]> git.ipfire.org Git - thirdparty/git.git/blame - object-file.c
reftable: implement refname validation
[thirdparty/git.git] / object-file.c
CommitLineData
0fcfd160
LT
1/*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
e5afd444 6 * This handles basic git object files - packing, unpacking,
0fcfd160
LT
7 * creation etc.
8 */
0fcfd160 9#include "cache.h"
b2141fc1 10#include "config.h"
6eac50d8 11#include "string-list.h"
697cc8ef 12#include "lockfile.h"
1f688557 13#include "delta.h"
a733cb60 14#include "pack.h"
8e440259
PE
15#include "blob.h"
16#include "commit.h"
4dd1fbc7 17#include "run-command.h"
8e440259
PE
18#include "tag.h"
19#include "tree.h"
c879daa2 20#include "tree-walk.h"
f35a6d3b 21#include "refs.h"
70f5d5d3 22#include "pack-revindex.h"
bc626927 23#include "hash-lookup.h"
568508e7 24#include "bulk-checkin.h"
031dc927 25#include "repository.h"
47f351e9 26#include "replace-object.h"
090ea126 27#include "streaming.h"
543c5caa 28#include "dir.h"
12d95ef6 29#include "list.h"
c4c6effa 30#include "mergesort.h"
cf3c6352 31#include "quote.h"
4f39cd82 32#include "packfile.h"
90c62155 33#include "object-store.h"
b14ed5ad 34#include "promisor-remote.h"
e05db0fd 35
1af64f73 36/* The maximum size for an object header. */
37#define MAX_HEADER_LEN 32
38
e1ccd7e2 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"
13eeedb5 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"
e1ccd7e2 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"
13eeedb5 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"
e1ccd7e2 57
e1ccd7e2 58static const struct object_id empty_tree_oid = {
5a6dce70 59 .hash = EMPTY_TREE_SHA1_BIN_LITERAL,
60 .algo = GIT_HASH_SHA1,
8576fde6 61};
e1ccd7e2 62static const struct object_id empty_blob_oid = {
5a6dce70 63 .hash = EMPTY_BLOB_SHA1_BIN_LITERAL,
64 .algo = GIT_HASH_SHA1,
8576fde6 65};
14228447 66static const struct object_id null_oid_sha1 = {
67 .hash = {0},
68 .algo = GIT_HASH_SHA1,
69};
13eeedb5 70static const struct object_id empty_tree_oid_sha256 = {
5a6dce70 71 .hash = EMPTY_TREE_SHA256_BIN_LITERAL,
72 .algo = GIT_HASH_SHA256,
13eeedb5 73};
74static const struct object_id empty_blob_oid_sha256 = {
5a6dce70 75 .hash = EMPTY_BLOB_SHA256_BIN_LITERAL,
76 .algo = GIT_HASH_SHA256,
13eeedb5 77};
14228447 78static const struct object_id null_oid_sha256 = {
79 .hash = {0},
80 .algo = GIT_HASH_SHA256,
81};
88cd621d 82
ac73cedf 83static void git_hash_sha1_init(git_hash_ctx *ctx)
f50e766b 84{
ac73cedf 85 git_SHA1_Init(&ctx->sha1);
f50e766b 86}
87
768e30ea 88static 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
ac73cedf 93static void git_hash_sha1_update(git_hash_ctx *ctx, const void *data, size_t len)
f50e766b 94{
ac73cedf 95 git_SHA1_Update(&ctx->sha1, data, len);
f50e766b 96}
97
ac73cedf 98static void git_hash_sha1_final(unsigned char *hash, git_hash_ctx *ctx)
f50e766b 99{
ac73cedf 100 git_SHA1_Final(hash, &ctx->sha1);
f50e766b 101}
102
ab795f0d 103static 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);
5a6dce70 107 oid->algo = GIT_HASH_SHA1;
ab795f0d 108}
109
13eeedb5 110
111static void git_hash_sha256_init(git_hash_ctx *ctx)
112{
113 git_SHA256_Init(&ctx->sha256);
114}
115
768e30ea 116static 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
13eeedb5 121static 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
126static void git_hash_sha256_final(unsigned char *hash, git_hash_ctx *ctx)
127{
128 git_SHA256_Final(hash, &ctx->sha256);
129}
130
ab795f0d 131static 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);
5a6dce70 139 oid->algo = GIT_HASH_SHA256;
ab795f0d 140}
141
ac73cedf 142static void git_hash_unknown_init(git_hash_ctx *ctx)
f50e766b 143{
1a07e59c 144 BUG("trying to init unknown hash");
f50e766b 145}
146
768e30ea 147static void git_hash_unknown_clone(git_hash_ctx *dst, const git_hash_ctx *src)
148{
149 BUG("trying to clone unknown hash");
150}
151
ac73cedf 152static void git_hash_unknown_update(git_hash_ctx *ctx, const void *data, size_t len)
f50e766b 153{
1a07e59c 154 BUG("trying to update unknown hash");
f50e766b 155}
156
ac73cedf 157static void git_hash_unknown_final(unsigned char *hash, git_hash_ctx *ctx)
f50e766b 158{
1a07e59c 159 BUG("trying to finalize unknown hash");
f50e766b 160}
161
ab795f0d 162static void git_hash_unknown_final_oid(struct object_id *oid, git_hash_ctx *ctx)
163{
164 BUG("trying to finalize unknown hash");
165}
166
f50e766b 167const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
168 {
169 NULL,
170 0x00000000,
171 0,
172 0,
a2ce0a75 173 0,
f50e766b 174 git_hash_unknown_init,
768e30ea 175 git_hash_unknown_clone,
f50e766b 176 git_hash_unknown_update,
177 git_hash_unknown_final,
ab795f0d 178 git_hash_unknown_final_oid,
f50e766b 179 NULL,
180 NULL,
14228447 181 NULL,
f50e766b 182 },
183 {
1ccf07cb 184 "sha1",
27f3796a 185 GIT_SHA1_FORMAT_ID,
f50e766b 186 GIT_SHA1_RAWSZ,
187 GIT_SHA1_HEXSZ,
a2ce0a75 188 GIT_SHA1_BLKSZ,
f50e766b 189 git_hash_sha1_init,
768e30ea 190 git_hash_sha1_clone,
f50e766b 191 git_hash_sha1_update,
192 git_hash_sha1_final,
ab795f0d 193 git_hash_sha1_final_oid,
f50e766b 194 &empty_tree_oid,
195 &empty_blob_oid,
14228447 196 &null_oid_sha1,
f50e766b 197 },
13eeedb5 198 {
199 "sha256",
27f3796a 200 GIT_SHA256_FORMAT_ID,
13eeedb5 201 GIT_SHA256_RAWSZ,
202 GIT_SHA256_HEXSZ,
203 GIT_SHA256_BLKSZ,
204 git_hash_sha256_init,
768e30ea 205 git_hash_sha256_clone,
13eeedb5 206 git_hash_sha256_update,
207 git_hash_sha256_final,
ab795f0d 208 git_hash_sha256_final_oid,
13eeedb5 209 &empty_tree_oid_sha256,
210 &empty_blob_oid_sha256,
14228447 211 &null_oid_sha256,
13eeedb5 212 }
f50e766b 213};
214
14228447 215const struct object_id *null_oid(void)
216{
217 return the_hash_algo->null_oid;
218}
219
d8a92ced 220const char *empty_tree_oid_hex(void)
221{
222 static char buf[GIT_MAX_HEXSZ + 1];
223 return oid_to_hex_r(buf, the_hash_algo->empty_tree);
224}
225
226const char *empty_blob_oid_hex(void)
227{
228 static char buf[GIT_MAX_HEXSZ + 1];
229 return oid_to_hex_r(buf, the_hash_algo->empty_blob);
230}
231
2f90b9d9 232int hash_algo_by_name(const char *name)
233{
234 int i;
235 if (!name)
236 return GIT_HASH_UNKNOWN;
237 for (i = 1; i < GIT_HASH_NALGOS; i++)
238 if (!strcmp(name, hash_algos[i].name))
239 return i;
240 return GIT_HASH_UNKNOWN;
241}
242
243int hash_algo_by_id(uint32_t format_id)
244{
245 int i;
246 for (i = 1; i < GIT_HASH_NALGOS; i++)
247 if (format_id == hash_algos[i].format_id)
248 return i;
249 return GIT_HASH_UNKNOWN;
250}
251
95399788 252int hash_algo_by_length(int len)
253{
254 int i;
255 for (i = 1; i < GIT_HASH_NALGOS; i++)
256 if (len == hash_algos[i].rawsz)
257 return i;
258 return GIT_HASH_UNKNOWN;
259}
2f90b9d9 260
c597ba80
NTND
261/*
262 * This is meant to hold a *small* number of objects that you would
cb1c8d1d 263 * want read_object_file() to be able to return, but yet you do not want
c597ba80
NTND
264 * to write them into the object store (e.g. a browse-only
265 * application).
266 */
267static struct cached_object {
62ba93ea 268 struct object_id oid;
c597ba80
NTND
269 enum object_type type;
270 void *buf;
271 unsigned long size;
272} *cached_objects;
273static int cached_object_nr, cached_object_alloc;
274
275static struct cached_object empty_tree = {
62ba93ea 276 { EMPTY_TREE_SHA1_BIN_LITERAL },
c597ba80
NTND
277 OBJ_TREE,
278 "",
279 0
280};
281
62ba93ea 282static struct cached_object *find_cached_object(const struct object_id *oid)
c597ba80
NTND
283{
284 int i;
285 struct cached_object *co = cached_objects;
286
287 for (i = 0; i < cached_object_nr; i++, co++) {
4a7e27e9 288 if (oideq(&co->oid, oid))
c597ba80
NTND
289 return co;
290 }
4a7e27e9 291 if (oideq(oid, the_hash_algo->empty_tree))
c597ba80
NTND
292 return &empty_tree;
293 return NULL;
294}
295
9472935d 296
8462ff43 297static int get_conv_flags(unsigned flags)
9472935d
TB
298{
299 if (flags & HASH_RENORMALIZE)
8462ff43 300 return CONV_EOL_RENORMALIZE;
9472935d 301 else if (flags & HASH_WRITE_OBJECT)
107642fe 302 return global_conv_flags_eol | CONV_WRITE_OBJECT;
9472935d 303 else
8462ff43 304 return 0;
9472935d
TB
305}
306
307
90a6464b
JH
308int mkdir_in_gitdir(const char *path)
309{
310 if (mkdir(path, 0777)) {
311 int saved_errno = errno;
312 struct stat st;
313 struct strbuf sb = STRBUF_INIT;
314
315 if (errno != EEXIST)
316 return -1;
317 /*
318 * Are we looking at a path in a symlinked worktree
319 * whose original repository does not yet have it?
320 * e.g. .git/rr-cache pointing at its original
321 * repository in which the user hasn't performed any
322 * conflict resolution yet?
323 */
324 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
325 strbuf_readlink(&sb, path, st.st_size) ||
326 !is_absolute_path(sb.buf) ||
327 mkdir(sb.buf, 0777)) {
328 strbuf_release(&sb);
329 errno = saved_errno;
330 return -1;
331 }
332 strbuf_release(&sb);
333 }
334 return adjust_shared_perm(path);
335}
336
eb3c027e 337static enum scld_error safe_create_leading_directories_1(char *path, int share)
b2cb9425 338{
26c8ae2a 339 char *next_component = path + offset_1st_component(path);
0be0521b 340 enum scld_error ret = SCLD_OK;
67d42212 341
0be0521b 342 while (ret == SCLD_OK && next_component) {
f0502332 343 struct stat st;
0f527403 344 char *slash = next_component, slash_character;
f0502332 345
0f527403
MH
346 while (*slash && !is_dir_sep(*slash))
347 slash++;
348
349 if (!*slash)
b2cb9425 350 break;
bf10cf70 351
26c8ae2a 352 next_component = slash + 1;
0f527403 353 while (is_dir_sep(*next_component))
bf10cf70 354 next_component++;
26c8ae2a 355 if (!*next_component)
5f0bdf50 356 break;
831651fd 357
0f527403 358 slash_character = *slash;
831651fd 359 *slash = '\0';
67d42212
JR
360 if (!stat(path, &st)) {
361 /* path exists */
204a047f
MH
362 if (!S_ISDIR(st.st_mode)) {
363 errno = ENOTDIR;
0be0521b 364 ret = SCLD_EXISTS;
204a047f 365 }
53a39721 366 } else if (mkdir(path, 0777)) {
928734d9 367 if (errno == EEXIST &&
9e6f885d 368 !stat(path, &st) && S_ISDIR(st.st_mode))
928734d9 369 ; /* somebody created it since we checked */
18d37e86
MH
370 else if (errno == ENOENT)
371 /*
372 * Either mkdir() failed because
373 * somebody just pruned the containing
374 * directory, or stat() failed because
375 * the file that was in our way was
376 * just removed. Either way, inform
377 * the caller that it might be worth
378 * trying again:
379 */
380 ret = SCLD_VANISHED;
9e6f885d 381 else
0be0521b 382 ret = SCLD_FAILED;
eb3c027e 383 } else if (share && adjust_shared_perm(path)) {
0be0521b 384 ret = SCLD_PERMS;
457f06d6 385 }
0f527403 386 *slash = slash_character;
b2cb9425 387 }
9e6f885d 388 return ret;
b2cb9425 389}
723c31fe 390
eb3c027e
MT
391enum scld_error safe_create_leading_directories(char *path)
392{
393 return safe_create_leading_directories_1(path, 1);
394}
395
396enum scld_error safe_create_leading_directories_no_share(char *path)
397{
398 return safe_create_leading_directories_1(path, 0);
399}
400
0be0521b 401enum scld_error safe_create_leading_directories_const(const char *path)
8e21d63b 402{
02944307 403 int save_errno;
8e21d63b
JK
404 /* path points to cache entries, so xstrdup before messing with it */
405 char *buf = xstrdup(path);
0be0521b 406 enum scld_error result = safe_create_leading_directories(buf);
02944307
MH
407
408 save_errno = errno;
8e21d63b 409 free(buf);
02944307 410 errno = save_errno;
8e21d63b
JK
411 return result;
412}
413
177978f5
MH
414int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
415{
416 /*
417 * The number of times we will try to remove empty directories
418 * in the way of path. This is only 1 because if another
419 * process is racily creating directories that conflict with
420 * us, we don't want to fight against them.
