delta_base_cache: drop special treatment of blobs
[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"
0fcfd160 28
144bde78
LT
29#ifndef O_NOATIME
30#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
31#define O_NOATIME 01000000
32#else
33#define O_NOATIME 0
34#endif
35#endif
36
28bd70d8
JN
37#define SZ_FMT PRIuMAX
38static inline uintmax_t sz_fmt(size_t s) { return s; }
e05db0fd 39
96f1e58f 40const unsigned char null_sha1[20];
3e56e724 41const struct object_id null_oid;
88cd621d 42
c597ba80
NTND
43/*
44 * This is meant to hold a *small* number of objects that you would
45 * want read_sha1_file() to be able to return, but yet you do not want
46 * to write them into the object store (e.g. a browse-only
47 * application).
48 */
49static struct cached_object {
50 unsigned char sha1[20];
51 enum object_type type;
52 void *buf;
53 unsigned long size;
54} *cached_objects;
55static int cached_object_nr, cached_object_alloc;
56
57static struct cached_object empty_tree = {
dab0d410 58 EMPTY_TREE_SHA1_BIN_LITERAL,
c597ba80
NTND
59 OBJ_TREE,
60 "",
61 0
62};
63
64static struct cached_object *find_cached_object(const unsigned char *sha1)
65{
66 int i;
67 struct cached_object *co = cached_objects;
68
69 for (i = 0; i < cached_object_nr; i++, co++) {
70 if (!hashcmp(co->sha1, sha1))
71 return co;
72 }
73 if (!hashcmp(sha1, empty_tree.sha1))
74 return &empty_tree;
75 return NULL;
76}
77
90a6464b
JH
78int mkdir_in_gitdir(const char *path)
79{
80 if (mkdir(path, 0777)) {
81 int saved_errno = errno;
82 struct stat st;
83 struct strbuf sb = STRBUF_INIT;
84
85 if (errno != EEXIST)
86 return -1;
87 /*
88 * Are we looking at a path in a symlinked worktree
89 * whose original repository does not yet have it?
90 * e.g. .git/rr-cache pointing at its original
91 * repository in which the user hasn't performed any
92 * conflict resolution yet?
93 */
94 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
95 strbuf_readlink(&sb, path, st.st_size) ||
96 !is_absolute_path(sb.buf) ||
97 mkdir(sb.buf, 0777)) {
98 strbuf_release(&sb);
99 errno = saved_errno;
100 return -1;
101 }
102 strbuf_release(&sb);
103 }
104 return adjust_shared_perm(path);
105}
106
0be0521b 107enum scld_error safe_create_leading_directories(char *path)
b2cb9425 108{
26c8ae2a 109 char *next_component = path + offset_1st_component(path);
0be0521b 110 enum scld_error ret = SCLD_OK;
67d42212 111
0be0521b 112 while (ret == SCLD_OK && next_component) {
f0502332 113 struct stat st;
0f527403 114 char *slash = next_component, slash_character;
f0502332 115
0f527403
MH
116 while (*slash && !is_dir_sep(*slash))
117 slash++;
118
119 if (!*slash)
b2cb9425 120 break;
bf10cf70 121
26c8ae2a 122 next_component = slash + 1;
0f527403 123 while (is_dir_sep(*next_component))
bf10cf70 124 next_component++;
26c8ae2a 125 if (!*next_component)
5f0bdf50 126 break;
831651fd 127
0f527403 128 slash_character = *slash;
831651fd 129 *slash = '\0';
67d42212
JR
130 if (!stat(path, &st)) {
131 /* path exists */
9e6f885d 132 if (!S_ISDIR(st.st_mode))
0be0521b 133 ret = SCLD_EXISTS;
53a39721 134 } else if (mkdir(path, 0777)) {
928734d9 135 if (errno == EEXIST &&
9e6f885d 136 !stat(path, &st) && S_ISDIR(st.st_mode))
928734d9 137 ; /* somebody created it since we checked */
18d37e86
MH
138 else if (errno == ENOENT)
139 /*
140 * Either mkdir() failed because
141 * somebody just pruned the containing
142 * directory, or stat() failed because
143 * the file that was in our way was
144 * just removed. Either way, inform
145 * the caller that it might be worth
146 * trying again:
147 */
148 ret = SCLD_VANISHED;
9e6f885d 149 else
0be0521b 150 ret = SCLD_FAILED;
53a39721 151 } else if (adjust_shared_perm(path)) {
0be0521b 152 ret = SCLD_PERMS;
457f06d6 153 }
0f527403 154 *slash = slash_character;
b2cb9425 155 }
9e6f885d 156 return ret;
b2cb9425 157}
723c31fe 158
0be0521b 159enum scld_error safe_create_leading_directories_const(const char *path)
8e21d63b
JK
160{
161 /* path points to cache entries, so xstrdup before messing with it */
162 char *buf = xstrdup(path);
0be0521b 163 enum scld_error result = safe_create_leading_directories(buf);
8e21d63b
JK
164 free(buf);
165 return result;
166}
167
ace1534d
JH
168static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
169{
170 int i;
171 for (i = 0; i < 20; i++) {
172 static char hex[] = "0123456789abcdef";
173 unsigned int val = sha1[i];
174 char *pos = pathbuf + i*2 + (i > 0);
175 *pos++ = hex[val >> 4];
176 *pos = hex[val & 0xf];
177 }
178}
179
30d6c6ea 180const char *sha1_file_name(const unsigned char *sha1)
0fcfd160 181{
560fb6a1
JK
182 static char buf[PATH_MAX];
183 const char *objdir;
184 int len;
0fcfd160 185
560fb6a1
JK
186 objdir = get_object_directory();
187 len = strlen(objdir);
188
189 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
190 if (len + 43 > PATH_MAX)
191 die("insanely long object directory %s", objdir);
192 memcpy(buf, objdir, len);
193 buf[len] = '/';
194 buf[len+3] = '/';
195 buf[len+42] = '\0';
196 fill_sha1_path(buf + len + 1, sha1);
197 return buf;
0fcfd160
LT
198}
199
d40d535b
MH
200/*
201 * Return the name of the pack or index file with the specified sha1
202 * in its filename. *base and *name are scratch space that must be
203 * provided by the caller. which should be "pack" or "idx".
204 */
633f43e1 205static char *sha1_get_pack_name(const unsigned char *sha1,
ac5190cc
JK
206 struct strbuf *buf,
207 const char *which)
bf592c50 208{
ac5190cc
JK
209 strbuf_reset(buf);
210 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
211 sha1_to_hex(sha1), which);
212 return buf->buf;
bf592c50
DB
213}
214
633f43e1 215char *sha1_pack_name(const unsigned char *sha1)
bf592c50 216{
ac5190cc
JK
217 static struct strbuf buf = STRBUF_INIT;
218 return sha1_get_pack_name(sha1, &buf, "pack");
633f43e1 219}
bf592c50 220
633f43e1
HO
221char *sha1_pack_index_name(const unsigned char *sha1)
222{
ac5190cc
JK
223 static struct strbuf buf = STRBUF_INIT;
224 return sha1_get_pack_name(sha1, &buf, "idx");
bf592c50
DB
225}
226
d5a63b99
JH
227struct alternate_object_database *alt_odb_list;
228static struct alternate_object_database **alt_odb_tail;
ace1534d 229
ddd5d056
JH
230/*
231 * Prepare alternate object database registry.
d5a63b99
JH
232 *
233 * The variable alt_odb_list points at the list of struct
234 * alternate_object_database. The elements on this list come from
235 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
236 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
237 * whose contents is similar to that environment variable but can be
238 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
239 * contains "/the/directory/corresponding/to/.git/objects/...", while
240 * its name points just after the slash at the end of ".git/objects/"
241 * in the example above, and has enough space to hold 40-byte hex
242 * SHA1, an extra slash for the first level indirection, and the
243 * terminating NUL.
ddd5d056 244 */
539e7506
EK
245static int link_alt_odb_entry(const char *entry, const char *relative_base,
246 int depth, const char *normalized_objdir)
ace1534d 247{
c2f493a4
MW
248 struct alternate_object_database *ent;
249 struct alternate_object_database *alt;
50a6c8ef 250 size_t pfxlen, entlen;
5bdf0a84 251 struct strbuf pathbuf = STRBUF_INIT;
d5a63b99 252
85dadc38 253 if (!is_absolute_path(entry) && relative_base) {
5bdf0a84
HW
254 strbuf_addstr(&pathbuf, real_path(relative_base));
255 strbuf_addch(&pathbuf, '/');
c2f493a4 256 }
6eac50d8 257 strbuf_addstr(&pathbuf, entry);
c2f493a4 258
5bdf0a84
HW
259 normalize_path_copy(pathbuf.buf, pathbuf.buf);
260
261 pfxlen = strlen(pathbuf.buf);
262
263 /*
264 * The trailing slash after the directory name is given by
265 * this function at the end. Remove duplicates.
266 */
267 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
268 pfxlen -= 1;
269
50a6c8ef
JK
270 entlen = st_add(pfxlen, 43); /* '/' + 2 hex + '/' + 38 hex + NUL */
271 ent = xmalloc(st_add(sizeof(*ent), entlen));
5bdf0a84
HW
272 memcpy(ent->base, pathbuf.buf, pfxlen);
273 strbuf_release(&pathbuf);
c2f493a4
MW
274
275 ent->name = ent->base + pfxlen + 1;
276 ent->base[pfxlen + 3] = '/';
277 ent->base[pfxlen] = ent->base[entlen-1] = 0;
278
279 /* Detect cases where alternate disappeared */
90b4a71c 280 if (!is_directory(ent->base)) {
c2f493a4
MW
281 error("object directory %s does not exist; "
282 "check .git/objects/info/alternates.",
283 ent->base);
284 free(ent);
285 return -1;
286 }
287
288 /* Prevent the common mistake of listing the same
289 * thing twice, or object directory itself.
290 */
291 for (alt = alt_odb_list; alt; alt = alt->next) {
80b47854
RS
292 if (pfxlen == alt->name - alt->base - 1 &&
293 !memcmp(ent->base, alt->base, pfxlen)) {
c2f493a4
MW
294 free(ent);
295 return -1;
296 }
297 }
ba0897e6 298 if (!fspathcmp(ent->base, normalized_objdir)) {
c2f493a4
MW
299 free(ent);
300 return -1;
301 }
302
303 /* add the alternate entry */
304 *alt_odb_tail = ent;
305 alt_odb_tail = &(ent->next);
306 ent->next = NULL;
307
308 /* recursively add alternates */
309 read_info_alternates(ent->base, depth + 1);
310
311 ent->base[pfxlen] = '/';
312
313 return 0;
314}
315
c5950164 316static void link_alt_odb_entries(const char *alt, int len, int sep,
c2f493a4
MW
317 const char *relative_base, int depth)
318{
6eac50d8
MH
319 struct string_list entries = STRING_LIST_INIT_NODUP;
320 char *alt_copy;
321 int i;
539e7506 322 struct strbuf objdirbuf = STRBUF_INIT;
c2f493a4
MW
323
324 if (depth > 5) {
325 error("%s: ignoring alternate object stores, nesting too deep.",
326 relative_base);
327 return;
328 }
329
f655651e 330 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
539e7506
EK
331 normalize_path_copy(objdirbuf.buf, objdirbuf.buf);
332
c5950164 333 alt_copy = xmemdupz(alt, len);
6eac50d8
MH
334 string_list_split_in_place(&entries, alt_copy, sep, -1);
335 for (i = 0; i < entries.nr; i++) {
336 const char *entry = entries.items[i].string;
337 if (entry[0] == '\0' || entry[0] == '#')
9577e7e3 338 continue;
6eac50d8
MH
339 if (!is_absolute_path(entry) && depth) {
340 error("%s: ignoring relative alternate object store %s",
341 relative_base, entry);
342 } else {
539e7506 343 link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
9577e7e3 344 }
9577e7e3 345 }
6eac50d8
MH
346 string_list_clear(&entries, 0);
347 free(alt_copy);
539e7506 348 strbuf_release(&objdirbuf);
d5a63b99
JH
349}
350
5e73633d 351void read_info_alternates(const char * relative_base, int depth)
d5a63b99 352{
9577e7e3 353 char *map;
dc49cd76 354 size_t mapsz;
d5a63b99 355 struct stat st;
5015f01c 356 char *path;
c2f493a4 357 int fd;
d5a63b99 358
5015f01c 359 path = xstrfmt("%s/info/alternates", relative_base);
f4e51683 360 fd = git_open_noatime(path);
5015f01c 361 free(path);
d5a63b99
JH
362 if (fd < 0)
363 return;
364 if (fstat(fd, &st) || (st.st_size == 0)) {
365 close(fd);
9a217f2a 366 return;
ace1534d 367 }
dc49cd76
SP
368 mapsz = xsize_t(st.st_size);
369 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 370 close(fd);
d5a63b99 371
c5950164 372 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
c2f493a4 373
dc49cd76 374 munmap(map, mapsz);
ace1534d
JH
375}
376
bef70b22
DB
377void add_to_alternates_file(const char *reference)
378{
379 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
77b9b1d1
JK
380 char *alts = git_pathdup("objects/info/alternates");
381 FILE *in, *out;
382
383 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
384 out = fdopen_lock_file(lock, "w");
385 if (!out)
386 die_errno("unable to fdopen alternates lockfile");
387
388 in = fopen(alts, "r");
389 if (in) {
390 struct strbuf line = STRBUF_INIT;
391 int found = 0;
392
3f163962 393 while (strbuf_getline(&line, in) != EOF) {
77b9b1d1
JK
394 if (!strcmp(reference, line.buf)) {
395 found = 1;
396 break;
397 }
398 fprintf_or_die(out, "%s\n", line.buf);
399 }
400
401 strbuf_release(&line);
402 fclose(in);
403
404 if (found) {
405 rollback_lock_file(lock);
406 lock = NULL;
407 }
408 }
409 else if (errno != ENOENT)
410 die_errno("unable to read alternates file");
411
412 if (lock) {
413 fprintf_or_die(out, "%s\n", reference);
414 if (commit_lock_file(lock))
415 die_errno("unable to move new alternates file into place");
416 if (alt_odb_tail)
417 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
418 }
419 free(alts);
bef70b22
DB
420}
421
fe1b2268 422int foreach_alt_odb(alt_odb_fn fn, void *cb)
d79796bc
JH
423{
424 struct alternate_object_database *ent;
fe1b2268 425 int r = 0;
d79796bc
JH
426
427 prepare_alt_odb();
fe1b2268
JK
428 for (ent = alt_odb_list; ent; ent = ent->next) {
429 r = fn(ent, cb);
430 if (r)
431 break;
432 }
433 return r;
d79796bc
JH
434}
435
c2f493a4
MW
436void prepare_alt_odb(void)
437{
554fe20d 438 const char *alt;
c2f493a4 439
7dc24aa5
SP
440 if (alt_odb_tail)
441 return;
442
c2f493a4
MW
443 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
444 if (!alt) alt = "";
445
c2f493a4 446 alt_odb_tail = &alt_odb_list;
c5950164 447 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
448
449 read_info_alternates(get_object_directory(), 0);
450}
451
3096b2ec 452/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
33d4221c 453static int freshen_file(const char *fn)
ace1534d 454{
33d4221c
JK
455 struct utimbuf t;
456 t.actime = t.modtime = time(NULL);
457 return !utime(fn, &t);
0f4dc14a 458}
ace1534d 459
3096b2ec
JK
460/*
461 * All of the check_and_freshen functions return 1 if the file exists and was
462 * freshened (if freshening was requested), 0 otherwise. If they return
463 * 0, you should not assume that it is safe to skip a write of the object (it
464 * either does not exist on disk, or has a stale mtime and may be subject to
465 * pruning).
