]> git.ipfire.org Git - thirdparty/git.git/blob - object-file.c
clone: allow "--bare" with "-o"
[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 .name = NULL,
171 .format_id = 0x00000000,
172 .rawsz = 0,
173 .hexsz = 0,
174 .blksz = 0,
175 .init_fn = git_hash_unknown_init,
176 .clone_fn = git_hash_unknown_clone,
177 .update_fn = git_hash_unknown_update,
178 .final_fn = git_hash_unknown_final,
179 .final_oid_fn = git_hash_unknown_final_oid,
180 .empty_tree = NULL,
181 .empty_blob = NULL,
182 .null_oid = NULL,
183 },
184 {
185 .name = "sha1",
186 .format_id = GIT_SHA1_FORMAT_ID,
187 .rawsz = GIT_SHA1_RAWSZ,
188 .hexsz = GIT_SHA1_HEXSZ,
189 .blksz = GIT_SHA1_BLKSZ,
190 .init_fn = git_hash_sha1_init,
191 .clone_fn = git_hash_sha1_clone,
192 .update_fn = git_hash_sha1_update,
193 .final_fn = git_hash_sha1_final,
194 .final_oid_fn = git_hash_sha1_final_oid,
195 .empty_tree = &empty_tree_oid,
196 .empty_blob = &empty_blob_oid,
197 .null_oid = &null_oid_sha1,
198 },
199 {
200 .name = "sha256",
201 .format_id = GIT_SHA256_FORMAT_ID,
202 .rawsz = GIT_SHA256_RAWSZ,
203 .hexsz = GIT_SHA256_HEXSZ,
204 .blksz = GIT_SHA256_BLKSZ,
205 .init_fn = git_hash_sha256_init,
206 .clone_fn = git_hash_sha256_clone,
207 .update_fn = git_hash_sha256_update,
208 .final_fn = git_hash_sha256_final,
209 .final_oid_fn = git_hash_sha256_final_oid,
210 .empty_tree = &empty_tree_oid_sha256,
211 .empty_blob = &empty_blob_oid_sha256,
212 .null_oid = &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 .oid = {
278 .hash = EMPTY_TREE_SHA1_BIN_LITERAL,
279 },
280 .type = OBJ_TREE,
281 .buf = "",
282 };
283
284 static struct cached_object *find_cached_object(const struct object_id *oid)
285 {
286 int i;
287 struct cached_object *co = cached_objects;
288
289 for (i = 0; i < cached_object_nr; i++, co++) {
290 if (oideq(&co->oid, oid))
291 return co;
292 }
293 if (oideq(oid, the_hash_algo->empty_tree))
294 return &empty_tree;
295 return NULL;
296 }
297
298
299 static int get_conv_flags(unsigned flags)
300 {
301 if (flags & HASH_RENORMALIZE)
302 return CONV_EOL_RENORMALIZE;
303 else if (flags & HASH_WRITE_OBJECT)
304 return global_conv_flags_eol | CONV_WRITE_OBJECT;
305 else
306 return 0;
307 }
308
309
310 int mkdir_in_gitdir(const char *path)
311 {
312 if (mkdir(path, 0777)) {
313 int saved_errno = errno;
314 struct stat st;
315 struct strbuf sb = STRBUF_INIT;
316
317 if (errno != EEXIST)
318 return -1;
319 /*
320 * Are we looking at a path in a symlinked worktree
321 * whose original repository does not yet have it?
322 * e.g. .git/rr-cache pointing at its original
323 * repository in which the user hasn't performed any
324 * conflict resolution yet?
325 */
326 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
327 strbuf_readlink(&sb, path, st.st_size) ||
328 !is_absolute_path(sb.buf) ||
329 mkdir(sb.buf, 0777)) {
330 strbuf_release(&sb);
331 errno = saved_errno;
332 return -1;
333 }
334 strbuf_release(&sb);
335 }
336 return adjust_shared_perm(path);
337 }
338
339 static enum scld_error safe_create_leading_directories_1(char *path, int share)
340 {
341 char *next_component = path + offset_1st_component(path);
342 enum scld_error ret = SCLD_OK;
343
344 while (ret == SCLD_OK && next_component) {
345 struct stat st;
346 char *slash = next_component, slash_character;
347
348 while (*slash && !is_dir_sep(*slash))
349 slash++;
350
351 if (!*slash)
352 break;
353
354 next_component = slash + 1;
355 while (is_dir_sep(*next_component))
356 next_component++;
357 if (!*next_component)
358 break;
359
360 slash_character = *slash;
361 *slash = '\0';
362 if (!stat(path, &st)) {
363 /* path exists */
364 if (!S_ISDIR(st.st_mode)) {
365 errno = ENOTDIR;
366 ret = SCLD_EXISTS;
367 }
368 } else if (mkdir(path, 0777)) {
369 if (errno == EEXIST &&
370 !stat(path, &st) && S_ISDIR(st.st_mode))
371 ; /* somebody created it since we checked */
372 else if (errno == ENOENT)
373 /*
374 * Either mkdir() failed because
375 * somebody just pruned the containing
376 * directory, or stat() failed because
377 * the file that was in our way was
378 * just removed. Either way, inform
379 * the caller that it might be worth
380 * trying again:
381 */
382 ret = SCLD_VANISHED;
383 else
384 ret = SCLD_FAILED;
385 } else if (share && adjust_shared_perm(path)) {
386 ret = SCLD_PERMS;
387 }
388 *slash = slash_character;
389 }
390 return ret;
391 }
392
393 enum scld_error safe_create_leading_directories(char *path)
394 {
395 return safe_create_leading_directories_1(path, 1);
396 }
397
398 enum scld_error safe_create_leading_directories_no_share(char *path)
399 {
400 return safe_create_leading_directories_1(path, 0);
401 }
402
403 enum scld_error safe_create_leading_directories_const(const char *path)
404 {
405 int save_errno;
406 /* path points to cache entries, so xstrdup before messing with it */
407 char *buf = xstrdup(path);
408 enum scld_error result = safe_create_leading_directories(buf);
409
410 save_errno = errno;
411 free(buf);
412 errno = save_errno;
413 return result;
414 }
415
416 static void fill_loose_path(struct strbuf *buf, const struct object_id *oid)
417 {
418 int i;
419 for (i = 0; i < the_hash_algo->rawsz; i++) {
420 static char hex[] = "0123456789abcdef";
421 unsigned int val = oid->hash[i];
422 strbuf_addch(buf, hex[val >> 4]);
423 strbuf_addch(buf, hex[val & 0xf]);
424 if (!i)
425 strbuf_addch(buf, '/');
426 }
427 }
428
429 static const char *odb_loose_path(struct object_directory *odb,
430 struct strbuf *buf,
431 const struct object_id *oid)
432 {
433 strbuf_reset(buf);
434 strbuf_addstr(buf, odb->path);
435 strbuf_addch(buf, '/');
436 fill_loose_path(buf, oid);
437 return buf->buf;
438 }
439
440 const char *loose_object_path(struct repository *r, struct strbuf *buf,
441 const struct object_id *oid)
442 {
443 return odb_loose_path(r->objects->odb, buf, oid);
444 }
445
446 /*
447 * Return non-zero iff the path is usable as an alternate object database.
448 */
449 static int alt_odb_usable(struct raw_object_store *o,
450 struct strbuf *path,
451 const char *normalized_objdir, khiter_t *pos)
452 {
453 int r;
454
455 /* Detect cases where alternate disappeared */
456 if (!is_directory(path->buf)) {
457 error(_("object directory %s does not exist; "
458 "check .git/objects/info/alternates"),
459 path->buf);
460 return 0;
461 }
462
463 /*
464 * Prevent the common mistake of listing the same
465 * thing twice, or object directory itself.
466 */
467 if (!o->odb_by_path) {
468 khiter_t p;
469
470 o->odb_by_path = kh_init_odb_path_map();
471 assert(!o->odb->next);
472 p = kh_put_odb_path_map(o->odb_by_path, o->odb->path, &r);
473 assert(r == 1); /* never used */
474 kh_value(o->odb_by_path, p) = o->odb;
475 }
476 if (fspatheq(path->buf, normalized_objdir))
477 return 0;
478 *pos = kh_put_odb_path_map(o->odb_by_path, path->buf, &r);
479 /* r: 0 = exists, 1 = never used, 2 = deleted */
480 return r == 0 ? 0 : 1;
481 }
482
483 /*
484 * Prepare alternate object database registry.
485 *
486 * The variable alt_odb_list points at the list of struct
487 * object_directory. The elements on this list come from
488 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
489 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
490 * whose contents is similar to that environment variable but can be
491 * LF separated. Its base points at a statically allocated buffer that
492 * contains "/the/directory/corresponding/to/.git/objects/...", while
493 * its name points just after the slash at the end of ".git/objects/"
494 * in the example above, and has enough space to hold all hex characters
495 * of the object ID, an extra slash for the first level indirection, and
496 * the terminating NUL.
497 */
498 static void read_info_alternates(struct repository *r,
499 const char *relative_base,
500 int depth);
501 static int link_alt_odb_entry(struct repository *r, const struct strbuf *entry,
502 const char *relative_base, int depth, const char *normalized_objdir)
503 {
504 struct object_directory *ent;
505 struct strbuf pathbuf = STRBUF_INIT;
506 khiter_t pos;
507
508 if (!is_absolute_path(entry->buf) && relative_base) {
509 strbuf_realpath(&pathbuf, relative_base, 1);
510 strbuf_addch(&pathbuf, '/');
511 }
512 strbuf_addbuf(&pathbuf, entry);
513
514 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
515 error(_("unable to normalize alternate object path: %s"),
516 pathbuf.buf);
517 strbuf_release(&pathbuf);
518 return -1;
519 }
520
521 /*
522 * The trailing slash after the directory name is given by
523 * this function at the end. Remove duplicates.
