builtin/unpack-objects: switch uses of SHA-1 to the_hash_algo
[git/git.git] / sha1_file.c
CommitLineData
0fcfd160
LT
1/*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This handles basic git sha1 object files - packing, unpacking,
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"
628522ec 23#include "sha1-lookup.h"
568508e7 24#include "bulk-checkin.h"
090ea126 25#include "streaming.h"
543c5caa 26#include "dir.h"
a73cdd21 27#include "mru.h"
12d95ef6 28#include "list.h"
c4c6effa 29#include "mergesort.h"
cf3c6352 30#include "quote.h"
4f39cd82 31#include "packfile.h"
e05db0fd 32
50c5cd58 33const unsigned char null_sha1[GIT_MAX_RAWSZ];
3e56e724 34const struct object_id null_oid;
8576fde6
JK
35const struct object_id empty_tree_oid = {
36 EMPTY_TREE_SHA1_BIN_LITERAL
37};
38const struct object_id empty_blob_oid = {
39 EMPTY_BLOB_SHA1_BIN_LITERAL
40};
88cd621d 41
ac73cedf 42static void git_hash_sha1_init(git_hash_ctx *ctx)
f50e766b 43{
ac73cedf 44 git_SHA1_Init(&ctx->sha1);
f50e766b 45}
46
ac73cedf 47static void git_hash_sha1_update(git_hash_ctx *ctx, const void *data, size_t len)
f50e766b 48{
ac73cedf 49 git_SHA1_Update(&ctx->sha1, data, len);
f50e766b 50}
51
ac73cedf 52static void git_hash_sha1_final(unsigned char *hash, git_hash_ctx *ctx)
f50e766b 53{
ac73cedf 54 git_SHA1_Final(hash, &ctx->sha1);
f50e766b 55}
56
ac73cedf 57static void git_hash_unknown_init(git_hash_ctx *ctx)
f50e766b 58{
59 die("trying to init unknown hash");
60}
61
ac73cedf 62static void git_hash_unknown_update(git_hash_ctx *ctx, const void *data, size_t len)
f50e766b 63{
64 die("trying to update unknown hash");
65}
66
ac73cedf 67static void git_hash_unknown_final(unsigned char *hash, git_hash_ctx *ctx)
f50e766b 68{
69 die("trying to finalize unknown hash");
70}
71
72const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
73 {
74 NULL,
75 0x00000000,
76 0,
77 0,
f50e766b 78 git_hash_unknown_init,
79 git_hash_unknown_update,
80 git_hash_unknown_final,
81 NULL,
82 NULL,
83 },
84 {
85 "sha-1",
86 /* "sha1", big-endian */
87 0x73686131,
f50e766b 88 GIT_SHA1_RAWSZ,
89 GIT_SHA1_HEXSZ,
90 git_hash_sha1_init,
91 git_hash_sha1_update,
92 git_hash_sha1_final,
93 &empty_tree_oid,
94 &empty_blob_oid,
95 },
96};
97
c597ba80
NTND
98/*
99 * This is meant to hold a *small* number of objects that you would
100 * want read_sha1_file() to be able to return, but yet you do not want
101 * to write them into the object store (e.g. a browse-only
102 * application).
103 */
104static struct cached_object {
105 unsigned char sha1[20];
106 enum object_type type;
107 void *buf;
108 unsigned long size;
109} *cached_objects;
110static int cached_object_nr, cached_object_alloc;
111
112static struct cached_object empty_tree = {
dab0d410 113 EMPTY_TREE_SHA1_BIN_LITERAL,
c597ba80
NTND
114 OBJ_TREE,
115 "",
116 0
117};
118
119static struct cached_object *find_cached_object(const unsigned char *sha1)
120{
121 int i;
122 struct cached_object *co = cached_objects;
123
124 for (i = 0; i < cached_object_nr; i++, co++) {
125 if (!hashcmp(co->sha1, sha1))
126 return co;
127 }
128 if (!hashcmp(sha1, empty_tree.sha1))
129 return &empty_tree;
130 return NULL;
131}
132
9472935d
TB
133
134static enum safe_crlf get_safe_crlf(unsigned flags)
135{
136 if (flags & HASH_RENORMALIZE)
137 return SAFE_CRLF_RENORMALIZE;
138 else if (flags & HASH_WRITE_OBJECT)
139 return safe_crlf;
140 else
141 return SAFE_CRLF_FALSE;
142}
143
144
90a6464b
JH
145int mkdir_in_gitdir(const char *path)
146{
147 if (mkdir(path, 0777)) {
148 int saved_errno = errno;
149 struct stat st;
150 struct strbuf sb = STRBUF_INIT;
151
152 if (errno != EEXIST)
153 return -1;
154 /*
155 * Are we looking at a path in a symlinked worktree
156 * whose original repository does not yet have it?
157 * e.g. .git/rr-cache pointing at its original
158 * repository in which the user hasn't performed any
159 * conflict resolution yet?
160 */
161 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
162 strbuf_readlink(&sb, path, st.st_size) ||
163 !is_absolute_path(sb.buf) ||
164 mkdir(sb.buf, 0777)) {
165 strbuf_release(&sb);
166 errno = saved_errno;
167 return -1;
168 }
169 strbuf_release(&sb);
170 }
171 return adjust_shared_perm(path);
172}
173
0be0521b 174enum scld_error safe_create_leading_directories(char *path)
b2cb9425 175{
26c8ae2a 176 char *next_component = path + offset_1st_component(path);
0be0521b 177 enum scld_error ret = SCLD_OK;
67d42212 178
0be0521b 179 while (ret == SCLD_OK && next_component) {
f0502332 180 struct stat st;
0f527403 181 char *slash = next_component, slash_character;
f0502332 182
0f527403
MH
183 while (*slash && !is_dir_sep(*slash))
184 slash++;
185
186 if (!*slash)
b2cb9425 187 break;
bf10cf70 188
26c8ae2a 189 next_component = slash + 1;
0f527403 190 while (is_dir_sep(*next_component))
bf10cf70 191 next_component++;
26c8ae2a 192 if (!*next_component)
5f0bdf50 193 break;
831651fd 194
0f527403 195 slash_character = *slash;
831651fd 196 *slash = '\0';
67d42212
JR
197 if (!stat(path, &st)) {
198 /* path exists */
204a047f
MH
199 if (!S_ISDIR(st.st_mode)) {
200 errno = ENOTDIR;
0be0521b 201 ret = SCLD_EXISTS;
204a047f 202 }
53a39721 203 } else if (mkdir(path, 0777)) {
928734d9 204 if (errno == EEXIST &&
9e6f885d 205 !stat(path, &st) && S_ISDIR(st.st_mode))
928734d9 206 ; /* somebody created it since we checked */
18d37e86
MH
207 else if (errno == ENOENT)
208 /*
209 * Either mkdir() failed because
210 * somebody just pruned the containing
211 * directory, or stat() failed because
212 * the file that was in our way was
213 * just removed. Either way, inform
214 * the caller that it might be worth
215 * trying again:
216 */
217 ret = SCLD_VANISHED;
9e6f885d 218 else
0be0521b 219 ret = SCLD_FAILED;
53a39721 220 } else if (adjust_shared_perm(path)) {
0be0521b 221 ret = SCLD_PERMS;
457f06d6 222 }
0f527403 223 *slash = slash_character;
b2cb9425 224 }
9e6f885d 225 return ret;
b2cb9425 226}
723c31fe 227
0be0521b 228enum scld_error safe_create_leading_directories_const(const char *path)
8e21d63b 229{
02944307 230 int save_errno;
8e21d63b
JK
231 /* path points to cache entries, so xstrdup before messing with it */
232 char *buf = xstrdup(path);
0be0521b 233 enum scld_error result = safe_create_leading_directories(buf);
02944307
MH
234
235 save_errno = errno;
8e21d63b 236 free(buf);
02944307 237 errno = save_errno;
8e21d63b
JK
238 return result;
239}
240
177978f5
MH
241int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
242{
243 /*
244 * The number of times we will try to remove empty directories
245 * in the way of path. This is only 1 because if another
246 * process is racily creating directories that conflict with
247 * us, we don't want to fight against them.
248 */
249 int remove_directories_remaining = 1;
250
251 /*
252 * The number of times that we will try to create the
253 * directories containing path. We are willing to attempt this
254 * more than once, because another process could be trying to
255 * clean up empty directories at the same time as we are
256 * trying to create them.
257 */
258 int create_directories_remaining = 3;
259
260 /* A scratch copy of path, filled lazily if we need it: */
261 struct strbuf path_copy = STRBUF_INIT;
262
263 int ret, save_errno;
264
265 /* Sanity check: */
266 assert(*path);
267
268retry_fn:
269 ret = fn(path, cb);
270 save_errno = errno;
271 if (!ret)
272 goto out;
273
274 if (errno == EISDIR && remove_directories_remaining-- > 0) {
275 /*
276 * A directory is in the way. Maybe it is empty; try
277 * to remove it:
278 */
279 if (!path_copy.len)
280 strbuf_addstr(&path_copy, path);
281
282 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
283 goto retry_fn;
284 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
285 /*
286 * Maybe the containing directory didn't exist, or
287 * maybe it was just deleted by a process that is
288 * racing with us to clean up empty directories. Try
289 * to create it:
290 */
291 enum scld_error scld_result;
292
293 if (!path_copy.len)
294 strbuf_addstr(&path_copy, path);
295
296 do {
297 scld_result = safe_create_leading_directories(path_copy.buf);
298 if (scld_result == SCLD_OK)
299 goto retry_fn;
300 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
301 }
302
303out:
304 strbuf_release(&path_copy);
305 errno = save_errno;
306 return ret;
307}
308
f7b7774f 309static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
ace1534d
JH
310{
311 int i;
312 for (i = 0; i < 20; i++) {
313 static char hex[] = "0123456789abcdef";
314 unsigned int val = sha1[i];
f7b7774f
JK
315 strbuf_addch(buf, hex[val >> 4]);
316 strbuf_addch(buf, hex[val & 0xf]);
afbba2f0 317 if (!i)
f7b7774f 318 strbuf_addch(buf, '/');
ace1534d
JH
319 }
320}
321
30d6c6ea 322const char *sha1_file_name(const unsigned char *sha1)
0fcfd160 323{
f7b7774f 324 static struct strbuf buf = STRBUF_INIT;
0fcfd160 325
f7b7774f
JK
326 strbuf_reset(&buf);
327 strbuf_addf(&buf, "%s/", get_object_directory());
560fb6a1 328
f7b7774f
JK
329 fill_sha1_path(&buf, sha1);
330 return buf.buf;
0fcfd160
LT
331}
332
38dbe5f0
JK
333struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
334{
335 strbuf_setlen(&alt->scratch, alt->base_len);
336 return &alt->scratch;
337}
338
29ec6af2
JK
339static const char *alt_sha1_path(struct alternate_object_database *alt,
340 const unsigned char *sha1)
341{
38dbe5f0 342 struct strbuf *buf = alt_scratch_buf(alt);
f7b7774f 343 fill_sha1_path(buf, sha1);
38dbe5f0 344 return buf->buf;
0fcfd160
LT
345}
346
d5a63b99
JH
347struct alternate_object_database *alt_odb_list;
348static struct alternate_object_database **alt_odb_tail;
ace1534d 349
4ea82473
JK
350/*
351 * Return non-zero iff the path is usable as an alternate object database.