466 */
33d4221c
JK
467static int check_and_freshen_file(const char *fn, int freshen)
468{
469 if (access(fn, F_OK))
470 return 0;
3096b2ec 471 if (freshen && !freshen_file(fn))
33d4221c
JK
472 return 0;
473 return 1;
474}
475
476static int check_and_freshen_local(const unsigned char *sha1, int freshen)
477{
478 return check_and_freshen_file(sha1_file_name(sha1), freshen);
479}
480
481static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
0f4dc14a
BC
482{
483 struct alternate_object_database *alt;
9a217f2a 484 prepare_alt_odb();
d5a63b99 485 for (alt = alt_odb_list; alt; alt = alt->next) {
0f4dc14a 486 fill_sha1_path(alt->name, sha1);
33d4221c 487 if (check_and_freshen_file(alt->base, freshen))
c529d75a 488 return 1;
ace1534d 489 }
c529d75a 490 return 0;
ace1534d
JH
491}
492
33d4221c
JK
493static int check_and_freshen(const unsigned char *sha1, int freshen)
494{
495 return check_and_freshen_local(sha1, freshen) ||
496 check_and_freshen_nonlocal(sha1, freshen);
497}
498
499int has_loose_object_nonlocal(const unsigned char *sha1)
500{
501 return check_and_freshen_nonlocal(sha1, 0);
502}
503
0f4dc14a
BC
504static int has_loose_object(const unsigned char *sha1)
505{
33d4221c 506 return check_and_freshen(sha1, 0);
0f4dc14a
BC
507}
508
60bb8b14 509static unsigned int pack_used_ctr;
a53128b6
SP
510static unsigned int pack_mmap_calls;
511static unsigned int peak_pack_open_windows;
512static unsigned int pack_open_windows;
c7934306
SP
513static unsigned int pack_open_fds;
514static unsigned int pack_max_fds;
a53128b6 515static size_t peak_pack_mapped;
60bb8b14 516static size_t pack_mapped;
9a217f2a 517struct packed_git *packed_git;
1f688557 518
a73cdd21
JK
519static struct mru packed_git_mru_storage;
520struct mru *packed_git_mru = &packed_git_mru_storage;
521
b79d18c9 522void pack_report(void)
a53128b6
SP
523{
524 fprintf(stderr,
e05db0fd
PR
525 "pack_report: getpagesize() = %10" SZ_FMT "\n"
526 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
527 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
528 sz_fmt(getpagesize()),
529 sz_fmt(packed_git_window_size),
530 sz_fmt(packed_git_limit));
a53128b6
SP
531 fprintf(stderr,
532 "pack_report: pack_used_ctr = %10u\n"
533 "pack_report: pack_mmap_calls = %10u\n"
534 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
535 "pack_report: pack_mapped = "
536 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
537 pack_used_ctr,
538 pack_mmap_calls,
539 pack_open_windows, peak_pack_open_windows,
9e42d6a1 540 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
541}
542
d40d535b
MH
543/*
544 * Open and mmap the index file at path, perform a couple of
545 * consistency checks, then record its information to p. Return 0 on
546 * success.
547 */
548static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
549{
550 void *idx_map;
42873078 551 struct pack_idx_header *hdr;
dc49cd76 552 size_t idx_size;
74e34e1f 553 uint32_t version, nr, i, *index;
f4e51683 554 int fd = git_open_noatime(path);
1f688557 555 struct stat st;
42873078 556
1f688557
JH
557 if (fd < 0)
558 return -1;
559 if (fstat(fd, &st)) {
560 close(fd);
561 return -1;
562 }
dc49cd76 563 idx_size = xsize_t(st.st_size);
2d88451b
SP
564 if (idx_size < 4 * 256 + 20 + 20) {
565 close(fd);
566 return error("index file %s is too small", path);
567 }
c4712e45 568 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 569 close(fd);
1f688557 570
42873078
NP
571 hdr = idx_map;
572 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
573 version = ntohl(hdr->idx_version);
574 if (version < 2 || version > 2) {
575 munmap(idx_map, idx_size);
6e1c2344 576 return error("index file %s is version %"PRIu32
74e34e1f
NP
577 " and is not supported by this binary"
578 " (try upgrading GIT to a newer version)",
579 path, version);
580 }
581 } else
582 version = 1;
df1b059d 583
1f688557 584 nr = 0;
42873078 585 index = idx_map;
74e34e1f
NP
586 if (version > 1)
587 index += 2; /* skip index header */
1f688557 588 for (i = 0; i < 256; i++) {
326bf396 589 uint32_t n = ntohl(index[i]);
2d88451b
SP
590 if (n < nr) {
591 munmap(idx_map, idx_size);
df1b059d 592 return error("non-monotonic index %s", path);
2d88451b 593 }
1f688557
JH
594 nr = n;
595 }
596
74e34e1f
NP
597 if (version == 1) {
598 /*
599 * Total size:
600 * - 256 index entries 4 bytes each
601 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
602 * - 20-byte SHA1 of the packfile
603 * - 20-byte SHA1 file checksum
604 */
605 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
606 munmap(idx_map, idx_size);
eef427a0 607 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
608 }
609 } else if (version == 2) {
610 /*
611 * Minimum size:
612 * - 8 bytes of header
613 * - 256 index entries 4 bytes each
614 * - 20-byte sha1 entry * nr
615 * - 4-byte crc entry * nr
616 * - 4-byte offset entry * nr
617 * - 20-byte SHA1 of the packfile
618 * - 20-byte SHA1 file checksum
619 * And after the 4-byte offset table might be a
620 * variable sized table containing 8-byte entries
621 * for offsets larger than 2^31.
622 */
623 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
624 unsigned long max_size = min_size;
625 if (nr)
626 max_size += (nr - 1)*8;
627 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 628 munmap(idx_map, idx_size);
eef427a0 629 return error("wrong index v2 file size in %s", path);
74e34e1f 630 }
7109c889
JH
631 if (idx_size != min_size &&
632 /*
633 * make sure we can deal with large pack offsets.
634 * 31-bit signed offset won't be enough, neither
635 * 32-bit unsigned one will be.
636 */
637 (sizeof(off_t) <= 4)) {
638 munmap(idx_map, idx_size);
639 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 640 }
2d88451b 641 }
1f688557 642
74e34e1f 643 p->index_version = version;
42873078
NP
644 p->index_data = idx_map;
645 p->index_size = idx_size;
57059091 646 p->num_objects = nr;
1f688557
JH
647 return 0;
648}
649
bc8e478a 650int open_pack_index(struct packed_git *p)
d079837e
SP
651{
652 char *idx_name;
9ae97018 653 size_t len;
d079837e
SP
654 int ret;
655
656 if (p->index_data)
657 return 0;
658
9ae97018
JK
659 if (!strip_suffix(p->pack_name, ".pack", &len))
660 die("BUG: pack_name does not end in .pack");
661 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
d079837e
SP
662 ret = check_packed_git_idx(idx_name, p);
663 free(idx_name);
664 return ret;
665}
666
11daf39b
SP
667static void scan_windows(struct packed_git *p,
668 struct packed_git **lru_p,
669 struct pack_window **lru_w,
670 struct pack_window **lru_l)
1f688557 671{
11daf39b
SP
672 struct pack_window *w, *w_l;
673
674 for (w_l = NULL, w = p->windows; w; w = w->next) {
675 if (!w->inuse_cnt) {
676 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
677 *lru_p = p;
678 *lru_w = w;
679 *lru_l = w_l;
54044bf8 680 }
54044bf8 681 }
11daf39b 682 w_l = w;
f9253394 683 }
11daf39b
SP
684}
685
7c3ecb32 686static int unuse_one_window(struct packed_git *current)
11daf39b
SP
687{
688 struct packed_git *p, *lru_p = NULL;
689 struct pack_window *lru_w = NULL, *lru_l = NULL;
690
691 if (current)
692 scan_windows(current, &lru_p, &lru_w, &lru_l);
693 for (p = packed_git; p; p = p->next)
694 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
695 if (lru_p) {
696 munmap(lru_w->base, lru_w->len);
697 pack_mapped -= lru_w->len;
698 if (lru_l)
699 lru_l->next = lru_w->next;
7c3ecb32 700 else
54044bf8 701 lru_p->windows = lru_w->next;
54044bf8 702 free(lru_w);
a53128b6 703 pack_open_windows--;
54044bf8
SP
704 return 1;
705 }
706 return 0;
f9253394
JH
707}
708
7c3ecb32 709void release_pack_memory(size_t need)
97bfeb34
SP
710{
711 size_t cur = pack_mapped;
7c3ecb32 712 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
97bfeb34
SP
713 ; /* nothing */
714}
715
02710228
SP
716static void mmap_limit_check(size_t length)
717{
718 static size_t limit = 0;
719 if (!limit) {
720 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
721 if (!limit)
722 limit = SIZE_MAX;
723 }
724 if (length > limit)
725 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
726 (uintmax_t)length, (uintmax_t)limit);
727}
728
1570856b
JK
729void *xmmap_gently(void *start, size_t length,
730 int prot, int flags, int fd, off_t offset)
58ecbd5e 731{
02710228
SP
732 void *ret;
733
734 mmap_limit_check(length);
735 ret = mmap(start, length, prot, flags, fd, offset);
58ecbd5e
JN
736 if (ret == MAP_FAILED) {
737 if (!length)
738 return NULL;
7c3ecb32 739 release_pack_memory(length);
58ecbd5e 740 ret = mmap(start, length, prot, flags, fd, offset);
58ecbd5e
JN
741 }
742 return ret;
743}
744
1570856b
JK
745void *xmmap(void *start, size_t length,
746 int prot, int flags, int fd, off_t offset)
747{
748 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
749 if (ret == MAP_FAILED)
9ca0aaf6 750 die_errno("mmap failed");
1570856b
JK
751 return ret;
752}
753
c9ced051
SP
754void close_pack_windows(struct packed_git *p)
755{
756 while (p->windows) {
757 struct pack_window *w = p->windows;
758
759 if (w->inuse_cnt)
760 die("pack '%s' still has open windows to it",
761 p->pack_name);
762 munmap(w->base, w->len);
763 pack_mapped -= w->len;
764 pack_open_windows--;
765 p->windows = w->next;
766 free(w);
767 }
768}
769
71fe5d7f
JS
770static int close_pack_fd(struct packed_git *p)
771{
772 if (p->pack_fd < 0)
773 return 0;
774
775 close(p->pack_fd);
776 pack_open_fds--;
777 p->pack_fd = -1;
778
779 return 1;
780}
781
38849a81
JS
782static void close_pack(struct packed_git *p)
783{
784 close_pack_windows(p);
785 close_pack_fd(p);
786 close_pack_index(p);
787}
788
789void close_all_packs(void)
790{
791 struct packed_git *p;
792
793 for (p = packed_git; p; p = p->next)
794 if (p->do_not_close)
ef1177d1 795 die("BUG: want to close pack marked 'do-not-close'");
38849a81
JS
796 else
797 close_pack(p);
798}
799
800
88d0db55
BC
801/*
802 * The LRU pack is the one with the oldest MRU window, preferring packs
803 * with no used windows, or the oldest mtime if it has no windows allocated.
804 */
805static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
806{
807 struct pack_window *w, *this_mru_w;
808 int has_windows_inuse = 0;
809
810 /*
811 * Reject this pack if it has windows and the previously selected
812 * one does not. If this pack does not have windows, reject
813 * it if the pack file is newer than the previously selected one.
814 */
815 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
816 return;
817
818 for (w = this_mru_w = p->windows; w; w = w->next) {
819 /*
820 * Reject this pack if any of its windows are in use,
821 * but the previously selected pack did not have any
822 * inuse windows. Otherwise, record that this pack
823 * has windows in use.
824 */
825 if (w->inuse_cnt) {
826 if (*accept_windows_inuse)
827 has_windows_inuse = 1;
828 else
829 return;
830 }
831
832 if (w->last_used > this_mru_w->last_used)
833 this_mru_w = w;
834
835 /*
836 * Reject this pack if it has windows that have been
837 * used more recently than the previously selected pack.
838 * If the previously selected pack had windows inuse and
839 * we have not encountered a window in this pack that is
840 * inuse, skip this check since we prefer a pack with no
841 * inuse windows to one that has inuse windows.
842 */
843 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
844 this_mru_w->last_used > (*mru_w)->last_used)
845 return;
846 }
847
848 /*
849 * Select this pack.
850 */
851 *mru_w = this_mru_w;
852 *lru_p = p;
853 *accept_windows_inuse = has_windows_inuse;
854}
855
856static int close_one_pack(void)
857{
858 struct packed_git *p, *lru_p = NULL;
859 struct pack_window *mru_w = NULL;
860 int accept_windows_inuse = 1;
861
862 for (p = packed_git; p; p = p->next) {
863 if (p->pack_fd == -1)
864 continue;
865 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
866 }
867
71fe5d7f
JS
868 if (lru_p)
869 return close_pack_fd(lru_p);
88d0db55
BC
870
871 return 0;
872}
873
03e79c88 874void unuse_pack(struct pack_window **w_cursor)
f9253394 875{
03e79c88
SP
876 struct pack_window *w = *w_cursor;
877 if (w) {
878 w->inuse_cnt--;
879 *w_cursor = NULL;
880 }
1f688557
JH
881}
882
fa5fc15d
SP
883void close_pack_index(struct packed_git *p)
884{
885 if (p->index_data) {
886 munmap((void *)p->index_data, p->index_size);
887 p->index_data = NULL;
888 }
889}
890
a0788266
JS
891static unsigned int get_max_fd_limit(void)
892{
893#ifdef RLIMIT_NOFILE
491a8dec
JH
894 {
895 struct rlimit lim;
a0788266 896
491a8dec
JH
897 if (!getrlimit(RLIMIT_NOFILE, &lim))
898 return lim.rlim_cur;
899 }
900#endif
901
902#ifdef _SC_OPEN_MAX
903 {
904 long open_max = sysconf(_SC_OPEN_MAX);
905 if (0 < open_max)
906 return open_max;
907 /*
908 * Otherwise, we got -1 for one of the two
909 * reasons:
910 *
911 * (1) sysconf() did not understand _SC_OPEN_MAX
912 * and signaled an error with -1; or
913 * (2) sysconf() said there is no limit.