524 */
525 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
526 strbuf_setlen(&pathbuf, pathbuf.len - 1);
527
528 if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir, &pos)) {
529 strbuf_release(&pathbuf);
530 return -1;
531 }
532
533 CALLOC_ARRAY(ent, 1);
534 /* pathbuf.buf is already in r->objects->odb_by_path */
535 ent->path = strbuf_detach(&pathbuf, NULL);
536
537 /* add the alternate entry */
538 *r->objects->odb_tail = ent;
539 r->objects->odb_tail = &(ent->next);
540 ent->next = NULL;
541 assert(r->objects->odb_by_path);
542 kh_value(r->objects->odb_by_path, pos) = ent;
543
544 /* recursively add alternates */
545 read_info_alternates(r, ent->path, depth + 1);
546
547 return 0;
548 }
549
550 static const char *parse_alt_odb_entry(const char *string,
551 int sep,
552 struct strbuf *out)
553 {
554 const char *end;
555
556 strbuf_reset(out);
557
558 if (*string == '#') {
559 /* comment; consume up to next separator */
560 end = strchrnul(string, sep);
561 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
562 /*
563 * quoted path; unquote_c_style has copied the
564 * data for us and set "end". Broken quoting (e.g.,
565 * an entry that doesn't end with a quote) falls
566 * back to the unquoted case below.
567 */
568 } else {
569 /* normal, unquoted path */
570 end = strchrnul(string, sep);
571 strbuf_add(out, string, end - string);
572 }
573
574 if (*end)
575 end++;
576 return end;
577 }
578
579 static void link_alt_odb_entries(struct repository *r, const char *alt,
580 int sep, const char *relative_base, int depth)
581 {
582 struct strbuf objdirbuf = STRBUF_INIT;
583 struct strbuf entry = STRBUF_INIT;
584
585 if (!alt || !*alt)
586 return;
587
588 if (depth > 5) {
589 error(_("%s: ignoring alternate object stores, nesting too deep"),
590 relative_base);
591 return;
592 }
593
594 strbuf_add_absolute_path(&objdirbuf, r->objects->odb->path);
595 if (strbuf_normalize_path(&objdirbuf) < 0)
596 die(_("unable to normalize object directory: %s"),
597 objdirbuf.buf);
598
599 while (*alt) {
600 alt = parse_alt_odb_entry(alt, sep, &entry);
601 if (!entry.len)
602 continue;
603 link_alt_odb_entry(r, &entry,
604 relative_base, depth, objdirbuf.buf);
605 }
606 strbuf_release(&entry);
607 strbuf_release(&objdirbuf);
608 }
609
610 static void read_info_alternates(struct repository *r,
611 const char *relative_base,
612 int depth)
613 {
614 char *path;
615 struct strbuf buf = STRBUF_INIT;
616
617 path = xstrfmt("%s/info/alternates", relative_base);
618 if (strbuf_read_file(&buf, path, 1024) < 0) {
619 warn_on_fopen_errors(path);
620 free(path);
621 return;
622 }
623
624 link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
625 strbuf_release(&buf);
626 free(path);
627 }
628
629 void add_to_alternates_file(const char *reference)
630 {
631 struct lock_file lock = LOCK_INIT;
632 char *alts = git_pathdup("objects/info/alternates");
633 FILE *in, *out;
634 int found = 0;
635
636 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
637 out = fdopen_lock_file(&lock, "w");
638 if (!out)
639 die_errno(_("unable to fdopen alternates lockfile"));
640
641 in = fopen(alts, "r");
642 if (in) {
643 struct strbuf line = STRBUF_INIT;
644
645 while (strbuf_getline(&line, in) != EOF) {
646 if (!strcmp(reference, line.buf)) {
647 found = 1;
648 break;
649 }
650 fprintf_or_die(out, "%s\n", line.buf);
651 }
652
653 strbuf_release(&line);
654 fclose(in);
655 }
656 else if (errno != ENOENT)
657 die_errno(_("unable to read alternates file"));
658
659 if (found) {
660 rollback_lock_file(&lock);
661 } else {
662 fprintf_or_die(out, "%s\n", reference);
663 if (commit_lock_file(&lock))
664 die_errno(_("unable to move new alternates file into place"));
665 if (the_repository->objects->loaded_alternates)
666 link_alt_odb_entries(the_repository, reference,
667 '\n', NULL, 0);
668 }
669 free(alts);
670 }
671
672 void add_to_alternates_memory(const char *reference)
673 {
674 /*
675 * Make sure alternates are initialized, or else our entry may be
676 * overwritten when they are.
677 */
678 prepare_alt_odb(the_repository);
679
680 link_alt_odb_entries(the_repository, reference,
681 '\n', NULL, 0);
682 }
683
684 struct object_directory *set_temporary_primary_odb(const char *dir, int will_destroy)
685 {
686 struct object_directory *new_odb;
687
688 /*
689 * Make sure alternates are initialized, or else our entry may be
690 * overwritten when they are.
691 */
692 prepare_alt_odb(the_repository);
693
694 /*
695 * Make a new primary odb and link the old primary ODB in as an
696 * alternate
697 */
698 new_odb = xcalloc(1, sizeof(*new_odb));
699 new_odb->path = xstrdup(dir);
700
701 /*
702 * Disable ref updates while a temporary odb is active, since
703 * the objects in the database may roll back.
704 */
705 new_odb->disable_ref_updates = 1;
706 new_odb->will_destroy = will_destroy;
707 new_odb->next = the_repository->objects->odb;
708 the_repository->objects->odb = new_odb;
709 return new_odb->next;
710 }
711
712 void restore_primary_odb(struct object_directory *restore_odb, const char *old_path)
713 {
714 struct object_directory *cur_odb = the_repository->objects->odb;
715
716 if (strcmp(old_path, cur_odb->path))
717 BUG("expected %s as primary object store; found %s",
718 old_path, cur_odb->path);
719
720 if (cur_odb->next != restore_odb)
721 BUG("we expect the old primary object store to be the first alternate");
722
723 the_repository->objects->odb = restore_odb;
724 free_object_directory(cur_odb);
725 }
726
727 /*
728 * Compute the exact path an alternate is at and returns it. In case of
729 * error NULL is returned and the human readable error is added to `err`
730 * `path` may be relative and should point to $GIT_DIR.
731 * `err` must not be null.
732 */
733 char *compute_alternate_path(const char *path, struct strbuf *err)
734 {
735 char *ref_git = NULL;
736 const char *repo;
737 int seen_error = 0;
738
739 ref_git = real_pathdup(path, 0);
740 if (!ref_git) {
741 seen_error = 1;
742 strbuf_addf(err, _("path '%s' does not exist"), path);
743 goto out;
744 }
745
746 repo = read_gitfile(ref_git);
747 if (!repo)
748 repo = read_gitfile(mkpath("%s/.git", ref_git));
749 if (repo) {
750 free(ref_git);
751 ref_git = xstrdup(repo);
752 }
753
754 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
755 char *ref_git_git = mkpathdup("%s/.git", ref_git);
756 free(ref_git);
757 ref_git = ref_git_git;
758 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
759 struct strbuf sb = STRBUF_INIT;
760 seen_error = 1;
761 if (get_common_dir(&sb, ref_git)) {
762 strbuf_addf(err,
763 _("reference repository '%s' as a linked "
764 "checkout is not supported yet."),
765 path);
766 goto out;
767 }
768
769 strbuf_addf(err, _("reference repository '%s' is not a "
770 "local repository."), path);
771 goto out;
772 }
773
774 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
775 strbuf_addf(err, _("reference repository '%s' is shallow"),
776 path);
777 seen_error = 1;
778 goto out;
779 }
780
781 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
782 strbuf_addf(err,
783 _("reference repository '%s' is grafted"),
784 path);
785 seen_error = 1;
786 goto out;
787 }
788
789 out:
790 if (seen_error) {
791 FREE_AND_NULL(ref_git);
792 }
793
794 return ref_git;
795 }
796
797 struct object_directory *find_odb(struct repository *r, const char *obj_dir)
798 {
799 struct object_directory *odb;
800 char *obj_dir_real = real_pathdup(obj_dir, 1);
801 struct strbuf odb_path_real = STRBUF_INIT;
802
803 prepare_alt_odb(r);
804 for (odb = r->objects->odb; odb; odb = odb->next) {
805 strbuf_realpath(&odb_path_real, odb->path, 1);
806 if (!strcmp(obj_dir_real, odb_path_real.buf))
807 break;
808 }
809
810 free(obj_dir_real);
811 strbuf_release(&odb_path_real);
812
813 if (!odb)
814 die(_("could not find object directory matching %s"), obj_dir);
815 return odb;
816 }
817
818 static void fill_alternate_refs_command(struct child_process *cmd,
819 const char *repo_path)
820 {
821 const char *value;
822
823 if (!git_config_get_value("core.alternateRefsCommand", &value)) {
824 cmd->use_shell = 1;
825
826 strvec_push(&cmd->args, value);
827 strvec_push(&cmd->args, repo_path);
828 } else {
829 cmd->git_cmd = 1;
830
831 strvec_pushf(&cmd->args, "--git-dir=%s", repo_path);
832 strvec_push(&cmd->args, "for-each-ref");
833 strvec_push(&cmd->args, "--format=%(objectname)");
834
835 if (!git_config_get_value("core.alternateRefsPrefixes", &value)) {
836 strvec_push(&cmd->args, "--");
837 strvec_split(&cmd->args, value);
838 }
839 }
840
841 strvec_pushv(&cmd->env, (const char **)local_repo_env);
842 cmd->out = -1;
843 }
844
845 static void read_alternate_refs(const char *path,
846 alternate_ref_fn *cb,
847 void *data)
848 {
849 struct child_process cmd = CHILD_PROCESS_INIT;
850 struct strbuf line = STRBUF_INIT;
851 FILE *fh;
852
853 fill_alternate_refs_command(&cmd, path);
854
855 if (start_command(&cmd))
856 return;
857
858 fh = xfdopen(cmd.out, "r");
859 while (strbuf_getline_lf(&line, fh) != EOF) {
860 struct object_id oid;
861 const char *p;
862
863 if (parse_oid_hex(line.buf, &oid, &p) || *p) {
864 warning(_("invalid line while parsing alternate refs: %s"),
865 line.buf);
866 break;
867 }
868
869 cb(&oid, data);
870 }
871
872 fclose(fh);
873 finish_command(&cmd);
874 strbuf_release(&line);
875 }
876
877 struct alternate_refs_data {
878 alternate_ref_fn *fn;
879 void *data;
880 };
881
882 static int refs_from_alternate_cb(struct object_directory *e,
883 void *data)
884 {
885 struct strbuf path = STRBUF_INIT;
886 size_t base_len;
887 struct alternate_refs_data *cb = data;
888
889 if (!strbuf_realpath(&path, e->path, 0))
890 goto out;
891 if (!strbuf_strip_suffix(&path, "/objects"))
892 goto out;
893 base_len = path.len;
894
895 /* Is this a git repository with refs? */
896 strbuf_addstr(&path, "/refs");
897 if (!is_directory(path.buf))
898 goto out;
899 strbuf_setlen(&path, base_len);
900
901 read_alternate_refs(path.buf, cb->fn, cb->data);
902
903 out:
904 strbuf_release(&path);
905 return 0;
906 }
907
908 void for_each_alternate_ref(alternate_ref_fn fn, void *data)
909 {
910 struct alternate_refs_data cb;
911 cb.fn = fn;
912 cb.data = data;
913 foreach_alt_odb(refs_from_alternate_cb, &cb);
914 }
915
916 int foreach_alt_odb(alt_odb_fn fn, void *cb)
917 {
918 struct object_directory *ent;
919 int r = 0;
920
921 prepare_alt_odb(the_repository);
922 for (ent = the_repository->objects->odb->next; ent; ent = ent->next) {
923 r = fn(ent, cb);
924 if (r)
925 break;
926 }
927 return r;
928 }
929
930 void prepare_alt_odb(struct repository *r)
931 {
932 if (r->objects->loaded_alternates)
933 return;
934
935 link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
936
937 read_info_alternates(r, r->objects->odb->path, 0);
938 r->objects->loaded_alternates = 1;
939 }
940
941 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
942 static int freshen_file(const char *fn)
943 {
944 return !utime(fn, NULL);
945 }
946
947 /*
948 * All of the check_and_freshen functions return 1 if the file exists and was
949 * freshened (if freshening was requested), 0 otherwise. If they return
950 * 0, you should not assume that it is safe to skip a write of the object (it
951 * either does not exist on disk, or has a stale mtime and may be subject to
952 * pruning).
