sha1_file: fast-path null sha1 as a missing object
[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"
0fcfd160 31
28bd70d8
JN
32#define SZ_FMT PRIuMAX
33static inline uintmax_t sz_fmt(size_t s) { return s; }
e05db0fd 34
96f1e58f 35const unsigned char null_sha1[20];
3e56e724 36const struct object_id null_oid;
8576fde6
JK
37const struct object_id empty_tree_oid = {
38 EMPTY_TREE_SHA1_BIN_LITERAL
39};
40const struct object_id empty_blob_oid = {
41 EMPTY_BLOB_SHA1_BIN_LITERAL
42};
88cd621d 43
c597ba80
NTND
44/*
45 * This is meant to hold a *small* number of objects that you would
46 * want read_sha1_file() to be able to return, but yet you do not want
47 * to write them into the object store (e.g. a browse-only
48 * application).
49 */
50static struct cached_object {
51 unsigned char sha1[20];
52 enum object_type type;
53 void *buf;
54 unsigned long size;
55} *cached_objects;
56static int cached_object_nr, cached_object_alloc;
57
58static struct cached_object empty_tree = {
dab0d410 59 EMPTY_TREE_SHA1_BIN_LITERAL,
c597ba80
NTND
60 OBJ_TREE,
61 "",
62 0
63};
64
65static struct cached_object *find_cached_object(const unsigned char *sha1)
66{
67 int i;
68 struct cached_object *co = cached_objects;
69
70 for (i = 0; i < cached_object_nr; i++, co++) {
71 if (!hashcmp(co->sha1, sha1))
72 return co;
73 }
74 if (!hashcmp(sha1, empty_tree.sha1))
75 return &empty_tree;
76 return NULL;
77}
78
90a6464b
JH
79int mkdir_in_gitdir(const char *path)
80{
81 if (mkdir(path, 0777)) {
82 int saved_errno = errno;
83 struct stat st;
84 struct strbuf sb = STRBUF_INIT;
85
86 if (errno != EEXIST)
87 return -1;
88 /*
89 * Are we looking at a path in a symlinked worktree
90 * whose original repository does not yet have it?
91 * e.g. .git/rr-cache pointing at its original
92 * repository in which the user hasn't performed any
93 * conflict resolution yet?
94 */
95 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
96 strbuf_readlink(&sb, path, st.st_size) ||
97 !is_absolute_path(sb.buf) ||
98 mkdir(sb.buf, 0777)) {
99 strbuf_release(&sb);
100 errno = saved_errno;
101 return -1;
102 }
103 strbuf_release(&sb);
104 }
105 return adjust_shared_perm(path);
106}
107
0be0521b 108enum scld_error safe_create_leading_directories(char *path)
b2cb9425 109{
26c8ae2a 110 char *next_component = path + offset_1st_component(path);
0be0521b 111 enum scld_error ret = SCLD_OK;
67d42212 112
0be0521b 113 while (ret == SCLD_OK && next_component) {
f0502332 114 struct stat st;
0f527403 115 char *slash = next_component, slash_character;
f0502332 116
0f527403
MH
117 while (*slash && !is_dir_sep(*slash))
118 slash++;
119
120 if (!*slash)
b2cb9425 121 break;
bf10cf70 122
26c8ae2a 123 next_component = slash + 1;
0f527403 124 while (is_dir_sep(*next_component))
bf10cf70 125 next_component++;
26c8ae2a 126 if (!*next_component)
5f0bdf50 127 break;
831651fd 128
0f527403 129 slash_character = *slash;
831651fd 130 *slash = '\0';
67d42212
JR
131 if (!stat(path, &st)) {
132 /* path exists */
204a047f
MH
133 if (!S_ISDIR(st.st_mode)) {
134 errno = ENOTDIR;
0be0521b 135 ret = SCLD_EXISTS;
204a047f 136 }
53a39721 137 } else if (mkdir(path, 0777)) {
928734d9 138 if (errno == EEXIST &&
9e6f885d 139 !stat(path, &st) && S_ISDIR(st.st_mode))
928734d9 140 ; /* somebody created it since we checked */
18d37e86
MH
141 else if (errno == ENOENT)
142 /*
143 * Either mkdir() failed because
144 * somebody just pruned the containing
145 * directory, or stat() failed because
146 * the file that was in our way was
147 * just removed. Either way, inform
148 * the caller that it might be worth
149 * trying again:
150 */
151 ret = SCLD_VANISHED;
9e6f885d 152 else
0be0521b 153 ret = SCLD_FAILED;
53a39721 154 } else if (adjust_shared_perm(path)) {
0be0521b 155 ret = SCLD_PERMS;
457f06d6 156 }
0f527403 157 *slash = slash_character;
b2cb9425 158 }
9e6f885d 159 return ret;
b2cb9425 160}
723c31fe 161
0be0521b 162enum scld_error safe_create_leading_directories_const(const char *path)
8e21d63b 163{
02944307 164 int save_errno;
8e21d63b
JK
165 /* path points to cache entries, so xstrdup before messing with it */
166 char *buf = xstrdup(path);
0be0521b 167 enum scld_error result = safe_create_leading_directories(buf);
02944307
MH
168
169 save_errno = errno;
8e21d63b 170 free(buf);
02944307 171 errno = save_errno;
8e21d63b
JK
172 return result;
173}
174
177978f5
MH
175int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
176{
177 /*
178 * The number of times we will try to remove empty directories
179 * in the way of path. This is only 1 because if another
180 * process is racily creating directories that conflict with
181 * us, we don't want to fight against them.
182 */
183 int remove_directories_remaining = 1;
184
185 /*
186 * The number of times that we will try to create the
187 * directories containing path. We are willing to attempt this
188 * more than once, because another process could be trying to
189 * clean up empty directories at the same time as we are
190 * trying to create them.
191 */
192 int create_directories_remaining = 3;
193
194 /* A scratch copy of path, filled lazily if we need it: */
195 struct strbuf path_copy = STRBUF_INIT;
196
197 int ret, save_errno;
198
199 /* Sanity check: */
200 assert(*path);
201
202retry_fn:
203 ret = fn(path, cb);
204 save_errno = errno;
205 if (!ret)
206 goto out;
207
208 if (errno == EISDIR && remove_directories_remaining-- > 0) {
209 /*
210 * A directory is in the way. Maybe it is empty; try
211 * to remove it:
212 */
213 if (!path_copy.len)
214 strbuf_addstr(&path_copy, path);
215
216 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
217 goto retry_fn;
218 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
219 /*
220 * Maybe the containing directory didn't exist, or
221 * maybe it was just deleted by a process that is
222 * racing with us to clean up empty directories. Try
223 * to create it:
224 */
225 enum scld_error scld_result;
226
227 if (!path_copy.len)
228 strbuf_addstr(&path_copy, path);
229
230 do {
231 scld_result = safe_create_leading_directories(path_copy.buf);
232 if (scld_result == SCLD_OK)
233 goto retry_fn;
234 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
235 }
236
237out:
238 strbuf_release(&path_copy);
239 errno = save_errno;
240 return ret;
241}
242
f7b7774f 243static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
ace1534d
JH
244{
245 int i;
246 for (i = 0; i < 20; i++) {
247 static char hex[] = "0123456789abcdef";
248 unsigned int val = sha1[i];
f7b7774f
JK
249 strbuf_addch(buf, hex[val >> 4]);
250 strbuf_addch(buf, hex[val & 0xf]);
afbba2f0 251 if (!i)
f7b7774f 252 strbuf_addch(buf, '/');
ace1534d
JH
253 }
254}
255
30d6c6ea 256const char *sha1_file_name(const unsigned char *sha1)
0fcfd160 257{
f7b7774f 258 static struct strbuf buf = STRBUF_INIT;
0fcfd160 259
f7b7774f
JK
260 strbuf_reset(&buf);
261 strbuf_addf(&buf, "%s/", get_object_directory());
560fb6a1 262
f7b7774f
JK
263 fill_sha1_path(&buf, sha1);
264 return buf.buf;
0fcfd160
LT
265}
266
38dbe5f0
JK
267struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
268{
269 strbuf_setlen(&alt->scratch, alt->base_len);
270 return &alt->scratch;
271}
272
29ec6af2
JK
273static const char *alt_sha1_path(struct alternate_object_database *alt,
274 const unsigned char *sha1)
275{
38dbe5f0 276 struct strbuf *buf = alt_scratch_buf(alt);
f7b7774f 277 fill_sha1_path(buf, sha1);
38dbe5f0 278 return buf->buf;
0fcfd160
LT
279}
280
1cec8c63
JK
281 char *odb_pack_name(struct strbuf *buf,
282 const unsigned char *sha1,
283 const char *ext)
bf592c50 284{
ac5190cc
JK
285 strbuf_reset(buf);
286 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
1cec8c63 287 sha1_to_hex(sha1), ext);
ac5190cc 288 return buf->buf;
bf592c50
DB
289}
290
633f43e1 291char *sha1_pack_name(const unsigned char *sha1)
bf592c50 292{
ac5190cc 293 static struct strbuf buf = STRBUF_INIT;
1cec8c63 294 return odb_pack_name(&buf, sha1, "pack");
633f43e1 295}
bf592c50 296
633f43e1
HO
297char *sha1_pack_index_name(const unsigned char *sha1)
298{
ac5190cc 299 static struct strbuf buf = STRBUF_INIT;
1cec8c63 300 return odb_pack_name(&buf, sha1, "idx");
bf592c50
DB
301}
302
d5a63b99
JH
303struct alternate_object_database *alt_odb_list;
304static struct alternate_object_database **alt_odb_tail;
ace1534d 305
4ea82473
JK
306/*
307 * Return non-zero iff the path is usable as an alternate object database.
308 */
309static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
310{
311 struct alternate_object_database *alt;
312
313 /* Detect cases where alternate disappeared */
314 if (!is_directory(path->buf)) {
315 error("object directory %s does not exist; "
316 "check .git/objects/info/alternates.",
317 path->buf);
318 return 0;
319 }
320
321 /*
322 * Prevent the common mistake of listing the same
323 * thing twice, or object directory itself.
324 */
325 for (alt = alt_odb_list; alt; alt = alt->next) {
ea0fc3b4 326 if (!fspathcmp(path->buf, alt->path))
4ea82473
JK
327 return 0;
328 }
329 if (!fspathcmp(path->buf, normalized_objdir))
330 return 0;
331
332 return 1;
333}
334
ddd5d056
JH
335/*
336 * Prepare alternate object database registry.
d5a63b99
JH
337 *
338 * The variable alt_odb_list points at the list of struct
339 * alternate_object_database. The elements on this list come from
340 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
341 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
342 * whose contents is similar to that environment variable but can be
343 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
344 * contains "/the/directory/corresponding/to/.git/objects/...", while
345 * its name points just after the slash at the end of ".git/objects/"
346 * in the example above, and has enough space to hold 40-byte hex
347 * SHA1, an extra slash for the first level indirection, and the
348 * terminating NUL.
ddd5d056 349 */
2456990d 350static void read_info_alternates(const char * relative_base, int depth);
539e7506
EK
351static int link_alt_odb_entry(const char *entry, const char *relative_base,
352 int depth, const char *normalized_objdir)
ace1534d 353{
c2f493a4 354 struct alternate_object_database *ent;
5bdf0a84 355 struct strbuf pathbuf = STRBUF_INIT;
d5a63b99 356
85dadc38 357 if (!is_absolute_path(entry) && relative_base) {
4ac9006f 358 strbuf_realpath(&pathbuf, relative_base, 1);
5bdf0a84 359 strbuf_addch(&pathbuf, '/');
c2f493a4 360 }
6eac50d8 361 strbuf_addstr(&pathbuf, entry);
c2f493a4 362
37a95862 363 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
670c359d
JK
364 error("unable to normalize alternate object path: %s",
365 pathbuf.buf);
366 strbuf_release(&pathbuf);
367 return -1;
368 }
5bdf0a84
HW
369
370 /*
371 * The trailing slash after the directory name is given by
372 * this function at the end. Remove duplicates.
373 */
4ea82473
JK
374 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
375 strbuf_setlen(&pathbuf, pathbuf.len - 1);
5bdf0a84 376
4ea82473
JK
377 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
378 strbuf_release(&pathbuf);
c2f493a4
MW
379 return -1;
380 }
381
7f0fa2c0 382 ent = alloc_alt_odb(pathbuf.buf);
c2f493a4
MW
383
384 /* add the alternate entry */
385 *alt_odb_tail = ent;
386 alt_odb_tail = &(ent->next);
387 ent->next = NULL;
388
389 /* recursively add alternates */
4ea82473 390 read_info_alternates(pathbuf.buf, depth + 1);
c2f493a4 391
4ea82473 392 strbuf_release(&pathbuf);
c2f493a4
MW
393 return 0;
394}
395
cf3c6352
JK
396static const char *parse_alt_odb_entry(const char *string,
397 int sep,
398 struct strbuf *out)
399{
400 const char *end;
401
402 strbuf_reset(out);
403
404 if (*string == '#') {
405 /* comment; consume up to next separator */
406 end = strchrnul(string, sep);
407 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
408 /*
409 * quoted path; unquote_c_style has copied the
410 * data for us and set "end". Broken quoting (e.g.,
411 * an entry that doesn't end with a quote) falls
412 * back to the unquoted case below.
413 */
414 } else {
415 /* normal, unquoted path */
416 end = strchrnul(string, sep);
417 strbuf_add(out, string, end - string);
418 }
419
420 if (*end)
421 end++;
422 return end;
423}
424
dc732bd5 425static void link_alt_odb_entries(const char *alt, int sep,
c2f493a4
MW
426 const char *relative_base, int depth)
427{
539e7506 428 struct strbuf objdirbuf = STRBUF_INIT;
cf3c6352 429 struct strbuf entry = STRBUF_INIT;
c2f493a4
MW
430
431 if (depth > 5) {
432 error("%s: ignoring alternate object stores, nesting too deep.",
433 relative_base);
434 return;
435 }
436
f655651e 437 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
670c359d
JK
438 if (strbuf_normalize_path(&objdirbuf) < 0)
439 die("unable to normalize object directory: %s",
440 objdirbuf.buf);
539e7506 441
cf3c6352
JK
442 while (*alt) {
443 alt = parse_alt_odb_entry(alt, sep, &entry);
444 if (!entry.len)
9577e7e3 445 continue;
cf3c6352 446 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
9577e7e3 447 }
cf3c6352 448 strbuf_release(&entry);
539e7506 449 strbuf_release(&objdirbuf);
d5a63b99
JH
450}
451
2456990d 452static void read_info_alternates(const char * relative_base, int depth)
d5a63b99 453{
5015f01c 454 char *path;
dc732bd5 455 struct strbuf buf = STRBUF_INIT;
d5a63b99 456
5015f01c 457 path = xstrfmt("%s/info/alternates", relative_base);
dc732bd5 458 if (strbuf_read_file(&buf, path, 1024) < 0) {
f0f7bebe 459 warn_on_fopen_errors(path);
dc732bd5 460 free(path);
9a217f2a 461 return;
ace1534d 462 }
d5a63b99 463
dc732bd5
JK
464 link_alt_odb_entries(buf.buf, '\n', relative_base, depth);
465 strbuf_release(&buf);
466 free(path);
ace1534d
JH
467}
468
7f0fa2c0
JK
469struct alternate_object_database *alloc_alt_odb(const char *dir)
470{
471 struct alternate_object_database *ent;
7f0fa2c0 472
597f9134 473 FLEX_ALLOC_STR(ent, path, dir);
38dbe5f0
JK
474 strbuf_init(&ent->scratch, 0);
475 strbuf_addf(&ent->scratch, "%s/", dir);
476 ent->base_len = ent->scratch.len;
7f0fa2c0
JK
477
478 return ent;
479}
480
bef70b22
DB
481void add_to_alternates_file(const char *reference)
482{
483 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
77b9b1d1
JK
484 char *alts = git_pathdup("objects/info/alternates");
485 FILE *in, *out;
486
487 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
488 out = fdopen_lock_file(lock, "w");
489 if (!out)
490 die_errno("unable to fdopen alternates lockfile");
491
492 in = fopen(alts, "r");
493 if (in) {
494 struct strbuf line = STRBUF_INIT;
495 int found = 0;
496
3f163962 497 while (strbuf_getline(&line, in) != EOF) {
77b9b1d1
JK
498 if (!strcmp(reference, line.buf)) {
499 found = 1;
500 break;
501 }
502 fprintf_or_die(out, "%s\n", line.buf);
503 }
504
505 strbuf_release(&line);
506 fclose(in);
507
508 if (found) {
509 rollback_lock_file(lock);
510 lock = NULL;
511 }
512 }
513 else if (errno != ENOENT)
514 die_errno("unable to read alternates file");
515
516 if (lock) {
517 fprintf_or_die(out, "%s\n", reference);
518 if (commit_lock_file(lock))
519 die_errno("unable to move new alternates file into place");
520 if (alt_odb_tail)
dc732bd5 521 link_alt_odb_entries(reference, '\n', NULL, 0);
77b9b1d1
JK
522 }
523 free(alts);
bef70b22
DB
524}
525
a5b34d21
JK
526void add_to_alternates_memory(const char *reference)
527{
528 /*
529 * Make sure alternates are initialized, or else our entry may be
530 * overwritten when they are.
