]> git.ipfire.org Git - thirdparty/git.git/blob - object-file.c
Merge branch 'jk/typofix'
[thirdparty/git.git] / object-file.c
1 /*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This handles basic git object files - packing, unpacking,
7 * creation etc.
8 */
9 #include "cache.h"
10 #include "config.h"
11 #include "string-list.h"
12 #include "lockfile.h"
13 #include "delta.h"
14 #include "pack.h"
15 #include "blob.h"
16 #include "commit.h"
17 #include "run-command.h"
18 #include "tag.h"
19 #include "tree.h"
20 #include "tree-walk.h"
21 #include "refs.h"
22 #include "pack-revindex.h"
23 #include "hash-lookup.h"
24 #include "bulk-checkin.h"
25 #include "repository.h"
26 #include "replace-object.h"
27 #include "streaming.h"
28 #include "dir.h"
29 #include "list.h"
30 #include "mergesort.h"
31 #include "quote.h"
32 #include "packfile.h"
33 #include "object-store.h"
34 #include "promisor-remote.h"
35
36 /* The maximum size for an object header. */
37 #define MAX_HEADER_LEN 32
38
39
40 #define EMPTY_TREE_SHA1_BIN_LITERAL \
41 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60" \
42 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04"
43 #define EMPTY_TREE_SHA256_BIN_LITERAL \
44 "\x6e\xf1\x9b\x41\x22\x5c\x53\x69\xf1\xc1" \
45 "\x04\xd4\x5d\x8d\x85\xef\xa9\xb0\x57\xb5" \
46 "\x3b\x14\xb4\xb9\xb9\x39\xdd\x74\xde\xcc" \
47 "\x53\x21"
48
49 #define EMPTY_BLOB_SHA1_BIN_LITERAL \
50 "\xe6\x9d\xe2\x9b\xb2\xd1\xd6\x43\x4b\x8b" \
51 "\x29\xae\x77\x5a\xd8\xc2\xe4\x8c\x53\x91"
52 #define EMPTY_BLOB_SHA256_BIN_LITERAL \
53 "\x47\x3a\x0f\x4c\x3b\xe8\xa9\x36\x81\xa2" \
54 "\x67\xe3\xb1\xe9\xa7\xdc\xda\x11\x85\x43" \
55 "\x6f\xe1\x41\xf7\x74\x91\x20\xa3\x03\x72" \
56 "\x18\x13"
57
58 static const struct object_id empty_tree_oid = {
59 .hash = EMPTY_TREE_SHA1_BIN_LITERAL,
60 .algo = GIT_HASH_SHA1,
61 };
62 static const struct object_id empty_blob_oid = {
63 .hash = EMPTY_BLOB_SHA1_BIN_LITERAL,
64 .algo = GIT_HASH_SHA1,
65 };
66 static const struct object_id null_oid_sha1 = {
67 .hash = {0},
68 .algo = GIT_HASH_SHA1,
69 };
70 static const struct object_id empty_tree_oid_sha256 = {
71 .hash = EMPTY_TREE_SHA256_BIN_LITERAL,
72 .algo = GIT_HASH_SHA256,
73 };
74 static const struct object_id empty_blob_oid_sha256 = {
75 .hash = EMPTY_BLOB_SHA256_BIN_LITERAL,
76 .algo = GIT_HASH_SHA256,
77 };
78 static const struct object_id null_oid_sha256 = {
79 .hash = {0},
80 .algo = GIT_HASH_SHA256,
81 };
82
83 static void git_hash_sha1_init(git_hash_ctx *ctx)
84 {
85 git_SHA1_Init(&ctx->sha1);
86 }
87
88 static void git_hash_sha1_clone(git_hash_ctx *dst, const git_hash_ctx *src)
89 {
90 git_SHA1_Clone(&dst->sha1, &src->sha1);
91 }
92
93 static void git_hash_sha1_update(git_hash_ctx *ctx, const void *data, size_t len)
94 {
95 git_SHA1_Update(&ctx->sha1, data, len);
96 }
97
98 static void git_hash_sha1_final(unsigned char *hash, git_hash_ctx *ctx)
99 {
100 git_SHA1_Final(hash, &ctx->sha1);
101 }
102
103 static void git_hash_sha1_final_oid(struct object_id *oid, git_hash_ctx *ctx)
104 {
105 git_SHA1_Final(oid->hash, &ctx->sha1);
106 memset(oid->hash + GIT_SHA1_RAWSZ, 0, GIT_MAX_RAWSZ - GIT_SHA1_RAWSZ);
107 oid->algo = GIT_HASH_SHA1;
108 }
109
110
111 static void git_hash_sha256_init(git_hash_ctx *ctx)
112 {
113 git_SHA256_Init(&ctx->sha256);
114 }
115
116 static void git_hash_sha256_clone(git_hash_ctx *dst, const git_hash_ctx *src)
117 {
118 git_SHA256_Clone(&dst->sha256, &src->sha256);
119 }
120
121 static void git_hash_sha256_update(git_hash_ctx *ctx, const void *data, size_t len)
122 {
123 git_SHA256_Update(&ctx->sha256, data, len);
124 }
125
126 static void git_hash_sha256_final(unsigned char *hash, git_hash_ctx *ctx)
127 {
128 git_SHA256_Final(hash, &ctx->sha256);
129 }
130
131 static void git_hash_sha256_final_oid(struct object_id *oid, git_hash_ctx *ctx)
132 {
133 git_SHA256_Final(oid->hash, &ctx->sha256);
134 /*
135 * This currently does nothing, so the compiler should optimize it out,
136 * but keep it in case we extend the hash size again.
137 */
138 memset(oid->hash + GIT_SHA256_RAWSZ, 0, GIT_MAX_RAWSZ - GIT_SHA256_RAWSZ);
139 oid->algo = GIT_HASH_SHA256;
140 }
141
142 static void git_hash_unknown_init(git_hash_ctx *ctx)
143 {
144 BUG("trying to init unknown hash");
145 }
146
147 static void git_hash_unknown_clone(git_hash_ctx *dst, const git_hash_ctx *src)
148 {
149 BUG("trying to clone unknown hash");
150 }
151
152 static void git_hash_unknown_update(git_hash_ctx *ctx, const void *data, size_t len)
153 {
154 BUG("trying to update unknown hash");
155 }
156
157 static void git_hash_unknown_final(unsigned char *hash, git_hash_ctx *ctx)
158 {
159 BUG("trying to finalize unknown hash");
160 }
161
162 static void git_hash_unknown_final_oid(struct object_id *oid, git_hash_ctx *ctx)
163 {
164 BUG("trying to finalize unknown hash");
165 }
166
167
168 const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
169 {
170 NULL,
171 0x00000000,
172 0,
173 0,
174 0,
175 git_hash_unknown_init,
176 git_hash_unknown_clone,
177 git_hash_unknown_update,
178 git_hash_unknown_final,
179 git_hash_unknown_final_oid,
180 NULL,
181 NULL,
182 NULL,
183 },
184 {
185 "sha1",
186 /* "sha1", big-endian */
187 0x73686131,
188 GIT_SHA1_RAWSZ,
189 GIT_SHA1_HEXSZ,
190 GIT_SHA1_BLKSZ,
191 git_hash_sha1_init,
192 git_hash_sha1_clone,
193 git_hash_sha1_update,
194 git_hash_sha1_final,
195 git_hash_sha1_final_oid,
196 &empty_tree_oid,
197 &empty_blob_oid,
198 &null_oid_sha1,
199 },
200 {
201 "sha256",
202 /* "s256", big-endian */
203 0x73323536,
204 GIT_SHA256_RAWSZ,
205 GIT_SHA256_HEXSZ,
206 GIT_SHA256_BLKSZ,
207 git_hash_sha256_init,
208 git_hash_sha256_clone,
209 git_hash_sha256_update,
210 git_hash_sha256_final,
211 git_hash_sha256_final_oid,
212 &empty_tree_oid_sha256,
213 &empty_blob_oid_sha256,
214 &null_oid_sha256,
215 }
216 };
217
218 const struct object_id *null_oid(void)
219 {
220 return the_hash_algo->null_oid;
221 }
222
223 const char *empty_tree_oid_hex(void)
224 {
225 static char buf[GIT_MAX_HEXSZ + 1];
226 return oid_to_hex_r(buf, the_hash_algo->empty_tree);
227 }
228
229 const char *empty_blob_oid_hex(void)
230 {
231 static char buf[GIT_MAX_HEXSZ + 1];
232 return oid_to_hex_r(buf, the_hash_algo->empty_blob);
233 }
234
235 int hash_algo_by_name(const char *name)
236 {
237 int i;
238 if (!name)
239 return GIT_HASH_UNKNOWN;
240 for (i = 1; i < GIT_HASH_NALGOS; i++)
241 if (!strcmp(name, hash_algos[i].name))
242 return i;
243 return GIT_HASH_UNKNOWN;
244 }
245
246 int hash_algo_by_id(uint32_t format_id)
247 {
248 int i;
249 for (i = 1; i < GIT_HASH_NALGOS; i++)
250 if (format_id == hash_algos[i].format_id)
251 return i;
252 return GIT_HASH_UNKNOWN;
253 }
254
255 int hash_algo_by_length(int len)
256 {
257 int i;
258 for (i = 1; i < GIT_HASH_NALGOS; i++)
259 if (len == hash_algos[i].rawsz)
260 return i;
261 return GIT_HASH_UNKNOWN;
262 }
263
264 /*
265 * This is meant to hold a *small* number of objects that you would
266 * want read_object_file() to be able to return, but yet you do not want
267 * to write them into the object store (e.g. a browse-only
268 * application).
269 */
270 static struct cached_object {
271 struct object_id oid;
272 enum object_type type;
273 void *buf;
274 unsigned long size;
275 } *cached_objects;
276 static int cached_object_nr, cached_object_alloc;
277
278 static struct cached_object empty_tree = {
279 { EMPTY_TREE_SHA1_BIN_LITERAL },
280 OBJ_TREE,
281 "",
282 0
283 };
284
285 static struct cached_object *find_cached_object(const struct object_id *oid)
286 {
287 int i;
288 struct cached_object *co = cached_objects;
289
290 for (i = 0; i < cached_object_nr; i++, co++) {
291 if (oideq(&co->oid, oid))
292 return co;
293 }
294 if (oideq(oid, the_hash_algo->empty_tree))
295 return &empty_tree;
296 return NULL;
297 }
298
299
300 static int get_conv_flags(unsigned flags)
301 {
302 if (flags & HASH_RENORMALIZE)
303 return CONV_EOL_RENORMALIZE;
304 else if (flags & HASH_WRITE_OBJECT)
305 return global_conv_flags_eol | CONV_WRITE_OBJECT;
306 else
307 return 0;
308 }
309
310
311 int mkdir_in_gitdir(const char *path)
312 {
313 if (mkdir(path, 0777)) {
314 int saved_errno = errno;
315 struct stat st;
316 struct strbuf sb = STRBUF_INIT;
317
318 if (errno != EEXIST)
319 return -1;
320 /*
321 * Are we looking at a path in a symlinked worktree
322 * whose original repository does not yet have it?