953 */
954 int check_and_freshen_file(const char *fn, int freshen)
955 {
956 if (access(fn, F_OK))
957 return 0;
958 if (freshen && !freshen_file(fn))
959 return 0;
960 return 1;
961 }
962
963 static int check_and_freshen_odb(struct object_directory *odb,
964 const struct object_id *oid,
965 int freshen)
966 {
967 static struct strbuf path = STRBUF_INIT;
968 odb_loose_path(odb, &path, oid);
969 return check_and_freshen_file(path.buf, freshen);
970 }
971
972 static int check_and_freshen_local(const struct object_id *oid, int freshen)
973 {
974 return check_and_freshen_odb(the_repository->objects->odb, oid, freshen);
975 }
976
977 static int check_and_freshen_nonlocal(const struct object_id *oid, int freshen)
978 {
979 struct object_directory *odb;
980
981 prepare_alt_odb(the_repository);
982 for (odb = the_repository->objects->odb->next; odb; odb = odb->next) {
983 if (check_and_freshen_odb(odb, oid, freshen))
984 return 1;
985 }
986 return 0;
987 }
988
989 static int check_and_freshen(const struct object_id *oid, int freshen)
990 {
991 return check_and_freshen_local(oid, freshen) ||
992 check_and_freshen_nonlocal(oid, freshen);
993 }
994
995 int has_loose_object_nonlocal(const struct object_id *oid)
996 {
997 return check_and_freshen_nonlocal(oid, 0);
998 }
999
1000 int has_loose_object(const struct object_id *oid)
1001 {
1002 return check_and_freshen(oid, 0);
1003 }
1004
1005 static void mmap_limit_check(size_t length)
1006 {
1007 static size_t limit = 0;
1008 if (!limit) {
1009 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
1010 if (!limit)
1011 limit = SIZE_MAX;
1012 }
1013 if (length > limit)
1014 die(_("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX),
1015 (uintmax_t)length, (uintmax_t)limit);
1016 }
1017
1018 void *xmmap_gently(void *start, size_t length,
1019 int prot, int flags, int fd, off_t offset)
1020 {
1021 void *ret;
1022
1023 mmap_limit_check(length);
1024 ret = mmap(start, length, prot, flags, fd, offset);
1025 if (ret == MAP_FAILED && !length)
1026 ret = NULL;
1027 return ret;
1028 }
1029
1030 const char *mmap_os_err(void)
1031 {
1032 static const char blank[] = "";
1033 #if defined(__linux__)
1034 if (errno == ENOMEM) {
1035 /* this continues an existing error message: */
1036 static const char enomem[] =
1037 ", check sys.vm.max_map_count and/or RLIMIT_DATA";
1038 return enomem;
1039 }
1040 #endif /* OS-specific bits */
1041 return blank;
1042 }
1043
1044 void *xmmap(void *start, size_t length,
1045 int prot, int flags, int fd, off_t offset)
1046 {
1047 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
1048 if (ret == MAP_FAILED)
1049 die_errno(_("mmap failed%s"), mmap_os_err());
1050 return ret;
1051 }
1052
1053 static int format_object_header_literally(char *str, size_t size,
1054 const char *type, size_t objsize)
1055 {
1056 return xsnprintf(str, size, "%s %"PRIuMAX, type, (uintmax_t)objsize) + 1;
1057 }
1058
1059 int format_object_header(char *str, size_t size, enum object_type type,
1060 size_t objsize)
1061 {
1062 const char *name = type_name(type);
1063
1064 if (!name)
1065 BUG("could not get a type name for 'enum object_type' value %d", type);
1066
1067 return format_object_header_literally(str, size, name, objsize);
1068 }
1069
1070 int check_object_signature(struct repository *r, const struct object_id *oid,
1071 void *buf, unsigned long size,
1072 enum object_type type)
1073 {
1074 struct object_id real_oid;
1075
1076 hash_object_file(r->hash_algo, buf, size, type, &real_oid);
1077
1078 return !oideq(oid, &real_oid) ? -1 : 0;
1079 }
1080
1081 int stream_object_signature(struct repository *r, const struct object_id *oid)
1082 {
1083 struct object_id real_oid;
1084 unsigned long size;
1085 enum object_type obj_type;
1086 struct git_istream *st;
1087 git_hash_ctx c;
1088 char hdr[MAX_HEADER_LEN];
1089 int hdrlen;
1090
1091 st = open_istream(r, oid, &obj_type, &size, NULL);
1092 if (!st)
1093 return -1;
1094
1095 /* Generate the header */
1096 hdrlen = format_object_header(hdr, sizeof(hdr), obj_type, size);
1097
1098 /* Sha1.. */
1099 r->hash_algo->init_fn(&c);
1100 r->hash_algo->update_fn(&c, hdr, hdrlen);
1101 for (;;) {
1102 char buf[1024 * 16];
1103 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1104
1105 if (readlen < 0) {
1106 close_istream(st);
1107 return -1;
1108 }
1109 if (!readlen)
1110 break;
1111 r->hash_algo->update_fn(&c, buf, readlen);
1112 }
1113 r->hash_algo->final_oid_fn(&real_oid, &c);
1114 close_istream(st);
1115 return !oideq(oid, &real_oid) ? -1 : 0;
1116 }
1117
1118 int git_open_cloexec(const char *name, int flags)
1119 {
1120 int fd;
1121 static int o_cloexec = O_CLOEXEC;
1122
1123 fd = open(name, flags | o_cloexec);
1124 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1125 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1126 o_cloexec &= ~O_CLOEXEC;
1127 fd = open(name, flags | o_cloexec);
1128 }
1129
1130 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1131 {
1132 static int fd_cloexec = FD_CLOEXEC;
1133
1134 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1135 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1136 int flags = fcntl(fd, F_GETFD);
1137 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1138 fd_cloexec = 0;
1139 }
1140 }
1141 #endif
1142 return fd;
1143 }
1144
1145 /*
1146 * Find "oid" as a loose object in the local repository or in an alternate.
1147 * Returns 0 on success, negative on failure.
1148 *
1149 * The "path" out-parameter will give the path of the object we found (if any).
1150 * Note that it may point to static storage and is only valid until another
1151 * call to stat_loose_object().
1152 */
1153 static int stat_loose_object(struct repository *r, const struct object_id *oid,
1154 struct stat *st, const char **path)
1155 {
1156 struct object_directory *odb;
1157 static struct strbuf buf = STRBUF_INIT;
1158
1159 prepare_alt_odb(r);
1160 for (odb = r->objects->odb; odb; odb = odb->next) {
1161 *path = odb_loose_path(odb, &buf, oid);
1162 if (!lstat(*path, st))
1163 return 0;
1164 }
1165
1166 return -1;
1167 }
1168
1169 /*
1170 * Like stat_loose_object(), but actually open the object and return the
1171 * descriptor. See the caveats on the "path" parameter above.