531 */
532 prepare_alt_odb();
533
dc732bd5 534 link_alt_odb_entries(reference, '\n', NULL, 0);
a5b34d21
JK
535}
536
9eeea7d2
SB
537/*
538 * Compute the exact path an alternate is at and returns it. In case of
539 * error NULL is returned and the human readable error is added to `err`
540 * `path` may be relative and should point to $GITDIR.
541 * `err` must not be null.
542 */
543char *compute_alternate_path(const char *path, struct strbuf *err)
544{
545 char *ref_git = NULL;
546 const char *repo, *ref_git_s;
547 int seen_error = 0;
548
549 ref_git_s = real_path_if_valid(path);
550 if (!ref_git_s) {
551 seen_error = 1;
552 strbuf_addf(err, _("path '%s' does not exist"), path);
553 goto out;
554 } else
555 /*
556 * Beware: read_gitfile(), real_path() and mkpath()
557 * return static buffer
558 */
559 ref_git = xstrdup(ref_git_s);
560
561 repo = read_gitfile(ref_git);
562 if (!repo)
563 repo = read_gitfile(mkpath("%s/.git", ref_git));
564 if (repo) {
565 free(ref_git);
566 ref_git = xstrdup(repo);
567 }
568
569 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
570 char *ref_git_git = mkpathdup("%s/.git", ref_git);
571 free(ref_git);
572 ref_git = ref_git_git;
573 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
574 struct strbuf sb = STRBUF_INIT;
575 seen_error = 1;
576 if (get_common_dir(&sb, ref_git)) {
577 strbuf_addf(err,
578 _("reference repository '%s' as a linked "
579 "checkout is not supported yet."),
580 path);
581 goto out;
582 }
583
584 strbuf_addf(err, _("reference repository '%s' is not a "
585 "local repository."), path);
586 goto out;
587 }
588
589 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
590 strbuf_addf(err, _("reference repository '%s' is shallow"),
591 path);
592 seen_error = 1;
593 goto out;
594 }
595
596 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
597 strbuf_addf(err,
598 _("reference repository '%s' is grafted"),
599 path);
600 seen_error = 1;
601 goto out;
602 }
603
604out:
605 if (seen_error) {
6a83d902 606 FREE_AND_NULL(ref_git);
9eeea7d2
SB
607 }
608
609 return ref_git;
610}
611
fe1b2268 612int foreach_alt_odb(alt_odb_fn fn, void *cb)
d79796bc
JH
613{
614 struct alternate_object_database *ent;
fe1b2268 615 int r = 0;
d79796bc
JH
616
617 prepare_alt_odb();
fe1b2268
JK
618 for (ent = alt_odb_list; ent; ent = ent->next) {
619 r = fn(ent, cb);
620 if (r)
621 break;
622 }
623 return r;
d79796bc
JH
624}
625
c2f493a4
MW
626void prepare_alt_odb(void)
627{
554fe20d 628 const char *alt;
c2f493a4 629
7dc24aa5
SP
630 if (alt_odb_tail)
631 return;
632
c2f493a4
MW
633 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
634 if (!alt) alt = "";
635
c2f493a4 636 alt_odb_tail = &alt_odb_list;
dc732bd5 637 link_alt_odb_entries(alt, PATH_SEP, NULL, 0);
c2f493a4
MW
638
639 read_info_alternates(get_object_directory(), 0);
640}
641
3096b2ec 642/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
33d4221c 643static int freshen_file(const char *fn)
ace1534d 644{
33d4221c
JK
645 struct utimbuf t;
646 t.actime = t.modtime = time(NULL);
647 return !utime(fn, &t);
0f4dc14a 648}
ace1534d 649
3096b2ec
JK
650/*
651 * All of the check_and_freshen functions return 1 if the file exists and was
652 * freshened (if freshening was requested), 0 otherwise. If they return
653 * 0, you should not assume that it is safe to skip a write of the object (it
654 * either does not exist on disk, or has a stale mtime and may be subject to
655 * pruning).
656 */
6a5e6f5e 657int check_and_freshen_file(const char *fn, int freshen)
33d4221c
JK
658{
659 if (access(fn, F_OK))
660 return 0;
3096b2ec 661 if (freshen && !freshen_file(fn))
33d4221c
JK
662 return 0;
663 return 1;
664}
665
666static int check_and_freshen_local(const unsigned char *sha1, int freshen)
667{
668 return check_and_freshen_file(sha1_file_name(sha1), freshen);
669}
670
671static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
0f4dc14a
BC
672{
673 struct alternate_object_database *alt;
9a217f2a 674 prepare_alt_odb();
d5a63b99 675 for (alt = alt_odb_list; alt; alt = alt->next) {
29ec6af2
JK
676 const char *path = alt_sha1_path(alt, sha1);
677 if (check_and_freshen_file(path, freshen))
c529d75a 678 return 1;
ace1534d 679 }
c529d75a 680 return 0;
ace1534d
JH
681}
682
33d4221c
JK
683static int check_and_freshen(const unsigned char *sha1, int freshen)
684{
685 return check_and_freshen_local(sha1, freshen) ||
686 check_and_freshen_nonlocal(sha1, freshen);
687}
688
689int has_loose_object_nonlocal(const unsigned char *sha1)
690{
691 return check_and_freshen_nonlocal(sha1, 0);
692}
693
0f4dc14a
BC
694static int has_loose_object(const unsigned char *sha1)
695{
33d4221c 696 return check_and_freshen(sha1, 0);
0f4dc14a
BC
697}
698
60bb8b14 699static unsigned int pack_used_ctr;
a53128b6
SP
700static unsigned int pack_mmap_calls;
701static unsigned int peak_pack_open_windows;
702static unsigned int pack_open_windows;
c7934306
SP
703static unsigned int pack_open_fds;
704static unsigned int pack_max_fds;
a53128b6 705static size_t peak_pack_mapped;
60bb8b14 706static size_t pack_mapped;
9a217f2a 707struct packed_git *packed_git;
1f688557 708
a73cdd21
JK
709static struct mru packed_git_mru_storage;
710struct mru *packed_git_mru = &packed_git_mru_storage;
711
b79d18c9 712void pack_report(void)
a53128b6
SP
713{
714 fprintf(stderr,
e05db0fd
PR
715 "pack_report: getpagesize() = %10" SZ_FMT "\n"
716 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
717 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
718 sz_fmt(getpagesize()),
719 sz_fmt(packed_git_window_size),
720 sz_fmt(packed_git_limit));
a53128b6
SP
721 fprintf(stderr,
722 "pack_report: pack_used_ctr = %10u\n"
723 "pack_report: pack_mmap_calls = %10u\n"
724 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
725 "pack_report: pack_mapped = "
726 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
727 pack_used_ctr,
728 pack_mmap_calls,
729 pack_open_windows, peak_pack_open_windows,
9e42d6a1 730 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
731}
732
d40d535b
MH
733/*
734 * Open and mmap the index file at path, perform a couple of
735 * consistency checks, then record its information to p. Return 0 on
736 * success.
737 */
738static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
739{
740 void *idx_map;
42873078 741 struct pack_idx_header *hdr;
dc49cd76 742 size_t idx_size;
74e34e1f 743 uint32_t version, nr, i, *index;
a5436b57 744 int fd = git_open(path);
1f688557 745 struct stat st;
42873078 746
1f688557
JH
747 if (fd < 0)
748 return -1;
749 if (fstat(fd, &st)) {
750 close(fd);
751 return -1;
752 }
dc49cd76 753 idx_size = xsize_t(st.st_size);
2d88451b
SP
754 if (idx_size < 4 * 256 + 20 + 20) {
755 close(fd);
756 return error("index file %s is too small", path);
757 }
c4712e45 758 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 759 close(fd);
1f688557 760
42873078
NP
761 hdr = idx_map;
762 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
763 version = ntohl(hdr->idx_version);
764 if (version < 2 || version > 2) {
765 munmap(idx_map, idx_size);
6e1c2344 766 return error("index file %s is version %"PRIu32
74e34e1f
NP
767 " and is not supported by this binary"
768 " (try upgrading GIT to a newer version)",
769 path, version);
770 }
771 } else
772 version = 1;
df1b059d 773
1f688557 774 nr = 0;
42873078 775 index = idx_map;
74e34e1f
NP
776 if (version > 1)
777 index += 2; /* skip index header */
1f688557 778 for (i = 0; i < 256; i++) {
326bf396 779 uint32_t n = ntohl(index[i]);
2d88451b
SP
780 if (n < nr) {
781 munmap(idx_map, idx_size);
df1b059d 782 return error("non-monotonic index %s", path);
2d88451b 783 }
1f688557
JH
784 nr = n;
785 }
786
74e34e1f
NP
787 if (version == 1) {
788 /*
789 * Total size:
790 * - 256 index entries 4 bytes each
791 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
792 * - 20-byte SHA1 of the packfile
793 * - 20-byte SHA1 file checksum
794 */
795 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
796 munmap(idx_map, idx_size);
eef427a0 797 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
798 }
799 } else if (version == 2) {
800 /*
801 * Minimum size:
802 * - 8 bytes of header
803 * - 256 index entries 4 bytes each
804 * - 20-byte sha1 entry * nr
805 * - 4-byte crc entry * nr
806 * - 4-byte offset entry * nr
807 * - 20-byte SHA1 of the packfile
808 * - 20-byte SHA1 file checksum
809 * And after the 4-byte offset table might be a
810 * variable sized table containing 8-byte entries
811 * for offsets larger than 2^31.
812 */
813 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
814 unsigned long max_size = min_size;
815 if (nr)
816 max_size += (nr - 1)*8;
817 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 818 munmap(idx_map, idx_size);
eef427a0 819 return error("wrong index v2 file size in %s", path);
74e34e1f 820 }
7109c889
JH
821 if (idx_size != min_size &&
822 /*
823 * make sure we can deal with large pack offsets.
824 * 31-bit signed offset won't be enough, neither
825 * 32-bit unsigned one will be.
826 */
827 (sizeof(off_t) <= 4)) {
828 munmap(idx_map, idx_size);
829 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 830 }
2d88451b 831 }
1f688557 832
74e34e1f 833 p->index_version = version;
42873078
NP
834 p->index_data = idx_map;
835 p->index_size = idx_size;
57059091 836 p->num_objects = nr;
1f688557
JH
837 return 0;
838}
839
bc8e478a 840int open_pack_index(struct packed_git *p)
d079837e
SP
841{
842 char *idx_name;
9ae97018 843 size_t len;
d079837e
SP
844 int ret;
845
846 if (p->index_data)
847 return 0;
848
9ae97018
JK
849 if (!strip_suffix(p->pack_name, ".pack", &len))
850 die("BUG: pack_name does not end in .pack");
851 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
d079837e
SP
852 ret = check_packed_git_idx(idx_name, p);
853 free(idx_name);
854 return ret;
855}
856
11daf39b
SP
857static void scan_windows(struct packed_git *p,
858 struct packed_git **lru_p,
859 struct pack_window **lru_w,
860 struct pack_window **lru_l)
1f688557 861{
11daf39b
SP
862 struct pack_window *w, *w_l;
863
864 for (w_l = NULL, w = p->windows; w; w = w->next) {
865 if (!w->inuse_cnt) {
866 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
867 *lru_p = p;
868 *lru_w = w;
869 *lru_l = w_l;
54044bf8 870 }
54044bf8 871 }
11daf39b 872 w_l = w;
f9253394 873 }
11daf39b
SP
874}
875
7c3ecb32 876static int unuse_one_window(struct packed_git *current)
11daf39b
SP
877{
878 struct packed_git *p, *lru_p = NULL;
879 struct pack_window *lru_w = NULL, *lru_l = NULL;
880
881 if (current)
882 scan_windows(current, &lru_p, &lru_w, &lru_l);
883 for (p = packed_git; p; p = p->next)
884 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
885 if (lru_p) {
886 munmap(lru_w->base, lru_w->len);
887 pack_mapped -= lru_w->len;
888 if (lru_l)
889 lru_l->next = lru_w->next;
7c3ecb32 890 else
54044bf8 891 lru_p->windows = lru_w->next;
54044bf8 892 free(lru_w);
a53128b6 893 pack_open_windows--;
54044bf8
SP
894 return 1;
895 }
896 return 0;
f9253394
JH
897}
898
7c3ecb32 899void release_pack_memory(size_t need)
97bfeb34
SP
900{
901 size_t cur = pack_mapped;
7c3ecb32 902 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
97bfeb34
SP
903 ; /* nothing */
904}
905
02710228
SP
906static void mmap_limit_check(size_t length)
907{
908 static size_t limit = 0;
909 if (!limit) {
910 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
911 if (!limit)
912 limit = SIZE_MAX;
913 }
914 if (length > limit)
915 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
916 (uintmax_t)length, (uintmax_t)limit);
917}
918
1570856b
JK
919void *xmmap_gently(void *start, size_t length,
920 int prot, int flags, int fd, off_t offset)
58ecbd5e 921{
02710228
SP
922 void *ret;
923
924 mmap_limit_check(length);
925 ret = mmap(start, length, prot, flags, fd, offset);
58ecbd5e
JN
926 if (ret == MAP_FAILED) {
927 if (!length)
928 return NULL;
7c3ecb32 929 release_pack_memory(length);
58ecbd5e 930 ret = mmap(start, length, prot, flags, fd, offset);
58ecbd5e
JN
931 }
932 return ret;
933}
934
1570856b
JK
935void *xmmap(void *start, size_t length,
936 int prot, int flags, int fd, off_t offset)
937{
938 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
939 if (ret == MAP_FAILED)
9ca0aaf6 940 die_errno("mmap failed");
1570856b
JK
941 return ret;
942}
943
c9ced051
SP
944void close_pack_windows(struct packed_git *p)
945{
946 while (p->windows) {
947 struct pack_window *w = p->windows;
948
949 if (w->inuse_cnt)
950 die("pack '%s' still has open windows to it",
951 p->pack_name);
952 munmap(w->base, w->len);
953 pack_mapped -= w->len;
954 pack_open_windows--;
955 p->windows = w->next;
956 free(w);
957 }
958}
959
71fe5d7f
JS
960static int close_pack_fd(struct packed_git *p)
961{
962 if (p->pack_fd < 0)
963 return 0;
964
965 close(p->pack_fd);
966 pack_open_fds--;
967 p->pack_fd = -1;
968
969 return 1;
970}
971
38849a81
JS
972static void close_pack(struct packed_git *p)
973{
974 close_pack_windows(p);
975 close_pack_fd(p);
976 close_pack_index(p);
977}
978
979void close_all_packs(void)
980{
981 struct packed_git *p;
982
983 for (p = packed_git; p; p = p->next)
984 if (p->do_not_close)
ef1177d1 985 die("BUG: want to close pack marked 'do-not-close'");
38849a81
JS
986 else
987 close_pack(p);
988}
989
990
88d0db55
BC
991/*
992 * The LRU pack is the one with the oldest MRU window, preferring packs
993 * with no used windows, or the oldest mtime if it has no windows allocated.
