapsize;
1409
1410 switch (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr),
1411 allow_unknown ? &hdrbuf : NULL)) {
1412 case ULHR_OK:
1413 if (parse_loose_header(hdrbuf.len ? hdrbuf.buf : hdr, oi) < 0)
1414 status = error(_("unable to parse %s header"), oid_to_hex(oid));
1415 else if (!allow_unknown && *oi->typep < 0)
1416 die(_("invalid object type"));
1417
1418 if (!oi->contentp)
1419 break;
1420 *oi->contentp = unpack_loose_rest(&stream, hdr, *oi->sizep, oid);
1421 if (*oi->contentp)
1422 goto cleanup;
1423
1424 status = -1;
1425 break;
1426 case ULHR_BAD:
1427 status = error(_("unable to unpack %s header"),
1428 oid_to_hex(oid));
1429 break;
1430 case ULHR_TOO_LONG:
1431 status = error(_("header for %s too long, exceeds %d bytes"),
1432 oid_to_hex(oid), MAX_HEADER_LEN);
1433 break;
1434 }
1435
1436 git_inflate_end(&stream);
1437 cleanup:
1438 munmap(map, mapsize);
1439 if (oi->sizep == &size_scratch)
1440 oi->sizep = NULL;
1441 strbuf_release(&hdrbuf);
1442 if (oi->typep == &type_scratch)
1443 oi->typep = NULL;
1444 oi->whence = OI_LOOSE;
1445 return status;
1446 }
1447
1448 int obj_read_use_lock = 0;
1449 pthread_mutex_t obj_read_mutex;
1450
1451 void enable_obj_read_lock(void)
1452 {
1453 if (obj_read_use_lock)
1454 return;
1455
1456 obj_read_use_lock = 1;
1457 init_recursive_mutex(&obj_read_mutex);
1458 }
1459
1460 void disable_obj_read_lock(void)
1461 {
1462 if (!obj_read_use_lock)
1463 return;
1464
1465 obj_read_use_lock = 0;
1466 pthread_mutex_destroy(&obj_read_mutex);
1467 }
1468
1469 int fetch_if_missing = 1;
1470
1471 static int do_oid_object_info_extended(struct repository *r,
1472 const struct object_id *oid,
1473 struct object_info *oi, unsigned flags)
1474 {
1475 static struct object_info blank_oi = OBJECT_INFO_INIT;
1476 struct cached_object *co;
1477 struct pack_entry e;
1478 int rtype;
1479 const struct object_id *real = oid;
1480 int already_retried = 0;
1481
1482
1483 if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1484 real = lookup_replace_object(r, oid);
1485
1486 if (is_null_oid(real))
1487 return -1;
1488
1489 if (!oi)
1490 oi = &blank_oi;
1491
1492 co = find_cached_object(real);
1493 if (co) {
1494 if (oi->typep)
1495 *(oi->typep) = co->type;
1496 if (oi->sizep)
1497 *(oi->sizep) = co->size;
1498 if (oi->disk_sizep)
1499 *(oi->disk_sizep) = 0;
1500 if (oi->delta_base_oid)
1501 oidclr(oi->delta_base_oid);
1502 if (oi->type_name)
1503 strbuf_addstr(oi->type_name, type_name(co->type));
1504 if (oi->contentp)
1505 *oi->contentp = xmemdupz(co->buf, co->size);
1506 oi->whence = OI_CACHED;
1507 return 0;
1508 }
1509
1510 while (1) {
1511 if (find_pack_entry(r, real, &e))
1512 break;
1513
1514 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1515 return -1;
1516
1517 /* Most likely it's a loose object. */
1518 if (!loose_object_info(r, real, oi, flags))
1519 return 0;
1520
1521 /* Not a loose object; someone else may have just packed it. */
1522 if (!(flags & OBJECT_INFO_QUICK)) {
1523 reprepare_packed_git(r);
1524 if (find_pack_entry(r, real, &e))
1525 break;
1526 }
1527
1528 /*
1529 * If r is the_repository, this might be an attempt at
1530 * accessing a submodule object as if it were in the_repository
1531 * (having called add_submodule_odb() on that submodule's ODB).
1532 * If any such ODBs exist, register them and try again.
1533 */
1534 if (r == the_repository &&
1535 register_all_submodule_odb_as_alternates())
1536 /* We added some alternates; retry */
1537 continue;
1538
1539 /* Check if it is a missing object */
1540 if (fetch_if_missing && repo_has_promisor_remote(r) &&
1541 !already_retried &&
1542 !(flags & OBJECT_INFO_SKIP_FETCH_OBJECT)) {
1543 /*
1544 * TODO Investigate checking promisor_remote_get_direct()
1545 * TODO return value and stopping on error here.
1546 */
1547 promisor_remote_get_direct(r, real, 1);
1548 already_retried = 1;
1549 continue;
1550 }
1551
1552 return -1;
1553 }
1554
1555 if (oi == &blank_oi)
1556 /*
1557 * We know that the caller doesn't actually need the
1558 * information below, so return early.