323 * e.g. .git/rr-cache pointing at its original
324 * repository in which the user hasn't performed any
325 * conflict resolution yet?
326 */
327 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
328 strbuf_readlink(&sb, path, st.st_size) ||
329 !is_absolute_path(sb.buf) ||
330 mkdir(sb.buf, 0777)) {
331 strbuf_release(&sb);
332 errno = saved_errno;
333 return -1;
334 }
335 strbuf_release(&sb);
336 }
337 return adjust_shared_perm(path);
338 }
339
340 static enum scld_error safe_create_leading_directories_1(char *path, int share)
341 {
342 char *next_component = path + offset_1st_component(path);
343 enum scld_error ret = SCLD_OK;
344
345 while (ret == SCLD_OK && next_component) {
346 struct stat st;
347 char *slash = next_component, slash_character;
348
349 while (*slash && !is_dir_sep(*slash))
350 slash++;
351
352 if (!*slash)
353 break;
354
355 next_component = slash + 1;
356 while (is_dir_sep(*next_component))
357 next_component++;
358 if (!*next_component)
359 break;
360
361 slash_character = *slash;
362 *slash = '\0';
363 if (!stat(path, &st)) {
364 /* path exists */
365 if (!S_ISDIR(st.st_mode)) {
366 errno = ENOTDIR;
367 ret = SCLD_EXISTS;
368 }
369 } else if (mkdir(path, 0777)) {
370 if (errno == EEXIST &&
371 !stat(path, &st) && S_ISDIR(st.st_mode))
372 ; /* somebody created it since we checked */
373 else if (errno == ENOENT)
374 /*
375 * Either mkdir() failed because
376 * somebody just pruned the containing
377 * directory, or stat() failed because
378 * the file that was in our way was
379 * just removed. Either way, inform
380 * the caller that it might be worth
381 * trying again:
382 */
383 ret = SCLD_VANISHED;
384 else
385 ret = SCLD_FAILED;
386 } else if (share && adjust_shared_perm(path)) {
387 ret = SCLD_PERMS;
388 }
389 *slash = slash_character;
390 }
391 return ret;
392 }
393
394 enum scld_error safe_create_leading_directories(char *path)
395 {
396 return safe_create_leading_directories_1(path, 1);
397 }
398
399 enum scld_error safe_create_leading_directories_no_share(char *path)
400 {
401 return safe_create_leading_directories_1(path, 0);
402 }
403
404 enum scld_error safe_create_leading_directories_const(const char *path)
405 {
406 int save_errno;
407 /* path points to cache entries, so xstrdup before messing with it */
408 char *buf = xstrdup(path);
409 enum scld_error result = safe_create_leading_directories(buf);
410
411 save_errno = errno;
412 free(buf);
413 errno = save_errno;
414 return result;
415 }
416
417 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
418 {
419 /*
420 * The number of times we will try to remove empty directories
421 * in the way of path. This is only 1 because if another
422 * process is racily creating directories that conflict with
423 * us, we don't want to fight against them.
424 */
425 int remove_directories_remaining = 1;
426
427 /*
428 * The number of times that we will try to create the
429 * directories containing path. We are willing to attempt this
430 * more than once, because another process could be trying to
431 * clean up empty directories at the same time as we are
432 * trying to create them.
433 */
434 int create_directories_remaining = 3;
435
436 /* A scratch copy of path, filled lazily if we need it: */
437 struct strbuf path_copy = STRBUF_INIT;
438
439 int ret, save_errno;
440
441 /* Sanity check: */
442 assert(*path);
443
444 retry_fn:
445 ret = fn(path, cb);
446 save_errno = errno;
447 if (!ret)
448 goto out;
449
450 if (errno == EISDIR && remove_directories_remaining-- > 0) {
451 /*
452 * A directory is in the way. Maybe it is empty; try
453 * to remove it:
454 */
455 if (!path_copy.len)
456 strbuf_addstr(&path_copy, path);
457
458 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
459 goto retry_fn;
460 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
461 /*
462 * Maybe the containing directory didn't exist, or
463 * maybe it was just deleted by a process that is
464 * racing with us to clean up empty directories. Try
465 * to create it:
466 */
467 enum scld_error scld_result;
468
469 if (!path_copy.len)
470 strbuf_addstr(&path_copy, path);
471
472 do {
473 scld_result = safe_create_leading_directories(path_copy.buf);
474 if (scld_result == SCLD_OK)
475 goto retry_fn;
476 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
477 }
478
479 out:
480 strbuf_release(&path_copy);
481 errno = save_errno;
482 return ret;
483 }
484
485 static void fill_loose_path(struct strbuf *buf, const struct object_id *oid)
486 {
487 int i;
488 for (i = 0; i < the_hash_algo->rawsz; i++) {
489 static char hex[] = "0123456789abcdef";
490 unsigned int val = oid->hash[i];
491 strbuf_addch(buf, hex[val >> 4]);
492 strbuf_addch(buf, hex[val & 0xf]);
493 if (!i)
494 strbuf_addch(buf, '/');
495 }
496 }
497
498 static const char *odb_loose_path(struct object_directory *odb,
499 struct strbuf *buf,
500 const struct object_id *oid)
501 {
502 strbuf_reset(buf);
503 strbuf_addstr(buf, odb->path);
504 strbuf_addch(buf, '/');
505 fill_loose_path(buf, oid);
506 return buf->buf;
507 }
508
509 const char *loose_object_path(struct repository *r, struct strbuf *buf,
510 const struct object_id *oid)
511 {
512 return odb_loose_path(r->objects->odb, buf, oid);
513 }
514
515 /*
516 * Return non-zero iff the path is usable as an alternate object database.
517 */
518 static int alt_odb_usable(struct raw_object_store *o,
519 struct strbuf *path,
520 const char *normalized_objdir)
521 {
522 struct object_directory *odb;
523
524 /* Detect cases where alternate disappeared */
525 if (!is_directory(path->buf)) {
526 error(_("object directory %s does not exist; "
527 "check .git/objects/info/alternates"),
528 path->buf);
529 return 0;
530 }
531
532 /*
533 * Prevent the common mistake of listing the same
534 * thing twice, or object directory itself.
535 */
536 for (odb = o->odb; odb; odb = odb->next) {
537 if (!fspathcmp(path->buf, odb->path))
538 return 0;
539 }
540 if (!fspathcmp(path->buf, normalized_objdir))
541 return 0;
542
543 return 1;
544 }
545
546 /*
547 * Prepare alternate object database registry.
548 *
549 * The variable alt_odb_list points at the list of struct
550 * object_directory. The elements on this list come from
551 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
552 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
553 * whose contents is similar to that environment variable but can be
554 * LF separated. Its base points at a statically allocated buffer that
555 * contains "/the/directory/corresponding/to/.git/objects/...", while
556 * its name points just after the slash at the end of ".git/objects/"
557 * in the example above, and has enough space to hold all hex characters
558 * of the object ID, an extra slash for the first level indirection, and
559 * the terminating NUL.
560 */
561 static void read_info_alternates(struct repository *r,
562 const char *relative_base,
563 int depth);
564 static int link_alt_odb_entry(struct repository *r, const char *entry,
565 const char *relative_base, int depth, const char *normalized_objdir)
566 {
567 struct object_directory *ent;
568 struct strbuf pathbuf = STRBUF_INIT;
569
570 if (!is_absolute_path(entry) && relative_base) {
571 strbuf_realpath(&pathbuf, relative_base, 1);
572 strbuf_addch(&pathbuf, '/');
573 }
574 strbuf_addstr(&pathbuf, entry);
575
576 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
577 error(_("unable to normalize alternate object path: %s"),
578 pathbuf.buf);
579 strbuf_release(&pathbuf);
580 return -1;
581 }
582
583 /*
584 * The trailing slash after the directory name is given by
585 * this function at the end. Remove duplicates.
586 */
587 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
588 strbuf_setlen(&pathbuf, pathbuf.len - 1);
589
590 if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir)) {
591 strbuf_release(&pathbuf);
592 return -1;
593 }
594
595 CALLOC_ARRAY(ent, 1);
596 ent->path = xstrdup(pathbuf.buf);
597
598 /* add the alternate entry */
599 *r->objects->odb_tail = ent;
600 r->objects->odb_tail = &(ent->next);
601 ent->next = NULL;
602
603 /* recursively add alternates */
604 read_info_alternates(r, pathbuf.buf, depth + 1);
605
606 strbuf_release(&pathbuf);
607 return 0;
608 }
609
610 static const char *parse_alt_odb_entry(const char *string,
611 int sep,
612 struct strbuf *out)
613 {
614 const char *end;
615
616 strbuf_reset(out);
617
618 if (*string == '#') {
619 /* comment; consume up to next separator */
620 end = strchrnul(string, sep);
621 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
622 /*
623 * quoted path; unquote_c_style has copied the
624 * data for us and set "end". Broken quoting (e.g.,
625 * an entry that doesn't end with a quote) falls
626 * back to the unquoted case below.