994 */
995static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
996{
997 struct pack_window *w, *this_mru_w;
998 int has_windows_inuse = 0;
999
1000 /*
1001 * Reject this pack if it has windows and the previously selected
1002 * one does not. If this pack does not have windows, reject
1003 * it if the pack file is newer than the previously selected one.
1004 */
1005 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
1006 return;
1007
1008 for (w = this_mru_w = p->windows; w; w = w->next) {
1009 /*
1010 * Reject this pack if any of its windows are in use,
1011 * but the previously selected pack did not have any
1012 * inuse windows. Otherwise, record that this pack
1013 * has windows in use.
1014 */
1015 if (w->inuse_cnt) {
1016 if (*accept_windows_inuse)
1017 has_windows_inuse = 1;
1018 else
1019 return;
1020 }
1021
1022 if (w->last_used > this_mru_w->last_used)
1023 this_mru_w = w;
1024
1025 /*
1026 * Reject this pack if it has windows that have been
1027 * used more recently than the previously selected pack.
1028 * If the previously selected pack had windows inuse and
1029 * we have not encountered a window in this pack that is
1030 * inuse, skip this check since we prefer a pack with no
1031 * inuse windows to one that has inuse windows.
1032 */
1033 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
1034 this_mru_w->last_used > (*mru_w)->last_used)
1035 return;
1036 }
1037
1038 /*
1039 * Select this pack.
1040 */
1041 *mru_w = this_mru_w;
1042 *lru_p = p;
1043 *accept_windows_inuse = has_windows_inuse;
1044}
1045
1046static int close_one_pack(void)
1047{
1048 struct packed_git *p, *lru_p = NULL;
1049 struct pack_window *mru_w = NULL;
1050 int accept_windows_inuse = 1;
1051
1052 for (p = packed_git; p; p = p->next) {
1053 if (p->pack_fd == -1)
1054 continue;
1055 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
1056 }
1057
71fe5d7f
JS
1058 if (lru_p)
1059 return close_pack_fd(lru_p);
88d0db55
BC
1060
1061 return 0;
1062}
1063
03e79c88 1064void unuse_pack(struct pack_window **w_cursor)
f9253394 1065{
03e79c88
SP
1066 struct pack_window *w = *w_cursor;
1067 if (w) {
1068 w->inuse_cnt--;
1069 *w_cursor = NULL;
1070 }
1f688557
JH
1071}
1072
fa5fc15d
SP
1073void close_pack_index(struct packed_git *p)
1074{
1075 if (p->index_data) {
1076 munmap((void *)p->index_data, p->index_size);
1077 p->index_data = NULL;
1078 }
1079}
1080
a0788266
JS
1081static unsigned int get_max_fd_limit(void)
1082{
1083#ifdef RLIMIT_NOFILE
491a8dec
JH
1084 {
1085 struct rlimit lim;
a0788266 1086
491a8dec
JH
1087 if (!getrlimit(RLIMIT_NOFILE, &lim))
1088 return lim.rlim_cur;
1089 }
1090#endif
1091
1092#ifdef _SC_OPEN_MAX
1093 {
1094 long open_max = sysconf(_SC_OPEN_MAX);
1095 if (0 < open_max)
1096 return open_max;
1097 /*
1098 * Otherwise, we got -1 for one of the two
1099 * reasons:
1100 *
1101 * (1) sysconf() did not understand _SC_OPEN_MAX
1102 * and signaled an error with -1; or
1103 * (2) sysconf() said there is no limit.
1104 *
1105 * We _could_ clear errno before calling sysconf() to
1106 * tell these two cases apart and return a huge number
1107 * in the latter case to let the caller cap it to a
1108 * value that is not so selfish, but letting the
1109 * fallback OPEN_MAX codepath take care of these cases
1110 * is a lot simpler.
1111 */
1112 }
1113#endif
a0788266 1114
491a8dec 1115#ifdef OPEN_MAX
a0788266
JS
1116 return OPEN_MAX;
1117#else
1118 return 1; /* see the caller ;-) */
1119#endif
1120}
1121
3cf8b462
SP
1122/*
1123 * Do not call this directly as this leaks p->pack_fd on error return;
1124 * call open_packed_git() instead.
1125 */
1126static int open_packed_git_1(struct packed_git *p)
1f688557 1127{
9bc879c1
SP
1128 struct stat st;
1129 struct pack_header hdr;
1130 unsigned char sha1[20];
1131 unsigned char *idx_sha1;
2c039da8 1132 long fd_flag;
9bc879c1 1133
d079837e
SP
1134 if (!p->index_data && open_pack_index(p))
1135 return error("packfile %s index unavailable", p->pack_name);
1136
c7934306 1137 if (!pack_max_fds) {
a0788266 1138 unsigned int max_fds = get_max_fd_limit();
c7934306
SP
1139
1140 /* Save 3 for stdin/stdout/stderr, 22 for work */
1141 if (25 < max_fds)
1142 pack_max_fds = max_fds - 25;
1143 else
1144 pack_max_fds = 1;
1145 }
1146
88d0db55 1147 while (pack_max_fds <= pack_open_fds && close_one_pack())
c7934306
SP
1148 ; /* nothing */
1149
a5436b57 1150 p->pack_fd = git_open(p->pack_name);
9bc879c1 1151 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 1152 return -1;
c7934306 1153 pack_open_fds++;
9bc879c1
SP
1154
1155 /* If we created the struct before we had the pack we lack size. */
bf592c50 1156 if (!p->pack_size) {
bf592c50 1157 if (!S_ISREG(st.st_mode))
072db278 1158 return error("packfile %s not a regular file", p->pack_name);
bf592c50 1159 p->pack_size = st.st_size;
9bc879c1 1160 } else if (p->pack_size != st.st_size)
072db278 1161 return error("packfile %s size changed", p->pack_name);
9bc879c1 1162
2c039da8
JH
1163 /* We leave these file descriptors open with sliding mmap;
1164 * there is no point keeping them open across exec(), though.
1165 */
1166 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
1167 if (fd_flag < 0)
072db278 1168 return error("cannot determine file descriptor flags");
2c039da8
JH
1169 fd_flag |= FD_CLOEXEC;
1170 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 1171 return error("cannot set FD_CLOEXEC");
2c039da8 1172
9bc879c1 1173 /* Verify we recognize this pack file format. */
e6e2bd62 1174 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 1175 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 1176 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 1177 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 1178 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
1179 return error("packfile %s is version %"PRIu32" and not"
1180 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
1181 p->pack_name, ntohl(hdr.hdr_version));
1182
1183 /* Verify the pack matches its index. */
57059091 1184 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
1185 return error("packfile %s claims to have %"PRIu32" objects"
1186 " while index indicates %"PRIu32" objects",
57059091
NP
1187 p->pack_name, ntohl(hdr.hdr_entries),
1188 p->num_objects);
9bc879c1 1189 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 1190 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 1191 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 1192 return error("packfile %s signature is unavailable", p->pack_name);
42873078 1193 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 1194 if (hashcmp(sha1, idx_sha1))
072db278
SP
1195 return error("packfile %s does not match index", p->pack_name);
1196 return 0;
9bc879c1
SP
1197}
1198
3cf8b462
SP
1199static int open_packed_git(struct packed_git *p)
1200{
1201 if (!open_packed_git_1(p))
1202 return 0;
71fe5d7f 1203 close_pack_fd(p);
3cf8b462
SP
1204 return -1;
1205}
1206
c4001d92 1207static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
1208{
1209 /* We must promise at least 20 bytes (one hash) after the
1210 * offset is available from this window, otherwise the offset
1211 * is not actually in this window and a different window (which
1212 * has that one hash excess) must be used. This is to support
1213 * the object header and delta base parsing routines below.
1214 */
1215 off_t win_off = win->offset;
1216 return win_off <= offset
1217 && (offset + 20) <= (win_off + win->len);
1218}
1219
4b25d091 1220unsigned char *use_pack(struct packed_git *p,
03e79c88 1221 struct pack_window **w_cursor,
c4001d92 1222 off_t offset,
ef49a7a0 1223 unsigned long *left)
9bc879c1 1224{
60bb8b14 1225 struct pack_window *win = *w_cursor;
03e79c88 1226
a9d98a14 1227 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
1228 * pointless to ask for an offset into the middle of that
1229 * hash, and the in_window function above wouldn't match
1230 * don't allow an offset too close to the end of the file.
1231 */
d131b7af
SP
1232 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1233 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
1234 if (offset > (p->pack_size - 20))
1235 die("offset beyond end of packfile (truncated pack?)");
13e0b0d3 1236 if (offset < 0)
7465feba 1237 die(_("offset before end of packfile (broken .idx?)"));
60bb8b14
SP
1238
1239 if (!win || !in_window(win, offset)) {
1240 if (win)
1241 win->inuse_cnt--;
1242 for (win = p->windows; win; win = win->next) {
1243 if (in_window(win, offset))
1244 break;
1245 }
1246 if (!win) {
78a28df9 1247 size_t window_align = packed_git_window_size / 2;
dc49cd76 1248 off_t len;
d131b7af
SP
1249
1250 if (p->pack_fd == -1 && open_packed_git(p))
1251 die("packfile %s cannot be accessed", p->pack_name);
1252
60bb8b14 1253 win = xcalloc(1, sizeof(*win));
78a28df9 1254 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
1255 len = p->pack_size - win->offset;
1256 if (len > packed_git_window_size)
1257 len = packed_git_window_size;
1258 win->len = (size_t)len;
60bb8b14 1259 pack_mapped += win->len;
11daf39b 1260 while (packed_git_limit < pack_mapped
7c3ecb32 1261 && unuse_one_window(p))
60bb8b14 1262 ; /* nothing */
c4712e45 1263 win->base = xmmap(NULL, win->len,
60bb8b14
SP
1264 PROT_READ, MAP_PRIVATE,
1265 p->pack_fd, win->offset);
1266 if (win->base == MAP_FAILED)
7616c6ca
NTND
1267 die_errno("packfile %s cannot be mapped",
1268 p->pack_name);
d131b7af 1269 if (!win->offset && win->len == p->pack_size
71fe5d7f
JS
1270 && !p->do_not_close)
1271 close_pack_fd(p);
a53128b6
SP
1272 pack_mmap_calls++;
1273 pack_open_windows++;
1274 if (pack_mapped > peak_pack_mapped)
1275 peak_pack_mapped = pack_mapped;
1276 if (pack_open_windows > peak_pack_open_windows)
1277 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
1278 win->next = p->windows;
1279 p->windows = win;
1280 }
1f688557 1281 }
03e79c88
SP
1282 if (win != *w_cursor) {
1283 win->last_used = pack_used_ctr++;
1284 win->inuse_cnt++;
1285 *w_cursor = win;
1286 }
60bb8b14 1287 offset -= win->offset;
03e79c88 1288 if (left)
dc49cd76 1289 *left = win->len - xsize_t(offset);
03e79c88 1290 return win->base + offset;
1f688557
JH
1291}
1292
27d69a46
NP
1293static struct packed_git *alloc_packed_git(int extra)
1294{
50a6c8ef 1295 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
27d69a46
NP
1296 memset(p, 0, sizeof(*p));
1297 p->pack_fd = -1;
1298 return p;
1299}
1300
e0500293
JN
1301static void try_to_free_pack_memory(size_t size)
1302{
7c3ecb32 1303 release_pack_memory(size);
e0500293
JN
1304}
1305
48bcc1c3 1306struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
1f688557 1307{
e0500293 1308 static int have_set_try_to_free_routine;
1f688557 1309 struct stat st;
48bcc1c3
JK
1310 size_t alloc;
1311 struct packed_git *p;
1f688557 1312
e0500293
JN
1313 if (!have_set_try_to_free_routine) {
1314 have_set_try_to_free_routine = 1;
1315 set_try_to_free_routine(try_to_free_pack_memory);
1316 }
1317
42873078
NP
1318 /*
1319 * Make sure a corresponding .pack file exists and that
1320 * the index looks sane.
1321 */
48bcc1c3 1322 if (!strip_suffix_mem(path, &path_len, ".idx"))
1f688557 1323 return NULL;
48bcc1c3
JK
1324
1325 /*
1326 * ".pack" is long enough to hold any suffix we're adding (and
1327 * the use xsnprintf double-checks that)
1328 */
50a6c8ef 1329 alloc = st_add3(path_len, strlen(".pack"), 1);
48bcc1c3 1330 p = alloc_packed_git(alloc);
42873078 1331 memcpy(p->pack_name, path, path_len);
8d25931d 1332
48bcc1c3 1333 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
8d25931d
BC
1334 if (!access(p->pack_name, F_OK))
1335 p->pack_keep = 1;
1336
48bcc1c3 1337 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
d079837e 1338 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 1339 free(p);
1f688557
JH
1340 return NULL;
1341 }
42873078 1342
1f688557
JH
1343 /* ok, it looks sane as far as we can check without
1344 * actually mapping the pack file.