1559 */
1560 return 0;
1561 rtype = packed_object_info(r, e.p, e.offset, oi);
1562 if (rtype < 0) {
1563 mark_bad_packed_object(e.p, real);
1564 return do_oid_object_info_extended(r, real, oi, 0);
1565 } else if (oi->whence == OI_PACKED) {
1566 oi->u.packed.offset = e.offset;
1567 oi->u.packed.pack = e.p;
1568 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1569 rtype == OBJ_OFS_DELTA);
1570 }
1571
1572 return 0;
1573 }
1574
1575 int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1576 struct object_info *oi, unsigned flags)
1577 {
1578 int ret;
1579 obj_read_lock();
1580 ret = do_oid_object_info_extended(r, oid, oi, flags);
1581 obj_read_unlock();
1582 return ret;
1583 }
1584
1585
1586 /* returns enum object_type or negative */
1587 int oid_object_info(struct repository *r,
1588 const struct object_id *oid,
1589 unsigned long *sizep)
1590 {
1591 enum object_type type;
1592 struct object_info oi = OBJECT_INFO_INIT;
1593
1594 oi.typep = &type;
1595 oi.sizep = sizep;
1596 if (oid_object_info_extended(r, oid, &oi,
1597 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1598 return -1;
1599 return type;
1600 }
1601
1602 static void *read_object(struct repository *r,
1603 const struct object_id *oid, enum object_type *type,
1604 unsigned long *size)
1605 {
1606 struct object_info oi = OBJECT_INFO_INIT;
1607 void *content;
1608 oi.typep = type;
1609 oi.sizep = size;
1610 oi.contentp = &content;
1611
1612 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1613 return NULL;
1614 return content;
1615 }
1616
1617 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1618 struct object_id *oid)
1619 {
1620 struct cached_object *co;
1621
1622 hash_object_file(the_hash_algo, buf, len, type_name(type), oid);
1623 if (has_object_file_with_flags(oid, OBJECT_INFO_QUICK | OBJECT_INFO_SKIP_FETCH_OBJECT) ||
1624 find_cached_object(oid))
1625 return 0;
1626 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1627 co = &cached_objects[cached_object_nr++];
1628 co->size = len;
1629 co->type = type;
1630 co->buf = xmalloc(len);
1631 memcpy(co->buf, buf, len);
1632 oidcpy(&co->oid, oid);
1633 return 0;
1634 }
1635
1636 /*
1637 * This function dies on corrupt objects; the callers who want to
1638 * deal with them should arrange to call read_object() and give error
1639 * messages themselves.
1640 */
1641 void *read_object_file_extended(struct repository *r,
1642 const struct object_id *oid,
1643 enum object_type *type,
1644 unsigned long *size,
1645 int lookup_replace)
1646 {
1647 void *data;
1648 const struct packed_git *p;
1649 const char *path;
1650 struct stat st;
1651 const struct object_id *repl = lookup_replace ?
1652 lookup_replace_object(r, oid) : oid;
1653
1654 errno = 0;
1655 data = read_object(r, repl, type, size);
1656 if (data)
1657 return data;
1658
1659 obj_read_lock();
1660 if (errno && errno != ENOENT)
1661 die_errno(_("failed to read object %s"), oid_to_hex(oid));
1662
1663 /* die if we replaced an object with one that does not exist */
1664 if (repl != oid)
1665 die(_("replacement %s not found for %s"),
1666 oid_to_hex(repl), oid_to_hex(oid));
1667
1668 if (!stat_loose_object(r, repl, &st, &path))
1669 die(_("loose object %s (stored in %s) is corrupt"),
1670 oid_to_hex(repl), path);
1671
1672 if ((p = has_packed_and_bad(r, repl)) != NULL)
1673 die(_("packed object %s (stored in %s) is corrupt"),
1674 oid_to_hex(repl), p->pack_name);
1675 obj_read_unlock();
1676
1677 return NULL;
1678 }
1679
1680 void *read_object_with_reference(struct repository *r,
1681 const struct object_id *oid,
1682 const char *required_type_name,
1683 unsigned long *size,
1684 struct object_id *actual_oid_return)
1685 {
1686 enum object_type type, required_type;
1687 void *buffer;
1688 unsigned long isize;
1689 struct object_id actual_oid;
1690
1691 required_type = type_from_string(required_type_name);
1692 oidcpy(&actual_oid, oid);
1693 while (1) {
1694 int ref_length = -1;
1695 const char *ref_type = NULL;
1696
1697 buffer = repo_read_object_file(r, &actual_oid, &type, &isize);
1698 if (!buffer)
1699 return NULL;
1700 if (type == required_type) {
1701 *size = isize;
1702 if (actual_oid_return)
1703 oidcpy(actual_oid_return, &actual_oid);
1704 return buffer;
1705 }
1706 /* Handle references */
1707 else if (type == OBJ_COMMIT)
1708 ref_type = "tree ";
1709 else if (type == OBJ_TAG)
1710 ref_type = "object ";
1711 else {
1712 free(buffer);
1713 return NULL;
1714 }
1715 ref_length = strlen(ref_type);
1716
1717 if (ref_length + the_hash_algo->hexsz > isize ||
1718 memcmp(buffer, ref_type, ref_length) ||
1719 get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1720 free(buffer);
1721 return NULL;
1722 }
1723 free(buffer);
1724 /* Now we have the ID of the referred-to object in
1725 * actual_oid. Check again. */
1726 }
1727 }
1728
1729 static void write_object_file_prepare(const struct git_hash_algo *algo,
1730 const void *buf, unsigned long len,
1731 const char *type, struct object_id *oid,
1732 char *hdr, int *hdrlen)
1733 {
1734 git_hash_ctx c;
1735
1736 /* Generate the header */
1737 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %"PRIuMAX , type, (uintmax_t)len)+1;
1738
1739 /* Sha1.. */
1740 algo->init_fn(&c);
1741 algo->update_fn(&c, hdr, *hdrlen);
1742 algo->update_fn(&c, buf, len);
1743 algo->final_oid_fn(oid, &c);
1744 }
1745
1746 /*
1747 * Move the just written object into its final resting place.