627 */
628 } else {
629 /* normal, unquoted path */
630 end = strchrnul(string, sep);
631 strbuf_add(out, string, end - string);
632 }
633
634 if (*end)
635 end++;
636 return end;
637 }
638
639 static void link_alt_odb_entries(struct repository *r, const char *alt,
640 int sep, const char *relative_base, int depth)
641 {
642 struct strbuf objdirbuf = STRBUF_INIT;
643 struct strbuf entry = STRBUF_INIT;
644
645 if (!alt || !*alt)
646 return;
647
648 if (depth > 5) {
649 error(_("%s: ignoring alternate object stores, nesting too deep"),
650 relative_base);
651 return;
652 }
653
654 strbuf_add_absolute_path(&objdirbuf, r->objects->odb->path);
655 if (strbuf_normalize_path(&objdirbuf) < 0)
656 die(_("unable to normalize object directory: %s"),
657 objdirbuf.buf);
658
659 while (*alt) {
660 alt = parse_alt_odb_entry(alt, sep, &entry);
661 if (!entry.len)
662 continue;
663 link_alt_odb_entry(r, entry.buf,
664 relative_base, depth, objdirbuf.buf);
665 }
666 strbuf_release(&entry);
667 strbuf_release(&objdirbuf);
668 }
669
670 static void read_info_alternates(struct repository *r,
671 const char *relative_base,
672 int depth)
673 {
674 char *path;
675 struct strbuf buf = STRBUF_INIT;
676
677 path = xstrfmt("%s/info/alternates", relative_base);
678 if (strbuf_read_file(&buf, path, 1024) < 0) {
679 warn_on_fopen_errors(path);
680 free(path);
681 return;
682 }
683
684 link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
685 strbuf_release(&buf);
686 free(path);
687 }
688
689 void add_to_alternates_file(const char *reference)
690 {
691 struct lock_file lock = LOCK_INIT;
692 char *alts = git_pathdup("objects/info/alternates");
693 FILE *in, *out;
694 int found = 0;
695
696 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
697 out = fdopen_lock_file(&lock, "w");
698 if (!out)
699 die_errno(_("unable to fdopen alternates lockfile"));
700
701 in = fopen(alts, "r");
702 if (in) {
703 struct strbuf line = STRBUF_INIT;
704
705 while (strbuf_getline(&line, in) != EOF) {
706 if (!strcmp(reference, line.buf)) {
707 found = 1;
708 break;
709 }
710 fprintf_or_die(out, "%s\n", line.buf);
711 }
712
713 strbuf_release(&line);
714 fclose(in);
715 }
716 else if (errno != ENOENT)
717 die_errno(_("unable to read alternates file"));
718
719 if (found) {
720 rollback_lock_file(&lock);
721 } else {
722 fprintf_or_die(out, "%s\n", reference);
723 if (commit_lock_file(&lock))
724 die_errno(_("unable to move new alternates file into place"));
725 if (the_repository->objects->loaded_alternates)
726 link_alt_odb_entries(the_repository, reference,
727 '\n', NULL, 0);
728 }
729 free(alts);
730 }
731
732 void add_to_alternates_memory(const char *reference)
733 {
734 /*
735 * Make sure alternates are initialized, or else our entry may be
736 * overwritten when they are.
737 */
738 prepare_alt_odb(the_repository);
739
740 link_alt_odb_entries(the_repository, reference,
741 '\n', NULL, 0);
742 }
743
744 /*
745 * Compute the exact path an alternate is at and returns it. In case of
746 * error NULL is returned and the human readable error is added to `err`
747 * `path` may be relative and should point to $GIT_DIR.
748 * `err` must not be null.
749 */
750 char *compute_alternate_path(const char *path, struct strbuf *err)
751 {
752 char *ref_git = NULL;
753 const char *repo;
754 int seen_error = 0;
755
756 ref_git = real_pathdup(path, 0);
757 if (!ref_git) {
758 seen_error = 1;
759 strbuf_addf(err, _("path '%s' does not exist"), path);
760 goto out;
761 }
762
763 repo = read_gitfile(ref_git);
764 if (!repo)
765 repo = read_gitfile(mkpath("%s/.git", ref_git));
766 if (repo) {
767 free(ref_git);
768 ref_git = xstrdup(repo);
769 }
770
771 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
772 char *ref_git_git = mkpathdup("%s/.git", ref_git);
773 free(ref_git);
774 ref_git = ref_git_git;
775 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
776 struct strbuf sb = STRBUF_INIT;
777 seen_error = 1;
778 if (get_common_dir(&sb, ref_git)) {
779 strbuf_addf(err,
780 _("reference repository '%s' as a linked "
781 "checkout is not supported yet."),
782 path);
783 goto out;
784 }
785
786 strbuf_addf(err, _("reference repository '%s' is not a "
787 "local repository."), path);
788 goto out;
789 }
790
791 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
792 strbuf_addf(err, _("reference repository '%s' is shallow"),
793 path);
794 seen_error = 1;
795 goto out;
796 }
797
798 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
799 strbuf_addf(err,
800 _("reference repository '%s' is grafted"),
801 path);
802 seen_error = 1;
803 goto out;
804 }
805
806 out:
807 if (seen_error) {
808 FREE_AND_NULL(ref_git);
809 }
810
811 return ref_git;
812 }
813
814 static void fill_alternate_refs_command(struct child_process *cmd,
815 const char *repo_path)
816 {
817 const char *value;
818
819 if (!git_config_get_value("core.alternateRefsCommand", &value)) {
820 cmd->use_shell = 1;
821
822 strvec_push(&cmd->args, value);
823 strvec_push(&cmd->args, repo_path);
824 } else {
825 cmd->git_cmd = 1;
826
827 strvec_pushf(&cmd->args, "--git-dir=%s", repo_path);
828 strvec_push(&cmd->args, "for-each-ref");
829 strvec_push(&cmd->args, "--format=%(objectname)");
830
831 if (!git_config_get_value("core.alternateRefsPrefixes", &value)) {
832 strvec_push(&cmd->args, "--");
833 strvec_split(&cmd->args, value);
834 }
835 }
836
837 cmd->env = local_repo_env;
838 cmd->out = -1;
839 }
840
841 static void read_alternate_refs(const char *path,
842 alternate_ref_fn *cb,
843 void *data)
844 {
845 struct child_process cmd = CHILD_PROCESS_INIT;
846 struct strbuf line = STRBUF_INIT;
847 FILE *fh;
848
849 fill_alternate_refs_command(&cmd, path);
850
851 if (start_command(&cmd))
852 return;
853
854 fh = xfdopen(cmd.out, "r");
855 while (strbuf_getline_lf(&line, fh) != EOF) {
856 struct object_id oid;
857 const char *p;
858
859 if (parse_oid_hex(line.buf, &oid, &p) || *p) {
860 warning(_("invalid line while parsing alternate refs: %s"),
861 line.buf);
862 break;
863 }
864
865 cb(&oid, data);
866 }
867
868 fclose(fh);
869 finish_command(&cmd);
870 strbuf_release(&line);
871 }
872
873 struct alternate_refs_data {
874 alternate_ref_fn *fn;
875 void *data;
876 };
877
878 static int refs_from_alternate_cb(struct object_directory *e,
879 void *data)
880 {
881 struct strbuf path = STRBUF_INIT;
882 size_t base_len;
883 struct alternate_refs_data *cb = data;
884
885 if (!strbuf_realpath(&path, e->path, 0))
886 goto out;
887 if (!strbuf_strip_suffix(&path, "/objects"))
888 goto out;
889 base_len = path.len;
890
891 /* Is this a git repository with refs? */
892 strbuf_addstr(&path, "/refs");
893 if (!is_directory(path.buf))
894 goto out;
895 strbuf_setlen(&path, base_len);
896
897 read_alternate_refs(path.buf, cb->fn, cb->data);
898
899 out:
900 strbuf_release(&path);
901 return 0;
902 }
903
904 void for_each_alternate_ref(alternate_ref_fn fn, void *data)
905 {
906 struct alternate_refs_data cb;
907 cb.fn = fn;
908 cb.data = data;
909 foreach_alt_odb(refs_from_alternate_cb, &cb);
910 }
911
912 int foreach_alt_odb(alt_odb_fn fn, void *cb)
913 {
914 struct object_directory *ent;
915 int r = 0;
916
917 prepare_alt_odb(the_repository);
918 for (ent = the_repository->objects->odb->next; ent; ent = ent->next) {
919 r = fn(ent, cb);
920 if (r)
921 break;
922 }
923 return r;
924 }
925
926 void prepare_alt_odb(struct repository *r)
927 {
928 if (r->objects->loaded_alternates)
929 return;
930
931 link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
932
933 read_info_alternates(r, r->objects->odb->path, 0);
934 r->objects->loaded_alternates = 1;
935 }
936
937 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
938 static int freshen_file(const char *fn)
939 {
940 return !utime(fn, NULL);
941 }
942
943 /*
944 * All of the check_and_freshen functions return 1 if the file exists and was
945 * freshened (if freshening was requested), 0 otherwise. If they return
946 * 0, you should not assume that it is safe to skip a write of the object (it
947 * either does not exist on disk, or has a stale mtime and may be subject to
948 * pruning).
949 */
950 int check_and_freshen_file(const char *fn, int freshen)
951 {
952 if (access(fn, F_OK))
953 return 0;
954 if (freshen && !freshen_file(fn))
955 return 0;
956 return 1;
957 }
958
959 static int check_and_freshen_odb(struct object_directory *odb,
960 const struct object_id *oid,
961 int freshen)
962 {
963 static struct strbuf path = STRBUF_INIT;
964 odb_loose_path(odb, &path, oid);
965 return check_and_freshen_file(path.buf, freshen);
966 }
967
968 static int check_and_freshen_local(const struct object_id *oid, int freshen)
969 {
970 return check_and_freshen_odb(the_repository->objects->odb, oid, freshen);
971 }
972
973 static int check_and_freshen_nonlocal(const struct object_id *oid, int freshen)
974 {
975 struct object_directory *odb;
976
977 prepare_alt_odb(the_repository);
978 for (odb = the_repository->objects->odb->next; odb; odb = odb->next) {
979 if (check_and_freshen_odb(odb, oid, freshen))
980 return 1;
981 }
982 return 0;
983 }
984
985 static int check_and_freshen(const struct object_id *oid, int freshen)
986 {
987 return check_and_freshen_local(oid, freshen) ||
988 check_and_freshen_nonlocal(oid, freshen);
989 }
990
991 int has_loose_object_nonlocal(const struct object_id *oid)
992 {
993 return check_and_freshen_nonlocal(oid, 0);
994 }
995
996 static int has_loose_object(const struct object_id *oid)
997 {
998 return check_and_freshen(oid, 0);
999 }
1000
1001 static void mmap_limit_check(size_t length)
1002 {
1003 static size_t limit = 0;
1004 if (!limit) {
1005 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
1006 if (!limit)
1007 limit = SIZE_MAX;
1008 }
1009 if (length > limit)
1010 die(_("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX),
1011 (uintmax_t)length, (uintmax_t)limit);
1012 }
1013
1014 void *xmmap_gently(void *start, size_t length,
1015 int prot, int flags, int fd, off_t offset)
1016 {
1017 void *ret;
1018
1019 mmap_limit_check(length);
1020 ret = mmap(start, length, prot, flags, fd, offset);
1021 if (ret == MAP_FAILED && !length)
1022 ret = NULL;
1023 return ret;
1024 }
1025
1026 const char *mmap_os_err(void)
1027 {
1028 static const char blank[] = "";
1029 #if defined(__linux__)
1030 if (errno == ENOMEM) {
1031 /* this continues an existing error message: */
1032 static const char enomem[] =
1033 ", check sys.vm.max_map_count and/or RLIMIT_DATA";
1034 return enomem;
1035 }
1036 #endif /* OS-specific bits */
1037 return blank;
1038 }
1039
1040 void *xmmap(void *start, size_t length,
1041 int prot, int flags, int fd, off_t offset)
1042 {
1043 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
1044 if (ret == MAP_FAILED)
1045 die_errno(_("mmap failed%s"), mmap_os_err());
1046 return ret;
1047 }
1048
1049 /*
1050 * With an in-core object data in "map", rehash it to make sure the
1051 * object name actually matches "oid" to detect object corruption.