1345 */
1f688557 1346 p->pack_size = st.st_size;
9d835df2 1347 p->pack_local = local;
b867092f 1348 p->mtime = st.st_mtime;
42873078
NP
1349 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1350 hashclr(p->sha1);
1f688557
JH
1351 return p;
1352}
1353
7b64469a 1354struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 1355{
42873078 1356 const char *path = sha1_pack_name(sha1);
50a6c8ef 1357 size_t alloc = st_add(strlen(path), 1);
c7ab0ba3 1358 struct packed_git *p = alloc_packed_git(alloc);
bf592c50 1359
c7ab0ba3 1360 memcpy(p->pack_name, path, alloc); /* includes NUL */
27d69a46 1361 hashcpy(p->sha1, sha1);
42873078
NP
1362 if (check_packed_git_idx(idx_path, p)) {
1363 free(p);
bf592c50 1364 return NULL;
42873078 1365 }
bf592c50 1366
bf592c50
DB
1367 return p;
1368}
1369
1370void install_packed_git(struct packed_git *pack)
1371{
c7934306
SP
1372 if (pack->pack_fd != -1)
1373 pack_open_fds++;
1374
bf592c50
DB
1375 pack->next = packed_git;
1376 packed_git = pack;
1377}
1378
0a489b06 1379void (*report_garbage)(unsigned seen_bits, const char *path);
543c5caa
NTND
1380
1381static void report_helper(const struct string_list *list,
1382 int seen_bits, int first, int last)
1383{
0a489b06 1384 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
543c5caa 1385 return;
0a489b06 1386
543c5caa 1387 for (; first < last; first++)
0a489b06 1388 report_garbage(seen_bits, list->items[first].string);
543c5caa
NTND
1389}
1390
1391static void report_pack_garbage(struct string_list *list)
1392{
1393 int i, baselen = -1, first = 0, seen_bits = 0;
1394
1395 if (!report_garbage)
1396 return;
1397
3383e199 1398 string_list_sort(list);
543c5caa
NTND
1399
1400 for (i = 0; i < list->nr; i++) {
1401 const char *path = list->items[i].string;
1402 if (baselen != -1 &&
1403 strncmp(path, list->items[first].string, baselen)) {
1404 report_helper(list, seen_bits, first, i);
1405 baselen = -1;
1406 seen_bits = 0;
1407 }
1408 if (baselen == -1) {
1409 const char *dot = strrchr(path, '.');
1410 if (!dot) {
0a489b06 1411 report_garbage(PACKDIR_FILE_GARBAGE, path);
543c5caa
NTND
1412 continue;
1413 }
1414 baselen = dot - path + 1;
1415 first = i;
1416 }
1417 if (!strcmp(path + baselen, "pack"))
1418 seen_bits |= 1;
1419 else if (!strcmp(path + baselen, "idx"))
1420 seen_bits |= 2;
1421 }
1422 report_helper(list, seen_bits, first, list->nr);
1423}
1424
9d835df2 1425static void prepare_packed_git_one(char *objdir, int local)
1f688557 1426{
880fb8de
RS
1427 struct strbuf path = STRBUF_INIT;
1428 size_t dirnamelen;
1f688557
JH
1429 DIR *dir;
1430 struct dirent *de;
543c5caa 1431 struct string_list garbage = STRING_LIST_INIT_DUP;
1f688557 1432
880fb8de
RS
1433 strbuf_addstr(&path, objdir);
1434 strbuf_addstr(&path, "/pack");
1435 dir = opendir(path.buf);
b5b16990 1436 if (!dir) {
26125f6b 1437 if (errno != ENOENT)
7616c6ca
NTND
1438 error_errno("unable to open object pack directory: %s",
1439 path.buf);
880fb8de 1440 strbuf_release(&path);
1f688557 1441 return;
b5b16990 1442 }
880fb8de
RS
1443 strbuf_addch(&path, '/');
1444 dirnamelen = path.len;
1f688557 1445 while ((de = readdir(dir)) != NULL) {
1f688557 1446 struct packed_git *p;
47bf4b0f 1447 size_t base_len;
543c5caa
NTND
1448
1449 if (is_dot_or_dotdot(de->d_name))
9cb18f56
JM
1450 continue;
1451
880fb8de
RS
1452 strbuf_setlen(&path, dirnamelen);
1453 strbuf_addstr(&path, de->d_name);
d90906a9 1454
47bf4b0f
JK
1455 base_len = path.len;
1456 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
d90906a9
NTND
1457 /* Don't reopen a pack we already have. */
1458 for (p = packed_git; p; p = p->next) {
47bf4b0f
JK
1459 size_t len;
1460 if (strip_suffix(p->pack_name, ".pack", &len) &&
1461 len == base_len &&
1462 !memcmp(p->pack_name, path.buf, len))
d90906a9
NTND
1463 break;
1464 }
1465 if (p == NULL &&
1466 /*
1467 * See if it really is a valid .idx file with
1468 * corresponding .pack file that we can map.
1469 */
880fb8de 1470 (p = add_packed_git(path.buf, path.len, local)) != NULL)
d90906a9 1471 install_packed_git(p);
86f7780c 1472 }
543c5caa
NTND
1473
1474 if (!report_garbage)
1475 continue;
1476
2975c770
JK
1477 if (ends_with(de->d_name, ".idx") ||
1478 ends_with(de->d_name, ".pack") ||
1479 ends_with(de->d_name, ".bitmap") ||
1480 ends_with(de->d_name, ".keep"))
880fb8de 1481 string_list_append(&garbage, path.buf);
543c5caa 1482 else
0a489b06 1483 report_garbage(PACKDIR_FILE_GARBAGE, path.buf);
1f688557 1484 }
5b35bcd5 1485 closedir(dir);
543c5caa
NTND
1486 report_pack_garbage(&garbage);
1487 string_list_clear(&garbage, 0);
880fb8de 1488 strbuf_release(&path);
1f688557
JH
1489}
1490
8e3f52d7
JK
1491static int approximate_object_count_valid;
1492
1493/*
1494 * Give a fast, rough count of the number of objects in the repository. This
1495 * ignores loose objects completely. If you have a lot of them, then either
1496 * you should repack because your performance will be awful, or they are
1497 * all unreachable objects about to be pruned, in which case they're not really
1498 * interesting as a measure of repo size in the first place.
1499 */
1500unsigned long approximate_object_count(void)
1501{
1502 static unsigned long count;
1503 if (!approximate_object_count_valid) {
1504 struct packed_git *p;
1505
1506 prepare_packed_git();
1507 count = 0;
1508 for (p = packed_git; p; p = p->next) {
1509 if (open_pack_index(p))
1510 continue;
1511 count += p->num_objects;
1512 }
1513 }
1514 return count;
1515}
1516
c4c6effa
RS
1517static void *get_next_packed_git(const void *p)
1518{
1519 return ((const struct packed_git *)p)->next;
1520}
1521
1522static void set_next_packed_git(void *p, void *next)
1523{
1524 ((struct packed_git *)p)->next = next;
1525}
1526
b867092f
JH
1527static int sort_pack(const void *a_, const void *b_)
1528{
c4c6effa
RS
1529 const struct packed_git *a = a_;
1530 const struct packed_git *b = b_;
b867092f
JH
1531 int st;
1532
1533 /*
1534 * Local packs tend to contain objects specific to our
1535 * variant of the project than remote ones. In addition,
1536 * remote ones could be on a network mounted filesystem.
1537 * Favor local ones for these reasons.
1538 */
1539 st = a->pack_local - b->pack_local;
1540 if (st)
1541 return -st;
1542
1543 /*
1544 * Younger packs tend to contain more recent objects,
1545 * and more recent objects tend to get accessed more
1546 * often.
1547 */
1548 if (a->mtime < b->mtime)
1549 return 1;
1550 else if (a->mtime == b->mtime)
1551 return 0;
1552 return -1;
1553}
1554
1555static void rearrange_packed_git(void)
1556{
c4c6effa
RS
1557 packed_git = llist_mergesort(packed_git, get_next_packed_git,
1558 set_next_packed_git, sort_pack);
b867092f
JH
1559}
1560
a73cdd21
JK
1561static void prepare_packed_git_mru(void)
1562{
1563 struct packed_git *p;
1564
1565 mru_clear(packed_git_mru);
1566 for (p = packed_git; p; p = p->next)
1567 mru_append(packed_git_mru, p);
1568}
1569
637cdd9d 1570static int prepare_packed_git_run_once = 0;
9a217f2a 1571void prepare_packed_git(void)
1f688557 1572{
d5a63b99 1573 struct alternate_object_database *alt;
1f688557 1574
637cdd9d 1575 if (prepare_packed_git_run_once)
1f688557 1576 return;
9d835df2 1577 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 1578 prepare_alt_odb();
597f9134
JK
1579 for (alt = alt_odb_list; alt; alt = alt->next)
1580 prepare_packed_git_one(alt->path, 0);
b867092f 1581 rearrange_packed_git();
a73cdd21 1582 prepare_packed_git_mru();
637cdd9d
JK
1583 prepare_packed_git_run_once = 1;
1584}
1585
fc04c412 1586void reprepare_packed_git(void)
637cdd9d 1587{
8e3f52d7 1588 approximate_object_count_valid = 0;
637cdd9d
JK
1589 prepare_packed_git_run_once = 0;
1590 prepare_packed_git();
1f688557
JH
1591}
1592
8eca0b47
NP
1593static void mark_bad_packed_object(struct packed_git *p,
1594 const unsigned char *sha1)
1595{
1596 unsigned i;
1597 for (i = 0; i < p->num_bad_objects; i++)
50a6c8ef 1598 if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
8eca0b47 1599 return;
50a6c8ef 1600 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
cd02599c 1601 st_mult(GIT_MAX_RAWSZ,
50a6c8ef
JK
1602 st_add(p->num_bad_objects, 1)));
1603 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
8eca0b47
NP
1604 p->num_bad_objects++;
1605}
1606
b6c4cecc 1607static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
ac939109
NP
1608{
1609 struct packed_git *p;
1610 unsigned i;
1611
1612 for (p = packed_git; p; p = p->next)
1613 for (i = 0; i < p->num_bad_objects; i++)
1614 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
b6c4cecc
JH
1615 return p;
1616 return NULL;
ac939109
NP
1617}
1618
090ea126
NTND
1619/*
1620 * With an in-core object data in "map", rehash it to make sure the
1621 * object name actually matches "sha1" to detect object corruption.
1622 * With "map" == NULL, try reading the object named with "sha1" using
1623 * the streaming interface and rehash it to do the same.
1624 */
1625int check_sha1_signature(const unsigned char *sha1, void *map,
1626 unsigned long size, const char *type)
0fcfd160
LT
1627{
1628 unsigned char real_sha1[20];
090ea126
NTND
1629 enum object_type obj_type;
1630 struct git_istream *st;
1631 git_SHA_CTX c;
1632 char hdr[32];
1633 int hdrlen;
1634
1635 if (map) {
1636 hash_sha1_file(map, size, type, real_sha1);
1637 return hashcmp(sha1, real_sha1) ? -1 : 0;
1638 }
1639
1640 st = open_istream(sha1, &obj_type, &size, NULL);
1641 if (!st)
1642 return -1;
1643
1644 /* Generate the header */
ef1286d3 1645 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
090ea126
NTND
1646
1647 /* Sha1.. */
1648 git_SHA1_Init(&c);
1649 git_SHA1_Update(&c, hdr, hdrlen);
1650 for (;;) {
1651 char buf[1024 * 16];
1652 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1653
f54fac53
JK
1654 if (readlen < 0) {
1655 close_istream(st);
1656 return -1;
1657 }
090ea126
NTND
1658 if (!readlen)
1659 break;
1660 git_SHA1_Update(&c, buf, readlen);
1661 }
1662 git_SHA1_Final(real_sha1, &c);
1663 close_istream(st);
a89fccd2 1664 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1665}
1666
1b8ac5ea 1667int git_open_cloexec(const char *name, int flags)
44d1c19e 1668{
1e3001a8
JH
1669 int fd;
1670 static int o_cloexec = O_CLOEXEC;
f2e872aa 1671
1e3001a8
JH
1672 fd = open(name, flags | o_cloexec);
1673 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
cd66ada0 1674 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1e3001a8
JH
1675 o_cloexec &= ~O_CLOEXEC;
1676 fd = open(name, flags | o_cloexec);
1b8ac5ea 1677 }
f2e872aa 1678
9fb9495d 1679#if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1e3001a8
JH
1680 {
1681 static int fd_cloexec = FD_CLOEXEC;
1682
1683 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1684 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
9fb9495d
EW
1685 int flags = fcntl(fd, F_GETFD);
1686 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1e3001a8 1687 fd_cloexec = 0;
cd66ada0 1688 }
44d1c19e 1689 }
1e3001a8 1690#endif
1b8ac5ea 1691 return fd;
44d1c19e
LT
1692}
1693
771e7d57
JK
1694/*
1695 * Find "sha1" as a loose object in the local repository or in an alternate.
1696 * Returns 0 on success, negative on failure.
1697 *
1698 * The "path" out-parameter will give the path of the object we found (if any).
1699 * Note that it may point to static storage and is only valid until another
1700 * call to sha1_file_name(), etc.
1701 */
1702static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
1703 const char **path)
052fe5ea 1704{
052fe5ea
JK
1705 struct alternate_object_database *alt;
1706
771e7d57
JK
1707 *path = sha1_file_name(sha1);
1708 if (!lstat(*path, st))
052fe5ea
JK
1709 return 0;
1710
1711 prepare_alt_odb();
1712 errno = ENOENT;
1713 for (alt = alt_odb_list; alt; alt = alt->next) {
771e7d57
JK
1714 *path = alt_sha1_path(alt, sha1);
1715 if (!lstat(*path, st))
052fe5ea
JK
1716 return 0;
1717 }
1718
1719 return -1;
1720}
1721
771e7d57
JK
1722/*
1723 * Like stat_sha1_file(), but actually open the object and return the
1724 * descriptor. See the caveats on the "path" parameter above.
1725 */
1726static int open_sha1_file(const unsigned char *sha1, const char **path)
44d1c19e
LT
1727{
1728 int fd;
44d1c19e 1729 struct alternate_object_database *alt;
d6c8a05b 1730 int most_interesting_errno;
44d1c19e 1731
771e7d57
JK
1732 *path = sha1_file_name(sha1);
1733 fd = git_open(*path);
44d1c19e
LT
1734 if (fd >= 0)
1735 return fd;
d6c8a05b 1736 most_interesting_errno = errno;
44d1c19e
LT
1737
1738 prepare_alt_odb();
44d1c19e 1739 for (alt = alt_odb_list; alt; alt = alt->next) {
771e7d57
JK
1740 *path = alt_sha1_path(alt, sha1);
1741 fd = git_open(*path);
44d1c19e
LT
1742 if (fd >= 0)
1743 return fd;
d6c8a05b
JK
1744 if (most_interesting_errno == ENOENT)
1745 most_interesting_errno = errno;
44d1c19e 1746 }
d6c8a05b 1747 errno = most_interesting_errno;
44d1c19e
LT
1748 return -1;
1749}
1750
f6371f92
JK
1751/*
1752 * Map the loose object at "path" if it is not NULL, or the path found by
1753 * searching for a loose object named "sha1".
1754 */
1755static void *map_sha1_file_1(const char *path,
1756 const unsigned char *sha1,
1757 unsigned long *size)
0fcfd160 1758{
0fcfd160 1759 void *map;
144bde78 1760 int fd;
ace1534d 1761
f6371f92
JK
1762 if (path)
1763 fd = git_open(path);
1764 else
1765 fd = open_sha1_file(sha1, &path);
44d1c19e
LT
1766 map = NULL;
1767 if (fd >= 0) {
1768 struct stat st;
0fcfd160 1769
44d1c19e
LT
1770 if (!fstat(fd, &st)) {
1771 *size = xsize_t(st.st_size);
33e42de0
MM
1772 if (!*size) {
1773 /* mmap() is forbidden on empty files */
771e7d57 1774 error("object file %s is empty", path);
33e42de0
MM
1775 return NULL;
1776 }
44d1c19e 1777 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1778 }
44d1c19e 1779 close(fd);
0fcfd160 1780 }
0fcfd160
LT
1781 return map;
1782}
1783
f6371f92
JK
1784void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1785{
1786 return map_sha1_file_1(NULL, sha1, size);
1787}
1788
09ded04b
NP
1789unsigned long unpack_object_header_buffer(const unsigned char *buf,
1790 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1791{
ad1ed5ee 1792 unsigned shift;
48fb7deb 1793 unsigned long size, c;
ad1ed5ee
JH
1794 unsigned long used = 0;
1795
1796 c = buf[used++];
1797 *type = (c >> 4) & 7;
1798 size = c & 15;
1799 shift = 4;
1800 while (c & 0x80) {
f630cfda 1801 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1802 error("bad object header");
ea4f9685
JH
1803 size = used = 0;
1804 break;
09ded04b 1805 }
ad1ed5ee
JH
1806 c = buf[used++];
1807 size += (c & 0x7f) << shift;
1808 shift += 7;
1809 }
1810 *sizep = size;
1811 return used;
1812}
1813
d21f8426
JH
1814static int unpack_sha1_short_header(git_zstream *stream,
1815 unsigned char *map, unsigned long mapsize,
1816 void *buffer, unsigned long bufsiz)
ad1ed5ee 1817{
c4483576
LT
1818 /* Get the data stream */
1819 memset(stream, 0, sizeof(*stream));
1820 stream->next_in = map;
1821 stream->avail_in = mapsize;
1822 stream->next_out = buffer;
93821bd9
LT
1823 stream->avail_out = bufsiz;
1824
39c68542 1825 git_inflate_init(stream);
cc5c54e7 1826 return git_inflate(stream, 0);
c4483576
LT
1827}
1828
d21f8426
JH
1829int unpack_sha1_header(git_zstream *stream,
1830 unsigned char *map, unsigned long mapsize,
1831 void *buffer, unsigned long bufsiz)
1832{
1833 int status = unpack_sha1_short_header(stream, map, mapsize,
1834 buffer, bufsiz);
1835
1836 if (status < Z_OK)
1837 return status;
1838
1839 /* Make sure we have the terminating NUL */
1840 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1841 return -1;
1842 return 0;
1843}
1844
46f03448
KN
1845static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1846 unsigned long mapsize, void *buffer,
1847 unsigned long bufsiz, struct strbuf *header)
1848{
1849 int status;
1850
d21f8426
JH
1851 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1852 if (status < Z_OK)
1853 return -1;
46f03448
KN
1854
1855 /*
1856 * Check if entire header is unpacked in the first iteration.