1748 */
1749 int finalize_object_file(const char *tmpfile, const char *filename)
1750 {
1751 int ret = 0;
1752
1753 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1754 goto try_rename;
1755 else if (link(tmpfile, filename))
1756 ret = errno;
1757
1758 /*
1759 * Coda hack - coda doesn't like cross-directory links,
1760 * so we fall back to a rename, which will mean that it
1761 * won't be able to check collisions, but that's not a
1762 * big deal.
1763 *
1764 * The same holds for FAT formatted media.
1765 *
1766 * When this succeeds, we just return. We have nothing
1767 * left to unlink.
1768 */
1769 if (ret && ret != EEXIST) {
1770 try_rename:
1771 if (!rename(tmpfile, filename))
1772 goto out;
1773 ret = errno;
1774 }
1775 unlink_or_warn(tmpfile);
1776 if (ret) {
1777 if (ret != EEXIST) {
1778 return error_errno(_("unable to write file %s"), filename);
1779 }
1780 /* FIXME!!! Collision check here ? */
1781 }
1782
1783 out:
1784 if (adjust_shared_perm(filename))
1785 return error(_("unable to set permission to '%s'"), filename);
1786 return 0;
1787 }
1788
1789 static int write_buffer(int fd, const void *buf, size_t len)
1790 {
1791 if (write_in_full(fd, buf, len) < 0)
1792 return error_errno(_("file write error"));
1793 return 0;
1794 }
1795
1796 int hash_object_file(const struct git_hash_algo *algo, const void *buf,
1797 unsigned long len, const char *type,
1798 struct object_id *oid)
1799 {
1800 char hdr[MAX_HEADER_LEN];
1801 int hdrlen = sizeof(hdr);
1802 write_object_file_prepare(algo, buf, len, type, oid, hdr, &hdrlen);
1803 return 0;
1804 }
1805
1806 /* Finalize a file on disk, and close it. */
1807 static void close_loose_object(int fd)
1808 {
1809 if (fsync_object_files)
1810 fsync_or_die(fd, "loose object file");
1811 if (close(fd) != 0)
1812 die_errno(_("error when closing loose object file"));
1813 }
1814
1815 /* Size of directory component, including the ending '/' */
1816 static inline int directory_size(const char *filename)
1817 {
1818 const char *s = strrchr(filename, '/');
1819 if (!s)
1820 return 0;
1821 return s - filename + 1;
1822 }
1823
1824 /*
1825 * This creates a temporary file in the same directory as the final
1826 * 'filename'
1827 *
1828 * We want to avoid cross-directory filename renames, because those
1829 * can have problems on various filesystems (FAT, NFS, Coda).
1830 */
1831 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1832 {
1833 int fd, dirlen = directory_size(filename);
1834
1835 strbuf_reset(tmp);
1836 strbuf_add(tmp, filename, dirlen);
1837 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1838 fd = git_mkstemp_mode(tmp->buf, 0444);
1839 if (fd < 0 && dirlen && errno == ENOENT) {
1840 /*
1841 * Make sure the directory exists; note that the contents
1842 * of the buffer are undefined after mkstemp returns an
1843 * error, so we have to rewrite the whole buffer from
1844 * scratch.