421 */
422 int remove_directories_remaining = 1;
423
424 /*
425 * The number of times that we will try to create the
426 * directories containing path. We are willing to attempt this
427 * more than once, because another process could be trying to
428 * clean up empty directories at the same time as we are
429 * trying to create them.
430 */
431 int create_directories_remaining = 3;
432
433 /* A scratch copy of path, filled lazily if we need it: */
434 struct strbuf path_copy = STRBUF_INIT;
435
436 int ret, save_errno;
437
438 /* Sanity check: */
439 assert(*path);
440
441retry_fn:
442 ret = fn(path, cb);
443 save_errno = errno;
444 if (!ret)
445 goto out;
446
447 if (errno == EISDIR && remove_directories_remaining-- > 0) {
448 /*
449 * A directory is in the way. Maybe it is empty; try
450 * to remove it:
451 */
452 if (!path_copy.len)
453 strbuf_addstr(&path_copy, path);
454
455 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
456 goto retry_fn;
457 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
458 /*
459 * Maybe the containing directory didn't exist, or
460 * maybe it was just deleted by a process that is
461 * racing with us to clean up empty directories. Try
462 * to create it:
463 */
464 enum scld_error scld_result;
465
466 if (!path_copy.len)
467 strbuf_addstr(&path_copy, path);
468
469 do {
470 scld_result = safe_create_leading_directories(path_copy.buf);
471 if (scld_result == SCLD_OK)
472 goto retry_fn;
473 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
474 }
475
476out:
477 strbuf_release(&path_copy);
478 errno = save_errno;
479 return ret;
480}
481
514c5fdd 482static void fill_loose_path(struct strbuf *buf, const struct object_id *oid)
ace1534d
JH
483{
484 int i;
94b5e093 485 for (i = 0; i < the_hash_algo->rawsz; i++) {
ace1534d 486 static char hex[] = "0123456789abcdef";
514c5fdd 487 unsigned int val = oid->hash[i];
f7b7774f
JK
488 strbuf_addch(buf, hex[val >> 4]);
489 strbuf_addch(buf, hex[val & 0xf]);
afbba2f0 490 if (!i)
f7b7774f 491 strbuf_addch(buf, '/');
ace1534d
JH
492 }
493}
494
f0eaf638
JK
495static const char *odb_loose_path(struct object_directory *odb,
496 struct strbuf *buf,
514c5fdd 497 const struct object_id *oid)
0fcfd160 498{
b69fb867 499 strbuf_reset(buf);
f0eaf638 500 strbuf_addstr(buf, odb->path);
34498471 501 strbuf_addch(buf, '/');
514c5fdd 502 fill_loose_path(buf, oid);
f3f043a1 503 return buf->buf;
0fcfd160
LT
504}
505
f3f043a1 506const char *loose_object_path(struct repository *r, struct strbuf *buf,
514c5fdd 507 const struct object_id *oid)
29ec6af2 508{
514c5fdd 509 return odb_loose_path(r->objects->odb, buf, oid);
0fcfd160
LT
510}
511
4ea82473
JK
512/*
513 * Return non-zero iff the path is usable as an alternate object database.
514 */
13313fc3
SB
515static int alt_odb_usable(struct raw_object_store *o,
516 struct strbuf *path,
cf2dc1c2 517 const char *normalized_objdir, khiter_t *pos)
4ea82473 518{
cf2dc1c2 519 int r;
4ea82473
JK
520
521 /* Detect cases where alternate disappeared */
522 if (!is_directory(path->buf)) {
259328b7
NTND
523 error(_("object directory %s does not exist; "
524 "check .git/objects/info/alternates"),
4ea82473
JK
525 path->buf);
526 return 0;
527 }
528
529 /*
530 * Prevent the common mistake of listing the same
531 * thing twice, or object directory itself.
532 */
cf2dc1c2
EW
533 if (!o->odb_by_path) {
534 khiter_t p;
535
536 o->odb_by_path = kh_init_odb_path_map();
537 assert(!o->odb->next);
538 p = kh_put_odb_path_map(o->odb_by_path, o->odb->path, &r);
539 assert(r == 1); /* never used */
540 kh_value(o->odb_by_path, p) = o->odb;
4ea82473 541 }
cf2dc1c2 542 if (fspatheq(path->buf, normalized_objdir))
4ea82473 543 return 0;
cf2dc1c2
EW
544 *pos = kh_put_odb_path_map(o->odb_by_path, path->buf, &r);
545 /* r: 0 = exists, 1 = never used, 2 = deleted */
546 return r == 0 ? 0 : 1;
4ea82473
JK
547}
548
ddd5d056
JH
549/*
550 * Prepare alternate object database registry.
d5a63b99
JH
551 *
552 * The variable alt_odb_list points at the list of struct
263db403 553 * object_directory. The elements on this list come from
d5a63b99
JH
554 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
555 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
556 * whose contents is similar to that environment variable but can be
557 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
558 * contains "/the/directory/corresponding/to/.git/objects/...", while
559 * its name points just after the slash at the end of ".git/objects/"
e5afd444
560 * in the example above, and has enough space to hold all hex characters
561 * of the object ID, an extra slash for the first level indirection, and
562 * the terminating NUL.
ddd5d056 563 */
77f012e8
SB
564static void read_info_alternates(struct repository *r,
565 const char *relative_base,
566 int depth);
407532f8 567static int link_alt_odb_entry(struct repository *r, const struct strbuf *entry,
cfc62fc9 568 const char *relative_base, int depth, const char *normalized_objdir)
ace1534d 569{
263db403 570 struct object_directory *ent;
5bdf0a84 571 struct strbuf pathbuf = STRBUF_INIT;
cf2dc1c2 572 khiter_t pos;
d5a63b99 573
407532f8 574 if (!is_absolute_path(entry->buf) && relative_base) {
4ac9006f 575 strbuf_realpath(&pathbuf, relative_base, 1);
5bdf0a84 576 strbuf_addch(&pathbuf, '/');
c2f493a4 577 }
407532f8 578 strbuf_addbuf(&pathbuf, entry);
c2f493a4 579
37a95862 580 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
259328b7 581 error(_("unable to normalize alternate object path: %s"),
670c359d
JK
582 pathbuf.buf);
583 strbuf_release(&pathbuf);
584 return -1;
585 }
5bdf0a84
HW
586
587 /*
588 * The trailing slash after the directory name is given by
589 * this function at the end. Remove duplicates.
590 */
4ea82473
JK
591 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
592 strbuf_setlen(&pathbuf, pathbuf.len - 1);
5bdf0a84 593
cf2dc1c2 594 if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir, &pos)) {
4ea82473 595 strbuf_release(&pathbuf);
c2f493a4
MW
596 return -1;
597 }
598
ca56dadb 599 CALLOC_ARRAY(ent, 1);
cf2dc1c2
EW
600 /* pathbuf.buf is already in r->objects->odb_by_path */
601 ent->path = strbuf_detach(&pathbuf, NULL);
c2f493a4
MW
602
603 /* add the alternate entry */
f0eaf638
JK
604 *r->objects->odb_tail = ent;
605 r->objects->odb_tail = &(ent->next);
c2f493a4 606 ent->next = NULL;
cf2dc1c2
EW
607 assert(r->objects->odb_by_path);
608 kh_value(r->objects->odb_by_path, pos) = ent;
c2f493a4
MW
609
610 /* recursively add alternates */
cf2dc1c2 611 read_info_alternates(r, ent->path, depth + 1);
c2f493a4
MW
612
613 return 0;
614}
615
cf3c6352
JK
616static const char *parse_alt_odb_entry(const char *string,
617 int sep,
618 struct strbuf *out)
619{
620 const char *end;
621
622 strbuf_reset(out);
623
624 if (*string == '#') {
625 /* comment; consume up to next separator */
626 end = strchrnul(string, sep);
627 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
628 /*
629 * quoted path; unquote_c_style has copied the
630 * data for us and set "end". Broken quoting (e.g.,
631 * an entry that doesn't end with a quote) falls
632 * back to the unquoted case below.
633 */
634 } else {
635 /* normal, unquoted path */
636 end = strchrnul(string, sep);
637 strbuf_add(out, string, end - string);
638 }
639
640 if (*end)
641 end++;
642 return end;
643}
644
77f012e8
SB
645static void link_alt_odb_entries(struct repository *r, const char *alt,
646 int sep, const char *relative_base, int depth)
c2f493a4 647{
539e7506 648 struct strbuf objdirbuf = STRBUF_INIT;
cf3c6352 649 struct strbuf entry = STRBUF_INIT;
c2f493a4 650
f28e3668
JK
651 if (!alt || !*alt)
652 return;
653
c2f493a4 654 if (depth > 5) {
259328b7 655 error(_("%s: ignoring alternate object stores, nesting too deep"),
c2f493a4
MW
656 relative_base);
657 return;
658 }
659
f0eaf638 660 strbuf_add_absolute_path(&objdirbuf, r->objects->odb->path);
670c359d 661 if (strbuf_normalize_path(&objdirbuf) < 0)
259328b7 662 die(_("unable to normalize object directory: %s"),
670c359d 663 objdirbuf.buf);
539e7506 664
cf3c6352
JK
665 while (*alt) {
666 alt = parse_alt_odb_entry(alt, sep, &entry);
667 if (!entry.len)
9577e7e3 668 continue;
407532f8 669 link_alt_odb_entry(r, &entry,
cfc62fc9 670 relative_base, depth, objdirbuf.buf);
9577e7e3 671 }
cf3c6352 672 strbuf_release(&entry);
539e7506 673 strbuf_release(&objdirbuf);
d5a63b99
JH
674}
675
77f012e8
SB
676static void read_info_alternates(struct repository *r,
677 const char *relative_base,
678 int depth)
d5a63b99 679{
5015f01c 680 char *path;
dc732bd5 681 struct strbuf buf = STRBUF_INIT;
d5a63b99 682
5015f01c 683 path = xstrfmt("%s/info/alternates", relative_base);
dc732bd5 684 if (strbuf_read_file(&buf, path, 1024) < 0) {
f0f7bebe 685 warn_on_fopen_errors(path);
dc732bd5 686 free(path);
9a217f2a 687 return;
ace1534d 688 }
d5a63b99 689
77f012e8 690 link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
dc732bd5
JK
691 strbuf_release(&buf);
692 free(path);
ace1534d
JH
693}
694
bef70b22
DB
695void add_to_alternates_file(const char *reference)
696{
f132a127 697 struct lock_file lock = LOCK_INIT;
77b9b1d1
JK
698 char *alts = git_pathdup("objects/info/alternates");
699 FILE *in, *out;
f132a127 700 int found = 0;
77b9b1d1 701
f132a127
702 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
703 out = fdopen_lock_file(&lock, "w");
77b9b1d1 704 if (!out)
259328b7 705 die_errno(_("unable to fdopen alternates lockfile"));
77b9b1d1
JK
706
707 in = fopen(alts, "r");
708 if (in) {
709 struct strbuf line = STRBUF_INIT;
77b9b1d1 710
3f163962 711 while (strbuf_getline(&line, in) != EOF) {
77b9b1d1
JK
712 if (!strcmp(reference, line.buf)) {
713 found = 1;
714 break;
715 }
716 fprintf_or_die(out, "%s\n", line.buf);
717 }
718
719 strbuf_release(&line);
720 fclose(in);
77b9b1d1
JK
721 }
722 else if (errno != ENOENT)
259328b7 723 die_errno(_("unable to read alternates file"));
77b9b1d1 724
f132a127
725 if (found) {
726 rollback_lock_file(&lock);
727 } else {
77b9b1d1 728 fprintf_or_die(out, "%s\n", reference);
f132a127 729 if (commit_lock_file(&lock))
259328b7 730 die_errno(_("unable to move new alternates file into place"));
f0eaf638 731 if (the_repository->objects->loaded_alternates)
93d8d1e2
SB
732 link_alt_odb_entries(the_repository, reference,
733 '\n', NULL, 0);
77b9b1d1
JK
734 }
735 free(alts);
bef70b22
DB
736}
737
a5b34d21
JK
738void add_to_alternates_memory(const char *reference)
739{
740 /*
741 * Make sure alternates are initialized, or else our entry may be
742 * overwritten when they are.