914 *
915 * We _could_ clear errno before calling sysconf() to
916 * tell these two cases apart and return a huge number
917 * in the latter case to let the caller cap it to a
918 * value that is not so selfish, but letting the
919 * fallback OPEN_MAX codepath take care of these cases
920 * is a lot simpler.
921 */
922 }
923#endif
a0788266 924
491a8dec 925#ifdef OPEN_MAX
a0788266
JS
926 return OPEN_MAX;
927#else
928 return 1; /* see the caller ;-) */
929#endif
930}
931
3cf8b462
SP
932/*
933 * Do not call this directly as this leaks p->pack_fd on error return;
934 * call open_packed_git() instead.
935 */
936static int open_packed_git_1(struct packed_git *p)
1f688557 937{
9bc879c1
SP
938 struct stat st;
939 struct pack_header hdr;
940 unsigned char sha1[20];
941 unsigned char *idx_sha1;
2c039da8 942 long fd_flag;
9bc879c1 943
d079837e
SP
944 if (!p->index_data && open_pack_index(p))
945 return error("packfile %s index unavailable", p->pack_name);
946
c7934306 947 if (!pack_max_fds) {
a0788266 948 unsigned int max_fds = get_max_fd_limit();
c7934306
SP
949
950 /* Save 3 for stdin/stdout/stderr, 22 for work */
951 if (25 < max_fds)
952 pack_max_fds = max_fds - 25;
953 else
954 pack_max_fds = 1;
955 }
956
88d0db55 957 while (pack_max_fds <= pack_open_fds && close_one_pack())
c7934306
SP
958 ; /* nothing */
959
f4e51683 960 p->pack_fd = git_open_noatime(p->pack_name);
9bc879c1 961 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 962 return -1;
c7934306 963 pack_open_fds++;
9bc879c1
SP
964
965 /* If we created the struct before we had the pack we lack size. */
bf592c50 966 if (!p->pack_size) {
bf592c50 967 if (!S_ISREG(st.st_mode))
072db278 968 return error("packfile %s not a regular file", p->pack_name);
bf592c50 969 p->pack_size = st.st_size;
9bc879c1 970 } else if (p->pack_size != st.st_size)
072db278 971 return error("packfile %s size changed", p->pack_name);
9bc879c1 972
2c039da8
JH
973 /* We leave these file descriptors open with sliding mmap;
974 * there is no point keeping them open across exec(), though.
975 */
976 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
977 if (fd_flag < 0)
072db278 978 return error("cannot determine file descriptor flags");
2c039da8
JH
979 fd_flag |= FD_CLOEXEC;
980 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 981 return error("cannot set FD_CLOEXEC");
2c039da8 982
9bc879c1 983 /* Verify we recognize this pack file format. */
e6e2bd62 984 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 985 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 986 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 987 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 988 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
989 return error("packfile %s is version %"PRIu32" and not"
990 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
991 p->pack_name, ntohl(hdr.hdr_version));
992
993 /* Verify the pack matches its index. */
57059091 994 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
995 return error("packfile %s claims to have %"PRIu32" objects"
996 " while index indicates %"PRIu32" objects",
57059091
NP
997 p->pack_name, ntohl(hdr.hdr_entries),
998 p->num_objects);
9bc879c1 999 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 1000 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 1001 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 1002 return error("packfile %s signature is unavailable", p->pack_name);
42873078 1003 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 1004 if (hashcmp(sha1, idx_sha1))
072db278
SP
1005 return error("packfile %s does not match index", p->pack_name);
1006 return 0;
9bc879c1
SP
1007}
1008
3cf8b462
SP
1009static int open_packed_git(struct packed_git *p)
1010{
1011 if (!open_packed_git_1(p))
1012 return 0;
71fe5d7f 1013 close_pack_fd(p);
3cf8b462
SP
1014 return -1;
1015}
1016
c4001d92 1017static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
1018{
1019 /* We must promise at least 20 bytes (one hash) after the
1020 * offset is available from this window, otherwise the offset
1021 * is not actually in this window and a different window (which
1022 * has that one hash excess) must be used. This is to support
1023 * the object header and delta base parsing routines below.
1024 */
1025 off_t win_off = win->offset;
1026 return win_off <= offset
1027 && (offset + 20) <= (win_off + win->len);
1028}
1029
4b25d091 1030unsigned char *use_pack(struct packed_git *p,
03e79c88 1031 struct pack_window **w_cursor,
c4001d92 1032 off_t offset,
ef49a7a0 1033 unsigned long *left)
9bc879c1 1034{
60bb8b14 1035 struct pack_window *win = *w_cursor;
03e79c88 1036
a9d98a14 1037 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
1038 * pointless to ask for an offset into the middle of that
1039 * hash, and the in_window function above wouldn't match
1040 * don't allow an offset too close to the end of the file.
1041 */
d131b7af
SP
1042 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1043 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
1044 if (offset > (p->pack_size - 20))
1045 die("offset beyond end of packfile (truncated pack?)");
13e0b0d3 1046 if (offset < 0)
7465feba 1047 die(_("offset before end of packfile (broken .idx?)"));
60bb8b14
SP
1048
1049 if (!win || !in_window(win, offset)) {
1050 if (win)
1051 win->inuse_cnt--;
1052 for (win = p->windows; win; win = win->next) {
1053 if (in_window(win, offset))
1054 break;
1055 }
1056 if (!win) {
78a28df9 1057 size_t window_align = packed_git_window_size / 2;
dc49cd76 1058 off_t len;
d131b7af
SP
1059
1060 if (p->pack_fd == -1 && open_packed_git(p))
1061 die("packfile %s cannot be accessed", p->pack_name);
1062
60bb8b14 1063 win = xcalloc(1, sizeof(*win));
78a28df9 1064 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
1065 len = p->pack_size - win->offset;
1066 if (len > packed_git_window_size)
1067 len = packed_git_window_size;
1068 win->len = (size_t)len;
60bb8b14 1069 pack_mapped += win->len;
11daf39b 1070 while (packed_git_limit < pack_mapped
7c3ecb32 1071 && unuse_one_window(p))
60bb8b14 1072 ; /* nothing */
c4712e45 1073 win->base = xmmap(NULL, win->len,
60bb8b14
SP
1074 PROT_READ, MAP_PRIVATE,
1075 p->pack_fd, win->offset);
1076 if (win->base == MAP_FAILED)
7616c6ca
NTND
1077 die_errno("packfile %s cannot be mapped",
1078 p->pack_name);
d131b7af 1079 if (!win->offset && win->len == p->pack_size
71fe5d7f
JS
1080 && !p->do_not_close)
1081 close_pack_fd(p);
a53128b6
SP
1082 pack_mmap_calls++;
1083 pack_open_windows++;
1084 if (pack_mapped > peak_pack_mapped)
1085 peak_pack_mapped = pack_mapped;
1086 if (pack_open_windows > peak_pack_open_windows)
1087 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
1088 win->next = p->windows;
1089 p->windows = win;
1090 }
1f688557 1091 }
03e79c88
SP
1092 if (win != *w_cursor) {
1093 win->last_used = pack_used_ctr++;
1094 win->inuse_cnt++;
1095 *w_cursor = win;
1096 }
60bb8b14 1097 offset -= win->offset;
03e79c88 1098 if (left)
dc49cd76 1099 *left = win->len - xsize_t(offset);
03e79c88 1100 return win->base + offset;
1f688557
JH
1101}
1102
27d69a46
NP
1103static struct packed_git *alloc_packed_git(int extra)
1104{
50a6c8ef 1105 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
27d69a46
NP
1106 memset(p, 0, sizeof(*p));
1107 p->pack_fd = -1;
1108 return p;
1109}
1110
e0500293
JN
1111static void try_to_free_pack_memory(size_t size)
1112{
7c3ecb32 1113 release_pack_memory(size);
e0500293
JN
1114}
1115
48bcc1c3 1116struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
1f688557 1117{
e0500293 1118 static int have_set_try_to_free_routine;
1f688557 1119 struct stat st;
48bcc1c3
JK
1120 size_t alloc;
1121 struct packed_git *p;
1f688557 1122
e0500293
JN
1123 if (!have_set_try_to_free_routine) {
1124 have_set_try_to_free_routine = 1;
1125 set_try_to_free_routine(try_to_free_pack_memory);
1126 }
1127
42873078
NP
1128 /*
1129 * Make sure a corresponding .pack file exists and that
1130 * the index looks sane.
1131 */
48bcc1c3 1132 if (!strip_suffix_mem(path, &path_len, ".idx"))
1f688557 1133 return NULL;
48bcc1c3
JK
1134
1135 /*
1136 * ".pack" is long enough to hold any suffix we're adding (and
1137 * the use xsnprintf double-checks that)
1138 */
50a6c8ef 1139 alloc = st_add3(path_len, strlen(".pack"), 1);
48bcc1c3 1140 p = alloc_packed_git(alloc);
42873078 1141 memcpy(p->pack_name, path, path_len);
8d25931d 1142
48bcc1c3 1143 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
8d25931d
BC
1144 if (!access(p->pack_name, F_OK))
1145 p->pack_keep = 1;
1146
48bcc1c3 1147 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
d079837e 1148 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 1149 free(p);
1f688557
JH
1150 return NULL;
1151 }
42873078 1152
1f688557
JH
1153 /* ok, it looks sane as far as we can check without
1154 * actually mapping the pack file.
1155 */
1f688557 1156 p->pack_size = st.st_size;
9d835df2 1157 p->pack_local = local;
b867092f 1158 p->mtime = st.st_mtime;
42873078
NP
1159 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1160 hashclr(p->sha1);
1f688557
JH
1161 return p;
1162}
1163
7b64469a 1164struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 1165{
42873078 1166 const char *path = sha1_pack_name(sha1);
50a6c8ef 1167 size_t alloc = st_add(strlen(path), 1);
c7ab0ba3 1168 struct packed_git *p = alloc_packed_git(alloc);
bf592c50 1169
c7ab0ba3 1170 memcpy(p->pack_name, path, alloc); /* includes NUL */
27d69a46 1171 hashcpy(p->sha1, sha1);
42873078
NP
1172 if (check_packed_git_idx(idx_path, p)) {
1173 free(p);
bf592c50 1174 return NULL;
42873078 1175 }
bf592c50 1176
bf592c50
DB
1177 return p;
1178}
1179
1180void install_packed_git(struct packed_git *pack)
1181{
c7934306
SP
1182 if (pack->pack_fd != -1)
1183 pack_open_fds++;
1184
bf592c50
DB
1185 pack->next = packed_git;
1186 packed_git = pack;
1187}
1188
0a489b06 1189void (*report_garbage)(unsigned seen_bits, const char *path);
543c5caa
NTND
1190
1191static void report_helper(const struct string_list *list,
1192 int seen_bits, int first, int last)
1193{
0a489b06 1194 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
543c5caa 1195 return;
0a489b06 1196
543c5caa 1197 for (; first < last; first++)
0a489b06 1198 report_garbage(seen_bits, list->items[first].string);
543c5caa
NTND
1199}
1200
1201static void report_pack_garbage(struct string_list *list)
1202{
1203 int i, baselen = -1, first = 0, seen_bits = 0;
1204
1205 if (!report_garbage)
1206 return;
1207
3383e199 1208 string_list_sort(list);
543c5caa
NTND
1209
1210 for (i = 0; i < list->nr; i++) {
1211 const char *path = list->items[i].string;
1212 if (baselen != -1 &&
1213 strncmp(path, list->items[first].string, baselen)) {
1214 report_helper(list, seen_bits, first, i);
1215 baselen = -1;
1216 seen_bits = 0;
1217 }
1218 if (baselen == -1) {
1219 const char *dot = strrchr(path, '.');
1220 if (!dot) {
0a489b06 1221 report_garbage(PACKDIR_FILE_GARBAGE, path);
543c5caa
NTND
1222 continue;
1223 }
1224 baselen = dot - path + 1;
1225 first = i;
1226 }
1227 if (!strcmp(path + baselen, "pack"))
1228 seen_bits |= 1;
1229 else if (!strcmp(path + baselen, "idx"))
1230 seen_bits |= 2;
1231 }
1232 report_helper(list, seen_bits, first, list->nr);
1233}
1234
9d835df2 1235static void prepare_packed_git_one(char *objdir, int local)
1f688557 1236{
880fb8de
RS
1237 struct strbuf path = STRBUF_INIT;
1238 size_t dirnamelen;
1f688557
JH
1239 DIR *dir;
1240 struct dirent *de;
543c5caa 1241 struct string_list garbage = STRING_LIST_INIT_DUP;
1f688557 1242
880fb8de
RS
1243 strbuf_addstr(&path, objdir);
1244 strbuf_addstr(&path, "/pack");
1245 dir = opendir(path.buf);
b5b16990 1246 if (!dir) {
26125f6b 1247 if (errno != ENOENT)
7616c6ca
NTND
1248 error_errno("unable to open object pack directory: %s",
1249 path.buf);
880fb8de 1250 strbuf_release(&path);
1f688557 1251 return;
b5b16990 1252 }
880fb8de
RS
1253 strbuf_addch(&path, '/');
1254 dirnamelen = path.len;
1f688557 1255 while ((de = readdir(dir)) != NULL) {
1f688557 1256 struct packed_git *p;
47bf4b0f 1257 size_t base_len;
543c5caa
NTND
1258
1259 if (is_dot_or_dotdot(de->d_name))
9cb18f56
JM
1260 continue;
1261
880fb8de
RS
1262 strbuf_setlen(&path, dirnamelen);
1263 strbuf_addstr(&path, de->d_name);
d90906a9 1264
47bf4b0f
JK
1265 base_len = path.len;
1266 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
d90906a9
NTND
1267 /* Don't reopen a pack we already have. */
1268 for (p = packed_git; p; p = p->next) {
47bf4b0f
JK
1269 size_t len;
1270 if (strip_suffix(p->pack_name, ".pack", &len) &&
1271 len == base_len &&
1272 !memcmp(p->pack_name, path.buf, len))
d90906a9
NTND
1273 break;
1274 }
1275 if (p == NULL &&
1276 /*
1277 * See if it really is a valid .idx file with
1278 * corresponding .pack file that we can map.