1845 */
1846 strbuf_reset(tmp);
1847 strbuf_add(tmp, filename, dirlen - 1);
1848 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1849 return -1;
1850 if (adjust_shared_perm(tmp->buf))
1851 return -1;
1852
1853 /* Try again */
1854 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1855 fd = git_mkstemp_mode(tmp->buf, 0444);
1856 }
1857 return fd;
1858 }
1859
1860 static int write_loose_object(const struct object_id *oid, char *hdr,
1861 int hdrlen, const void *buf, unsigned long len,
1862 time_t mtime, unsigned flags)
1863 {
1864 int fd, ret;
1865 unsigned char compressed[4096];
1866 git_zstream stream;
1867 git_hash_ctx c;
1868 struct object_id parano_oid;
1869 static struct strbuf tmp_file = STRBUF_INIT;
1870 static struct strbuf filename = STRBUF_INIT;
1871
1872 loose_object_path(the_repository, &filename, oid);
1873
1874 fd = create_tmpfile(&tmp_file, filename.buf);
1875 if (fd < 0) {
1876 if (flags & HASH_SILENT)
1877 return -1;
1878 else if (errno == EACCES)
1879 return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
1880 else
1881 return error_errno(_("unable to create temporary file"));
1882 }
1883
1884 /* Set it up */
1885 git_deflate_init(&stream, zlib_compression_level);
1886 stream.next_out = compressed;
1887 stream.avail_out = sizeof(compressed);
1888 the_hash_algo->init_fn(&c);
1889
1890 /* First header.. */
1891 stream.next_in = (unsigned char *)hdr;
1892 stream.avail_in = hdrlen;
1893 while (git_deflate(&stream, 0) == Z_OK)
1894 ; /* nothing */
1895 the_hash_algo->update_fn(&c, hdr, hdrlen);
1896
1897 /* Then the data itself.. */
1898 stream.next_in = (void *)buf;
1899 stream.avail_in = len;
1900 do {
1901 unsigned char *in0 = stream.next_in;
1902 ret = git_deflate(&stream, Z_FINISH);
1903 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1904 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1905 die(_("unable to write loose object file"));
1906 stream.next_out = compressed;
1907 stream.avail_out = sizeof(compressed);
1908 } while (ret == Z_OK);
1909
1910 if (ret != Z_STREAM_END)
1911 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
1912 ret);
1913 ret = git_deflate_end_gently(&stream);
1914 if (ret != Z_OK)
1915 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
1916 ret);
1917 the_hash_algo->final_oid_fn(&parano_oid, &c);
1918 if (!oideq(oid, &parano_oid))
1919 die(_("confused by unstable object source data for %s"),
1920 oid_to_hex(oid));
1921
1922 close_loose_object(fd);
1923
1924 if (mtime) {
1925 struct utimbuf utb;
1926 utb.actime = mtime;
1927 utb.modtime = mtime;
1928 if (utime(tmp_file.buf, &utb) < 0 &&
1929 !(flags & HASH_SILENT))
1930 warning_errno(_("failed utime() on %s"), tmp_file.buf);
1931 }
1932
1933 return finalize_object_file(tmp_file.buf, filename.buf);
1934 }
1935
1936 static int freshen_loose_object(const struct object_id *oid)
1937 {
1938 return check_and_freshen(oid, 1);
1939 }
1940
1941 static int freshen_packed_object(const struct object_id *oid)
1942 {
1943 struct pack_entry e;
1944 if (!find_pack_entry(the_repository, oid, &e))
1945 return 0;
1946 if (e.p->freshened)
1947 return 1;
1948 if (!freshen_file(e.p->pack_name))
1949 return 0;
1950 e.p->freshened = 1;
1951 return 1;
1952 }
1953
1954 int write_object_file_flags(const void *buf, unsigned long len,
1955 const char *type, struct object_id *oid,
1956 unsigned flags)
1957 {
1958 char hdr[MAX_HEADER_LEN];
1959 int hdrlen = sizeof(hdr);
1960
1961 /* Normally if we have it in the pack then we do not bother writing
1962 * it out into .git/objects/??/?{38} file.