743 */
0b209034 744 prepare_alt_odb(the_repository);
a5b34d21 745
93d8d1e2
SB
746 link_alt_odb_entries(the_repository, reference,
747 '\n', NULL, 0);
a5b34d21
JK
748}
749
9eeea7d2
SB
750/*
751 * Compute the exact path an alternate is at and returns it. In case of
752 * error NULL is returned and the human readable error is added to `err`
efde7b72 753 * `path` may be relative and should point to $GIT_DIR.
9eeea7d2
SB
754 * `err` must not be null.
755 */
756char *compute_alternate_path(const char *path, struct strbuf *err)
757{
758 char *ref_git = NULL;
4530a85b 759 const char *repo;
9eeea7d2
SB
760 int seen_error = 0;
761
4530a85b
AM
762 ref_git = real_pathdup(path, 0);
763 if (!ref_git) {
9eeea7d2
SB
764 seen_error = 1;
765 strbuf_addf(err, _("path '%s' does not exist"), path);
766 goto out;
4530a85b 767 }
9eeea7d2
SB
768
769 repo = read_gitfile(ref_git);
770 if (!repo)
771 repo = read_gitfile(mkpath("%s/.git", ref_git));
772 if (repo) {
773 free(ref_git);
774 ref_git = xstrdup(repo);
775 }
776
777 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
778 char *ref_git_git = mkpathdup("%s/.git", ref_git);
779 free(ref_git);
780 ref_git = ref_git_git;
781 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
782 struct strbuf sb = STRBUF_INIT;
783 seen_error = 1;
784 if (get_common_dir(&sb, ref_git)) {
785 strbuf_addf(err,
786 _("reference repository '%s' as a linked "
787 "checkout is not supported yet."),
788 path);
789 goto out;
790 }
791
792 strbuf_addf(err, _("reference repository '%s' is not a "
793 "local repository."), path);
794 goto out;
795 }
796
797 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
798 strbuf_addf(err, _("reference repository '%s' is shallow"),
799 path);
800 seen_error = 1;
801 goto out;
802 }
803
804 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
805 strbuf_addf(err,
806 _("reference repository '%s' is grafted"),
807 path);
808 seen_error = 1;
809 goto out;
810 }
811
812out:
813 if (seen_error) {
6a83d902 814 FREE_AND_NULL(ref_git);
9eeea7d2
SB
815 }
816
817 return ref_git;
818}
819
709dfa69
JK
820static void fill_alternate_refs_command(struct child_process *cmd,
821 const char *repo_path)
822{
823 const char *value;
824
825 if (!git_config_get_value("core.alternateRefsCommand", &value)) {
826 cmd->use_shell = 1;
827
c972bf4c
JK
828 strvec_push(&cmd->args, value);
829 strvec_push(&cmd->args, repo_path);
709dfa69
JK
830 } else {
831 cmd->git_cmd = 1;
832
c972bf4c
JK
833 strvec_pushf(&cmd->args, "--git-dir=%s", repo_path);
834 strvec_push(&cmd->args, "for-each-ref");
835 strvec_push(&cmd->args, "--format=%(objectname)");
709dfa69
JK
836
837 if (!git_config_get_value("core.alternateRefsPrefixes", &value)) {
c972bf4c
JK
838 strvec_push(&cmd->args, "--");
839 strvec_split(&cmd->args, value);
709dfa69
JK
840 }
841 }
842
843 cmd->env = local_repo_env;
844 cmd->out = -1;
845}
846
847static void read_alternate_refs(const char *path,
848 alternate_ref_fn *cb,
849 void *data)
850{
851 struct child_process cmd = CHILD_PROCESS_INIT;
852 struct strbuf line = STRBUF_INIT;
853 FILE *fh;
854
855 fill_alternate_refs_command(&cmd, path);
856
857 if (start_command(&cmd))
858 return;
859
860 fh = xfdopen(cmd.out, "r");
861 while (strbuf_getline_lf(&line, fh) != EOF) {
862 struct object_id oid;
863 const char *p;
864
865 if (parse_oid_hex(line.buf, &oid, &p) || *p) {
866 warning(_("invalid line while parsing alternate refs: %s"),
867 line.buf);
868 break;
869 }
870
871 cb(&oid, data);
872 }
873
874 fclose(fh);
875 finish_command(&cmd);
86ad3ea5 876 strbuf_release(&line);
709dfa69
JK
877}
878
879struct alternate_refs_data {
880 alternate_ref_fn *fn;
881 void *data;
882};
883
884static int refs_from_alternate_cb(struct object_directory *e,
885 void *data)
886{
887 struct strbuf path = STRBUF_INIT;
888 size_t base_len;
889 struct alternate_refs_data *cb = data;
890
891 if (!strbuf_realpath(&path, e->path, 0))
892 goto out;
893 if (!strbuf_strip_suffix(&path, "/objects"))
894 goto out;
895 base_len = path.len;
896
897 /* Is this a git repository with refs? */
898 strbuf_addstr(&path, "/refs");
899 if (!is_directory(path.buf))
900 goto out;
901 strbuf_setlen(&path, base_len);
902
903 read_alternate_refs(path.buf, cb->fn, cb->data);
904
905out:
906 strbuf_release(&path);
907 return 0;
908}
909
910void for_each_alternate_ref(alternate_ref_fn fn, void *data)
911{
912 struct alternate_refs_data cb;
913 cb.fn = fn;
914 cb.data = data;
915 foreach_alt_odb(refs_from_alternate_cb, &cb);
916}
917
fe1b2268 918int foreach_alt_odb(alt_odb_fn fn, void *cb)
d79796bc 919{
263db403 920 struct object_directory *ent;
fe1b2268 921 int r = 0;
d79796bc 922
0b209034 923 prepare_alt_odb(the_repository);
f0eaf638 924 for (ent = the_repository->objects->odb->next; ent; ent = ent->next) {
fe1b2268
JK
925 r = fn(ent, cb);
926 if (r)
927 break;
928 }
929 return r;
d79796bc
JH
930}
931
13068bf0 932void prepare_alt_odb(struct repository *r)
c2f493a4 933{
f0eaf638 934 if (r->objects->loaded_alternates)
7dc24aa5
SP
935 return;
936
13068bf0 937 link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
c2f493a4 938
f0eaf638
JK
939 read_info_alternates(r, r->objects->odb->path, 0);
940 r->objects->loaded_alternates = 1;
c2f493a4
MW
941}
942
3096b2ec 943/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
33d4221c 944static int freshen_file(const char *fn)
ace1534d 945{
312cd761 946 return !utime(fn, NULL);
0f4dc14a 947}
ace1534d 948
3096b2ec
JK
949/*
950 * All of the check_and_freshen functions return 1 if the file exists and was
951 * freshened (if freshening was requested), 0 otherwise. If they return
952 * 0, you should not assume that it is safe to skip a write of the object (it
953 * either does not exist on disk, or has a stale mtime and may be subject to
954 * pruning).
955 */
6a5e6f5e 956int check_and_freshen_file(const char *fn, int freshen)
33d4221c
JK
957{
958 if (access(fn, F_OK))
959 return 0;
3096b2ec 960 if (freshen && !freshen_file(fn))
33d4221c
JK
961 return 0;
962 return 1;
963}
964
f0eaf638
JK
965static int check_and_freshen_odb(struct object_directory *odb,
966 const struct object_id *oid,
967 int freshen)
33d4221c 968{
f0eaf638 969 static struct strbuf path = STRBUF_INIT;
514c5fdd 970 odb_loose_path(odb, &path, oid);
f0eaf638
JK
971 return check_and_freshen_file(path.buf, freshen);
972}
ea657730 973
f0eaf638
JK
974static int check_and_freshen_local(const struct object_id *oid, int freshen)
975{
976 return check_and_freshen_odb(the_repository->objects->odb, oid, freshen);
33d4221c
JK
977}
978
6862ebbf 979static int check_and_freshen_nonlocal(const struct object_id *oid, int freshen)
0f4dc14a 980{
263db403 981 struct object_directory *odb;
f3f043a1 982
0b209034 983 prepare_alt_odb(the_repository);
f0eaf638
JK
984 for (odb = the_repository->objects->odb->next; odb; odb = odb->next) {
985 if (check_and_freshen_odb(odb, oid, freshen))
c529d75a 986 return 1;
ace1534d 987 }
c529d75a 988 return 0;
ace1534d
JH
989}
990
6862ebbf 991static int check_and_freshen(const struct object_id *oid, int freshen)
33d4221c 992{
6862ebbf 993 return check_and_freshen_local(oid, freshen) ||
994 check_and_freshen_nonlocal(oid, freshen);
33d4221c
JK
995}
996
6862ebbf 997int has_loose_object_nonlocal(const struct object_id *oid)
33d4221c 998{
6862ebbf 999 return check_and_freshen_nonlocal(oid, 0);
33d4221c
JK
1000}
1001
6862ebbf 1002static int has_loose_object(const struct object_id *oid)
0f4dc14a 1003{
6862ebbf 1004 return check_and_freshen(oid, 0);
0f4dc14a
BC
1005}
1006
02710228
SP
1007static void mmap_limit_check(size_t length)
1008{
1009 static size_t limit = 0;
1010 if (!limit) {
1011 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
1012 if (!limit)
1013 limit = SIZE_MAX;
1014 }
1015 if (length > limit)
259328b7 1016 die(_("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX),
02710228
SP
1017 (uintmax_t)length, (uintmax_t)limit);
1018}
1019
1570856b
JK
1020void *xmmap_gently(void *start, size_t length,
1021 int prot, int flags, int fd, off_t offset)
58ecbd5e 1022{
02710228
SP
1023 void *ret;
1024
1025 mmap_limit_check(length);
1026 ret = mmap(start, length, prot, flags, fd, offset);
9827d4c1
JK
1027 if (ret == MAP_FAILED && !length)
1028 ret = NULL;
58ecbd5e
JN
1029 return ret;
1030}
1031
dc059294
EW
1032const char *mmap_os_err(void)
1033{
1034 static const char blank[] = "";
1035#if defined(__linux__)
1036 if (errno == ENOMEM) {
1037 /* this continues an existing error message: */
1038 static const char enomem[] =
1039", check sys.vm.max_map_count and/or RLIMIT_DATA";
1040 return enomem;
1041 }
1042#endif /* OS-specific bits */
1043 return blank;
1044}
1045
1570856b
JK
1046void *xmmap(void *start, size_t length,
1047 int prot, int flags, int fd, off_t offset)
1048{
1049 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
1050 if (ret == MAP_FAILED)
dc059294 1051 die_errno(_("mmap failed%s"), mmap_os_err());
1570856b
JK
1052 return ret;
1053}
1054
88d0db55 1055/*
090ea126 1056 * With an in-core object data in "map", rehash it to make sure the
cb1c8d1d
JK
1057 * object name actually matches "oid" to detect object corruption.
1058 * With "map" == NULL, try reading the object named with "oid" using
090ea126 1059 * the streaming interface and rehash it to do the same.
88d0db55 1060 */
b98d1885
MT
1061int check_object_signature(struct repository *r, const struct object_id *oid,
1062 void *map, unsigned long size, const char *type)
88d0db55 1063{
f070facc 1064 struct object_id real_oid;
090ea126
NTND
1065 enum object_type obj_type;
1066 struct git_istream *st;
18e2588e 1067 git_hash_ctx c;
1af64f73 1068 char hdr[MAX_HEADER_LEN];
090ea126 1069 int hdrlen;
88d0db55 1070
090ea126 1071 if (map) {
b98d1885 1072 hash_object_file(r->hash_algo, map, size, type, &real_oid);
9001dc2a 1073 return !oideq(oid, &real_oid) ? -1 : 0;
88d0db55
BC
1074 }
1075
b98d1885 1076 st = open_istream(r, oid, &obj_type, &size, NULL);
090ea126
NTND
1077 if (!st)
1078 return -1;
88d0db55 1079
090ea126 1080 /* Generate the header */
ca473cef 1081 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(obj_type), (uintmax_t)size) + 1;
88d0db55 1082
090ea126 1083 /* Sha1.. */
b98d1885
MT
1084 r->hash_algo->init_fn(&c);
1085 r->hash_algo->update_fn(&c, hdr, hdrlen);
090ea126
NTND
1086 for (;;) {
1087 char buf[1024 * 16];
1088 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1f688557 1089
f54fac53
JK
1090 if (readlen < 0) {
1091 close_istream(st);
1092 return -1;
1093 }
090ea126
NTND
1094 if (!readlen)
1095 break;
b98d1885 1096 r->hash_algo->update_fn(&c, buf, readlen);
fa5fc15d 1097 }
5951bf46 1098 r->hash_algo->final_oid_fn(&real_oid, &c);
090ea126 1099 close_istream(st);
9001dc2a 1100 return !oideq(oid, &real_oid) ? -1 : 0;
fa5fc15d
SP
1101}
1102
1b8ac5ea 1103int git_open_cloexec(const char *name, int flags)
a0788266 1104{
1e3001a8
JH
1105 int fd;
1106 static int o_cloexec = O_CLOEXEC;
a0788266 1107
1e3001a8
JH
1108 fd = open(name, flags | o_cloexec);
1109 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
cd66ada0 1110 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1e3001a8
JH
1111 o_cloexec &= ~O_CLOEXEC;
1112 fd = open(name, flags | o_cloexec);
491a8dec 1113 }
491a8dec 1114
9fb9495d 1115#if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
491a8dec 1116 {
1e3001a8 1117 static int fd_cloexec = FD_CLOEXEC;
a0788266 1118
1e3001a8
JH
1119 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1120 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
9fb9495d
EW
1121 int flags = fcntl(fd, F_GETFD);
1122 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1e3001a8 1123 fd_cloexec = 0;
cd66ada0 1124 }
44d1c19e 1125 }
a0788266 1126#endif
1b8ac5ea 1127 return fd;
a0788266
JS
1128}
1129
3cf8b462 1130/*
514c5fdd 1131 * Find "oid" as a loose object in the local repository or in an alternate.