1279 */
880fb8de 1280 (p = add_packed_git(path.buf, path.len, local)) != NULL)
d90906a9 1281 install_packed_git(p);
86f7780c 1282 }
543c5caa
NTND
1283
1284 if (!report_garbage)
1285 continue;
1286
2975c770
JK
1287 if (ends_with(de->d_name, ".idx") ||
1288 ends_with(de->d_name, ".pack") ||
1289 ends_with(de->d_name, ".bitmap") ||
1290 ends_with(de->d_name, ".keep"))
880fb8de 1291 string_list_append(&garbage, path.buf);
543c5caa 1292 else
0a489b06 1293 report_garbage(PACKDIR_FILE_GARBAGE, path.buf);
1f688557 1294 }
5b35bcd5 1295 closedir(dir);
543c5caa
NTND
1296 report_pack_garbage(&garbage);
1297 string_list_clear(&garbage, 0);
880fb8de 1298 strbuf_release(&path);
1f688557
JH
1299}
1300
b867092f
JH
1301static int sort_pack(const void *a_, const void *b_)
1302{
1303 struct packed_git *a = *((struct packed_git **)a_);
1304 struct packed_git *b = *((struct packed_git **)b_);
1305 int st;
1306
1307 /*
1308 * Local packs tend to contain objects specific to our
1309 * variant of the project than remote ones. In addition,
1310 * remote ones could be on a network mounted filesystem.
1311 * Favor local ones for these reasons.
1312 */
1313 st = a->pack_local - b->pack_local;
1314 if (st)
1315 return -st;
1316
1317 /*
1318 * Younger packs tend to contain more recent objects,
1319 * and more recent objects tend to get accessed more
1320 * often.
1321 */
1322 if (a->mtime < b->mtime)
1323 return 1;
1324 else if (a->mtime == b->mtime)
1325 return 0;
1326 return -1;
1327}
1328
1329static void rearrange_packed_git(void)
1330{
1331 struct packed_git **ary, *p;
1332 int i, n;
1333
1334 for (n = 0, p = packed_git; p; p = p->next)
1335 n++;
1336 if (n < 2)
1337 return;
1338
1339 /* prepare an array of packed_git for easier sorting */
1340 ary = xcalloc(n, sizeof(struct packed_git *));
1341 for (n = 0, p = packed_git; p; p = p->next)
1342 ary[n++] = p;
1343
1344 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1345
1346 /* link them back again */
1347 for (i = 0; i < n - 1; i++)
1348 ary[i]->next = ary[i + 1];
1349 ary[n - 1]->next = NULL;
1350 packed_git = ary[0];
1351
1352 free(ary);
1353}
1354
a73cdd21
JK
1355static void prepare_packed_git_mru(void)
1356{
1357 struct packed_git *p;
1358
1359 mru_clear(packed_git_mru);
1360 for (p = packed_git; p; p = p->next)
1361 mru_append(packed_git_mru, p);
1362}
1363
637cdd9d 1364static int prepare_packed_git_run_once = 0;
9a217f2a 1365void prepare_packed_git(void)
1f688557 1366{
d5a63b99 1367 struct alternate_object_database *alt;
1f688557 1368
637cdd9d 1369 if (prepare_packed_git_run_once)
1f688557 1370 return;
9d835df2 1371 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 1372 prepare_alt_odb();
d5a63b99 1373 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 1374 alt->name[-1] = 0;
9d835df2 1375 prepare_packed_git_one(alt->base, 0);
1494e038 1376 alt->name[-1] = '/';
1f688557 1377 }
b867092f 1378 rearrange_packed_git();
a73cdd21 1379 prepare_packed_git_mru();
637cdd9d
JK
1380 prepare_packed_git_run_once = 1;
1381}
1382
fc04c412 1383void reprepare_packed_git(void)
637cdd9d
JK
1384{
1385 prepare_packed_git_run_once = 0;
1386 prepare_packed_git();
1f688557
JH
1387}
1388
8eca0b47
NP
1389static void mark_bad_packed_object(struct packed_git *p,
1390 const unsigned char *sha1)
1391{
1392 unsigned i;
1393 for (i = 0; i < p->num_bad_objects; i++)
50a6c8ef 1394 if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
8eca0b47 1395 return;
50a6c8ef
JK
1396 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1397 st_mult(GIT_SHA1_RAWSZ,
1398 st_add(p->num_bad_objects, 1)));
1399 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
8eca0b47
NP
1400 p->num_bad_objects++;
1401}
1402
b6c4cecc 1403static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
ac939109
NP
1404{
1405 struct packed_git *p;
1406 unsigned i;
1407
1408 for (p = packed_git; p; p = p->next)
1409 for (i = 0; i < p->num_bad_objects; i++)
1410 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
b6c4cecc
JH
1411 return p;
1412 return NULL;
ac939109
NP
1413}
1414
090ea126
NTND
1415/*
1416 * With an in-core object data in "map", rehash it to make sure the
1417 * object name actually matches "sha1" to detect object corruption.
1418 * With "map" == NULL, try reading the object named with "sha1" using
1419 * the streaming interface and rehash it to do the same.
1420 */
1421int check_sha1_signature(const unsigned char *sha1, void *map,
1422 unsigned long size, const char *type)
0fcfd160
LT
1423{
1424 unsigned char real_sha1[20];
090ea126
NTND
1425 enum object_type obj_type;
1426 struct git_istream *st;
1427 git_SHA_CTX c;
1428 char hdr[32];
1429 int hdrlen;
1430
1431 if (map) {
1432 hash_sha1_file(map, size, type, real_sha1);
1433 return hashcmp(sha1, real_sha1) ? -1 : 0;
1434 }
1435
1436 st = open_istream(sha1, &obj_type, &size, NULL);
1437 if (!st)
1438 return -1;
1439
1440 /* Generate the header */
ef1286d3 1441 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
090ea126
NTND
1442
1443 /* Sha1.. */
1444 git_SHA1_Init(&c);
1445 git_SHA1_Update(&c, hdr, hdrlen);
1446 for (;;) {
1447 char buf[1024 * 16];
1448 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1449
f54fac53
JK
1450 if (readlen < 0) {
1451 close_istream(st);
1452 return -1;
1453 }
090ea126
NTND
1454 if (!readlen)
1455 break;
1456 git_SHA1_Update(&c, buf, readlen);
1457 }
1458 git_SHA1_Final(real_sha1, &c);
1459 close_istream(st);
a89fccd2 1460 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1461}
1462
ec73f580 1463int git_open_noatime(const char *name)
44d1c19e
LT
1464{
1465 static int sha1_file_open_flag = O_NOATIME;
44d1c19e 1466
f2e872aa 1467 for (;;) {
dff6f280
CB
1468 int fd;
1469
1470 errno = 0;
1471 fd = open(name, O_RDONLY | sha1_file_open_flag);
44d1c19e 1472 if (fd >= 0)
f2e872aa
SP
1473 return fd;
1474
f2e872aa
SP
1475 /* Might the failure be due to O_NOATIME? */
1476 if (errno != ENOENT && sha1_file_open_flag) {
44d1c19e 1477 sha1_file_open_flag = 0;
f2e872aa
SP
1478 continue;
1479 }
1480
1481 return -1;
44d1c19e 1482 }
44d1c19e
LT
1483}
1484
052fe5ea
JK
1485static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1486{
052fe5ea
JK
1487 struct alternate_object_database *alt;
1488
30d6c6ea 1489 if (!lstat(sha1_file_name(sha1), st))
052fe5ea
JK
1490 return 0;
1491
1492 prepare_alt_odb();
1493 errno = ENOENT;
1494 for (alt = alt_odb_list; alt; alt = alt->next) {
30d6c6ea 1495 fill_sha1_path(alt->name, sha1);
052fe5ea
JK
1496 if (!lstat(alt->base, st))
1497 return 0;
1498 }
1499
1500 return -1;
1501}
1502
44d1c19e
LT
1503static int open_sha1_file(const unsigned char *sha1)
1504{
1505 int fd;
44d1c19e 1506 struct alternate_object_database *alt;
d6c8a05b 1507 int most_interesting_errno;
44d1c19e 1508
30d6c6ea 1509 fd = git_open_noatime(sha1_file_name(sha1));
44d1c19e
LT
1510 if (fd >= 0)
1511 return fd;
d6c8a05b 1512 most_interesting_errno = errno;
44d1c19e
LT
1513
1514 prepare_alt_odb();
44d1c19e 1515 for (alt = alt_odb_list; alt; alt = alt->next) {
30d6c6ea 1516 fill_sha1_path(alt->name, sha1);
f4e51683 1517 fd = git_open_noatime(alt->base);
44d1c19e
LT
1518 if (fd >= 0)
1519 return fd;
d6c8a05b
JK
1520 if (most_interesting_errno == ENOENT)
1521 most_interesting_errno = errno;
44d1c19e 1522 }
d6c8a05b 1523 errno = most_interesting_errno;
44d1c19e
LT
1524 return -1;
1525}
1526
f0270efd 1527void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1528{
0fcfd160 1529 void *map;
144bde78 1530 int fd;
ace1534d 1531
44d1c19e
LT
1532 fd = open_sha1_file(sha1);
1533 map = NULL;
1534 if (fd >= 0) {
1535 struct stat st;
0fcfd160 1536
44d1c19e
LT
1537 if (!fstat(fd, &st)) {
1538 *size = xsize_t(st.st_size);
33e42de0
MM
1539 if (!*size) {
1540 /* mmap() is forbidden on empty files */
1541 error("object file %s is empty", sha1_file_name(sha1));
1542 return NULL;
1543 }
44d1c19e 1544 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1545 }
44d1c19e 1546 close(fd);
0fcfd160 1547 }
0fcfd160
LT
1548 return map;
1549}
1550
09ded04b
NP
1551unsigned long unpack_object_header_buffer(const unsigned char *buf,
1552 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1553{
ad1ed5ee 1554 unsigned shift;
48fb7deb 1555 unsigned long size, c;
ad1ed5ee
JH
1556 unsigned long used = 0;
1557
1558 c = buf[used++];
1559 *type = (c >> 4) & 7;
1560 size = c & 15;
1561 shift = 4;
1562 while (c & 0x80) {
f630cfda 1563 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1564 error("bad object header");
ea4f9685
JH
1565 size = used = 0;
1566 break;
09ded04b 1567 }
ad1ed5ee
JH
1568 c = buf[used++];
1569 size += (c & 0x7f) << shift;
1570 shift += 7;
1571 }
1572 *sizep = size;
1573 return used;
1574}
1575
eb4f4076 1576int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
ad1ed5ee 1577{
c4483576
LT
1578 /* Get the data stream */
1579 memset(stream, 0, sizeof(*stream));
1580 stream->next_in = map;
1581 stream->avail_in = mapsize;
1582 stream->next_out = buffer;
93821bd9
LT
1583 stream->avail_out = bufsiz;
1584
39c68542 1585 git_inflate_init(stream);
cc5c54e7 1586 return git_inflate(stream, 0);
c4483576
LT
1587}
1588
46f03448
KN
1589static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1590 unsigned long mapsize, void *buffer,
1591 unsigned long bufsiz, struct strbuf *header)
1592{
1593 int status;
1594
1595 status = unpack_sha1_header(stream, map, mapsize, buffer, bufsiz);
1596
1597 /*
1598 * Check if entire header is unpacked in the first iteration.
1599 */
1600 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1601 return 0;
1602
1603 /*
1604 * buffer[0..bufsiz] was not large enough. Copy the partial
1605 * result out to header, and then append the result of further
1606 * reading the stream.
1607 */
1608 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1609 stream->next_out = buffer;
1610 stream->avail_out = bufsiz;
1611
1612 do {
1613 status = git_inflate(stream, 0);
1614 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1615 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1616 return 0;
1617 stream->next_out = buffer;
1618 stream->avail_out = bufsiz;
1619 } while (status != Z_STREAM_END);
1620 return -1;
1621}
1622
ef49a7a0 1623static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1624{
1625 int bytes = strlen(buffer) + 1;
3aee68aa 1626 unsigned char *buf = xmallocz(size);
93821bd9 1627 unsigned long n;
7efbff75 1628 int status = Z_OK;
5180cacc 1629
93821bd9
LT
1630 n = stream->total_out - bytes;
1631 if (n > size)
1632 n = size;
1633 memcpy(buf, (char *) buffer + bytes, n);
1634 bytes = n;
456cdf6e
LT
1635 if (bytes <= size) {
1636 /*
1637 * The above condition must be (bytes <= size), not
1638 * (bytes < size). In other words, even though we
ccf5ace0 1639 * expect no more output and set avail_out to zero,
456cdf6e
LT
1640 * the input zlib stream may have bytes that express
1641 * "this concludes the stream", and we *do* want to
1642 * eat that input.
1643 *
1644 * Otherwise we would not be able to test that we
1645 * consumed all the input to reach the expected size;
1646 * we also want to check that zlib tells us that all
1647 * went well with status == Z_STREAM_END at the end.
1648 */
5180cacc
LT
1649 stream->next_out = buf + bytes;
1650 stream->avail_out = size - bytes;
7efbff75 1651 while (status == Z_OK)
39c68542 1652 status = git_inflate(stream, Z_FINISH);
5180cacc 1653 }
456cdf6e 1654 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1655 git_inflate_end(stream);
7efbff75
JH
1656 return buf;
1657 }
1658
1659 if (status < 0)
1660 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1661 else if (stream->avail_in)
1662 error("garbage at end of loose object '%s'",
1663 sha1_to_hex(sha1));
1664 free(buf);
1665 return NULL;
5180cacc
LT
1666}
1667
1668/*
1669 * We used to just use "sscanf()", but that's actually way
1670 * too permissive for what we want to check. So do an anal
1671 * object header parse by hand.
1672 */
46f03448
KN
1673static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1674 unsigned int flags)
5180cacc 1675{
46f03448 1676 const char *type_buf = hdr;
5180cacc 1677 unsigned long size;
46f03448 1678 int type, type_len = 0;
5180cacc
LT
1679
1680 /*
46f03448 1681 * The type can be of any size but is followed by
21666f1a 1682 * a space.
5180cacc 1683 */
5180cacc
LT
1684 for (;;) {
1685 char c = *hdr++;
1686 if (c == ' ')
1687 break;
46f03448 1688 type_len++;
5180cacc 1689 }
46f03448
KN
1690
1691 type = type_from_string_gently(type_buf, type_len, 1);
1692 if (oi->typename)
1693 strbuf_add(oi->typename, type_buf, type_len);
1694 /*
1695 * Set type to 0 if its an unknown object and
2e3a16b2 1696 * we're obtaining the type using '--allow-unknown-type'
46f03448
KN
1697 * option.
