t1450: refactor loose-object removal
[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"
6eac50d8 10#include "string-list.h"
697cc8ef 11#include "lockfile.h"
1f688557 12#include "delta.h"
a733cb60 13#include "pack.h"
8e440259
PE
14#include "blob.h"
15#include "commit.h"
4dd1fbc7 16#include "run-command.h"
8e440259
PE
17#include "tag.h"
18#include "tree.h"
c879daa2 19#include "tree-walk.h"
f35a6d3b 20#include "refs.h"
70f5d5d3 21#include "pack-revindex.h"
628522ec 22#include "sha1-lookup.h"
568508e7 23#include "bulk-checkin.h"
090ea126 24#include "streaming.h"
543c5caa 25#include "dir.h"
a73cdd21 26#include "mru.h"
12d95ef6 27#include "list.h"
c4c6effa 28#include "mergesort.h"
0fcfd160 29
144bde78
LT
30#ifndef O_NOATIME
31#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
32#define O_NOATIME 01000000
33#else
34#define O_NOATIME 0
35#endif
36#endif
37
28bd70d8
JN
38#define SZ_FMT PRIuMAX
39static inline uintmax_t sz_fmt(size_t s) { return s; }
e05db0fd 40
96f1e58f 41const unsigned char null_sha1[20];
3e56e724 42const struct object_id null_oid;
8576fde6
JK
43const struct object_id empty_tree_oid = {
44 EMPTY_TREE_SHA1_BIN_LITERAL
45};
46const struct object_id empty_blob_oid = {
47 EMPTY_BLOB_SHA1_BIN_LITERAL
48};
88cd621d 49
c597ba80
NTND
50/*
51 * This is meant to hold a *small* number of objects that you would
52 * want read_sha1_file() to be able to return, but yet you do not want
53 * to write them into the object store (e.g. a browse-only
54 * application).
55 */
56static struct cached_object {
57 unsigned char sha1[20];
58 enum object_type type;
59 void *buf;
60 unsigned long size;
61} *cached_objects;
62static int cached_object_nr, cached_object_alloc;
63
64static struct cached_object empty_tree = {
dab0d410 65 EMPTY_TREE_SHA1_BIN_LITERAL,
c597ba80
NTND
66 OBJ_TREE,
67 "",
68 0
69};
70
71static struct cached_object *find_cached_object(const unsigned char *sha1)
72{
73 int i;
74 struct cached_object *co = cached_objects;
75
76 for (i = 0; i < cached_object_nr; i++, co++) {
77 if (!hashcmp(co->sha1, sha1))
78 return co;
79 }
80 if (!hashcmp(sha1, empty_tree.sha1))
81 return &empty_tree;
82 return NULL;
83}
84
90a6464b
JH
85int mkdir_in_gitdir(const char *path)
86{
87 if (mkdir(path, 0777)) {
88 int saved_errno = errno;
89 struct stat st;
90 struct strbuf sb = STRBUF_INIT;
91
92 if (errno != EEXIST)
93 return -1;
94 /*
95 * Are we looking at a path in a symlinked worktree
96 * whose original repository does not yet have it?
97 * e.g. .git/rr-cache pointing at its original
98 * repository in which the user hasn't performed any
99 * conflict resolution yet?
100 */
101 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
102 strbuf_readlink(&sb, path, st.st_size) ||
103 !is_absolute_path(sb.buf) ||
104 mkdir(sb.buf, 0777)) {
105 strbuf_release(&sb);
106 errno = saved_errno;
107 return -1;
108 }
109 strbuf_release(&sb);
110 }
111 return adjust_shared_perm(path);
112}
113
0be0521b 114enum scld_error safe_create_leading_directories(char *path)
b2cb9425 115{
26c8ae2a 116 char *next_component = path + offset_1st_component(path);
0be0521b 117 enum scld_error ret = SCLD_OK;
67d42212 118
0be0521b 119 while (ret == SCLD_OK && next_component) {
f0502332 120 struct stat st;
0f527403 121 char *slash = next_component, slash_character;
f0502332 122
0f527403
MH
123 while (*slash && !is_dir_sep(*slash))
124 slash++;
125
126 if (!*slash)
b2cb9425 127 break;
bf10cf70 128
26c8ae2a 129 next_component = slash + 1;
0f527403 130 while (is_dir_sep(*next_component))
bf10cf70 131 next_component++;
26c8ae2a 132 if (!*next_component)
5f0bdf50 133 break;
831651fd 134
0f527403 135 slash_character = *slash;
831651fd 136 *slash = '\0';
67d42212
JR
137 if (!stat(path, &st)) {
138 /* path exists */
9e6f885d 139 if (!S_ISDIR(st.st_mode))
0be0521b 140 ret = SCLD_EXISTS;
53a39721 141 } else if (mkdir(path, 0777)) {
928734d9 142 if (errno == EEXIST &&
9e6f885d 143 !stat(path, &st) && S_ISDIR(st.st_mode))
928734d9 144 ; /* somebody created it since we checked */
18d37e86
MH
145 else if (errno == ENOENT)
146 /*
147 * Either mkdir() failed because
148 * somebody just pruned the containing
149 * directory, or stat() failed because
150 * the file that was in our way was
151 * just removed. Either way, inform
152 * the caller that it might be worth
153 * trying again:
154 */
155 ret = SCLD_VANISHED;
9e6f885d 156 else
0be0521b 157 ret = SCLD_FAILED;
53a39721 158 } else if (adjust_shared_perm(path)) {
0be0521b 159 ret = SCLD_PERMS;
457f06d6 160 }
0f527403 161 *slash = slash_character;
b2cb9425 162 }
9e6f885d 163 return ret;
b2cb9425 164}
723c31fe 165
0be0521b 166enum scld_error safe_create_leading_directories_const(const char *path)
8e21d63b
JK
167{
168 /* path points to cache entries, so xstrdup before messing with it */
169 char *buf = xstrdup(path);
0be0521b 170 enum scld_error result = safe_create_leading_directories(buf);
8e21d63b
JK
171 free(buf);
172 return result;
173}
174
f7b7774f 175static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
ace1534d
JH
176{
177 int i;
178 for (i = 0; i < 20; i++) {
179 static char hex[] = "0123456789abcdef";
180 unsigned int val = sha1[i];
f7b7774f
JK
181 strbuf_addch(buf, hex[val >> 4]);
182 strbuf_addch(buf, hex[val & 0xf]);
afbba2f0 183 if (!i)
f7b7774f 184 strbuf_addch(buf, '/');
ace1534d
JH
185 }
186}
187
30d6c6ea 188const char *sha1_file_name(const unsigned char *sha1)
0fcfd160 189{
f7b7774f 190 static struct strbuf buf = STRBUF_INIT;
0fcfd160 191
f7b7774f
JK
192 strbuf_reset(&buf);
193 strbuf_addf(&buf, "%s/", get_object_directory());
560fb6a1 194
f7b7774f
JK
195 fill_sha1_path(&buf, sha1);
196 return buf.buf;
0fcfd160
LT
197}
198
38dbe5f0
JK
199struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
200{
201 strbuf_setlen(&alt->scratch, alt->base_len);
202 return &alt->scratch;
203}
204
29ec6af2
JK
205static const char *alt_sha1_path(struct alternate_object_database *alt,
206 const unsigned char *sha1)
207{
38dbe5f0 208 struct strbuf *buf = alt_scratch_buf(alt);
f7b7774f 209 fill_sha1_path(buf, sha1);
38dbe5f0 210 return buf->buf;
0fcfd160
LT
211}
212
d40d535b
MH
213/*
214 * Return the name of the pack or index file with the specified sha1
215 * in its filename. *base and *name are scratch space that must be
216 * provided by the caller. which should be "pack" or "idx".
217 */
633f43e1 218static char *sha1_get_pack_name(const unsigned char *sha1,
ac5190cc
JK
219 struct strbuf *buf,
220 const char *which)
bf592c50 221{
ac5190cc
JK
222 strbuf_reset(buf);
223 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
224 sha1_to_hex(sha1), which);
225 return buf->buf;
bf592c50
DB
226}
227
633f43e1 228char *sha1_pack_name(const unsigned char *sha1)
bf592c50 229{
ac5190cc
JK
230 static struct strbuf buf = STRBUF_INIT;
231 return sha1_get_pack_name(sha1, &buf, "pack");
633f43e1 232}
bf592c50 233
633f43e1
HO
234char *sha1_pack_index_name(const unsigned char *sha1)
235{
ac5190cc
JK
236 static struct strbuf buf = STRBUF_INIT;
237 return sha1_get_pack_name(sha1, &buf, "idx");
bf592c50
DB
238}
239
d5a63b99
JH
240struct alternate_object_database *alt_odb_list;
241static struct alternate_object_database **alt_odb_tail;
ace1534d 242
4ea82473
JK
243/*
244 * Return non-zero iff the path is usable as an alternate object database.
245 */
246static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
247{
248 struct alternate_object_database *alt;
249
250 /* Detect cases where alternate disappeared */
251 if (!is_directory(path->buf)) {
252 error("object directory %s does not exist; "
253 "check .git/objects/info/alternates.",
254 path->buf);
255 return 0;
256 }
257
258 /*
259 * Prevent the common mistake of listing the same
260 * thing twice, or object directory itself.
261 */
262 for (alt = alt_odb_list; alt; alt = alt->next) {
ea0fc3b4 263 if (!fspathcmp(path->buf, alt->path))
4ea82473
JK
264 return 0;
265 }
266 if (!fspathcmp(path->buf, normalized_objdir))
267 return 0;
268
269 return 1;
270}
271
ddd5d056
JH
272/*
273 * Prepare alternate object database registry.
d5a63b99
JH
274 *
275 * The variable alt_odb_list points at the list of struct
276 * alternate_object_database. The elements on this list come from
277 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
278 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
279 * whose contents is similar to that environment variable but can be
280 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
281 * contains "/the/directory/corresponding/to/.git/objects/...", while
282 * its name points just after the slash at the end of ".git/objects/"
283 * in the example above, and has enough space to hold 40-byte hex
284 * SHA1, an extra slash for the first level indirection, and the
285 * terminating NUL.
ddd5d056 286 */
539e7506
EK
287static int link_alt_odb_entry(const char *entry, const char *relative_base,
288 int depth, const char *normalized_objdir)
ace1534d 289{
c2f493a4 290 struct alternate_object_database *ent;
5bdf0a84 291 struct strbuf pathbuf = STRBUF_INIT;
d5a63b99 292
85dadc38 293 if (!is_absolute_path(entry) && relative_base) {
5bdf0a84
HW
294 strbuf_addstr(&pathbuf, real_path(relative_base));
295 strbuf_addch(&pathbuf, '/');
c2f493a4 296 }
6eac50d8 297 strbuf_addstr(&pathbuf, entry);
c2f493a4 298
37a95862 299 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
670c359d
JK
300 error("unable to normalize alternate object path: %s",
301 pathbuf.buf);
302 strbuf_release(&pathbuf);
303 return -1;
304 }
5bdf0a84
HW
305
306 /*
307 * The trailing slash after the directory name is given by
308 * this function at the end. Remove duplicates.
309 */
4ea82473
JK
310 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
311 strbuf_setlen(&pathbuf, pathbuf.len - 1);
5bdf0a84 312
4ea82473
JK
313 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
314 strbuf_release(&pathbuf);
c2f493a4
MW
315 return -1;
316 }
317
7f0fa2c0 318 ent = alloc_alt_odb(pathbuf.buf);
c2f493a4
MW
319
320 /* add the alternate entry */
321 *alt_odb_tail = ent;
322 alt_odb_tail = &(ent->next);
323 ent->next = NULL;
324
325 /* recursively add alternates */
4ea82473 326 read_info_alternates(pathbuf.buf, depth + 1);
c2f493a4 327
4ea82473 328 strbuf_release(&pathbuf);
c2f493a4
MW
329 return 0;
330}
331
c5950164 332static void link_alt_odb_entries(const char *alt, int len, int sep,
c2f493a4
MW
333 const char *relative_base, int depth)
334{
6eac50d8
MH
335 struct string_list entries = STRING_LIST_INIT_NODUP;
336 char *alt_copy;
337 int i;
539e7506 338 struct strbuf objdirbuf = STRBUF_INIT;
c2f493a4
MW
339
340 if (depth > 5) {
341 error("%s: ignoring alternate object stores, nesting too deep.",
342 relative_base);
343 return;
344 }
345
f655651e 346 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
670c359d
JK
347 if (strbuf_normalize_path(&objdirbuf) < 0)
348 die("unable to normalize object directory: %s",
349 objdirbuf.buf);
539e7506 350
c5950164 351 alt_copy = xmemdupz(alt, len);
6eac50d8
MH
352 string_list_split_in_place(&entries, alt_copy, sep, -1);
353 for (i = 0; i < entries.nr; i++) {
354 const char *entry = entries.items[i].string;
355 if (entry[0] == '\0' || entry[0] == '#')
9577e7e3 356 continue;
087b6d58 357 link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
9577e7e3 358 }
6eac50d8
MH
359 string_list_clear(&entries, 0);
360 free(alt_copy);
539e7506 361 strbuf_release(&objdirbuf);
d5a63b99
JH
362}
363
5e73633d 364void read_info_alternates(const char * relative_base, int depth)
d5a63b99 365{
9577e7e3 366 char *map;
dc49cd76 367 size_t mapsz;
d5a63b99 368 struct stat st;
5015f01c 369 char *path;
c2f493a4 370 int fd;
d5a63b99 371
5015f01c 372 path = xstrfmt("%s/info/alternates", relative_base);
a5436b57 373 fd = git_open(path);
5015f01c 374 free(path);
d5a63b99
JH
375 if (fd < 0)
376 return;
377 if (fstat(fd, &st) || (st.st_size == 0)) {
378 close(fd);
9a217f2a 379 return;
ace1534d 380 }
dc49cd76
SP
381 mapsz = xsize_t(st.st_size);
382 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 383 close(fd);
d5a63b99 384
c5950164 385 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
c2f493a4 386
dc49cd76 387 munmap(map, mapsz);
ace1534d
JH
388}
389
7f0fa2c0
JK
390struct alternate_object_database *alloc_alt_odb(const char *dir)
391{
392 struct alternate_object_database *ent;
7f0fa2c0 393
597f9134 394 FLEX_ALLOC_STR(ent, path, dir);
38dbe5f0
JK
395 strbuf_init(&ent->scratch, 0);
396 strbuf_addf(&ent->scratch, "%s/", dir);
397 ent->base_len = ent->scratch.len;
7f0fa2c0
JK
398
399 return ent;
400}
401
bef70b22
DB
402void add_to_alternates_file(const char *reference)
403{
404 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
77b9b1d1
JK
405 char *alts = git_pathdup("objects/info/alternates");
406 FILE *in, *out;
407
408 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
409 out = fdopen_lock_file(lock, "w");
410 if (!out)
411 die_errno("unable to fdopen alternates lockfile");
412
413 in = fopen(alts, "r");
414 if (in) {
415 struct strbuf line = STRBUF_INIT;
416 int found = 0;
417
3f163962 418 while (strbuf_getline(&line, in) != EOF) {
77b9b1d1
JK
419 if (!strcmp(reference, line.buf)) {
420 found = 1;
421 break;
422 }
423 fprintf_or_die(out, "%s\n", line.buf);
424 }
425
426 strbuf_release(&line);
427 fclose(in);
428
429 if (found) {
430 rollback_lock_file(lock);
431 lock = NULL;
432 }
433 }
434 else if (errno != ENOENT)
435 die_errno("unable to read alternates file");
436
437 if (lock) {
438 fprintf_or_die(out, "%s\n", reference);
439 if (commit_lock_file(lock))
440 die_errno("unable to move new alternates file into place");
441 if (alt_odb_tail)
442 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
443 }
444 free(alts);
bef70b22
DB
445}
446
a5b34d21
JK
447void add_to_alternates_memory(const char *reference)
448{
449 /*
450 * Make sure alternates are initialized, or else our entry may be
451 * overwritten when they are.