771e7d57
JK
1132 * Returns 0 on success, negative on failure.
1133 *
1134 * The "path" out-parameter will give the path of the object we found (if any).
1135 * Note that it may point to static storage and is only valid until another
514c5fdd 1136 * call to stat_loose_object().
3cf8b462 1137 */
514c5fdd
JK
1138static int stat_loose_object(struct repository *r, const struct object_id *oid,
1139 struct stat *st, const char **path)
1f688557 1140{
263db403 1141 struct object_directory *odb;
ea657730
CC
1142 static struct strbuf buf = STRBUF_INIT;
1143
d2607fa0 1144 prepare_alt_odb(r);
f0eaf638 1145 for (odb = r->objects->odb; odb; odb = odb->next) {
514c5fdd 1146 *path = odb_loose_path(odb, &buf, oid);
771e7d57 1147 if (!lstat(*path, st))
052fe5ea 1148 return 0;
c7934306
SP
1149 }
1150
3cf8b462
SP
1151 return -1;
1152}
1153
771e7d57 1154/*
514c5fdd 1155 * Like stat_loose_object(), but actually open the object and return the
771e7d57
JK
1156 * descriptor. See the caveats on the "path" parameter above.
1157 */
514c5fdd
JK
1158static int open_loose_object(struct repository *r,
1159 const struct object_id *oid, const char **path)
60bb8b14 1160{
44d1c19e 1161 int fd;
263db403 1162 struct object_directory *odb;
f0eaf638 1163 int most_interesting_errno = ENOENT;
ea657730
CC
1164 static struct strbuf buf = STRBUF_INIT;
1165
ec7283e5 1166 prepare_alt_odb(r);
f0eaf638 1167 for (odb = r->objects->odb; odb; odb = odb->next) {
514c5fdd 1168 *path = odb_loose_path(odb, &buf, oid);
771e7d57 1169 fd = git_open(*path);
44d1c19e
LT
1170 if (fd >= 0)
1171 return fd;
f0eaf638 1172
d6c8a05b
JK
1173 if (most_interesting_errno == ENOENT)
1174 most_interesting_errno = errno;
03e79c88 1175 }
d6c8a05b 1176 errno = most_interesting_errno;
44d1c19e 1177 return -1;
1f688557
JH
1178}
1179
61c7711c 1180static int quick_has_loose(struct repository *r,
d7a24573 1181 const struct object_id *oid)
61c7711c 1182{
61c7711c
JK
1183 struct object_directory *odb;
1184
61c7711c
JK
1185 prepare_alt_odb(r);
1186 for (odb = r->objects->odb; odb; odb = odb->next) {
92d8ed8a 1187 if (oidtree_contains(odb_loose_cache(odb, oid), oid))
61c7711c
JK
1188 return 1;
1189 }
1190 return 0;
1191}
1192
f6371f92
JK
1193/*
1194 * Map the loose object at "path" if it is not NULL, or the path found by
514c5fdd 1195 * searching for a loose object named "oid".
f6371f92 1196 */
514c5fdd
JK
1197static void *map_loose_object_1(struct repository *r, const char *path,
1198 const struct object_id *oid, unsigned long *size)
27d69a46 1199{
0fcfd160 1200 void *map;
144bde78 1201 int fd;
84dd81c1 1202
f6371f92
JK
1203 if (path)
1204 fd = git_open(path);
8261e1f1 1205 else
514c5fdd 1206 fd = open_loose_object(r, oid, &path);
44d1c19e
LT
1207 map = NULL;
1208 if (fd >= 0) {
1209 struct stat st;
1f688557 1210
44d1c19e
LT
1211 if (!fstat(fd, &st)) {
1212 *size = xsize_t(st.st_size);
33e42de0
MM
1213 if (!*size) {
1214 /* mmap() is forbidden on empty files */
259328b7 1215 error(_("object file %s is empty"), path);
6881925e 1216 close(fd);
33e42de0
MM
1217 return NULL;
1218 }
44d1c19e 1219 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1220 }
44d1c19e 1221 close(fd);
74e34e1f 1222 }
0fcfd160 1223 return map;
74e34e1f
NP
1224}
1225
514c5fdd
JK
1226void *map_loose_object(struct repository *r,
1227 const struct object_id *oid,
1228 unsigned long *size)
068f85e3 1229{
514c5fdd 1230 return map_loose_object_1(r, NULL, oid, size);
068f85e3 1231}
1232
00a7760e
JK
1233static int unpack_loose_short_header(git_zstream *stream,
1234 unsigned char *map, unsigned long mapsize,
1235 void *buffer, unsigned long bufsiz)
47fe3f6e 1236{
31877c9a
MT
1237 int ret;
1238
c4483576
LT
1239 /* Get the data stream */
1240 memset(stream, 0, sizeof(*stream));
1241 stream->next_in = map;
1242 stream->avail_in = mapsize;
1243 stream->next_out = buffer;
93821bd9 1244 stream->avail_out = bufsiz;
47fe3f6e 1245
39c68542 1246 git_inflate_init(stream);
31877c9a
MT
1247 obj_read_unlock();
1248 ret = git_inflate(stream, 0);
1249 obj_read_lock();
1250
1251 return ret;
9a217f2a
JH
1252}
1253
00a7760e
JK
1254int unpack_loose_header(git_zstream *stream,
1255 unsigned char *map, unsigned long mapsize,
1256 void *buffer, unsigned long bufsiz)
1f688557 1257{
00a7760e
JK
1258 int status = unpack_loose_short_header(stream, map, mapsize,
1259 buffer, bufsiz);
42873078 1260
d21f8426
JH
1261 if (status < Z_OK)
1262 return status;
628522ec 1263
d21f8426
JH
1264 /* Make sure we have the terminating NUL */
1265 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1266 return -1;
1f688557
JH
1267 return 0;
1268}
1269
00a7760e
JK
1270static int unpack_loose_header_to_strbuf(git_zstream *stream, unsigned char *map,
1271 unsigned long mapsize, void *buffer,
1272 unsigned long bufsiz, struct strbuf *header)
d131b7af 1273{
46f03448
KN
1274 int status;
1275
00a7760e 1276 status = unpack_loose_short_header(stream, map, mapsize, buffer, bufsiz);
d21f8426
JH
1277 if (status < Z_OK)
1278 return -1;
d131b7af 1279
46f03448
KN
1280 /*
1281 * Check if entire header is unpacked in the first iteration.
d131b7af 1282 */
46f03448
KN
1283 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1284 return 0;
d131b7af 1285
46f03448
KN
1286 /*
1287 * buffer[0..bufsiz] was not large enough. Copy the partial
1288 * result out to header, and then append the result of further
1289 * reading the stream.
1290 */
1291 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1292 stream->next_out = buffer;
1293 stream->avail_out = bufsiz;
d131b7af 1294
46f03448 1295 do {
31877c9a 1296 obj_read_unlock();
46f03448 1297 status = git_inflate(stream, 0);
31877c9a 1298 obj_read_lock();
46f03448
KN
1299 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1300 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1301 return 0;
1302 stream->next_out = buffer;
1303 stream->avail_out = bufsiz;
1304 } while (status != Z_STREAM_END);
1305 return -1;
d131b7af
SP
1306}
1307
00a7760e
JK
1308static void *unpack_loose_rest(git_zstream *stream,
1309 void *buffer, unsigned long size,
1310 const struct object_id *oid)
95099731 1311{
5180cacc 1312 int bytes = strlen(buffer) + 1;
3aee68aa 1313 unsigned char *buf = xmallocz(size);
93821bd9 1314 unsigned long n;
7efbff75 1315 int status = Z_OK;
95099731 1316
93821bd9
LT
1317 n = stream->total_out - bytes;
1318 if (n > size)
1319 n = size;
1320 memcpy(buf, (char *) buffer + bytes, n);
1321 bytes = n;
456cdf6e
LT
1322 if (bytes <= size) {
1323 /*
1324 * The above condition must be (bytes <= size), not
1325 * (bytes < size). In other words, even though we
ccf5ace0 1326 * expect no more output and set avail_out to zero,
456cdf6e
LT
1327 * the input zlib stream may have bytes that express
1328 * "this concludes the stream", and we *do* want to
1329 * eat that input.
1330 *
1331 * Otherwise we would not be able to test that we
1332 * consumed all the input to reach the expected size;
1333 * we also want to check that zlib tells us that all
1334 * went well with status == Z_STREAM_END at the end.
1335 */
5180cacc
LT
1336 stream->next_out = buf + bytes;
1337 stream->avail_out = size - bytes;
31877c9a
MT
1338 while (status == Z_OK) {
1339 obj_read_unlock();
39c68542 1340 status = git_inflate(stream, Z_FINISH);
31877c9a
MT
1341 obj_read_lock();
1342 }
5180cacc 1343 }
456cdf6e 1344 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1345 git_inflate_end(stream);
7efbff75 1346 return buf;
95099731
NTND
1347 }
1348
7efbff75 1349 if (status < 0)
00a7760e 1350 error(_("corrupt loose object '%s'"), oid_to_hex(oid));
7efbff75 1351 else if (stream->avail_in)
259328b7 1352 error(_("garbage at end of loose object '%s'"),
00a7760e 1353 oid_to_hex(oid));
7efbff75
JH
1354 free(buf);
1355 return NULL;
95099731
NTND
1356}
1357
d40d535b 1358/*
5180cacc
LT
1359 * We used to just use "sscanf()", but that's actually way
1360 * too permissive for what we want to check. So do an anal
1361 * object header parse by hand.
d40d535b 1362 */
00a7760e
JK
1363static int parse_loose_header_extended(const char *hdr, struct object_info *oi,
1364 unsigned int flags)
1f688557 1365{
46f03448 1366 const char *type_buf = hdr;
5180cacc 1367 unsigned long size;
46f03448 1368 int type, type_len = 0;
43057304 1369
5180cacc 1370 /*
46f03448 1371 * The type can be of any size but is followed by
21666f1a 1372 * a space.
5180cacc 1373 */
5180cacc
LT
1374 for (;;) {
1375 char c = *hdr++;
d21f8426
JH
1376 if (!c)
1377 return -1;
5180cacc
LT
1378 if (c == ' ')
1379 break;
46f03448 1380 type_len++;
5180cacc 1381 }
1f688557 1382
46f03448 1383 type = type_from_string_gently(type_buf, type_len, 1);
6ca32f47
BW
1384 if (oi->type_name)
1385 strbuf_add(oi->type_name, type_buf, type_len);
46f03448
KN
1386 /*
1387 * Set type to 0 if its an unknown object and
2e3a16b2 1388 * we're obtaining the type using '--allow-unknown-type'
46f03448
KN
1389 * option.
1390 */
19fc5e84 1391 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
46f03448
KN
1392 type = 0;
1393 else if (type < 0)
259328b7 1394 die(_("invalid object type"));
46f03448
KN
1395 if (oi->typep)
1396 *oi->typep = type;
5180cacc
LT
1397
1398 /*
1399 * The length must follow immediately, and be in canonical
1400 * decimal format (ie "010" is not valid).