352 */
353static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
354{
355 struct alternate_object_database *alt;
356
357 /* Detect cases where alternate disappeared */
358 if (!is_directory(path->buf)) {
359 error("object directory %s does not exist; "
360 "check .git/objects/info/alternates.",
361 path->buf);
362 return 0;
363 }
364
365 /*
366 * Prevent the common mistake of listing the same
367 * thing twice, or object directory itself.
368 */
369 for (alt = alt_odb_list; alt; alt = alt->next) {
ea0fc3b4 370 if (!fspathcmp(path->buf, alt->path))
4ea82473
JK
371 return 0;
372 }
373 if (!fspathcmp(path->buf, normalized_objdir))
374 return 0;
375
376 return 1;
377}
378
ddd5d056
JH
379/*
380 * Prepare alternate object database registry.
d5a63b99
JH
381 *
382 * The variable alt_odb_list points at the list of struct
383 * alternate_object_database. The elements on this list come from
384 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
385 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
386 * whose contents is similar to that environment variable but can be
387 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
388 * contains "/the/directory/corresponding/to/.git/objects/...", while
389 * its name points just after the slash at the end of ".git/objects/"
390 * in the example above, and has enough space to hold 40-byte hex
391 * SHA1, an extra slash for the first level indirection, and the
392 * terminating NUL.
ddd5d056 393 */
2456990d 394static void read_info_alternates(const char * relative_base, int depth);
539e7506
EK
395static int link_alt_odb_entry(const char *entry, const char *relative_base,
396 int depth, const char *normalized_objdir)
ace1534d 397{
c2f493a4 398 struct alternate_object_database *ent;
5bdf0a84 399 struct strbuf pathbuf = STRBUF_INIT;
d5a63b99 400
85dadc38 401 if (!is_absolute_path(entry) && relative_base) {
4ac9006f 402 strbuf_realpath(&pathbuf, relative_base, 1);
5bdf0a84 403 strbuf_addch(&pathbuf, '/');
c2f493a4 404 }
6eac50d8 405 strbuf_addstr(&pathbuf, entry);
c2f493a4 406
37a95862 407 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
670c359d
JK
408 error("unable to normalize alternate object path: %s",
409 pathbuf.buf);
410 strbuf_release(&pathbuf);
411 return -1;
412 }
5bdf0a84
HW
413
414 /*
415 * The trailing slash after the directory name is given by
416 * this function at the end. Remove duplicates.
417 */
4ea82473
JK
418 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
419 strbuf_setlen(&pathbuf, pathbuf.len - 1);
5bdf0a84 420
4ea82473
JK
421 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
422 strbuf_release(&pathbuf);
c2f493a4
MW
423 return -1;
424 }
425
7f0fa2c0 426 ent = alloc_alt_odb(pathbuf.buf);
c2f493a4
MW
427
428 /* add the alternate entry */
429 *alt_odb_tail = ent;
430 alt_odb_tail = &(ent->next);
431 ent->next = NULL;
432
433 /* recursively add alternates */
4ea82473 434 read_info_alternates(pathbuf.buf, depth + 1);
c2f493a4 435
4ea82473 436 strbuf_release(&pathbuf);
c2f493a4
MW
437 return 0;
438}
439
cf3c6352
JK
440static const char *parse_alt_odb_entry(const char *string,
441 int sep,
442 struct strbuf *out)
443{
444 const char *end;
445
446 strbuf_reset(out);
447
448 if (*string == '#') {
449 /* comment; consume up to next separator */
450 end = strchrnul(string, sep);
451 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
452 /*
453 * quoted path; unquote_c_style has copied the
454 * data for us and set "end". Broken quoting (e.g.,
455 * an entry that doesn't end with a quote) falls
456 * back to the unquoted case below.
457 */
458 } else {
459 /* normal, unquoted path */
460 end = strchrnul(string, sep);
461 strbuf_add(out, string, end - string);
462 }
463
464 if (*end)
465 end++;
466 return end;
467}
468
dc732bd5 469static void link_alt_odb_entries(const char *alt, int sep,
c2f493a4
MW
470 const char *relative_base, int depth)
471{
539e7506 472 struct strbuf objdirbuf = STRBUF_INIT;
cf3c6352 473 struct strbuf entry = STRBUF_INIT;
c2f493a4 474
f28e3668
JK
475 if (!alt || !*alt)
476 return;
477
c2f493a4
MW
478 if (depth > 5) {
479 error("%s: ignoring alternate object stores, nesting too deep.",
480 relative_base);
481 return;
482 }
483
f655651e 484 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
670c359d
JK
485 if (strbuf_normalize_path(&objdirbuf) < 0)
486 die("unable to normalize object directory: %s",
487 objdirbuf.buf);
539e7506 488
cf3c6352
JK
489 while (*alt) {
490 alt = parse_alt_odb_entry(alt, sep, &entry);
491 if (!entry.len)
9577e7e3 492 continue;
cf3c6352 493 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
9577e7e3 494 }
cf3c6352 495 strbuf_release(&entry);
539e7506 496 strbuf_release(&objdirbuf);
d5a63b99
JH
497}
498
2456990d 499static void read_info_alternates(const char * relative_base, int depth)
d5a63b99 500{
5015f01c 501 char *path;
dc732bd5 502 struct strbuf buf = STRBUF_INIT;
d5a63b99 503
5015f01c 504 path = xstrfmt("%s/info/alternates", relative_base);
dc732bd5 505 if (strbuf_read_file(&buf, path, 1024) < 0) {
f0f7bebe 506 warn_on_fopen_errors(path);
dc732bd5 507 free(path);
9a217f2a 508 return;
ace1534d 509 }
d5a63b99 510
dc732bd5
JK
511 link_alt_odb_entries(buf.buf, '\n', relative_base, depth);
512 strbuf_release(&buf);
513 free(path);
ace1534d
JH
514}
515
7f0fa2c0
JK
516struct alternate_object_database *alloc_alt_odb(const char *dir)
517{
518 struct alternate_object_database *ent;
7f0fa2c0 519
597f9134 520 FLEX_ALLOC_STR(ent, path, dir);
38dbe5f0
JK
521 strbuf_init(&ent->scratch, 0);
522 strbuf_addf(&ent->scratch, "%s/", dir);
523 ent->base_len = ent->scratch.len;
7f0fa2c0
JK
524
525 return ent;
526}
527
bef70b22
DB
528void add_to_alternates_file(const char *reference)
529{
f132a127 530 struct lock_file lock = LOCK_INIT;
77b9b1d1
JK
531 char *alts = git_pathdup("objects/info/alternates");
532 FILE *in, *out;
f132a127 533 int found = 0;
77b9b1d1 534
f132a127
535 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
536 out = fdopen_lock_file(&lock, "w");
77b9b1d1
JK
537 if (!out)
538 die_errno("unable to fdopen alternates lockfile");
539
540 in = fopen(alts, "r");
541 if (in) {
542 struct strbuf line = STRBUF_INIT;
77b9b1d1 543
3f163962 544 while (strbuf_getline(&line, in) != EOF) {
77b9b1d1
JK
545 if (!strcmp(reference, line.buf)) {
546 found = 1;
547 break;
548 }
549 fprintf_or_die(out, "%s\n", line.buf);
550 }
551
552 strbuf_release(&line);
553 fclose(in);
77b9b1d1
JK
554 }
555 else if (errno != ENOENT)
556 die_errno("unable to read alternates file");
557
f132a127
558 if (found) {
559 rollback_lock_file(&lock);
560 } else {
77b9b1d1 561 fprintf_or_die(out, "%s\n", reference);
f132a127 562 if (commit_lock_file(&lock))
77b9b1d1
JK
563 die_errno("unable to move new alternates file into place");
564 if (alt_odb_tail)
dc732bd5 565 link_alt_odb_entries(reference, '\n', NULL, 0);
77b9b1d1
JK
566 }
567 free(alts);
bef70b22
DB
568}
569
a5b34d21
JK
570void add_to_alternates_memory(const char *reference)
571{
572 /*
573 * Make sure alternates are initialized, or else our entry may be
574 * overwritten when they are.
575 */
576 prepare_alt_odb();
577
dc732bd5 578 link_alt_odb_entries(reference, '\n', NULL, 0);
a5b34d21
JK
579}
580
9eeea7d2
SB
581/*
582 * Compute the exact path an alternate is at and returns it. In case of
583 * error NULL is returned and the human readable error is added to `err`
584 * `path` may be relative and should point to $GITDIR.
585 * `err` must not be null.