1963 */
1964 write_object_file_prepare(the_hash_algo, buf, len, type, oid, hdr,
1965 &hdrlen);
1966 if (freshen_packed_object(oid) || freshen_loose_object(oid))
1967 return 0;
1968 return write_loose_object(oid, hdr, hdrlen, buf, len, 0, flags);
1969 }
1970
1971 int hash_object_file_literally(const void *buf, unsigned long len,
1972 const char *type, struct object_id *oid,
1973 unsigned flags)
1974 {
1975 char *header;
1976 int hdrlen, status = 0;
1977
1978 /* type string, SP, %lu of the length plus NUL must fit this */
1979 hdrlen = strlen(type) + MAX_HEADER_LEN;
1980 header = xmalloc(hdrlen);
1981 write_object_file_prepare(the_hash_algo, buf, len, type, oid, header,
1982 &hdrlen);
1983
1984 if (!(flags & HASH_WRITE_OBJECT))
1985 goto cleanup;
1986 if (freshen_packed_object(oid) || freshen_loose_object(oid))
1987 goto cleanup;
1988 status = write_loose_object(oid, header, hdrlen, buf, len, 0, 0);
1989
1990 cleanup:
1991 free(header);
1992 return status;
1993 }
1994
1995 int force_object_loose(const struct object_id *oid, time_t mtime)
1996 {
1997 void *buf;
1998 unsigned long len;
1999 enum object_type type;
2000 char hdr[MAX_HEADER_LEN];
2001 int hdrlen;
2002 int ret;
2003
2004 if (has_loose_object(oid))
2005 return 0;
2006 buf = read_object(the_repository, oid, &type, &len);
2007 if (!buf)
2008 return error(_("cannot read object for %s"), oid_to_hex(oid));
2009 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(type), (uintmax_t)len) + 1;
2010 ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime, 0);
2011 free(buf);
2012
2013 return ret;
2014 }
2015
2016 int has_object(struct repository *r, const struct object_id *oid,
2017 unsigned flags)
2018 {
2019 int quick = !(flags & HAS_OBJECT_RECHECK_PACKED);
2020 unsigned object_info_flags = OBJECT_INFO_SKIP_FETCH_OBJECT |
2021 (quick ? OBJECT_INFO_QUICK : 0);
2022
2023 if (!startup_info->have_repository)
2024 return 0;
2025 return oid_object_info_extended(r, oid, NULL, object_info_flags) >= 0;
2026 }
2027
2028 int repo_has_object_file_with_flags(struct repository *r,
2029 const struct object_id *oid, int flags)
2030 {
2031 if (!startup_info->have_repository)
2032 return 0;
2033 return oid_object_info_extended(r, oid, NULL, flags) >= 0;
2034 }
2035
2036 int repo_has_object_file(struct repository *r,
2037 const struct object_id *oid)
2038 {
2039 return repo_has_object_file_with_flags(r, oid, 0);
2040 }
2041
2042 static void check_tree(const void *buf, size_t size)
2043 {
2044 struct tree_desc desc;
2045 struct name_entry entry;
2046
2047 init_tree_desc(&desc, buf, size);
2048 while (tree_entry(&desc, &entry))
2049 /* do nothing
2050 * tree_entry() will die() on malformed entries */
2051 ;
2052 }
2053
2054 static void check_commit(const void *buf, size_t size)
2055 {
2056 struct commit c;
2057 memset(&c, 0, sizeof(c));
2058 if (parse_commit_buffer(the_repository, &c, buf, size, 0))
2059 die(_("corrupt commit"));
2060 }
2061
2062 static void check_tag(const void *buf, size_t size)
2063 {
2064 struct tag t;
2065 memset(&t, 0, sizeof(t));
2066 if (parse_tag_buffer(the_repository, &t, buf, size))
2067 die(_("corrupt tag"));
2068 }
2069
2070 static int index_mem(struct index_state *istate,
2071 struct object_id *oid, void *buf, size_t size,
2072 enum object_type type,
2073 const char *path, unsigned flags)
2074 {
2075 int ret, re_allocated = 0;
2076 int write_object = flags & HASH_WRITE_OBJECT;
2077
2078 if (!type)
2079 type = OBJ_BLOB;
2080
2081 /*
2082 * Convert blobs to git internal format
2083 */
2084 if ((type == OBJ_BLOB) && path) {
2085 struct strbuf nbuf = STRBUF_INIT;
2086 if (convert_to_git(istate, path, buf, size, &nbuf,
2087 get_conv_flags(flags))) {
2088 buf = strbuf_detach(&nbuf, &size);
2089 re_allocated = 1;
2090 }
2091 }
2092 if (flags & HASH_FORMAT_CHECK) {
2093 if (type == OBJ_TREE)
2094 check_tree(buf, size);
2095 if (type == OBJ_COMMIT)
2096 check_commit(buf, size);
2097 if (type == OBJ_TAG)
2098 check_tag(buf, size);
2099 }
2100
2101 if (write_object)
2102 ret = write_object_file(buf, size, type_name(type), oid);
2103 else
2104 ret = hash_object_file(the_hash_algo, buf, size,
2105 type_name(type), oid);
2106 if (re_allocated)
2107 free(buf);
2108 return ret;
2109 }
2110
2111 static int index_stream_convert_blob(struct index_state *istate,
2112 struct object_id *oid,
2113 int fd,
2114 const char *path,
2115 unsigned flags)
2116 {
2117 int ret;
2118 const int write_object = flags & HASH_WRITE_OBJECT;