452 */
453 prepare_alt_odb();
454
455 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
456}
457
9eeea7d2
SB
458/*
459 * Compute the exact path an alternate is at and returns it. In case of
460 * error NULL is returned and the human readable error is added to `err`
461 * `path` may be relative and should point to $GITDIR.
462 * `err` must not be null.
463 */
464char *compute_alternate_path(const char *path, struct strbuf *err)
465{
466 char *ref_git = NULL;
467 const char *repo, *ref_git_s;
468 int seen_error = 0;
469
470 ref_git_s = real_path_if_valid(path);
471 if (!ref_git_s) {
472 seen_error = 1;
473 strbuf_addf(err, _("path '%s' does not exist"), path);
474 goto out;
475 } else
476 /*
477 * Beware: read_gitfile(), real_path() and mkpath()
478 * return static buffer
479 */
480 ref_git = xstrdup(ref_git_s);
481
482 repo = read_gitfile(ref_git);
483 if (!repo)
484 repo = read_gitfile(mkpath("%s/.git", ref_git));
485 if (repo) {
486 free(ref_git);
487 ref_git = xstrdup(repo);
488 }
489
490 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
491 char *ref_git_git = mkpathdup("%s/.git", ref_git);
492 free(ref_git);
493 ref_git = ref_git_git;
494 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
495 struct strbuf sb = STRBUF_INIT;
496 seen_error = 1;
497 if (get_common_dir(&sb, ref_git)) {
498 strbuf_addf(err,
499 _("reference repository '%s' as a linked "
500 "checkout is not supported yet."),
501 path);
502 goto out;
503 }
504
505 strbuf_addf(err, _("reference repository '%s' is not a "
506 "local repository."), path);
507 goto out;
508 }
509
510 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
511 strbuf_addf(err, _("reference repository '%s' is shallow"),
512 path);
513 seen_error = 1;
514 goto out;
515 }
516
517 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
518 strbuf_addf(err,
519 _("reference repository '%s' is grafted"),
520 path);
521 seen_error = 1;
522 goto out;
523 }
524
525out:
526 if (seen_error) {
527 free(ref_git);
528 ref_git = NULL;
529 }
530
531 return ref_git;
532}
533
fe1b2268 534int foreach_alt_odb(alt_odb_fn fn, void *cb)
d79796bc
JH
535{
536 struct alternate_object_database *ent;
fe1b2268 537 int r = 0;
d79796bc
JH
538
539 prepare_alt_odb();
fe1b2268
JK
540 for (ent = alt_odb_list; ent; ent = ent->next) {
541 r = fn(ent, cb);
542 if (r)
543 break;
544 }
545 return r;
d79796bc
JH
546}
547
c2f493a4
MW
548void prepare_alt_odb(void)
549{
554fe20d 550 const char *alt;
c2f493a4 551
7dc24aa5
SP
552 if (alt_odb_tail)
553 return;
554
c2f493a4
MW
555 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
556 if (!alt) alt = "";
557
c2f493a4 558 alt_odb_tail = &alt_odb_list;
c5950164 559 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
560
561 read_info_alternates(get_object_directory(), 0);
562}
563
3096b2ec 564/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
33d4221c 565static int freshen_file(const char *fn)
ace1534d 566{
33d4221c
JK
567 struct utimbuf t;
568 t.actime = t.modtime = time(NULL);
569 return !utime(fn, &t);
0f4dc14a 570}
ace1534d 571
3096b2ec
JK
572/*
573 * All of the check_and_freshen functions return 1 if the file exists and was
574 * freshened (if freshening was requested), 0 otherwise. If they return
575 * 0, you should not assume that it is safe to skip a write of the object (it
576 * either does not exist on disk, or has a stale mtime and may be subject to
577 * pruning).
578 */
33d4221c
JK
579static int check_and_freshen_file(const char *fn, int freshen)
580{
581 if (access(fn, F_OK))
582 return 0;
3096b2ec 583 if (freshen && !freshen_file(fn))
33d4221c
JK
584 return 0;
585 return 1;
586}
587
588static int check_and_freshen_local(const unsigned char *sha1, int freshen)
589{
590 return check_and_freshen_file(sha1_file_name(sha1), freshen);
591}
592
593static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
0f4dc14a
BC
594{
595 struct alternate_object_database *alt;
9a217f2a 596 prepare_alt_odb();
d5a63b99 597 for (alt = alt_odb_list; alt; alt = alt->next) {
29ec6af2
JK
598 const char *path = alt_sha1_path(alt, sha1);
599 if (check_and_freshen_file(path, freshen))
c529d75a 600 return 1;
ace1534d 601 }
c529d75a 602 return 0;
ace1534d
JH
603}
604
33d4221c
JK
605static int check_and_freshen(const unsigned char *sha1, int freshen)
606{
607 return check_and_freshen_local(sha1, freshen) ||
608 check_and_freshen_nonlocal(sha1, freshen);
609}
610
611int has_loose_object_nonlocal(const unsigned char *sha1)
612{
613 return check_and_freshen_nonlocal(sha1, 0);
614}
615
0f4dc14a
BC
616static int has_loose_object(const unsigned char *sha1)
617{
33d4221c 618 return check_and_freshen(sha1, 0);
0f4dc14a
BC
619}
620
60bb8b14 621static unsigned int pack_used_ctr;
a53128b6
SP
622static unsigned int pack_mmap_calls;
623static unsigned int peak_pack_open_windows;
624static unsigned int pack_open_windows;
c7934306
SP
625static unsigned int pack_open_fds;
626static unsigned int pack_max_fds;
a53128b6 627static size_t peak_pack_mapped;
60bb8b14 628static size_t pack_mapped;
9a217f2a 629struct packed_git *packed_git;
1f688557 630
a73cdd21
JK
631static struct mru packed_git_mru_storage;
632struct mru *packed_git_mru = &packed_git_mru_storage;
633
b79d18c9 634void pack_report(void)
a53128b6
SP
635{
636 fprintf(stderr,
e05db0fd
PR
637 "pack_report: getpagesize() = %10" SZ_FMT "\n"
638 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
639 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
640 sz_fmt(getpagesize()),
641 sz_fmt(packed_git_window_size),
642 sz_fmt(packed_git_limit));
a53128b6
SP
643 fprintf(stderr,
644 "pack_report: pack_used_ctr = %10u\n"
645 "pack_report: pack_mmap_calls = %10u\n"
646 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
647 "pack_report: pack_mapped = "
648 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
649 pack_used_ctr,
650 pack_mmap_calls,
651 pack_open_windows, peak_pack_open_windows,
9e42d6a1 652 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
653}
654
d40d535b
MH
655/*
656 * Open and mmap the index file at path, perform a couple of
657 * consistency checks, then record its information to p. Return 0 on
658 * success.
659 */
660static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
661{
662 void *idx_map;
42873078 663 struct pack_idx_header *hdr;
dc49cd76 664 size_t idx_size;
74e34e1f 665 uint32_t version, nr, i, *index;
a5436b57 666 int fd = git_open(path);
1f688557 667 struct stat st;
42873078 668
1f688557
JH
669 if (fd < 0)
670 return -1;
671 if (fstat(fd, &st)) {
672 close(fd);
673 return -1;
674 }
dc49cd76 675 idx_size = xsize_t(st.st_size);
2d88451b
SP
676 if (idx_size < 4 * 256 + 20 + 20) {
677 close(fd);
678 return error("index file %s is too small", path);
679 }
c4712e45 680 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 681 close(fd);
1f688557 682
42873078
NP
683 hdr = idx_map;
684 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
685 version = ntohl(hdr->idx_version);
686 if (version < 2 || version > 2) {
687 munmap(idx_map, idx_size);
6e1c2344 688 return error("index file %s is version %"PRIu32
74e34e1f
NP
689 " and is not supported by this binary"
690 " (try upgrading GIT to a newer version)",
691 path, version);
692 }
693 } else
694 version = 1;
df1b059d 695
1f688557 696 nr = 0;
42873078 697 index = idx_map;
74e34e1f
NP
698 if (version > 1)
699 index += 2; /* skip index header */
1f688557 700 for (i = 0; i < 256; i++) {
326bf396 701 uint32_t n = ntohl(index[i]);
2d88451b
SP
702 if (n < nr) {
703 munmap(idx_map, idx_size);
df1b059d 704 return error("non-monotonic index %s", path);
2d88451b 705 }
1f688557
JH
706 nr = n;
707 }
708
74e34e1f
NP
709 if (version == 1) {
710 /*
711 * Total size:
712 * - 256 index entries 4 bytes each
713 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
714 * - 20-byte SHA1 of the packfile
715 * - 20-byte SHA1 file checksum
716 */
717 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
718 munmap(idx_map, idx_size);
eef427a0 719 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
720 }
721 } else if (version == 2) {
722 /*
723 * Minimum size:
724 * - 8 bytes of header
725 * - 256 index entries 4 bytes each
726 * - 20-byte sha1 entry * nr
727 * - 4-byte crc entry * nr
728 * - 4-byte offset entry * nr
729 * - 20-byte SHA1 of the packfile
730 * - 20-byte SHA1 file checksum
731 * And after the 4-byte offset table might be a
732 * variable sized table containing 8-byte entries
733 * for offsets larger than 2^31.
734 */
735 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
736 unsigned long max_size = min_size;
737 if (nr)
738 max_size += (nr - 1)*8;
739 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 740 munmap(idx_map, idx_size);
eef427a0 741 return error("wrong index v2 file size in %s", path);
74e34e1f 742 }
7109c889
JH
743 if (idx_size != min_size &&
744 /*
745 * make sure we can deal with large pack offsets.
746 * 31-bit signed offset won't be enough, neither
747 * 32-bit unsigned one will be.
748 */
749 (sizeof(off_t) <= 4)) {
750 munmap(idx_map, idx_size);
751 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 752 }
2d88451b 753 }
1f688557 754
74e34e1f 755 p->index_version = version;
42873078
NP
756 p->index_data = idx_map;
757 p->index_size = idx_size;
57059091 758 p->num_objects = nr;
1f688557
JH
759 return 0;
760}
761
bc8e478a 762int open_pack_index(struct packed_git *p)
d079837e
SP
763{
764 char *idx_name;
9ae97018 765 size_t len;
d079837e
SP
766 int ret;
767
768 if (p->index_data)
769 return 0;
770
9ae97018
JK
771 if (!strip_suffix(p->pack_name, ".pack", &len))
772 die("BUG: pack_name does not end in .pack");
773 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
d079837e
SP
774 ret = check_packed_git_idx(idx_name, p);
775 free(idx_name);
776 return ret;
777}
778
11daf39b
SP
779static void scan_windows(struct packed_git *p,
780 struct packed_git **lru_p,
781 struct pack_window **lru_w,
782 struct pack_window **lru_l)
1f688557 783{
11daf39b
SP
784 struct pack_window *w, *w_l;
785
786 for (w_l = NULL, w = p->windows; w; w = w->next) {
787 if (!w->inuse_cnt) {
788 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
789 *lru_p = p;
790 *lru_w = w;
791 *lru_l = w_l;
54044bf8 792 }
54044bf8 793 }
11daf39b 794 w_l = w;
f9253394 795 }
11daf39b
SP
796}
797
7c3ecb32 798static int unuse_one_window(struct packed_git *current)
11daf39b
SP
799{
800 struct packed_git *p, *lru_p = NULL;
801 struct pack_window *lru_w = NULL, *lru_l = NULL;
802
803 if (current)
804 scan_windows(current, &lru_p, &lru_w, &lru_l);
805 for (p = packed_git; p; p = p->next)
806 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
807 if (lru_p) {
808 munmap(lru_w->base, lru_w->len);
809 pack_mapped -= lru_w->len;
810 if (lru_l)
811 lru_l->next = lru_w->next;
7c3ecb32 812 else
54044bf8 813 lru_p->windows = lru_w->next;
54044bf8 814 free(lru_w);
a53128b6 815 pack_open_windows--;
54044bf8
SP
816 return 1;
817 }
818 return 0;
f9253394
JH
819}
820
7c3ecb32 821void release_pack_memory(size_t need)
97bfeb34
SP
822{
823 size_t cur = pack_mapped;
7c3ecb32 824 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
97bfeb34
SP
825 ; /* nothing */
826}
827
02710228
SP
828static void mmap_limit_check(size_t length)
829{
830 static size_t limit = 0;
831 if (!limit) {
832 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
833 if (!limit)
834 limit = SIZE_MAX;
835 }
836 if (length > limit)
837 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
838 (uintmax_t)length, (uintmax_t)limit);
839}
840
1570856b
JK
841void *xmmap_gently(void *start, size_t length,
842 int prot, int flags, int fd, off_t offset)
58ecbd5e 843{
02710228
SP
844 void *ret;
845
846 mmap_limit_check(length);
847 ret = mmap(start, length, prot, flags, fd, offset);
58ecbd5e
JN
848 if (ret == MAP_FAILED) {
849 if (!length)
850 return NULL;
7c3ecb32 851 release_pack_memory(length);
58ecbd5e 852 ret = mmap(start, length, prot, flags, fd, offset);
58ecbd5e
JN
853 }
854 return ret;
855}
856
1570856b
JK
857void *xmmap(void *start, size_t length,
858 int prot, int flags, int fd, off_t offset)
859{
860 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
861 if (ret == MAP_FAILED)
9ca0aaf6 862 die_errno("mmap failed");
1570856b
JK
863 return ret;
864}
865
c9ced051
SP
866void close_pack_windows(struct packed_git *p)
867{
868 while (p->windows) {
869 struct pack_window *w = p->windows;
870
871 if (w->inuse_cnt)
872 die("pack '%s' still has open windows to it",
873 p->pack_name);
874 munmap(w->base, w->len);
875 pack_mapped -= w->len;
876 pack_open_windows--;
877 p->windows = w->next;
878 free(w);
879 }
880}
881
71fe5d7f
JS
882static int close_pack_fd(struct packed_git *p)
883{
884 if (p->pack_fd < 0)
885 return 0;
886
887 close(p->pack_fd);
888 pack_open_fds--;
889 p->pack_fd = -1;
890
891 return 1;
892}
893
38849a81
JS
894static void close_pack(struct packed_git *p)
895{
896 close_pack_windows(p);
897 close_pack_fd(p);
898 close_pack_index(p);
899}
900
901void close_all_packs(void)
902{
903 struct packed_git *p;
904
905 for (p = packed_git; p; p = p->next)
906 if (p->do_not_close)
ef1177d1 907 die("BUG: want to close pack marked 'do-not-close'");
38849a81
JS
908 else
909 close_pack(p);
910}
911
912
88d0db55
BC
913/*
914 * The LRU pack is the one with the oldest MRU window, preferring packs
915 * with no used windows, or the oldest mtime if it has no windows allocated.
916 */
917static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
918{
919 struct pack_window *w, *this_mru_w;
920 int has_windows_inuse = 0;
921
922 /*
923 * Reject this pack if it has windows and the previously selected
924 * one does not. If this pack does not have windows, reject
925 * it if the pack file is newer than the previously selected one.
926 */
927 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
928 return;
929
930 for (w = this_mru_w = p->windows; w; w = w->next) {
931 /*
932 * Reject this pack if any of its windows are in use,
933 * but the previously selected pack did not have any
934 * inuse windows. Otherwise, record that this pack
935 * has windows in use.