586 */
587char *compute_alternate_path(const char *path, struct strbuf *err)
588{
589 char *ref_git = NULL;
590 const char *repo, *ref_git_s;
591 int seen_error = 0;
592
593 ref_git_s = real_path_if_valid(path);
594 if (!ref_git_s) {
595 seen_error = 1;
596 strbuf_addf(err, _("path '%s' does not exist"), path);
597 goto out;
598 } else
599 /*
600 * Beware: read_gitfile(), real_path() and mkpath()
601 * return static buffer
602 */
603 ref_git = xstrdup(ref_git_s);
604
605 repo = read_gitfile(ref_git);
606 if (!repo)
607 repo = read_gitfile(mkpath("%s/.git", ref_git));
608 if (repo) {
609 free(ref_git);
610 ref_git = xstrdup(repo);
611 }
612
613 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
614 char *ref_git_git = mkpathdup("%s/.git", ref_git);
615 free(ref_git);
616 ref_git = ref_git_git;
617 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
618 struct strbuf sb = STRBUF_INIT;
619 seen_error = 1;
620 if (get_common_dir(&sb, ref_git)) {
621 strbuf_addf(err,
622 _("reference repository '%s' as a linked "
623 "checkout is not supported yet."),
624 path);
625 goto out;
626 }
627
628 strbuf_addf(err, _("reference repository '%s' is not a "
629 "local repository."), path);
630 goto out;
631 }
632
633 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
634 strbuf_addf(err, _("reference repository '%s' is shallow"),
635 path);
636 seen_error = 1;
637 goto out;
638 }
639
640 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
641 strbuf_addf(err,
642 _("reference repository '%s' is grafted"),
643 path);
644 seen_error = 1;
645 goto out;
646 }
647
648out:
649 if (seen_error) {
6a83d902 650 FREE_AND_NULL(ref_git);
9eeea7d2
SB
651 }
652
653 return ref_git;
654}
655
fe1b2268 656int foreach_alt_odb(alt_odb_fn fn, void *cb)
d79796bc
JH
657{
658 struct alternate_object_database *ent;
fe1b2268 659 int r = 0;
d79796bc
JH
660
661 prepare_alt_odb();
fe1b2268
JK
662 for (ent = alt_odb_list; ent; ent = ent->next) {
663 r = fn(ent, cb);
664 if (r)
665 break;
666 }
667 return r;
d79796bc
JH
668}
669
c2f493a4
MW
670void prepare_alt_odb(void)
671{
554fe20d 672 const char *alt;
c2f493a4 673
7dc24aa5
SP
674 if (alt_odb_tail)
675 return;
676
c2f493a4 677 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
c2f493a4 678
c2f493a4 679 alt_odb_tail = &alt_odb_list;
dc732bd5 680 link_alt_odb_entries(alt, PATH_SEP, NULL, 0);
c2f493a4
MW
681
682 read_info_alternates(get_object_directory(), 0);
683}
684
3096b2ec 685/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
33d4221c 686static int freshen_file(const char *fn)
ace1534d 687{
33d4221c
JK
688 struct utimbuf t;
689 t.actime = t.modtime = time(NULL);
690 return !utime(fn, &t);
0f4dc14a 691}
ace1534d 692
3096b2ec
JK
693/*
694 * All of the check_and_freshen functions return 1 if the file exists and was
695 * freshened (if freshening was requested), 0 otherwise. If they return
696 * 0, you should not assume that it is safe to skip a write of the object (it
697 * either does not exist on disk, or has a stale mtime and may be subject to
698 * pruning).
699 */
6a5e6f5e 700int check_and_freshen_file(const char *fn, int freshen)
33d4221c
JK
701{
702 if (access(fn, F_OK))
703 return 0;
3096b2ec 704 if (freshen && !freshen_file(fn))
33d4221c
JK
705 return 0;
706 return 1;
707}
708
709static int check_and_freshen_local(const unsigned char *sha1, int freshen)
710{
711 return check_and_freshen_file(sha1_file_name(sha1), freshen);
712}
713
714static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
0f4dc14a
BC
715{
716 struct alternate_object_database *alt;
9a217f2a 717 prepare_alt_odb();
d5a63b99 718 for (alt = alt_odb_list; alt; alt = alt->next) {
29ec6af2
JK
719 const char *path = alt_sha1_path(alt, sha1);
720 if (check_and_freshen_file(path, freshen))
c529d75a 721 return 1;
ace1534d 722 }
c529d75a 723 return 0;
ace1534d
JH
724}
725
33d4221c
JK
726static int check_and_freshen(const unsigned char *sha1, int freshen)
727{
728 return check_and_freshen_local(sha1, freshen) ||
729 check_and_freshen_nonlocal(sha1, freshen);
730}
731
732int has_loose_object_nonlocal(const unsigned char *sha1)
733{
734 return check_and_freshen_nonlocal(sha1, 0);
735}
736
0f4dc14a
BC
737static int has_loose_object(const unsigned char *sha1)
738{
33d4221c 739 return check_and_freshen(sha1, 0);
0f4dc14a
BC
740}
741
02710228
SP
742static void mmap_limit_check(size_t length)
743{
744 static size_t limit = 0;
745 if (!limit) {
746 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
747 if (!limit)
748 limit = SIZE_MAX;
749 }
750 if (length > limit)
751 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
752 (uintmax_t)length, (uintmax_t)limit);
753}
754
1570856b
JK
755void *xmmap_gently(void *start, size_t length,
756 int prot, int flags, int fd, off_t offset)
58ecbd5e 757{
02710228
SP
758 void *ret;
759
760 mmap_limit_check(length);
761 ret = mmap(start, length, prot, flags, fd, offset);
58ecbd5e
JN
762 if (ret == MAP_FAILED) {
763 if (!length)
764 return NULL;
7c3ecb32 765 release_pack_memory(length);
58ecbd5e 766 ret = mmap(start, length, prot, flags, fd, offset);
58ecbd5e
JN
767 }
768 return ret;
769}
770
1570856b
JK
771void *xmmap(void *start, size_t length,
772 int prot, int flags, int fd, off_t offset)
773{
774 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
775 if (ret == MAP_FAILED)
9ca0aaf6 776 die_errno("mmap failed");
1570856b
JK
777 return ret;
778}
779
88d0db55 780/*
090ea126
NTND
781 * With an in-core object data in "map", rehash it to make sure the
782 * object name actually matches "sha1" to detect object corruption.
783 * With "map" == NULL, try reading the object named with "sha1" using
784 * the streaming interface and rehash it to do the same.
88d0db55 785 */
090ea126
NTND
786int check_sha1_signature(const unsigned char *sha1, void *map,
787 unsigned long size, const char *type)
88d0db55 788{
0fcfd160 789 unsigned char real_sha1[20];
090ea126
NTND
790 enum object_type obj_type;
791 struct git_istream *st;
792 git_SHA_CTX c;
793 char hdr[32];
794 int hdrlen;
88d0db55 795
090ea126
NTND
796 if (map) {
797 hash_sha1_file(map, size, type, real_sha1);
798 return hashcmp(sha1, real_sha1) ? -1 : 0;
88d0db55
BC
799 }
800
090ea126
NTND
801 st = open_istream(sha1, &obj_type, &size, NULL);
802 if (!st)
803 return -1;
88d0db55 804
090ea126 805 /* Generate the header */
ef1286d3 806 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
88d0db55 807
090ea126
NTND
808 /* Sha1.. */
809 git_SHA1_Init(&c);
810 git_SHA1_Update(&c, hdr, hdrlen);
811 for (;;) {
812 char buf[1024 * 16];
813 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1f688557 814
f54fac53
JK
815 if (readlen < 0) {
816 close_istream(st);
817 return -1;
818 }
090ea126
NTND
819 if (!readlen)
820 break;
821 git_SHA1_Update(&c, buf, readlen);
fa5fc15d 822 }
090ea126
NTND
823 git_SHA1_Final(real_sha1, &c);
824 close_istream(st);
a89fccd2 825 return hashcmp(sha1, real_sha1) ? -1 : 0;
fa5fc15d
SP
826}
827
1b8ac5ea 828int git_open_cloexec(const char *name, int flags)
a0788266 829{
1e3001a8
JH
830 int fd;
831 static int o_cloexec = O_CLOEXEC;
a0788266 832
1e3001a8
JH
833 fd = open(name, flags | o_cloexec);
834 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
cd66ada0 835 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1e3001a8
JH
836 o_cloexec &= ~O_CLOEXEC;
837 fd = open(name, flags | o_cloexec);
491a8dec 838 }
491a8dec 839
9fb9495d 840#if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
491a8dec 841 {
1e3001a8 842 static int fd_cloexec = FD_CLOEXEC;
a0788266 843
1e3001a8
JH
844 if (!o_cloexec && 0 <= fd && fd_cloexec) {
845 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
9fb9495d
EW
846 int flags = fcntl(fd, F_GETFD);
847 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1e3001a8 848 fd_cloexec = 0;
cd66ada0 849 }
44d1c19e 850 }
a0788266 851#endif
1b8ac5ea 852 return fd;
a0788266
JS
853}
854
3cf8b462 855/*
771e7d57
JK
856 * Find "sha1" as a loose object in the local repository or in an alternate.
857 * Returns 0 on success, negative on failure.
858 *
859 * The "path" out-parameter will give the path of the object we found (if any).
860 * Note that it may point to static storage and is only valid until another
861 * call to sha1_file_name(), etc.
3cf8b462 862 */
771e7d57
JK
863static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
864 const char **path)
1f688557 865{
052fe5ea 866 struct alternate_object_database *alt;
d079837e 867
771e7d57
JK
868 *path = sha1_file_name(sha1);
869 if (!lstat(*path, st))
052fe5ea 870 return 0;
c7934306 871
052fe5ea
JK
872 prepare_alt_odb();
873 errno = ENOENT;
874 for (alt = alt_odb_list; alt; alt = alt->next) {
771e7d57
JK
875 *path = alt_sha1_path(alt, sha1);
876 if (!lstat(*path, st))
052fe5ea 877 return 0;
c7934306
SP
878 }
879
3cf8b462
SP
880 return -1;
881}
882
771e7d57
JK
883/*
884 * Like stat_sha1_file(), but actually open the object and return the
885 * descriptor. See the caveats on the "path" parameter above.