1857 */
1858 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1859 return 0;
1860
1861 /*
1862 * buffer[0..bufsiz] was not large enough. Copy the partial
1863 * result out to header, and then append the result of further
1864 * reading the stream.
1865 */
1866 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1867 stream->next_out = buffer;
1868 stream->avail_out = bufsiz;
1869
1870 do {
1871 status = git_inflate(stream, 0);
1872 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1873 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1874 return 0;
1875 stream->next_out = buffer;
1876 stream->avail_out = bufsiz;
1877 } while (status != Z_STREAM_END);
1878 return -1;
1879}
1880
ef49a7a0 1881static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1882{
1883 int bytes = strlen(buffer) + 1;
3aee68aa 1884 unsigned char *buf = xmallocz(size);
93821bd9 1885 unsigned long n;
7efbff75 1886 int status = Z_OK;
5180cacc 1887
93821bd9
LT
1888 n = stream->total_out - bytes;
1889 if (n > size)
1890 n = size;
1891 memcpy(buf, (char *) buffer + bytes, n);
1892 bytes = n;
456cdf6e
LT
1893 if (bytes <= size) {
1894 /*
1895 * The above condition must be (bytes <= size), not
1896 * (bytes < size). In other words, even though we
ccf5ace0 1897 * expect no more output and set avail_out to zero,
456cdf6e
LT
1898 * the input zlib stream may have bytes that express
1899 * "this concludes the stream", and we *do* want to
1900 * eat that input.
1901 *
1902 * Otherwise we would not be able to test that we
1903 * consumed all the input to reach the expected size;
1904 * we also want to check that zlib tells us that all
1905 * went well with status == Z_STREAM_END at the end.
1906 */
5180cacc
LT
1907 stream->next_out = buf + bytes;
1908 stream->avail_out = size - bytes;
7efbff75 1909 while (status == Z_OK)
39c68542 1910 status = git_inflate(stream, Z_FINISH);
5180cacc 1911 }
456cdf6e 1912 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1913 git_inflate_end(stream);
7efbff75
JH
1914 return buf;
1915 }
1916
1917 if (status < 0)
1918 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1919 else if (stream->avail_in)
1920 error("garbage at end of loose object '%s'",
1921 sha1_to_hex(sha1));
1922 free(buf);
1923 return NULL;
5180cacc
LT
1924}
1925
1926/*
1927 * We used to just use "sscanf()", but that's actually way
1928 * too permissive for what we want to check. So do an anal
1929 * object header parse by hand.
1930 */
46f03448
KN
1931static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1932 unsigned int flags)
5180cacc 1933{
46f03448 1934 const char *type_buf = hdr;
5180cacc 1935 unsigned long size;
46f03448 1936 int type, type_len = 0;
5180cacc
LT
1937
1938 /*
46f03448 1939 * The type can be of any size but is followed by
21666f1a 1940 * a space.
5180cacc 1941 */
5180cacc
LT
1942 for (;;) {
1943 char c = *hdr++;
d21f8426
JH
1944 if (!c)
1945 return -1;
5180cacc
LT
1946 if (c == ' ')
1947 break;
46f03448 1948 type_len++;
5180cacc 1949 }
46f03448
KN
1950
1951 type = type_from_string_gently(type_buf, type_len, 1);
1952 if (oi->typename)
1953 strbuf_add(oi->typename, type_buf, type_len);
1954 /*
1955 * Set type to 0 if its an unknown object and
2e3a16b2 1956 * we're obtaining the type using '--allow-unknown-type'
46f03448
KN
1957 * option.
1958 */
19fc5e84 1959 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
46f03448
KN
1960 type = 0;
1961 else if (type < 0)
1962 die("invalid object type");
1963 if (oi->typep)
1964 *oi->typep = type;
5180cacc
LT
1965
1966 /*
1967 * The length must follow immediately, and be in canonical
1968 * decimal format (ie "010" is not valid).
1969 */
1970 size = *hdr++ - '0';
1971 if (size > 9)
1972 return -1;
1973 if (size) {
1974 for (;;) {
1975 unsigned long c = *hdr - '0';
1976 if (c > 9)
1977 break;
1978 hdr++;
1979 size = size * 10 + c;
1980 }
1981 }
46f03448
KN
1982
1983 if (oi->sizep)
1984 *oi->sizep = size;
5180cacc
LT
1985
1986 /*
1987 * The length must be followed by a zero byte
1988 */
46f03448
KN
1989 return *hdr ? -1 : type;
1990}
1991
1992int parse_sha1_header(const char *hdr, unsigned long *sizep)
1993{
27b5c1a0 1994 struct object_info oi = OBJECT_INFO_INIT;
46f03448
KN
1995
1996 oi.sizep = sizep;
1f0c0d36 1997 return parse_sha1_header_extended(hdr, &oi, 0);
0fcfd160
LT
1998}
1999
54dab52a
NP
2000unsigned long get_size_from_delta(struct packed_git *p,
2001 struct pack_window **w_curs,
2002 off_t curpos)
2003{
2004 const unsigned char *data;
2005 unsigned char delta_head[20], *in;
ef49a7a0 2006 git_zstream stream;
54dab52a
NP
2007 int st;
2008
2009 memset(&stream, 0, sizeof(stream));
2010 stream.next_out = delta_head;
2011 stream.avail_out = sizeof(delta_head);
2012
39c68542 2013 git_inflate_init(&stream);
54dab52a
NP
2014 do {
2015 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2016 stream.next_in = in;
39c68542 2017 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
2018 curpos += stream.next_in - in;
2019 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
2020 stream.total_out < sizeof(delta_head));
39c68542 2021 git_inflate_end(&stream);
3d77d877
NP
2022 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
2023 error("delta data unpack-initial failed");
2024 return 0;
2025 }
54dab52a
NP
2026
2027 /* Examine the initial part of the delta to figure out
2028 * the result size.
2029 */
2030 data = delta_head;
2031
2032 /* ignore base size */
2033 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
2034
2035 /* Read the result size */
2036 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
2037}
2038
c4001d92 2039static off_t get_delta_base(struct packed_git *p,
03e79c88 2040 struct pack_window **w_curs,
c4001d92 2041 off_t *curpos,
21666f1a 2042 enum object_type type,
c4001d92 2043 off_t delta_obj_offset)
eb32d236 2044{
2b87c45b 2045 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 2046 off_t base_offset;
eb32d236 2047
8d8a4ea5
SP
2048 /* use_pack() assured us we have [base_info, base_info + 20)
2049 * as a range that we can look at without walking off the
2050 * end of the mapped window. Its actually the hash size
2051 * that is assured. An OFS_DELTA longer than the hash size
2052 * is stupid, as then a REF_DELTA would be smaller to store.
2053 */
21666f1a 2054 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
2055 unsigned used = 0;
2056 unsigned char c = base_info[used++];
2057 base_offset = c & 127;
2058 while (c & 128) {
2059 base_offset += 1;
8723f216 2060 if (!base_offset || MSB(base_offset, 7))
8eca0b47 2061 return 0; /* overflow */
eb32d236
NP
2062 c = base_info[used++];
2063 base_offset = (base_offset << 7) + (c & 127);
2064 }
2065 base_offset = delta_obj_offset - base_offset;
d8f32556 2066 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 2067 return 0; /* out of bound */
2b87c45b 2068 *curpos += used;
21666f1a 2069 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
2070 /* The base entry _must_ be in the same pack */
2071 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 2072 *curpos += 20;
eb32d236
NP
2073 } else
2074 die("I am totally screwed");
2b87c45b 2075 return base_offset;
eb32d236
NP
2076}
2077
5d642e75
JK
2078/*
2079 * Like get_delta_base above, but we return the sha1 instead of the pack
2080 * offset. This means it is cheaper for REF deltas (we do not have to do
2081 * the final object lookup), but more expensive for OFS deltas (we
2082 * have to load the revidx to convert the offset back into a sha1).
2083 */
2084static const unsigned char *get_delta_base_sha1(struct packed_git *p,
2085 struct pack_window **w_curs,
2086 off_t curpos,
2087 enum object_type type,
2088 off_t delta_obj_offset)
2089{
2090 if (type == OBJ_REF_DELTA) {
2091 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
2092 return base;
2093 } else if (type == OBJ_OFS_DELTA) {
2094 struct revindex_entry *revidx;
2095 off_t base_offset = get_delta_base(p, w_curs, &curpos,
2096 type, delta_obj_offset);
2097
2098 if (!base_offset)
2099 return NULL;
2100
2101 revidx = find_pack_revindex(p, base_offset);
2102 if (!revidx)
2103 return NULL;
2104
2105 return nth_packed_object_sha1(p, revidx->nr);
2106 } else
2107 return NULL;
2108}
2109
f8c8abc5
JH
2110int unpack_object_header(struct packed_git *p,
2111 struct pack_window **w_curs,
2112 off_t *curpos,
2113 unsigned long *sizep)
a733cb60 2114{
03e79c88 2115 unsigned char *base;
ef49a7a0 2116 unsigned long left;
ad1ed5ee 2117 unsigned long used;
2b87c45b 2118 enum object_type type;
a733cb60 2119
8d8a4ea5 2120 /* use_pack() assures us we have [base, base + 20) available
0353a0c4 2121 * as a range that we can look at. (Its actually the hash
3dff5379 2122 * size that is assured.) With our object header encoding
8d8a4ea5
SP
2123 * the maximum deflated object size is 2^137, which is just
2124 * insane, so we know won't exceed what we have been given.
2125 */
2b87c45b 2126 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
2127 used = unpack_object_header_buffer(base, left, &type, sizep);
2128 if (!used) {
2129 type = OBJ_BAD;
2130 } else
2131 *curpos += used;
ad1ed5ee 2132
2b87c45b 2133 return type;
a733cb60
LT
2134}
2135
790d96c0
TR
2136static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
2137{
2138 int type;
2139 struct revindex_entry *revidx;
2140 const unsigned char *sha1;
2141 revidx = find_pack_revindex(p, obj_offset);
2142 if (!revidx)
2143 return OBJ_BAD;
2144 sha1 = nth_packed_object_sha1(p, revidx->nr);
2145 mark_bad_packed_object(p, sha1);
2146 type = sha1_object_info(sha1, NULL);
2147 if (type <= OBJ_NONE)
2148 return OBJ_BAD;
2149 return type;
2150}
2151
790d96c0
TR
2152#define POI_STACK_PREALLOC 64
2153
90191d37
JK
2154static enum object_type packed_to_object_type(struct packed_git *p,
2155 off_t obj_offset,
2156 enum object_type type,
2157 struct pack_window **w_curs,
2158 off_t curpos)
1f688557 2159{
790d96c0
TR
2160 off_t small_poi_stack[POI_STACK_PREALLOC];
2161 off_t *poi_stack = small_poi_stack;
2162 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
5db47c2b 2163
790d96c0
TR
2164 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2165 off_t base_offset;
90191d37 2166 unsigned long size;
790d96c0
TR
2167 /* Push the object we're going to leave behind */
2168 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
2169 poi_stack_alloc = alloc_nr(poi_stack_nr);
b32fa95f 2170 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
790d96c0
TR
2171 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
2172 } else {
2173 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
2174 }
2175 poi_stack[poi_stack_nr++] = obj_offset;
2176 /* If parsing the base offset fails, just unwind */
90191d37 2177 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
790d96c0
TR
2178 if (!base_offset)
2179 goto unwind;
2180 curpos = obj_offset = base_offset;
90191d37 2181 type = unpack_object_header(p, w_curs, &curpos, &size);
790d96c0
TR
2182 if (type <= OBJ_NONE) {
2183 /* If getting the base itself fails, we first
2184 * retry the base, otherwise unwind */
2185 type = retry_bad_packed_offset(p, base_offset);
2186 if (type > OBJ_NONE)
2187 goto out;
2188 goto unwind;
2189 }
2190 }
2191
21666f1a 2192 switch (type) {
790d96c0 2193 case OBJ_BAD:
a733cb60 2194 case OBJ_COMMIT:
a733cb60 2195 case OBJ_TREE:
a733cb60 2196 case OBJ_BLOB:
a733cb60 2197 case OBJ_TAG:
a69d0943 2198 break;
1f688557 2199 default:
3d77d877
NP
2200 error("unknown object type %i at offset %"PRIuMAX" in %s",
2201 type, (uintmax_t)obj_offset, p->pack_name);
2202 type = OBJ_BAD;
1f688557 2203 }
790d96c0
TR
2204
2205out:
2206 if (poi_stack != small_poi_stack)
2207 free(poi_stack);
21666f1a 2208 return type;
790d96c0
TR
2209
2210unwind:
2211 while (poi_stack_nr) {
2212 obj_offset = poi_stack[--poi_stack_nr];
2213 type = retry_bad_packed_offset(p, obj_offset);
2214 if (type > OBJ_NONE)
2215 goto out;
2216 }
2217 type = OBJ_BAD;
2218 goto out;
1f688557
JH
2219}
2220
8261e1f1 2221static struct hashmap delta_base_cache;
18bdec11 2222static size_t delta_base_cached;
5e08ecbf 2223
12d95ef6 2224static LIST_HEAD(delta_base_cache_lru);
5e08ecbf 2225
8261e1f1 2226struct delta_base_cache_key {
e5e01619
LT
2227 struct packed_git *p;
2228 off_t base_offset;
8261e1f1
JK
2229};
2230
2231struct delta_base_cache_entry {
2232 struct hashmap hash;
2233 struct delta_base_cache_key key;
2234 struct list_head lru;
2235 void *data;
e5e01619 2236 unsigned long size;
e5e01619 2237 enum object_type type;
8261e1f1 2238};
e5e01619 2239
8261e1f1 2240static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
e5e01619 2241{
8261e1f1 2242 unsigned int hash;
e5e01619 2243
8261e1f1 2244 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
e5e01619 2245 hash += (hash >> 8) + (hash >> 16);
8261e1f1 2246 return hash;
e5e01619
LT
2247}
2248
84dd81c1
TR
2249static struct delta_base_cache_entry *
2250get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
5266d369 2251{
8261e1f1
JK
2252 struct hashmap_entry entry;
2253 struct delta_base_cache_key key;
2254
2255 if (!delta_base_cache.cmpfn)
2256 return NULL;
2257
2258 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
2259 key.p = p;
2260 key.base_offset = base_offset;
2261 return hashmap_get(&delta_base_cache, &entry, &key);
2262}
2263
2264static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
2265 const struct delta_base_cache_key *b)
2266{
2267 return a->p == b->p && a->base_offset == b->base_offset;
84dd81c1
TR
2268}
2269
7663cdc8
SB
2270static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
2271 const void *va, const void *vb,
8261e1f1 2272 const void *vkey)
84dd81c1 2273{
8261e1f1
JK
2274 const struct delta_base_cache_entry *a = va, *b = vb;
2275 const struct delta_base_cache_key *key = vkey;
2276 if (key)
2277 return !delta_base_cache_key_eq(&a->key, key);
2278 else
2279 return !delta_base_cache_key_eq(&a->key, &b->key);
5266d369
JH
2280}
2281
84dd81c1
TR
2282static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2283{
8261e1f1 2284 return !!get_delta_base_cache_entry(p, base_offset);
84dd81c1
TR
2285}
2286
4a5397ca
JK
2287/*
2288 * Remove the entry from the cache, but do _not_ free the associated
2289 * entry data. The caller takes ownership of the "data" buffer, and
2290 * should copy out any fields it wants before detaching.