936 */
937 if (w->inuse_cnt) {
938 if (*accept_windows_inuse)
939 has_windows_inuse = 1;
940 else
941 return;
942 }
943
944 if (w->last_used > this_mru_w->last_used)
945 this_mru_w = w;
946
947 /*
948 * Reject this pack if it has windows that have been
949 * used more recently than the previously selected pack.
950 * If the previously selected pack had windows inuse and
951 * we have not encountered a window in this pack that is
952 * inuse, skip this check since we prefer a pack with no
953 * inuse windows to one that has inuse windows.
954 */
955 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
956 this_mru_w->last_used > (*mru_w)->last_used)
957 return;
958 }
959
960 /*
961 * Select this pack.
962 */
963 *mru_w = this_mru_w;
964 *lru_p = p;
965 *accept_windows_inuse = has_windows_inuse;
966}
967
968static int close_one_pack(void)
969{
970 struct packed_git *p, *lru_p = NULL;
971 struct pack_window *mru_w = NULL;
972 int accept_windows_inuse = 1;
973
974 for (p = packed_git; p; p = p->next) {
975 if (p->pack_fd == -1)
976 continue;
977 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
978 }
979
71fe5d7f
JS
980 if (lru_p)
981 return close_pack_fd(lru_p);
88d0db55
BC
982
983 return 0;
984}
985
03e79c88 986void unuse_pack(struct pack_window **w_cursor)
f9253394 987{
03e79c88
SP
988 struct pack_window *w = *w_cursor;
989 if (w) {
990 w->inuse_cnt--;
991 *w_cursor = NULL;
992 }
1f688557
JH
993}
994
fa5fc15d
SP
995void close_pack_index(struct packed_git *p)
996{
997 if (p->index_data) {
998 munmap((void *)p->index_data, p->index_size);
999 p->index_data = NULL;
1000 }
1001}
1002
a0788266
JS
1003static unsigned int get_max_fd_limit(void)
1004{
1005#ifdef RLIMIT_NOFILE
491a8dec
JH
1006 {
1007 struct rlimit lim;
a0788266 1008
491a8dec
JH
1009 if (!getrlimit(RLIMIT_NOFILE, &lim))
1010 return lim.rlim_cur;
1011 }
1012#endif
1013
1014#ifdef _SC_OPEN_MAX
1015 {
1016 long open_max = sysconf(_SC_OPEN_MAX);
1017 if (0 < open_max)
1018 return open_max;
1019 /*
1020 * Otherwise, we got -1 for one of the two
1021 * reasons:
1022 *
1023 * (1) sysconf() did not understand _SC_OPEN_MAX
1024 * and signaled an error with -1; or
1025 * (2) sysconf() said there is no limit.
1026 *
1027 * We _could_ clear errno before calling sysconf() to
1028 * tell these two cases apart and return a huge number
1029 * in the latter case to let the caller cap it to a
1030 * value that is not so selfish, but letting the
1031 * fallback OPEN_MAX codepath take care of these cases
1032 * is a lot simpler.
1033 */
1034 }
1035#endif
a0788266 1036
491a8dec 1037#ifdef OPEN_MAX
a0788266
JS
1038 return OPEN_MAX;
1039#else
1040 return 1; /* see the caller ;-) */
1041#endif
1042}
1043
3cf8b462
SP
1044/*
1045 * Do not call this directly as this leaks p->pack_fd on error return;
1046 * call open_packed_git() instead.
1047 */
1048static int open_packed_git_1(struct packed_git *p)
1f688557 1049{
9bc879c1
SP
1050 struct stat st;
1051 struct pack_header hdr;
1052 unsigned char sha1[20];
1053 unsigned char *idx_sha1;
2c039da8 1054 long fd_flag;
9bc879c1 1055
d079837e
SP
1056 if (!p->index_data && open_pack_index(p))
1057 return error("packfile %s index unavailable", p->pack_name);
1058
c7934306 1059 if (!pack_max_fds) {
a0788266 1060 unsigned int max_fds = get_max_fd_limit();
c7934306
SP
1061
1062 /* Save 3 for stdin/stdout/stderr, 22 for work */
1063 if (25 < max_fds)
1064 pack_max_fds = max_fds - 25;
1065 else
1066 pack_max_fds = 1;
1067 }
1068
88d0db55 1069 while (pack_max_fds <= pack_open_fds && close_one_pack())
c7934306
SP
1070 ; /* nothing */
1071
a5436b57 1072 p->pack_fd = git_open(p->pack_name);
9bc879c1 1073 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 1074 return -1;
c7934306 1075 pack_open_fds++;
9bc879c1
SP
1076
1077 /* If we created the struct before we had the pack we lack size. */
bf592c50 1078 if (!p->pack_size) {
bf592c50 1079 if (!S_ISREG(st.st_mode))
072db278 1080 return error("packfile %s not a regular file", p->pack_name);
bf592c50 1081 p->pack_size = st.st_size;
9bc879c1 1082 } else if (p->pack_size != st.st_size)
072db278 1083 return error("packfile %s size changed", p->pack_name);
9bc879c1 1084
2c039da8
JH
1085 /* We leave these file descriptors open with sliding mmap;
1086 * there is no point keeping them open across exec(), though.
1087 */
1088 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
1089 if (fd_flag < 0)
072db278 1090 return error("cannot determine file descriptor flags");
2c039da8
JH
1091 fd_flag |= FD_CLOEXEC;
1092 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 1093 return error("cannot set FD_CLOEXEC");
2c039da8 1094
9bc879c1 1095 /* Verify we recognize this pack file format. */
e6e2bd62 1096 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 1097 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 1098 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 1099 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 1100 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
1101 return error("packfile %s is version %"PRIu32" and not"
1102 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
1103 p->pack_name, ntohl(hdr.hdr_version));
1104
1105 /* Verify the pack matches its index. */
57059091 1106 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
1107 return error("packfile %s claims to have %"PRIu32" objects"
1108 " while index indicates %"PRIu32" objects",
57059091
NP
1109 p->pack_name, ntohl(hdr.hdr_entries),
1110 p->num_objects);
9bc879c1 1111 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 1112 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 1113 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 1114 return error("packfile %s signature is unavailable", p->pack_name);
42873078 1115 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 1116 if (hashcmp(sha1, idx_sha1))
072db278
SP
1117 return error("packfile %s does not match index", p->pack_name);
1118 return 0;
9bc879c1
SP
1119}
1120
3cf8b462
SP
1121static int open_packed_git(struct packed_git *p)
1122{
1123 if (!open_packed_git_1(p))
1124 return 0;
71fe5d7f 1125 close_pack_fd(p);
3cf8b462
SP
1126 return -1;
1127}
1128
c4001d92 1129static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
1130{
1131 /* We must promise at least 20 bytes (one hash) after the
1132 * offset is available from this window, otherwise the offset
1133 * is not actually in this window and a different window (which
1134 * has that one hash excess) must be used. This is to support
1135 * the object header and delta base parsing routines below.
1136 */
1137 off_t win_off = win->offset;
1138 return win_off <= offset
1139 && (offset + 20) <= (win_off + win->len);
1140}
1141
4b25d091 1142unsigned char *use_pack(struct packed_git *p,
03e79c88 1143 struct pack_window **w_cursor,
c4001d92 1144 off_t offset,
ef49a7a0 1145 unsigned long *left)
9bc879c1 1146{
60bb8b14 1147 struct pack_window *win = *w_cursor;
03e79c88 1148
a9d98a14 1149 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
1150 * pointless to ask for an offset into the middle of that
1151 * hash, and the in_window function above wouldn't match
1152 * don't allow an offset too close to the end of the file.
1153 */
d131b7af
SP
1154 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1155 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
1156 if (offset > (p->pack_size - 20))
1157 die("offset beyond end of packfile (truncated pack?)");
13e0b0d3 1158 if (offset < 0)
7465feba 1159 die(_("offset before end of packfile (broken .idx?)"));
60bb8b14
SP
1160
1161 if (!win || !in_window(win, offset)) {
1162 if (win)
1163 win->inuse_cnt--;
1164 for (win = p->windows; win; win = win->next) {
1165 if (in_window(win, offset))
1166 break;
1167 }
1168 if (!win) {
78a28df9 1169 size_t window_align = packed_git_window_size / 2;
dc49cd76 1170 off_t len;
d131b7af
SP
1171
1172 if (p->pack_fd == -1 && open_packed_git(p))
1173 die("packfile %s cannot be accessed", p->pack_name);
1174
60bb8b14 1175 win = xcalloc(1, sizeof(*win));
78a28df9 1176 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
1177 len = p->pack_size - win->offset;
1178 if (len > packed_git_window_size)
1179 len = packed_git_window_size;
1180 win->len = (size_t)len;
60bb8b14 1181 pack_mapped += win->len;
11daf39b 1182 while (packed_git_limit < pack_mapped
7c3ecb32 1183 && unuse_one_window(p))
60bb8b14 1184 ; /* nothing */
c4712e45 1185 win->base = xmmap(NULL, win->len,
60bb8b14
SP
1186 PROT_READ, MAP_PRIVATE,
1187 p->pack_fd, win->offset);
1188 if (win->base == MAP_FAILED)
7616c6ca
NTND
1189 die_errno("packfile %s cannot be mapped",
1190 p->pack_name);
d131b7af 1191 if (!win->offset && win->len == p->pack_size
71fe5d7f
JS
1192 && !p->do_not_close)
1193 close_pack_fd(p);
a53128b6
SP
1194 pack_mmap_calls++;
1195 pack_open_windows++;
1196 if (pack_mapped > peak_pack_mapped)
1197 peak_pack_mapped = pack_mapped;
1198 if (pack_open_windows > peak_pack_open_windows)
1199 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
1200 win->next = p->windows;
1201 p->windows = win;
1202 }
1f688557 1203 }
03e79c88
SP
1204 if (win != *w_cursor) {
1205 win->last_used = pack_used_ctr++;
1206 win->inuse_cnt++;
1207 *w_cursor = win;
1208 }
60bb8b14 1209 offset -= win->offset;
03e79c88 1210 if (left)
dc49cd76 1211 *left = win->len - xsize_t(offset);
03e79c88 1212 return win->base + offset;
1f688557
JH
1213}
1214
27d69a46
NP
1215static struct packed_git *alloc_packed_git(int extra)
1216{
50a6c8ef 1217 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
27d69a46
NP
1218 memset(p, 0, sizeof(*p));
1219 p->pack_fd = -1;
1220 return p;
1221}
1222
e0500293
JN
1223static void try_to_free_pack_memory(size_t size)
1224{
7c3ecb32 1225 release_pack_memory(size);
e0500293
JN
1226}
1227
48bcc1c3 1228struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
1f688557 1229{
e0500293 1230 static int have_set_try_to_free_routine;
1f688557 1231 struct stat st;
48bcc1c3
JK
1232 size_t alloc;
1233 struct packed_git *p;
1f688557 1234
e0500293
JN
1235 if (!have_set_try_to_free_routine) {
1236 have_set_try_to_free_routine = 1;
1237 set_try_to_free_routine(try_to_free_pack_memory);
1238 }
1239
42873078
NP
1240 /*
1241 * Make sure a corresponding .pack file exists and that
1242 * the index looks sane.
1243 */
48bcc1c3 1244 if (!strip_suffix_mem(path, &path_len, ".idx"))
1f688557 1245 return NULL;
48bcc1c3
JK
1246
1247 /*
1248 * ".pack" is long enough to hold any suffix we're adding (and
1249 * the use xsnprintf double-checks that)
1250 */
50a6c8ef 1251 alloc = st_add3(path_len, strlen(".pack"), 1);
48bcc1c3 1252 p = alloc_packed_git(alloc);
42873078 1253 memcpy(p->pack_name, path, path_len);
8d25931d 1254
48bcc1c3 1255 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
8d25931d
BC
1256 if (!access(p->pack_name, F_OK))
1257 p->pack_keep = 1;
1258
48bcc1c3 1259 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
d079837e 1260 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 1261 free(p);
1f688557
JH
1262 return NULL;
1263 }
42873078 1264
1f688557
JH
1265 /* ok, it looks sane as far as we can check without
1266 * actually mapping the pack file.
1267 */
1f688557 1268 p->pack_size = st.st_size;
9d835df2 1269 p->pack_local = local;
b867092f 1270 p->mtime = st.st_mtime;
42873078
NP
1271 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1272 hashclr(p->sha1);
1f688557
JH
1273 return p;
1274}
1275
7b64469a 1276struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 1277{
42873078 1278 const char *path = sha1_pack_name(sha1);
50a6c8ef 1279 size_t alloc = st_add(strlen(path), 1);
c7ab0ba3 1280 struct packed_git *p = alloc_packed_git(alloc);
bf592c50 1281
c7ab0ba3 1282 memcpy(p->pack_name, path, alloc); /* includes NUL */
27d69a46 1283 hashcpy(p->sha1, sha1);
42873078
NP
1284 if (check_packed_git_idx(idx_path, p)) {
1285 free(p);
bf592c50 1286 return NULL;
42873078 1287 }
bf592c50 1288
bf592c50
DB
1289 return p;
1290}
1291
1292void install_packed_git(struct packed_git *pack)
1293{
c7934306
SP
1294 if (pack->pack_fd != -1)
1295 pack_open_fds++;
1296
bf592c50
DB
1297 pack->next = packed_git;
1298 packed_git = pack;
1299}
1300
0a489b06 1301void (*report_garbage)(unsigned seen_bits, const char *path);
543c5caa
NTND
1302
1303static void report_helper(const struct string_list *list,
1304 int seen_bits, int first, int last)
1305{
0a489b06 1306 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
543c5caa 1307 return;
0a489b06 1308
543c5caa 1309 for (; first < last; first++)
0a489b06 1310 report_garbage(seen_bits, list->items[first].string);
543c5caa
NTND
1311}
1312
1313static void report_pack_garbage(struct string_list *list)
1314{
1315 int i, baselen = -1, first = 0, seen_bits = 0;
1316
1317 if (!report_garbage)
1318 return;
1319
3383e199 1320 string_list_sort(list);
543c5caa
NTND
1321
1322 for (i = 0; i < list->nr; i++) {
1323 const char *path = list->items[i].string;
1324 if (baselen != -1 &&
1325 strncmp(path, list->items[first].string, baselen)) {
1326 report_helper(list, seen_bits, first, i);
1327 baselen = -1;
1328 seen_bits = 0;
1329 }
1330 if (baselen == -1) {
1331 const char *dot = strrchr(path, '.');
1332 if (!dot) {
0a489b06 1333 report_garbage(PACKDIR_FILE_GARBAGE, path);
543c5caa
NTND
1334 continue;
1335 }
1336 baselen = dot - path + 1;
1337 first = i;
1338 }
1339 if (!strcmp(path + baselen, "pack"))
1340 seen_bits |= 1;
1341 else if (!strcmp(path + baselen, "idx"))
1342 seen_bits |= 2;
1343 }
1344 report_helper(list, seen_bits, first, list->nr);
1345}
1346
9d835df2 1347static void prepare_packed_git_one(char *objdir, int local)
1f688557 1348{
880fb8de
RS
1349 struct strbuf path = STRBUF_INIT;
1350 size_t dirnamelen;
1f688557
JH
1351 DIR *dir;
1352 struct dirent *de;
543c5caa 1353 struct string_list garbage = STRING_LIST_INIT_DUP;
1f688557 1354
880fb8de
RS
1355 strbuf_addstr(&path, objdir);
1356 strbuf_addstr(&path, "/pack");
1357 dir = opendir(path.buf);
b5b16990 1358 if (!dir) {
26125f6b 1359 if (errno != ENOENT)
7616c6ca
NTND
1360 error_errno("unable to open object pack directory: %s",
1361 path.buf);
880fb8de 1362 strbuf_release(&path);
1f688557 1363 return;
b5b16990 1364 }
880fb8de
RS
1365 strbuf_addch(&path, '/');
1366 dirnamelen = path.len;
1f688557 1367 while ((de = readdir(dir)) != NULL) {
1f688557 1368 struct packed_git *p;
47bf4b0f 1369 size_t base_len;
543c5caa
NTND
1370
1371 if (is_dot_or_dotdot(de->d_name))
9cb18f56
JM
1372 continue;
1373
880fb8de
RS
1374 strbuf_setlen(&path, dirnamelen);
1375 strbuf_addstr(&path, de->d_name);
d90906a9 1376
47bf4b0f
JK
1377 base_len = path.len;
1378 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
d90906a9
NTND
1379 /* Don't reopen a pack we already have. */
1380 for (p = packed_git; p; p = p->next) {
47bf4b0f
JK
1381 size_t len;
1382 if (strip_suffix(p->pack_name, ".pack", &len) &&
1383 len == base_len &&
1384 !memcmp(p->pack_name, path.buf, len))
d90906a9
NTND
1385 break;
1386 }
1387 if (p == NULL &&
1388 /*
1389 * See if it really is a valid .idx file with
1390 * corresponding .pack file that we can map.