2119 struct strbuf sbuf = STRBUF_INIT;
2120
2121 assert(path);
2122 assert(would_convert_to_git_filter_fd(istate, path));
2123
2124 convert_to_git_filter_fd(istate, path, fd, &sbuf,
2125 get_conv_flags(flags));
2126
2127 if (write_object)
2128 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
2129 oid);
2130 else
2131 ret = hash_object_file(the_hash_algo, sbuf.buf, sbuf.len,
2132 type_name(OBJ_BLOB), oid);
2133 strbuf_release(&sbuf);
2134 return ret;
2135 }
2136
2137 static int index_pipe(struct index_state *istate, struct object_id *oid,
2138 int fd, enum object_type type,
2139 const char *path, unsigned flags)
2140 {
2141 struct strbuf sbuf = STRBUF_INIT;
2142 int ret;
2143
2144 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2145 ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags);
2146 else
2147 ret = -1;
2148 strbuf_release(&sbuf);
2149 return ret;
2150 }
2151
2152 #define SMALL_FILE_SIZE (32*1024)
2153
2154 static int index_core(struct index_state *istate,
2155 struct object_id *oid, int fd, size_t size,
2156 enum object_type type, const char *path,
2157 unsigned flags)
2158 {
2159 int ret;
2160
2161 if (!size) {
2162 ret = index_mem(istate, oid, "", size, type, path, flags);
2163 } else if (size <= SMALL_FILE_SIZE) {
2164 char *buf = xmalloc(size);
2165 ssize_t read_result = read_in_full(fd, buf, size);
2166 if (read_result < 0)
2167 ret = error_errno(_("read error while indexing %s"),
2168 path ? path : "<unknown>");
2169 else if (read_result != size)
2170 ret = error(_("short read while indexing %s"),
2171 path ? path : "<unknown>");
2172 else
2173 ret = index_mem(istate, oid, buf, size, type, path, flags);
2174 free(buf);
2175 } else {
2176 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2177 ret = index_mem(istate, oid, buf, size, type, path, flags);
2178 munmap(buf, size);
2179 }
2180 return ret;
2181 }
2182
2183 /*
2184 * This creates one packfile per large blob unless bulk-checkin
2185 * machinery is "plugged".
2186 *
2187 * This also bypasses the usual "convert-to-git" dance, and that is on
2188 * purpose. We could write a streaming version of the converting
2189 * functions and insert that before feeding the data to fast-import
2190 * (or equivalent in-core API described above). However, that is
2191 * somewhat complicated, as we do not know the size of the filter
2192 * result, which we need to know beforehand when writing a git object.
2193 * Since the primary motivation for trying to stream from the working
2194 * tree file and to avoid mmaping it in core is to deal with large
2195 * binary blobs, they generally do not want to get any conversion, and
2196 * callers should avoid this code path when filters are requested.
2197 */
2198 static int index_stream(struct object_id *oid, int fd, size_t size,
2199 enum object_type type, const char *path,
2200 unsigned flags)
2201 {
2202 return index_bulk_checkin(oid, fd, size, type, path, flags);
2203 }
2204
2205 int index_fd(struct index_state *istate, struct object_id *oid,
2206 int fd, struct stat *st,
2207 enum object_type type, const char *path, unsigned flags)
2208 {
2209 int ret;
2210
2211 /*
2212 * Call xsize_t() only when needed to avoid potentially unnecessary
2213 * die() for large files.
2214 */
2215 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, path))
2216 ret = index_stream_convert_blob(istate, oid, fd, path, flags);
2217 else if (!S_ISREG(st->st_mode))
2218 ret = index_pipe(istate, oid, fd, type, path, flags);
2219 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
2220 (path && would_convert_to_git(istate, path)))
2221 ret = index_core(istate, oid, fd, xsize_t(st->st_size),
2222 type, path, flags);
2223 else
2224 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
2225 flags);
2226 close(fd);
2227 return ret;
2228 }
2229
2230 int index_path(struct index_state *istate, struct object_id *oid,
2231 const char *path, struct stat *st, unsigned flags)
2232 {
2233 int fd;
2234 struct strbuf sb = STRBUF_INIT;
2235 int rc = 0;
2236
2237 switch (st->st_mode & S_IFMT) {
2238 case S_IFREG:
2239 fd = open(path, O_RDONLY);
2240 if (fd < 0)
2241 return error_errno("open(\"%s\")", path);
2242 if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0)
2243 return error(_("%s: failed to insert into database"),
2244 path);
2245 break;
2246 case S_IFLNK:
2247 if (strbuf_readlink(&sb, path, st->st_size))
2248 return error_errno("readlink(\"%s\")", path);