1401 */
1402 size = *hdr++ - '0';
1403 if (size > 9)
1404 return -1;
1405 if (size) {
1406 for (;;) {
1407 unsigned long c = *hdr - '0';
1408 if (c > 9)
1409 break;
1410 hdr++;
1411 size = size * 10 + c;
1b1005d1 1412 }
c01f51cc 1413 }
46f03448
KN
1414
1415 if (oi->sizep)
1416 *oi->sizep = size;
5180cacc
LT
1417
1418 /*
1419 * The length must be followed by a zero byte
1420 */
46f03448 1421 return *hdr ? -1 : type;
1f688557
JH
1422}
1423
00a7760e 1424int parse_loose_header(const char *hdr, unsigned long *sizep)
bf592c50 1425{
27b5c1a0 1426 struct object_info oi = OBJECT_INFO_INIT;
21666f1a 1427
46f03448 1428 oi.sizep = sizep;
00a7760e 1429 return parse_loose_header_extended(hdr, &oi, 0);
bf592c50
DB
1430}
1431
514c5fdd
JK
1432static int loose_object_info(struct repository *r,
1433 const struct object_id *oid,
1434 struct object_info *oi, int flags)
65c2e0c3 1435{
46f03448
KN
1436 int status = 0;
1437 unsigned long mapsize;
65c2e0c3 1438 void *map;
ef49a7a0 1439 git_zstream stream;
1af64f73 1440 char hdr[MAX_HEADER_LEN];
46f03448 1441 struct strbuf hdrbuf = STRBUF_INIT;
c84a1f3e 1442 unsigned long size_scratch;
65c2e0c3 1443
b99b6bcc
JK
1444 if (oi->delta_base_oid)
1445 oidclr(oi->delta_base_oid);
5d642e75 1446
052fe5ea
JK
1447 /*
1448 * If we don't care about type or size, then we don't
4ef8d1dd
JH
1449 * need to look inside the object at all. Note that we
1450 * do not optimize out the stat call, even if the
1451 * caller doesn't care about the disk-size, since our
1452 * return value implicitly indicates whether the
1453 * object even exists.
052fe5ea 1454 */
6ca32f47 1455 if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
771e7d57 1456 const char *path;
4ef8d1dd 1457 struct stat st;
61c7711c 1458 if (!oi->disk_sizep && (flags & OBJECT_INFO_QUICK))
d7a24573 1459 return quick_has_loose(r, oid) ? 0 : -1;
514c5fdd 1460 if (stat_loose_object(r, oid, &st, &path) < 0)
4ef8d1dd
JH
1461 return -1;
1462 if (oi->disk_sizep)
23c339c0 1463 *oi->disk_sizep = st.st_size;
052fe5ea
JK
1464 return 0;
1465 }
1466
514c5fdd 1467 map = map_loose_object(r, oid, &mapsize);
f0df4ed5 1468 if (!map)
dbea72a8 1469 return -1;
c84a1f3e
JT
1470
1471 if (!oi->sizep)
1472 oi->sizep = &size_scratch;
1473
23c339c0
JK
1474 if (oi->disk_sizep)
1475 *oi->disk_sizep = mapsize;
19fc5e84 1476 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
00a7760e 1477 if (unpack_loose_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
259328b7 1478 status = error(_("unable to unpack %s header with --allow-unknown-type"),
514c5fdd 1479 oid_to_hex(oid));
00a7760e 1480 } else if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
259328b7 1481 status = error(_("unable to unpack %s header"),
514c5fdd 1482 oid_to_hex(oid));
46f03448
KN
1483 if (status < 0)
1484 ; /* Do nothing */
1485 else if (hdrbuf.len) {
00a7760e 1486 if ((status = parse_loose_header_extended(hdrbuf.buf, oi, flags)) < 0)
259328b7 1487 status = error(_("unable to parse %s header with --allow-unknown-type"),
514c5fdd 1488 oid_to_hex(oid));
00a7760e 1489 } else if ((status = parse_loose_header_extended(hdr, oi, flags)) < 0)
514c5fdd 1490 status = error(_("unable to parse %s header"), oid_to_hex(oid));
c84a1f3e 1491
b3ea7dd3 1492 if (status >= 0 && oi->contentp) {
00a7760e
JK
1493 *oi->contentp = unpack_loose_rest(&stream, hdr,
1494 *oi->sizep, oid);
b3ea7dd3
JK
1495 if (!*oi->contentp) {
1496 git_inflate_end(&stream);
1497 status = -1;
1498 }
1499 } else
c84a1f3e
JT
1500 git_inflate_end(&stream);
1501
65c2e0c3 1502 munmap(map, mapsize);
46f03448 1503 if (status && oi->typep)
23c339c0 1504 *oi->typep = status;
c84a1f3e
JT
1505 if (oi->sizep == &size_scratch)
1506 oi->sizep = NULL;
46f03448 1507 strbuf_release(&hdrbuf);
3ab0fb06 1508 oi->whence = OI_LOOSE;
93cff9a9 1509 return (status < 0) ? status : 0;
65c2e0c3
JH
1510}
1511
31877c9a
MT
1512int obj_read_use_lock = 0;
1513pthread_mutex_t obj_read_mutex;
1514
1515void enable_obj_read_lock(void)
1516{
1517 if (obj_read_use_lock)
1518 return;
1519
1520 obj_read_use_lock = 1;
1521 init_recursive_mutex(&obj_read_mutex);
1522}
1523
1524void disable_obj_read_lock(void)
1525{
1526 if (!obj_read_use_lock)
1527 return;
1528
1529 obj_read_use_lock = 0;
1530 pthread_mutex_destroy(&obj_read_mutex);
1531}
1532
8b4c0103
JT
1533int fetch_if_missing = 1;
1534
31877c9a
MT
1535static int do_oid_object_info_extended(struct repository *r,
1536 const struct object_id *oid,
1537 struct object_info *oi, unsigned flags)
f0df4ed5 1538{
cd585e2a 1539 static struct object_info blank_oi = OBJECT_INFO_INIT;
9c8a294a 1540 struct cached_object *co;
f0df4ed5 1541 struct pack_entry e;
5b086407 1542 int rtype;
b383a13c 1543 const struct object_id *real = oid;
8b4c0103 1544 int already_retried = 0;
f0df4ed5 1545
31877c9a 1546
b383a13c 1547 if (flags & OBJECT_INFO_LOOKUP_REPLACE)
9d98354f 1548 real = lookup_replace_object(r, oid);
f0df4ed5 1549
b383a13c 1550 if (is_null_oid(real))
87b5e236
JK
1551 return -1;
1552
cd585e2a
JT
1553 if (!oi)
1554 oi = &blank_oi;
1555
9c8a294a
JT
1556 co = find_cached_object(real);
1557 if (co) {
1558 if (oi->typep)
1559 *(oi->typep) = co->type;
1560 if (oi->sizep)
1561 *(oi->sizep) = co->size;
1562 if (oi->disk_sizep)
1563 *(oi->disk_sizep) = 0;
b99b6bcc
JK
1564 if (oi->delta_base_oid)
1565 oidclr(oi->delta_base_oid);
9c8a294a
JT
1566 if (oi->type_name)
1567 strbuf_addstr(oi->type_name, type_name(co->type));
1568 if (oi->contentp)
1569 *oi->contentp = xmemdupz(co->buf, co->size);
1570 oi->whence = OI_CACHED;
1571 return 0;
c4d9986f
NTND
1572 }
1573
8b4c0103 1574 while (1) {
42c8ce1c 1575 if (find_pack_entry(r, real, &e))
8b4c0103
JT
1576 break;
1577
024aa469
TI
1578 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1579 return -1;
1580
ddd63e64 1581 /* Most likely it's a loose object. */
514c5fdd 1582 if (!loose_object_info(r, real, oi, flags))
5b086407 1583 return 0;
ddd63e64
SG
1584
1585 /* Not a loose object; someone else may have just packed it. */
2b7750c9 1586 if (!(flags & OBJECT_INFO_QUICK)) {
9d98354f 1587 reprepare_packed_git(r);
42c8ce1c 1588 if (find_pack_entry(r, real, &e))
2b7750c9
JT
1589 break;
1590 }
8b4c0103
JT
1591
1592 /* Check if it is a missing object */
ef830cc4
JT
1593 if (fetch_if_missing && repo_has_promisor_remote(r) &&
1594 !already_retried &&
31f5256c 1595 !(flags & OBJECT_INFO_SKIP_FETCH_OBJECT)) {
8b4c0103 1596 /*
b14ed5ad
CC
1597 * TODO Investigate checking promisor_remote_get_direct()
1598 * TODO return value and stopping on error here.
8b4c0103 1599 */
b14ed5ad 1600 promisor_remote_get_direct(r, real, 1);
8b4c0103
JT
1601 already_retried = 1;
1602 continue;
dfdd4afc 1603 }
8b4c0103
JT
1604
1605 return -1;
f0df4ed5 1606 }
3d77d877 1607
cd585e2a
JT
1608 if (oi == &blank_oi)
1609 /*
1610 * We know that the caller doesn't actually need the
1611 * information below, so return early.
1612 */
1613 return 0;
9d98354f 1614 rtype = packed_object_info(r, e.p, e.offset, oi);
412916ee 1615 if (rtype < 0) {
b383a13c 1616 mark_bad_packed_object(e.p, real->hash);
31877c9a 1617 return do_oid_object_info_extended(r, real, oi, 0);
3ab0fb06 1618 } else if (oi->whence == OI_PACKED) {
9a490590
JH
1619 oi->u.packed.offset = e.offset;
1620 oi->u.packed.pack = e.p;
1621 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1622 rtype == OBJ_OFS_DELTA);
3d77d877
NP
1623 }
1624
5b086407 1625 return 0;
f0df4ed5
JS
1626}
1627
31877c9a
MT
1628int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1629 struct object_info *oi, unsigned flags)
1630{
1631 int ret;
1632 obj_read_lock();
1633 ret = do_oid_object_info_extended(r, oid, oi, flags);
1634 obj_read_unlock();
1635 return ret;
1636}
1637
1638
3fc0dca9 1639/* returns enum object_type or negative */
9d98354f
SB
1640int oid_object_info(struct repository *r,
1641 const struct object_id *oid,
1642 unsigned long *sizep)
9a490590 1643{
5b086407 1644 enum object_type type;
27b5c1a0 1645 struct object_info oi = OBJECT_INFO_INIT;
9a490590 1646
5b086407 1647 oi.typep = &type;
9a490590 1648 oi.sizep = sizep;
9d98354f
SB
1649 if (oid_object_info_extended(r, oid, &oi,
1650 OBJECT_INFO_LOOKUP_REPLACE) < 0)
5b086407
JK
1651 return -1;
1652 return type;
9a490590
JH
1653}
1654
1b9b5c69 1655static void *read_object(struct repository *r,
cba595ab 1656 const struct object_id *oid, enum object_type *type,
f1d8130b
JT
1657 unsigned long *size)
1658{
1659 struct object_info oi = OBJECT_INFO_INIT;
1660 void *content;
1661 oi.typep = type;
1662 oi.sizep = size;
1663 oi.contentp = &content;
1664
cba595ab 1665 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
f1d8130b
JT
1666 return NULL;
1667 return content;
1668}
1669
829e5c3b
PO
1670int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1671 struct object_id *oid)
d66b37bb
JH
1672{
1673 struct cached_object *co;
1674
2dcde20e 1675 hash_object_file(the_hash_algo, buf, len, type_name(type), oid);
a64d2aae
JT
1676 if (has_object_file_with_flags(oid, OBJECT_INFO_QUICK | OBJECT_INFO_SKIP_FETCH_OBJECT) ||
1677 find_cached_object(oid))
d66b37bb 1678 return 0;
c7353967 1679 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
d66b37bb
JH
1680 co = &cached_objects[cached_object_nr++];
1681 co->size = len;
21666f1a 1682 co->type = type;
efa13f7b
JH
1683 co->buf = xmalloc(len);
1684 memcpy(co->buf, buf, len);
62ba93ea 1685 oidcpy(&co->oid, oid);
d66b37bb
JH
1686 return 0;
1687}
1688
b6c4cecc
JH
1689/*
1690 * This function dies on corrupt objects; the callers who want to
1691 * deal with them should arrange to call read_object() and give error
1692 * messages themselves.