1172 */
1173 static int open_loose_object(struct repository *r,
1174 const struct object_id *oid, const char **path)
1175 {
1176 int fd;
1177 struct object_directory *odb;
1178 int most_interesting_errno = ENOENT;
1179 static struct strbuf buf = STRBUF_INIT;
1180
1181 prepare_alt_odb(r);
1182 for (odb = r->objects->odb; odb; odb = odb->next) {
1183 *path = odb_loose_path(odb, &buf, oid);
1184 fd = git_open(*path);
1185 if (fd >= 0)
1186 return fd;
1187
1188 if (most_interesting_errno == ENOENT)
1189 most_interesting_errno = errno;
1190 }
1191 errno = most_interesting_errno;
1192 return -1;
1193 }
1194
1195 static int quick_has_loose(struct repository *r,
1196 const struct object_id *oid)
1197 {
1198 struct object_directory *odb;
1199
1200 prepare_alt_odb(r);
1201 for (odb = r->objects->odb; odb; odb = odb->next) {
1202 if (oidtree_contains(odb_loose_cache(odb, oid), oid))
1203 return 1;
1204 }
1205 return 0;
1206 }
1207
1208 /*
1209 * Map the loose object at "path" if it is not NULL, or the path found by
1210 * searching for a loose object named "oid".
1211 */
1212 static void *map_loose_object_1(struct repository *r, const char *path,
1213 const struct object_id *oid, unsigned long *size)
1214 {
1215 void *map;
1216 int fd;
1217
1218 if (path)
1219 fd = git_open(path);
1220 else
1221 fd = open_loose_object(r, oid, &path);
1222 map = NULL;
1223 if (fd >= 0) {
1224 struct stat st;
1225
1226 if (!fstat(fd, &st)) {
1227 *size = xsize_t(st.st_size);
1228 if (!*size) {
1229 /* mmap() is forbidden on empty files */
1230 error(_("object file %s is empty"), path);
1231 close(fd);
1232 return NULL;
1233 }
1234 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1235 }
1236 close(fd);
1237 }
1238 return map;
1239 }
1240
1241 void *map_loose_object(struct repository *r,
1242 const struct object_id *oid,
1243 unsigned long *size)
1244 {
1245 return map_loose_object_1(r, NULL, oid, size);
1246 }
1247
1248 enum unpack_loose_header_result unpack_loose_header(git_zstream *stream,
1249 unsigned char *map,
1250 unsigned long mapsize,
1251 void *buffer,
1252 unsigned long bufsiz,
1253 struct strbuf *header)
1254 {
1255 int status;
1256
1257 /* Get the data stream */
1258 memset(stream, 0, sizeof(*stream));
1259 stream->next_in = map;
1260 stream->avail_in = mapsize;
1261 stream->next_out = buffer;
1262 stream->avail_out = bufsiz;
1263
1264 git_inflate_init(stream);
1265 obj_read_unlock();
1266 status = git_inflate(stream, 0);
1267 obj_read_lock();
1268 if (status < Z_OK)
1269 return ULHR_BAD;
1270
1271 /*
1272 * Check if entire header is unpacked in the first iteration.
1273 */
1274 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1275 return ULHR_OK;
1276
1277 /*
1278 * We have a header longer than MAX_HEADER_LEN. The "header"
1279 * here is only non-NULL when we run "cat-file
1280 * --allow-unknown-type".
1281 */
1282 if (!header)
1283 return ULHR_TOO_LONG;
1284
1285 /*
1286 * buffer[0..bufsiz] was not large enough. Copy the partial
1287 * result out to header, and then append the result of further
1288 * reading the stream.
1289 */
1290 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1291 stream->next_out = buffer;
1292 stream->avail_out = bufsiz;
1293
1294 do {
1295 obj_read_unlock();
1296 status = git_inflate(stream, 0);
1297 obj_read_lock();
1298 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1299 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1300 return 0;
1301 stream->next_out = buffer;
1302 stream->avail_out = bufsiz;
1303 } while (status != Z_STREAM_END);
1304 return ULHR_TOO_LONG;
1305 }
1306
1307 static void *unpack_loose_rest(git_zstream *stream,
1308 void *buffer, unsigned long size,
1309 const struct object_id *oid)
1310 {
1311 int bytes = strlen(buffer) + 1;
1312 unsigned char *buf = xmallocz(size);
1313 unsigned long n;
1314 int status = Z_OK;
1315
1316 n = stream->total_out - bytes;
1317 if (n > size)
1318 n = size;
1319 memcpy(buf, (char *) buffer + bytes, n);
1320 bytes = n;
1321 if (bytes <= size) {
1322 /*
1323 * The above condition must be (bytes <= size), not
1324 * (bytes < size). In other words, even though we
1325 * expect no more output and set avail_out to zero,
1326 * the input zlib stream may have bytes that express
1327 * "this concludes the stream", and we *do* want to
1328 * eat that input.
1329 *
1330 * Otherwise we would not be able to test that we
1331 * consumed all the input to reach the expected size;
1332 * we also want to check that zlib tells us that all
1333 * went well with status == Z_STREAM_END at the end.
1334 */
1335 stream->next_out = buf + bytes;
1336 stream->avail_out = size - bytes;
1337 while (status == Z_OK) {
1338 obj_read_unlock();
1339 status = git_inflate(stream, Z_FINISH);
1340 obj_read_lock();
1341 }
1342 }
1343 if (status == Z_STREAM_END && !stream->avail_in) {
1344 git_inflate_end(stream);
1345 return buf;
1346 }
1347
1348 if (status < 0)
1349 error(_("corrupt loose object '%s'"), oid_to_hex(oid));
1350 else if (stream->avail_in)
1351 error(_("garbage at end of loose object '%s'"),
1352 oid_to_hex(oid));
1353 free(buf);
1354 return NULL;
1355 }
1356
1357 /*
1358 * We used to just use "sscanf()", but that's actually way
1359 * too permissive for what we want to check. So do an anal
1360 * object header parse by hand.
1361 */
1362 int parse_loose_header(const char *hdr, struct object_info *oi)
1363 {
1364 const char *type_buf = hdr;
1365 size_t size;
1366 int type, type_len = 0;
1367
1368 /*
1369 * The type can be of any size but is followed by
1370 * a space.
1371 */
1372 for (;;) {
1373 char c = *hdr++;
1374 if (!c)
1375 return -1;
1376 if (c == ' ')
1377 break;
1378 type_len++;
1379 }
1380
1381 type = type_from_string_gently(type_buf, type_len, 1);
1382 if (oi->type_name)
1383 strbuf_add(oi->type_name, type_buf, type_len);
1384 if (oi->typep)
1385 *oi->typep = type;
1386
1387 /*
1388 * The length must follow immediately, and be in canonical
1389 * decimal format (ie "010" is not valid).
1390 */
1391 size = *hdr++ - '0';
1392 if (size > 9)
1393 return -1;
1394 if (size) {
1395 for (;;) {
1396 unsigned long c = *hdr - '0';
1397 if (c > 9)
1398 break;
1399 hdr++;
1400 size = st_add(st_mult(size, 10), c);
1401 }
1402 }
1403
1404 if (oi->sizep)
1405 *oi->sizep = cast_size_t_to_ulong(size);
1406
1407 /*
1408 * The length must be followed by a zero byte
1409 */
1410 if (*hdr)
1411 return -1;
1412
1413 /*
1414 * The format is valid, but the type may still be bogus. The
1415 * Caller needs to check its oi->typep.
1416 */
1417 return 0;
1418 }
1419
1420 static int loose_object_info(struct repository *r,
1421 const struct object_id *oid,
1422 struct object_info *oi, int flags)
1423 {
1424 int status = 0;
1425 unsigned long mapsize;
1426 void *map;
1427 git_zstream stream;
1428 char hdr[MAX_HEADER_LEN];
1429 struct strbuf hdrbuf = STRBUF_INIT;
1430 unsigned long size_scratch;
1431 enum object_type type_scratch;
1432 int allow_unknown = flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE;
1433
1434 if (oi->delta_base_oid)
1435 oidclr(oi->delta_base_oid);
1436
1437 /*
1438 * If we don't care about type or size, then we don't
1439 * need to look inside the object at all. Note that we
1440 * do not optimize out the stat call, even if the
1441 * caller doesn't care about the disk-size, since our
1442 * return value implicitly indicates whether the
1443 * object even exists.