2249 if (!(flags & HASH_WRITE_OBJECT))
2250 hash_object_file(the_hash_algo, sb.buf, sb.len,
2251 blob_type, oid);
2252 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
2253 rc = error(_("%s: failed to insert into database"), path);
2254 strbuf_release(&sb);
2255 break;
2256 case S_IFDIR:
2257 return resolve_gitlink_ref(path, "HEAD", oid);
2258 default:
2259 return error(_("%s: unsupported file type"), path);
2260 }
2261 return rc;
2262 }
2263
2264 int read_pack_header(int fd, struct pack_header *header)
2265 {
2266 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
2267 /* "eof before pack header was fully read" */
2268 return PH_ERROR_EOF;
2269
2270 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2271 /* "protocol error (pack signature mismatch detected)" */
2272 return PH_ERROR_PACK_SIGNATURE;
2273 if (!pack_version_ok(header->hdr_version))
2274 /* "protocol error (pack version unsupported)" */
2275 return PH_ERROR_PROTOCOL;
2276 return 0;
2277 }
2278
2279 void assert_oid_type(const struct object_id *oid, enum object_type expect)
2280 {
2281 enum object_type type = oid_object_info(the_repository, oid, NULL);
2282 if (type < 0)
2283 die(_("%s is not a valid object"), oid_to_hex(oid));
2284 if (type != expect)
2285 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
2286 type_name(expect));
2287 }
2288
2289 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2290 struct strbuf *path,
2291 each_loose_object_fn obj_cb,
2292 each_loose_cruft_fn cruft_cb,
2293 each_loose_subdir_fn subdir_cb,
2294 void *data)
2295 {
2296 size_t origlen, baselen;
2297 DIR *dir;
2298 struct dirent *de;
2299 int r = 0;
2300 struct object_id oid;
2301
2302 if (subdir_nr > 0xff)
2303 BUG("invalid loose object subdirectory: %x", subdir_nr);
2304
2305 origlen = path->len;
2306 strbuf_complete(path, '/');
2307 strbuf_addf(path, "%02x", subdir_nr);
2308
2309 dir = opendir(path->buf);
2310 if (!dir) {
2311 if (errno != ENOENT)
2312 r = error_errno(_("unable to open %s"), path->buf);
2313 strbuf_setlen(path, origlen);
2314 return r;
2315 }
2316
2317 oid.hash[0] = subdir_nr;
2318 strbuf_addch(path, '/');
2319 baselen = path->len;
2320
2321 while ((de = readdir_skip_dot_and_dotdot(dir))) {
2322 size_t namelen;
2323
2324 namelen = strlen(de->d_name);
2325 strbuf_setlen(path, baselen);
2326 strbuf_add(path, de->d_name, namelen);
2327 if (namelen == the_hash_algo->hexsz - 2 &&
2328 !hex_to_bytes(oid.hash + 1, de->d_name,
2329 the_hash_algo->rawsz - 1)) {
2330 oid_set_algo(&oid, the_hash_algo);
2331 if (obj_cb) {
2332 r = obj_cb(&oid, path->buf, data);
2333 if (r)
2334 break;
2335 }
2336 continue;
2337 }
2338
2339 if (cruft_cb) {
2340 r = cruft_cb(de->d_name, path->buf, data);
2341 if (r)
2342 break;
2343 }
2344 }
2345 closedir(dir);
2346
2347 strbuf_setlen(path, baselen - 1);
2348 if (!r && subdir_cb)
2349 r = subdir_cb(subdir_nr, path->buf, data);
2350
2351 strbuf_setlen(path, origlen);
2352
2353 return r;
2354 }
2355
2356 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2357 each_loose_object_fn obj_cb,
2358 each_loose_cruft_fn cruft_cb,
2359 each_loose_subdir_fn subdir_cb,
2360 void *data)
2361 {
2362 int r = 0;
2363 int i;
2364
2365 for (i = 0; i < 256; i++) {
2366 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2367 subdir_cb, data);
2368 if (r)
2369 break;
2370 }
2371
2372 return r;
2373 }
2374
2375 int for_each_loose_file_in_objdir(const char *path,
2376 each_loose_object_fn obj_cb,
2377 each_loose_cruft_fn cruft_cb,
2378 each_loose_subdir_fn subdir_cb,
2379 void *data)
2380 {
2381 struct strbuf buf = STRBUF_INIT;
2382 int r;
2383
2384 strbuf_addstr(&buf, path);
2385 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2386 subdir_cb, data);
2387 strbuf_release(&buf);
2388
2389 return r;
2390 }
2391
2392 int for_each_loose_object(each_loose_object_fn cb, void *data,
2393 enum for_each_object_flags flags)
2394 {
2395 struct object_directory *odb;
2396
2397 prepare_alt_odb(the_repository);
2398 for (odb = the_repository->objects->odb; odb; odb = odb->next) {
2399 int r = for_each_loose_file_in_objdir(odb->path, cb, NULL,
2400 NULL, data);
2401 if (r)
2402 return r;
2403
2404 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2405 break;
2406 }
2407
2408 return 0;
2409 }
2410
2411 static int append_loose_object(const struct object_id *oid, const char *path,
2412 void *data)
2413 {
2414 oidtree_insert(data, oid);
2415 return 0;
2416 }
2417
2418 struct oidtree *odb_loose_cache(struct object_directory *odb,