1698 */
1699 if ((flags & LOOKUP_UNKNOWN_OBJECT) && (type < 0))
1700 type = 0;
1701 else if (type < 0)
1702 die("invalid object type");
1703 if (oi->typep)
1704 *oi->typep = type;
5180cacc
LT
1705
1706 /*
1707 * The length must follow immediately, and be in canonical
1708 * decimal format (ie "010" is not valid).
1709 */
1710 size = *hdr++ - '0';
1711 if (size > 9)
1712 return -1;
1713 if (size) {
1714 for (;;) {
1715 unsigned long c = *hdr - '0';
1716 if (c > 9)
1717 break;
1718 hdr++;
1719 size = size * 10 + c;
1720 }
1721 }
46f03448
KN
1722
1723 if (oi->sizep)
1724 *oi->sizep = size;
5180cacc
LT
1725
1726 /*
1727 * The length must be followed by a zero byte
1728 */
46f03448
KN
1729 return *hdr ? -1 : type;
1730}
1731
1732int parse_sha1_header(const char *hdr, unsigned long *sizep)
1733{
1734 struct object_info oi;
1735
1736 oi.sizep = sizep;
1737 oi.typename = NULL;
1738 oi.typep = NULL;
1739 return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
5180cacc
LT
1740}
1741
7efbff75 1742static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1743{
5180cacc 1744 int ret;
ef49a7a0 1745 git_zstream stream;
5180cacc 1746 char hdr[8192];
0fcfd160 1747
5180cacc 1748 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1749 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1750 return NULL;
1751
7efbff75 1752 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1753}
1754
54dab52a
NP
1755unsigned long get_size_from_delta(struct packed_git *p,
1756 struct pack_window **w_curs,
1757 off_t curpos)
1758{
1759 const unsigned char *data;
1760 unsigned char delta_head[20], *in;
ef49a7a0 1761 git_zstream stream;
54dab52a
NP
1762 int st;
1763
1764 memset(&stream, 0, sizeof(stream));
1765 stream.next_out = delta_head;
1766 stream.avail_out = sizeof(delta_head);
1767
39c68542 1768 git_inflate_init(&stream);
54dab52a
NP
1769 do {
1770 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1771 stream.next_in = in;
39c68542 1772 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1773 curpos += stream.next_in - in;
1774 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1775 stream.total_out < sizeof(delta_head));
39c68542 1776 git_inflate_end(&stream);
3d77d877
NP
1777 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1778 error("delta data unpack-initial failed");
1779 return 0;
1780 }
54dab52a
NP
1781
1782 /* Examine the initial part of the delta to figure out
1783 * the result size.
1784 */
1785 data = delta_head;
1786
1787 /* ignore base size */
1788 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1789
1790 /* Read the result size */
1791 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1792}
1793
c4001d92 1794static off_t get_delta_base(struct packed_git *p,
03e79c88 1795 struct pack_window **w_curs,
c4001d92 1796 off_t *curpos,
21666f1a 1797 enum object_type type,
c4001d92 1798 off_t delta_obj_offset)
eb32d236 1799{
2b87c45b 1800 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1801 off_t base_offset;
eb32d236 1802
8d8a4ea5
SP
1803 /* use_pack() assured us we have [base_info, base_info + 20)
1804 * as a range that we can look at without walking off the
1805 * end of the mapped window. Its actually the hash size
1806 * that is assured. An OFS_DELTA longer than the hash size
1807 * is stupid, as then a REF_DELTA would be smaller to store.
1808 */
21666f1a 1809 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1810 unsigned used = 0;
1811 unsigned char c = base_info[used++];
1812 base_offset = c & 127;
1813 while (c & 128) {
1814 base_offset += 1;
8723f216 1815 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1816 return 0; /* overflow */
eb32d236
NP
1817 c = base_info[used++];
1818 base_offset = (base_offset << 7) + (c & 127);
1819 }
1820 base_offset = delta_obj_offset - base_offset;
d8f32556 1821 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1822 return 0; /* out of bound */
2b87c45b 1823 *curpos += used;
21666f1a 1824 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1825 /* The base entry _must_ be in the same pack */
1826 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1827 *curpos += 20;
eb32d236
NP
1828 } else
1829 die("I am totally screwed");
2b87c45b 1830 return base_offset;
eb32d236
NP
1831}
1832
5d642e75
JK
1833/*
1834 * Like get_delta_base above, but we return the sha1 instead of the pack
1835 * offset. This means it is cheaper for REF deltas (we do not have to do
1836 * the final object lookup), but more expensive for OFS deltas (we
1837 * have to load the revidx to convert the offset back into a sha1).
1838 */
1839static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1840 struct pack_window **w_curs,
1841 off_t curpos,
1842 enum object_type type,
1843 off_t delta_obj_offset)
1844{
1845 if (type == OBJ_REF_DELTA) {
1846 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1847 return base;
1848 } else if (type == OBJ_OFS_DELTA) {
1849 struct revindex_entry *revidx;
1850 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1851 type, delta_obj_offset);
1852
1853 if (!base_offset)
1854 return NULL;
1855
1856 revidx = find_pack_revindex(p, base_offset);
1857 if (!revidx)
1858 return NULL;
1859
1860 return nth_packed_object_sha1(p, revidx->nr);
1861 } else
1862 return NULL;
1863}
1864
f8c8abc5
JH
1865int unpack_object_header(struct packed_git *p,
1866 struct pack_window **w_curs,
1867 off_t *curpos,
1868 unsigned long *sizep)
a733cb60 1869{
03e79c88 1870 unsigned char *base;
ef49a7a0 1871 unsigned long left;
ad1ed5ee 1872 unsigned long used;
2b87c45b 1873 enum object_type type;
a733cb60 1874
8d8a4ea5 1875 /* use_pack() assures us we have [base, base + 20) available
0353a0c4 1876 * as a range that we can look at. (Its actually the hash
3dff5379 1877 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1878 * the maximum deflated object size is 2^137, which is just
1879 * insane, so we know won't exceed what we have been given.
1880 */
2b87c45b 1881 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1882 used = unpack_object_header_buffer(base, left, &type, sizep);
1883 if (!used) {
1884 type = OBJ_BAD;
1885 } else
1886 *curpos += used;
ad1ed5ee 1887
2b87c45b 1888 return type;
a733cb60
LT
1889}
1890
790d96c0
TR
1891static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1892{
1893 int type;
1894 struct revindex_entry *revidx;
1895 const unsigned char *sha1;
1896 revidx = find_pack_revindex(p, obj_offset);
1897 if (!revidx)
1898 return OBJ_BAD;
1899 sha1 = nth_packed_object_sha1(p, revidx->nr);
1900 mark_bad_packed_object(p, sha1);
1901 type = sha1_object_info(sha1, NULL);
1902 if (type <= OBJ_NONE)
1903 return OBJ_BAD;
1904 return type;
1905}
1906
790d96c0
TR
1907#define POI_STACK_PREALLOC 64
1908
90191d37
JK
1909static enum object_type packed_to_object_type(struct packed_git *p,
1910 off_t obj_offset,
1911 enum object_type type,
1912 struct pack_window **w_curs,
1913 off_t curpos)
1f688557 1914{
790d96c0
TR
1915 off_t small_poi_stack[POI_STACK_PREALLOC];
1916 off_t *poi_stack = small_poi_stack;
1917 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
5db47c2b 1918
790d96c0
TR
1919 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1920 off_t base_offset;
90191d37 1921 unsigned long size;
790d96c0
TR
1922 /* Push the object we're going to leave behind */
1923 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1924 poi_stack_alloc = alloc_nr(poi_stack_nr);
b32fa95f 1925 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
790d96c0
TR
1926 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1927 } else {
1928 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1929 }
1930 poi_stack[poi_stack_nr++] = obj_offset;
1931 /* If parsing the base offset fails, just unwind */
90191d37 1932 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
790d96c0
TR
1933 if (!base_offset)
1934 goto unwind;
1935 curpos = obj_offset = base_offset;
90191d37 1936 type = unpack_object_header(p, w_curs, &curpos, &size);
790d96c0
TR
1937 if (type <= OBJ_NONE) {
1938 /* If getting the base itself fails, we first
1939 * retry the base, otherwise unwind */
1940 type = retry_bad_packed_offset(p, base_offset);
1941 if (type > OBJ_NONE)
1942 goto out;
1943 goto unwind;
1944 }
1945 }
1946
21666f1a 1947 switch (type) {
790d96c0 1948 case OBJ_BAD:
a733cb60 1949 case OBJ_COMMIT:
a733cb60 1950 case OBJ_TREE:
a733cb60 1951 case OBJ_BLOB:
a733cb60 1952 case OBJ_TAG:
a69d0943 1953 break;
1f688557 1954 default:
3d77d877
NP
1955 error("unknown object type %i at offset %"PRIuMAX" in %s",
1956 type, (uintmax_t)obj_offset, p->pack_name);
1957 type = OBJ_BAD;
1f688557 1958 }
790d96c0
TR
1959
1960out:
1961 if (poi_stack != small_poi_stack)
1962 free(poi_stack);
21666f1a 1963 return type;
790d96c0
TR
1964
1965unwind:
1966 while (poi_stack_nr) {
1967 obj_offset = poi_stack[--poi_stack_nr];
1968 type = retry_bad_packed_offset(p, obj_offset);
1969 if (type > OBJ_NONE)
1970 goto out;
1971 }
1972 type = OBJ_BAD;
1973 goto out;
1f688557
JH
1974}
1975
90191d37 1976static int packed_object_info(struct packed_git *p, off_t obj_offset,
23c339c0 1977 struct object_info *oi)
90191d37
JK
1978{
1979 struct pack_window *w_curs = NULL;
1980 unsigned long size;
1981 off_t curpos = obj_offset;
1982 enum object_type type;
1983
412916ee
JK
1984 /*
1985 * We always get the representation type, but only convert it to
1986 * a "real" type later if the caller is interested.
1987 */
90191d37
JK
1988 type = unpack_object_header(p, &w_curs, &curpos, &size);
1989
23c339c0 1990 if (oi->sizep) {
90191d37
JK
1991 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1992 off_t tmp_pos = curpos;
1993 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1994 type, obj_offset);
1995 if (!base_offset) {
1996 type = OBJ_BAD;
1997 goto out;
1998 }
23c339c0
JK
1999 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2000 if (*oi->sizep == 0) {
90191d37
JK
2001 type = OBJ_BAD;
2002 goto out;
2003 }
2004 } else {
23c339c0 2005 *oi->sizep = size;
90191d37
JK
2006 }
2007 }
2008
23c339c0 2009 if (oi->disk_sizep) {
90191d37 2010 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
23c339c0 2011 *oi->disk_sizep = revidx[1].offset - obj_offset;
90191d37
JK
2012 }
2013
23c339c0
JK
2014 if (oi->typep) {
2015 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
2016 if (*oi->typep < 0) {
412916ee
JK
2017 type = OBJ_BAD;
2018 goto out;
2019 }
2020 }
90191d37 2021
5d642e75
JK
2022 if (oi->delta_base_sha1) {
2023 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2024 const unsigned char *base;
2025
2026 base = get_delta_base_sha1(p, &w_curs, curpos,
2027 type, obj_offset);
2028 if (!base) {
2029 type = OBJ_BAD;
2030 goto out;
2031 }
2032
2033 hashcpy(oi->delta_base_sha1, base);
2034 } else
2035 hashclr(oi->delta_base_sha1);
2036 }
2037
90191d37
JK
2038out:
2039 unuse_pack(&w_curs);
2040 return type;
2041}
2042
eb950c19 2043static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 2044 struct pack_window **w_curs,
c4001d92 2045 off_t curpos,
eb950c19 2046 unsigned long size)
de530aaa
SP
2047{
2048 int st;
ef49a7a0 2049 git_zstream stream;
079afb18 2050 unsigned char *buffer, *in;
de530aaa 2051
735efde8
NTND
2052 buffer = xmallocz_gently(size);
2053 if (!buffer)
2054 return NULL;
de530aaa 2055 memset(&stream, 0, sizeof(stream));
de530aaa 2056 stream.next_out = buffer;
39eea7bd 2057 stream.avail_out = size + 1;
de530aaa 2058
39c68542 2059 git_inflate_init(&stream);
079afb18 2060 do {
2b87c45b 2061 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 2062 stream.next_in = in;
39c68542 2063 st = git_inflate(&stream, Z_FINISH);
39eea7bd
JH
2064 if (!stream.avail_out)
2065 break; /* the payload is larger than it should be */
2b87c45b 2066 curpos += stream.next_in - in;
079afb18 2067 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 2068 git_inflate_end(&stream);
de530aaa
SP
2069 if ((st != Z_STREAM_END) || stream.total_out != size) {
2070 free(buffer);
2071 return NULL;
2072 }
2073
2074 return buffer;
2075}
2076
e5e01619
LT
2077#define MAX_DELTA_CACHE (256)
2078
18bdec11 2079static size_t delta_base_cached;
5e08ecbf 2080
12d95ef6 2081static LIST_HEAD(delta_base_cache_lru);
5e08ecbf 2082
e5e01619 2083static struct delta_base_cache_entry {
12d95ef6 2084 struct list_head lru;
5e08ecbf 2085 void *data;
e5e01619
LT
2086 struct packed_git *p;
2087 off_t base_offset;
2088 unsigned long size;
e5e01619
LT
2089 enum object_type type;
2090} delta_base_cache[MAX_DELTA_CACHE];
2091
2092static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
2093{
2094 unsigned long hash;
2095
56a1a3ab 2096 hash = (unsigned long)(intptr_t)p + (unsigned long)base_offset;
e5e01619 2097 hash += (hash >> 8) + (hash >> 16);
3358004a 2098 return hash % MAX_DELTA_CACHE;
e5e01619
LT
2099}
2100
84dd81c1
TR
2101static struct delta_base_cache_entry *
2102get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
5266d369
JH
2103{
2104 unsigned long hash = pack_entry_hash(p, base_offset);
84dd81c1
TR
2105 return delta_base_cache + hash;
2106}
2107
2108static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
2109 struct packed_git *p, off_t base_offset)
2110{
5266d369
JH
2111 return (ent->data && ent->p == p && ent->base_offset == base_offset);
2112}
2113
84dd81c1
TR
2114static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2115{
2116 struct delta_base_cache_entry *ent;
2117 ent = get_delta_base_cache_entry(p, base_offset);
2118 return eq_delta_base_cache_entry(ent, p, base_offset);
2119}
2120
4a5397ca
JK
2121/*
2122 * Remove the entry from the cache, but do _not_ free the associated
2123 * entry data. The caller takes ownership of the "data" buffer, and
2124 * should copy out any fields it wants before detaching.