886 */
887static int open_sha1_file(const unsigned char *sha1, const char **path)
60bb8b14 888{
44d1c19e 889 int fd;
44d1c19e 890 struct alternate_object_database *alt;
d6c8a05b 891 int most_interesting_errno;
60bb8b14 892
771e7d57
JK
893 *path = sha1_file_name(sha1);
894 fd = git_open(*path);
44d1c19e
LT
895 if (fd >= 0)
896 return fd;
d6c8a05b 897 most_interesting_errno = errno;
03e79c88 898
44d1c19e 899 prepare_alt_odb();
44d1c19e 900 for (alt = alt_odb_list; alt; alt = alt->next) {
771e7d57
JK
901 *path = alt_sha1_path(alt, sha1);
902 fd = git_open(*path);
44d1c19e
LT
903 if (fd >= 0)
904 return fd;
d6c8a05b
JK
905 if (most_interesting_errno == ENOENT)
906 most_interesting_errno = errno;
03e79c88 907 }
d6c8a05b 908 errno = most_interesting_errno;
44d1c19e 909 return -1;
1f688557
JH
910}
911
f6371f92
JK
912/*
913 * Map the loose object at "path" if it is not NULL, or the path found by
914 * searching for a loose object named "sha1".
915 */
916static void *map_sha1_file_1(const char *path,
917 const unsigned char *sha1,
918 unsigned long *size)
27d69a46 919{
0fcfd160 920 void *map;
144bde78 921 int fd;
84dd81c1 922
f6371f92
JK
923 if (path)
924 fd = git_open(path);
8261e1f1 925 else
f6371f92 926 fd = open_sha1_file(sha1, &path);
44d1c19e
LT
927 map = NULL;
928 if (fd >= 0) {
929 struct stat st;
1f688557 930
44d1c19e
LT
931 if (!fstat(fd, &st)) {
932 *size = xsize_t(st.st_size);
33e42de0
MM
933 if (!*size) {
934 /* mmap() is forbidden on empty files */
771e7d57 935 error("object file %s is empty", path);
33e42de0
MM
936 return NULL;
937 }
44d1c19e 938 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 939 }
44d1c19e 940 close(fd);
74e34e1f 941 }
0fcfd160 942 return map;
74e34e1f
NP
943}
944
f6371f92 945void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
068f85e3 946{
f6371f92 947 return map_sha1_file_1(NULL, sha1, size);
068f85e3 948}
949
d21f8426
JH
950static int unpack_sha1_short_header(git_zstream *stream,
951 unsigned char *map, unsigned long mapsize,
952 void *buffer, unsigned long bufsiz)
47fe3f6e 953{
c4483576
LT
954 /* Get the data stream */
955 memset(stream, 0, sizeof(*stream));
956 stream->next_in = map;
957 stream->avail_in = mapsize;
958 stream->next_out = buffer;
93821bd9 959 stream->avail_out = bufsiz;
47fe3f6e 960
39c68542 961 git_inflate_init(stream);
cc5c54e7 962 return git_inflate(stream, 0);
9a217f2a
JH
963}
964
d21f8426
JH
965int unpack_sha1_header(git_zstream *stream,
966 unsigned char *map, unsigned long mapsize,
967 void *buffer, unsigned long bufsiz)
1f688557 968{
d21f8426
JH
969 int status = unpack_sha1_short_header(stream, map, mapsize,
970 buffer, bufsiz);
42873078 971
d21f8426
JH
972 if (status < Z_OK)
973 return status;
628522ec 974
d21f8426
JH
975 /* Make sure we have the terminating NUL */
976 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
977 return -1;
1f688557
JH
978 return 0;
979}
980
46f03448
KN
981static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
982 unsigned long mapsize, void *buffer,
983 unsigned long bufsiz, struct strbuf *header)
d131b7af 984{
46f03448
KN
985 int status;
986
d21f8426
JH
987 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
988 if (status < Z_OK)
989 return -1;
d131b7af 990
46f03448
KN
991 /*
992 * Check if entire header is unpacked in the first iteration.
d131b7af 993 */
46f03448
KN
994 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
995 return 0;
d131b7af 996
46f03448
KN
997 /*
998 * buffer[0..bufsiz] was not large enough. Copy the partial
999 * result out to header, and then append the result of further
1000 * reading the stream.
1001 */
1002 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1003 stream->next_out = buffer;
1004 stream->avail_out = bufsiz;
d131b7af 1005
46f03448
KN
1006 do {
1007 status = git_inflate(stream, 0);
1008 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1009 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1010 return 0;
1011 stream->next_out = buffer;
1012 stream->avail_out = bufsiz;
1013 } while (status != Z_STREAM_END);
1014 return -1;
d131b7af
SP
1015}
1016
ef49a7a0 1017static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
95099731 1018{
5180cacc 1019 int bytes = strlen(buffer) + 1;
3aee68aa 1020 unsigned char *buf = xmallocz(size);
93821bd9 1021 unsigned long n;
7efbff75 1022 int status = Z_OK;
95099731 1023
93821bd9
LT
1024 n = stream->total_out - bytes;
1025 if (n > size)
1026 n = size;
1027 memcpy(buf, (char *) buffer + bytes, n);
1028 bytes = n;
456cdf6e
LT
1029 if (bytes <= size) {
1030 /*
1031 * The above condition must be (bytes <= size), not
1032 * (bytes < size). In other words, even though we
ccf5ace0 1033 * expect no more output and set avail_out to zero,
456cdf6e
LT
1034 * the input zlib stream may have bytes that express
1035 * "this concludes the stream", and we *do* want to
1036 * eat that input.
1037 *
1038 * Otherwise we would not be able to test that we
1039 * consumed all the input to reach the expected size;
1040 * we also want to check that zlib tells us that all
1041 * went well with status == Z_STREAM_END at the end.
1042 */
5180cacc
LT
1043 stream->next_out = buf + bytes;
1044 stream->avail_out = size - bytes;
7efbff75 1045 while (status == Z_OK)
39c68542 1046 status = git_inflate(stream, Z_FINISH);
5180cacc 1047 }
456cdf6e 1048 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1049 git_inflate_end(stream);
7efbff75 1050 return buf;
95099731
NTND
1051 }
1052
7efbff75
JH
1053 if (status < 0)
1054 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1055 else if (stream->avail_in)
1056 error("garbage at end of loose object '%s'",
1057 sha1_to_hex(sha1));
1058 free(buf);
1059 return NULL;
95099731
NTND
1060}
1061
d40d535b 1062/*
5180cacc
LT
1063 * We used to just use "sscanf()", but that's actually way
1064 * too permissive for what we want to check. So do an anal
1065 * object header parse by hand.
d40d535b 1066 */
46f03448
KN
1067static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1068 unsigned int flags)
1f688557 1069{
46f03448 1070 const char *type_buf = hdr;
5180cacc 1071 unsigned long size;
46f03448 1072 int type, type_len = 0;
43057304 1073
5180cacc 1074 /*
46f03448 1075 * The type can be of any size but is followed by
21666f1a 1076 * a space.
5180cacc 1077 */
5180cacc
LT
1078 for (;;) {
1079 char c = *hdr++;
d21f8426
JH
1080 if (!c)
1081 return -1;
5180cacc
LT
1082 if (c == ' ')
1083 break;
46f03448 1084 type_len++;
5180cacc 1085 }
1f688557 1086
46f03448
KN
1087 type = type_from_string_gently(type_buf, type_len, 1);
1088 if (oi->typename)
1089 strbuf_add(oi->typename, type_buf, type_len);
1090 /*
1091 * Set type to 0 if its an unknown object and
2e3a16b2 1092 * we're obtaining the type using '--allow-unknown-type'
46f03448
KN
1093 * option.
1094 */
19fc5e84 1095 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
46f03448
KN
1096 type = 0;
1097 else if (type < 0)
1098 die("invalid object type");
1099 if (oi->typep)
1100 *oi->typep = type;
5180cacc
LT
1101
1102 /*
1103 * The length must follow immediately, and be in canonical
1104 * decimal format (ie "010" is not valid).
1105 */
1106 size = *hdr++ - '0';
1107 if (size > 9)
1108 return -1;
1109 if (size) {
1110 for (;;) {
1111 unsigned long c = *hdr - '0';
1112 if (c > 9)
1113 break;
1114 hdr++;
1115 size = size * 10 + c;
1b1005d1 1116 }
c01f51cc 1117 }
46f03448
KN
1118
1119 if (oi->sizep)
1120 *oi->sizep = size;
5180cacc
LT
1121
1122 /*
1123 * The length must be followed by a zero byte
1124 */
46f03448 1125 return *hdr ? -1 : type;
1f688557
JH
1126}
1127
46f03448 1128int parse_sha1_header(const char *hdr, unsigned long *sizep)
bf592c50 1129{
27b5c1a0 1130 struct object_info oi = OBJECT_INFO_INIT;
21666f1a 1131
46f03448 1132 oi.sizep = sizep;
1f0c0d36 1133 return parse_sha1_header_extended(hdr, &oi, 0);
bf592c50
DB
1134}
1135
052fe5ea 1136static int sha1_loose_object_info(const unsigned char *sha1,
46f03448
KN
1137 struct object_info *oi,
1138 int flags)
65c2e0c3 1139{
46f03448
KN
1140 int status = 0;
1141 unsigned long mapsize;
65c2e0c3 1142 void *map;
ef49a7a0 1143 git_zstream stream;
d65a16f6 1144 char hdr[32];
46f03448 1145 struct strbuf hdrbuf = STRBUF_INIT;
c84a1f3e 1146 unsigned long size_scratch;
65c2e0c3 1147
5d642e75
JK
1148 if (oi->delta_base_sha1)
1149 hashclr(oi->delta_base_sha1);
1150
052fe5ea
JK
1151 /*
1152 * If we don't care about type or size, then we don't
4ef8d1dd
JH
1153 * need to look inside the object at all. Note that we
1154 * do not optimize out the stat call, even if the
1155 * caller doesn't care about the disk-size, since our
1156 * return value implicitly indicates whether the
1157 * object even exists.