1444 */
1445 if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1446 const char *path;
1447 struct stat st;
1448 if (!oi->disk_sizep && (flags & OBJECT_INFO_QUICK))
1449 return quick_has_loose(r, oid) ? 0 : -1;
1450 if (stat_loose_object(r, oid, &st, &path) < 0)
1451 return -1;
1452 if (oi->disk_sizep)
1453 *oi->disk_sizep = st.st_size;
1454 return 0;
1455 }
1456
1457 map = map_loose_object(r, oid, &mapsize);
1458 if (!map)
1459 return -1;
1460
1461 if (!oi->sizep)
1462 oi->sizep = &size_scratch;
1463 if (!oi->typep)
1464 oi->typep = &type_scratch;
1465
1466 if (oi->disk_sizep)
1467 *oi->disk_sizep = mapsize;
1468
1469 switch (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr),
1470 allow_unknown ? &hdrbuf : NULL)) {
1471 case ULHR_OK:
1472 if (parse_loose_header(hdrbuf.len ? hdrbuf.buf : hdr, oi) < 0)
1473 status = error(_("unable to parse %s header"), oid_to_hex(oid));
1474 else if (!allow_unknown && *oi->typep < 0)
1475 die(_("invalid object type"));
1476
1477 if (!oi->contentp)
1478 break;
1479 *oi->contentp = unpack_loose_rest(&stream, hdr, *oi->sizep, oid);
1480 if (*oi->contentp)
1481 goto cleanup;
1482
1483 status = -1;
1484 break;
1485 case ULHR_BAD:
1486 status = error(_("unable to unpack %s header"),
1487 oid_to_hex(oid));
1488 break;
1489 case ULHR_TOO_LONG:
1490 status = error(_("header for %s too long, exceeds %d bytes"),
1491 oid_to_hex(oid), MAX_HEADER_LEN);
1492 break;
1493 }
1494
1495 git_inflate_end(&stream);
1496 cleanup:
1497 munmap(map, mapsize);
1498 if (oi->sizep == &size_scratch)
1499 oi->sizep = NULL;
1500 strbuf_release(&hdrbuf);
1501 if (oi->typep == &type_scratch)
1502 oi->typep = NULL;
1503 oi->whence = OI_LOOSE;
1504 return status;
1505 }
1506
1507 int obj_read_use_lock = 0;
1508 pthread_mutex_t obj_read_mutex;
1509
1510 void enable_obj_read_lock(void)
1511 {
1512 if (obj_read_use_lock)
1513 return;
1514
1515 obj_read_use_lock = 1;
1516 init_recursive_mutex(&obj_read_mutex);
1517 }
1518
1519 void disable_obj_read_lock(void)
1520 {
1521 if (!obj_read_use_lock)
1522 return;
1523
1524 obj_read_use_lock = 0;
1525 pthread_mutex_destroy(&obj_read_mutex);
1526 }
1527
1528 int fetch_if_missing = 1;
1529
1530 static int do_oid_object_info_extended(struct repository *r,
1531 const struct object_id *oid,
1532 struct object_info *oi, unsigned flags)
1533 {
1534 static struct object_info blank_oi = OBJECT_INFO_INIT;
1535 struct cached_object *co;
1536 struct pack_entry e;
1537 int rtype;
1538 const struct object_id *real = oid;
1539 int already_retried = 0;
1540
1541
1542 if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1543 real = lookup_replace_object(r, oid);
1544
1545 if (is_null_oid(real))
1546 return -1;
1547
1548 if (!oi)
1549 oi = &blank_oi;
1550
1551 co = find_cached_object(real);
1552 if (co) {
1553 if (oi->typep)
1554 *(oi->typep) = co->type;
1555 if (oi->sizep)
1556 *(oi->sizep) = co->size;
1557 if (oi->disk_sizep)
1558 *(oi->disk_sizep) = 0;
1559 if (oi->delta_base_oid)
1560 oidclr(oi->delta_base_oid);
1561 if (oi->type_name)
1562 strbuf_addstr(oi->type_name, type_name(co->type));
1563 if (oi->contentp)
1564 *oi->contentp = xmemdupz(co->buf, co->size);
1565 oi->whence = OI_CACHED;
1566 return 0;
1567 }
1568
1569 while (1) {
1570 if (find_pack_entry(r, real, &e))
1571 break;
1572
1573 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1574 return -1;
1575
1576 /* Most likely it's a loose object. */
1577 if (!loose_object_info(r, real, oi, flags))
1578 return 0;
1579
1580 /* Not a loose object; someone else may have just packed it. */
1581 if (!(flags & OBJECT_INFO_QUICK)) {
1582 reprepare_packed_git(r);
1583 if (find_pack_entry(r, real, &e))
1584 break;
1585 }
1586
1587 /*
1588 * If r is the_repository, this might be an attempt at
1589 * accessing a submodule object as if it were in the_repository
1590 * (having called add_submodule_odb() on that submodule's ODB).
1591 * If any such ODBs exist, register them and try again.
1592 */
1593 if (r == the_repository &&
1594 register_all_submodule_odb_as_alternates())
1595 /* We added some alternates; retry */
1596 continue;
1597
1598 /* Check if it is a missing object */
1599 if (fetch_if_missing && repo_has_promisor_remote(r) &&
1600 !already_retried &&
1601 !(flags & OBJECT_INFO_SKIP_FETCH_OBJECT)) {
1602 /*
1603 * TODO Investigate checking promisor_remote_get_direct()
1604 * TODO return value and stopping on error here.
1605 */
1606 promisor_remote_get_direct(r, real, 1);
1607 already_retried = 1;
1608 continue;
1609 }
1610
1611 return -1;
1612 }
1613
1614 if (oi == &blank_oi)
1615 /*
1616 * We know that the caller doesn't actually need the
1617 * information below, so return early.
1618 */
1619 return 0;
1620 rtype = packed_object_info(r, e.p, e.offset, oi);
1621 if (rtype < 0) {
1622 mark_bad_packed_object(e.p, real);
1623 return do_oid_object_info_extended(r, real, oi, 0);
1624 } else if (oi->whence == OI_PACKED) {
1625 oi->u.packed.offset = e.offset;
1626 oi->u.packed.pack = e.p;
1627 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1628 rtype == OBJ_OFS_DELTA);
1629 }
1630
1631 return 0;
1632 }
1633
1634 int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1635 struct object_info *oi, unsigned flags)
1636 {
1637 int ret;
1638 obj_read_lock();
1639 ret = do_oid_object_info_extended(r, oid, oi, flags);
1640 obj_read_unlock();
1641 return ret;
1642 }
1643
1644
1645 /* returns enum object_type or negative */
1646 int oid_object_info(struct repository *r,
1647 const struct object_id *oid,
1648 unsigned long *sizep)
1649 {
1650 enum object_type type;
1651 struct object_info oi = OBJECT_INFO_INIT;
1652
1653 oi.typep = &type;
1654 oi.sizep = sizep;
1655 if (oid_object_info_extended(r, oid, &oi,
1656 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1657 return -1;
1658 return type;
1659 }
1660
1661 static void *read_object(struct repository *r,
1662 const struct object_id *oid, enum object_type *type,
1663 unsigned long *size)
1664 {
1665 struct object_info oi = OBJECT_INFO_INIT;
1666 void *content;
1667 oi.typep = type;
1668 oi.sizep = size;
1669 oi.contentp = &content;
1670
1671 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1672 return NULL;
1673 return content;
1674 }
1675
1676 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1677 struct object_id *oid)
1678 {
1679 struct cached_object *co;
1680
1681 hash_object_file(the_hash_algo, buf, len, type, oid);
1682 if (has_object_file_with_flags(oid, OBJECT_INFO_QUICK | OBJECT_INFO_SKIP_FETCH_OBJECT) ||
1683 find_cached_object(oid))
1684 return 0;
1685 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1686 co = &cached_objects[cached_object_nr++];
1687 co->size = len;
1688 co->type = type;
1689 co->buf = xmalloc(len);
1690 memcpy(co->buf, buf, len);
1691 oidcpy(&co->oid, oid);
1692 return 0;
1693 }
1694
1695 /*
1696 * This function dies on corrupt objects; the callers who want to
1697 * deal with them should arrange to call read_object() and give error
1698 * messages themselves.
1699 */
1700 void *read_object_file_extended(struct repository *r,
1701 const struct object_id *oid,
1702 enum object_type *type,
1703 unsigned long *size,
1704 int lookup_replace)
1705 {
1706 void *data;
1707 const struct packed_git *p;
1708 const char *path;
1709 struct stat st;
1710 const struct object_id *repl = lookup_replace ?
1711 lookup_replace_object(r, oid) : oid;
1712
1713 errno = 0;
1714 data = read_object(r, repl, type, size);
1715 if (data)
1716 return data;
1717
1718 obj_read_lock();
1719 if (errno && errno != ENOENT)
1720 die_errno(_("failed to read object %s"), oid_to_hex(oid));
1721
1722 /* die if we replaced an object with one that does not exist */
1723 if (repl != oid)
1724 die(_("replacement %s not found for %s"),
1725 oid_to_hex(repl), oid_to_hex(oid));
1726
1727 if (!stat_loose_object(r, repl, &st, &path))
1728 die(_("loose object %s (stored in %s) is corrupt"),
1729 oid_to_hex(repl), path);
1730
1731 if ((p = has_packed_and_bad(r, repl)))
1732 die(_("packed object %s (stored in %s) is corrupt"),
1733 oid_to_hex(repl), p->pack_name);
1734 obj_read_unlock();
1735
1736 return NULL;
1737 }
1738
1739 void *read_object_with_reference(struct repository *r,
1740 const struct object_id *oid,
1741 enum object_type required_type,
1742 unsigned long *size,
1743 struct object_id *actual_oid_return)
1744 {
1745 enum object_type type;
1746 void *buffer;
1747 unsigned long isize;
1748 struct object_id actual_oid;
1749
1750 oidcpy(&actual_oid, oid);
1751 while (1) {
1752 int ref_length = -1;
1753 const char *ref_type = NULL;
1754
1755 buffer = repo_read_object_file(r, &actual_oid, &type, &isize);
1756 if (!buffer)
1757 return NULL;
1758 if (type == required_type) {
1759 *size = isize;
1760 if (actual_oid_return)
1761 oidcpy(actual_oid_return, &actual_oid);
1762 return buffer;
1763 }
1764 /* Handle references */
1765 else if (type == OBJ_COMMIT)
1766 ref_type = "tree ";
1767 else if (type == OBJ_TAG)
1768 ref_type = "object ";
1769 else {
1770 free(buffer);
1771 return NULL;
1772 }
1773 ref_length = strlen(ref_type);
1774
1775 if (ref_length + the_hash_algo->hexsz > isize ||
1776 memcmp(buffer, ref_type, ref_length) ||
1777 get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1778 free(buffer);
1779 return NULL;
1780 }
1781 free(buffer);
1782 /* Now we have the ID of the referred-to object in
1783 * actual_oid. Check again. */
1784 }
1785 }
1786
1787 static void hash_object_body(const struct git_hash_algo *algo, git_hash_ctx *c,
1788 const void *buf, unsigned long len,
1789 struct object_id *oid,
1790 char *hdr, int *hdrlen)
1791 {
1792 algo->init_fn(c);
1793 algo->update_fn(c, hdr, *hdrlen);
1794 algo->update_fn(c, buf, len);
1795 algo->final_oid_fn(oid, c);
1796 }
1797
1798 static void write_object_file_prepare(const struct git_hash_algo *algo,
1799 const void *buf, unsigned long len,
1800 enum object_type type, struct object_id *oid,
1801 char *hdr, int *hdrlen)
1802 {
1803 git_hash_ctx c;
1804
1805 /* Generate the header */
1806 *hdrlen = format_object_header(hdr, *hdrlen, type, len);
1807
1808 /* Sha1.. */
1809 hash_object_body(algo, &c, buf, len, oid, hdr, hdrlen);
1810 }
1811
1812 static void write_object_file_prepare_literally(const struct git_hash_algo *algo,
1813 const void *buf, unsigned long len,
1814 const char *type, struct object_id *oid,
1815 char *hdr, int *hdrlen)
1816 {
1817 git_hash_ctx c;
1818
1819 *hdrlen = format_object_header_literally(hdr, *hdrlen, type, len);
1820 hash_object_body(algo, &c, buf, len, oid, hdr, hdrlen);
1821 }
1822
1823 /*
1824 * Move the just written object into its final resting place.