2125 */
2126static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
84dd81c1
TR
2127{
2128 ent->data = NULL;
12d95ef6 2129 list_del(&ent->lru);
84dd81c1
TR
2130 delta_base_cached -= ent->size;
2131}
2132
62f255ad 2133static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
85fe35ab 2134 unsigned long *base_size, enum object_type *type)
62f255ad 2135{
84dd81c1 2136 struct delta_base_cache_entry *ent;
e5e01619 2137
84dd81c1
TR
2138 ent = get_delta_base_cache_entry(p, base_offset);
2139
2140 if (!eq_delta_base_cache_entry(ent, p, base_offset))
749bc58c 2141 return unpack_entry(p, base_offset, type, base_size);
e5e01619 2142
e5e01619
LT
2143 *type = ent->type;
2144 *base_size = ent->size;
85fe35ab 2145 return xmemdupz(ent->data, ent->size);
62f255ad
LT
2146}
2147
18bdec11
SP
2148static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2149{
2150 if (ent->data) {
2151 free(ent->data);
f92dd60f 2152 detach_delta_base_cache_entry(ent);
18bdec11
SP
2153 }
2154}
2155
3d20c636
SP
2156void clear_delta_base_cache(void)
2157{
2158 unsigned long p;
2159 for (p = 0; p < MAX_DELTA_CACHE; p++)
2160 release_delta_base_cache(&delta_base_cache[p]);
2161}
2162
62f255ad
LT
2163static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2164 void *base, unsigned long base_size, enum object_type type)
2165{
5e08ecbf 2166 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 2167 struct delta_base_cache_entry *ent = delta_base_cache + hash;
12d95ef6 2168 struct list_head *lru;
e5e01619 2169
18bdec11
SP
2170 release_delta_base_cache(ent);
2171 delta_base_cached += base_size;
5e08ecbf 2172
12d95ef6
JK
2173 list_for_each(lru, &delta_base_cache_lru) {
2174 struct delta_base_cache_entry *f =
2175 list_entry(lru, struct delta_base_cache_entry, lru);
2176 if (delta_base_cached <= delta_base_cache_limit)
2177 break;
5e08ecbf
NP
2178 release_delta_base_cache(f);
2179 }
18bdec11 2180
e5e01619
LT
2181 ent->p = p;
2182 ent->base_offset = base_offset;
2183 ent->type = type;
2184 ent->data = base;
2185 ent->size = base_size;
12d95ef6 2186 list_add_tail(&ent->lru, &delta_base_cache_lru);
62f255ad
LT
2187}
2188
c2c5b270
CC
2189static void *read_object(const unsigned char *sha1, enum object_type *type,
2190 unsigned long *size);
2191
5f44324d
JH
2192static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2193{
67dc598e
KB
2194 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2195 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2196 p->pack_name, (uintmax_t)obj_offset);
5f44324d
JH
2197}
2198
0e8189e2
NP
2199int do_check_packed_object_crc;
2200
abe601bb
TR
2201#define UNPACK_ENTRY_STACK_PREALLOC 64
2202struct unpack_entry_stack_ent {
2203 off_t obj_offset;
2204 off_t curpos;
2205 unsigned long size;
2206};
2207
c4001d92 2208void *unpack_entry(struct packed_git *p, off_t obj_offset,
abe601bb 2209 enum object_type *final_type, unsigned long *final_size)
1f688557 2210{
03e79c88 2211 struct pack_window *w_curs = NULL;
c4001d92 2212 off_t curpos = obj_offset;
abe601bb
TR
2213 void *data = NULL;
2214 unsigned long size;
2215 enum object_type type;
2216 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2217 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2218 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2219 int base_from_cache = 0;
1f688557 2220
67dc598e 2221 write_pack_access_log(p, obj_offset);
5f44324d 2222
abe601bb
TR
2223 /* PHASE 1: drill down to the innermost base object */
2224 for (;;) {
2225 off_t base_offset;
2226 int i;
2227 struct delta_base_cache_entry *ent;
2228
77965f8b
NTND
2229 ent = get_delta_base_cache_entry(p, curpos);
2230 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2231 type = ent->type;
2232 data = ent->data;
2233 size = ent->size;
4a5397ca 2234 detach_delta_base_cache_entry(ent);
77965f8b
NTND
2235 base_from_cache = 1;
2236 break;
2237 }
2238
abe601bb
TR
2239 if (do_check_packed_object_crc && p->index_version > 1) {
2240 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
211c61c6 2241 off_t len = revidx[1].offset - obj_offset;
abe601bb
TR
2242 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2243 const unsigned char *sha1 =
2244 nth_packed_object_sha1(p, revidx->nr);
2245 error("bad packed object CRC for %s",
2246 sha1_to_hex(sha1));
2247 mark_bad_packed_object(p, sha1);
2248 unuse_pack(&w_curs);
2249 return NULL;
2250 }
2251 }
2252
abe601bb
TR
2253 type = unpack_object_header(p, &w_curs, &curpos, &size);
2254 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2255 break;
2256
2257 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2258 if (!base_offset) {
2259 error("failed to validate delta base reference "
2260 "at offset %"PRIuMAX" from %s",
2261 (uintmax_t)curpos, p->pack_name);
2262 /* bail to phase 2, in hopes of recovery */
2263 data = NULL;
2264 break;
2265 }
2266
2267 /* push object, proceed to base */
2268 if (delta_stack_nr >= delta_stack_alloc
2269 && delta_stack == small_delta_stack) {
2270 delta_stack_alloc = alloc_nr(delta_stack_nr);
b32fa95f 2271 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
abe601bb
TR
2272 memcpy(delta_stack, small_delta_stack,
2273 sizeof(*delta_stack)*delta_stack_nr);
2274 } else {
2275 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2276 }
2277 i = delta_stack_nr++;
2278 delta_stack[i].obj_offset = obj_offset;
2279 delta_stack[i].curpos = curpos;
2280 delta_stack[i].size = size;
2281
2282 curpos = obj_offset = base_offset;
0e8189e2
NP
2283 }
2284
abe601bb
TR
2285 /* PHASE 2: handle the base */
2286 switch (type) {
eb32d236
NP
2287 case OBJ_OFS_DELTA:
2288 case OBJ_REF_DELTA:
abe601bb 2289 if (data)
ef1177d1 2290 die("BUG: unpack_entry: left loop at a valid delta");
4d703a1a 2291 break;
a733cb60 2292 case OBJ_COMMIT:
a733cb60 2293 case OBJ_TREE:
a733cb60 2294 case OBJ_BLOB:
a733cb60 2295 case OBJ_TAG:
abe601bb
TR
2296 if (!base_from_cache)
2297 data = unpack_compressed_entry(p, &w_curs, curpos, size);
4d703a1a 2298 break;
1f688557 2299 default:
8eca0b47
NP
2300 data = NULL;
2301 error("unknown object type %i at offset %"PRIuMAX" in %s",
abe601bb
TR
2302 type, (uintmax_t)obj_offset, p->pack_name);
2303 }
2304
2305 /* PHASE 3: apply deltas in order */
2306
2307 /* invariants:
2308 * 'data' holds the base data, or NULL if there was corruption
2309 */
2310 while (delta_stack_nr) {
2311 void *delta_data;
2312 void *base = data;
2313 unsigned long delta_size, base_size = size;
2314 int i;
2315
2316 data = NULL;
2317
2318 if (base)
2319 add_delta_base_cache(p, obj_offset, base, base_size, type);
2320
2321 if (!base) {
2322 /*
2323 * We're probably in deep shit, but let's try to fetch
2324 * the required base anyway from another pack or loose.
2325 * This is costly but should happen only in the presence
2326 * of a corrupted pack, and is better than failing outright.
2327 */
2328 struct revindex_entry *revidx;
2329 const unsigned char *base_sha1;
2330 revidx = find_pack_revindex(p, obj_offset);
2331 if (revidx) {
2332 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2333 error("failed to read delta base object %s"
2334 " at offset %"PRIuMAX" from %s",
2335 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2336 p->pack_name);
2337 mark_bad_packed_object(p, base_sha1);
2338 base = read_object(base_sha1, &type, &base_size);
2339 }
2340 }
2341
2342 i = --delta_stack_nr;
2343 obj_offset = delta_stack[i].obj_offset;
2344 curpos = delta_stack[i].curpos;
2345 delta_size = delta_stack[i].size;
2346
2347 if (!base)
2348 continue;
2349
2350 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2351
2352 if (!delta_data) {
2353 error("failed to unpack compressed delta "
2354 "at offset %"PRIuMAX" from %s",
2355 (uintmax_t)curpos, p->pack_name);
abe601bb
TR
2356 data = NULL;
2357 continue;
2358 }
2359
2360 data = patch_delta(base, base_size,
2361 delta_data, delta_size,
2362 &size);
1ee886c1
JK
2363
2364 /*
2365 * We could not apply the delta; warn the user, but keep going.
2366 * Our failure will be noticed either in the next iteration of
2367 * the loop, or if this is the final delta, in the caller when
2368 * we return NULL. Those code paths will take care of making
2369 * a more explicit warning and retrying with another copy of
2370 * the object.
2371 */
abe601bb 2372 if (!data)
1ee886c1 2373 error("failed to apply delta");
abe601bb 2374
4b8f772c 2375 free(delta_data);
1f688557 2376 }
abe601bb
TR
2377
2378 *final_type = type;
2379 *final_size = size;
2380
03e79c88 2381 unuse_pack(&w_curs);
019d1e65
NTND
2382
2383 if (delta_stack != small_delta_stack)
2384 free(delta_stack);
2385
21666f1a 2386 return data;
1f688557
JH
2387}
2388
d079837e 2389const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 2390 uint32_t n)
9a217f2a 2391{
42873078 2392 const unsigned char *index = p->index_data;
d079837e
SP
2393 if (!index) {
2394 if (open_pack_index(p))
2395 return NULL;
2396 index = p->index_data;
2397 }
57059091 2398 if (n >= p->num_objects)
d72308e0 2399 return NULL;
74e34e1f
NP
2400 index += 4 * 256;
2401 if (p->index_version == 1) {
2402 return index + 24 * n + 4;
2403 } else {
2404 index += 8;
2405 return index + 20 * n;
2406 }
2407}
2408
47fe3f6e
JK
2409void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
2410{
2411 const unsigned char *ptr = vptr;
2412 const unsigned char *start = p->index_data;
2413 const unsigned char *end = start + p->index_size;
2414 if (ptr < start)
7465feba 2415 die(_("offset before start of pack index for %s (corrupt index?)"),
47fe3f6e
JK
2416 p->pack_name);
2417 /* No need to check for underflow; .idx files must be at least 8 bytes */
2418 if (ptr >= end - 8)
7465feba 2419 die(_("offset beyond end of pack index for %s (truncated index?)"),
47fe3f6e
JK
2420 p->pack_name);
2421}
2422
99093238 2423off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
2424{
2425 const unsigned char *index = p->index_data;
2426 index += 4 * 256;
2427 if (p->index_version == 1) {
2428 return ntohl(*((uint32_t *)(index + 24 * n)));
2429 } else {
2430 uint32_t off;
2431 index += 8 + p->num_objects * (20 + 4);
2432 off = ntohl(*((uint32_t *)(index + 4 * n)));
2433 if (!(off & 0x80000000))
2434 return off;
2435 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
47fe3f6e 2436 check_pack_index_ptr(p, index);
74e34e1f
NP
2437 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2438 ntohl(*((uint32_t *)(index + 4)));
2439 }
9a217f2a
JH
2440}
2441
c4001d92 2442off_t find_pack_entry_one(const unsigned char *sha1,
43057304 2443 struct packed_git *p)
1f688557 2444{
42873078 2445 const uint32_t *level1_ofs = p->index_data;
42873078 2446 const unsigned char *index = p->index_data;
628522ec
JH
2447 unsigned hi, lo, stride;
2448 static int use_lookup = -1;
2449 static int debug_lookup = -1;
2450
2451 if (debug_lookup < 0)
2452 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 2453
d079837e
SP
2454 if (!index) {
2455 if (open_pack_index(p))
2456 return 0;
2457 level1_ofs = p->index_data;
2458 index = p->index_data;
2459 }
74e34e1f
NP
2460 if (p->index_version > 1) {
2461 level1_ofs += 2;
2462 index += 8;
2463 }
42873078 2464 index += 4 * 256;
74e34e1f
NP
2465 hi = ntohl(level1_ofs[*sha1]);
2466 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
2467 if (p->index_version > 1) {
2468 stride = 20;
2469 } else {
2470 stride = 24;
2471 index += 4;
2472 }
2473
2474 if (debug_lookup)
6e1c2344 2475 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
2476 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2477
2478 if (use_lookup < 0)
2479 use_lookup = !!getenv("GIT_USE_LOOKUP");
2480 if (use_lookup) {
2481 int pos = sha1_entry_pos(index, stride, 0,
2482 lo, hi, p->num_objects, sha1);
2483 if (pos < 0)
2484 return 0;
2485 return nth_packed_object_offset(p, pos);
2486 }
1f688557
JH
2487
2488 do {
74e34e1f 2489 unsigned mi = (lo + hi) / 2;
628522ec
JH
2490 int cmp = hashcmp(index + mi * stride, sha1);
2491
2492 if (debug_lookup)
2493 printf("lo %u hi %u rg %u mi %u\n",
2494 lo, hi, hi - lo, mi);
43057304 2495 if (!cmp)
74e34e1f 2496 return nth_packed_object_offset(p, mi);
1f688557
JH
2497 if (cmp > 0)
2498 hi = mi;
2499 else
2500 lo = mi+1;
2501 } while (lo < hi);
2502 return 0;
2503}
2504
4c080182 2505int is_pack_valid(struct packed_git *p)
d131b7af
SP
2506{
2507 /* An already open pack is known to be valid. */
2508 if (p->pack_fd != -1)
2509 return 1;
2510
2511 /* If the pack has one window completely covering the
2512 * file size, the pack is known to be valid even if
2513 * the descriptor is not currently open.