052fe5ea 1158 */
c84a1f3e 1159 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
771e7d57 1160 const char *path;
4ef8d1dd 1161 struct stat st;
771e7d57 1162 if (stat_sha1_file(sha1, &st, &path) < 0)
4ef8d1dd
JH
1163 return -1;
1164 if (oi->disk_sizep)
23c339c0 1165 *oi->disk_sizep = st.st_size;
052fe5ea
JK
1166 return 0;
1167 }
1168
bb6b8e4f 1169 map = map_sha1_file(sha1, &mapsize);
f0df4ed5 1170 if (!map)
dbea72a8 1171 return -1;
c84a1f3e
JT
1172
1173 if (!oi->sizep)
1174 oi->sizep = &size_scratch;
1175
23c339c0
JK
1176 if (oi->disk_sizep)
1177 *oi->disk_sizep = mapsize;
19fc5e84 1178 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
46f03448
KN
1179 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1180 status = error("unable to unpack %s header with --allow-unknown-type",
1181 sha1_to_hex(sha1));
1182 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
36e4d74a
JH
1183 status = error("unable to unpack %s header",
1184 sha1_to_hex(sha1));
46f03448
KN
1185 if (status < 0)
1186 ; /* Do nothing */
1187 else if (hdrbuf.len) {
1188 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1189 status = error("unable to parse %s header with --allow-unknown-type",
1190 sha1_to_hex(sha1));
1191 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
36e4d74a 1192 status = error("unable to parse %s header", sha1_to_hex(sha1));
c84a1f3e 1193
b3ea7dd3 1194 if (status >= 0 && oi->contentp) {
c84a1f3e
JT
1195 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1196 *oi->sizep, sha1);
b3ea7dd3
JK
1197 if (!*oi->contentp) {
1198 git_inflate_end(&stream);
1199 status = -1;
1200 }
1201 } else
c84a1f3e
JT
1202 git_inflate_end(&stream);
1203
65c2e0c3 1204 munmap(map, mapsize);
46f03448 1205 if (status && oi->typep)
23c339c0 1206 *oi->typep = status;
c84a1f3e
JT
1207 if (oi->sizep == &size_scratch)
1208 oi->sizep = NULL;
46f03448 1209 strbuf_release(&hdrbuf);
3ab0fb06 1210 oi->whence = OI_LOOSE;
93cff9a9 1211 return (status < 0) ? status : 0;
65c2e0c3
JH
1212}
1213
de7b5d62 1214int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
f0df4ed5 1215{
cd585e2a 1216 static struct object_info blank_oi = OBJECT_INFO_INIT;
f0df4ed5 1217 struct pack_entry e;
5b086407 1218 int rtype;
1f0c0d36
JT
1219 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
1220 lookup_replace_object(sha1) :
1221 sha1;
f0df4ed5 1222
87b5e236
JK
1223 if (is_null_sha1(real))
1224 return -1;
1225
cd585e2a
JT
1226 if (!oi)
1227 oi = &blank_oi;
1228
dfdd4afc
JT
1229 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1230 struct cached_object *co = find_cached_object(real);
1231 if (co) {
1232 if (oi->typep)
1233 *(oi->typep) = co->type;
1234 if (oi->sizep)
1235 *(oi->sizep) = co->size;
1236 if (oi->disk_sizep)
1237 *(oi->disk_sizep) = 0;
1238 if (oi->delta_base_sha1)
1239 hashclr(oi->delta_base_sha1);
1240 if (oi->typename)
1241 strbuf_addstr(oi->typename, typename(co->type));
1242 if (oi->contentp)
1243 *oi->contentp = xmemdupz(co->buf, co->size);
1244 oi->whence = OI_CACHED;
1245 return 0;
1246 }
c4d9986f
NTND
1247 }
1248
1f7117ef 1249 if (!find_pack_entry(real, &e)) {
ddd63e64 1250 /* Most likely it's a loose object. */
3ab0fb06 1251 if (!sha1_loose_object_info(real, oi, flags))
5b086407 1252 return 0;
ddd63e64
SG
1253
1254 /* Not a loose object; someone else may have just packed it. */
dfdd4afc 1255 if (flags & OBJECT_INFO_QUICK) {
052fe5ea 1256 return -1;
dfdd4afc
JT
1257 } else {
1258 reprepare_packed_git();
1259 if (!find_pack_entry(real, &e))
1260 return -1;
1261 }
f0df4ed5 1262 }
3d77d877 1263
cd585e2a
JT
1264 if (oi == &blank_oi)
1265 /*
1266 * We know that the caller doesn't actually need the
1267 * information below, so return early.
1268 */
1269 return 0;
46f03448 1270
23c339c0 1271 rtype = packed_object_info(e.p, e.offset, oi);
412916ee 1272 if (rtype < 0) {
1f7117ef
CC
1273 mark_bad_packed_object(e.p, real);
1274 return sha1_object_info_extended(real, oi, 0);
3ab0fb06 1275 } else if (oi->whence == OI_PACKED) {
9a490590
JH
1276 oi->u.packed.offset = e.offset;
1277 oi->u.packed.pack = e.p;
1278 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1279 rtype == OBJ_OFS_DELTA);
3d77d877
NP
1280 }
1281
5b086407 1282 return 0;
f0df4ed5
JS
1283}
1284
3fc0dca9 1285/* returns enum object_type or negative */
9a490590
JH
1286int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1287{
5b086407 1288 enum object_type type;
27b5c1a0 1289 struct object_info oi = OBJECT_INFO_INIT;
9a490590 1290
5b086407 1291 oi.typep = &type;
9a490590 1292 oi.sizep = sizep;
1f0c0d36
JT
1293 if (sha1_object_info_extended(sha1, &oi,
1294 OBJECT_INFO_LOOKUP_REPLACE) < 0)
5b086407
JK
1295 return -1;
1296 return type;
9a490590
JH
1297}
1298
f1d8130b
JT
1299static void *read_object(const unsigned char *sha1, enum object_type *type,
1300 unsigned long *size)
1301{
1302 struct object_info oi = OBJECT_INFO_INIT;
1303 void *content;
1304 oi.typep = type;
1305 oi.sizep = size;
1306 oi.contentp = &content;
1307
1308 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
1309 return NULL;
1310 return content;
1311}
1312
21666f1a
NP
1313int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1314 unsigned char *sha1)
d66b37bb
JH
1315{
1316 struct cached_object *co;
1317
21666f1a 1318 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
1319 if (has_sha1_file(sha1) || find_cached_object(sha1))
1320 return 0;
c7353967 1321 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
d66b37bb
JH
1322 co = &cached_objects[cached_object_nr++];
1323 co->size = len;
21666f1a 1324 co->type = type;
efa13f7b
JH
1325 co->buf = xmalloc(len);
1326 memcpy(co->buf, buf, len);
d66b37bb
JH
1327 hashcpy(co->sha1, sha1);
1328 return 0;
1329}
1330
b6c4cecc
JH
1331/*
1332 * This function dies on corrupt objects; the callers who want to
1333 * deal with them should arrange to call read_object() and give error
1334 * messages themselves.
1335 */
5bf29b95
JH
1336void *read_sha1_file_extended(const unsigned char *sha1,
1337 enum object_type *type,
1338 unsigned long *size,
1f0c0d36 1339 int lookup_replace)
ac939109 1340{
3ba7a065 1341 void *data;
b6c4cecc 1342 const struct packed_git *p;
771e7d57
JK
1343 const char *path;
1344 struct stat st;
1f0c0d36
JT
1345 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
1346 : sha1;
b6c4cecc 1347
3ba7a065
JH
1348 errno = 0;
1349 data = read_object(repl, type, size);
4bbf5a26 1350 if (data)
b6c4cecc 1351 return data;
68095570 1352
25f3af3f 1353 if (errno && errno != ENOENT)
3ba7a065
JH
1354 die_errno("failed to read object %s", sha1_to_hex(sha1));
1355
68095570 1356 /* die if we replaced an object with one that does not exist */
b6c4cecc 1357 if (repl != sha1)
68095570
CC
1358 die("replacement %s not found for %s",
1359 sha1_to_hex(repl), sha1_to_hex(sha1));
1360
771e7d57 1361 if (!stat_sha1_file(repl, &st, &path))
b6c4cecc
JH
1362 die("loose object %s (stored in %s) is corrupt",
1363 sha1_to_hex(repl), path);
68095570 1364
b6c4cecc
JH
1365 if ((p = has_packed_and_bad(repl)) != NULL)
1366 die("packed object %s (stored in %s) is corrupt",
1367 sha1_to_hex(repl), p->pack_name);
f5552aee 1368
b6c4cecc 1369 return NULL;
ac939109
NP
1370}
1371
40469ee9 1372void *read_object_with_reference(const unsigned char *sha1,
21666f1a 1373 const char *required_type_name,
40469ee9
JH
1374 unsigned long *size,
1375 unsigned char *actual_sha1_return)
f4913f91 1376{
21666f1a 1377 enum object_type type, required_type;
f4913f91
JH
1378 void *buffer;
1379 unsigned long isize;
40469ee9 1380 unsigned char actual_sha1[20];
f4913f91 1381
21666f1a 1382 required_type = type_from_string(required_type_name);
e702496e 1383 hashcpy(actual_sha1, sha1);
40469ee9
JH
1384 while (1) {
1385 int ref_length = -1;
1386 const char *ref_type = NULL;
f4913f91 1387
21666f1a 1388 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
1389 if (!buffer)
1390 return NULL;
21666f1a 1391 if (type == required_type) {
40469ee9
JH
1392 *size = isize;
1393 if (actual_sha1_return)
e702496e 1394 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
1395 return buffer;
1396 }
1397 /* Handle references */
21666f1a 1398 else if (type == OBJ_COMMIT)
40469ee9 1399 ref_type = "tree ";
21666f1a 1400 else if (type == OBJ_TAG)
40469ee9
JH
1401 ref_type = "object ";
1402 else {
1403 free(buffer);
1404 return NULL;
1405 }
1406 ref_length = strlen(ref_type);
f4913f91 1407
50974ec9
MK
1408 if (ref_length + 40 > isize ||
1409 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 1410 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
1411 free(buffer);
1412 return NULL;
1413 }
1cf58e72 1414 free(buffer);
40469ee9
JH
1415 /* Now we have the ID of the referred-to object in
1416 * actual_sha1. Check again. */
f4913f91 1417 }
f4913f91
JH
1418}
1419
ce9fbf16 1420static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 1421 const char *type, unsigned char *sha1,
d65a16f6 1422 char *hdr, int *hdrlen)
d410c0f5 1423{
9126f009 1424 git_SHA_CTX c;
d410c0f5
JH
1425
1426 /* Generate the header */
ef1286d3 1427 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
d410c0f5
JH
1428
1429 /* Sha1.. */
9126f009
NP
1430 git_SHA1_Init(&c);
1431 git_SHA1_Update(&c, hdr, *hdrlen);
1432 git_SHA1_Update(&c, buf, len);
1433 git_SHA1_Final(sha1, &c);
d410c0f5
JH
1434}
1435
230f1322 1436/*
5a688fe4 1437 * Move the just written object into its final resting place.