1052 * With "map" == NULL, try reading the object named with "oid" using
1053 * the streaming interface and rehash it to do the same.
1054 */
1055 int check_object_signature(struct repository *r, const struct object_id *oid,
1056 void *map, unsigned long size, const char *type)
1057 {
1058 struct object_id real_oid;
1059 enum object_type obj_type;
1060 struct git_istream *st;
1061 git_hash_ctx c;
1062 char hdr[MAX_HEADER_LEN];
1063 int hdrlen;
1064
1065 if (map) {
1066 hash_object_file(r->hash_algo, map, size, type, &real_oid);
1067 return !oideq(oid, &real_oid) ? -1 : 0;
1068 }
1069
1070 st = open_istream(r, oid, &obj_type, &size, NULL);
1071 if (!st)
1072 return -1;
1073
1074 /* Generate the header */
1075 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(obj_type), (uintmax_t)size) + 1;
1076
1077 /* Sha1.. */
1078 r->hash_algo->init_fn(&c);
1079 r->hash_algo->update_fn(&c, hdr, hdrlen);
1080 for (;;) {
1081 char buf[1024 * 16];
1082 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1083
1084 if (readlen < 0) {
1085 close_istream(st);
1086 return -1;
1087 }
1088 if (!readlen)
1089 break;
1090 r->hash_algo->update_fn(&c, buf, readlen);
1091 }
1092 r->hash_algo->final_oid_fn(&real_oid, &c);
1093 close_istream(st);
1094 return !oideq(oid, &real_oid) ? -1 : 0;
1095 }
1096
1097 int git_open_cloexec(const char *name, int flags)
1098 {
1099 int fd;
1100 static int o_cloexec = O_CLOEXEC;
1101
1102 fd = open(name, flags | o_cloexec);
1103 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1104 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1105 o_cloexec &= ~O_CLOEXEC;
1106 fd = open(name, flags | o_cloexec);
1107 }
1108
1109 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1110 {
1111 static int fd_cloexec = FD_CLOEXEC;
1112
1113 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1114 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1115 int flags = fcntl(fd, F_GETFD);
1116 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1117 fd_cloexec = 0;
1118 }
1119 }
1120 #endif
1121 return fd;
1122 }
1123
1124 /*
1125 * Find "oid" as a loose object in the local repository or in an alternate.
1126 * Returns 0 on success, negative on failure.
1127 *
1128 * The "path" out-parameter will give the path of the object we found (if any).
1129 * Note that it may point to static storage and is only valid until another
1130 * call to stat_loose_object().
1131 */
1132 static int stat_loose_object(struct repository *r, const struct object_id *oid,
1133 struct stat *st, const char **path)
1134 {
1135 struct object_directory *odb;
1136 static struct strbuf buf = STRBUF_INIT;
1137
1138 prepare_alt_odb(r);
1139 for (odb = r->objects->odb; odb; odb = odb->next) {
1140 *path = odb_loose_path(odb, &buf, oid);
1141 if (!lstat(*path, st))
1142 return 0;
1143 }
1144
1145 return -1;
1146 }
1147
1148 /*
1149 * Like stat_loose_object(), but actually open the object and return the
1150 * descriptor. See the caveats on the "path" parameter above.
1151 */
1152 static int open_loose_object(struct repository *r,
1153 const struct object_id *oid, const char **path)
1154 {
1155 int fd;
1156 struct object_directory *odb;
1157 int most_interesting_errno = ENOENT;
1158 static struct strbuf buf = STRBUF_INIT;
1159
1160 prepare_alt_odb(r);
1161 for (odb = r->objects->odb; odb; odb = odb->next) {
1162 *path = odb_loose_path(odb, &buf, oid);
1163 fd = git_open(*path);
1164 if (fd >= 0)
1165 return fd;
1166
1167 if (most_interesting_errno == ENOENT)
1168 most_interesting_errno = errno;
1169 }
1170 errno = most_interesting_errno;
1171 return -1;
1172 }
1173
1174 static int quick_has_loose(struct repository *r,
1175 const struct object_id *oid)
1176 {
1177 struct object_directory *odb;
1178
1179 prepare_alt_odb(r);
1180 for (odb = r->objects->odb; odb; odb = odb->next) {
1181 if (oid_array_lookup(odb_loose_cache(odb, oid), oid) >= 0)
1182 return 1;
1183 }
1184 return 0;
1185 }
1186
1187 /*
1188 * Map the loose object at "path" if it is not NULL, or the path found by
1189 * searching for a loose object named "oid".
1190 */
1191 static void *map_loose_object_1(struct repository *r, const char *path,
1192 const struct object_id *oid, unsigned long *size)
1193 {
1194 void *map;
1195 int fd;
1196
1197 if (path)
1198 fd = git_open(path);
1199 else
1200 fd = open_loose_object(r, oid, &path);
1201 map = NULL;
1202 if (fd >= 0) {
1203 struct stat st;
1204
1205 if (!fstat(fd, &st)) {
1206 *size = xsize_t(st.st_size);
1207 if (!*size) {
1208 /* mmap() is forbidden on empty files */
1209 error(_("object file %s is empty"), path);
1210 close(fd);
1211 return NULL;
1212 }
1213 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1214 }
1215 close(fd);
1216 }
1217 return map;
1218 }
1219
1220 void *map_loose_object(struct repository *r,
1221 const struct object_id *oid,
1222 unsigned long *size)
1223 {
1224 return map_loose_object_1(r, NULL, oid, size);
1225 }
1226
1227 static int unpack_loose_short_header(git_zstream *stream,
1228 unsigned char *map, unsigned long mapsize,
1229 void *buffer, unsigned long bufsiz)
1230 {
1231 int ret;
1232
1233 /* Get the data stream */
1234 memset(stream, 0, sizeof(*stream));
1235 stream->next_in = map;
1236 stream->avail_in = mapsize;
1237 stream->next_out = buffer;
1238 stream->avail_out = bufsiz;
1239
1240 git_inflate_init(stream);
1241 obj_read_unlock();
1242 ret = git_inflate(stream, 0);
1243 obj_read_lock();
1244
1245 return ret;
1246 }
1247
1248 int unpack_loose_header(git_zstream *stream,
1249 unsigned char *map, unsigned long mapsize,
1250 void *buffer, unsigned long bufsiz)
1251 {
1252 int status = unpack_loose_short_header(stream, map, mapsize,
1253 buffer, bufsiz);
1254
1255 if (status < Z_OK)
1256 return status;
1257
1258 /* Make sure we have the terminating NUL */
1259 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1260 return -1;
1261 return 0;
1262 }
1263
1264 static int unpack_loose_header_to_strbuf(git_zstream *stream, unsigned char *map,
1265 unsigned long mapsize, void *buffer,
1266 unsigned long bufsiz, struct strbuf *header)
1267 {
1268 int status;
1269
1270 status = unpack_loose_short_header(stream, map, mapsize, buffer, bufsiz);
1271 if (status < Z_OK)
1272 return -1;
1273
1274 /*
1275 * Check if entire header is unpacked in the first iteration.
1276 */
1277 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1278 return 0;
1279
1280 /*
1281 * buffer[0..bufsiz] was not large enough. Copy the partial
1282 * result out to header, and then append the result of further
1283 * reading the stream.
1284 */
1285 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1286 stream->next_out = buffer;
1287 stream->avail_out = bufsiz;
1288
1289 do {
1290 obj_read_unlock();
1291 status = git_inflate(stream, 0);
1292 obj_read_lock();
1293 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1294 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1295 return 0;
1296 stream->next_out = buffer;
1297 stream->avail_out = bufsiz;
1298 } while (status != Z_STREAM_END);
1299 return -1;
1300 }
1301
1302 static void *unpack_loose_rest(git_zstream *stream,
1303 void *buffer, unsigned long size,
1304 const struct object_id *oid)
1305 {
1306 int bytes = strlen(buffer) + 1;
1307 unsigned char *buf = xmallocz(size);
1308 unsigned long n;
1309 int status = Z_OK;
1310
1311 n = stream->total_out - bytes;
1312 if (n > size)
1313 n = size;
1314 memcpy(buf, (char *) buffer + bytes, n);
1315 bytes = n;
1316 if (bytes <= size) {
1317 /*
1318 * The above condition must be (bytes <= size), not
1319 * (bytes < size). In other words, even though we
1320 * expect no more output and set avail_out to zero,
1321 * the input zlib stream may have bytes that express
1322 * "this concludes the stream", and we *do* want to
1323 * eat that input.
1324 *
1325 * Otherwise we would not be able to test that we
1326 * consumed all the input to reach the expected size;
1327 * we also want to check that zlib tells us that all
1328 * went well with status == Z_STREAM_END at the end.
1329 */
1330 stream->next_out = buf + bytes;
1331 stream->avail_out = size - bytes;
1332 while (status == Z_OK) {
1333 obj_read_unlock();
1334 status = git_inflate(stream, Z_FINISH);
1335 obj_read_lock();
1336 }
1337 }
1338 if (status == Z_STREAM_END && !stream->avail_in) {
1339 git_inflate_end(stream);
1340 return buf;
1341 }
1342
1343 if (status < 0)
1344 error(_("corrupt loose object '%s'"), oid_to_hex(oid));
1345 else if (stream->avail_in)
1346 error(_("garbage at end of loose object '%s'"),
1347 oid_to_hex(oid));
1348 free(buf);
1349 return NULL;
1350 }
1351
1352 /*
1353 * We used to just use "sscanf()", but that's actually way
1354 * too permissive for what we want to check. So do an anal
1355 * object header parse by hand.