1693 */
a3b72c89
SB
1694void *read_object_file_extended(struct repository *r,
1695 const struct object_id *oid,
b4f5aca4 1696 enum object_type *type,
1697 unsigned long *size,
1698 int lookup_replace)
ac939109 1699{
3ba7a065 1700 void *data;
b6c4cecc 1701 const struct packed_git *p;
771e7d57
JK
1702 const char *path;
1703 struct stat st;
1f2e7cea 1704 const struct object_id *repl = lookup_replace ?
a3b72c89 1705 lookup_replace_object(r, oid) : oid;
b6c4cecc 1706
3ba7a065 1707 errno = 0;
cba595ab 1708 data = read_object(r, repl, type, size);
4bbf5a26 1709 if (data)
b6c4cecc 1710 return data;
68095570 1711
31877c9a 1712 obj_read_lock();
25f3af3f 1713 if (errno && errno != ENOENT)
259328b7 1714 die_errno(_("failed to read object %s"), oid_to_hex(oid));
3ba7a065 1715
68095570 1716 /* die if we replaced an object with one that does not exist */
b383a13c 1717 if (repl != oid)
259328b7 1718 die(_("replacement %s not found for %s"),
b383a13c 1719 oid_to_hex(repl), oid_to_hex(oid));
68095570 1720
cba595ab 1721 if (!stat_loose_object(r, repl, &st, &path))
259328b7 1722 die(_("loose object %s (stored in %s) is corrupt"),
b383a13c 1723 oid_to_hex(repl), path);
68095570 1724
a3b72c89 1725 if ((p = has_packed_and_bad(r, repl->hash)) != NULL)
259328b7 1726 die(_("packed object %s (stored in %s) is corrupt"),
b383a13c 1727 oid_to_hex(repl), p->pack_name);
31877c9a 1728 obj_read_unlock();
f5552aee 1729
b6c4cecc 1730 return NULL;
ac939109
NP
1731}
1732
d3b4705a
NTND
1733void *read_object_with_reference(struct repository *r,
1734 const struct object_id *oid,
21666f1a 1735 const char *required_type_name,
40469ee9 1736 unsigned long *size,
02f0547e 1737 struct object_id *actual_oid_return)
f4913f91 1738{
21666f1a 1739 enum object_type type, required_type;
f4913f91
JH
1740 void *buffer;
1741 unsigned long isize;
02f0547e 1742 struct object_id actual_oid;
f4913f91 1743
21666f1a 1744 required_type = type_from_string(required_type_name);
02f0547e 1745 oidcpy(&actual_oid, oid);
40469ee9
JH
1746 while (1) {
1747 int ref_length = -1;
1748 const char *ref_type = NULL;
f4913f91 1749
d3b4705a 1750 buffer = repo_read_object_file(r, &actual_oid, &type, &isize);
40469ee9
JH
1751 if (!buffer)
1752 return NULL;
21666f1a 1753 if (type == required_type) {
40469ee9 1754 *size = isize;
02f0547e 1755 if (actual_oid_return)
1756 oidcpy(actual_oid_return, &actual_oid);
40469ee9
JH
1757 return buffer;
1758 }
1759 /* Handle references */
21666f1a 1760 else if (type == OBJ_COMMIT)
40469ee9 1761 ref_type = "tree ";
21666f1a 1762 else if (type == OBJ_TAG)
40469ee9
JH
1763 ref_type = "object ";
1764 else {
1765 free(buffer);
1766 return NULL;
1767 }
1768 ref_length = strlen(ref_type);
f4913f91 1769
94b5e093 1770 if (ref_length + the_hash_algo->hexsz > isize ||
50974ec9 1771 memcmp(buffer, ref_type, ref_length) ||
02f0547e 1772 get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
40469ee9
JH
1773 free(buffer);
1774 return NULL;
1775 }
1cf58e72 1776 free(buffer);
40469ee9 1777 /* Now we have the ID of the referred-to object in
02f0547e 1778 * actual_oid. Check again. */
f4913f91 1779 }
f4913f91
JH
1780}
1781
7ad5c44d
MT
1782static void write_object_file_prepare(const struct git_hash_algo *algo,
1783 const void *buf, unsigned long len,
a09c985e
PO
1784 const char *type, struct object_id *oid,
1785 char *hdr, int *hdrlen)
d410c0f5 1786{
18e2588e 1787 git_hash_ctx c;
d410c0f5
JH
1788
1789 /* Generate the header */
ca473cef 1790 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %"PRIuMAX , type, (uintmax_t)len)+1;
d410c0f5
JH
1791
1792 /* Sha1.. */
7ad5c44d
MT
1793 algo->init_fn(&c);
1794 algo->update_fn(&c, hdr, *hdrlen);
1795 algo->update_fn(&c, buf, len);
5951bf46 1796 algo->final_oid_fn(oid, &c);
d410c0f5
JH
1797}
1798
230f1322 1799/*
5a688fe4 1800 * Move the just written object into its final resting place.
230f1322 1801 */
cb5add58 1802int finalize_object_file(const char *tmpfile, const char *filename)
230f1322 1803{
e32c0a9c 1804 int ret = 0;
5a688fe4 1805
348df166 1806 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
1807 goto try_rename;
1808 else if (link(tmpfile, filename))
e32c0a9c 1809 ret = errno;
7ebb6fca
LT
1810
1811 /*
1812 * Coda hack - coda doesn't like cross-directory links,
1813 * so we fall back to a rename, which will mean that it
1814 * won't be able to check collisions, but that's not a
1815 * big deal.
1816 *
1817 * The same holds for FAT formatted media.
1818 *
3be1f18e 1819 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
1820 * left to unlink.
1821 */
1822 if (ret && ret != EEXIST) {
be66a6c4 1823 try_rename:
7ebb6fca 1824 if (!rename(tmpfile, filename))
3be1f18e 1825 goto out;
9e48b389 1826 ret = errno;
230f1322 1827 }
691f1a28 1828 unlink_or_warn(tmpfile);
230f1322
LT
1829 if (ret) {
1830 if (ret != EEXIST) {
2c319886 1831 return error_errno(_("unable to write file %s"), filename);
230f1322
LT
1832 }
1833 /* FIXME!!! Collision check here ? */
1834 }
1835
3be1f18e 1836out:
5256b006 1837 if (adjust_shared_perm(filename))
259328b7 1838 return error(_("unable to set permission to '%s'"), filename);
230f1322
LT
1839 return 0;
1840}
1841
4d548150
LT
1842static int write_buffer(int fd, const void *buf, size_t len)
1843{
d34cf19b 1844 if (write_in_full(fd, buf, len) < 0)
259328b7 1845 return error_errno(_("file write error"));
4d548150
LT
1846 return 0;
1847}
1848
2dcde20e
MT
1849int hash_object_file(const struct git_hash_algo *algo, const void *buf,
1850 unsigned long len, const char *type,
f070facc 1851 struct object_id *oid)
abdc3fc8 1852{
1af64f73 1853 char hdr[MAX_HEADER_LEN];
ef1286d3 1854 int hdrlen = sizeof(hdr);
2dcde20e 1855 write_object_file_prepare(algo, buf, len, type, oid, hdr, &hdrlen);
abdc3fc8
RS
1856 return 0;
1857}
1858
e9039dd3 1859/* Finalize a file on disk, and close it. */
514c5fdd 1860static void close_loose_object(int fd)
e9039dd3 1861{
aafe9fba 1862 if (fsync_object_files)
76011357 1863 fsync_or_die(fd, "loose object file");
e9039dd3 1864 if (close(fd) != 0)
76011357 1865 die_errno(_("error when closing loose object file"));
e9039dd3
LT
1866}
1867
5723fe7e
LT
1868/* Size of directory component, including the ending '/' */
1869static inline int directory_size(const char *filename)
1870{
1871 const char *s = strrchr(filename, '/');
1872 if (!s)
1873 return 0;
1874 return s - filename + 1;
1875}
1876
1877/*
1878 * This creates a temporary file in the same directory as the final
1879 * 'filename'
1880 *
1881 * We want to avoid cross-directory filename renames, because those
1882 * can have problems on various filesystems (FAT, NFS, Coda).
1883 */
d4b3d11a 1884static int create_tmpfile(struct strbuf *tmp, const char *filename)
5723fe7e
LT
1885{
1886 int fd, dirlen = directory_size(filename);
1887
d4b3d11a
JK
1888 strbuf_reset(tmp);
1889 strbuf_add(tmp, filename, dirlen);
1890 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1891 fd = git_mkstemp_mode(tmp->buf, 0444);
cbacbf4e 1892 if (fd < 0 && dirlen && errno == ENOENT) {
d4b3d11a
JK
1893 /*
1894 * Make sure the directory exists; note that the contents
1895 * of the buffer are undefined after mkstemp returns an
1896 * error, so we have to rewrite the whole buffer from
1897 * scratch.
1898 */
1899 strbuf_reset(tmp);
1900 strbuf_add(tmp, filename, dirlen - 1);
1901 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
b2476a60 1902 return -1;
d4b3d11a 1903 if (adjust_shared_perm(tmp->buf))
5723fe7e
LT
1904 return -1;
1905
1906 /* Try again */
d4b3d11a
JK
1907 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1908 fd = git_mkstemp_mode(tmp->buf, 0444);
5723fe7e
LT
1909 }
1910 return fd;
1911}
1912
3fc7281f
PO
1913static int write_loose_object(const struct object_id *oid, char *hdr,
1914 int hdrlen, const void *buf, unsigned long len,
1915 time_t mtime)
0fcfd160 1916{
915308b1 1917 int fd, ret;
9892beba 1918 unsigned char compressed[4096];
ef49a7a0 1919 git_zstream stream;
18e2588e 1920 git_hash_ctx c;
3fc7281f 1921 struct object_id parano_oid;
d4b3d11a 1922 static struct strbuf tmp_file = STRBUF_INIT;
ea657730
CC
1923 static struct strbuf filename = STRBUF_INIT;
1924
514c5fdd 1925 loose_object_path(the_repository, &filename, oid);
a44c9a5e 1926
ea657730 1927 fd = create_tmpfile(&tmp_file, filename.buf);
aac17941 1928 if (fd < 0) {
35243577 1929 if (errno == EACCES)
259328b7 1930 return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
916d081b 1931 else
259328b7 1932 return error_errno(_("unable to create temporary file"));
aac17941
LT
1933 }
1934
0fcfd160 1935 /* Set it up */
55bb5c91 1936 git_deflate_init(&stream, zlib_compression_level);
0fcfd160 1937 stream.next_out = compressed;
9892beba 1938 stream.avail_out = sizeof(compressed);
18e2588e 1939 the_hash_algo->init_fn(&c);
a44c9a5e
LT
1940
1941 /* First header.. */
d65a16f6 1942 stream.next_in = (unsigned char *)hdr;
a44c9a5e 1943 stream.avail_in = hdrlen;
55bb5c91
JH
1944 while (git_deflate(&stream, 0) == Z_OK)
1945 ; /* nothing */
18e2588e 1946 the_hash_algo->update_fn(&c, hdr, hdrlen);
a44c9a5e
LT
1947
1948 /* Then the data itself.. */
c00e657d 1949 stream.next_in = (void *)buf;
a44c9a5e 1950 stream.avail_in = len;
9892beba 1951 do {
748af44c 1952 unsigned char *in0 = stream.next_in;
55bb5c91 1953 ret = git_deflate(&stream, Z_FINISH);
18e2588e 1954 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
9892beba 1955 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
76011357 1956 die(_("unable to write loose object file"));
9892beba
NP
1957 stream.next_out = compressed;
1958 stream.avail_out = sizeof(compressed);
1959 } while (ret == Z_OK);
1960
ac54c277 1961 if (ret != Z_STREAM_END)
259328b7 1962 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
3fc7281f 1963 ret);
55bb5c91 1964 ret = git_deflate_end_gently(&stream);
ac54c277 1965 if (ret != Z_OK)
259328b7 1966 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
3fc7281f 1967 ret);
5951bf46 1968 the_hash_algo->final_oid_fn(&parano_oid, &c);
9001dc2a 1969 if (!oideq(oid, &parano_oid))
259328b7 1970 die(_("confused by unstable object source data for %s"),
3fc7281f 1971 oid_to_hex(oid));
ac54c277 1972
514c5fdd 1973 close_loose_object(fd);
0fcfd160 1974
bbac7311
NP
1975 if (mtime) {
1976 struct utimbuf utb;
1977 utb.actime = mtime;
1978 utb.modtime = mtime;
d4b3d11a 1979 if (utime(tmp_file.buf, &utb) < 0)
259328b7 1980 warning_errno(_("failed utime() on %s"), tmp_file.buf);
bbac7311
NP
1981 }
1982
ea657730 1983 return finalize_object_file(tmp_file.buf, filename.buf);
0fcfd160 1984}
8237b185 1985
6862ebbf 1986static int freshen_loose_object(const struct object_id *oid)
33d4221c 1987{
6862ebbf 1988 return check_and_freshen(oid, 1);
33d4221c
JK
1989}
1990
6862ebbf 1991static int freshen_packed_object(const struct object_id *oid)
33d4221c
JK
1992{
1993 struct pack_entry e;
544443cb 1994 if (!find_pack_entry(the_repository, oid, &e))
ee1c6c34
JK
1995 return 0;
1996 if (e.p->freshened)
1997 return 1;
1998 if (!freshen_file(e.p->pack_name))
1999 return 0;
2000 e.p->freshened = 1;
2001 return 1;
33d4221c
JK
2002}
2003
a09c985e
PO
2004int write_object_file(const void *buf, unsigned long len, const char *type,
2005 struct object_id *oid)
bbac7311 2006{
1af64f73 2007 char hdr[MAX_HEADER_LEN];
ef1286d3 2008 int hdrlen = sizeof(hdr);
bbac7311
NP
2009
2010 /* Normally if we have it in the pack then we do not bother writing
2011 * it out into .git/objects/??/?{38} file.