230f1322 1438 */
cb5add58 1439int finalize_object_file(const char *tmpfile, const char *filename)
230f1322 1440{
e32c0a9c 1441 int ret = 0;
5a688fe4 1442
348df166 1443 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
1444 goto try_rename;
1445 else if (link(tmpfile, filename))
e32c0a9c 1446 ret = errno;
7ebb6fca
LT
1447
1448 /*
1449 * Coda hack - coda doesn't like cross-directory links,
1450 * so we fall back to a rename, which will mean that it
1451 * won't be able to check collisions, but that's not a
1452 * big deal.
1453 *
1454 * The same holds for FAT formatted media.
1455 *
3be1f18e 1456 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
1457 * left to unlink.
1458 */
1459 if (ret && ret != EEXIST) {
be66a6c4 1460 try_rename:
7ebb6fca 1461 if (!rename(tmpfile, filename))
3be1f18e 1462 goto out;
9e48b389 1463 ret = errno;
230f1322 1464 }
691f1a28 1465 unlink_or_warn(tmpfile);
230f1322
LT
1466 if (ret) {
1467 if (ret != EEXIST) {
7616c6ca 1468 return error_errno("unable to write sha1 filename %s", filename);
230f1322
LT
1469 }
1470 /* FIXME!!! Collision check here ? */
1471 }
1472
3be1f18e 1473out:
5256b006 1474 if (adjust_shared_perm(filename))
5a688fe4 1475 return error("unable to set permission to '%s'", filename);
230f1322
LT
1476 return 0;
1477}
1478
4d548150
LT
1479static int write_buffer(int fd, const void *buf, size_t len)
1480{
d34cf19b 1481 if (write_in_full(fd, buf, len) < 0)
7616c6ca 1482 return error_errno("file write error");
4d548150
LT
1483 return 0;
1484}
1485
ce9fbf16 1486int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
1487 unsigned char *sha1)
1488{
d65a16f6 1489 char hdr[32];
ef1286d3 1490 int hdrlen = sizeof(hdr);
abdc3fc8
RS
1491 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1492 return 0;
1493}
1494
e9039dd3
LT
1495/* Finalize a file on disk, and close it. */
1496static void close_sha1_file(int fd)
1497{
aafe9fba
LT
1498 if (fsync_object_files)
1499 fsync_or_die(fd, "sha1 file");
e9039dd3 1500 if (close(fd) != 0)
d824cbba 1501 die_errno("error when closing sha1 file");
e9039dd3
LT
1502}
1503
5723fe7e
LT
1504/* Size of directory component, including the ending '/' */
1505static inline int directory_size(const char *filename)
1506{
1507 const char *s = strrchr(filename, '/');
1508 if (!s)
1509 return 0;
1510 return s - filename + 1;
1511}
1512
1513/*
1514 * This creates a temporary file in the same directory as the final
1515 * 'filename'
1516 *
1517 * We want to avoid cross-directory filename renames, because those
1518 * can have problems on various filesystems (FAT, NFS, Coda).
1519 */
d4b3d11a 1520static int create_tmpfile(struct strbuf *tmp, const char *filename)
5723fe7e
LT
1521{
1522 int fd, dirlen = directory_size(filename);
1523
d4b3d11a
JK
1524 strbuf_reset(tmp);
1525 strbuf_add(tmp, filename, dirlen);
1526 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1527 fd = git_mkstemp_mode(tmp->buf, 0444);
cbacbf4e 1528 if (fd < 0 && dirlen && errno == ENOENT) {
d4b3d11a
JK
1529 /*
1530 * Make sure the directory exists; note that the contents
1531 * of the buffer are undefined after mkstemp returns an
1532 * error, so we have to rewrite the whole buffer from
1533 * scratch.
1534 */
1535 strbuf_reset(tmp);
1536 strbuf_add(tmp, filename, dirlen - 1);
1537 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
b2476a60 1538 return -1;
d4b3d11a 1539 if (adjust_shared_perm(tmp->buf))
5723fe7e
LT
1540 return -1;
1541
1542 /* Try again */
d4b3d11a
JK
1543 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1544 fd = git_mkstemp_mode(tmp->buf, 0444);
5723fe7e
LT
1545 }
1546 return fd;
1547}
1548
bbac7311 1549static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 1550 const void *buf, unsigned long len, time_t mtime)
0fcfd160 1551{
915308b1 1552 int fd, ret;
9892beba 1553 unsigned char compressed[4096];
ef49a7a0 1554 git_zstream stream;
748af44c
NP
1555 git_SHA_CTX c;
1556 unsigned char parano_sha1[20];
d4b3d11a 1557 static struct strbuf tmp_file = STRBUF_INIT;
30d6c6ea 1558 const char *filename = sha1_file_name(sha1);
a44c9a5e 1559
d4b3d11a 1560 fd = create_tmpfile(&tmp_file, filename);
aac17941 1561 if (fd < 0) {
35243577 1562 if (errno == EACCES)
82247e9b 1563 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
916d081b 1564 else
7616c6ca 1565 return error_errno("unable to create temporary file");
aac17941
LT
1566 }
1567
0fcfd160 1568 /* Set it up */
55bb5c91 1569 git_deflate_init(&stream, zlib_compression_level);
0fcfd160 1570 stream.next_out = compressed;
9892beba 1571 stream.avail_out = sizeof(compressed);
748af44c 1572 git_SHA1_Init(&c);
a44c9a5e
LT
1573
1574 /* First header.. */
d65a16f6 1575 stream.next_in = (unsigned char *)hdr;
a44c9a5e 1576 stream.avail_in = hdrlen;
55bb5c91
JH
1577 while (git_deflate(&stream, 0) == Z_OK)
1578 ; /* nothing */
748af44c 1579 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
1580
1581 /* Then the data itself.. */
c00e657d 1582 stream.next_in = (void *)buf;
a44c9a5e 1583 stream.avail_in = len;
9892beba 1584 do {
748af44c 1585 unsigned char *in0 = stream.next_in;
55bb5c91 1586 ret = git_deflate(&stream, Z_FINISH);
748af44c 1587 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
1588 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1589 die("unable to write sha1 file");
1590 stream.next_out = compressed;
1591 stream.avail_out = sizeof(compressed);
1592 } while (ret == Z_OK);
1593
ac54c277
LT
1594 if (ret != Z_STREAM_END)
1595 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
55bb5c91 1596 ret = git_deflate_end_gently(&stream);
ac54c277
LT
1597 if (ret != Z_OK)
1598 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
1599 git_SHA1_Final(parano_sha1, &c);
1600 if (hashcmp(sha1, parano_sha1) != 0)
1601 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 1602
e9039dd3 1603 close_sha1_file(fd);
0fcfd160 1604
bbac7311
NP
1605 if (mtime) {
1606 struct utimbuf utb;
1607 utb.actime = mtime;
1608 utb.modtime = mtime;
d4b3d11a 1609 if (utime(tmp_file.buf, &utb) < 0)
7616c6ca 1610 warning_errno("failed utime() on %s", tmp_file.buf);
bbac7311
NP
1611 }
1612
d4b3d11a 1613 return finalize_object_file(tmp_file.buf, filename);
0fcfd160 1614}
8237b185 1615
33d4221c
JK
1616static int freshen_loose_object(const unsigned char *sha1)
1617{
1618 return check_and_freshen(sha1, 1);
1619}
1620
1621static int freshen_packed_object(const unsigned char *sha1)
1622{
1623 struct pack_entry e;
ee1c6c34
JK
1624 if (!find_pack_entry(sha1, &e))
1625 return 0;
1626 if (e.p->freshened)
1627 return 1;
1628 if (!freshen_file(e.p->pack_name))
1629 return 0;
1630 e.p->freshened = 1;
1631 return 1;
33d4221c
JK
1632}
1633
1427a7ff 1634int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
bbac7311 1635{
bbac7311 1636 char hdr[32];
ef1286d3 1637 int hdrlen = sizeof(hdr);
bbac7311
NP
1638
1639 /* Normally if we have it in the pack then we do not bother writing
1640 * it out into .git/objects/??/?{38} file.