1356 */
1357 static int parse_loose_header_extended(const char *hdr, struct object_info *oi,
1358 unsigned int flags)
1359 {
1360 const char *type_buf = hdr;
1361 unsigned long size;
1362 int type, type_len = 0;
1363
1364 /*
1365 * The type can be of any size but is followed by
1366 * a space.
1367 */
1368 for (;;) {
1369 char c = *hdr++;
1370 if (!c)
1371 return -1;
1372 if (c == ' ')
1373 break;
1374 type_len++;
1375 }
1376
1377 type = type_from_string_gently(type_buf, type_len, 1);
1378 if (oi->type_name)
1379 strbuf_add(oi->type_name, type_buf, type_len);
1380 /*
1381 * Set type to 0 if its an unknown object and
1382 * we're obtaining the type using '--allow-unknown-type'
1383 * option.
1384 */
1385 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1386 type = 0;
1387 else if (type < 0)
1388 die(_("invalid object type"));
1389 if (oi->typep)
1390 *oi->typep = type;
1391
1392 /*
1393 * The length must follow immediately, and be in canonical
1394 * decimal format (ie "010" is not valid).
1395 */
1396 size = *hdr++ - '0';
1397 if (size > 9)
1398 return -1;
1399 if (size) {
1400 for (;;) {
1401 unsigned long c = *hdr - '0';
1402 if (c > 9)
1403 break;
1404 hdr++;
1405 size = size * 10 + c;
1406 }
1407 }
1408
1409 if (oi->sizep)
1410 *oi->sizep = size;
1411
1412 /*
1413 * The length must be followed by a zero byte
1414 */
1415 return *hdr ? -1 : type;
1416 }
1417
1418 int parse_loose_header(const char *hdr, unsigned long *sizep)
1419 {
1420 struct object_info oi = OBJECT_INFO_INIT;
1421
1422 oi.sizep = sizep;
1423 return parse_loose_header_extended(hdr, &oi, 0);
1424 }
1425
1426 static int loose_object_info(struct repository *r,
1427 const struct object_id *oid,
1428 struct object_info *oi, int flags)
1429 {
1430 int status = 0;
1431 unsigned long mapsize;
1432 void *map;
1433 git_zstream stream;
1434 char hdr[MAX_HEADER_LEN];
1435 struct strbuf hdrbuf = STRBUF_INIT;
1436 unsigned long size_scratch;
1437
1438 if (oi->delta_base_oid)
1439 oidclr(oi->delta_base_oid);
1440
1441 /*
1442 * If we don't care about type or size, then we don't
1443 * need to look inside the object at all. Note that we
1444 * do not optimize out the stat call, even if the
1445 * caller doesn't care about the disk-size, since our
1446 * return value implicitly indicates whether the
1447 * object even exists.
1448 */
1449 if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1450 const char *path;
1451 struct stat st;
1452 if (!oi->disk_sizep && (flags & OBJECT_INFO_QUICK))
1453 return quick_has_loose(r, oid) ? 0 : -1;
1454 if (stat_loose_object(r, oid, &st, &path) < 0)
1455 return -1;
1456 if (oi->disk_sizep)
1457 *oi->disk_sizep = st.st_size;
1458 return 0;
1459 }
1460
1461 map = map_loose_object(r, oid, &mapsize);
1462 if (!map)
1463 return -1;
1464
1465 if (!oi->sizep)
1466 oi->sizep = &size_scratch;
1467
1468 if (oi->disk_sizep)
1469 *oi->disk_sizep = mapsize;
1470 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1471 if (unpack_loose_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1472 status = error(_("unable to unpack %s header with --allow-unknown-type"),
1473 oid_to_hex(oid));
1474 } else if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1475 status = error(_("unable to unpack %s header"),
1476 oid_to_hex(oid));
1477 if (status < 0)
1478 ; /* Do nothing */
1479 else if (hdrbuf.len) {
1480 if ((status = parse_loose_header_extended(hdrbuf.buf, oi, flags)) < 0)
1481 status = error(_("unable to parse %s header with --allow-unknown-type"),
1482 oid_to_hex(oid));
1483 } else if ((status = parse_loose_header_extended(hdr, oi, flags)) < 0)
1484 status = error(_("unable to parse %s header"), oid_to_hex(oid));
1485
1486 if (status >= 0 && oi->contentp) {
1487 *oi->contentp = unpack_loose_rest(&stream, hdr,
1488 *oi->sizep, oid);
1489 if (!*oi->contentp) {
1490 git_inflate_end(&stream);
1491 status = -1;
1492 }
1493 } else
1494 git_inflate_end(&stream);
1495
1496 munmap(map, mapsize);
1497 if (status && oi->typep)
1498 *oi->typep = status;
1499 if (oi->sizep == &size_scratch)
1500 oi->sizep = NULL;
1501 strbuf_release(&hdrbuf);
1502 oi->whence = OI_LOOSE;
1503 return (status < 0) ? status : 0;
1504 }
1505
1506 int obj_read_use_lock = 0;
1507 pthread_mutex_t obj_read_mutex;
1508
1509 void enable_obj_read_lock(void)
1510 {
1511 if (obj_read_use_lock)
1512 return;
1513
1514 obj_read_use_lock = 1;
1515 init_recursive_mutex(&obj_read_mutex);
1516 }
1517
1518 void disable_obj_read_lock(void)
1519 {
1520 if (!obj_read_use_lock)
1521 return;
1522
1523 obj_read_use_lock = 0;
1524 pthread_mutex_destroy(&obj_read_mutex);
1525 }
1526
1527 int fetch_if_missing = 1;
1528
1529 static int do_oid_object_info_extended(struct repository *r,
1530 const struct object_id *oid,
1531 struct object_info *oi, unsigned flags)
1532 {
1533 static struct object_info blank_oi = OBJECT_INFO_INIT;
1534 struct cached_object *co;
1535 struct pack_entry e;
1536 int rtype;
1537 const struct object_id *real = oid;
1538 int already_retried = 0;
1539
1540
1541 if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1542 real = lookup_replace_object(r, oid);
1543
1544 if (is_null_oid(real))
1545 return -1;
1546
1547 if (!oi)
1548 oi = &blank_oi;
1549
1550 co = find_cached_object(real);
1551 if (co) {
1552 if (oi->typep)
1553 *(oi->typep) = co->type;
1554 if (oi->sizep)
1555 *(oi->sizep) = co->size;
1556 if (oi->disk_sizep)
1557 *(oi->disk_sizep) = 0;
1558 if (oi->delta_base_oid)
1559 oidclr(oi->delta_base_oid);
1560 if (oi->type_name)
1561 strbuf_addstr(oi->type_name, type_name(co->type));
1562 if (oi->contentp)
1563 *oi->contentp = xmemdupz(co->buf, co->size);
1564 oi->whence = OI_CACHED;
1565 return 0;
1566 }
1567
1568 while (1) {
1569 if (find_pack_entry(r, real, &e))
1570 break;
1571
1572 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1573 return -1;
1574
1575 /* Most likely it's a loose object. */
1576 if (!loose_object_info(r, real, oi, flags))
1577 return 0;
1578
1579 /* Not a loose object; someone else may have just packed it. */
1580 if (!(flags & OBJECT_INFO_QUICK)) {
1581 reprepare_packed_git(r);
1582 if (find_pack_entry(r, real, &e))
1583 break;
1584 }
1585
1586 /* Check if it is a missing object */
1587 if (fetch_if_missing && repo_has_promisor_remote(r) &&
1588 !already_retried &&
1589 !(flags & OBJECT_INFO_SKIP_FETCH_OBJECT)) {
1590 /*
1591 * TODO Investigate checking promisor_remote_get_direct()
1592 * TODO return value and stopping on error here.
1593 */
1594 promisor_remote_get_direct(r, real, 1);
1595 already_retried = 1;
1596 continue;
1597 }
1598
1599 return -1;
1600 }
1601
1602 if (oi == &blank_oi)
1603 /*
1604 * We know that the caller doesn't actually need the
1605 * information below, so return early.
1606 */
1607 return 0;
1608 rtype = packed_object_info(r, e.p, e.offset, oi);
1609 if (rtype < 0) {
1610 mark_bad_packed_object(e.p, real->hash);
1611 return do_oid_object_info_extended(r, real, oi, 0);
1612 } else if (oi->whence == OI_PACKED) {
1613 oi->u.packed.offset = e.offset;
1614 oi->u.packed.pack = e.p;
1615 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1616 rtype == OBJ_OFS_DELTA);
1617 }
1618
1619 return 0;
1620 }
1621
1622 int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1623 struct object_info *oi, unsigned flags)
1624 {
1625 int ret;
1626 obj_read_lock();
1627 ret = do_oid_object_info_extended(r, oid, oi, flags);
1628 obj_read_unlock();
1629 return ret;
1630 }
1631
1632
1633 /* returns enum object_type or negative */
1634 int oid_object_info(struct repository *r,
1635 const struct object_id *oid,
1636 unsigned long *sizep)
1637 {
1638 enum object_type type;
1639 struct object_info oi = OBJECT_INFO_INIT;
1640
1641 oi.typep = &type;
1642 oi.sizep = sizep;
1643 if (oid_object_info_extended(r, oid, &oi,
1644 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1645 return -1;
1646 return type;
1647 }
1648
1649 static void *read_object(struct repository *r,
1650 const struct object_id *oid, enum object_type *type,
1651 unsigned long *size)
1652 {
1653 struct object_info oi = OBJECT_INFO_INIT;
1654 void *content;
1655 oi.typep = type;
1656 oi.sizep = size;
1657 oi.contentp = &content;
1658
1659 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1660 return NULL;
1661 return content;
1662 }
1663
1664 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1665 struct object_id *oid)
1666 {
1667 struct cached_object *co;
1668
1669 hash_object_file(the_hash_algo, buf, len, type_name(type), oid);
1670 if (has_object_file_with_flags(oid, OBJECT_INFO_QUICK | OBJECT_INFO_SKIP_FETCH_OBJECT) ||
1671 find_cached_object(oid))
1672 return 0;
1673 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1674 co = &cached_objects[cached_object_nr++];
1675 co->size = len;
1676 co->type = type;
1677 co->buf = xmalloc(len);
1678 memcpy(co->buf, buf, len);
1679 oidcpy(&co->oid, oid);
1680 return 0;
1681 }
1682
1683 /*
1684 * This function dies on corrupt objects; the callers who want to
1685 * deal with them should arrange to call read_object() and give error
1686 * messages themselves.