2514 */
2515 if (p->windows) {
2516 struct pack_window *w = p->windows;
2517
2518 if (!w->offset && w->len == p->pack_size)
2519 return 1;
2520 }
2521
2522 /* Force the pack to open to prove its valid. */
2523 return !open_packed_git(p);
2524}
2525
95099731
NTND
2526static int fill_pack_entry(const unsigned char *sha1,
2527 struct pack_entry *e,
2528 struct packed_git *p)
2529{
2530 off_t offset;
2531
2532 if (p->num_bad_objects) {
2533 unsigned i;
2534 for (i = 0; i < p->num_bad_objects; i++)
2535 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2536 return 0;
2537 }
2538
2539 offset = find_pack_entry_one(sha1, p);
2540 if (!offset)
2541 return 0;
2542
2543 /*
2544 * We are about to tell the caller where they can locate the
2545 * requested object. We better make sure the packfile is
2546 * still here and can be accessed before supplying that
2547 * answer, as it may have been deleted since the index was
2548 * loaded!
2549 */
319b678a 2550 if (!is_pack_valid(p))
95099731 2551 return 0;
95099731
NTND
2552 e->offset = offset;
2553 e->p = p;
2554 hashcpy(e->sha1, sha1);
2555 return 1;
2556}
2557
d40d535b
MH
2558/*
2559 * Iff a pack file contains the object named by sha1, return true and
2560 * store its location to e.
2561 */
4d6acb70 2562static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557 2563{
a73cdd21 2564 struct mru_entry *p;
43057304 2565
1f688557 2566 prepare_packed_git();
f7c22cc6
NP
2567 if (!packed_git)
2568 return 0;
1f688557 2569
a73cdd21
JK
2570 for (p = packed_git_mru->head; p; p = p->next) {
2571 if (fill_pack_entry(sha1, e, p->item)) {
2572 mru_mark(packed_git_mru, p);
1b1005d1
MH
2573 return 1;
2574 }
c01f51cc 2575 }
1f688557
JH
2576 return 0;
2577}
2578
a6080a0a 2579struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
2580 struct packed_git *packs)
2581{
2582 struct packed_git *p;
bf592c50
DB
2583
2584 for (p = packs; p; p = p->next) {
43057304 2585 if (find_pack_entry_one(sha1, p))
bf592c50
DB
2586 return p;
2587 }
2588 return NULL;
21666f1a 2589
bf592c50
DB
2590}
2591
052fe5ea 2592static int sha1_loose_object_info(const unsigned char *sha1,
46f03448
KN
2593 struct object_info *oi,
2594 int flags)
65c2e0c3 2595{
46f03448
KN
2596 int status = 0;
2597 unsigned long mapsize;
65c2e0c3 2598 void *map;
ef49a7a0 2599 git_zstream stream;
d65a16f6 2600 char hdr[32];
46f03448 2601 struct strbuf hdrbuf = STRBUF_INIT;
65c2e0c3 2602
5d642e75
JK
2603 if (oi->delta_base_sha1)
2604 hashclr(oi->delta_base_sha1);
2605
052fe5ea
JK
2606 /*
2607 * If we don't care about type or size, then we don't
4ef8d1dd
JH
2608 * need to look inside the object at all. Note that we
2609 * do not optimize out the stat call, even if the
2610 * caller doesn't care about the disk-size, since our
2611 * return value implicitly indicates whether the
2612 * object even exists.
052fe5ea 2613 */
46f03448 2614 if (!oi->typep && !oi->typename && !oi->sizep) {
4ef8d1dd
JH
2615 struct stat st;
2616 if (stat_sha1_file(sha1, &st) < 0)
2617 return -1;
2618 if (oi->disk_sizep)
23c339c0 2619 *oi->disk_sizep = st.st_size;
052fe5ea
JK
2620 return 0;
2621 }
2622
bb6b8e4f 2623 map = map_sha1_file(sha1, &mapsize);
f0df4ed5 2624 if (!map)
dbea72a8 2625 return -1;
23c339c0
JK
2626 if (oi->disk_sizep)
2627 *oi->disk_sizep = mapsize;
46f03448
KN
2628 if ((flags & LOOKUP_UNKNOWN_OBJECT)) {
2629 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2630 status = error("unable to unpack %s header with --allow-unknown-type",
2631 sha1_to_hex(sha1));
2632 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
36e4d74a
JH
2633 status = error("unable to unpack %s header",
2634 sha1_to_hex(sha1));
46f03448
KN
2635 if (status < 0)
2636 ; /* Do nothing */
2637 else if (hdrbuf.len) {
2638 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2639 status = error("unable to parse %s header with --allow-unknown-type",
2640 sha1_to_hex(sha1));
2641 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
36e4d74a 2642 status = error("unable to parse %s header", sha1_to_hex(sha1));
39c68542 2643 git_inflate_end(&stream);
65c2e0c3 2644 munmap(map, mapsize);
46f03448 2645 if (status && oi->typep)
23c339c0 2646 *oi->typep = status;
46f03448 2647 strbuf_release(&hdrbuf);
052fe5ea 2648 return 0;
65c2e0c3
JH
2649}
2650
de7b5d62 2651int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
f0df4ed5 2652{
c4d9986f 2653 struct cached_object *co;
f0df4ed5 2654 struct pack_entry e;
5b086407 2655 int rtype;
46f03448 2656 enum object_type real_type;
1f7117ef 2657 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
f0df4ed5 2658
1f7117ef 2659 co = find_cached_object(real);
c4d9986f 2660 if (co) {
5b086407
JK
2661 if (oi->typep)
2662 *(oi->typep) = co->type;
9a490590
JH
2663 if (oi->sizep)
2664 *(oi->sizep) = co->size;
161f00e7
JK
2665 if (oi->disk_sizep)
2666 *(oi->disk_sizep) = 0;
5d642e75
JK
2667 if (oi->delta_base_sha1)
2668 hashclr(oi->delta_base_sha1);
46f03448
KN
2669 if (oi->typename)
2670 strbuf_addstr(oi->typename, typename(co->type));
9a490590 2671 oi->whence = OI_CACHED;
5b086407 2672 return 0;
c4d9986f
NTND
2673 }
2674
1f7117ef 2675 if (!find_pack_entry(real, &e)) {
ddd63e64 2676 /* Most likely it's a loose object. */
46f03448 2677 if (!sha1_loose_object_info(real, oi, flags)) {
9a490590 2678 oi->whence = OI_LOOSE;
5b086407 2679 return 0;
9a490590 2680 }
ddd63e64
SG
2681
2682 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 2683 reprepare_packed_git();
1f7117ef 2684 if (!find_pack_entry(real, &e))
052fe5ea 2685 return -1;
f0df4ed5 2686 }
3d77d877 2687
46f03448
KN
2688 /*
2689 * packed_object_info() does not follow the delta chain to
2690 * find out the real type, unless it is given oi->typep.
2691 */
2692 if (oi->typename && !oi->typep)
2693 oi->typep = &real_type;
2694
23c339c0 2695 rtype = packed_object_info(e.p, e.offset, oi);
412916ee 2696 if (rtype < 0) {
1f7117ef 2697 mark_bad_packed_object(e.p, real);
46f03448
KN
2698 if (oi->typep == &real_type)
2699 oi->typep = NULL;
1f7117ef 2700 return sha1_object_info_extended(real, oi, 0);
5266d369
JH
2701 } else if (in_delta_base_cache(e.p, e.offset)) {
2702 oi->whence = OI_DBCACHED;
9a490590
JH
2703 } else {
2704 oi->whence = OI_PACKED;
2705 oi->u.packed.offset = e.offset;
2706 oi->u.packed.pack = e.p;
2707 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2708 rtype == OBJ_OFS_DELTA);
3d77d877 2709 }
46f03448
KN
2710 if (oi->typename)
2711 strbuf_addstr(oi->typename, typename(*oi->typep));
2712 if (oi->typep == &real_type)
2713 oi->typep = NULL;
3d77d877 2714
5b086407 2715 return 0;
f0df4ed5
JS
2716}
2717
3fc0dca9 2718/* returns enum object_type or negative */
9a490590
JH
2719int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2720{
5b086407 2721 enum object_type type;
d099b717 2722 struct object_info oi = {NULL};
9a490590 2723
5b086407 2724 oi.typep = &type;
9a490590 2725 oi.sizep = sizep;
de7b5d62 2726 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
5b086407
JK
2727 return -1;
2728 return type;
9a490590
JH
2729}
2730
21666f1a
NP
2731static void *read_packed_sha1(const unsigned char *sha1,
2732 enum object_type *type, unsigned long *size)
1f688557
JH
2733{
2734 struct pack_entry e;
8eca0b47 2735 void *data;
1f688557 2736
cd673c1f 2737 if (!find_pack_entry(sha1, &e))
1f688557 2738 return NULL;
85fe35ab 2739 data = cache_or_unpack_entry(e.p, e.offset, size, type);
8eca0b47
NP
2740 if (!data) {
2741 /*
2742 * We're probably in deep shit, but let's try to fetch
2743 * the required object anyway from another pack or loose.
2744 * This should happen only in the presence of a corrupted
2745 * pack, and is better than failing outright.
2746 */
2747 error("failed to read object %s at offset %"PRIuMAX" from %s",
2748 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2749 mark_bad_packed_object(e.p, sha1);
ac939109 2750 data = read_object(sha1, type, size);
8eca0b47
NP
2751 }
2752 return data;
1f688557
JH
2753}
2754
21666f1a
NP
2755int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2756 unsigned char *sha1)
d66b37bb
JH
2757{
2758 struct cached_object *co;
2759
21666f1a 2760 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2761 if (has_sha1_file(sha1) || find_cached_object(sha1))
2762 return 0;
c7353967 2763 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
d66b37bb
JH
2764 co = &cached_objects[cached_object_nr++];
2765 co->size = len;
21666f1a 2766 co->type = type;
efa13f7b
JH
2767 co->buf = xmalloc(len);
2768 memcpy(co->buf, buf, len);
d66b37bb
JH
2769 hashcpy(co->sha1, sha1);
2770 return 0;
2771}
2772
c2c5b270
CC
2773static void *read_object(const unsigned char *sha1, enum object_type *type,
2774 unsigned long *size)
0fcfd160
LT
2775{
2776 unsigned long mapsize;
2777 void *map, *buf;
d66b37bb
JH
2778 struct cached_object *co;
2779
2780 co = find_cached_object(sha1);
2781 if (co) {
21666f1a 2782 *type = co->type;
d66b37bb 2783 *size = co->size;
182af834 2784 return xmemdupz(co->buf, co->size);
d66b37bb 2785 }
0fcfd160 2786
8276c007
PE
2787 buf = read_packed_sha1(sha1, type, size);
2788 if (buf)
2789 return buf;
bb6b8e4f 2790 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2791 if (map) {
7efbff75 2792 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2793 munmap(map, mapsize);
2794 return buf;
2795 }
637cdd9d 2796 reprepare_packed_git();
8276c007 2797 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2798}
2799
b6c4cecc
JH
2800/*
2801 * This function dies on corrupt objects; the callers who want to
2802 * deal with them should arrange to call read_object() and give error
2803 * messages themselves.
2804 */
5bf29b95
JH
2805void *read_sha1_file_extended(const unsigned char *sha1,
2806 enum object_type *type,
2807 unsigned long *size,
2808 unsigned flag)
ac939109 2809{
3ba7a065 2810 void *data;
b6c4cecc 2811 const struct packed_git *p;
bf93eea0 2812 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
b6c4cecc 2813
3ba7a065
JH
2814 errno = 0;
2815 data = read_object(repl, type, size);
4bbf5a26 2816 if (data)
b6c4cecc 2817 return data;
68095570 2818
25f3af3f 2819 if (errno && errno != ENOENT)
3ba7a065
JH
2820 die_errno("failed to read object %s", sha1_to_hex(sha1));
2821
68095570 2822 /* die if we replaced an object with one that does not exist */
b6c4cecc 2823 if (repl != sha1)
68095570
CC
2824 die("replacement %s not found for %s",
2825 sha1_to_hex(repl), sha1_to_hex(sha1));
2826
b6c4cecc 2827 if (has_loose_object(repl)) {
30d6c6ea
MH
2828 const char *path = sha1_file_name(sha1);
2829
b6c4cecc
JH
2830 die("loose object %s (stored in %s) is corrupt",
2831 sha1_to_hex(repl), path);
e8b15e61 2832 }
68095570 2833
b6c4cecc
JH
2834 if ((p = has_packed_and_bad(repl)) != NULL)
2835 die("packed object %s (stored in %s) is corrupt",
2836 sha1_to_hex(repl), p->pack_name);
f5552aee 2837
b6c4cecc 2838 return NULL;
ac939109
NP
2839}
2840
40469ee9 2841void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2842 const char *required_type_name,
40469ee9
JH
2843 unsigned long *size,
2844 unsigned char *actual_sha1_return)
f4913f91 2845{
21666f1a 2846 enum object_type type, required_type;
f4913f91
JH
2847 void *buffer;
2848 unsigned long isize;
40469ee9 2849 unsigned char actual_sha1[20];
f4913f91 2850
21666f1a 2851 required_type = type_from_string(required_type_name);
e702496e 2852 hashcpy(actual_sha1, sha1);
40469ee9
JH
2853 while (1) {
2854 int ref_length = -1;
2855 const char *ref_type = NULL;
f4913f91 2856
21666f1a 2857 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2858 if (!buffer)
2859 return NULL;
21666f1a 2860 if (type == required_type) {
40469ee9
JH
2861 *size = isize;
2862 if (actual_sha1_return)
e702496e 2863 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2864 return buffer;
2865 }
2866 /* Handle references */
21666f1a 2867 else if (type == OBJ_COMMIT)
40469ee9 2868 ref_type = "tree ";
21666f1a 2869 else if (type == OBJ_TAG)
40469ee9
JH
2870 ref_type = "object ";
2871 else {
2872 free(buffer);
2873 return NULL;
2874 }
2875 ref_length = strlen(ref_type);
f4913f91 2876
50974ec9
MK
2877 if (ref_length + 40 > isize ||
2878 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2879 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2880 free(buffer);
2881 return NULL;
2882 }
1cf58e72 2883 free(buffer);
40469ee9
JH
2884 /* Now we have the ID of the referred-to object in
2885 * actual_sha1. Check again. */
f4913f91 2886 }
f4913f91
JH
2887}
2888
ce9fbf16 2889static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2890 const char *type, unsigned char *sha1,
d65a16f6 2891 char *hdr, int *hdrlen)
d410c0f5 2892{
9126f009 2893 git_SHA_CTX c;
d410c0f5
JH
2894
2895 /* Generate the header */
ef1286d3 2896 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
d410c0f5
JH
2897
2898 /* Sha1.. */
9126f009
NP
2899 git_SHA1_Init(&c);
2900 git_SHA1_Update(&c, hdr, *hdrlen);
2901 git_SHA1_Update(&c, buf, len);
2902 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2903}
2904
230f1322 2905/*
5a688fe4 2906 * Move the just written object into its final resting place.