1825 */
1826 int finalize_object_file(const char *tmpfile, const char *filename)
1827 {
1828 int ret = 0;
1829
1830 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1831 goto try_rename;
1832 else if (link(tmpfile, filename))
1833 ret = errno;
1834
1835 /*
1836 * Coda hack - coda doesn't like cross-directory links,
1837 * so we fall back to a rename, which will mean that it
1838 * won't be able to check collisions, but that's not a
1839 * big deal.
1840 *
1841 * The same holds for FAT formatted media.
1842 *
1843 * When this succeeds, we just return. We have nothing
1844 * left to unlink.
1845 */
1846 if (ret && ret != EEXIST) {
1847 try_rename:
1848 if (!rename(tmpfile, filename))
1849 goto out;
1850 ret = errno;
1851 }
1852 unlink_or_warn(tmpfile);
1853 if (ret) {
1854 if (ret != EEXIST) {
1855 return error_errno(_("unable to write file %s"), filename);
1856 }
1857 /* FIXME!!! Collision check here ? */
1858 }
1859
1860 out:
1861 if (adjust_shared_perm(filename))
1862 return error(_("unable to set permission to '%s'"), filename);
1863 return 0;
1864 }
1865
1866 static int write_buffer(int fd, const void *buf, size_t len)
1867 {
1868 if (write_in_full(fd, buf, len) < 0)
1869 return error_errno(_("file write error"));
1870 return 0;
1871 }
1872
1873 static void hash_object_file_literally(const struct git_hash_algo *algo,
1874 const void *buf, unsigned long len,
1875 const char *type, struct object_id *oid)
1876 {
1877 char hdr[MAX_HEADER_LEN];
1878 int hdrlen = sizeof(hdr);
1879
1880 write_object_file_prepare_literally(algo, buf, len, type, oid, hdr, &hdrlen);
1881 }
1882
1883 void hash_object_file(const struct git_hash_algo *algo, const void *buf,
1884 unsigned long len, enum object_type type,
1885 struct object_id *oid)
1886 {
1887 hash_object_file_literally(algo, buf, len, type_name(type), oid);
1888 }
1889
1890 /* Finalize a file on disk, and close it. */
1891 static void close_loose_object(int fd, const char *filename)
1892 {
1893 if (the_repository->objects->odb->will_destroy)
1894 goto out;
1895
1896 if (batch_fsync_enabled(FSYNC_COMPONENT_LOOSE_OBJECT))
1897 fsync_loose_object_bulk_checkin(fd, filename);
1898 else if (fsync_object_files > 0)
1899 fsync_or_die(fd, filename);
1900 else
1901 fsync_component_or_die(FSYNC_COMPONENT_LOOSE_OBJECT, fd,
1902 filename);
1903
1904 out:
1905 if (close(fd) != 0)
1906 die_errno(_("error when closing loose object file"));
1907 }
1908
1909 /* Size of directory component, including the ending '/' */
1910 static inline int directory_size(const char *filename)
1911 {
1912 const char *s = strrchr(filename, '/');
1913 if (!s)
1914 return 0;
1915 return s - filename + 1;
1916 }
1917
1918 /*
1919 * This creates a temporary file in the same directory as the final
1920 * 'filename'
1921 *
1922 * We want to avoid cross-directory filename renames, because those
1923 * can have problems on various filesystems (FAT, NFS, Coda).
1924 */
1925 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1926 {
1927 int fd, dirlen = directory_size(filename);
1928
1929 strbuf_reset(tmp);
1930 strbuf_add(tmp, filename, dirlen);
1931 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1932 fd = git_mkstemp_mode(tmp->buf, 0444);
1933 if (fd < 0 && dirlen && errno == ENOENT) {
1934 /*
1935 * Make sure the directory exists; note that the contents
1936 * of the buffer are undefined after mkstemp returns an
1937 * error, so we have to rewrite the whole buffer from
1938 * scratch.
1939 */
1940 strbuf_reset(tmp);
1941 strbuf_add(tmp, filename, dirlen - 1);
1942 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1943 return -1;
1944 if (adjust_shared_perm(tmp->buf))
1945 return -1;
1946
1947 /* Try again */
1948 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1949 fd = git_mkstemp_mode(tmp->buf, 0444);
1950 }
1951 return fd;
1952 }
1953
1954 static int write_loose_object(const struct object_id *oid, char *hdr,
1955 int hdrlen, const void *buf, unsigned long len,
1956 time_t mtime, unsigned flags)
1957 {
1958 int fd, ret;
1959 unsigned char compressed[4096];
1960 git_zstream stream;
1961 git_hash_ctx c;
1962 struct object_id parano_oid;
1963 static struct strbuf tmp_file = STRBUF_INIT;
1964 static struct strbuf filename = STRBUF_INIT;
1965
1966 if (batch_fsync_enabled(FSYNC_COMPONENT_LOOSE_OBJECT))
1967 prepare_loose_object_bulk_checkin();
1968
1969 loose_object_path(the_repository, &filename, oid);
1970
1971 fd = create_tmpfile(&tmp_file, filename.buf);
1972 if (fd < 0) {
1973 if (flags & HASH_SILENT)
1974 return -1;
1975 else if (errno == EACCES)
1976 return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
1977 else
1978 return error_errno(_("unable to create temporary file"));
1979 }
1980
1981 /* Set it up */
1982 git_deflate_init(&stream, zlib_compression_level);
1983 stream.next_out = compressed;
1984 stream.avail_out = sizeof(compressed);
1985 the_hash_algo->init_fn(&c);
1986
1987 /* First header.. */
1988 stream.next_in = (unsigned char *)hdr;
1989 stream.avail_in = hdrlen;
1990 while (git_deflate(&stream, 0) == Z_OK)
1991 ; /* nothing */
1992 the_hash_algo->update_fn(&c, hdr, hdrlen);
1993
1994 /* Then the data itself.. */
1995 stream.next_in = (void *)buf;
1996 stream.avail_in = len;
1997 do {
1998 unsigned char *in0 = stream.next_in;
1999 ret = git_deflate(&stream, Z_FINISH);
2000 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
2001 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2002 die(_("unable to write loose object file"));
2003 stream.next_out = compressed;
2004 stream.avail_out = sizeof(compressed);
2005 } while (ret == Z_OK);
2006
2007 if (ret != Z_STREAM_END)
2008 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
2009 ret);
2010 ret = git_deflate_end_gently(&stream);
2011 if (ret != Z_OK)
2012 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
2013 ret);
2014 the_hash_algo->final_oid_fn(&parano_oid, &c);
2015 if (!oideq(oid, &parano_oid))
2016 die(_("confused by unstable object source data for %s"),
2017 oid_to_hex(oid));
2018
2019 close_loose_object(fd, tmp_file.buf);
2020
2021 if (mtime) {
2022 struct utimbuf utb;
2023 utb.actime = mtime;
2024 utb.modtime = mtime;
2025 if (utime(tmp_file.buf, &utb) < 0 &&
2026 !(flags & HASH_SILENT))
2027 warning_errno(_("failed utime() on %s"), tmp_file.buf);
2028 }
2029
2030 return finalize_object_file(tmp_file.buf, filename.buf);
2031 }
2032
2033 static int freshen_loose_object(const struct object_id *oid)
2034 {
2035 return check_and_freshen(oid, 1);
2036 }
2037
2038 static int freshen_packed_object(const struct object_id *oid)
2039 {
2040 struct pack_entry e;
2041 if (!find_pack_entry(the_repository, oid, &e))
2042 return 0;
2043 if (e.p->is_cruft)
2044 return 0;
2045 if (e.p->freshened)
2046 return 1;
2047 if (!freshen_file(e.p->pack_name))
2048 return 0;
2049 e.p->freshened = 1;
2050 return 1;
2051 }
2052
2053 int write_object_file_flags(const void *buf, unsigned long len,
2054 enum object_type type, struct object_id *oid,
2055 unsigned flags)
2056 {
2057 char hdr[MAX_HEADER_LEN];
2058 int hdrlen = sizeof(hdr);
2059
2060 /* Normally if we have it in the pack then we do not bother writing
2061 * it out into .git/objects/??/?{38} file.