2291 */
2292static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
84dd81c1 2293{
8261e1f1 2294 hashmap_remove(&delta_base_cache, ent, &ent->key);
12d95ef6 2295 list_del(&ent->lru);
84dd81c1 2296 delta_base_cached -= ent->size;
8261e1f1 2297 free(ent);
84dd81c1
TR
2298}
2299
62f255ad 2300static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
85fe35ab 2301 unsigned long *base_size, enum object_type *type)
62f255ad 2302{
84dd81c1 2303 struct delta_base_cache_entry *ent;
e5e01619 2304
84dd81c1 2305 ent = get_delta_base_cache_entry(p, base_offset);
8261e1f1 2306 if (!ent)
749bc58c 2307 return unpack_entry(p, base_offset, type, base_size);
e5e01619 2308
c84a1f3e
JT
2309 if (type)
2310 *type = ent->type;
2311 if (base_size)
2312 *base_size = ent->size;
85fe35ab 2313 return xmemdupz(ent->data, ent->size);
62f255ad
LT
2314}
2315
18bdec11
SP
2316static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2317{
8261e1f1
JK
2318 free(ent->data);
2319 detach_delta_base_cache_entry(ent);
18bdec11
SP
2320}
2321
3d20c636
SP
2322void clear_delta_base_cache(void)
2323{
abd5a002
JK
2324 struct list_head *lru, *tmp;
2325 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
2326 struct delta_base_cache_entry *entry =
2327 list_entry(lru, struct delta_base_cache_entry, lru);
8261e1f1
JK
2328 release_delta_base_cache(entry);
2329 }
3d20c636
SP
2330}
2331
62f255ad
LT
2332static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2333 void *base, unsigned long base_size, enum object_type type)
2334{
8261e1f1 2335 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
4b92bae7 2336 struct list_head *lru, *tmp;
e5e01619 2337
18bdec11 2338 delta_base_cached += base_size;
5e08ecbf 2339
4b92bae7 2340 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
12d95ef6
JK
2341 struct delta_base_cache_entry *f =
2342 list_entry(lru, struct delta_base_cache_entry, lru);
2343 if (delta_base_cached <= delta_base_cache_limit)
2344 break;
5e08ecbf
NP
2345 release_delta_base_cache(f);
2346 }
18bdec11 2347
8261e1f1
JK
2348 ent->key.p = p;
2349 ent->key.base_offset = base_offset;
e5e01619
LT
2350 ent->type = type;
2351 ent->data = base;
2352 ent->size = base_size;
12d95ef6 2353 list_add_tail(&ent->lru, &delta_base_cache_lru);
8261e1f1
JK
2354
2355 if (!delta_base_cache.cmpfn)
7663cdc8 2356 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
8261e1f1
JK
2357 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
2358 hashmap_add(&delta_base_cache, ent);
62f255ad
LT
2359}
2360
845b102b
JT
2361int packed_object_info(struct packed_git *p, off_t obj_offset,
2362 struct object_info *oi)
2363{
2364 struct pack_window *w_curs = NULL;
2365 unsigned long size;
2366 off_t curpos = obj_offset;
2367 enum object_type type;
2368
2369 /*
2370 * We always get the representation type, but only convert it to
2371 * a "real" type later if the caller is interested.
2372 */
c84a1f3e
JT
2373 if (oi->contentp) {
2374 *oi->contentp = cache_or_unpack_entry(p, obj_offset, oi->sizep,
2375 &type);
2376 if (!*oi->contentp)
2377 type = OBJ_BAD;
2378 } else {
2379 type = unpack_object_header(p, &w_curs, &curpos, &size);
2380 }
845b102b 2381
c84a1f3e 2382 if (!oi->contentp && oi->sizep) {
845b102b
JT
2383 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2384 off_t tmp_pos = curpos;
2385 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
2386 type, obj_offset);
2387 if (!base_offset) {
2388 type = OBJ_BAD;
2389 goto out;
2390 }
2391 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2392 if (*oi->sizep == 0) {
2393 type = OBJ_BAD;
2394 goto out;
2395 }
2396 } else {
2397 *oi->sizep = size;
2398 }
2399 }
2400
2401 if (oi->disk_sizep) {
2402 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2403 *oi->disk_sizep = revidx[1].offset - obj_offset;
2404 }
2405
2406 if (oi->typep || oi->typename) {
2407 enum object_type ptot;
2408 ptot = packed_to_object_type(p, obj_offset, type, &w_curs,
2409 curpos);
2410 if (oi->typep)
2411 *oi->typep = ptot;
2412 if (oi->typename) {
2413 const char *tn = typename(ptot);
2414 if (tn)
2415 strbuf_addstr(oi->typename, tn);
2416 }
2417 if (ptot < 0) {
2418 type = OBJ_BAD;
2419 goto out;
2420 }
2421 }
2422
2423 if (oi->delta_base_sha1) {
2424 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2425 const unsigned char *base;
2426
2427 base = get_delta_base_sha1(p, &w_curs, curpos,
2428 type, obj_offset);
2429 if (!base) {
2430 type = OBJ_BAD;
2431 goto out;
2432 }
2433
2434 hashcpy(oi->delta_base_sha1, base);
2435 } else
2436 hashclr(oi->delta_base_sha1);
2437 }
2438
2439out:
2440 unuse_pack(&w_curs);
2441 return type;
2442}
2443
2444static void *unpack_compressed_entry(struct packed_git *p,
2445 struct pack_window **w_curs,
2446 off_t curpos,
2447 unsigned long size)
2448{
2449 int st;
2450 git_zstream stream;
2451 unsigned char *buffer, *in;
2452
2453 buffer = xmallocz_gently(size);
2454 if (!buffer)
2455 return NULL;
2456 memset(&stream, 0, sizeof(stream));
2457 stream.next_out = buffer;
2458 stream.avail_out = size + 1;
2459
2460 git_inflate_init(&stream);
2461 do {
2462 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2463 stream.next_in = in;
2464 st = git_inflate(&stream, Z_FINISH);
2465 if (!stream.avail_out)
2466 break; /* the payload is larger than it should be */
2467 curpos += stream.next_in - in;
2468 } while (st == Z_OK || st == Z_BUF_ERROR);
2469 git_inflate_end(&stream);
2470 if ((st != Z_STREAM_END) || stream.total_out != size) {
2471 free(buffer);
2472 return NULL;
2473 }
2474
2475 return buffer;
2476}
2477
c2c5b270
CC
2478static void *read_object(const unsigned char *sha1, enum object_type *type,
2479 unsigned long *size);
2480
5f44324d
JH
2481static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2482{
67dc598e
KB
2483 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2484 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2485 p->pack_name, (uintmax_t)obj_offset);
5f44324d
JH
2486}
2487
0e8189e2
NP
2488int do_check_packed_object_crc;
2489
abe601bb
TR
2490#define UNPACK_ENTRY_STACK_PREALLOC 64
2491struct unpack_entry_stack_ent {
2492 off_t obj_offset;
2493 off_t curpos;
2494 unsigned long size;
2495};
2496
c4001d92 2497void *unpack_entry(struct packed_git *p, off_t obj_offset,
abe601bb 2498 enum object_type *final_type, unsigned long *final_size)
1f688557 2499{
03e79c88 2500 struct pack_window *w_curs = NULL;
c4001d92 2501 off_t curpos = obj_offset;
abe601bb
TR
2502 void *data = NULL;
2503 unsigned long size;
2504 enum object_type type;
2505 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2506 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2507 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2508 int base_from_cache = 0;
1f688557 2509
67dc598e 2510 write_pack_access_log(p, obj_offset);
5f44324d 2511
abe601bb
TR
2512 /* PHASE 1: drill down to the innermost base object */
2513 for (;;) {
2514 off_t base_offset;
2515 int i;
2516 struct delta_base_cache_entry *ent;
2517
77965f8b 2518 ent = get_delta_base_cache_entry(p, curpos);
8261e1f1 2519 if (ent) {
77965f8b
NTND
2520 type = ent->type;
2521 data = ent->data;
2522 size = ent->size;
4a5397ca 2523 detach_delta_base_cache_entry(ent);
77965f8b
NTND
2524 base_from_cache = 1;
2525 break;
2526 }
2527
abe601bb
TR
2528 if (do_check_packed_object_crc && p->index_version > 1) {
2529 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
211c61c6 2530 off_t len = revidx[1].offset - obj_offset;
abe601bb
TR
2531 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2532 const unsigned char *sha1 =
2533 nth_packed_object_sha1(p, revidx->nr);
2534 error("bad packed object CRC for %s",
2535 sha1_to_hex(sha1));
2536 mark_bad_packed_object(p, sha1);
896dca3a
RS
2537 data = NULL;
2538 goto out;
abe601bb
TR
2539 }
2540 }
2541
abe601bb
TR
2542 type = unpack_object_header(p, &w_curs, &curpos, &size);
2543 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2544 break;
2545
2546 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2547 if (!base_offset) {
2548 error("failed to validate delta base reference "
2549 "at offset %"PRIuMAX" from %s",
2550 (uintmax_t)curpos, p->pack_name);
2551 /* bail to phase 2, in hopes of recovery */
2552 data = NULL;
2553 break;
2554 }
2555
2556 /* push object, proceed to base */
2557 if (delta_stack_nr >= delta_stack_alloc
2558 && delta_stack == small_delta_stack) {
2559 delta_stack_alloc = alloc_nr(delta_stack_nr);
b32fa95f 2560 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
abe601bb
TR
2561 memcpy(delta_stack, small_delta_stack,
2562 sizeof(*delta_stack)*delta_stack_nr);
2563 } else {
2564 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2565 }
2566 i = delta_stack_nr++;
2567 delta_stack[i].obj_offset = obj_offset;
2568 delta_stack[i].curpos = curpos;
2569 delta_stack[i].size = size;
2570
2571 curpos = obj_offset = base_offset;
0e8189e2
NP
2572 }
2573
abe601bb
TR
2574 /* PHASE 2: handle the base */
2575 switch (type) {
eb32d236
NP
2576 case OBJ_OFS_DELTA:
2577 case OBJ_REF_DELTA:
abe601bb 2578 if (data)
ef1177d1 2579 die("BUG: unpack_entry: left loop at a valid delta");
4d703a1a 2580 break;
a733cb60 2581 case OBJ_COMMIT:
a733cb60 2582 case OBJ_TREE:
a733cb60 2583 case OBJ_BLOB:
a733cb60 2584 case OBJ_TAG:
abe601bb
TR
2585 if (!base_from_cache)
2586 data = unpack_compressed_entry(p, &w_curs, curpos, size);
4d703a1a 2587 break;
1f688557 2588 default:
8eca0b47
NP
2589 data = NULL;
2590 error("unknown object type %i at offset %"PRIuMAX" in %s",
abe601bb
TR
2591 type, (uintmax_t)obj_offset, p->pack_name);
2592 }
2593
2594 /* PHASE 3: apply deltas in order */
2595
2596 /* invariants:
2597 * 'data' holds the base data, or NULL if there was corruption
2598 */
2599 while (delta_stack_nr) {
2600 void *delta_data;
2601 void *base = data;
886ddf47 2602 void *external_base = NULL;
abe601bb
TR
2603 unsigned long delta_size, base_size = size;
2604 int i;
2605
2606 data = NULL;
2607
2608 if (base)
2609 add_delta_base_cache(p, obj_offset, base, base_size, type);
2610
2611 if (!base) {
2612 /*
2613 * We're probably in deep shit, but let's try to fetch
2614 * the required base anyway from another pack or loose.
2615 * This is costly but should happen only in the presence
2616 * of a corrupted pack, and is better than failing outright.
2617 */
2618 struct revindex_entry *revidx;
2619 const unsigned char *base_sha1;
2620 revidx = find_pack_revindex(p, obj_offset);
2621 if (revidx) {
2622 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2623 error("failed to read delta base object %s"
2624 " at offset %"PRIuMAX" from %s",
2625 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2626 p->pack_name);
2627 mark_bad_packed_object(p, base_sha1);
2628 base = read_object(base_sha1, &type, &base_size);
886ddf47 2629 external_base = base;
abe601bb
TR
2630 }
2631 }
2632
2633 i = --delta_stack_nr;
2634 obj_offset = delta_stack[i].obj_offset;
2635 curpos = delta_stack[i].curpos;
2636 delta_size = delta_stack[i].size;
2637
2638 if (!base)
2639 continue;
2640
2641 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2642
2643 if (!delta_data) {
2644 error("failed to unpack compressed delta "
2645 "at offset %"PRIuMAX" from %s",
2646 (uintmax_t)curpos, p->pack_name);
abe601bb 2647 data = NULL;
886ddf47 2648 free(external_base);
abe601bb
TR
2649 continue;
2650 }
2651
2652 data = patch_delta(base, base_size,
2653 delta_data, delta_size,
2654 &size);
1ee886c1
JK
2655
2656 /*
2657 * We could not apply the delta; warn the user, but keep going.
2658 * Our failure will be noticed either in the next iteration of
2659 * the loop, or if this is the final delta, in the caller when
2660 * we return NULL. Those code paths will take care of making
2661 * a more explicit warning and retrying with another copy of
2662 * the object.