1687 */
1688 void *read_object_file_extended(struct repository *r,
1689 const struct object_id *oid,
1690 enum object_type *type,
1691 unsigned long *size,
1692 int lookup_replace)
1693 {
1694 void *data;
1695 const struct packed_git *p;
1696 const char *path;
1697 struct stat st;
1698 const struct object_id *repl = lookup_replace ?
1699 lookup_replace_object(r, oid) : oid;
1700
1701 errno = 0;
1702 data = read_object(r, repl, type, size);
1703 if (data)
1704 return data;
1705
1706 obj_read_lock();
1707 if (errno && errno != ENOENT)
1708 die_errno(_("failed to read object %s"), oid_to_hex(oid));
1709
1710 /* die if we replaced an object with one that does not exist */
1711 if (repl != oid)
1712 die(_("replacement %s not found for %s"),
1713 oid_to_hex(repl), oid_to_hex(oid));
1714
1715 if (!stat_loose_object(r, repl, &st, &path))
1716 die(_("loose object %s (stored in %s) is corrupt"),
1717 oid_to_hex(repl), path);
1718
1719 if ((p = has_packed_and_bad(r, repl->hash)) != NULL)
1720 die(_("packed object %s (stored in %s) is corrupt"),
1721 oid_to_hex(repl), p->pack_name);
1722 obj_read_unlock();
1723
1724 return NULL;
1725 }
1726
1727 void *read_object_with_reference(struct repository *r,
1728 const struct object_id *oid,
1729 const char *required_type_name,
1730 unsigned long *size,
1731 struct object_id *actual_oid_return)
1732 {
1733 enum object_type type, required_type;
1734 void *buffer;
1735 unsigned long isize;
1736 struct object_id actual_oid;
1737
1738 required_type = type_from_string(required_type_name);
1739 oidcpy(&actual_oid, oid);
1740 while (1) {
1741 int ref_length = -1;
1742 const char *ref_type = NULL;
1743
1744 buffer = repo_read_object_file(r, &actual_oid, &type, &isize);
1745 if (!buffer)
1746 return NULL;
1747 if (type == required_type) {
1748 *size = isize;
1749 if (actual_oid_return)
1750 oidcpy(actual_oid_return, &actual_oid);
1751 return buffer;
1752 }
1753 /* Handle references */
1754 else if (type == OBJ_COMMIT)
1755 ref_type = "tree ";
1756 else if (type == OBJ_TAG)
1757 ref_type = "object ";
1758 else {
1759 free(buffer);
1760 return NULL;
1761 }
1762 ref_length = strlen(ref_type);
1763
1764 if (ref_length + the_hash_algo->hexsz > isize ||
1765 memcmp(buffer, ref_type, ref_length) ||
1766 get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1767 free(buffer);
1768 return NULL;
1769 }
1770 free(buffer);
1771 /* Now we have the ID of the referred-to object in
1772 * actual_oid. Check again. */
1773 }
1774 }
1775
1776 static void write_object_file_prepare(const struct git_hash_algo *algo,
1777 const void *buf, unsigned long len,
1778 const char *type, struct object_id *oid,
1779 char *hdr, int *hdrlen)
1780 {
1781 git_hash_ctx c;
1782
1783 /* Generate the header */
1784 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %"PRIuMAX , type, (uintmax_t)len)+1;
1785
1786 /* Sha1.. */
1787 algo->init_fn(&c);
1788 algo->update_fn(&c, hdr, *hdrlen);
1789 algo->update_fn(&c, buf, len);
1790 algo->final_oid_fn(oid, &c);
1791 }
1792
1793 /*
1794 * Move the just written object into its final resting place.
1795 */
1796 int finalize_object_file(const char *tmpfile, const char *filename)
1797 {
1798 int ret = 0;
1799
1800 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1801 goto try_rename;
1802 else if (link(tmpfile, filename))
1803 ret = errno;
1804
1805 /*
1806 * Coda hack - coda doesn't like cross-directory links,
1807 * so we fall back to a rename, which will mean that it
1808 * won't be able to check collisions, but that's not a
1809 * big deal.
1810 *
1811 * The same holds for FAT formatted media.
1812 *
1813 * When this succeeds, we just return. We have nothing
1814 * left to unlink.
1815 */
1816 if (ret && ret != EEXIST) {
1817 try_rename:
1818 if (!rename(tmpfile, filename))
1819 goto out;
1820 ret = errno;
1821 }
1822 unlink_or_warn(tmpfile);
1823 if (ret) {
1824 if (ret != EEXIST) {
1825 return error_errno(_("unable to write file %s"), filename);
1826 }
1827 /* FIXME!!! Collision check here ? */
1828 }
1829
1830 out:
1831 if (adjust_shared_perm(filename))
1832 return error(_("unable to set permission to '%s'"), filename);
1833 return 0;
1834 }
1835
1836 static int write_buffer(int fd, const void *buf, size_t len)
1837 {
1838 if (write_in_full(fd, buf, len) < 0)
1839 return error_errno(_("file write error"));
1840 return 0;
1841 }
1842
1843 int hash_object_file(const struct git_hash_algo *algo, const void *buf,
1844 unsigned long len, const char *type,
1845 struct object_id *oid)
1846 {
1847 char hdr[MAX_HEADER_LEN];
1848 int hdrlen = sizeof(hdr);
1849 write_object_file_prepare(algo, buf, len, type, oid, hdr, &hdrlen);
1850 return 0;
1851 }
1852
1853 /* Finalize a file on disk, and close it. */
1854 static void close_loose_object(int fd)
1855 {
1856 if (fsync_object_files)
1857 fsync_or_die(fd, "loose object file");
1858 if (close(fd) != 0)
1859 die_errno(_("error when closing loose object file"));
1860 }
1861
1862 /* Size of directory component, including the ending '/' */
1863 static inline int directory_size(const char *filename)
1864 {
1865 const char *s = strrchr(filename, '/');
1866 if (!s)
1867 return 0;
1868 return s - filename + 1;
1869 }
1870
1871 /*
1872 * This creates a temporary file in the same directory as the final
1873 * 'filename'
1874 *
1875 * We want to avoid cross-directory filename renames, because those
1876 * can have problems on various filesystems (FAT, NFS, Coda).
1877 */
1878 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1879 {
1880 int fd, dirlen = directory_size(filename);
1881
1882 strbuf_reset(tmp);
1883 strbuf_add(tmp, filename, dirlen);
1884 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1885 fd = git_mkstemp_mode(tmp->buf, 0444);
1886 if (fd < 0 && dirlen && errno == ENOENT) {
1887 /*
1888 * Make sure the directory exists; note that the contents
1889 * of the buffer are undefined after mkstemp returns an
1890 * error, so we have to rewrite the whole buffer from
1891 * scratch.
1892 */
1893 strbuf_reset(tmp);
1894 strbuf_add(tmp, filename, dirlen - 1);
1895 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1896 return -1;
1897 if (adjust_shared_perm(tmp->buf))
1898 return -1;
1899
1900 /* Try again */
1901 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1902 fd = git_mkstemp_mode(tmp->buf, 0444);
1903 }
1904 return fd;
1905 }
1906
1907 static int write_loose_object(const struct object_id *oid, char *hdr,
1908 int hdrlen, const void *buf, unsigned long len,
1909 time_t mtime)
1910 {
1911 int fd, ret;
1912 unsigned char compressed[4096];
1913 git_zstream stream;
1914 git_hash_ctx c;
1915 struct object_id parano_oid;
1916 static struct strbuf tmp_file = STRBUF_INIT;
1917 static struct strbuf filename = STRBUF_INIT;
1918
1919 loose_object_path(the_repository, &filename, oid);
1920
1921 fd = create_tmpfile(&tmp_file, filename.buf);
1922 if (fd < 0) {
1923 if (errno == EACCES)
1924 return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
1925 else
1926 return error_errno(_("unable to create temporary file"));
1927 }
1928
1929 /* Set it up */
1930 git_deflate_init(&stream, zlib_compression_level);
1931 stream.next_out = compressed;
1932 stream.avail_out = sizeof(compressed);
1933 the_hash_algo->init_fn(&c);
1934
1935 /* First header.. */
1936 stream.next_in = (unsigned char *)hdr;
1937 stream.avail_in = hdrlen;
1938 while (git_deflate(&stream, 0) == Z_OK)
1939 ; /* nothing */
1940 the_hash_algo->update_fn(&c, hdr, hdrlen);
1941
1942 /* Then the data itself.. */
1943 stream.next_in = (void *)buf;
1944 stream.avail_in = len;
1945 do {
1946 unsigned char *in0 = stream.next_in;
1947 ret = git_deflate(&stream, Z_FINISH);
1948 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1949 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1950 die(_("unable to write loose object file"));
1951 stream.next_out = compressed;
1952 stream.avail_out = sizeof(compressed);
1953 } while (ret == Z_OK);
1954
1955 if (ret != Z_STREAM_END)
1956 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
1957 ret);
1958 ret = git_deflate_end_gently(&stream);
1959 if (ret != Z_OK)
1960 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
1961 ret);
1962 the_hash_algo->final_oid_fn(&parano_oid, &c);
1963 if (!oideq(oid, &parano_oid))
1964 die(_("confused by unstable object source data for %s"),
1965 oid_to_hex(oid));
1966
1967 close_loose_object(fd);
1968
1969 if (mtime) {
1970 struct utimbuf utb;
1971 utb.actime = mtime;
1972 utb.modtime = mtime;
1973 if (utime(tmp_file.buf, &utb) < 0)
1974 warning_errno(_("failed utime() on %s"), tmp_file.buf);
1975 }
1976
1977 return finalize_object_file(tmp_file.buf, filename.buf);
1978 }
1979
1980 static int freshen_loose_object(const struct object_id *oid)
1981 {
1982 return check_and_freshen(oid, 1);
1983 }
1984
1985 static int freshen_packed_object(const struct object_id *oid)
1986 {
1987 struct pack_entry e;
1988 if (!find_pack_entry(the_repository, oid, &e))
1989 return 0;
1990 if (e.p->freshened)
1991 return 1;
1992 if (!freshen_file(e.p->pack_name))
1993 return 0;
1994 e.p->freshened = 1;
1995 return 1;
1996 }
1997
1998 int write_object_file(const void *buf, unsigned long len, const char *type,
1999 struct object_id *oid)
2000 {
2001 char hdr[MAX_HEADER_LEN];
2002 int hdrlen = sizeof(hdr);
2003
2004 /* Normally if we have it in the pack then we do not bother writing
2005 * it out into .git/objects/??/?{38} file.