2062 */
2063 write_object_file_prepare(the_hash_algo, buf, len, type, oid, hdr,
2064 &hdrlen);
2065 if (freshen_packed_object(oid) || freshen_loose_object(oid))
2066 return 0;
2067 return write_loose_object(oid, hdr, hdrlen, buf, len, 0, flags);
2068 }
2069
2070 int write_object_file_literally(const void *buf, unsigned long len,
2071 const char *type, struct object_id *oid,
2072 unsigned flags)
2073 {
2074 char *header;
2075 int hdrlen, status = 0;
2076
2077 /* type string, SP, %lu of the length plus NUL must fit this */
2078 hdrlen = strlen(type) + MAX_HEADER_LEN;
2079 header = xmalloc(hdrlen);
2080 write_object_file_prepare_literally(the_hash_algo, buf, len, type,
2081 oid, header, &hdrlen);
2082
2083 if (!(flags & HASH_WRITE_OBJECT))
2084 goto cleanup;
2085 if (freshen_packed_object(oid) || freshen_loose_object(oid))
2086 goto cleanup;
2087 status = write_loose_object(oid, header, hdrlen, buf, len, 0, 0);
2088
2089 cleanup:
2090 free(header);
2091 return status;
2092 }
2093
2094 int force_object_loose(const struct object_id *oid, time_t mtime)
2095 {
2096 void *buf;
2097 unsigned long len;
2098 enum object_type type;
2099 char hdr[MAX_HEADER_LEN];
2100 int hdrlen;
2101 int ret;
2102
2103 if (has_loose_object(oid))
2104 return 0;
2105 buf = read_object(the_repository, oid, &type, &len);
2106 if (!buf)
2107 return error(_("cannot read object for %s"), oid_to_hex(oid));
2108 hdrlen = format_object_header(hdr, sizeof(hdr), type, len);
2109 ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime, 0);
2110 free(buf);
2111
2112 return ret;
2113 }
2114
2115 int has_object(struct repository *r, const struct object_id *oid,
2116 unsigned flags)
2117 {
2118 int quick = !(flags & HAS_OBJECT_RECHECK_PACKED);
2119 unsigned object_info_flags = OBJECT_INFO_SKIP_FETCH_OBJECT |
2120 (quick ? OBJECT_INFO_QUICK : 0);
2121
2122 if (!startup_info->have_repository)
2123 return 0;
2124 return oid_object_info_extended(r, oid, NULL, object_info_flags) >= 0;
2125 }
2126
2127 int repo_has_object_file_with_flags(struct repository *r,
2128 const struct object_id *oid, int flags)
2129 {
2130 if (!startup_info->have_repository)
2131 return 0;
2132 return oid_object_info_extended(r, oid, NULL, flags) >= 0;
2133 }
2134
2135 int repo_has_object_file(struct repository *r,
2136 const struct object_id *oid)
2137 {
2138 return repo_has_object_file_with_flags(r, oid, 0);
2139 }
2140
2141 static void check_tree(const void *buf, size_t size)
2142 {
2143 struct tree_desc desc;
2144 struct name_entry entry;
2145
2146 init_tree_desc(&desc, buf, size);
2147 while (tree_entry(&desc, &entry))
2148 /* do nothing
2149 * tree_entry() will die() on malformed entries */
2150 ;
2151 }
2152
2153 static void check_commit(const void *buf, size_t size)
2154 {
2155 struct commit c;
2156 memset(&c, 0, sizeof(c));
2157 if (parse_commit_buffer(the_repository, &c, buf, size, 0))
2158 die(_("corrupt commit"));
2159 }
2160
2161 static void check_tag(const void *buf, size_t size)
2162 {
2163 struct tag t;
2164 memset(&t, 0, sizeof(t));
2165 if (parse_tag_buffer(the_repository, &t, buf, size))
2166 die(_("corrupt tag"));
2167 }
2168
2169 static int index_mem(struct index_state *istate,
2170 struct object_id *oid, void *buf, size_t size,
2171 enum object_type type,
2172 const char *path, unsigned flags)
2173 {
2174 int ret = 0;
2175 int re_allocated = 0;
2176 int write_object = flags & HASH_WRITE_OBJECT;
2177
2178 if (!type)
2179 type = OBJ_BLOB;
2180
2181 /*
2182 * Convert blobs to git internal format
2183 */
2184 if ((type == OBJ_BLOB) && path) {
2185 struct strbuf nbuf = STRBUF_INIT;
2186 if (convert_to_git(istate, path, buf, size, &nbuf,
2187 get_conv_flags(flags))) {
2188 buf = strbuf_detach(&nbuf, &size);
2189 re_allocated = 1;
2190 }
2191 }
2192 if (flags & HASH_FORMAT_CHECK) {
2193 if (type == OBJ_TREE)
2194 check_tree(buf, size);
2195 if (type == OBJ_COMMIT)
2196 check_commit(buf, size);
2197 if (type == OBJ_TAG)
2198 check_tag(buf, size);
2199 }
2200
2201 if (write_object)
2202 ret = write_object_file(buf, size, type, oid);
2203 else
2204 hash_object_file(the_hash_algo, buf, size, type, oid);
2205 if (re_allocated)
2206 free(buf);
2207 return ret;
2208 }
2209
2210 static int index_stream_convert_blob(struct index_state *istate,
2211 struct object_id *oid,
2212 int fd,
2213 const char *path,
2214 unsigned flags)
2215 {
2216 int ret = 0;
2217 const int write_object = flags & HASH_WRITE_OBJECT;
2218 struct strbuf sbuf = STRBUF_INIT;
2219
2220 assert(path);
2221 assert(would_convert_to_git_filter_fd(istate, path));
2222
2223 convert_to_git_filter_fd(istate, path, fd, &sbuf,
2224 get_conv_flags(flags));
2225
2226 if (write_object)
2227 ret = write_object_file(sbuf.buf, sbuf.len, OBJ_BLOB,
2228 oid);
2229 else
2230 hash_object_file(the_hash_algo, sbuf.buf, sbuf.len, OBJ_BLOB,
2231 oid);
2232 strbuf_release(&sbuf);
2233 return ret;
2234 }
2235
2236 static int index_pipe(struct index_state *istate, struct object_id *oid,
2237 int fd, enum object_type type,
2238 const char *path, unsigned flags)
2239 {
2240 struct strbuf sbuf = STRBUF_INIT;
2241 int ret;
2242
2243 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2244 ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags);
2245 else
2246 ret = -1;
2247 strbuf_release(&sbuf);
2248 return ret;
2249 }
2250
2251 #define SMALL_FILE_SIZE (32*1024)
2252
2253 static int index_core(struct index_state *istate,
2254 struct object_id *oid, int fd, size_t size,
2255 enum object_type type, const char *path,
2256 unsigned flags)
2257 {
2258 int ret;
2259
2260 if (!size) {
2261 ret = index_mem(istate, oid, "", size, type, path, flags);
2262 } else if (size <= SMALL_FILE_SIZE) {
2263 char *buf = xmalloc(size);
2264 ssize_t read_result = read_in_full(fd, buf, size);
2265 if (read_result < 0)
2266 ret = error_errno(_("read error while indexing %s"),
2267 path ? path : "<unknown>");
2268 else if (read_result != size)
2269 ret = error(_("short read while indexing %s"),
2270 path ? path : "<unknown>");
2271 else
2272 ret = index_mem(istate, oid, buf, size, type, path, flags);
2273 free(buf);
2274 } else {
2275 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2276 ret = index_mem(istate, oid, buf, size, type, path, flags);
2277 munmap(buf, size);
2278 }
2279 return ret;
2280 }
2281
2282 /*
2283 * This creates one packfile per large blob unless bulk-checkin
2284 * machinery is "plugged".
2285 *
2286 * This also bypasses the usual "convert-to-git" dance, and that is on
2287 * purpose. We could write a streaming version of the converting
2288 * functions and insert that before feeding the data to fast-import
2289 * (or equivalent in-core API described above). However, that is
2290 * somewhat complicated, as we do not know the size of the filter
2291 * result, which we need to know beforehand when writing a git object.
2292 * Since the primary motivation for trying to stream from the working
2293 * tree file and to avoid mmaping it in core is to deal with large
2294 * binary blobs, they generally do not want to get any conversion, and
2295 * callers should avoid this code path when filters are requested.
2296 */
2297 static int index_stream(struct object_id *oid, int fd, size_t size,
2298 enum object_type type, const char *path,
2299 unsigned flags)
2300 {
2301 return index_bulk_checkin(oid, fd, size, type, path, flags);
2302 }
2303
2304 int index_fd(struct index_state *istate, struct object_id *oid,
2305 int fd, struct stat *st,
2306 enum object_type type, const char *path, unsigned flags)
2307 {
2308 int ret;
2309
2310 /*
2311 * Call xsize_t() only when needed to avoid potentially unnecessary
2312 * die() for large files.