1391 */
880fb8de 1392 (p = add_packed_git(path.buf, path.len, local)) != NULL)
d90906a9 1393 install_packed_git(p);
86f7780c 1394 }
543c5caa
NTND
1395
1396 if (!report_garbage)
1397 continue;
1398
2975c770
JK
1399 if (ends_with(de->d_name, ".idx") ||
1400 ends_with(de->d_name, ".pack") ||
1401 ends_with(de->d_name, ".bitmap") ||
1402 ends_with(de->d_name, ".keep"))
880fb8de 1403 string_list_append(&garbage, path.buf);
543c5caa 1404 else
0a489b06 1405 report_garbage(PACKDIR_FILE_GARBAGE, path.buf);
1f688557 1406 }
5b35bcd5 1407 closedir(dir);
543c5caa
NTND
1408 report_pack_garbage(&garbage);
1409 string_list_clear(&garbage, 0);
880fb8de 1410 strbuf_release(&path);
1f688557
JH
1411}
1412
8e3f52d7
JK
1413static int approximate_object_count_valid;
1414
1415/*
1416 * Give a fast, rough count of the number of objects in the repository. This
1417 * ignores loose objects completely. If you have a lot of them, then either
1418 * you should repack because your performance will be awful, or they are
1419 * all unreachable objects about to be pruned, in which case they're not really
1420 * interesting as a measure of repo size in the first place.
1421 */
1422unsigned long approximate_object_count(void)
1423{
1424 static unsigned long count;
1425 if (!approximate_object_count_valid) {
1426 struct packed_git *p;
1427
1428 prepare_packed_git();
1429 count = 0;
1430 for (p = packed_git; p; p = p->next) {
1431 if (open_pack_index(p))
1432 continue;
1433 count += p->num_objects;
1434 }
1435 }
1436 return count;
1437}
1438
c4c6effa
RS
1439static void *get_next_packed_git(const void *p)
1440{
1441 return ((const struct packed_git *)p)->next;
1442}
1443
1444static void set_next_packed_git(void *p, void *next)
1445{
1446 ((struct packed_git *)p)->next = next;
1447}
1448
b867092f
JH
1449static int sort_pack(const void *a_, const void *b_)
1450{
c4c6effa
RS
1451 const struct packed_git *a = a_;
1452 const struct packed_git *b = b_;
b867092f
JH
1453 int st;
1454
1455 /*
1456 * Local packs tend to contain objects specific to our
1457 * variant of the project than remote ones. In addition,
1458 * remote ones could be on a network mounted filesystem.
1459 * Favor local ones for these reasons.
1460 */
1461 st = a->pack_local - b->pack_local;
1462 if (st)
1463 return -st;
1464
1465 /*
1466 * Younger packs tend to contain more recent objects,
1467 * and more recent objects tend to get accessed more
1468 * often.
1469 */
1470 if (a->mtime < b->mtime)
1471 return 1;
1472 else if (a->mtime == b->mtime)
1473 return 0;
1474 return -1;
1475}
1476
1477static void rearrange_packed_git(void)
1478{
c4c6effa
RS
1479 packed_git = llist_mergesort(packed_git, get_next_packed_git,
1480 set_next_packed_git, sort_pack);
b867092f
JH
1481}
1482
a73cdd21
JK
1483static void prepare_packed_git_mru(void)
1484{
1485 struct packed_git *p;
1486
1487 mru_clear(packed_git_mru);
1488 for (p = packed_git; p; p = p->next)
1489 mru_append(packed_git_mru, p);
1490}
1491
637cdd9d 1492static int prepare_packed_git_run_once = 0;
9a217f2a 1493void prepare_packed_git(void)
1f688557 1494{
d5a63b99 1495 struct alternate_object_database *alt;
1f688557 1496
637cdd9d 1497 if (prepare_packed_git_run_once)
1f688557 1498 return;
9d835df2 1499 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 1500 prepare_alt_odb();
597f9134
JK
1501 for (alt = alt_odb_list; alt; alt = alt->next)
1502 prepare_packed_git_one(alt->path, 0);
b867092f 1503 rearrange_packed_git();
a73cdd21 1504 prepare_packed_git_mru();
637cdd9d
JK
1505 prepare_packed_git_run_once = 1;
1506}
1507
fc04c412 1508void reprepare_packed_git(void)
637cdd9d 1509{
8e3f52d7 1510 approximate_object_count_valid = 0;
637cdd9d
JK
1511 prepare_packed_git_run_once = 0;
1512 prepare_packed_git();
1f688557
JH
1513}
1514
8eca0b47
NP
1515static void mark_bad_packed_object(struct packed_git *p,
1516 const unsigned char *sha1)
1517{
1518 unsigned i;
1519 for (i = 0; i < p->num_bad_objects; i++)
50a6c8ef 1520 if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
8eca0b47 1521 return;
50a6c8ef
JK
1522 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1523 st_mult(GIT_SHA1_RAWSZ,
1524 st_add(p->num_bad_objects, 1)));
1525 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
8eca0b47
NP
1526 p->num_bad_objects++;
1527}
1528
b6c4cecc 1529static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
ac939109
NP
1530{
1531 struct packed_git *p;
1532 unsigned i;
1533
1534 for (p = packed_git; p; p = p->next)
1535 for (i = 0; i < p->num_bad_objects; i++)
1536 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
b6c4cecc
JH
1537 return p;
1538 return NULL;
ac939109
NP
1539}
1540
090ea126
NTND
1541/*
1542 * With an in-core object data in "map", rehash it to make sure the
1543 * object name actually matches "sha1" to detect object corruption.
1544 * With "map" == NULL, try reading the object named with "sha1" using
1545 * the streaming interface and rehash it to do the same.
1546 */
1547int check_sha1_signature(const unsigned char *sha1, void *map,
1548 unsigned long size, const char *type)
0fcfd160
LT
1549{
1550 unsigned char real_sha1[20];
090ea126
NTND
1551 enum object_type obj_type;
1552 struct git_istream *st;
1553 git_SHA_CTX c;
1554 char hdr[32];
1555 int hdrlen;
1556
1557 if (map) {
1558 hash_sha1_file(map, size, type, real_sha1);
1559 return hashcmp(sha1, real_sha1) ? -1 : 0;
1560 }
1561
1562 st = open_istream(sha1, &obj_type, &size, NULL);
1563 if (!st)
1564 return -1;
1565
1566 /* Generate the header */
ef1286d3 1567 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
090ea126
NTND
1568
1569 /* Sha1.. */
1570 git_SHA1_Init(&c);
1571 git_SHA1_Update(&c, hdr, hdrlen);
1572 for (;;) {
1573 char buf[1024 * 16];
1574 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1575
f54fac53
JK
1576 if (readlen < 0) {
1577 close_istream(st);
1578 return -1;
1579 }
090ea126
NTND
1580 if (!readlen)
1581 break;
1582 git_SHA1_Update(&c, buf, readlen);
1583 }
1584 git_SHA1_Final(real_sha1, &c);
1585 close_istream(st);
a89fccd2 1586 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1587}
1588
a5436b57 1589int git_open(const char *name)
44d1c19e 1590{
cd66ada0 1591 static int sha1_file_open_flag = O_NOATIME | O_CLOEXEC;
44d1c19e 1592
f2e872aa 1593 for (;;) {
dff6f280
CB
1594 int fd;
1595
1596 errno = 0;
1597 fd = open(name, O_RDONLY | sha1_file_open_flag);
44d1c19e 1598 if (fd >= 0)
f2e872aa
SP
1599 return fd;
1600
cd66ada0
LS
1601 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1602 if ((sha1_file_open_flag & O_CLOEXEC) && errno == EINVAL) {
1603 sha1_file_open_flag &= ~O_CLOEXEC;
f2e872aa
SP
1604 continue;
1605 }
1606
cd66ada0
LS
1607 /* Might the failure be due to O_NOATIME? */
1608 if (errno != ENOENT && (sha1_file_open_flag & O_NOATIME)) {
1609 sha1_file_open_flag &= ~O_NOATIME;
1610 continue;
1611 }
f2e872aa 1612 return -1;
44d1c19e 1613 }
44d1c19e
LT
1614}
1615
052fe5ea
JK
1616static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1617{
052fe5ea
JK
1618 struct alternate_object_database *alt;
1619
30d6c6ea 1620 if (!lstat(sha1_file_name(sha1), st))
052fe5ea
JK
1621 return 0;
1622
1623 prepare_alt_odb();
1624 errno = ENOENT;
1625 for (alt = alt_odb_list; alt; alt = alt->next) {
29ec6af2
JK
1626 const char *path = alt_sha1_path(alt, sha1);
1627 if (!lstat(path, st))
052fe5ea
JK
1628 return 0;
1629 }
1630
1631 return -1;
1632}
1633
44d1c19e
LT
1634static int open_sha1_file(const unsigned char *sha1)
1635{
1636 int fd;
44d1c19e 1637 struct alternate_object_database *alt;
d6c8a05b 1638 int most_interesting_errno;
44d1c19e 1639
a5436b57 1640 fd = git_open(sha1_file_name(sha1));
44d1c19e
LT
1641 if (fd >= 0)
1642 return fd;
d6c8a05b 1643 most_interesting_errno = errno;
44d1c19e
LT
1644
1645 prepare_alt_odb();
44d1c19e 1646 for (alt = alt_odb_list; alt; alt = alt->next) {
29ec6af2 1647 const char *path = alt_sha1_path(alt, sha1);
a5436b57 1648 fd = git_open(path);
44d1c19e
LT
1649 if (fd >= 0)
1650 return fd;
d6c8a05b
JK
1651 if (most_interesting_errno == ENOENT)
1652 most_interesting_errno = errno;
44d1c19e 1653 }
d6c8a05b 1654 errno = most_interesting_errno;
44d1c19e
LT
1655 return -1;
1656}
1657
f0270efd 1658void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1659{
0fcfd160 1660 void *map;
144bde78 1661 int fd;
ace1534d 1662
44d1c19e
LT
1663 fd = open_sha1_file(sha1);
1664 map = NULL;
1665 if (fd >= 0) {
1666 struct stat st;
0fcfd160 1667
44d1c19e
LT
1668 if (!fstat(fd, &st)) {
1669 *size = xsize_t(st.st_size);
33e42de0
MM
1670 if (!*size) {
1671 /* mmap() is forbidden on empty files */
1672 error("object file %s is empty", sha1_file_name(sha1));
1673 return NULL;
1674 }
44d1c19e 1675 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1676 }
44d1c19e 1677 close(fd);
0fcfd160 1678 }
0fcfd160
LT
1679 return map;
1680}
1681
09ded04b
NP
1682unsigned long unpack_object_header_buffer(const unsigned char *buf,
1683 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1684{
ad1ed5ee 1685 unsigned shift;
48fb7deb 1686 unsigned long size, c;
ad1ed5ee
JH
1687 unsigned long used = 0;
1688
1689 c = buf[used++];
1690 *type = (c >> 4) & 7;
1691 size = c & 15;
1692 shift = 4;
1693 while (c & 0x80) {
f630cfda 1694 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1695 error("bad object header");
ea4f9685
JH
1696 size = used = 0;
1697 break;
09ded04b 1698 }
ad1ed5ee
JH
1699 c = buf[used++];
1700 size += (c & 0x7f) << shift;
1701 shift += 7;
1702 }
1703 *sizep = size;
1704 return used;
1705}
1706
d21f8426
JH
1707static int unpack_sha1_short_header(git_zstream *stream,
1708 unsigned char *map, unsigned long mapsize,
1709 void *buffer, unsigned long bufsiz)
ad1ed5ee 1710{
c4483576
LT
1711 /* Get the data stream */
1712 memset(stream, 0, sizeof(*stream));
1713 stream->next_in = map;
1714 stream->avail_in = mapsize;
1715 stream->next_out = buffer;
93821bd9
LT
1716 stream->avail_out = bufsiz;
1717
39c68542 1718 git_inflate_init(stream);
cc5c54e7 1719 return git_inflate(stream, 0);
c4483576
LT
1720}
1721
d21f8426
JH
1722int unpack_sha1_header(git_zstream *stream,
1723 unsigned char *map, unsigned long mapsize,
1724 void *buffer, unsigned long bufsiz)
1725{
1726 int status = unpack_sha1_short_header(stream, map, mapsize,
1727 buffer, bufsiz);
1728
1729 if (status < Z_OK)
1730 return status;
1731
1732 /* Make sure we have the terminating NUL */
1733 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1734 return -1;
1735 return 0;
1736}
1737
46f03448
KN
1738static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1739 unsigned long mapsize, void *buffer,
1740 unsigned long bufsiz, struct strbuf *header)
1741{
1742 int status;
1743
d21f8426
JH
1744 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1745 if (status < Z_OK)
1746 return -1;
46f03448
KN
1747
1748 /*
1749 * Check if entire header is unpacked in the first iteration.
1750 */
1751 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1752 return 0;
1753
1754 /*
1755 * buffer[0..bufsiz] was not large enough. Copy the partial
1756 * result out to header, and then append the result of further
1757 * reading the stream.
1758 */
1759 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1760 stream->next_out = buffer;
1761 stream->avail_out = bufsiz;
1762
1763 do {
1764 status = git_inflate(stream, 0);
1765 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1766 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1767 return 0;
1768 stream->next_out = buffer;
1769 stream->avail_out = bufsiz;
1770 } while (status != Z_STREAM_END);
1771 return -1;
1772}
1773
ef49a7a0 1774static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1775{
1776 int bytes = strlen(buffer) + 1;
3aee68aa 1777 unsigned char *buf = xmallocz(size);
93821bd9 1778 unsigned long n;
7efbff75 1779 int status = Z_OK;
5180cacc 1780
93821bd9
LT
1781 n = stream->total_out - bytes;
1782 if (n > size)
1783 n = size;
1784 memcpy(buf, (char *) buffer + bytes, n);
1785 bytes = n;
456cdf6e
LT
1786 if (bytes <= size) {
1787 /*
1788 * The above condition must be (bytes <= size), not
1789 * (bytes < size). In other words, even though we
ccf5ace0 1790 * expect no more output and set avail_out to zero,
456cdf6e
LT
1791 * the input zlib stream may have bytes that express
1792 * "this concludes the stream", and we *do* want to
1793 * eat that input.
1794 *
1795 * Otherwise we would not be able to test that we
1796 * consumed all the input to reach the expected size;
1797 * we also want to check that zlib tells us that all
1798 * went well with status == Z_STREAM_END at the end.