2663 */
abe601bb 2664 if (!data)
1ee886c1 2665 error("failed to apply delta");
abe601bb 2666
4b8f772c 2667 free(delta_data);
886ddf47 2668 free(external_base);
1f688557 2669 }
abe601bb 2670
c84a1f3e
JT
2671 if (final_type)
2672 *final_type = type;
2673 if (final_size)
2674 *final_size = size;
abe601bb 2675
896dca3a 2676out:
03e79c88 2677 unuse_pack(&w_curs);
019d1e65
NTND
2678
2679 if (delta_stack != small_delta_stack)
2680 free(delta_stack);
2681
21666f1a 2682 return data;
1f688557
JH
2683}
2684
d079837e 2685const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 2686 uint32_t n)
9a217f2a 2687{
42873078 2688 const unsigned char *index = p->index_data;
d079837e
SP
2689 if (!index) {
2690 if (open_pack_index(p))
2691 return NULL;
2692 index = p->index_data;
2693 }
57059091 2694 if (n >= p->num_objects)
d72308e0 2695 return NULL;
74e34e1f
NP
2696 index += 4 * 256;
2697 if (p->index_version == 1) {
2698 return index + 24 * n + 4;
2699 } else {
2700 index += 8;
2701 return index + 20 * n;
2702 }
2703}
2704
068f85e3 2705const struct object_id *nth_packed_object_oid(struct object_id *oid,
2706 struct packed_git *p,
2707 uint32_t n)
2708{
2709 const unsigned char *hash = nth_packed_object_sha1(p, n);
2710 if (!hash)
2711 return NULL;
2712 hashcpy(oid->hash, hash);
2713 return oid;
2714}
2715
47fe3f6e
JK
2716void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
2717{
2718 const unsigned char *ptr = vptr;
2719 const unsigned char *start = p->index_data;
2720 const unsigned char *end = start + p->index_size;
2721 if (ptr < start)
7465feba 2722 die(_("offset before start of pack index for %s (corrupt index?)"),
47fe3f6e
JK
2723 p->pack_name);
2724 /* No need to check for underflow; .idx files must be at least 8 bytes */
2725 if (ptr >= end - 8)
7465feba 2726 die(_("offset beyond end of pack index for %s (truncated index?)"),
47fe3f6e
JK
2727 p->pack_name);
2728}
2729
99093238 2730off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
2731{
2732 const unsigned char *index = p->index_data;
2733 index += 4 * 256;
2734 if (p->index_version == 1) {
2735 return ntohl(*((uint32_t *)(index + 24 * n)));
2736 } else {
2737 uint32_t off;
2738 index += 8 + p->num_objects * (20 + 4);
2739 off = ntohl(*((uint32_t *)(index + 4 * n)));
2740 if (!(off & 0x80000000))
2741 return off;
2742 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
47fe3f6e 2743 check_pack_index_ptr(p, index);
74e34e1f
NP
2744 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2745 ntohl(*((uint32_t *)(index + 4)));
2746 }
9a217f2a
JH
2747}
2748
c4001d92 2749off_t find_pack_entry_one(const unsigned char *sha1,
43057304 2750 struct packed_git *p)
1f688557 2751{
42873078 2752 const uint32_t *level1_ofs = p->index_data;
42873078 2753 const unsigned char *index = p->index_data;
628522ec 2754 unsigned hi, lo, stride;
628522ec
JH
2755 static int debug_lookup = -1;
2756
2757 if (debug_lookup < 0)
2758 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 2759
d079837e
SP
2760 if (!index) {
2761 if (open_pack_index(p))
2762 return 0;
2763 level1_ofs = p->index_data;
2764 index = p->index_data;
2765 }
74e34e1f
NP
2766 if (p->index_version > 1) {
2767 level1_ofs += 2;
2768 index += 8;
2769 }
42873078 2770 index += 4 * 256;
74e34e1f
NP
2771 hi = ntohl(level1_ofs[*sha1]);
2772 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
2773 if (p->index_version > 1) {
2774 stride = 20;
2775 } else {
2776 stride = 24;
2777 index += 4;
2778 }
2779
2780 if (debug_lookup)
6e1c2344 2781 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
2782 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2783
6355a768 2784 while (lo < hi) {
74e34e1f 2785 unsigned mi = (lo + hi) / 2;
628522ec
JH
2786 int cmp = hashcmp(index + mi * stride, sha1);
2787
2788 if (debug_lookup)
2789 printf("lo %u hi %u rg %u mi %u\n",
2790 lo, hi, hi - lo, mi);
43057304 2791 if (!cmp)
74e34e1f 2792 return nth_packed_object_offset(p, mi);
1f688557
JH
2793 if (cmp > 0)
2794 hi = mi;
2795 else
2796 lo = mi+1;
6355a768 2797 }
1f688557
JH
2798 return 0;
2799}
2800
4c080182 2801int is_pack_valid(struct packed_git *p)
d131b7af
SP
2802{
2803 /* An already open pack is known to be valid. */
2804 if (p->pack_fd != -1)
2805 return 1;
2806
2807 /* If the pack has one window completely covering the
2808 * file size, the pack is known to be valid even if
2809 * the descriptor is not currently open.
2810 */
2811 if (p->windows) {
2812 struct pack_window *w = p->windows;
2813
2814 if (!w->offset && w->len == p->pack_size)
2815 return 1;
2816 }
2817
2818 /* Force the pack to open to prove its valid. */
2819 return !open_packed_git(p);
2820}
2821
95099731
NTND
2822static int fill_pack_entry(const unsigned char *sha1,
2823 struct pack_entry *e,
2824 struct packed_git *p)
2825{
2826 off_t offset;
2827
2828 if (p->num_bad_objects) {
2829 unsigned i;
2830 for (i = 0; i < p->num_bad_objects; i++)
2831 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2832 return 0;
2833 }
2834
2835 offset = find_pack_entry_one(sha1, p);
2836 if (!offset)
2837 return 0;
2838
2839 /*
2840 * We are about to tell the caller where they can locate the
2841 * requested object. We better make sure the packfile is
2842 * still here and can be accessed before supplying that
2843 * answer, as it may have been deleted since the index was
2844 * loaded!
2845 */
319b678a 2846 if (!is_pack_valid(p))
95099731 2847 return 0;
95099731
NTND
2848 e->offset = offset;
2849 e->p = p;
2850 hashcpy(e->sha1, sha1);
2851 return 1;
2852}
2853
d40d535b
MH
2854/*
2855 * Iff a pack file contains the object named by sha1, return true and
2856 * store its location to e.
2857 */
4d6acb70 2858static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557 2859{
a73cdd21 2860 struct mru_entry *p;
43057304 2861
1f688557 2862 prepare_packed_git();
f7c22cc6
NP
2863 if (!packed_git)
2864 return 0;
1f688557 2865
a73cdd21
JK
2866 for (p = packed_git_mru->head; p; p = p->next) {
2867 if (fill_pack_entry(sha1, e, p->item)) {
2868 mru_mark(packed_git_mru, p);
1b1005d1
MH
2869 return 1;
2870 }
c01f51cc 2871 }
1f688557
JH
2872 return 0;
2873}
2874
a6080a0a 2875struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
2876 struct packed_git *packs)
2877{
2878 struct packed_git *p;
bf592c50
DB
2879
2880 for (p = packs; p; p = p->next) {
43057304 2881 if (find_pack_entry_one(sha1, p))
bf592c50
DB
2882 return p;
2883 }
2884 return NULL;
21666f1a 2885
bf592c50
DB
2886}
2887
052fe5ea 2888static int sha1_loose_object_info(const unsigned char *sha1,
46f03448
KN
2889 struct object_info *oi,
2890 int flags)
65c2e0c3 2891{
46f03448
KN
2892 int status = 0;
2893 unsigned long mapsize;
65c2e0c3 2894 void *map;
ef49a7a0 2895 git_zstream stream;
d65a16f6 2896 char hdr[32];
46f03448 2897 struct strbuf hdrbuf = STRBUF_INIT;
c84a1f3e 2898 unsigned long size_scratch;
65c2e0c3 2899
5d642e75
JK
2900 if (oi->delta_base_sha1)
2901 hashclr(oi->delta_base_sha1);
2902
052fe5ea
JK
2903 /*
2904 * If we don't care about type or size, then we don't
4ef8d1dd
JH
2905 * need to look inside the object at all. Note that we
2906 * do not optimize out the stat call, even if the
2907 * caller doesn't care about the disk-size, since our
2908 * return value implicitly indicates whether the
2909 * object even exists.
052fe5ea 2910 */
c84a1f3e 2911 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
771e7d57 2912 const char *path;
4ef8d1dd 2913 struct stat st;
771e7d57 2914 if (stat_sha1_file(sha1, &st, &path) < 0)
4ef8d1dd
JH
2915 return -1;
2916 if (oi->disk_sizep)
23c339c0 2917 *oi->disk_sizep = st.st_size;
052fe5ea
JK
2918 return 0;
2919 }
2920
bb6b8e4f 2921 map = map_sha1_file(sha1, &mapsize);
f0df4ed5 2922 if (!map)
dbea72a8 2923 return -1;
c84a1f3e
JT
2924
2925 if (!oi->sizep)
2926 oi->sizep = &size_scratch;
2927
23c339c0
JK
2928 if (oi->disk_sizep)
2929 *oi->disk_sizep = mapsize;
19fc5e84 2930 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
46f03448
KN
2931 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2932 status = error("unable to unpack %s header with --allow-unknown-type",
2933 sha1_to_hex(sha1));
2934 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
36e4d74a
JH
2935 status = error("unable to unpack %s header",
2936 sha1_to_hex(sha1));
46f03448
KN
2937 if (status < 0)
2938 ; /* Do nothing */
2939 else if (hdrbuf.len) {
2940 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2941 status = error("unable to parse %s header with --allow-unknown-type",
2942 sha1_to_hex(sha1));
2943 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
36e4d74a 2944 status = error("unable to parse %s header", sha1_to_hex(sha1));
c84a1f3e 2945
b3ea7dd3 2946 if (status >= 0 && oi->contentp) {
c84a1f3e
JT
2947 *oi->contentp = unpack_sha1_rest(&stream, hdr,
2948 *oi->sizep, sha1);
b3ea7dd3
JK
2949 if (!*oi->contentp) {
2950 git_inflate_end(&stream);
2951 status = -1;
2952 }
2953 } else
c84a1f3e
JT
2954 git_inflate_end(&stream);
2955
65c2e0c3 2956 munmap(map, mapsize);
46f03448 2957 if (status && oi->typep)
23c339c0 2958 *oi->typep = status;
c84a1f3e
JT
2959 if (oi->sizep == &size_scratch)
2960 oi->sizep = NULL;
46f03448 2961 strbuf_release(&hdrbuf);
93cff9a9 2962 return (status < 0) ? status : 0;
65c2e0c3
JH
2963}
2964
de7b5d62 2965int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
f0df4ed5 2966{
cd585e2a 2967 static struct object_info blank_oi = OBJECT_INFO_INIT;
f0df4ed5 2968 struct pack_entry e;
5b086407 2969 int rtype;
1f0c0d36
JT
2970 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
2971 lookup_replace_object(sha1) :
2972 sha1;
f0df4ed5 2973
87b5e236
JK
2974 if (is_null_sha1(real))
2975 return -1;
2976
cd585e2a
JT
2977 if (!oi)
2978 oi = &blank_oi;
2979
dfdd4afc
JT
2980 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
2981 struct cached_object *co = find_cached_object(real);
2982 if (co) {
2983 if (oi->typep)
2984 *(oi->typep) = co->type;
2985 if (oi->sizep)
2986 *(oi->sizep) = co->size;
2987 if (oi->disk_sizep)
2988 *(oi->disk_sizep) = 0;
2989 if (oi->delta_base_sha1)
2990 hashclr(oi->delta_base_sha1);
2991 if (oi->typename)
2992 strbuf_addstr(oi->typename, typename(co->type));
2993 if (oi->contentp)
2994 *oi->contentp = xmemdupz(co->buf, co->size);
2995 oi->whence = OI_CACHED;
2996 return 0;
2997 }
c4d9986f
NTND
2998 }
2999
1f7117ef 3000 if (!find_pack_entry(real, &e)) {
ddd63e64 3001 /* Most likely it's a loose object. */
46f03448 3002 if (!sha1_loose_object_info(real, oi, flags)) {
9a490590 3003 oi->whence = OI_LOOSE;
5b086407 3004 return 0;
9a490590 3005 }
ddd63e64
SG
3006
3007 /* Not a loose object; someone else may have just packed it. */
dfdd4afc 3008 if (flags & OBJECT_INFO_QUICK) {
052fe5ea 3009 return -1;
dfdd4afc
JT
3010 } else {
3011 reprepare_packed_git();
3012 if (!find_pack_entry(real, &e))
3013 return -1;
3014 }
f0df4ed5 3015 }
3d77d877 3016
cd585e2a
JT
3017 if (oi == &blank_oi)
3018 /*
3019 * We know that the caller doesn't actually need the
3020 * information below, so return early.
3021 */
3022 return 0;
46f03448 3023
23c339c0 3024 rtype = packed_object_info(e.p, e.offset, oi);
412916ee 3025 if (rtype < 0) {
1f7117ef
CC
3026 mark_bad_packed_object(e.p, real);
3027 return sha1_object_info_extended(real, oi, 0);
5266d369
JH
3028 } else if (in_delta_base_cache(e.p, e.offset)) {
3029 oi->whence = OI_DBCACHED;
9a490590
JH
3030 } else {
3031 oi->whence = OI_PACKED;
3032 oi->u.packed.offset = e.offset;
3033 oi->u.packed.pack = e.p;
3034 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
3035 rtype == OBJ_OFS_DELTA);
3d77d877
NP
3036 }
3037
5b086407 3038 return 0;
f0df4ed5
JS
3039}
3040
3fc0dca9 3041/* returns enum object_type or negative */
9a490590
JH
3042int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
3043{
5b086407 3044 enum object_type type;
27b5c1a0 3045 struct object_info oi = OBJECT_INFO_INIT;
9a490590 3046
5b086407 3047 oi.typep = &type;
9a490590 3048 oi.sizep = sizep;
1f0c0d36
JT
3049 if (sha1_object_info_extended(sha1, &oi,
3050 OBJECT_INFO_LOOKUP_REPLACE) < 0)
5b086407
JK
3051 return -1;
3052 return type;
9a490590
JH
3053}
3054
21666f1a
NP
3055static void *read_packed_sha1(const unsigned char *sha1,
3056 enum object_type *type, unsigned long *size)
1f688557
JH
3057{
3058 struct pack_entry e;
8eca0b47 3059 void *data;
1f688557 3060
cd673c1f 3061 if (!find_pack_entry(sha1, &e))
1f688557 3062 return NULL;
85fe35ab 3063 data = cache_or_unpack_entry(e.p, e.offset, size, type);
8eca0b47
NP
3064 if (!data) {
3065 /*
3066 * We're probably in deep shit, but let's try to fetch
3067 * the required object anyway from another pack or loose.
3068 * This should happen only in the presence of a corrupted
3069 * pack, and is better than failing outright.
3070 */
3071 error("failed to read object %s at offset %"PRIuMAX" from %s",
3072 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
3073 mark_bad_packed_object(e.p, sha1);
ac939109 3074 data = read_object(sha1, type, size);
8eca0b47
NP
3075 }
3076 return data;
1f688557
JH
3077}
3078
21666f1a
NP
3079int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
3080 unsigned char *sha1)
d66b37bb
JH
3081{
3082 struct cached_object *co;
3083
21666f1a 3084 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
3085 if (has_sha1_file(sha1) || find_cached_object(sha1))
3086 return 0;
c7353967 3087 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
d66b37bb
JH
3088 co = &cached_objects[cached_object_nr++];
3089 co->size = len;
21666f1a 3090 co->type = type;
efa13f7b
JH
3091 co->buf = xmalloc(len);
3092 memcpy(co->buf, buf, len);
d66b37bb
JH
3093 hashcpy(co->sha1, sha1);
3094 return 0;
3095}
3096
c2c5b270
CC
3097static void *read_object(const unsigned char *sha1, enum object_type *type,
3098 unsigned long *size)
0fcfd160 3099{
c84a1f3e
JT
3100 struct object_info oi = OBJECT_INFO_INIT;
3101 void *content;
3102 oi.typep = type;
3103 oi.sizep = size;
3104 oi.contentp = &content;
0fcfd160 3105
c84a1f3e
JT
3106 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
3107 return NULL;
3108 return content;
0fcfd160
LT
3109}
3110
b6c4cecc
JH
3111/*
3112 * This function dies on corrupt objects; the callers who want to
3113 * deal with them should arrange to call read_object() and give error
3114 * messages themselves.