1641 */
1642 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
b5f52f37 1643 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
bbac7311
NP
1644 return 0;
1645 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
1646}
1647
0c3db67c 1648int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
da77611d 1649 struct object_id *oid, unsigned flags)
0c3db67c
ES
1650{
1651 char *header;
1652 int hdrlen, status = 0;
1653
1654 /* type string, SP, %lu of the length plus NUL must fit this */
ef1286d3
JK
1655 hdrlen = strlen(type) + 32;
1656 header = xmalloc(hdrlen);
da77611d 1657 write_sha1_file_prepare(buf, len, type, oid->hash, header, &hdrlen);
0c3db67c
ES
1658
1659 if (!(flags & HASH_WRITE_OBJECT))
1660 goto cleanup;
da77611d 1661 if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
0c3db67c 1662 goto cleanup;
da77611d 1663 status = write_loose_object(oid->hash, header, hdrlen, buf, len, 0);
0c3db67c
ES
1664
1665cleanup:
1666 free(header);
1667 return status;
1668}
1669
bbac7311
NP
1670int force_object_loose(const unsigned char *sha1, time_t mtime)
1671{
bbac7311
NP
1672 void *buf;
1673 unsigned long len;
1674 enum object_type type;
1675 char hdr[32];
1676 int hdrlen;
1fb23e65 1677 int ret;
bbac7311 1678
c529d75a 1679 if (has_loose_object(sha1))
bbac7311 1680 return 0;
789bf26b 1681 buf = read_object(sha1, &type, &len);
bbac7311
NP
1682 if (!buf)
1683 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
ef1286d3 1684 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1fb23e65
BS
1685 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
1686 free(buf);
1687
1688 return ret;
bbac7311
NP
1689}
1690
0eeb077b 1691int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
8237b185 1692{
3e8b7d3c
JN
1693 if (!startup_info->have_repository)
1694 return 0;
e83e71c5
JT
1695 return sha1_object_info_extended(sha1, NULL,
1696 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
b419aa25 1697}
1698
1699int has_object_file(const struct object_id *oid)
1700{
1701 return has_sha1_file(oid->hash);
8237b185 1702}
74400e71 1703
5827a035
JK
1704int has_object_file_with_flags(const struct object_id *oid, int flags)
1705{
1706 return has_sha1_file_with_flags(oid->hash, flags);
1707}
1708
c879daa2
NTND
1709static void check_tree(const void *buf, size_t size)
1710{
1711 struct tree_desc desc;
1712 struct name_entry entry;
1713
1714 init_tree_desc(&desc, buf, size);
1715 while (tree_entry(&desc, &entry))
1716 /* do nothing
1717 * tree_entry() will die() on malformed entries */
1718 ;
1719}
1720
1721static void check_commit(const void *buf, size_t size)
1722{
1723 struct commit c;
1724 memset(&c, 0, sizeof(c));
1725 if (parse_commit_buffer(&c, buf, size))
1726 die("corrupt commit");
1727}
1728
1729static void check_tag(const void *buf, size_t size)
1730{
1731 struct tag t;
1732 memset(&t, 0, sizeof(t));
1733 if (parse_tag_buffer(&t, buf, size))
1734 die("corrupt tag");
1735}
1736
bcd29864 1737static int index_mem(struct object_id *oid, void *buf, size_t size,
c4ce46fc
JH
1738 enum object_type type,
1739 const char *path, unsigned flags)
e7332f96 1740{
6c510bee 1741 int ret, re_allocated = 0;
c4ce46fc 1742 int write_object = flags & HASH_WRITE_OBJECT;
74400e71 1743
7672db20 1744 if (!type)
edaec3fb 1745 type = OBJ_BLOB;
6c510bee
LT
1746
1747 /*
1748 * Convert blobs to git internal format
1749 */
43df4f86 1750 if ((type == OBJ_BLOB) && path) {
f285a2d7 1751 struct strbuf nbuf = STRBUF_INIT;
82b474e0 1752 if (convert_to_git(&the_index, path, buf, size, &nbuf,
9472935d 1753 get_safe_crlf(flags))) {
b315c5c0 1754 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
1755 re_allocated = 1;
1756 }
1757 }
c4ce46fc 1758 if (flags & HASH_FORMAT_CHECK) {
c879daa2
NTND
1759 if (type == OBJ_TREE)
1760 check_tree(buf, size);
1761 if (type == OBJ_COMMIT)
1762 check_commit(buf, size);
1763 if (type == OBJ_TAG)
1764 check_tag(buf, size);
1765 }
6c510bee 1766
7672db20 1767 if (write_object)
bcd29864 1768 ret = write_sha1_file(buf, size, typename(type), oid->hash);
abdc3fc8 1769 else
bcd29864 1770 ret = hash_sha1_file(buf, size, typename(type), oid->hash);
43df4f86 1771 if (re_allocated)
6c510bee 1772 free(buf);
43df4f86
DP
1773 return ret;
1774}
1775
bcd29864 1776static int index_stream_convert_blob(struct object_id *oid, int fd,
9035d75a
SP
1777 const char *path, unsigned flags)
1778{
1779 int ret;
1780 const int write_object = flags & HASH_WRITE_OBJECT;
1781 struct strbuf sbuf = STRBUF_INIT;
1782
1783 assert(path);
1784 assert(would_convert_to_git_filter_fd(path));
1785
d6c41c20 1786 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
9472935d 1787 get_safe_crlf(flags));
9035d75a
SP
1788
1789 if (write_object)
1790 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
bcd29864 1791 oid->hash);
9035d75a
SP
1792 else
1793 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
bcd29864 1794 oid->hash);
9035d75a
SP
1795 strbuf_release(&sbuf);
1796 return ret;
1797}
1798
bcd29864 1799static int index_pipe(struct object_id *oid, int fd, enum object_type type,
7b41e1e1
JH
1800 const char *path, unsigned flags)
1801{
1802 struct strbuf sbuf = STRBUF_INIT;
1803 int ret;
1804
1805 if (strbuf_read(&sbuf, fd, 4096) >= 0)
bcd29864 1806 ret = index_mem(oid, sbuf.buf, sbuf.len, type, path, flags);
7b41e1e1
JH
1807 else
1808 ret = -1;
1809 strbuf_release(&sbuf);
1810 return ret;
1811}
1812
ea68b0ce
DP
1813#define SMALL_FILE_SIZE (32*1024)
1814
bcd29864 1815static int index_core(struct object_id *oid, int fd, size_t size,
7b41e1e1
JH
1816 enum object_type type, const char *path,
1817 unsigned flags)
43df4f86
DP
1818{
1819 int ret;
43df4f86 1820
7b41e1e1 1821 if (!size) {
bcd29864 1822 ret = index_mem(oid, "", size, type, path, flags);
ea68b0ce
DP
1823 } else if (size <= SMALL_FILE_SIZE) {
1824 char *buf = xmalloc(size);
90dca671
JK
1825 ssize_t read_result = read_in_full(fd, buf, size);
1826 if (read_result < 0)
1827 ret = error_errno("read error while indexing %s",
1828 path ? path : "<unknown>");
1829 else if (read_result != size)
1830 ret = error("short read while indexing %s",
1831 path ? path : "<unknown>");
ea68b0ce 1832 else
bcd29864 1833 ret = index_mem(oid, buf, size, type, path, flags);
ea68b0ce 1834 free(buf);
08bda208 1835 } else {
43df4f86 1836 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
bcd29864 1837 ret = index_mem(oid, buf, size, type, path, flags);
aac17941 1838 munmap(buf, size);
08bda208 1839 }
7b41e1e1
JH
1840 return ret;
1841}
1842
4dd1fbc7 1843/*
568508e7
JH
1844 * This creates one packfile per large blob unless bulk-checkin
1845 * machinery is "plugged".
4dd1fbc7
JH
1846 *
1847 * This also bypasses the usual "convert-to-git" dance, and that is on
1848 * purpose. We could write a streaming version of the converting
1849 * functions and insert that before feeding the data to fast-import
4f22b101
JK
1850 * (or equivalent in-core API described above). However, that is
1851 * somewhat complicated, as we do not know the size of the filter
1852 * result, which we need to know beforehand when writing a git object.
1853 * Since the primary motivation for trying to stream from the working
1854 * tree file and to avoid mmaping it in core is to deal with large
1855 * binary blobs, they generally do not want to get any conversion, and
1856 * callers should avoid this code path when filters are requested.
4dd1fbc7 1857 */
7d5e1dc3 1858static int index_stream(struct object_id *oid, int fd, size_t size,
4dd1fbc7
JH
1859 enum object_type type, const char *path,
1860 unsigned flags)
1861{
7d5e1dc3 1862 return index_bulk_checkin(oid->hash, fd, size, type, path, flags);
4dd1fbc7
JH
1863}
1864
e3506559 1865int index_fd(struct object_id *oid, int fd, struct stat *st,
7b41e1e1
JH
1866 enum object_type type, const char *path, unsigned flags)
1867{
1868 int ret;
7b41e1e1 1869
9079ab7c
SP
1870 /*
1871 * Call xsize_t() only when needed to avoid potentially unnecessary
1872 * die() for large files.