1799 */
5180cacc
LT
1800 stream->next_out = buf + bytes;
1801 stream->avail_out = size - bytes;
7efbff75 1802 while (status == Z_OK)
39c68542 1803 status = git_inflate(stream, Z_FINISH);
5180cacc 1804 }
456cdf6e 1805 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1806 git_inflate_end(stream);
7efbff75
JH
1807 return buf;
1808 }
1809
1810 if (status < 0)
1811 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1812 else if (stream->avail_in)
1813 error("garbage at end of loose object '%s'",
1814 sha1_to_hex(sha1));
1815 free(buf);
1816 return NULL;
5180cacc
LT
1817}
1818
1819/*
1820 * We used to just use "sscanf()", but that's actually way
1821 * too permissive for what we want to check. So do an anal
1822 * object header parse by hand.
1823 */
46f03448
KN
1824static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1825 unsigned int flags)
5180cacc 1826{
46f03448 1827 const char *type_buf = hdr;
5180cacc 1828 unsigned long size;
46f03448 1829 int type, type_len = 0;
5180cacc
LT
1830
1831 /*
46f03448 1832 * The type can be of any size but is followed by
21666f1a 1833 * a space.
5180cacc 1834 */
5180cacc
LT
1835 for (;;) {
1836 char c = *hdr++;
d21f8426
JH
1837 if (!c)
1838 return -1;
5180cacc
LT
1839 if (c == ' ')
1840 break;
46f03448 1841 type_len++;
5180cacc 1842 }
46f03448
KN
1843
1844 type = type_from_string_gently(type_buf, type_len, 1);
1845 if (oi->typename)
1846 strbuf_add(oi->typename, type_buf, type_len);
1847 /*
1848 * Set type to 0 if its an unknown object and
2e3a16b2 1849 * we're obtaining the type using '--allow-unknown-type'
46f03448
KN
1850 * option.
1851 */
1852 if ((flags & LOOKUP_UNKNOWN_OBJECT) && (type < 0))
1853 type = 0;
1854 else if (type < 0)
1855 die("invalid object type");
1856 if (oi->typep)
1857 *oi->typep = type;
5180cacc
LT
1858
1859 /*
1860 * The length must follow immediately, and be in canonical
1861 * decimal format (ie "010" is not valid).
1862 */
1863 size = *hdr++ - '0';
1864 if (size > 9)
1865 return -1;
1866 if (size) {
1867 for (;;) {
1868 unsigned long c = *hdr - '0';
1869 if (c > 9)
1870 break;
1871 hdr++;
1872 size = size * 10 + c;
1873 }
1874 }
46f03448
KN
1875
1876 if (oi->sizep)
1877 *oi->sizep = size;
5180cacc
LT
1878
1879 /*
1880 * The length must be followed by a zero byte
1881 */
46f03448
KN
1882 return *hdr ? -1 : type;
1883}
1884
1885int parse_sha1_header(const char *hdr, unsigned long *sizep)
1886{
27b5c1a0 1887 struct object_info oi = OBJECT_INFO_INIT;
46f03448
KN
1888
1889 oi.sizep = sizep;
46f03448 1890 return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
5180cacc
LT
1891}
1892
7efbff75 1893static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1894{
5180cacc 1895 int ret;
ef49a7a0 1896 git_zstream stream;
5180cacc 1897 char hdr[8192];
0fcfd160 1898
5180cacc 1899 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1900 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1901 return NULL;
1902
7efbff75 1903 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1904}
1905
54dab52a
NP
1906unsigned long get_size_from_delta(struct packed_git *p,
1907 struct pack_window **w_curs,
1908 off_t curpos)
1909{
1910 const unsigned char *data;
1911 unsigned char delta_head[20], *in;
ef49a7a0 1912 git_zstream stream;
54dab52a
NP
1913 int st;
1914
1915 memset(&stream, 0, sizeof(stream));
1916 stream.next_out = delta_head;
1917 stream.avail_out = sizeof(delta_head);
1918
39c68542 1919 git_inflate_init(&stream);
54dab52a
NP
1920 do {
1921 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1922 stream.next_in = in;
39c68542 1923 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1924 curpos += stream.next_in - in;
1925 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1926 stream.total_out < sizeof(delta_head));
39c68542 1927 git_inflate_end(&stream);
3d77d877
NP
1928 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1929 error("delta data unpack-initial failed");
1930 return 0;
1931 }
54dab52a
NP
1932
1933 /* Examine the initial part of the delta to figure out
1934 * the result size.
1935 */
1936 data = delta_head;
1937
1938 /* ignore base size */
1939 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1940
1941 /* Read the result size */
1942 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1943}
1944
c4001d92 1945static off_t get_delta_base(struct packed_git *p,
03e79c88 1946 struct pack_window **w_curs,
c4001d92 1947 off_t *curpos,
21666f1a 1948 enum object_type type,
c4001d92 1949 off_t delta_obj_offset)
eb32d236 1950{
2b87c45b 1951 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1952 off_t base_offset;
eb32d236 1953
8d8a4ea5
SP
1954 /* use_pack() assured us we have [base_info, base_info + 20)
1955 * as a range that we can look at without walking off the
1956 * end of the mapped window. Its actually the hash size
1957 * that is assured. An OFS_DELTA longer than the hash size
1958 * is stupid, as then a REF_DELTA would be smaller to store.
1959 */
21666f1a 1960 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1961 unsigned used = 0;
1962 unsigned char c = base_info[used++];
1963 base_offset = c & 127;
1964 while (c & 128) {
1965 base_offset += 1;
8723f216 1966 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1967 return 0; /* overflow */
eb32d236
NP
1968 c = base_info[used++];
1969 base_offset = (base_offset << 7) + (c & 127);
1970 }
1971 base_offset = delta_obj_offset - base_offset;
d8f32556 1972 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1973 return 0; /* out of bound */
2b87c45b 1974 *curpos += used;
21666f1a 1975 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1976 /* The base entry _must_ be in the same pack */
1977 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1978 *curpos += 20;
eb32d236
NP
1979 } else
1980 die("I am totally screwed");
2b87c45b 1981 return base_offset;
eb32d236
NP
1982}
1983
5d642e75
JK
1984/*
1985 * Like get_delta_base above, but we return the sha1 instead of the pack
1986 * offset. This means it is cheaper for REF deltas (we do not have to do
1987 * the final object lookup), but more expensive for OFS deltas (we
1988 * have to load the revidx to convert the offset back into a sha1).
1989 */
1990static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1991 struct pack_window **w_curs,
1992 off_t curpos,
1993 enum object_type type,
1994 off_t delta_obj_offset)
1995{
1996 if (type == OBJ_REF_DELTA) {
1997 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1998 return base;
1999 } else if (type == OBJ_OFS_DELTA) {
2000 struct revindex_entry *revidx;
2001 off_t base_offset = get_delta_base(p, w_curs, &curpos,
2002 type, delta_obj_offset);
2003
2004 if (!base_offset)
2005 return NULL;
2006
2007 revidx = find_pack_revindex(p, base_offset);
2008 if (!revidx)
2009 return NULL;
2010
2011 return nth_packed_object_sha1(p, revidx->nr);
2012 } else
2013 return NULL;
2014}
2015
f8c8abc5
JH
2016int unpack_object_header(struct packed_git *p,
2017 struct pack_window **w_curs,
2018 off_t *curpos,
2019 unsigned long *sizep)
a733cb60 2020{
03e79c88 2021 unsigned char *base;
ef49a7a0 2022 unsigned long left;
ad1ed5ee 2023 unsigned long used;
2b87c45b 2024 enum object_type type;
a733cb60 2025
8d8a4ea5 2026 /* use_pack() assures us we have [base, base + 20) available
0353a0c4 2027 * as a range that we can look at. (Its actually the hash
3dff5379 2028 * size that is assured.) With our object header encoding
8d8a4ea5
SP
2029 * the maximum deflated object size is 2^137, which is just
2030 * insane, so we know won't exceed what we have been given.
2031 */
2b87c45b 2032 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
2033 used = unpack_object_header_buffer(base, left, &type, sizep);
2034 if (!used) {
2035 type = OBJ_BAD;
2036 } else
2037 *curpos += used;
ad1ed5ee 2038
2b87c45b 2039 return type;
a733cb60
LT
2040}
2041
790d96c0
TR
2042static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
2043{
2044 int type;
2045 struct revindex_entry *revidx;
2046 const unsigned char *sha1;
2047 revidx = find_pack_revindex(p, obj_offset);
2048 if (!revidx)
2049 return OBJ_BAD;
2050 sha1 = nth_packed_object_sha1(p, revidx->nr);
2051 mark_bad_packed_object(p, sha1);
2052 type = sha1_object_info(sha1, NULL);
2053 if (type <= OBJ_NONE)
2054 return OBJ_BAD;
2055 return type;
2056}
2057
790d96c0
TR
2058#define POI_STACK_PREALLOC 64
2059
90191d37
JK
2060static enum object_type packed_to_object_type(struct packed_git *p,
2061 off_t obj_offset,
2062 enum object_type type,
2063 struct pack_window **w_curs,
2064 off_t curpos)
1f688557 2065{
790d96c0
TR
2066 off_t small_poi_stack[POI_STACK_PREALLOC];
2067 off_t *poi_stack = small_poi_stack;
2068 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
5db47c2b 2069
790d96c0
TR
2070 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2071 off_t base_offset;
90191d37 2072 unsigned long size;
790d96c0
TR
2073 /* Push the object we're going to leave behind */
2074 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
2075 poi_stack_alloc = alloc_nr(poi_stack_nr);
b32fa95f 2076 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
790d96c0
TR
2077 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
2078 } else {
2079 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
2080 }
2081 poi_stack[poi_stack_nr++] = obj_offset;
2082 /* If parsing the base offset fails, just unwind */
90191d37 2083 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
790d96c0
TR
2084 if (!base_offset)
2085 goto unwind;
2086 curpos = obj_offset = base_offset;
90191d37 2087 type = unpack_object_header(p, w_curs, &curpos, &size);
790d96c0
TR
2088 if (type <= OBJ_NONE) {
2089 /* If getting the base itself fails, we first
2090 * retry the base, otherwise unwind */
2091 type = retry_bad_packed_offset(p, base_offset);
2092 if (type > OBJ_NONE)
2093 goto out;
2094 goto unwind;
2095 }
2096 }
2097
21666f1a 2098 switch (type) {
790d96c0 2099 case OBJ_BAD:
a733cb60 2100 case OBJ_COMMIT:
a733cb60 2101 case OBJ_TREE:
a733cb60 2102 case OBJ_BLOB:
a733cb60 2103 case OBJ_TAG:
a69d0943 2104 break;
1f688557 2105 default:
3d77d877
NP
2106 error("unknown object type %i at offset %"PRIuMAX" in %s",
2107 type, (uintmax_t)obj_offset, p->pack_name);
2108 type = OBJ_BAD;
1f688557 2109 }
790d96c0
TR
2110
2111out:
2112 if (poi_stack != small_poi_stack)
2113 free(poi_stack);
21666f1a 2114 return type;
790d96c0
TR
2115
2116unwind:
2117 while (poi_stack_nr) {
2118 obj_offset = poi_stack[--poi_stack_nr];
2119 type = retry_bad_packed_offset(p, obj_offset);
2120 if (type > OBJ_NONE)
2121 goto out;
2122 }
2123 type = OBJ_BAD;
2124 goto out;
1f688557
JH
2125}
2126
ca79c985
JK
2127int packed_object_info(struct packed_git *p, off_t obj_offset,
2128 struct object_info *oi)
90191d37
JK
2129{
2130 struct pack_window *w_curs = NULL;
2131 unsigned long size;
2132 off_t curpos = obj_offset;
2133 enum object_type type;
2134
412916ee
JK
2135 /*
2136 * We always get the representation type, but only convert it to
2137 * a "real" type later if the caller is interested.
2138 */
90191d37
JK
2139 type = unpack_object_header(p, &w_curs, &curpos, &size);
2140
23c339c0 2141 if (oi->sizep) {
90191d37
JK
2142 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2143 off_t tmp_pos = curpos;
2144 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
2145 type, obj_offset);
2146 if (!base_offset) {
2147 type = OBJ_BAD;
2148 goto out;
2149 }
23c339c0
JK
2150 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2151 if (*oi->sizep == 0) {
90191d37
JK
2152 type = OBJ_BAD;
2153 goto out;
2154 }
2155 } else {
23c339c0 2156 *oi->sizep = size;
90191d37
JK
2157 }
2158 }
2159
23c339c0 2160 if (oi->disk_sizep) {
90191d37 2161 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
23c339c0 2162 *oi->disk_sizep = revidx[1].offset - obj_offset;
90191d37
JK
2163 }
2164
23c339c0
JK
2165 if (oi->typep) {
2166 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
2167 if (*oi->typep < 0) {
412916ee
JK
2168 type = OBJ_BAD;
2169 goto out;
2170 }
2171 }
90191d37 2172
5d642e75
JK
2173 if (oi->delta_base_sha1) {
2174 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2175 const unsigned char *base;
2176
2177 base = get_delta_base_sha1(p, &w_curs, curpos,
2178 type, obj_offset);
2179 if (!base) {
2180 type = OBJ_BAD;
2181 goto out;
2182 }
2183
2184 hashcpy(oi->delta_base_sha1, base);
2185 } else
2186 hashclr(oi->delta_base_sha1);
2187 }
2188
90191d37
JK
2189out:
2190 unuse_pack(&w_curs);
2191 return type;
2192}
2193
eb950c19 2194static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 2195 struct pack_window **w_curs,
c4001d92 2196 off_t curpos,
eb950c19 2197 unsigned long size)
de530aaa
SP
2198{
2199 int st;
ef49a7a0 2200 git_zstream stream;
079afb18 2201 unsigned char *buffer, *in;
de530aaa 2202
735efde8
NTND
2203 buffer = xmallocz_gently(size);
2204 if (!buffer)
2205 return NULL;
de530aaa 2206 memset(&stream, 0, sizeof(stream));
de530aaa 2207 stream.next_out = buffer;
39eea7bd 2208 stream.avail_out = size + 1;
de530aaa 2209
39c68542 2210 git_inflate_init(&stream);
079afb18 2211 do {
2b87c45b 2212 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 2213 stream.next_in = in;
39c68542 2214 st = git_inflate(&stream, Z_FINISH);
39eea7bd
JH
2215 if (!stream.avail_out)
2216 break; /* the payload is larger than it should be */
2b87c45b 2217 curpos += stream.next_in - in;
079afb18 2218 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 2219 git_inflate_end(&stream);
de530aaa
SP
2220 if ((st != Z_STREAM_END) || stream.total_out != size) {
2221 free(buffer);
2222 return NULL;
2223 }
2224
2225 return buffer;
2226}
2227
8261e1f1 2228static struct hashmap delta_base_cache;
18bdec11 2229static size_t delta_base_cached;
5e08ecbf 2230
12d95ef6 2231static LIST_HEAD(delta_base_cache_lru);
5e08ecbf 2232
8261e1f1 2233struct delta_base_cache_key {
e5e01619
LT
2234 struct packed_git *p;
2235 off_t base_offset;
8261e1f1
JK
2236};
2237
2238struct delta_base_cache_entry {
2239 struct hashmap hash;
2240 struct delta_base_cache_key key;
2241 struct list_head lru;
2242 void *data;
e5e01619 2243 unsigned long size;
e5e01619 2244 enum object_type type;
8261e1f1 2245};
e5e01619 2246
8261e1f1 2247static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
e5e01619 2248{
8261e1f1 2249 unsigned int hash;
e5e01619 2250
8261e1f1 2251 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
e5e01619 2252 hash += (hash >> 8) + (hash >> 16);
8261e1f1 2253 return hash;
e5e01619
LT
2254}
2255
84dd81c1
TR
2256static struct delta_base_cache_entry *
2257get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
5266d369 2258{
8261e1f1
JK
2259 struct hashmap_entry entry;
2260 struct delta_base_cache_key key;
2261
2262 if (!delta_base_cache.cmpfn)
2263 return NULL;
2264
2265 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
2266 key.p = p;
2267 key.base_offset = base_offset;
2268 return hashmap_get(&delta_base_cache, &entry, &key);
2269}
2270
2271static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
2272 const struct delta_base_cache_key *b)
2273{
2274 return a->p == b->p && a->base_offset == b->base_offset;
84dd81c1
TR
2275}
2276
8261e1f1
JK
2277static int delta_base_cache_hash_cmp(const void *va, const void *vb,
2278 const void *vkey)
84dd81c1 2279{
8261e1f1
JK
2280 const struct delta_base_cache_entry *a = va, *b = vb;
2281 const struct delta_base_cache_key *key = vkey;
2282 if (key)
2283 return !delta_base_cache_key_eq(&a->key, key);
2284 else
2285 return !delta_base_cache_key_eq(&a->key, &b->key);
5266d369
JH
2286}
2287
84dd81c1
TR
2288static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2289{
8261e1f1 2290 return !!get_delta_base_cache_entry(p, base_offset);
84dd81c1
TR
2291}
2292
4a5397ca
JK
2293/*
2294 * Remove the entry from the cache, but do _not_ free the associated
2295 * entry data. The caller takes ownership of the "data" buffer, and
2296 * should copy out any fields it wants before detaching.