2006 */
2007 write_object_file_prepare(the_hash_algo, buf, len, type, oid, hdr,
2008 &hdrlen);
2009 if (freshen_packed_object(oid) || freshen_loose_object(oid))
2010 return 0;
2011 return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
2012 }
2013
2014 int hash_object_file_literally(const void *buf, unsigned long len,
2015 const char *type, struct object_id *oid,
2016 unsigned flags)
2017 {
2018 char *header;
2019 int hdrlen, status = 0;
2020
2021 /* type string, SP, %lu of the length plus NUL must fit this */
2022 hdrlen = strlen(type) + MAX_HEADER_LEN;
2023 header = xmalloc(hdrlen);
2024 write_object_file_prepare(the_hash_algo, buf, len, type, oid, header,
2025 &hdrlen);
2026
2027 if (!(flags & HASH_WRITE_OBJECT))
2028 goto cleanup;
2029 if (freshen_packed_object(oid) || freshen_loose_object(oid))
2030 goto cleanup;
2031 status = write_loose_object(oid, header, hdrlen, buf, len, 0);
2032
2033 cleanup:
2034 free(header);
2035 return status;
2036 }
2037
2038 int force_object_loose(const struct object_id *oid, time_t mtime)
2039 {
2040 void *buf;
2041 unsigned long len;
2042 enum object_type type;
2043 char hdr[MAX_HEADER_LEN];
2044 int hdrlen;
2045 int ret;
2046
2047 if (has_loose_object(oid))
2048 return 0;
2049 buf = read_object(the_repository, oid, &type, &len);
2050 if (!buf)
2051 return error(_("cannot read object for %s"), oid_to_hex(oid));
2052 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(type), (uintmax_t)len) + 1;
2053 ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
2054 free(buf);
2055
2056 return ret;
2057 }
2058
2059 int has_object(struct repository *r, const struct object_id *oid,
2060 unsigned flags)
2061 {
2062 int quick = !(flags & HAS_OBJECT_RECHECK_PACKED);
2063 unsigned object_info_flags = OBJECT_INFO_SKIP_FETCH_OBJECT |
2064 (quick ? OBJECT_INFO_QUICK : 0);
2065
2066 if (!startup_info->have_repository)
2067 return 0;
2068 return oid_object_info_extended(r, oid, NULL, object_info_flags) >= 0;
2069 }
2070
2071 int repo_has_object_file_with_flags(struct repository *r,
2072 const struct object_id *oid, int flags)
2073 {
2074 if (!startup_info->have_repository)
2075 return 0;
2076 return oid_object_info_extended(r, oid, NULL, flags) >= 0;
2077 }
2078
2079 int repo_has_object_file(struct repository *r,
2080 const struct object_id *oid)
2081 {
2082 return repo_has_object_file_with_flags(r, oid, 0);
2083 }
2084
2085 static void check_tree(const void *buf, size_t size)
2086 {
2087 struct tree_desc desc;
2088 struct name_entry entry;
2089
2090 init_tree_desc(&desc, buf, size);
2091 while (tree_entry(&desc, &entry))
2092 /* do nothing
2093 * tree_entry() will die() on malformed entries */
2094 ;
2095 }
2096
2097 static void check_commit(const void *buf, size_t size)
2098 {
2099 struct commit c;
2100 memset(&c, 0, sizeof(c));
2101 if (parse_commit_buffer(the_repository, &c, buf, size, 0))
2102 die(_("corrupt commit"));
2103 }
2104
2105 static void check_tag(const void *buf, size_t size)
2106 {
2107 struct tag t;
2108 memset(&t, 0, sizeof(t));
2109 if (parse_tag_buffer(the_repository, &t, buf, size))
2110 die(_("corrupt tag"));
2111 }
2112
2113 static int index_mem(struct index_state *istate,
2114 struct object_id *oid, void *buf, size_t size,
2115 enum object_type type,
2116 const char *path, unsigned flags)
2117 {
2118 int ret, re_allocated = 0;
2119 int write_object = flags & HASH_WRITE_OBJECT;
2120
2121 if (!type)
2122 type = OBJ_BLOB;
2123
2124 /*
2125 * Convert blobs to git internal format
2126 */
2127 if ((type == OBJ_BLOB) && path) {
2128 struct strbuf nbuf = STRBUF_INIT;
2129 if (convert_to_git(istate, path, buf, size, &nbuf,
2130 get_conv_flags(flags))) {
2131 buf = strbuf_detach(&nbuf, &size);
2132 re_allocated = 1;
2133 }
2134 }
2135 if (flags & HASH_FORMAT_CHECK) {
2136 if (type == OBJ_TREE)
2137 check_tree(buf, size);
2138 if (type == OBJ_COMMIT)
2139 check_commit(buf, size);
2140 if (type == OBJ_TAG)
2141 check_tag(buf, size);
2142 }
2143
2144 if (write_object)
2145 ret = write_object_file(buf, size, type_name(type), oid);
2146 else
2147 ret = hash_object_file(the_hash_algo, buf, size,
2148 type_name(type), oid);
2149 if (re_allocated)
2150 free(buf);
2151 return ret;
2152 }
2153
2154 static int index_stream_convert_blob(struct index_state *istate,
2155 struct object_id *oid,
2156 int fd,
2157 const char *path,
2158 unsigned flags)
2159 {
2160 int ret;
2161 const int write_object = flags & HASH_WRITE_OBJECT;
2162 struct strbuf sbuf = STRBUF_INIT;
2163
2164 assert(path);
2165 assert(would_convert_to_git_filter_fd(istate, path));
2166
2167 convert_to_git_filter_fd(istate, path, fd, &sbuf,
2168 get_conv_flags(flags));
2169
2170 if (write_object)
2171 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
2172 oid);
2173 else
2174 ret = hash_object_file(the_hash_algo, sbuf.buf, sbuf.len,
2175 type_name(OBJ_BLOB), oid);
2176 strbuf_release(&sbuf);
2177 return ret;
2178 }
2179
2180 static int index_pipe(struct index_state *istate, struct object_id *oid,
2181 int fd, enum object_type type,
2182 const char *path, unsigned flags)
2183 {
2184 struct strbuf sbuf = STRBUF_INIT;
2185 int ret;
2186
2187 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2188 ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags);
2189 else
2190 ret = -1;
2191 strbuf_release(&sbuf);
2192 return ret;
2193 }
2194
2195 #define SMALL_FILE_SIZE (32*1024)
2196
2197 static int index_core(struct index_state *istate,
2198 struct object_id *oid, int fd, size_t size,
2199 enum object_type type, const char *path,
2200 unsigned flags)
2201 {
2202 int ret;
2203
2204 if (!size) {
2205 ret = index_mem(istate, oid, "", size, type, path, flags);
2206 } else if (size <= SMALL_FILE_SIZE) {
2207 char *buf = xmalloc(size);
2208 ssize_t read_result = read_in_full(fd, buf, size);
2209 if (read_result < 0)
2210 ret = error_errno(_("read error while indexing %s"),
2211 path ? path : "<unknown>");
2212 else if (read_result != size)
2213 ret = error(_("short read while indexing %s"),
2214 path ? path : "<unknown>");
2215 else
2216 ret = index_mem(istate, oid, buf, size, type, path, flags);
2217 free(buf);
2218 } else {
2219 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2220 ret = index_mem(istate, oid, buf, size, type, path, flags);
2221 munmap(buf, size);
2222 }
2223 return ret;
2224 }
2225
2226 /*
2227 * This creates one packfile per large blob unless bulk-checkin
2228 * machinery is "plugged".
2229 *
2230 * This also bypasses the usual "convert-to-git" dance, and that is on
2231 * purpose. We could write a streaming version of the converting
2232 * functions and insert that before feeding the data to fast-import
2233 * (or equivalent in-core API described above). However, that is
2234 * somewhat complicated, as we do not know the size of the filter
2235 * result, which we need to know beforehand when writing a git object.
2236 * Since the primary motivation for trying to stream from the working
2237 * tree file and to avoid mmaping it in core is to deal with large
2238 * binary blobs, they generally do not want to get any conversion, and
2239 * callers should avoid this code path when filters are requested.
2240 */
2241 static int index_stream(struct object_id *oid, int fd, size_t size,
2242 enum object_type type, const char *path,
2243 unsigned flags)
2244 {
2245 return index_bulk_checkin(oid, fd, size, type, path, flags);
2246 }
2247
2248 int index_fd(struct index_state *istate, struct object_id *oid,
2249 int fd, struct stat *st,
2250 enum object_type type, const char *path, unsigned flags)
2251 {
2252 int ret;
2253
2254 /*
2255 * Call xsize_t() only when needed to avoid potentially unnecessary
2256 * die() for large files.