2313 */
2314 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, path))
2315 ret = index_stream_convert_blob(istate, oid, fd, path, flags);
2316 else if (!S_ISREG(st->st_mode))
2317 ret = index_pipe(istate, oid, fd, type, path, flags);
2318 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
2319 (path && would_convert_to_git(istate, path)))
2320 ret = index_core(istate, oid, fd, xsize_t(st->st_size),
2321 type, path, flags);
2322 else
2323 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
2324 flags);
2325 close(fd);
2326 return ret;
2327 }
2328
2329 int index_path(struct index_state *istate, struct object_id *oid,
2330 const char *path, struct stat *st, unsigned flags)
2331 {
2332 int fd;
2333 struct strbuf sb = STRBUF_INIT;
2334 int rc = 0;
2335
2336 switch (st->st_mode & S_IFMT) {
2337 case S_IFREG:
2338 fd = open(path, O_RDONLY);
2339 if (fd < 0)
2340 return error_errno("open(\"%s\")", path);
2341 if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0)
2342 return error(_("%s: failed to insert into database"),
2343 path);
2344 break;
2345 case S_IFLNK:
2346 if (strbuf_readlink(&sb, path, st->st_size))
2347 return error_errno("readlink(\"%s\")", path);
2348 if (!(flags & HASH_WRITE_OBJECT))
2349 hash_object_file(the_hash_algo, sb.buf, sb.len,
2350 OBJ_BLOB, oid);
2351 else if (write_object_file(sb.buf, sb.len, OBJ_BLOB, oid))
2352 rc = error(_("%s: failed to insert into database"), path);
2353 strbuf_release(&sb);
2354 break;
2355 case S_IFDIR:
2356 return resolve_gitlink_ref(path, "HEAD", oid);
2357 default:
2358 return error(_("%s: unsupported file type"), path);
2359 }
2360 return rc;
2361 }
2362
2363 int read_pack_header(int fd, struct pack_header *header)
2364 {
2365 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
2366 /* "eof before pack header was fully read" */
2367 return PH_ERROR_EOF;
2368
2369 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2370 /* "protocol error (pack signature mismatch detected)" */
2371 return PH_ERROR_PACK_SIGNATURE;
2372 if (!pack_version_ok(header->hdr_version))
2373 /* "protocol error (pack version unsupported)" */
2374 return PH_ERROR_PROTOCOL;
2375 return 0;
2376 }
2377
2378 void assert_oid_type(const struct object_id *oid, enum object_type expect)
2379 {
2380 enum object_type type = oid_object_info(the_repository, oid, NULL);
2381 if (type < 0)
2382 die(_("%s is not a valid object"), oid_to_hex(oid));
2383 if (type != expect)
2384 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
2385 type_name(expect));
2386 }
2387
2388 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2389 struct strbuf *path,
2390 each_loose_object_fn obj_cb,
2391 each_loose_cruft_fn cruft_cb,
2392 each_loose_subdir_fn subdir_cb,
2393 void *data)
2394 {
2395 size_t origlen, baselen;
2396 DIR *dir;
2397 struct dirent *de;
2398 int r = 0;
2399 struct object_id oid;
2400
2401 if (subdir_nr > 0xff)
2402 BUG("invalid loose object subdirectory: %x", subdir_nr);
2403
2404 origlen = path->len;
2405 strbuf_complete(path, '/');
2406 strbuf_addf(path, "%02x", subdir_nr);
2407
2408 dir = opendir(path->buf);
2409 if (!dir) {
2410 if (errno != ENOENT)
2411 r = error_errno(_("unable to open %s"), path->buf);
2412 strbuf_setlen(path, origlen);
2413 return r;
2414 }
2415
2416 oid.hash[0] = subdir_nr;
2417 strbuf_addch(path, '/');
2418 baselen = path->len;
2419
2420 while ((de = readdir_skip_dot_and_dotdot(dir))) {
2421 size_t namelen;
2422
2423 namelen = strlen(de->d_name);
2424 strbuf_setlen(path, baselen);
2425 strbuf_add(path, de->d_name, namelen);
2426 if (namelen == the_hash_algo->hexsz - 2 &&
2427 !hex_to_bytes(oid.hash + 1, de->d_name,
2428 the_hash_algo->rawsz - 1)) {
2429 oid_set_algo(&oid, the_hash_algo);
2430 if (obj_cb) {
2431 r = obj_cb(&oid, path->buf, data);
2432 if (r)
2433 break;
2434 }
2435 continue;
2436 }
2437
2438 if (cruft_cb) {
2439 r = cruft_cb(de->d_name, path->buf, data);
2440 if (r)
2441 break;
2442 }
2443 }
2444 closedir(dir);
2445
2446 strbuf_setlen(path, baselen - 1);
2447 if (!r && subdir_cb)
2448 r = subdir_cb(subdir_nr, path->buf, data);
2449
2450 strbuf_setlen(path, origlen);
2451
2452 return r;
2453 }
2454
2455 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2456 each_loose_object_fn obj_cb,
2457 each_loose_cruft_fn cruft_cb,
2458 each_loose_subdir_fn subdir_cb,
2459 void *data)
2460 {
2461 int r = 0;
2462 int i;
2463
2464 for (i = 0; i < 256; i++) {
2465 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2466 subdir_cb, data);
2467 if (r)
2468 break;
2469 }
2470
2471 return r;
2472 }
2473
2474 int for_each_loose_file_in_objdir(const char *path,
2475 each_loose_object_fn obj_cb,
2476 each_loose_cruft_fn cruft_cb,
2477 each_loose_subdir_fn subdir_cb,
2478 void *data)
2479 {
2480 struct strbuf buf = STRBUF_INIT;
2481 int r;
2482
2483 strbuf_addstr(&buf, path);
2484 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2485 subdir_cb, data);
2486 strbuf_release(&buf);
2487
2488 return r;
2489 }
2490
2491 int for_each_loose_object(each_loose_object_fn cb, void *data,
2492 enum for_each_object_flags flags)
2493 {
2494 struct object_directory *odb;
2495
2496 prepare_alt_odb(the_repository);
2497 for (odb = the_repository->objects->odb; odb; odb = odb->next) {
2498 int r = for_each_loose_file_in_objdir(odb->path, cb, NULL,
2499 NULL, data);
2500 if (r)
2501 return r;
2502
2503 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2504 break;
2505 }
2506
2507 return 0;
2508 }
2509
2510 static int append_loose_object(const struct object_id *oid, const char *path,
2511 void *data)
2512 {
2513 oidtree_insert(data, oid);
2514 return 0;
2515 }
2516
2517 struct oidtree *odb_loose_cache(struct object_directory *odb,
2518 const struct object_id *oid)
2519 {
2520 int subdir_nr = oid->hash[0];
2521 struct strbuf buf = STRBUF_INIT;
2522 size_t word_bits = bitsizeof(odb->loose_objects_subdir_seen[0]);
2523 size_t word_index = subdir_nr / word_bits;
2524 size_t mask = (size_t)1u << (subdir_nr % word_bits);
2525 uint32_t *bitmap;
2526
2527 if (subdir_nr < 0 ||
2528 subdir_nr >= bitsizeof(odb->loose_objects_subdir_seen))
2529 BUG("subdir_nr out of range");
2530
2531 bitmap = &odb->loose_objects_subdir_seen[word_index];
2532 if (*bitmap & mask)
2533 return odb->loose_objects_cache;
2534 if (!odb->loose_objects_cache) {
2535 ALLOC_ARRAY(odb->loose_objects_cache, 1);
2536 oidtree_init(odb->loose_objects_cache);
2537 }
2538 strbuf_addstr(&buf, odb->path);
2539 for_each_file_in_obj_subdir(subdir_nr, &buf,
2540 append_loose_object,
2541 NULL, NULL,
2542 odb->loose_objects_cache);
2543 *bitmap |= mask;
2544 strbuf_release(&buf);
2545 return odb->loose_objects_cache;
2546 }
2547
2548 void odb_clear_loose_cache(struct object_directory *odb)
2549 {
2550 oidtree_clear(odb->loose_objects_cache);
2551 FREE_AND_NULL(odb->loose_objects_cache);
2552 memset(&odb->loose_objects_subdir_seen, 0,
2553 sizeof(odb->loose_objects_subdir_seen));
2554 }
2555
2556 static int check_stream_oid(git_zstream *stream,
2557 const char *hdr,
2558 unsigned long size,
2559 const char *path,
2560 const struct object_id *expected_oid)
2561 {
2562 git_hash_ctx c;
2563 struct object_id real_oid;
2564 unsigned char buf[4096];
2565 unsigned long total_read;
2566 int status = Z_OK;
2567
2568 the_hash_algo->init_fn(&c);
2569 the_hash_algo->update_fn(&c, hdr, stream->total_out);
2570
2571 /*
2572 * We already read some bytes into hdr, but the ones up to the NUL
2573 * do not count against the object's content size.
2574 */
2575 total_read = stream->total_out - strlen(hdr) - 1;
2576
2577 /*
2578 * This size comparison must be "<=" to read the final zlib packets;
2579 * see the comment in unpack_loose_rest for details.
2580 */
2581 while (total_read <= size &&
2582 (status == Z_OK ||
2583 (status == Z_BUF_ERROR && !stream->avail_out))) {
2584 stream->next_out = buf;
2585 stream->avail_out = sizeof(buf);
2586 if (size - total_read < stream->avail_out)
2587 stream->avail_out = size - total_read;
2588 status = git_inflate(stream, Z_FINISH);
2589 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2590 total_read += stream->next_out - buf;
2591 }
2592 git_inflate_end(stream);
2593
2594 if (status != Z_STREAM_END) {
2595 error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid));
2596 return -1;
2597 }
2598 if (stream->avail_in) {
2599 error(_("garbage at end of loose object '%s'"),
2600 oid_to_hex(expected_oid));
2601 return -1;
2602 }
2603
2604 the_hash_algo->final_oid_fn(&real_oid, &c);
2605 if (!oideq(expected_oid, &real_oid)) {
2606 error(_("hash mismatch for %s (expected %s)"), path,
2607 oid_to_hex(expected_oid));
2608 return -1;
2609 }
2610
2611 return 0;
2612 }
2613
2614 int read_loose_object(const char *path,
2615 const struct object_id *expected_oid,
2616 struct object_id *real_oid,
2617 void **contents,
2618 struct object_info *oi)
2619 {
2620 int ret = -1;
2621 void *map = NULL;
2622 unsigned long mapsize;
2623 git_zstream stream;
2624 char hdr[MAX_HEADER_LEN];
2625 unsigned long *size = oi->sizep;
2626
2627 map = map_loose_object_1(the_repository, path, NULL, &mapsize);
2628 if (!map) {
2629 error_errno(_("unable to mmap %s"), path);
2630 goto out;
2631 }
2632
2633 if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr),
2634 NULL) != ULHR_OK) {
2635 error(_("unable to unpack header of %s"), path);
2636 goto out;
2637 }
2638
2639 if (parse_loose_header(hdr, oi) < 0) {
2640 error(_("unable to parse header of %s"), path);
2641 git_inflate_end(&stream);
2642 goto out;
2643 }
2644
2645 if (*oi->typep == OBJ_BLOB && *size > big_file_threshold) {
2646 if (check_stream_oid(&stream, hdr, *size, path, expected_oid) < 0)
2647 goto out;
2648 } else {
2649 *contents = unpack_loose_rest(&stream, hdr, *size, expected_oid);
2650 if (!*contents) {
2651 error(_("unable to unpack contents of %s"), path);
2652 git_inflate_end(&stream);
2653 goto out;
2654 }
2655 hash_object_file_literally(the_repository->hash_algo,
2656 *contents, *size,
2657 oi->type_name->buf, real_oid);
2658 if (!oideq(expected_oid, real_oid))
2659 goto out;
2660 }
2661
2662 ret = 0; /* everything checks out */
2663
2664 out:
2665 if (map)
2666 munmap(map, mapsize);
2667 return ret;
2668 }