3115 */
5bf29b95
JH
3116void *read_sha1_file_extended(const unsigned char *sha1,
3117 enum object_type *type,
3118 unsigned long *size,
1f0c0d36 3119 int lookup_replace)
ac939109 3120{
3ba7a065 3121 void *data;
b6c4cecc 3122 const struct packed_git *p;
771e7d57
JK
3123 const char *path;
3124 struct stat st;
1f0c0d36
JT
3125 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
3126 : sha1;
b6c4cecc 3127
3ba7a065
JH
3128 errno = 0;
3129 data = read_object(repl, type, size);
4bbf5a26 3130 if (data)
b6c4cecc 3131 return data;
68095570 3132
25f3af3f 3133 if (errno && errno != ENOENT)
3ba7a065
JH
3134 die_errno("failed to read object %s", sha1_to_hex(sha1));
3135
68095570 3136 /* die if we replaced an object with one that does not exist */
b6c4cecc 3137 if (repl != sha1)
68095570
CC
3138 die("replacement %s not found for %s",
3139 sha1_to_hex(repl), sha1_to_hex(sha1));
3140
771e7d57 3141 if (!stat_sha1_file(repl, &st, &path))
b6c4cecc
JH
3142 die("loose object %s (stored in %s) is corrupt",
3143 sha1_to_hex(repl), path);
68095570 3144
b6c4cecc
JH
3145 if ((p = has_packed_and_bad(repl)) != NULL)
3146 die("packed object %s (stored in %s) is corrupt",
3147 sha1_to_hex(repl), p->pack_name);
f5552aee 3148
b6c4cecc 3149 return NULL;
ac939109
NP
3150}
3151
40469ee9 3152void *read_object_with_reference(const unsigned char *sha1,
21666f1a 3153 const char *required_type_name,
40469ee9
JH
3154 unsigned long *size,
3155 unsigned char *actual_sha1_return)
f4913f91 3156{
21666f1a 3157 enum object_type type, required_type;
f4913f91
JH
3158 void *buffer;
3159 unsigned long isize;
40469ee9 3160 unsigned char actual_sha1[20];
f4913f91 3161
21666f1a 3162 required_type = type_from_string(required_type_name);
e702496e 3163 hashcpy(actual_sha1, sha1);
40469ee9
JH
3164 while (1) {
3165 int ref_length = -1;
3166 const char *ref_type = NULL;
f4913f91 3167
21666f1a 3168 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
3169 if (!buffer)
3170 return NULL;
21666f1a 3171 if (type == required_type) {
40469ee9
JH
3172 *size = isize;
3173 if (actual_sha1_return)
e702496e 3174 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
3175 return buffer;
3176 }
3177 /* Handle references */
21666f1a 3178 else if (type == OBJ_COMMIT)
40469ee9 3179 ref_type = "tree ";
21666f1a 3180 else if (type == OBJ_TAG)
40469ee9
JH
3181 ref_type = "object ";
3182 else {
3183 free(buffer);
3184 return NULL;
3185 }
3186 ref_length = strlen(ref_type);
f4913f91 3187
50974ec9
MK
3188 if (ref_length + 40 > isize ||
3189 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 3190 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
3191 free(buffer);
3192 return NULL;
3193 }
1cf58e72 3194 free(buffer);
40469ee9
JH
3195 /* Now we have the ID of the referred-to object in
3196 * actual_sha1. Check again. */
f4913f91 3197 }
f4913f91
JH
3198}
3199
ce9fbf16 3200static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 3201 const char *type, unsigned char *sha1,
d65a16f6 3202 char *hdr, int *hdrlen)
d410c0f5 3203{
9126f009 3204 git_SHA_CTX c;
d410c0f5
JH
3205
3206 /* Generate the header */
ef1286d3 3207 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
d410c0f5
JH
3208
3209 /* Sha1.. */
9126f009
NP
3210 git_SHA1_Init(&c);
3211 git_SHA1_Update(&c, hdr, *hdrlen);
3212 git_SHA1_Update(&c, buf, len);
3213 git_SHA1_Final(sha1, &c);
d410c0f5
JH
3214}
3215
230f1322 3216/*
5a688fe4 3217 * Move the just written object into its final resting place.
230f1322 3218 */
cb5add58 3219int finalize_object_file(const char *tmpfile, const char *filename)
230f1322 3220{
e32c0a9c 3221 int ret = 0;
5a688fe4 3222
348df166 3223 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
3224 goto try_rename;
3225 else if (link(tmpfile, filename))
e32c0a9c 3226 ret = errno;
7ebb6fca
LT
3227
3228 /*
3229 * Coda hack - coda doesn't like cross-directory links,
3230 * so we fall back to a rename, which will mean that it
3231 * won't be able to check collisions, but that's not a
3232 * big deal.
3233 *
3234 * The same holds for FAT formatted media.
3235 *
3be1f18e 3236 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
3237 * left to unlink.
3238 */
3239 if (ret && ret != EEXIST) {
be66a6c4 3240 try_rename:
7ebb6fca 3241 if (!rename(tmpfile, filename))
3be1f18e 3242 goto out;
9e48b389 3243 ret = errno;
230f1322 3244 }
691f1a28 3245 unlink_or_warn(tmpfile);
230f1322
LT
3246 if (ret) {
3247 if (ret != EEXIST) {
7616c6ca 3248 return error_errno("unable to write sha1 filename %s", filename);
230f1322
LT
3249 }
3250 /* FIXME!!! Collision check here ? */
3251 }
3252
3be1f18e 3253out:
5256b006 3254 if (adjust_shared_perm(filename))
5a688fe4 3255 return error("unable to set permission to '%s'", filename);
230f1322
LT
3256 return 0;
3257}
3258
4d548150
LT
3259static int write_buffer(int fd, const void *buf, size_t len)
3260{
d34cf19b 3261 if (write_in_full(fd, buf, len) < 0)
7616c6ca 3262 return error_errno("file write error");
4d548150
LT
3263 return 0;
3264}
3265
ce9fbf16 3266int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
3267 unsigned char *sha1)
3268{
d65a16f6 3269 char hdr[32];
ef1286d3 3270 int hdrlen = sizeof(hdr);
abdc3fc8
RS
3271 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3272 return 0;
3273}
3274
e9039dd3
LT
3275/* Finalize a file on disk, and close it. */
3276static void close_sha1_file(int fd)
3277{
aafe9fba
LT
3278 if (fsync_object_files)
3279 fsync_or_die(fd, "sha1 file");
e9039dd3 3280 if (close(fd) != 0)
d824cbba 3281 die_errno("error when closing sha1 file");
e9039dd3
LT
3282}
3283
5723fe7e
LT
3284/* Size of directory component, including the ending '/' */
3285static inline int directory_size(const char *filename)
3286{
3287 const char *s = strrchr(filename, '/');
3288 if (!s)
3289 return 0;
3290 return s - filename + 1;
3291}
3292
3293/*
3294 * This creates a temporary file in the same directory as the final
3295 * 'filename'
3296 *
3297 * We want to avoid cross-directory filename renames, because those
3298 * can have problems on various filesystems (FAT, NFS, Coda).
3299 */
d4b3d11a 3300static int create_tmpfile(struct strbuf *tmp, const char *filename)
5723fe7e
LT
3301{
3302 int fd, dirlen = directory_size(filename);
3303
d4b3d11a
JK
3304 strbuf_reset(tmp);
3305 strbuf_add(tmp, filename, dirlen);
3306 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
3307 fd = git_mkstemp_mode(tmp->buf, 0444);
cbacbf4e 3308 if (fd < 0 && dirlen && errno == ENOENT) {
d4b3d11a
JK
3309 /*
3310 * Make sure the directory exists; note that the contents
3311 * of the buffer are undefined after mkstemp returns an
3312 * error, so we have to rewrite the whole buffer from
3313 * scratch.
3314 */
3315 strbuf_reset(tmp);
3316 strbuf_add(tmp, filename, dirlen - 1);
3317 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
b2476a60 3318 return -1;
d4b3d11a 3319 if (adjust_shared_perm(tmp->buf))
5723fe7e
LT
3320 return -1;
3321
3322 /* Try again */
d4b3d11a
JK
3323 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
3324 fd = git_mkstemp_mode(tmp->buf, 0444);
5723fe7e
LT
3325 }
3326 return fd;
3327}
3328
bbac7311 3329static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 3330 const void *buf, unsigned long len, time_t mtime)
0fcfd160 3331{
915308b1 3332 int fd, ret;
9892beba 3333 unsigned char compressed[4096];
ef49a7a0 3334 git_zstream stream;
748af44c
NP
3335 git_SHA_CTX c;
3336 unsigned char parano_sha1[20];
d4b3d11a 3337 static struct strbuf tmp_file = STRBUF_INIT;
30d6c6ea 3338 const char *filename = sha1_file_name(sha1);
a44c9a5e 3339
d4b3d11a 3340 fd = create_tmpfile(&tmp_file, filename);
aac17941 3341 if (fd < 0) {
35243577 3342 if (errno == EACCES)
82247e9b 3343 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
916d081b 3344 else
7616c6ca 3345 return error_errno("unable to create temporary file");
aac17941
LT
3346 }
3347
0fcfd160 3348 /* Set it up */
55bb5c91 3349 git_deflate_init(&stream, zlib_compression_level);
0fcfd160 3350 stream.next_out = compressed;
9892beba 3351 stream.avail_out = sizeof(compressed);
748af44c 3352 git_SHA1_Init(&c);
a44c9a5e
LT
3353
3354 /* First header.. */
d65a16f6 3355 stream.next_in = (unsigned char *)hdr;
a44c9a5e 3356 stream.avail_in = hdrlen;
55bb5c91
JH
3357 while (git_deflate(&stream, 0) == Z_OK)
3358 ; /* nothing */
748af44c 3359 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
3360
3361 /* Then the data itself.. */
c00e657d 3362 stream.next_in = (void *)buf;
a44c9a5e 3363 stream.avail_in = len;
9892beba 3364 do {
748af44c 3365 unsigned char *in0 = stream.next_in;
55bb5c91 3366 ret = git_deflate(&stream, Z_FINISH);
748af44c 3367 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
3368 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3369 die("unable to write sha1 file");
3370 stream.next_out = compressed;
3371 stream.avail_out = sizeof(compressed);
3372 } while (ret == Z_OK);
3373
ac54c277
LT
3374 if (ret != Z_STREAM_END)
3375 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
55bb5c91 3376 ret = git_deflate_end_gently(&stream);
ac54c277
LT
3377 if (ret != Z_OK)
3378 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
3379 git_SHA1_Final(parano_sha1, &c);
3380 if (hashcmp(sha1, parano_sha1) != 0)
3381 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 3382
e9039dd3 3383 close_sha1_file(fd);
0fcfd160 3384
bbac7311
NP
3385 if (mtime) {
3386 struct utimbuf utb;
3387 utb.actime = mtime;
3388 utb.modtime = mtime;
d4b3d11a 3389 if (utime(tmp_file.buf, &utb) < 0)
7616c6ca 3390 warning_errno("failed utime() on %s", tmp_file.buf);
bbac7311
NP
3391 }
3392
d4b3d11a 3393 return finalize_object_file(tmp_file.buf, filename);
0fcfd160 3394}
8237b185 3395
33d4221c
JK
3396static int freshen_loose_object(const unsigned char *sha1)
3397{
3398 return check_and_freshen(sha1, 1);
3399}
3400
3401static int freshen_packed_object(const unsigned char *sha1)
3402{
3403 struct pack_entry e;
ee1c6c34
JK
3404 if (!find_pack_entry(sha1, &e))
3405 return 0;
3406 if (e.p->freshened)
3407 return 1;
3408 if (!freshen_file(e.p->pack_name))
3409 return 0;
3410 e.p->freshened = 1;
3411 return 1;
33d4221c
JK
3412}
3413
1427a7ff 3414int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
bbac7311 3415{
bbac7311 3416 char hdr[32];
ef1286d3 3417 int hdrlen = sizeof(hdr);
bbac7311
NP
3418
3419 /* Normally if we have it in the pack then we do not bother writing
3420 * it out into .git/objects/??/?{38} file.
3421 */
3422 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
b5f52f37 3423 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
bbac7311
NP
3424 return 0;
3425 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3426}
3427
0c3db67c
ES
3428int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3429 unsigned char *sha1, unsigned flags)
3430{
3431 char *header;
3432 int hdrlen, status = 0;
3433
3434 /* type string, SP, %lu of the length plus NUL must fit this */
ef1286d3
JK
3435 hdrlen = strlen(type) + 32;
3436 header = xmalloc(hdrlen);
0c3db67c
ES
3437 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3438
3439 if (!(flags & HASH_WRITE_OBJECT))
3440 goto cleanup;
051086b9 3441 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
0c3db67c
ES
3442 goto cleanup;
3443 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3444
3445cleanup:
3446 free(header);
3447 return status;
3448}
3449
bbac7311
NP
3450int force_object_loose(const unsigned char *sha1, time_t mtime)
3451{
bbac7311
NP
3452 void *buf;
3453 unsigned long len;
3454 enum object_type type;
3455 char hdr[32];
3456 int hdrlen;
1fb23e65 3457 int ret;
bbac7311 3458
c529d75a 3459 if (has_loose_object(sha1))
bbac7311
NP
3460 return 0;
3461 buf = read_packed_sha1(sha1, &type, &len);
3462 if (!buf)
3463 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
ef1286d3 3464 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1fb23e65
BS
3465 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3466 free(buf);
3467
3468 return ret;
bbac7311
NP
3469}
3470
bf592c50
DB
3471int has_pack_index(const unsigned char *sha1)
3472{
3473 struct stat st;
3474 if (stat(sha1_pack_index_name(sha1), &st))
3475 return 0;
3476 return 1;
bf592c50
DB
3477}
3478
cd673c1f
JH
3479int has_sha1_pack(const unsigned char *sha1)
3480{
3481 struct pack_entry e;
3482 return find_pack_entry(sha1, &e);
3483}
3484
0eeb077b 3485int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
8237b185 3486{
3e8b7d3c
JN
3487 if (!startup_info->have_repository)
3488 return 0;
e83e71c5
JT
3489 return sha1_object_info_extended(sha1, NULL,
3490 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
b419aa25 3491}
3492
3493int has_object_file(const struct object_id *oid)
3494{
3495 return has_sha1_file(oid->hash);
8237b185 3496}
74400e71 3497
5827a035
JK
3498int has_object_file_with_flags(const struct object_id *oid, int flags)
3499{
3500 return has_sha1_file_with_flags(oid->hash, flags);
3501}
3502
c879daa2
NTND
3503static void check_tree(const void *buf, size_t size)
3504{
3505 struct tree_desc desc;
3506 struct name_entry entry;
3507
3508 init_tree_desc(&desc, buf, size);
3509 while (tree_entry(&desc, &entry))
3510 /* do nothing
3511 * tree_entry() will die() on malformed entries */
3512 ;
3513}
3514
3515static void check_commit(const void *buf, size_t size)
3516{
3517 struct commit c;
3518 memset(&c, 0, sizeof(c));
3519 if (parse_commit_buffer(&c, buf, size))
3520 die("corrupt commit");
3521}
3522
3523static void check_tag(const void *buf, size_t size)
3524{
3525 struct tag t;
3526 memset(&t, 0, sizeof(t));
3527 if (parse_tag_buffer(&t, buf, size))
3528 die("corrupt tag");
3529}
3530
43df4f86 3531static int index_mem(unsigned char *sha1, void *buf, size_t size,
c4ce46fc
JH
3532 enum object_type type,
3533 const char *path, unsigned flags)
e7332f96 3534{
6c510bee 3535 int ret, re_allocated = 0;
c4ce46fc 3536 int write_object = flags & HASH_WRITE_OBJECT;
74400e71 3537
7672db20 3538 if (!type)
edaec3fb 3539 type = OBJ_BLOB;
6c510bee
LT
3540
3541 /*
3542 * Convert blobs to git internal format
3543 */
43df4f86 3544 if ((type == OBJ_BLOB) && path) {
f285a2d7 3545 struct strbuf nbuf = STRBUF_INIT;
82b474e0 3546 if (convert_to_git(&the_index, path, buf, size, &nbuf,
5e12e78e 3547 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
b315c5c0 3548 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
3549 re_allocated = 1;
3550 }
3551 }
c4ce46fc 3552 if (flags & HASH_FORMAT_CHECK) {
c879daa2
NTND
3553 if (type == OBJ_TREE)
3554 check_tree(buf, size);
3555 if (type == OBJ_COMMIT)
3556 check_commit(buf, size);
3557 if (type == OBJ_TAG)
3558 check_tag(buf, size);
3559 }
6c510bee