2012 */
7ad5c44d
MT
2013 write_object_file_prepare(the_hash_algo, buf, len, type, oid, hdr,
2014 &hdrlen);
6862ebbf 2015 if (freshen_packed_object(oid) || freshen_loose_object(oid))
bbac7311 2016 return 0;
3fc7281f 2017 return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
bbac7311
NP
2018}
2019
1752cbbc
PO
2020int hash_object_file_literally(const void *buf, unsigned long len,
2021 const char *type, struct object_id *oid,
2022 unsigned flags)
0c3db67c
ES
2023{
2024 char *header;
2025 int hdrlen, status = 0;
2026
2027 /* type string, SP, %lu of the length plus NUL must fit this */
1af64f73 2028 hdrlen = strlen(type) + MAX_HEADER_LEN;
ef1286d3 2029 header = xmalloc(hdrlen);
7ad5c44d
MT
2030 write_object_file_prepare(the_hash_algo, buf, len, type, oid, header,
2031 &hdrlen);
0c3db67c
ES
2032
2033 if (!(flags & HASH_WRITE_OBJECT))
2034 goto cleanup;
6862ebbf 2035 if (freshen_packed_object(oid) || freshen_loose_object(oid))
0c3db67c 2036 goto cleanup;
3fc7281f 2037 status = write_loose_object(oid, header, hdrlen, buf, len, 0);
0c3db67c
ES
2038
2039cleanup:
2040 free(header);
2041 return status;
2042}
2043
4bdb70a4 2044int force_object_loose(const struct object_id *oid, time_t mtime)
bbac7311 2045{
bbac7311
NP
2046 void *buf;
2047 unsigned long len;
2048 enum object_type type;
1af64f73 2049 char hdr[MAX_HEADER_LEN];
bbac7311 2050 int hdrlen;
1fb23e65 2051 int ret;
bbac7311 2052
6862ebbf 2053 if (has_loose_object(oid))
bbac7311 2054 return 0;
cba595ab 2055 buf = read_object(the_repository, oid, &type, &len);
bbac7311 2056 if (!buf)
2c319886 2057 return error(_("cannot read object for %s"), oid_to_hex(oid));
ca473cef 2058 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(type), (uintmax_t)len) + 1;
3fc7281f 2059 ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
1fb23e65
BS
2060 free(buf);
2061
2062 return ret;
bbac7311
NP
2063}
2064
1d8d9cb6
JT
2065int has_object(struct repository *r, const struct object_id *oid,
2066 unsigned flags)
2067{
2068 int quick = !(flags & HAS_OBJECT_RECHECK_PACKED);
2069 unsigned object_info_flags = OBJECT_INFO_SKIP_FETCH_OBJECT |
2070 (quick ? OBJECT_INFO_QUICK : 0);
2071
2072 if (!startup_info->have_repository)
2073 return 0;
2074 return oid_object_info_extended(r, oid, NULL, object_info_flags) >= 0;
2075}
2076
cba595ab
JH
2077int repo_has_object_file_with_flags(struct repository *r,
2078 const struct object_id *oid, int flags)
8237b185 2079{
3e8b7d3c
JN
2080 if (!startup_info->have_repository)
2081 return 0;
9c8a294a 2082 return oid_object_info_extended(r, oid, NULL, flags) >= 0;
b419aa25 2083}
2084
9b45f499
SB
2085int repo_has_object_file(struct repository *r,
2086 const struct object_id *oid)
b419aa25 2087{
cba595ab 2088 return repo_has_object_file_with_flags(r, oid, 0);
5827a035
JK
2089}
2090
c879daa2
NTND
2091static void check_tree(const void *buf, size_t size)
2092{
2093 struct tree_desc desc;
2094 struct name_entry entry;
2095
2096 init_tree_desc(&desc, buf, size);
2097 while (tree_entry(&desc, &entry))
2098 /* do nothing
2099 * tree_entry() will die() on malformed entries */
2100 ;
2101}
2102
2103static void check_commit(const void *buf, size_t size)
2104{
2105 struct commit c;
2106 memset(&c, 0, sizeof(c));
08f4f445 2107 if (parse_commit_buffer(the_repository, &c, buf, size, 0))
259328b7 2108 die(_("corrupt commit"));
c879daa2
NTND
2109}
2110
2111static void check_tag(const void *buf, size_t size)
2112{
2113 struct tag t;
2114 memset(&t, 0, sizeof(t));
0e740fed 2115 if (parse_tag_buffer(the_repository, &t, buf, size))
259328b7 2116 die(_("corrupt tag"));
c879daa2
NTND
2117}
2118
58bf2a4c
NTND
2119static int index_mem(struct index_state *istate,
2120 struct object_id *oid, void *buf, size_t size,
c4ce46fc
JH
2121 enum object_type type,
2122 const char *path, unsigned flags)
e7332f96 2123{
6c510bee 2124 int ret, re_allocated = 0;
c4ce46fc 2125 int write_object = flags & HASH_WRITE_OBJECT;
74400e71 2126
7672db20 2127 if (!type)
edaec3fb 2128 type = OBJ_BLOB;
6c510bee
LT
2129
2130 /*
2131 * Convert blobs to git internal format
2132 */
43df4f86 2133 if ((type == OBJ_BLOB) && path) {
f285a2d7 2134 struct strbuf nbuf = STRBUF_INIT;
58bf2a4c 2135 if (convert_to_git(istate, path, buf, size, &nbuf,
8462ff43 2136 get_conv_flags(flags))) {
b315c5c0 2137 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
2138 re_allocated = 1;
2139 }
2140 }
c4ce46fc 2141 if (flags & HASH_FORMAT_CHECK) {
c879daa2
NTND
2142 if (type == OBJ_TREE)
2143 check_tree(buf, size);
2144 if (type == OBJ_COMMIT)
2145 check_commit(buf, size);
2146 if (type == OBJ_TAG)
2147 check_tag(buf, size);
2148 }
6c510bee 2149
7672db20 2150 if (write_object)
169c9c01 2151 ret = write_object_file(buf, size, type_name(type), oid);
abdc3fc8 2152 else
2dcde20e
MT
2153 ret = hash_object_file(the_hash_algo, buf, size,
2154 type_name(type), oid);
43df4f86 2155 if (re_allocated)
6c510bee 2156 free(buf);
43df4f86
DP
2157 return ret;
2158}
2159
58bf2a4c
NTND
2160static int index_stream_convert_blob(struct index_state *istate,
2161 struct object_id *oid,
2162 int fd,
2163 const char *path,
2164 unsigned flags)
9035d75a
SP
2165{
2166 int ret;
2167 const int write_object = flags & HASH_WRITE_OBJECT;
2168 struct strbuf sbuf = STRBUF_INIT;
2169
2170 assert(path);
58bf2a4c 2171 assert(would_convert_to_git_filter_fd(istate, path));
9035d75a 2172
58bf2a4c 2173 convert_to_git_filter_fd(istate, path, fd, &sbuf,
8462ff43 2174 get_conv_flags(flags));
9035d75a
SP
2175
2176 if (write_object)
169c9c01 2177 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
a09c985e 2178 oid);
9035d75a 2179 else
2dcde20e
MT
2180 ret = hash_object_file(the_hash_algo, sbuf.buf, sbuf.len,
2181 type_name(OBJ_BLOB), oid);
9035d75a
SP
2182 strbuf_release(&sbuf);
2183 return ret;
2184}
2185
58bf2a4c
NTND
2186static int index_pipe(struct index_state *istate, struct object_id *oid,
2187 int fd, enum object_type type,
7b41e1e1
JH
2188 const char *path, unsigned flags)
2189{
2190 struct strbuf sbuf = STRBUF_INIT;
2191 int ret;
2192
2193 if (strbuf_read(&sbuf, fd, 4096) >= 0)
58bf2a4c 2194 ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags);
7b41e1e1
JH
2195 else
2196 ret = -1;
2197 strbuf_release(&sbuf);
2198 return ret;
2199}
2200
ea68b0ce
DP
2201#define SMALL_FILE_SIZE (32*1024)
2202
58bf2a4c
NTND
2203static int index_core(struct index_state *istate,
2204 struct object_id *oid, int fd, size_t size,
7b41e1e1
JH
2205 enum object_type type, const char *path,
2206 unsigned flags)
43df4f86
DP
2207{
2208 int ret;
43df4f86 2209
7b41e1e1 2210 if (!size) {
58bf2a4c 2211 ret = index_mem(istate, oid, "", size, type, path, flags);
ea68b0ce
DP
2212 } else if (size <= SMALL_FILE_SIZE) {
2213 char *buf = xmalloc(size);
90dca671
JK
2214 ssize_t read_result = read_in_full(fd, buf, size);
2215 if (read_result < 0)
259328b7 2216 ret = error_errno(_("read error while indexing %s"),
90dca671
JK
2217 path ? path : "<unknown>");
2218 else if (read_result != size)
259328b7 2219 ret = error(_("short read while indexing %s"),
90dca671 2220 path ? path : "<unknown>");
ea68b0ce 2221 else
58bf2a4c 2222 ret = index_mem(istate, oid, buf, size, type, path, flags);
ea68b0ce 2223 free(buf);
08bda208 2224 } else {
43df4f86 2225 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
58bf2a4c 2226 ret = index_mem(istate, oid, buf, size, type, path, flags);
aac17941 2227 munmap(buf, size);
08bda208 2228 }
7b41e1e1
JH
2229 return ret;
2230}
2231
4dd1fbc7 2232/*
568508e7
JH
2233 * This creates one packfile per large blob unless bulk-checkin
2234 * machinery is "plugged".
4dd1fbc7
JH
2235 *
2236 * This also bypasses the usual "convert-to-git" dance, and that is on
2237 * purpose. We could write a streaming version of the converting
2238 * functions and insert that before feeding the data to fast-import
4f22b101
JK
2239 * (or equivalent in-core API described above). However, that is
2240 * somewhat complicated, as we do not know the size of the filter
2241 * result, which we need to know beforehand when writing a git object.
2242 * Since the primary motivation for trying to stream from the working
2243 * tree file and to avoid mmaping it in core is to deal with large
2244 * binary blobs, they generally do not want to get any conversion, and
2245 * callers should avoid this code path when filters are requested.
4dd1fbc7 2246 */
7d5e1dc3 2247static int index_stream(struct object_id *oid, int fd, size_t size,
4dd1fbc7
JH
2248 enum object_type type, const char *path,
2249 unsigned flags)
2250{
68ee6dfc 2251 return index_bulk_checkin(oid, fd, size, type, path, flags);
4dd1fbc7
JH
2252}
2253
58bf2a4c
NTND
2254int index_fd(struct index_state *istate, struct object_id *oid,
2255 int fd, struct stat *st,
7b41e1e1
JH
2256 enum object_type type, const char *path, unsigned flags)
2257{
2258 int ret;
7b41e1e1 2259
9079ab7c
SP
2260 /*
2261 * Call xsize_t() only when needed to avoid potentially unnecessary
2262 * die() for large files.