1873 */
9035d75a 1874 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
bcd29864 1875 ret = index_stream_convert_blob(oid, fd, path, flags);
9035d75a 1876 else if (!S_ISREG(st->st_mode))
bcd29864 1877 ret = index_pipe(oid, fd, type, path, flags);
9079ab7c 1878 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
82b474e0 1879 (path && would_convert_to_git(&the_index, path)))
bcd29864 1880 ret = index_core(oid, fd, xsize_t(st->st_size), type, path,
9079ab7c 1881 flags);
4dd1fbc7 1882 else
7d5e1dc3 1883 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
9079ab7c 1884 flags);
43df4f86 1885 close(fd);
aac17941 1886 return ret;
74400e71 1887}
ec1fcc16 1888
98e019b0 1889int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
ec1fcc16
JH
1890{
1891 int fd;
b760d3aa 1892 struct strbuf sb = STRBUF_INIT;
ea8e0297 1893 int rc = 0;
ec1fcc16
JH
1894
1895 switch (st->st_mode & S_IFMT) {
1896 case S_IFREG:
1897 fd = open(path, O_RDONLY);
1898 if (fd < 0)
7616c6ca 1899 return error_errno("open(\"%s\")", path);
e3506559 1900 if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
ec1fcc16
JH
1901 return error("%s: failed to insert into database",
1902 path);
1903 break;
1904 case S_IFLNK:
7616c6ca
NTND
1905 if (strbuf_readlink(&sb, path, st->st_size))
1906 return error_errno("readlink(\"%s\")", path);
c4ce46fc 1907 if (!(flags & HASH_WRITE_OBJECT))
98e019b0
PO
1908 hash_sha1_file(sb.buf, sb.len, blob_type, oid->hash);
1909 else if (write_sha1_file(sb.buf, sb.len, blob_type, oid->hash))
ea8e0297 1910 rc = error("%s: failed to insert into database", path);
b760d3aa 1911 strbuf_release(&sb);
ec1fcc16 1912 break;
f35a6d3b 1913 case S_IFDIR:
a98e6101 1914 return resolve_gitlink_ref(path, "HEAD", oid);
ec1fcc16
JH
1915 default:
1916 return error("%s: unsupported file type", path);
1917 }
ea8e0297 1918 return rc;
ec1fcc16 1919}
a69e5429
JH
1920
1921int read_pack_header(int fd, struct pack_header *header)
1922{
f48ecd38 1923 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
c697ad14
HO
1924 /* "eof before pack header was fully read" */
1925 return PH_ERROR_EOF;
1926
a69e5429
JH
1927 if (header->hdr_signature != htonl(PACK_SIGNATURE))
1928 /* "protocol error (pack signature mismatch detected)" */
1929 return PH_ERROR_PACK_SIGNATURE;
1930 if (!pack_version_ok(header->hdr_version))
1931 /* "protocol error (pack version unsupported)" */
1932 return PH_ERROR_PROTOCOL;
1933 return 0;
1934}
40d52ff7
JK
1935
1936void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
1937{
1938 enum object_type type = sha1_object_info(sha1, NULL);
1939 if (type < 0)
1940 die("%s is not a valid object", sha1_to_hex(sha1));
1941 if (type != expect)
1942 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
1943 typename(expect));
1944}
27e1e22d 1945
70c49050 1946int for_each_file_in_obj_subdir(unsigned int subdir_nr,
cc817ca3
RS
1947 struct strbuf *path,
1948 each_loose_object_fn obj_cb,
1949 each_loose_cruft_fn cruft_cb,
1950 each_loose_subdir_fn subdir_cb,
1951 void *data)
27e1e22d 1952{
0375f472
RS
1953 size_t origlen, baselen;
1954 DIR *dir;
27e1e22d
JK
1955 struct dirent *de;
1956 int r = 0;
62a24c89 1957 struct object_id oid;
27e1e22d 1958
70c49050
RS
1959 if (subdir_nr > 0xff)
1960 BUG("invalid loose object subdirectory: %x", subdir_nr);
1961
0375f472
RS
1962 origlen = path->len;
1963 strbuf_complete(path, '/');
1964 strbuf_addf(path, "%02x", subdir_nr);
0375f472
RS
1965
1966 dir = opendir(path->buf);
27e1e22d 1967 if (!dir) {
0375f472
RS
1968 if (errno != ENOENT)
1969 r = error_errno("unable to open %s", path->buf);
1970 strbuf_setlen(path, origlen);
1971 return r;
27e1e22d
JK
1972 }
1973
62a24c89 1974 oid.hash[0] = subdir_nr;
163ee5e6
DS
1975 strbuf_addch(path, '/');
1976 baselen = path->len;
62a24c89 1977
27e1e22d 1978 while ((de = readdir(dir))) {
163ee5e6 1979 size_t namelen;
27e1e22d
JK
1980 if (is_dot_or_dotdot(de->d_name))
1981 continue;
1982
163ee5e6 1983 namelen = strlen(de->d_name);
27e1e22d 1984 strbuf_setlen(path, baselen);
163ee5e6
DS
1985 strbuf_add(path, de->d_name, namelen);
1986 if (namelen == GIT_SHA1_HEXSZ - 2 &&
62a24c89
RS
1987 !hex_to_bytes(oid.hash + 1, de->d_name,
1988 GIT_SHA1_RAWSZ - 1)) {
1989 if (obj_cb) {
1990 r = obj_cb(&oid, path->buf, data);
1991 if (r)
1992 break;
27e1e22d 1993 }
62a24c89 1994 continue;
27e1e22d
JK
1995 }
1996
1997 if (cruft_cb) {
1998 r = cruft_cb(de->d_name, path->buf, data);
1999 if (r)
2000 break;
2001 }
2002 }
094c7e63 2003 closedir(dir);
27e1e22d 2004
163ee5e6 2005 strbuf_setlen(path, baselen - 1);
27e1e22d
JK
2006 if (!r && subdir_cb)
2007 r = subdir_cb(subdir_nr, path->buf, data);
2008
0375f472
RS
2009 strbuf_setlen(path, origlen);
2010
27e1e22d
JK
2011 return r;
2012}
2013
e6f875e0 2014int for_each_loose_file_in_objdir_buf(struct strbuf *path,
27e1e22d
JK
2015 each_loose_object_fn obj_cb,
2016 each_loose_cruft_fn cruft_cb,
2017 each_loose_subdir_fn subdir_cb,
2018 void *data)
2019{
27e1e22d
JK
2020 int r = 0;
2021 int i;
2022
27e1e22d 2023 for (i = 0; i < 256; i++) {
e6f875e0 2024 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
27e1e22d 2025 subdir_cb, data);
27e1e22d
JK
2026 if (r)
2027 break;
2028 }
2029
e6f875e0
JK
2030 return r;
2031}
2032
2033int for_each_loose_file_in_objdir(const char *path,
2034 each_loose_object_fn obj_cb,
2035 each_loose_cruft_fn cruft_cb,
2036 each_loose_subdir_fn subdir_cb,
2037 void *data)
2038{
2039 struct strbuf buf = STRBUF_INIT;
2040 int r;
2041
2042 strbuf_addstr(&buf, path);
2043 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2044 subdir_cb, data);
27e1e22d 2045 strbuf_release(&buf);
e6f875e0 2046
27e1e22d
JK
2047 return r;
2048}
660c889e
JK
2049
2050struct loose_alt_odb_data {
2051 each_loose_object_fn *cb;
2052 void *data;
2053};
2054
2055static int loose_from_alt_odb(struct alternate_object_database *alt,
2056 void *vdata)
2057{
2058 struct loose_alt_odb_data *data = vdata;
b0a42642
JM
2059 struct strbuf buf = STRBUF_INIT;
2060 int r;
2061
597f9134 2062 strbuf_addstr(&buf, alt->path);
b0a42642
JM
2063 r = for_each_loose_file_in_objdir_buf(&buf,
2064 data->cb, NULL, NULL,
2065 data->data);
2066 strbuf_release(&buf);
2067 return r;
660c889e
JK
2068}
2069
1385bb7b 2070int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
660c889e
JK
2071{
2072 struct loose_alt_odb_data alt;
2073 int r;
2074
2075 r = for_each_loose_file_in_objdir(get_object_directory(),
2076 cb, NULL, NULL, data);
2077 if (r)
2078 return r;
2079
1385bb7b
JK
2080 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2081 return 0;
2082
660c889e
JK
2083 alt.cb = cb;
2084 alt.data = data;
2085 return foreach_alt_odb(loose_from_alt_odb, &alt);
2086}
2087
f6371f92
JK
2088static int check_stream_sha1(git_zstream *stream,
2089 const char *hdr,
2090 unsigned long size,
2091 const char *path,
2092 const unsigned char *expected_sha1)
2093{
2094 git_SHA_CTX c;
cd02599c 2095 unsigned char real_sha1[GIT_MAX_RAWSZ];
f6371f92
JK
2096 unsigned char buf[4096];
2097 unsigned long total_read;
2098 int status = Z_OK;
2099
2100 git_SHA1_Init(&c);
2101 git_SHA1_Update(&c, hdr, stream->total_out);
2102
2103 /*
2104 * We already read some bytes into hdr, but the ones up to the NUL
2105 * do not count against the object's content size.
2106 */
2107 total_read = stream->total_out - strlen(hdr) - 1;
2108
2109 /*
2110 * This size comparison must be "<=" to read the final zlib packets;
2111 * see the comment in unpack_sha1_rest for details.
2112 */
2113 while (total_read <= size &&
2114 (status == Z_OK || status == Z_BUF_ERROR)) {
2115 stream->next_out = buf;
2116 stream->avail_out = sizeof(buf);
2117 if (size - total_read < stream->avail_out)
2118 stream->avail_out = size - total_read;
2119 status = git_inflate(stream, Z_FINISH);
2120 git_SHA1_Update(&c, buf, stream->next_out - buf);
2121 total_read += stream->next_out - buf;
2122 }
2123 git_inflate_end(stream);
2124
2125 if (status != Z_STREAM_END) {
2126 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2127 return -1;
2128 }
cce044df
JK
2129 if (stream->avail_in) {
2130 error("garbage at end of loose object '%s'",
2131 sha1_to_hex(expected_sha1));
2132 return -1;
2133 }
f6371f92
JK
2134
2135 git_SHA1_Final(real_sha1, &c);
2136 if (hashcmp(expected_sha1, real_sha1)) {
2137 error("sha1 mismatch for %s (expected %s)", path,
2138 sha1_to_hex(expected_sha1));
2139 return -1;
2140 }
2141
2142 return 0;
2143}
2144
2145int read_loose_object(const char *path,
2146 const unsigned char *expected_sha1,
2147 enum object_type *type,
2148 unsigned long *size,
2149 void **contents)
2150{
2151 int ret = -1;
f6371f92
JK
2152 void *map = NULL;
2153 unsigned long mapsize;
2154 git_zstream stream;
2155 char hdr[32];
2156
2157 *contents = NULL;
2158
2159 map = map_sha1_file_1(path, NULL, &mapsize);
2160 if (!map) {
2161 error_errno("unable to mmap %s", path);
2162 goto out;
2163 }
2164
2165 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2166 error("unable to unpack header of %s", path);
2167 goto out;
2168 }
2169
2170 *type = parse_sha1_header(hdr, size);
2171 if (*type < 0) {
2172 error("unable to parse header of %s", path);
2173 git_inflate_end(&stream);
2174 goto out;
2175 }
2176
2177 if (*type == OBJ_BLOB) {
2178 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
2179 goto out;
2180 } else {
2181 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
2182 if (!*contents) {
2183 error("unable to unpack contents of %s", path);
2184 git_inflate_end(&stream);
2185 goto out;
2186 }
2187 if (check_sha1_signature(expected_sha1, *contents,
2188 *size, typename(*type))) {
2189 error("sha1 mismatch for %s (expected %s)", path,
2190 sha1_to_hex(expected_sha1));
2191 free(*contents);
2192 goto out;
2193 }
2194 }
2195
2196 ret = 0; /* everything checks out */
2197
2198out:
2199 if (map)
2200 munmap(map, mapsize);
f6371f92
JK
2201 return ret;
2202}