2297 */
2298static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
84dd81c1 2299{
8261e1f1 2300 hashmap_remove(&delta_base_cache, ent, &ent->key);
12d95ef6 2301 list_del(&ent->lru);
84dd81c1 2302 delta_base_cached -= ent->size;
8261e1f1 2303 free(ent);
84dd81c1
TR
2304}
2305
62f255ad 2306static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
85fe35ab 2307 unsigned long *base_size, enum object_type *type)
62f255ad 2308{
84dd81c1 2309 struct delta_base_cache_entry *ent;
e5e01619 2310
84dd81c1 2311 ent = get_delta_base_cache_entry(p, base_offset);
8261e1f1 2312 if (!ent)
749bc58c 2313 return unpack_entry(p, base_offset, type, base_size);
e5e01619 2314
e5e01619
LT
2315 *type = ent->type;
2316 *base_size = ent->size;
85fe35ab 2317 return xmemdupz(ent->data, ent->size);
62f255ad
LT
2318}
2319
18bdec11
SP
2320static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2321{
8261e1f1
JK
2322 free(ent->data);
2323 detach_delta_base_cache_entry(ent);
18bdec11
SP
2324}
2325
3d20c636
SP
2326void clear_delta_base_cache(void)
2327{
8261e1f1
JK
2328 struct hashmap_iter iter;
2329 struct delta_base_cache_entry *entry;
2330 for (entry = hashmap_iter_first(&delta_base_cache, &iter);
2331 entry;
2332 entry = hashmap_iter_next(&iter)) {
2333 release_delta_base_cache(entry);
2334 }
3d20c636
SP
2335}
2336
62f255ad
LT
2337static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2338 void *base, unsigned long base_size, enum object_type type)
2339{
8261e1f1 2340 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
4b92bae7 2341 struct list_head *lru, *tmp;
e5e01619 2342
18bdec11 2343 delta_base_cached += base_size;
5e08ecbf 2344
4b92bae7 2345 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
12d95ef6
JK
2346 struct delta_base_cache_entry *f =
2347 list_entry(lru, struct delta_base_cache_entry, lru);
2348 if (delta_base_cached <= delta_base_cache_limit)
2349 break;
5e08ecbf
NP
2350 release_delta_base_cache(f);
2351 }
18bdec11 2352
8261e1f1
JK
2353 ent->key.p = p;
2354 ent->key.base_offset = base_offset;
e5e01619
LT
2355 ent->type = type;
2356 ent->data = base;
2357 ent->size = base_size;
12d95ef6 2358 list_add_tail(&ent->lru, &delta_base_cache_lru);
8261e1f1
JK
2359
2360 if (!delta_base_cache.cmpfn)
2361 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, 0);
2362 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
2363 hashmap_add(&delta_base_cache, ent);
62f255ad
LT
2364}
2365
c2c5b270
CC
2366static void *read_object(const unsigned char *sha1, enum object_type *type,
2367 unsigned long *size);
2368
5f44324d
JH
2369static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2370{
67dc598e
KB
2371 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2372 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2373 p->pack_name, (uintmax_t)obj_offset);
5f44324d
JH
2374}
2375
0e8189e2
NP
2376int do_check_packed_object_crc;
2377
abe601bb
TR
2378#define UNPACK_ENTRY_STACK_PREALLOC 64
2379struct unpack_entry_stack_ent {
2380 off_t obj_offset;
2381 off_t curpos;
2382 unsigned long size;
2383};
2384
c4001d92 2385void *unpack_entry(struct packed_git *p, off_t obj_offset,
abe601bb 2386 enum object_type *final_type, unsigned long *final_size)
1f688557 2387{
03e79c88 2388 struct pack_window *w_curs = NULL;
c4001d92 2389 off_t curpos = obj_offset;
abe601bb
TR
2390 void *data = NULL;
2391 unsigned long size;
2392 enum object_type type;
2393 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2394 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2395 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2396 int base_from_cache = 0;
1f688557 2397
67dc598e 2398 write_pack_access_log(p, obj_offset);
5f44324d 2399
abe601bb
TR
2400 /* PHASE 1: drill down to the innermost base object */
2401 for (;;) {
2402 off_t base_offset;
2403 int i;
2404 struct delta_base_cache_entry *ent;
2405
77965f8b 2406 ent = get_delta_base_cache_entry(p, curpos);
8261e1f1 2407 if (ent) {
77965f8b
NTND
2408 type = ent->type;
2409 data = ent->data;
2410 size = ent->size;
4a5397ca 2411 detach_delta_base_cache_entry(ent);
77965f8b
NTND
2412 base_from_cache = 1;
2413 break;
2414 }
2415
abe601bb
TR
2416 if (do_check_packed_object_crc && p->index_version > 1) {
2417 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
211c61c6 2418 off_t len = revidx[1].offset - obj_offset;
abe601bb
TR
2419 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2420 const unsigned char *sha1 =
2421 nth_packed_object_sha1(p, revidx->nr);
2422 error("bad packed object CRC for %s",
2423 sha1_to_hex(sha1));
2424 mark_bad_packed_object(p, sha1);
2425 unuse_pack(&w_curs);
2426 return NULL;
2427 }
2428 }
2429
abe601bb
TR
2430 type = unpack_object_header(p, &w_curs, &curpos, &size);
2431 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2432 break;
2433
2434 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2435 if (!base_offset) {
2436 error("failed to validate delta base reference "
2437 "at offset %"PRIuMAX" from %s",
2438 (uintmax_t)curpos, p->pack_name);
2439 /* bail to phase 2, in hopes of recovery */
2440 data = NULL;
2441 break;
2442 }
2443
2444 /* push object, proceed to base */
2445 if (delta_stack_nr >= delta_stack_alloc
2446 && delta_stack == small_delta_stack) {
2447 delta_stack_alloc = alloc_nr(delta_stack_nr);
b32fa95f 2448 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
abe601bb
TR
2449 memcpy(delta_stack, small_delta_stack,
2450 sizeof(*delta_stack)*delta_stack_nr);
2451 } else {
2452 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2453 }
2454 i = delta_stack_nr++;
2455 delta_stack[i].obj_offset = obj_offset;
2456 delta_stack[i].curpos = curpos;
2457 delta_stack[i].size = size;
2458
2459 curpos = obj_offset = base_offset;
0e8189e2
NP
2460 }
2461
abe601bb
TR
2462 /* PHASE 2: handle the base */
2463 switch (type) {
eb32d236
NP
2464 case OBJ_OFS_DELTA:
2465 case OBJ_REF_DELTA:
abe601bb 2466 if (data)
ef1177d1 2467 die("BUG: unpack_entry: left loop at a valid delta");
4d703a1a 2468 break;
a733cb60 2469 case OBJ_COMMIT:
a733cb60 2470 case OBJ_TREE:
a733cb60 2471 case OBJ_BLOB:
a733cb60 2472 case OBJ_TAG:
abe601bb
TR
2473 if (!base_from_cache)
2474 data = unpack_compressed_entry(p, &w_curs, curpos, size);
4d703a1a 2475 break;
1f688557 2476 default:
8eca0b47
NP
2477 data = NULL;
2478 error("unknown object type %i at offset %"PRIuMAX" in %s",
abe601bb
TR
2479 type, (uintmax_t)obj_offset, p->pack_name);
2480 }
2481
2482 /* PHASE 3: apply deltas in order */
2483
2484 /* invariants:
2485 * 'data' holds the base data, or NULL if there was corruption
2486 */
2487 while (delta_stack_nr) {
2488 void *delta_data;
2489 void *base = data;
2490 unsigned long delta_size, base_size = size;
2491 int i;
2492
2493 data = NULL;
2494
2495 if (base)
2496 add_delta_base_cache(p, obj_offset, base, base_size, type);
2497
2498 if (!base) {
2499 /*
2500 * We're probably in deep shit, but let's try to fetch
2501 * the required base anyway from another pack or loose.
2502 * This is costly but should happen only in the presence
2503 * of a corrupted pack, and is better than failing outright.
2504 */
2505 struct revindex_entry *revidx;
2506 const unsigned char *base_sha1;
2507 revidx = find_pack_revindex(p, obj_offset);
2508 if (revidx) {
2509 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2510 error("failed to read delta base object %s"
2511 " at offset %"PRIuMAX" from %s",
2512 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2513 p->pack_name);
2514 mark_bad_packed_object(p, base_sha1);
2515 base = read_object(base_sha1, &type, &base_size);
2516 }
2517 }
2518
2519 i = --delta_stack_nr;
2520 obj_offset = delta_stack[i].obj_offset;
2521 curpos = delta_stack[i].curpos;
2522 delta_size = delta_stack[i].size;
2523
2524 if (!base)
2525 continue;
2526
2527 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2528
2529 if (!delta_data) {
2530 error("failed to unpack compressed delta "
2531 "at offset %"PRIuMAX" from %s",
2532 (uintmax_t)curpos, p->pack_name);
abe601bb
TR
2533 data = NULL;
2534 continue;
2535 }
2536
2537 data = patch_delta(base, base_size,
2538 delta_data, delta_size,
2539 &size);
1ee886c1
JK
2540
2541 /*
2542 * We could not apply the delta; warn the user, but keep going.
2543 * Our failure will be noticed either in the next iteration of
2544 * the loop, or if this is the final delta, in the caller when
2545 * we return NULL. Those code paths will take care of making
2546 * a more explicit warning and retrying with another copy of
2547 * the object.
2548 */
abe601bb 2549 if (!data)
1ee886c1 2550 error("failed to apply delta");
abe601bb 2551
4b8f772c 2552 free(delta_data);
1f688557 2553 }
abe601bb
TR
2554
2555 *final_type = type;
2556 *final_size = size;
2557
03e79c88 2558 unuse_pack(&w_curs);
019d1e65
NTND
2559
2560 if (delta_stack != small_delta_stack)
2561 free(delta_stack);
2562
21666f1a 2563 return data;
1f688557
JH
2564}
2565
d079837e 2566const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 2567 uint32_t n)
9a217f2a 2568{
42873078 2569 const unsigned char *index = p->index_data;
d079837e
SP
2570 if (!index) {
2571 if (open_pack_index(p))
2572 return NULL;
2573 index = p->index_data;
2574 }
57059091 2575 if (n >= p->num_objects)
d72308e0 2576 return NULL;
74e34e1f
NP
2577 index += 4 * 256;
2578 if (p->index_version == 1) {
2579 return index + 24 * n + 4;
2580 } else {
2581 index += 8;
2582 return index + 20 * n;
2583 }
2584}
2585
47fe3f6e
JK
2586void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
2587{
2588 const unsigned char *ptr = vptr;
2589 const unsigned char *start = p->index_data;
2590 const unsigned char *end = start + p->index_size;
2591 if (ptr < start)
7465feba 2592 die(_("offset before start of pack index for %s (corrupt index?)"),
47fe3f6e
JK
2593 p->pack_name);
2594 /* No need to check for underflow; .idx files must be at least 8 bytes */
2595 if (ptr >= end - 8)
7465feba 2596 die(_("offset beyond end of pack index for %s (truncated index?)"),
47fe3f6e
JK
2597 p->pack_name);
2598}
2599
99093238 2600off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
2601{
2602 const unsigned char *index = p->index_data;
2603 index += 4 * 256;
2604 if (p->index_version == 1) {
2605 return ntohl(*((uint32_t *)(index + 24 * n)));
2606 } else {
2607 uint32_t off;
2608 index += 8 + p->num_objects * (20 + 4);
2609 off = ntohl(*((uint32_t *)(index + 4 * n)));
2610 if (!(off & 0x80000000))
2611 return off;
2612 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
47fe3f6e 2613 check_pack_index_ptr(p, index);
74e34e1f
NP
2614 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2615 ntohl(*((uint32_t *)(index + 4)));
2616 }
9a217f2a
JH
2617}
2618
c4001d92 2619off_t find_pack_entry_one(const unsigned char *sha1,
43057304 2620 struct packed_git *p)
1f688557 2621{
42873078 2622 const uint32_t *level1_ofs = p->index_data;
42873078 2623 const unsigned char *index = p->index_data;
628522ec
JH
2624 unsigned hi, lo, stride;
2625 static int use_lookup = -1;
2626 static int debug_lookup = -1;
2627
2628 if (debug_lookup < 0)
2629 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 2630
d079837e
SP
2631 if (!index) {
2632 if (open_pack_index(p))
2633 return 0;
2634 level1_ofs = p->index_data;
2635 index = p->index_data;
2636 }
74e34e1f
NP
2637 if (p->index_version > 1) {
2638 level1_ofs += 2;
2639 index += 8;
2640 }
42873078 2641 index += 4 * 256;
74e34e1f
NP
2642 hi = ntohl(level1_ofs[*sha1]);
2643 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
2644 if (p->index_version > 1) {
2645 stride = 20;
2646 } else {
2647 stride = 24;
2648 index += 4;
2649 }
2650
2651 if (debug_lookup)
6e1c2344 2652 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
2653 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2654
2655 if (use_lookup < 0)
2656 use_lookup = !!getenv("GIT_USE_LOOKUP");
2657 if (use_lookup) {
2658 int pos = sha1_entry_pos(index, stride, 0,
2659 lo, hi, p->num_objects, sha1);
2660 if (pos < 0)
2661 return 0;
2662 return nth_packed_object_offset(p, pos);
2663 }
1f688557
JH
2664
2665 do {
74e34e1f 2666 unsigned mi = (lo + hi) / 2;
628522ec
JH
2667 int cmp = hashcmp(index + mi * stride, sha1);
2668
2669 if (debug_lookup)
2670 printf("lo %u hi %u rg %u mi %u\n",
2671 lo, hi, hi - lo, mi);
43057304 2672 if (!cmp)
74e34e1f 2673 return nth_packed_object_offset(p, mi);
1f688557
JH
2674 if (cmp > 0)
2675 hi = mi;
2676 else
2677 lo = mi+1;
2678 } while (lo < hi);
2679 return 0;
2680}
2681
4c080182 2682int is_pack_valid(struct packed_git *p)
d131b7af
SP
2683{
2684 /* An already open pack is known to be valid. */
2685 if (p->pack_fd != -1)
2686 return 1;
2687
2688 /* If the pack has one window completely covering the
2689 * file size, the pack is known to be valid even if
2690 * the descriptor is not currently open.