2263 */
58bf2a4c
NTND
2264 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, path))
2265 ret = index_stream_convert_blob(istate, oid, fd, path, flags);
9035d75a 2266 else if (!S_ISREG(st->st_mode))
58bf2a4c 2267 ret = index_pipe(istate, oid, fd, type, path, flags);
9079ab7c 2268 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
58bf2a4c
NTND
2269 (path && would_convert_to_git(istate, path)))
2270 ret = index_core(istate, oid, fd, xsize_t(st->st_size),
2271 type, path, flags);
4dd1fbc7 2272 else
7d5e1dc3 2273 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
9079ab7c 2274 flags);
43df4f86 2275 close(fd);
aac17941 2276 return ret;
74400e71 2277}
ec1fcc16 2278
58bf2a4c
NTND
2279int index_path(struct index_state *istate, struct object_id *oid,
2280 const char *path, struct stat *st, unsigned flags)
ec1fcc16
JH
2281{
2282 int fd;
b760d3aa 2283 struct strbuf sb = STRBUF_INIT;
ea8e0297 2284 int rc = 0;
ec1fcc16
JH
2285
2286 switch (st->st_mode & S_IFMT) {
2287 case S_IFREG:
2288 fd = open(path, O_RDONLY);
2289 if (fd < 0)
7616c6ca 2290 return error_errno("open(\"%s\")", path);
58bf2a4c 2291 if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0)
259328b7 2292 return error(_("%s: failed to insert into database"),
ec1fcc16
JH
2293 path);
2294 break;
2295 case S_IFLNK:
7616c6ca
NTND
2296 if (strbuf_readlink(&sb, path, st->st_size))
2297 return error_errno("readlink(\"%s\")", path);
c4ce46fc 2298 if (!(flags & HASH_WRITE_OBJECT))
2dcde20e
MT
2299 hash_object_file(the_hash_algo, sb.buf, sb.len,
2300 blob_type, oid);
a09c985e 2301 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
259328b7 2302 rc = error(_("%s: failed to insert into database"), path);
b760d3aa 2303 strbuf_release(&sb);
ec1fcc16 2304 break;
f35a6d3b 2305 case S_IFDIR:
a98e6101 2306 return resolve_gitlink_ref(path, "HEAD", oid);
ec1fcc16 2307 default:
259328b7 2308 return error(_("%s: unsupported file type"), path);
ec1fcc16 2309 }
ea8e0297 2310 return rc;
ec1fcc16 2311}
a69e5429
JH
2312
2313int read_pack_header(int fd, struct pack_header *header)
2314{
f48ecd38 2315 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
c697ad14
HO
2316 /* "eof before pack header was fully read" */
2317 return PH_ERROR_EOF;
2318
a69e5429
JH
2319 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2320 /* "protocol error (pack signature mismatch detected)" */
2321 return PH_ERROR_PACK_SIGNATURE;
2322 if (!pack_version_ok(header->hdr_version))
2323 /* "protocol error (pack version unsupported)" */
2324 return PH_ERROR_PROTOCOL;
2325 return 0;
2326}
40d52ff7 2327
e816caa0 2328void assert_oid_type(const struct object_id *oid, enum object_type expect)
40d52ff7 2329{
0df8e965 2330 enum object_type type = oid_object_info(the_repository, oid, NULL);
40d52ff7 2331 if (type < 0)
259328b7 2332 die(_("%s is not a valid object"), oid_to_hex(oid));
40d52ff7 2333 if (type != expect)
259328b7 2334 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
debca9d2 2335 type_name(expect));
40d52ff7 2336}
27e1e22d 2337
70c49050 2338int for_each_file_in_obj_subdir(unsigned int subdir_nr,
cc817ca3
RS
2339 struct strbuf *path,
2340 each_loose_object_fn obj_cb,
2341 each_loose_cruft_fn cruft_cb,
2342 each_loose_subdir_fn subdir_cb,
2343 void *data)
27e1e22d 2344{
0375f472
RS
2345 size_t origlen, baselen;
2346 DIR *dir;
27e1e22d
JK
2347 struct dirent *de;
2348 int r = 0;
62a24c89 2349 struct object_id oid;
27e1e22d 2350
70c49050
RS
2351 if (subdir_nr > 0xff)
2352 BUG("invalid loose object subdirectory: %x", subdir_nr);
2353
0375f472
RS
2354 origlen = path->len;
2355 strbuf_complete(path, '/');
2356 strbuf_addf(path, "%02x", subdir_nr);
0375f472
RS
2357
2358 dir = opendir(path->buf);
27e1e22d 2359 if (!dir) {
0375f472 2360 if (errno != ENOENT)
259328b7 2361 r = error_errno(_("unable to open %s"), path->buf);
0375f472
RS
2362 strbuf_setlen(path, origlen);
2363 return r;
27e1e22d
JK
2364 }
2365
62a24c89 2366 oid.hash[0] = subdir_nr;
163ee5e6
DS
2367 strbuf_addch(path, '/');
2368 baselen = path->len;
62a24c89 2369
b548f0f1 2370 while ((de = readdir_skip_dot_and_dotdot(dir))) {
163ee5e6 2371 size_t namelen;
27e1e22d 2372
163ee5e6 2373 namelen = strlen(de->d_name);
27e1e22d 2374 strbuf_setlen(path, baselen);
163ee5e6 2375 strbuf_add(path, de->d_name, namelen);
94b5e093 2376 if (namelen == the_hash_algo->hexsz - 2 &&
62a24c89 2377 !hex_to_bytes(oid.hash + 1, de->d_name,
94b5e093 2378 the_hash_algo->rawsz - 1)) {
5a6dce70 2379 oid_set_algo(&oid, the_hash_algo);
62a24c89
RS
2380 if (obj_cb) {
2381 r = obj_cb(&oid, path->buf, data);
2382 if (r)
2383 break;
27e1e22d 2384 }
62a24c89 2385 continue;
27e1e22d
JK
2386 }
2387
2388 if (cruft_cb) {
2389 r = cruft_cb(de->d_name, path->buf, data);
2390 if (r)
2391 break;
2392 }
2393 }
094c7e63 2394 closedir(dir);
27e1e22d 2395
163ee5e6 2396 strbuf_setlen(path, baselen - 1);
27e1e22d
JK
2397 if (!r && subdir_cb)
2398 r = subdir_cb(subdir_nr, path->buf, data);
2399
0375f472
RS
2400 strbuf_setlen(path, origlen);
2401
27e1e22d
JK
2402 return r;
2403}
2404
e6f875e0 2405int for_each_loose_file_in_objdir_buf(struct strbuf *path,
27e1e22d
JK
2406 each_loose_object_fn obj_cb,
2407 each_loose_cruft_fn cruft_cb,
2408 each_loose_subdir_fn subdir_cb,
2409 void *data)
2410{
27e1e22d
JK
2411 int r = 0;
2412 int i;
2413
27e1e22d 2414 for (i = 0; i < 256; i++) {
e6f875e0 2415 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
27e1e22d 2416 subdir_cb, data);
27e1e22d
JK
2417 if (r)
2418 break;
2419 }
2420
e6f875e0
JK
2421 return r;
2422}
2423
2424int for_each_loose_file_in_objdir(const char *path,
2425 each_loose_object_fn obj_cb,
2426 each_loose_cruft_fn cruft_cb,
2427 each_loose_subdir_fn subdir_cb,
2428 void *data)
2429{
2430 struct strbuf buf = STRBUF_INIT;
2431 int r;
2432
2433 strbuf_addstr(&buf, path);
2434 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2435 subdir_cb, data);
27e1e22d 2436 strbuf_release(&buf);
e6f875e0 2437
27e1e22d
JK
2438 return r;
2439}
660c889e 2440
a7ff6f5a
JK
2441int for_each_loose_object(each_loose_object_fn cb, void *data,
2442 enum for_each_object_flags flags)
660c889e 2443{
f0eaf638 2444 struct object_directory *odb;
b0a42642 2445
f0eaf638
JK
2446 prepare_alt_odb(the_repository);
2447 for (odb = the_repository->objects->odb; odb; odb = odb->next) {
2448 int r = for_each_loose_file_in_objdir(odb->path, cb, NULL,
2449 NULL, data);
2450 if (r)
2451 return r;
660c889e 2452
f0eaf638
JK
2453 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2454 break;
2455 }
1385bb7b 2456
f0eaf638 2457 return 0;
660c889e
JK
2458}
2459
3a2e0824
JK
2460static int append_loose_object(const struct object_id *oid, const char *path,
2461 void *data)
660c889e 2462{
92d8ed8a 2463 oidtree_insert(data, oid);
3a2e0824
JK
2464 return 0;
2465}
660c889e 2466
92d8ed8a 2467struct oidtree *odb_loose_cache(struct object_directory *odb,
0000d654
RS
2468 const struct object_id *oid)
2469{
2470 int subdir_nr = oid->hash[0];
3a2e0824 2471 struct strbuf buf = STRBUF_INIT;
33f379ee
EW
2472 size_t word_bits = bitsizeof(odb->loose_objects_subdir_seen[0]);
2473 size_t word_index = subdir_nr / word_bits;
581a3bb1 2474 size_t mask = 1u << (subdir_nr % word_bits);
33f379ee 2475 uint32_t *bitmap;
660c889e 2476
3a2e0824 2477 if (subdir_nr < 0 ||
33f379ee 2478 subdir_nr >= bitsizeof(odb->loose_objects_subdir_seen))
3a2e0824 2479 BUG("subdir_nr out of range");
1385bb7b 2480
33f379ee
EW
2481 bitmap = &odb->loose_objects_subdir_seen[word_index];
2482 if (*bitmap & mask)
92d8ed8a
EW
2483 return odb->loose_objects_cache;
2484 if (!odb->loose_objects_cache) {
2485 ALLOC_ARRAY(odb->loose_objects_cache, 1);
2486 oidtree_init(odb->loose_objects_cache);
2487 }
3a2e0824
JK
2488 strbuf_addstr(&buf, odb->path);
2489 for_each_file_in_obj_subdir(subdir_nr, &buf,
2490 append_loose_object,
2491 NULL, NULL,
92d8ed8a 2492 odb->loose_objects_cache);
33f379ee 2493 *bitmap |= mask;
7317aa71 2494 strbuf_release(&buf);
92d8ed8a 2495 return odb->loose_objects_cache;
660c889e
JK
2496}
2497
d4e19e51
RS
2498void odb_clear_loose_cache(struct object_directory *odb)
2499{
92d8ed8a
EW
2500 oidtree_clear(odb->loose_objects_cache);
2501 FREE_AND_NULL(odb->loose_objects_cache);
d4e19e51
RS
2502 memset(&odb->loose_objects_subdir_seen, 0,
2503 sizeof(odb->loose_objects_subdir_seen));
2504}
2505
00a7760e
JK
2506static int check_stream_oid(git_zstream *stream,
2507 const char *hdr,
2508 unsigned long size,
2509 const char *path,
2510 const struct object_id *expected_oid)
f6371f92 2511{
18e2588e 2512 git_hash_ctx c;
00a7760e 2513 struct object_id real_oid;
f6371f92
JK
2514 unsigned char buf[4096];
2515 unsigned long total_read;
2516 int status = Z_OK;
2517
18e2588e 2518 the_hash_algo->init_fn(&c);
2519 the_hash_algo->update_fn(&c, hdr, stream->total_out);
f6371f92
JK
2520
2521 /*
2522 * We already read some bytes into hdr, but the ones up to the NUL
2523 * do not count against the object's content size.
2524 */
2525 total_read = stream->total_out - strlen(hdr) - 1;
2526
2527 /*
2528 * This size comparison must be "<=" to read the final zlib packets;
00a7760e 2529 * see the comment in unpack_loose_rest for details.
f6371f92
JK
2530 */
2531 while (total_read <= size &&
18ad13e5
JH
2532 (status == Z_OK ||
2533 (status == Z_BUF_ERROR && !stream->avail_out))) {
f6371f92
JK
2534 stream->next_out = buf;
2535 stream->avail_out = sizeof(buf);
2536 if (size - total_read < stream->avail_out)
2537 stream->avail_out = size - total_read;
2538 status = git_inflate(stream, Z_FINISH);
18e2588e 2539 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
f6371f92
JK
2540 total_read += stream->next_out - buf;
2541 }
2542 git_inflate_end(stream);
2543
2544 if (status != Z_STREAM_END) {
00a7760e 2545 error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid));
f6371f92
JK
2546 return -1;
2547 }
cce044df 2548 if (stream->avail_in) {
259328b7 2549 error(_("garbage at end of loose object '%s'"),
00a7760e 2550 oid_to_hex(expected_oid));
cce044df
JK
2551 return -1;
2552 }
f6371f92 2553
5951bf46 2554 the_hash_algo->final_oid_fn(&real_oid, &c);
00a7760e 2555 if (!oideq(expected_oid, &real_oid)) {
01f8d594 2556 error(_("hash mismatch for %s (expected %s)"), path,
00a7760e 2557 oid_to_hex(expected_oid));
f6371f92
JK
2558 return -1;
2559 }
2560
2561 return 0;
2562}
2563
2564int read_loose_object(const char *path,
d61d87bd 2565 const struct object_id *expected_oid,
f6371f92
JK
2566 enum object_type *type,
2567 unsigned long *size,
2568 void **contents)
2569{
2570 int ret = -1;
f6371f92
JK
2571 void *map = NULL;
2572 unsigned long mapsize;
2573 git_zstream stream;
1af64f73 2574 char hdr[MAX_HEADER_LEN];
f6371f92
JK
2575
2576 *contents = NULL;
2577
514c5fdd 2578 map = map_loose_object_1(the_repository, path, NULL, &mapsize);
f6371f92 2579 if (!map) {
259328b7 2580 error_errno(_("unable to mmap %s"), path);
f6371f92
JK
2581 goto out;
2582 }
2583
00a7760e 2584 if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
259328b7 2585 error(_("unable to unpack header of %s"), path);
f6371f92
JK
2586 goto out;
2587 }
2588
00a7760e 2589 *type = parse_loose_header(hdr, size);
f6371f92 2590 if (*type < 0) {
259328b7 2591 error(_("unable to parse header of %s"), path);
f6371f92
JK
2592 git_inflate_end(&stream);
2593 goto out;
2594 }
2595
7ac4f3a0 2596 if (*type == OBJ_BLOB && *size > big_file_threshold) {
00a7760e 2597 if (check_stream_oid(&stream, hdr, *size, path, expected_oid) < 0)
f6371f92
JK
2598 goto out;
2599 } else {
00a7760e 2600 *contents = unpack_loose_rest(&stream, hdr, *size, expected_oid);
f6371f92 2601 if (!*contents) {
259328b7 2602 error(_("unable to unpack contents of %s"), path);
f6371f92
JK
2603 git_inflate_end(&stream);
2604 goto out;
2605 }
b98d1885
MT
2606 if (check_object_signature(the_repository, expected_oid,
2607 *contents, *size,
2608 type_name(*type))) {
01f8d594 2609 error(_("hash mismatch for %s (expected %s)"), path,
d61d87bd 2610 oid_to_hex(expected_oid));
f6371f92
JK
2611 free(*contents);
2612 goto out;
2613 }
2614 }
2615
2616 ret = 0; /* everything checks out */
2617
2618out:
2619 if (map)
2620 munmap(map, mapsize);
f6371f92
JK
2621 return ret;
2622}