2257 */
2258 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, path))
2259 ret = index_stream_convert_blob(istate, oid, fd, path, flags);
2260 else if (!S_ISREG(st->st_mode))
2261 ret = index_pipe(istate, oid, fd, type, path, flags);
2262 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
2263 (path && would_convert_to_git(istate, path)))
2264 ret = index_core(istate, oid, fd, xsize_t(st->st_size),
2265 type, path, flags);
2266 else
2267 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
2268 flags);
2269 close(fd);
2270 return ret;
2271 }
2272
2273 int index_path(struct index_state *istate, struct object_id *oid,
2274 const char *path, struct stat *st, unsigned flags)
2275 {
2276 int fd;
2277 struct strbuf sb = STRBUF_INIT;
2278 int rc = 0;
2279
2280 switch (st->st_mode & S_IFMT) {
2281 case S_IFREG:
2282 fd = open(path, O_RDONLY);
2283 if (fd < 0)
2284 return error_errno("open(\"%s\")", path);
2285 if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0)
2286 return error(_("%s: failed to insert into database"),
2287 path);
2288 break;
2289 case S_IFLNK:
2290 if (strbuf_readlink(&sb, path, st->st_size))
2291 return error_errno("readlink(\"%s\")", path);
2292 if (!(flags & HASH_WRITE_OBJECT))
2293 hash_object_file(the_hash_algo, sb.buf, sb.len,
2294 blob_type, oid);
2295 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
2296 rc = error(_("%s: failed to insert into database"), path);
2297 strbuf_release(&sb);
2298 break;
2299 case S_IFDIR:
2300 return resolve_gitlink_ref(path, "HEAD", oid);
2301 default:
2302 return error(_("%s: unsupported file type"), path);
2303 }
2304 return rc;
2305 }
2306
2307 int read_pack_header(int fd, struct pack_header *header)
2308 {
2309 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
2310 /* "eof before pack header was fully read" */
2311 return PH_ERROR_EOF;
2312
2313 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2314 /* "protocol error (pack signature mismatch detected)" */
2315 return PH_ERROR_PACK_SIGNATURE;
2316 if (!pack_version_ok(header->hdr_version))
2317 /* "protocol error (pack version unsupported)" */
2318 return PH_ERROR_PROTOCOL;
2319 return 0;
2320 }
2321
2322 void assert_oid_type(const struct object_id *oid, enum object_type expect)
2323 {
2324 enum object_type type = oid_object_info(the_repository, oid, NULL);
2325 if (type < 0)
2326 die(_("%s is not a valid object"), oid_to_hex(oid));
2327 if (type != expect)
2328 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
2329 type_name(expect));
2330 }
2331
2332 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2333 struct strbuf *path,
2334 each_loose_object_fn obj_cb,
2335 each_loose_cruft_fn cruft_cb,
2336 each_loose_subdir_fn subdir_cb,
2337 void *data)
2338 {
2339 size_t origlen, baselen;
2340 DIR *dir;
2341 struct dirent *de;
2342 int r = 0;
2343 struct object_id oid;
2344
2345 if (subdir_nr > 0xff)
2346 BUG("invalid loose object subdirectory: %x", subdir_nr);
2347
2348 origlen = path->len;
2349 strbuf_complete(path, '/');
2350 strbuf_addf(path, "%02x", subdir_nr);
2351
2352 dir = opendir(path->buf);
2353 if (!dir) {
2354 if (errno != ENOENT)
2355 r = error_errno(_("unable to open %s"), path->buf);
2356 strbuf_setlen(path, origlen);
2357 return r;
2358 }
2359
2360 oid.hash[0] = subdir_nr;
2361 strbuf_addch(path, '/');
2362 baselen = path->len;
2363
2364 while ((de = readdir_skip_dot_and_dotdot(dir))) {
2365 size_t namelen;
2366
2367 namelen = strlen(de->d_name);
2368 strbuf_setlen(path, baselen);
2369 strbuf_add(path, de->d_name, namelen);
2370 if (namelen == the_hash_algo->hexsz - 2 &&
2371 !hex_to_bytes(oid.hash + 1, de->d_name,
2372 the_hash_algo->rawsz - 1)) {
2373 oid_set_algo(&oid, the_hash_algo);
2374 if (obj_cb) {
2375 r = obj_cb(&oid, path->buf, data);
2376 if (r)
2377 break;
2378 }
2379 continue;
2380 }
2381
2382 if (cruft_cb) {
2383 r = cruft_cb(de->d_name, path->buf, data);
2384 if (r)
2385 break;
2386 }
2387 }
2388 closedir(dir);
2389
2390 strbuf_setlen(path, baselen - 1);
2391 if (!r && subdir_cb)
2392 r = subdir_cb(subdir_nr, path->buf, data);
2393
2394 strbuf_setlen(path, origlen);
2395
2396 return r;
2397 }
2398
2399 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2400 each_loose_object_fn obj_cb,
2401 each_loose_cruft_fn cruft_cb,
2402 each_loose_subdir_fn subdir_cb,
2403 void *data)
2404 {
2405 int r = 0;
2406 int i;
2407
2408 for (i = 0; i < 256; i++) {
2409 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2410 subdir_cb, data);
2411 if (r)
2412 break;
2413 }
2414
2415 return r;
2416 }
2417
2418 int for_each_loose_file_in_objdir(const char *path,
2419 each_loose_object_fn obj_cb,
2420 each_loose_cruft_fn cruft_cb,
2421 each_loose_subdir_fn subdir_cb,
2422 void *data)
2423 {
2424 struct strbuf buf = STRBUF_INIT;
2425 int r;
2426
2427 strbuf_addstr(&buf, path);
2428 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2429 subdir_cb, data);
2430 strbuf_release(&buf);
2431
2432 return r;
2433 }
2434
2435 int for_each_loose_object(each_loose_object_fn cb, void *data,
2436 enum for_each_object_flags flags)
2437 {
2438 struct object_directory *odb;
2439
2440 prepare_alt_odb(the_repository);
2441 for (odb = the_repository->objects->odb; odb; odb = odb->next) {
2442 int r = for_each_loose_file_in_objdir(odb->path, cb, NULL,
2443 NULL, data);
2444 if (r)
2445 return r;
2446
2447 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2448 break;
2449 }
2450
2451 return 0;
2452 }
2453
2454 static int append_loose_object(const struct object_id *oid, const char *path,
2455 void *data)
2456 {
2457 oid_array_append(data, oid);
2458 return 0;
2459 }
2460
2461 struct oid_array *odb_loose_cache(struct object_directory *odb,
2462 const struct object_id *oid)
2463 {
2464 int subdir_nr = oid->hash[0];
2465 struct strbuf buf = STRBUF_INIT;
2466
2467 if (subdir_nr < 0 ||
2468 subdir_nr >= ARRAY_SIZE(odb->loose_objects_subdir_seen))
2469 BUG("subdir_nr out of range");
2470
2471 if (odb->loose_objects_subdir_seen[subdir_nr])
2472 return &odb->loose_objects_cache[subdir_nr];
2473
2474 strbuf_addstr(&buf, odb->path);
2475 for_each_file_in_obj_subdir(subdir_nr, &buf,
2476 append_loose_object,
2477 NULL, NULL,
2478 &odb->loose_objects_cache[subdir_nr]);
2479 odb->loose_objects_subdir_seen[subdir_nr] = 1;
2480 strbuf_release(&buf);
2481 return &odb->loose_objects_cache[subdir_nr];
2482 }
2483
2484 void odb_clear_loose_cache(struct object_directory *odb)
2485 {
2486 int i;
2487
2488 for (i = 0; i < ARRAY_SIZE(odb->loose_objects_cache); i++)
2489 oid_array_clear(&odb->loose_objects_cache[i]);
2490 memset(&odb->loose_objects_subdir_seen, 0,
2491 sizeof(odb->loose_objects_subdir_seen));
2492 }
2493
2494 static int check_stream_oid(git_zstream *stream,
2495 const char *hdr,
2496 unsigned long size,
2497 const char *path,
2498 const struct object_id *expected_oid)
2499 {
2500 git_hash_ctx c;
2501 struct object_id real_oid;
2502 unsigned char buf[4096];
2503 unsigned long total_read;
2504 int status = Z_OK;
2505
2506 the_hash_algo->init_fn(&c);
2507 the_hash_algo->update_fn(&c, hdr, stream->total_out);
2508
2509 /*
2510 * We already read some bytes into hdr, but the ones up to the NUL
2511 * do not count against the object's content size.
2512 */
2513 total_read = stream->total_out - strlen(hdr) - 1;
2514
2515 /*
2516 * This size comparison must be "<=" to read the final zlib packets;
2517 * see the comment in unpack_loose_rest for details.
2518 */
2519 while (total_read <= size &&
2520 (status == Z_OK ||
2521 (status == Z_BUF_ERROR && !stream->avail_out))) {
2522 stream->next_out = buf;
2523 stream->avail_out = sizeof(buf);
2524 if (size - total_read < stream->avail_out)
2525 stream->avail_out = size - total_read;
2526 status = git_inflate(stream, Z_FINISH);
2527 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2528 total_read += stream->next_out - buf;
2529 }
2530 git_inflate_end(stream);
2531
2532 if (status != Z_STREAM_END) {
2533 error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid));
2534 return -1;
2535 }
2536 if (stream->avail_in) {
2537 error(_("garbage at end of loose object '%s'"),
2538 oid_to_hex(expected_oid));
2539 return -1;
2540 }
2541
2542 the_hash_algo->final_oid_fn(&real_oid, &c);
2543 if (!oideq(expected_oid, &real_oid)) {
2544 error(_("hash mismatch for %s (expected %s)"), path,
2545 oid_to_hex(expected_oid));
2546 return -1;
2547 }
2548
2549 return 0;
2550 }
2551
2552 int read_loose_object(const char *path,
2553 const struct object_id *expected_oid,
2554 enum object_type *type,
2555 unsigned long *size,
2556 void **contents)
2557 {
2558 int ret = -1;
2559 void *map = NULL;
2560 unsigned long mapsize;
2561 git_zstream stream;
2562 char hdr[MAX_HEADER_LEN];
2563
2564 *contents = NULL;
2565
2566 map = map_loose_object_1(the_repository, path, NULL, &mapsize);
2567 if (!map) {
2568 error_errno(_("unable to mmap %s"), path);
2569 goto out;
2570 }
2571
2572 if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2573 error(_("unable to unpack header of %s"), path);
2574 goto out;
2575 }
2576
2577 *type = parse_loose_header(hdr, size);
2578 if (*type < 0) {
2579 error(_("unable to parse header of %s"), path);
2580 git_inflate_end(&stream);
2581 goto out;
2582 }
2583
2584 if (*type == OBJ_BLOB && *size > big_file_threshold) {
2585 if (check_stream_oid(&stream, hdr, *size, path, expected_oid) < 0)
2586 goto out;
2587 } else {
2588 *contents = unpack_loose_rest(&stream, hdr, *size, expected_oid);
2589 if (!*contents) {
2590 error(_("unable to unpack contents of %s"), path);
2591 git_inflate_end(&stream);
2592 goto out;
2593 }
2594 if (check_object_signature(the_repository, expected_oid,
2595 *contents, *size,
2596 type_name(*type))) {
2597 error(_("hash mismatch for %s (expected %s)"), path,
2598 oid_to_hex(expected_oid));
2599 free(*contents);
2600 goto out;
2601 }
2602 }
2603
2604 ret = 0; /* everything checks out */
2605
2606 out:
2607 if (map)
2608 munmap(map, mapsize);
2609 return ret;
2610 }