2419 const struct object_id *oid)
2420 {
2421 int subdir_nr = oid->hash[0];
2422 struct strbuf buf = STRBUF_INIT;
2423 size_t word_bits = bitsizeof(odb->loose_objects_subdir_seen[0]);
2424 size_t word_index = subdir_nr / word_bits;
2425 size_t mask = (size_t)1u << (subdir_nr % word_bits);
2426 uint32_t *bitmap;
2427
2428 if (subdir_nr < 0 ||
2429 subdir_nr >= bitsizeof(odb->loose_objects_subdir_seen))
2430 BUG("subdir_nr out of range");
2431
2432 bitmap = &odb->loose_objects_subdir_seen[word_index];
2433 if (*bitmap & mask)
2434 return odb->loose_objects_cache;
2435 if (!odb->loose_objects_cache) {
2436 ALLOC_ARRAY(odb->loose_objects_cache, 1);
2437 oidtree_init(odb->loose_objects_cache);
2438 }
2439 strbuf_addstr(&buf, odb->path);
2440 for_each_file_in_obj_subdir(subdir_nr, &buf,
2441 append_loose_object,
2442 NULL, NULL,
2443 odb->loose_objects_cache);
2444 *bitmap |= mask;
2445 strbuf_release(&buf);
2446 return odb->loose_objects_cache;
2447 }
2448
2449 void odb_clear_loose_cache(struct object_directory *odb)
2450 {
2451 oidtree_clear(odb->loose_objects_cache);
2452 FREE_AND_NULL(odb->loose_objects_cache);
2453 memset(&odb->loose_objects_subdir_seen, 0,
2454 sizeof(odb->loose_objects_subdir_seen));
2455 }
2456
2457 static int check_stream_oid(git_zstream *stream,
2458 const char *hdr,
2459 unsigned long size,
2460 const char *path,
2461 const struct object_id *expected_oid)
2462 {
2463 git_hash_ctx c;
2464 struct object_id real_oid;
2465 unsigned char buf[4096];
2466 unsigned long total_read;
2467 int status = Z_OK;
2468
2469 the_hash_algo->init_fn(&c);
2470 the_hash_algo->update_fn(&c, hdr, stream->total_out);
2471
2472 /*
2473 * We already read some bytes into hdr, but the ones up to the NUL
2474 * do not count against the object's content size.
2475 */
2476 total_read = stream->total_out - strlen(hdr) - 1;
2477
2478 /*
2479 * This size comparison must be "<=" to read the final zlib packets;
2480 * see the comment in unpack_loose_rest for details.
2481 */
2482 while (total_read <= size &&
2483 (status == Z_OK ||
2484 (status == Z_BUF_ERROR && !stream->avail_out))) {
2485 stream->next_out = buf;
2486 stream->avail_out = sizeof(buf);
2487 if (size - total_read < stream->avail_out)
2488 stream->avail_out = size - total_read;
2489 status = git_inflate(stream, Z_FINISH);
2490 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2491 total_read += stream->next_out - buf;
2492 }
2493 git_inflate_end(stream);
2494
2495 if (status != Z_STREAM_END) {
2496 error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid));
2497 return -1;
2498 }
2499 if (stream->avail_in) {
2500 error(_("garbage at end of loose object '%s'"),
2501 oid_to_hex(expected_oid));
2502 return -1;
2503 }
2504
2505 the_hash_algo->final_oid_fn(&real_oid, &c);
2506 if (!oideq(expected_oid, &real_oid)) {
2507 error(_("hash mismatch for %s (expected %s)"), path,
2508 oid_to_hex(expected_oid));
2509 return -1;
2510 }
2511
2512 return 0;
2513 }
2514
2515 int read_loose_object(const char *path,
2516 const struct object_id *expected_oid,
2517 struct object_id *real_oid,
2518 void **contents,
2519 struct object_info *oi)
2520 {
2521 int ret = -1;
2522 void *map = NULL;
2523 unsigned long mapsize;
2524 git_zstream stream;
2525 char hdr[MAX_HEADER_LEN];
2526 unsigned long *size = oi->sizep;
2527
2528 map = map_loose_object_1(the_repository, path, NULL, &mapsize);
2529 if (!map) {
2530 error_errno(_("unable to mmap %s"), path);
2531 goto out;
2532 }
2533
2534 if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr),
2535 NULL) < 0) {
2536 error(_("unable to unpack header of %s"), path);
2537 goto out;
2538 }
2539
2540 if (parse_loose_header(hdr, oi) < 0) {
2541 error(_("unable to parse header of %s"), path);
2542 git_inflate_end(&stream);
2543 goto out;
2544 }
2545
2546 if (*oi->typep == OBJ_BLOB && *size > big_file_threshold) {
2547 if (check_stream_oid(&stream, hdr, *size, path, expected_oid) < 0)
2548 goto out;
2549 } else {
2550 *contents = unpack_loose_rest(&stream, hdr, *size, expected_oid);
2551 if (!*contents) {
2552 error(_("unable to unpack contents of %s"), path);
2553 git_inflate_end(&stream);
2554 goto out;
2555 }
2556 if (check_object_signature(the_repository, expected_oid,
2557 *contents, *size,
2558 oi->type_name->buf, real_oid))
2559 goto out;
2560 }
2561
2562 ret = 0; /* everything checks out */
2563
2564 out:
2565 if (map)
2566 munmap(map, mapsize);
2567 return ret;
2568 }