2691 */
2692 if (p->windows) {
2693 struct pack_window *w = p->windows;
2694
2695 if (!w->offset && w->len == p->pack_size)
2696 return 1;
2697 }
2698
2699 /* Force the pack to open to prove its valid. */
2700 return !open_packed_git(p);
2701}
2702
95099731
NTND
2703static int fill_pack_entry(const unsigned char *sha1,
2704 struct pack_entry *e,
2705 struct packed_git *p)
2706{
2707 off_t offset;
2708
2709 if (p->num_bad_objects) {
2710 unsigned i;
2711 for (i = 0; i < p->num_bad_objects; i++)
2712 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2713 return 0;
2714 }
2715
2716 offset = find_pack_entry_one(sha1, p);
2717 if (!offset)
2718 return 0;
2719
2720 /*
2721 * We are about to tell the caller where they can locate the
2722 * requested object. We better make sure the packfile is
2723 * still here and can be accessed before supplying that
2724 * answer, as it may have been deleted since the index was
2725 * loaded!
2726 */
319b678a 2727 if (!is_pack_valid(p))
95099731 2728 return 0;
95099731
NTND
2729 e->offset = offset;
2730 e->p = p;
2731 hashcpy(e->sha1, sha1);
2732 return 1;
2733}
2734
d40d535b
MH
2735/*
2736 * Iff a pack file contains the object named by sha1, return true and
2737 * store its location to e.
2738 */
4d6acb70 2739static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557 2740{
a73cdd21 2741 struct mru_entry *p;
43057304 2742
1f688557 2743 prepare_packed_git();
f7c22cc6
NP
2744 if (!packed_git)
2745 return 0;
1f688557 2746
a73cdd21
JK
2747 for (p = packed_git_mru->head; p; p = p->next) {
2748 if (fill_pack_entry(sha1, e, p->item)) {
2749 mru_mark(packed_git_mru, p);
1b1005d1
MH
2750 return 1;
2751 }
c01f51cc 2752 }
1f688557
JH
2753 return 0;
2754}
2755
a6080a0a 2756struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
2757 struct packed_git *packs)
2758{
2759 struct packed_git *p;
bf592c50
DB
2760
2761 for (p = packs; p; p = p->next) {
43057304 2762 if (find_pack_entry_one(sha1, p))
bf592c50
DB
2763 return p;
2764 }
2765 return NULL;
21666f1a 2766
bf592c50
DB
2767}
2768
052fe5ea 2769static int sha1_loose_object_info(const unsigned char *sha1,
46f03448
KN
2770 struct object_info *oi,
2771 int flags)
65c2e0c3 2772{
46f03448
KN
2773 int status = 0;
2774 unsigned long mapsize;
65c2e0c3 2775 void *map;
ef49a7a0 2776 git_zstream stream;
d65a16f6 2777 char hdr[32];
46f03448 2778 struct strbuf hdrbuf = STRBUF_INIT;
65c2e0c3 2779
5d642e75
JK
2780 if (oi->delta_base_sha1)
2781 hashclr(oi->delta_base_sha1);
2782
052fe5ea
JK
2783 /*
2784 * If we don't care about type or size, then we don't
4ef8d1dd
JH
2785 * need to look inside the object at all. Note that we
2786 * do not optimize out the stat call, even if the
2787 * caller doesn't care about the disk-size, since our
2788 * return value implicitly indicates whether the
2789 * object even exists.
052fe5ea 2790 */
46f03448 2791 if (!oi->typep && !oi->typename && !oi->sizep) {
4ef8d1dd
JH
2792 struct stat st;
2793 if (stat_sha1_file(sha1, &st) < 0)
2794 return -1;
2795 if (oi->disk_sizep)
23c339c0 2796 *oi->disk_sizep = st.st_size;
052fe5ea
JK
2797 return 0;
2798 }
2799
bb6b8e4f 2800 map = map_sha1_file(sha1, &mapsize);
f0df4ed5 2801 if (!map)
dbea72a8 2802 return -1;
23c339c0
JK
2803 if (oi->disk_sizep)
2804 *oi->disk_sizep = mapsize;
46f03448
KN
2805 if ((flags & LOOKUP_UNKNOWN_OBJECT)) {
2806 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2807 status = error("unable to unpack %s header with --allow-unknown-type",
2808 sha1_to_hex(sha1));
2809 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
36e4d74a
JH
2810 status = error("unable to unpack %s header",
2811 sha1_to_hex(sha1));
46f03448
KN
2812 if (status < 0)
2813 ; /* Do nothing */
2814 else if (hdrbuf.len) {
2815 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2816 status = error("unable to parse %s header with --allow-unknown-type",
2817 sha1_to_hex(sha1));
2818 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
36e4d74a 2819 status = error("unable to parse %s header", sha1_to_hex(sha1));
39c68542 2820 git_inflate_end(&stream);
65c2e0c3 2821 munmap(map, mapsize);
46f03448 2822 if (status && oi->typep)
23c339c0 2823 *oi->typep = status;
46f03448 2824 strbuf_release(&hdrbuf);
052fe5ea 2825 return 0;
65c2e0c3
JH
2826}
2827
de7b5d62 2828int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
f0df4ed5 2829{
c4d9986f 2830 struct cached_object *co;
f0df4ed5 2831 struct pack_entry e;
5b086407 2832 int rtype;
46f03448 2833 enum object_type real_type;
1f7117ef 2834 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
f0df4ed5 2835
1f7117ef 2836 co = find_cached_object(real);
c4d9986f 2837 if (co) {
5b086407
JK
2838 if (oi->typep)
2839 *(oi->typep) = co->type;
9a490590
JH
2840 if (oi->sizep)
2841 *(oi->sizep) = co->size;
161f00e7
JK
2842 if (oi->disk_sizep)
2843 *(oi->disk_sizep) = 0;
5d642e75
JK
2844 if (oi->delta_base_sha1)
2845 hashclr(oi->delta_base_sha1);
46f03448
KN
2846 if (oi->typename)
2847 strbuf_addstr(oi->typename, typename(co->type));
9a490590 2848 oi->whence = OI_CACHED;
5b086407 2849 return 0;
c4d9986f
NTND
2850 }
2851
1f7117ef 2852 if (!find_pack_entry(real, &e)) {
ddd63e64 2853 /* Most likely it's a loose object. */
46f03448 2854 if (!sha1_loose_object_info(real, oi, flags)) {
9a490590 2855 oi->whence = OI_LOOSE;
5b086407 2856 return 0;
9a490590 2857 }
ddd63e64
SG
2858
2859 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 2860 reprepare_packed_git();
1f7117ef 2861 if (!find_pack_entry(real, &e))
052fe5ea 2862 return -1;
f0df4ed5 2863 }
3d77d877 2864
46f03448
KN
2865 /*
2866 * packed_object_info() does not follow the delta chain to
2867 * find out the real type, unless it is given oi->typep.
2868 */
2869 if (oi->typename && !oi->typep)
2870 oi->typep = &real_type;
2871
23c339c0 2872 rtype = packed_object_info(e.p, e.offset, oi);
412916ee 2873 if (rtype < 0) {
1f7117ef 2874 mark_bad_packed_object(e.p, real);
46f03448
KN
2875 if (oi->typep == &real_type)
2876 oi->typep = NULL;
1f7117ef 2877 return sha1_object_info_extended(real, oi, 0);
5266d369
JH
2878 } else if (in_delta_base_cache(e.p, e.offset)) {
2879 oi->whence = OI_DBCACHED;
9a490590
JH
2880 } else {
2881 oi->whence = OI_PACKED;
2882 oi->u.packed.offset = e.offset;
2883 oi->u.packed.pack = e.p;
2884 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2885 rtype == OBJ_OFS_DELTA);
3d77d877 2886 }
46f03448
KN
2887 if (oi->typename)
2888 strbuf_addstr(oi->typename, typename(*oi->typep));
2889 if (oi->typep == &real_type)
2890 oi->typep = NULL;
3d77d877 2891
5b086407 2892 return 0;
f0df4ed5
JS
2893}
2894
3fc0dca9 2895/* returns enum object_type or negative */
9a490590
JH
2896int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2897{
5b086407 2898 enum object_type type;
27b5c1a0 2899 struct object_info oi = OBJECT_INFO_INIT;
9a490590 2900
5b086407 2901 oi.typep = &type;
9a490590 2902 oi.sizep = sizep;
de7b5d62 2903 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
5b086407
JK
2904 return -1;
2905 return type;
9a490590
JH
2906}
2907
21666f1a
NP
2908static void *read_packed_sha1(const unsigned char *sha1,
2909 enum object_type *type, unsigned long *size)
1f688557
JH
2910{
2911 struct pack_entry e;
8eca0b47 2912 void *data;
1f688557 2913
cd673c1f 2914 if (!find_pack_entry(sha1, &e))
1f688557 2915 return NULL;
85fe35ab 2916 data = cache_or_unpack_entry(e.p, e.offset, size, type);
8eca0b47
NP
2917 if (!data) {
2918 /*
2919 * We're probably in deep shit, but let's try to fetch
2920 * the required object anyway from another pack or loose.
2921 * This should happen only in the presence of a corrupted
2922 * pack, and is better than failing outright.
2923 */
2924 error("failed to read object %s at offset %"PRIuMAX" from %s",
2925 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2926 mark_bad_packed_object(e.p, sha1);
ac939109 2927 data = read_object(sha1, type, size);
8eca0b47
NP
2928 }
2929 return data;
1f688557
JH
2930}
2931
21666f1a
NP
2932int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2933 unsigned char *sha1)
d66b37bb
JH
2934{
2935 struct cached_object *co;
2936
21666f1a 2937 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2938 if (has_sha1_file(sha1) || find_cached_object(sha1))
2939 return 0;
c7353967 2940 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
d66b37bb
JH
2941 co = &cached_objects[cached_object_nr++];
2942 co->size = len;
21666f1a 2943 co->type = type;
efa13f7b
JH
2944 co->buf = xmalloc(len);
2945 memcpy(co->buf, buf, len);
d66b37bb
JH
2946 hashcpy(co->sha1, sha1);
2947 return 0;
2948}
2949
c2c5b270
CC
2950static void *read_object(const unsigned char *sha1, enum object_type *type,
2951 unsigned long *size)
0fcfd160
LT
2952{
2953 unsigned long mapsize;
2954 void *map, *buf;
d66b37bb
JH
2955 struct cached_object *co;
2956
2957 co = find_cached_object(sha1);
2958 if (co) {
21666f1a 2959 *type = co->type;
d66b37bb 2960 *size = co->size;
182af834 2961 return xmemdupz(co->buf, co->size);
d66b37bb 2962 }
0fcfd160 2963
8276c007
PE
2964 buf = read_packed_sha1(sha1, type, size);
2965 if (buf)
2966 return buf;
bb6b8e4f 2967 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2968 if (map) {
7efbff75 2969 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2970 munmap(map, mapsize);
2971 return buf;
2972 }
637cdd9d 2973 reprepare_packed_git();
8276c007 2974 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2975}
2976
b6c4cecc
JH
2977/*
2978 * This function dies on corrupt objects; the callers who want to
2979 * deal with them should arrange to call read_object() and give error
2980 * messages themselves.
2981 */
5bf29b95
JH
2982void *read_sha1_file_extended(const unsigned char *sha1,
2983 enum object_type *type,
2984 unsigned long *size,
2985 unsigned flag)
ac939109 2986{
3ba7a065 2987 void *data;
b6c4cecc 2988 const struct packed_git *p;
bf93eea0 2989 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
b6c4cecc 2990
3ba7a065
JH
2991 errno = 0;
2992 data = read_object(repl, type, size);
4bbf5a26 2993 if (data)
b6c4cecc 2994 return data;
68095570 2995
25f3af3f 2996 if (errno && errno != ENOENT)
3ba7a065
JH
2997 die_errno("failed to read object %s", sha1_to_hex(sha1));
2998
68095570 2999 /* die if we replaced an object with one that does not exist */
b6c4cecc 3000 if (repl != sha1)
68095570
CC
3001 die("replacement %s not found for %s",
3002 sha1_to_hex(repl), sha1_to_hex(sha1));
3003
b6c4cecc 3004 if (has_loose_object(repl)) {
30d6c6ea
MH
3005 const char *path = sha1_file_name(sha1);
3006
b6c4cecc
JH
3007 die("loose object %s (stored in %s) is corrupt",
3008 sha1_to_hex(repl), path);
e8b15e61 3009 }
68095570 3010
b6c4cecc
JH
3011 if ((p = has_packed_and_bad(repl)) != NULL)
3012 die("packed object %s (stored in %s) is corrupt",
3013 sha1_to_hex(repl), p->pack_name);
f5552aee 3014
b6c4cecc 3015 return NULL;
ac939109
NP
3016}
3017
40469ee9 3018void *read_object_with_reference(const unsigned char *sha1,
21666f1a 3019 const char *required_type_name,
40469ee9
JH
3020 unsigned long *size,
3021 unsigned char *actual_sha1_return)
f4913f91 3022{
21666f1a 3023 enum object_type type, required_type;
f4913f91
JH
3024 void *buffer;
3025 unsigned long isize;
40469ee9 3026 unsigned char actual_sha1[20];
f4913f91 3027
21666f1a 3028 required_type = type_from_string(required_type_name);
e702496e 3029 hashcpy(actual_sha1, sha1);
40469ee9
JH
3030 while (1) {
3031 int ref_length = -1;
3032 const char *ref_type = NULL;
f4913f91 3033
21666f1a 3034 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
3035 if (!buffer)
3036 return NULL;
21666f1a 3037 if (type == required_type) {
40469ee9
JH
3038 *size = isize;
3039 if (actual_sha1_return)
e702496e 3040 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
3041 return buffer;
3042 }
3043 /* Handle references */
21666f1a 3044 else if (type == OBJ_COMMIT)
40469ee9 3045 ref_type = "tree ";
21666f1a 3046 else if (type == OBJ_TAG)
40469ee9
JH
3047 ref_type = "object ";
3048 else {
3049 free(buffer);
3050 return NULL;
3051 }
3052 ref_length = strlen(ref_type);
f4913f91 3053
50974ec9
MK
3054 if (ref_length + 40 > isize ||
3055 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 3056 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
3057 free(buffer);
3058 return NULL;
3059 }
1cf58e72 3060 free(buffer);
40469ee9
JH
3061 /* Now we have the ID of the referred-to object in
3062 * actual_sha1. Check again. */
f4913f91 3063 }
f4913f91
JH
3064}
3065
ce9fbf16 3066static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 3067 const char *type, unsigned char *sha1,
d65a16f6 3068 char *hdr, int *hdrlen)
d410c0f5 3069{
9126f009 3070 git_SHA_CTX c;
d410c0f5
JH
3071
3072 /* Generate the header */
ef1286d3 3073 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
d410c0f5
JH
3074
3075 /* Sha1.. */
9126f009
NP
3076 git_SHA1_Init(&c);
3077 git_SHA1_Update(&c, hdr, *hdrlen);
3078 git_SHA1_Update(&c, buf, len);
3079 git_SHA1_Final(sha1, &c);
d410c0f5
JH
3080}
3081
230f1322 3082/*
5a688fe4 3083 * Move the just written object into its final resting place.
230f1322 3084 */
cb5add58 3085int finalize_object_file(const char *tmpfile, const char *filename)
230f1322 3086{
e32c0a9c 3087 int ret = 0;
5a688fe4 3088
348df166 3089 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
3090 goto try_rename;
3091 else if (link(tmpfile, filename))
e32c0a9c 3092 ret = errno;
7ebb6fca
LT
3093
3094 /*
3095 * Coda hack - coda doesn't like cross-directory links,
3096 * so we fall back to a rename, which will mean that it
3097 * won't be able to check collisions, but that's not a
3098 * big deal.
3099 *
3100 * The same holds for FAT formatted media.
3101 *
3be1f18e 3102 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
3103 * left to unlink.
3104 */
3105 if (ret && ret != EEXIST) {
be66a6c4 3106 try_rename:
7ebb6fca 3107 if (!rename(tmpfile, filename))
3be1f18e 3108 goto out;
9e48b389 3109 ret = errno;
230f1322 3110 }
691f1a28 3111 unlink_or_warn(tmpfile);
230f1322
LT
3112 if (ret) {
3113 if (ret != EEXIST) {
7616c6ca 3114 return error_errno("unable to write sha1 filename %s", filename);
230f1322
LT
3115 }
3116 /* FIXME!!! Collision check here ? */
3117 }
3118
3be1f18e 3119out:
5256b006 3120 if (adjust_shared_perm(filename))
5a688fe4 3121 return error("unable to set permission to '%s'", filename);
230f1322
LT
3122 return 0;
3123}
3124
4d548150
LT
3125static int write_buffer(int fd, const void *buf, size_t len)
3126{
d34cf19b 3127 if (write_in_full(fd, buf, len) < 0)
7616c6ca 3128 return error_errno("file write error");
4d548150
LT
3129 return 0;
3130}
3131
ce9fbf16 3132int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
3133 unsigned char *sha1)
3134{
d65a16f6 3135 char hdr[32];
ef1286d3 3136 int hdrlen = sizeof(hdr);
abdc3fc8
RS
3137 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3138 return 0;
3139}
3140
e9039dd3
LT
3141/* Finalize a file on disk, and close it. */
3142static void close_sha1_file(int fd)
3143{
aafe9fba
LT
3144 if (fsync_object_files)
3145 fsync_or_die(fd, "sha1 file");
e9039dd3 3146 if (close(fd) != 0)
d824cbba 3147 die_errno("error when closing sha1 file");
e9039dd3
LT
3148}
3149
5723fe7e
LT
3150/* Size of directory component, including the ending '/' */
3151static inline int directory_size(const char *filename)
3152{
3153 const char *s = strrchr(filename, '/');
3154 if (!s)
3155 return 0;
3156 return s - filename + 1;
3157}
3158
3159/*
3160 * This creates a temporary file in the same directory as the final
3161 * 'filename'
3162 *
3163 * We want to avoid cross-directory filename renames, because those
3164 * can have problems on various filesystems (FAT, NFS, Coda).
3165 */
d4b3d11a 3166static int create_tmpfile(struct strbuf *tmp, const char *filename)
5723fe7e
LT
3167{
3168 int fd, dirlen = directory_size(filename);
3169
d4b3d11a
JK
3170 strbuf_reset(tmp);
3171 strbuf_add(tmp, filename, dirlen);
3172 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
3173 fd = git_mkstemp_mode(tmp->buf, 0444);
cbacbf4e 3174 if (fd < 0 && dirlen && errno == ENOENT) {
d4b3d11a
JK
3175 /*
3176 * Make sure the directory exists; note that the contents
3177 * of the buffer are undefined after mkstemp returns an
3178 * error, so we have to rewrite the whole buffer from
3179 * scratch.
3180 */
3181 strbuf_reset(tmp);
3182 strbuf_add(tmp, filename, dirlen - 1);
3183 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
b2476a60 3184 return -1;
d4b3d11a 3185 if (adjust_shared_perm(tmp->buf))
5723fe7e
LT
3186 return -1;
3187
3188 /* Try again */
d4b3d11a
JK
3189 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
3190 fd = git_mkstemp_mode(tmp->buf, 0444);
5723fe7e
LT
3191 }
3192 return fd;
3193}
3194
bbac7311 3195static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 3196 const void *buf, unsigned long len, time_t mtime)
0fcfd160 3197{
915308b1 3198 int fd, ret;
9892beba 3199 unsigned char compressed[4096];
ef49a7a0 3200 git_zstream stream;
748af44c
NP
3201 git_SHA_CTX c;
3202 unsigned char parano_sha1[20];
d4b3d11a 3203 static struct strbuf tmp_file = STRBUF_INIT;
30d6c6ea 3204 const char *filename = sha1_file_name(sha1);
a44c9a5e 3205
d4b3d11a 3206 fd = create_tmpfile(&tmp_file, filename);
aac17941 3207 if (fd < 0) {
35243577 3208 if (errno == EACCES)
82247e9b 3209 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
916d081b 3210 else
7616c6ca 3211 return error_errno("unable to create temporary file");
aac17941
LT
3212 }
3213
0fcfd160 3214 /* Set it up */
55bb5c91 3215 git_deflate_init(&stream, zlib_compression_level);
0fcfd160 3216 stream.next_out = compressed;
9892beba 3217 stream.avail_out = sizeof(compressed);
748af44c 3218 git_SHA1_Init(&c);
a44c9a5e
LT
3219
3220 /* First header.. */
d65a16f6 3221 stream.next_in = (unsigned char *)hdr;
a44c9a5e 3222 stream.avail_in = hdrlen;
55bb5c91
JH
3223 while (git_deflate(&stream, 0) == Z_OK)
3224 ; /* nothing */
748af44c 3225 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
3226
3227 /* Then the data itself.. */
c00e657d 3228 stream.next_in = (void *)buf;
a44c9a5e 3229 stream.avail_in = len;
9892beba 3230 do {
748af44c 3231 unsigned char *in0 = stream.next_in;
55bb5c91 3232 ret = git_deflate(&stream, Z_FINISH);
748af44c 3233 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
3234 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3235 die("unable to write sha1 file");
3236 stream.next_out = compressed;
3237 stream.avail_out = sizeof(compressed);
3238 } while (ret == Z_OK);
3239
ac54c277
LT
3240 if (ret != Z_STREAM_END)
3241 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
55bb5c91 3242 ret = git_deflate_end_gently(&stream);
ac54c277
LT
3243 if (ret != Z_OK)
3244 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
3245 git_SHA1_Final(parano_sha1, &c);
3246 if (hashcmp(sha1, parano_sha1) != 0)
3247 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 3248
e9039dd3 3249 close_sha1_file(fd);
0fcfd160 3250
bbac7311
NP
3251 if (mtime) {
3252 struct utimbuf utb;
3253 utb.actime = mtime;
3254 utb.modtime = mtime;
d4b3d11a 3255 if (utime(tmp_file.buf, &utb) < 0)
7616c6ca 3256 warning_errno("failed utime() on %s", tmp_file.buf);
bbac7311
NP
3257 }
3258
d4b3d11a 3259 return finalize_object_file(tmp_file.buf, filename);
0fcfd160 3260}
8237b185 3261
33d4221c
JK
3262static int freshen_loose_object(const unsigned char *sha1)
3263{
3264 return check_and_freshen(sha1, 1);
3265}
3266
3267static int freshen_packed_object(const unsigned char *sha1)
3268{
3269 struct pack_entry e;
ee1c6c34
JK
3270 if (!find_pack_entry(sha1, &e))
3271 return 0;
3272 if (e.p->freshened)
3273 return 1;
3274 if (!freshen_file(e.p->pack_name))
3275 return 0;
3276 e.p->freshened = 1;
3277 return 1;
33d4221c
JK
3278}
3279
1427a7ff 3280int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
bbac7311 3281{
bbac7311 3282 char hdr[32];
ef1286d3 3283 int hdrlen = sizeof(hdr);
bbac7311
NP
3284
3285 /* Normally if we have it in the pack then we do not bother writing
3286 * it out into .git/objects/??/?{38} file.
3287 */
3288 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
b5f52f37 3289 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
bbac7311
NP
3290 return 0;
3291 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3292}
3293
0c3db67c
ES
3294int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3295 unsigned char *sha1, unsigned flags)
3296{
3297 char *header;
3298 int hdrlen, status = 0;
3299
3300 /* type string, SP, %lu of the length plus NUL must fit this */
ef1286d3
JK
3301 hdrlen = strlen(type) + 32;
3302 header = xmalloc(hdrlen);
0c3db67c
ES
3303 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3304
3305 if (!(flags & HASH_WRITE_OBJECT))
3306 goto cleanup;
051086b9 3307 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
0c3db67c
ES
3308 goto cleanup;
3309 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3310
3311cleanup:
3312 free(header);
3313 return status;
3314}
3315
bbac7311
NP
3316int force_object_loose(const unsigned char *sha1, time_t mtime)
3317{
bbac7311
NP
3318 void *buf;
3319 unsigned long len;
3320 enum object_type type;
3321 char hdr[32];
3322 int hdrlen;
1fb23e65 3323 int ret;
bbac7311 3324
c529d75a 3325 if (has_loose_object(sha1))
bbac7311
NP
3326 return 0;
3327 buf = read_packed_sha1(sha1, &type, &len);
3328 if (!buf)
3329 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
ef1286d3 3330 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1fb23e65
BS
3331 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3332 free(buf);
3333
3334 return ret;
bbac7311
NP
3335}
3336
bf592c50
DB
3337int has_pack_index(const unsigned char *sha1)
3338{
3339 struct stat st;
3340 if (stat(sha1_pack_index_name(sha1), &st))
3341 return 0;
3342 return 1;
bf592c50
DB
3343}
3344
cd673c1f
JH
3345int has_sha1_pack(const unsigned char *sha1)
3346{
3347 struct pack_entry e;
3348 return find_pack_entry(sha1, &e);
3349}
3350
0eeb077b 3351int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
8237b185 3352{
1f688557
JH
3353 struct pack_entry e;
3354
cd673c1f 3355 if (find_pack_entry(sha1, &e))
1f688557 3356 return 1;
45e8a748
JK
3357 if (has_loose_object(sha1))
3358 return 1;
0eeb077b
JK
3359 if (flags & HAS_SHA1_QUICK)
3360 return 0;
45e8a748
JK
3361 reprepare_packed_git();
3362 return find_pack_entry(sha1, &e);
b419aa25 3363}
3364
3365int has_object_file(const struct object_id *oid)
3366{
3367 return has_sha1_file(oid->hash);
8237b185 3368}
74400e71 3369
5827a035
JK
3370int has_object_file_with_flags(const struct object_id *oid, int flags)
3371{
3372 return has_sha1_file_with_flags(oid->hash, flags);
3373}
3374
c879daa2
NTND
3375static void check_tree(const void *buf, size_t size)
3376{
3377 struct tree_desc desc;
3378 struct name_entry entry;
3379
3380 init_tree_desc(&desc, buf, size);
3381 while (tree_entry(&desc, &entry))
3382 /* do nothing
3383 * tree_entry() will die() on malformed entries */
3384 ;
3385}
3386
3387static void check_commit(const void *buf, size_t size)
3388{
3389 struct commit c;
3390 memset(&c, 0, sizeof(c));
3391 if (parse_commit_buffer(&c, buf, size))
3392 die("corrupt commit");
3393}
3394
3395static void check_tag(const void *buf, size_t size)
3396{
3397 struct tag t;
3398 memset(&t, 0, sizeof(t));
3399 if (parse_tag_buffer(&t, buf, size))
3400 die("corrupt tag");
3401}
3402
43df4f86 3403static int index_mem(unsigned char *sha1, void *buf, size_t size,
c4ce46fc
JH
3404 enum object_type type,
3405 const char *path, unsigned flags)
e7332f96 3406{
6c510bee 3407 int ret, re_allocated = 0;
c4ce46fc 3408 int write_object = flags & HASH_WRITE_OBJECT;
74400e71 3409
7672db20 3410 if (!type)
edaec3fb 3411 type = OBJ_BLOB;
6c510bee
LT
3412
3413 /*
3414 * Convert blobs to git internal format
3415 */
43df4f86 3416 if ((type == OBJ_BLOB) && path) {
f285a2d7 3417 struct strbuf nbuf = STRBUF_INIT;
21e5ad50 3418 if (convert_to_git(path, buf, size, &nbuf,
5e12e78e 3419 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
b315c5c0 3420 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
3421 re_allocated = 1;
3422 }
3423 }
c4ce46fc 3424 if (flags & HASH_FORMAT_CHECK) {
c879daa2
NTND
3425 if (type == OBJ_TREE)
3426 check_tree(buf, size);
3427 if (type == OBJ_COMMIT)
3428 check_commit(buf, size);
3429 if (type == OBJ_TAG)
3430 check_tag(buf, size);
3431 }
6c510bee 3432
7672db20 3433 if (write_object)
edaec3fb 3434 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 3435 else
edaec3fb 3436 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 3437 if (re_allocated)
6c510bee 3438 free(buf);
43df4f86
DP
3439 return ret;
3440}
3441
9035d75a
SP
3442static int index_stream_convert_blob(unsigned char *sha1, int fd,
3443 const char *path, unsigned flags)
3444{
3445 int ret;
3446 const int write_object = flags & HASH_WRITE_OBJECT;
3447 struct strbuf sbuf = STRBUF_INIT;
3448
3449 assert(path);
3450 assert(would_convert_to_git_filter_fd(path));
3451
3452 convert_to_git_filter_fd(path, fd, &sbuf,
3453 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3454
3455 if (write_object)
3456 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3457 sha1);
3458 else
3459 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3460 sha1);
3461 strbuf_release(&sbuf);
3462 return ret;
3463}
3464
7b41e1e1
JH
3465static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3466 const char *path, unsigned flags)
3467{
3468 struct strbuf sbuf = STRBUF_INIT;
3469 int ret;
3470
3471 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3472 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3473 else
3474 ret = -1;
3475 strbuf_release(&sbuf);
3476 return ret;
3477}
3478
ea68b0ce
DP
3479#define SMALL_FILE_SIZE (32*1024)
3480
7b41e1e1
JH
3481static int index_core(unsigned char *sha1, int fd, size_t size,
3482 enum object_type type, const char *path,
3483 unsigned flags)
43df4f86
DP
3484{
3485 int ret;
43df4f86 3486
7b41e1e1 3487 if (!size) {
f6a1e1e2 3488 ret = index_mem(sha1, "", size, type, path, flags);
ea68b0ce
DP
3489 } else if (size <= SMALL_FILE_SIZE) {
3490 char *buf = xmalloc(size);
3491 if (size == read_in_full(fd, buf, size))
c4ce46fc 3492 ret = index_mem(sha1, buf, size, type, path, flags);
ea68b0ce 3493 else
7616c6ca 3494 ret = error_errno("short read");
ea68b0ce 3495 free(buf);
08bda208 3496 } else {
43df4f86 3497 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
c4ce46fc 3498 ret = index_mem(sha1, buf, size, type, path, flags);
aac17941 3499 munmap(buf, size);
08bda208 3500 }
7b41e1e1
JH
3501 return ret;
3502}
3503
4dd1fbc7 3504/*
568508e7
JH
3505 * This creates one packfile per large blob unless bulk-checkin
3506 * machinery is "plugged".
4dd1fbc7
JH
3507 *
3508 * This also bypasses the usual "convert-to-git" dance, and that is on
3509 * purpose. We could write a streaming version of the converting
3510 * functions and insert that before feeding the data to fast-import
4f22b101
JK
3511 * (or equivalent in-core API described above). However, that is
3512 * somewhat complicated, as we do not know the size of the filter
3513 * result, which we need to know beforehand when writing a git object.
3514 * Since the primary motivation for trying to stream from the working
3515 * tree file and to avoid mmaping it in core is to deal with large
3516 * binary blobs, they generally do not want to get any conversion, and
3517 * callers should avoid this code path when filters are requested.
4dd1fbc7
JH
3518 */