230f1322 2907 */
cb5add58 2908int finalize_object_file(const char *tmpfile, const char *filename)
230f1322 2909{
e32c0a9c 2910 int ret = 0;
5a688fe4 2911
348df166 2912 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
2913 goto try_rename;
2914 else if (link(tmpfile, filename))
e32c0a9c 2915 ret = errno;
7ebb6fca
LT
2916
2917 /*
2918 * Coda hack - coda doesn't like cross-directory links,
2919 * so we fall back to a rename, which will mean that it
2920 * won't be able to check collisions, but that's not a
2921 * big deal.
2922 *
2923 * The same holds for FAT formatted media.
2924 *
3be1f18e 2925 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
2926 * left to unlink.
2927 */
2928 if (ret && ret != EEXIST) {
be66a6c4 2929 try_rename:
7ebb6fca 2930 if (!rename(tmpfile, filename))
3be1f18e 2931 goto out;
9e48b389 2932 ret = errno;
230f1322 2933 }
691f1a28 2934 unlink_or_warn(tmpfile);
230f1322
LT
2935 if (ret) {
2936 if (ret != EEXIST) {
7616c6ca 2937 return error_errno("unable to write sha1 filename %s", filename);
230f1322
LT
2938 }
2939 /* FIXME!!! Collision check here ? */
2940 }
2941
3be1f18e 2942out:
5256b006 2943 if (adjust_shared_perm(filename))
5a688fe4 2944 return error("unable to set permission to '%s'", filename);
230f1322
LT
2945 return 0;
2946}
2947
4d548150
LT
2948static int write_buffer(int fd, const void *buf, size_t len)
2949{
d34cf19b 2950 if (write_in_full(fd, buf, len) < 0)
7616c6ca 2951 return error_errno("file write error");
4d548150
LT
2952 return 0;
2953}
2954
ce9fbf16 2955int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2956 unsigned char *sha1)
2957{
d65a16f6 2958 char hdr[32];
ef1286d3 2959 int hdrlen = sizeof(hdr);
abdc3fc8
RS
2960 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2961 return 0;
2962}
2963
e9039dd3
LT
2964/* Finalize a file on disk, and close it. */
2965static void close_sha1_file(int fd)
2966{
aafe9fba
LT
2967 if (fsync_object_files)
2968 fsync_or_die(fd, "sha1 file");
e9039dd3 2969 if (close(fd) != 0)
d824cbba 2970 die_errno("error when closing sha1 file");
e9039dd3
LT
2971}
2972
5723fe7e
LT
2973/* Size of directory component, including the ending '/' */
2974static inline int directory_size(const char *filename)
2975{
2976 const char *s = strrchr(filename, '/');
2977 if (!s)
2978 return 0;
2979 return s - filename + 1;
2980}
2981
2982/*
2983 * This creates a temporary file in the same directory as the final
2984 * 'filename'
2985 *
2986 * We want to avoid cross-directory filename renames, because those
2987 * can have problems on various filesystems (FAT, NFS, Coda).
2988 */
d4b3d11a 2989static int create_tmpfile(struct strbuf *tmp, const char *filename)
5723fe7e
LT
2990{
2991 int fd, dirlen = directory_size(filename);
2992
d4b3d11a
JK
2993 strbuf_reset(tmp);
2994 strbuf_add(tmp, filename, dirlen);
2995 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
2996 fd = git_mkstemp_mode(tmp->buf, 0444);
cbacbf4e 2997 if (fd < 0 && dirlen && errno == ENOENT) {
d4b3d11a
JK
2998 /*
2999 * Make sure the directory exists; note that the contents
3000 * of the buffer are undefined after mkstemp returns an
3001 * error, so we have to rewrite the whole buffer from
3002 * scratch.
3003 */
3004 strbuf_reset(tmp);
3005 strbuf_add(tmp, filename, dirlen - 1);
3006 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
b2476a60 3007 return -1;
d4b3d11a 3008 if (adjust_shared_perm(tmp->buf))
5723fe7e
LT
3009 return -1;
3010
3011 /* Try again */
d4b3d11a
JK
3012 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
3013 fd = git_mkstemp_mode(tmp->buf, 0444);
5723fe7e
LT
3014 }
3015 return fd;
3016}
3017
bbac7311 3018static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 3019 const void *buf, unsigned long len, time_t mtime)
0fcfd160 3020{
915308b1 3021 int fd, ret;
9892beba 3022 unsigned char compressed[4096];
ef49a7a0 3023 git_zstream stream;
748af44c
NP
3024 git_SHA_CTX c;
3025 unsigned char parano_sha1[20];
d4b3d11a 3026 static struct strbuf tmp_file = STRBUF_INIT;
30d6c6ea 3027 const char *filename = sha1_file_name(sha1);
a44c9a5e 3028
d4b3d11a 3029 fd = create_tmpfile(&tmp_file, filename);
aac17941 3030 if (fd < 0) {
35243577 3031 if (errno == EACCES)
82247e9b 3032 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
916d081b 3033 else
7616c6ca 3034 return error_errno("unable to create temporary file");
aac17941
LT
3035 }
3036
0fcfd160 3037 /* Set it up */
55bb5c91 3038 git_deflate_init(&stream, zlib_compression_level);
0fcfd160 3039 stream.next_out = compressed;
9892beba 3040 stream.avail_out = sizeof(compressed);
748af44c 3041 git_SHA1_Init(&c);
a44c9a5e
LT
3042
3043 /* First header.. */
d65a16f6 3044 stream.next_in = (unsigned char *)hdr;
a44c9a5e 3045 stream.avail_in = hdrlen;
55bb5c91
JH
3046 while (git_deflate(&stream, 0) == Z_OK)
3047 ; /* nothing */
748af44c 3048 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
3049
3050 /* Then the data itself.. */
c00e657d 3051 stream.next_in = (void *)buf;
a44c9a5e 3052 stream.avail_in = len;
9892beba 3053 do {
748af44c 3054 unsigned char *in0 = stream.next_in;
55bb5c91 3055 ret = git_deflate(&stream, Z_FINISH);
748af44c 3056 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
3057 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3058 die("unable to write sha1 file");
3059 stream.next_out = compressed;
3060 stream.avail_out = sizeof(compressed);
3061 } while (ret == Z_OK);
3062
ac54c277
LT
3063 if (ret != Z_STREAM_END)
3064 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
55bb5c91 3065 ret = git_deflate_end_gently(&stream);
ac54c277
LT
3066 if (ret != Z_OK)
3067 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
3068 git_SHA1_Final(parano_sha1, &c);
3069 if (hashcmp(sha1, parano_sha1) != 0)
3070 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 3071
e9039dd3 3072 close_sha1_file(fd);
0fcfd160 3073
bbac7311
NP
3074 if (mtime) {
3075 struct utimbuf utb;
3076 utb.actime = mtime;
3077 utb.modtime = mtime;
d4b3d11a 3078 if (utime(tmp_file.buf, &utb) < 0)
7616c6ca 3079 warning_errno("failed utime() on %s", tmp_file.buf);
bbac7311
NP
3080 }
3081
d4b3d11a 3082 return finalize_object_file(tmp_file.buf, filename);
0fcfd160 3083}
8237b185 3084
33d4221c
JK
3085static int freshen_loose_object(const unsigned char *sha1)
3086{
3087 return check_and_freshen(sha1, 1);
3088}
3089
3090static int freshen_packed_object(const unsigned char *sha1)
3091{
3092 struct pack_entry e;
ee1c6c34
JK
3093 if (!find_pack_entry(sha1, &e))
3094 return 0;
3095 if (e.p->freshened)
3096 return 1;
3097 if (!freshen_file(e.p->pack_name))
3098 return 0;
3099 e.p->freshened = 1;
3100 return 1;
33d4221c
JK
3101}
3102
1427a7ff 3103int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
bbac7311 3104{
bbac7311 3105 char hdr[32];
ef1286d3 3106 int hdrlen = sizeof(hdr);
bbac7311
NP
3107
3108 /* Normally if we have it in the pack then we do not bother writing
3109 * it out into .git/objects/??/?{38} file.
3110 */
3111 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
b5f52f37 3112 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
bbac7311
NP
3113 return 0;
3114 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3115}
3116
0c3db67c
ES
3117int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3118 unsigned char *sha1, unsigned flags)
3119{
3120 char *header;
3121 int hdrlen, status = 0;
3122
3123 /* type string, SP, %lu of the length plus NUL must fit this */
ef1286d3
JK
3124 hdrlen = strlen(type) + 32;
3125 header = xmalloc(hdrlen);
0c3db67c
ES
3126 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3127
3128 if (!(flags & HASH_WRITE_OBJECT))
3129 goto cleanup;
051086b9 3130 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
0c3db67c
ES
3131 goto cleanup;
3132 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3133
3134cleanup:
3135 free(header);
3136 return status;
3137}
3138
bbac7311
NP
3139int force_object_loose(const unsigned char *sha1, time_t mtime)
3140{
bbac7311
NP
3141 void *buf;
3142 unsigned long len;
3143 enum object_type type;
3144 char hdr[32];
3145 int hdrlen;
1fb23e65 3146 int ret;
bbac7311 3147
c529d75a 3148 if (has_loose_object(sha1))
bbac7311
NP
3149 return 0;
3150 buf = read_packed_sha1(sha1, &type, &len);
3151 if (!buf)
3152 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
ef1286d3 3153 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1fb23e65
BS
3154 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3155 free(buf);
3156
3157 return ret;
bbac7311
NP
3158}
3159
bf592c50
DB
3160int has_pack_index(const unsigned char *sha1)
3161{
3162 struct stat st;
3163 if (stat(sha1_pack_index_name(sha1), &st))
3164 return 0;
3165 return 1;
bf592c50
DB
3166}
3167
cd673c1f
JH
3168int has_sha1_pack(const unsigned char *sha1)
3169{
3170 struct pack_entry e;
3171 return find_pack_entry(sha1, &e);
3172}
3173
0eeb077b 3174int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
8237b185 3175{
1f688557
JH
3176 struct pack_entry e;
3177
cd673c1f 3178 if (find_pack_entry(sha1, &e))
1f688557 3179 return 1;
45e8a748
JK
3180 if (has_loose_object(sha1))
3181 return 1;
0eeb077b
JK
3182 if (flags & HAS_SHA1_QUICK)
3183 return 0;
45e8a748
JK
3184 reprepare_packed_git();
3185 return find_pack_entry(sha1, &e);
b419aa25 3186}
3187
3188int has_object_file(const struct object_id *oid)
3189{
3190 return has_sha1_file(oid->hash);
8237b185 3191}
74400e71 3192
c879daa2
NTND
3193static void check_tree(const void *buf, size_t size)
3194{
3195 struct tree_desc desc;
3196 struct name_entry entry;
3197
3198 init_tree_desc(&desc, buf, size);
3199 while (tree_entry(&desc, &entry))
3200 /* do nothing
3201 * tree_entry() will die() on malformed entries */
3202 ;
3203}
3204
3205static void check_commit(const void *buf, size_t size)
3206{
3207 struct commit c;
3208 memset(&c, 0, sizeof(c));
3209 if (parse_commit_buffer(&c, buf, size))
3210 die("corrupt commit");
3211}
3212
3213static void check_tag(const void *buf, size_t size)
3214{
3215 struct tag t;
3216 memset(&t, 0, sizeof(t));
3217 if (parse_tag_buffer(&t, buf, size))
3218 die("corrupt tag");
3219}
3220
43df4f86 3221static int index_mem(unsigned char *sha1, void *buf, size_t size,
c4ce46fc
JH
3222 enum object_type type,
3223 const char *path, unsigned flags)
e7332f96 3224{
6c510bee 3225 int ret, re_allocated = 0;
c4ce46fc 3226 int write_object = flags & HASH_WRITE_OBJECT;
74400e71 3227
7672db20 3228 if (!type)
edaec3fb 3229 type = OBJ_BLOB;
6c510bee
LT
3230
3231 /*
3232 * Convert blobs to git internal format
3233 */
43df4f86 3234 if ((type == OBJ_BLOB) && path) {
f285a2d7 3235 struct strbuf nbuf = STRBUF_INIT;
21e5ad50 3236 if (convert_to_git(path, buf, size, &nbuf,
5e12e78e 3237 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
b315c5c0 3238 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
3239 re_allocated = 1;
3240 }
3241 }
c4ce46fc 3242 if (flags & HASH_FORMAT_CHECK) {
c879daa2
NTND
3243 if (type == OBJ_TREE)
3244 check_tree(buf, size);
3245 if (type == OBJ_COMMIT)
3246 check_commit(buf, size);
3247 if (type == OBJ_TAG)
3248 check_tag(buf, size);
3249 }
6c510bee 3250
7672db20 3251 if (write_object)
edaec3fb 3252 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 3253 else
edaec3fb 3254 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 3255 if (re_allocated)
6c510bee 3256 free(buf);
43df4f86
DP
3257 return ret;
3258}
3259
9035d75a
SP
3260static int index_stream_convert_blob(unsigned char *sha1, int fd,
3261 const char *path, unsigned flags)
3262{
3263 int ret;
3264 const int write_object = flags & HASH_WRITE_OBJECT;
3265 struct strbuf sbuf = STRBUF_INIT;
3266
3267 assert(path);
3268 assert(would_convert_to_git_filter_fd(path));
3269
3270 convert_to_git_filter_fd(path, fd, &sbuf,
3271 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3272
3273 if (write_object)
3274 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3275 sha1);
3276 else
3277 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3278 sha1);
3279 strbuf_release(&sbuf);
3280 return ret;
3281}
3282
7b41e1e1
JH
3283static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3284 const char *path, unsigned flags)
3285{
3286 struct strbuf sbuf = STRBUF_INIT;
3287 int ret;
3288
3289 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3290 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3291 else
3292 ret = -1;
3293 strbuf_release(&sbuf);
3294 return ret;
3295}
3296
ea68b0ce
DP
3297#define SMALL_FILE_SIZE (32*1024)
3298
7b41e1e1
JH
3299static int index_core(unsigned char *sha1, int fd, size_t size,
3300 enum object_type type, const char *path,
3301 unsigned flags)
43df4f86
DP
3302{
3303 int ret;
43df4f86 3304
7b41e1e1 3305 if (!size) {
f6a1e1e2 3306 ret = index_mem(sha1, "", size, type, path, flags);
ea68b0ce
DP
3307 } else if (size <= SMALL_FILE_SIZE) {
3308 char *buf = xmalloc(size);
3309 if (size == read_in_full(fd, buf, size))
c4ce46fc 3310 ret = index_mem(sha1, buf, size, type, path, flags);
ea68b0ce 3311 else
7616c6ca 3312 ret = error_errno("short read");
ea68b0ce 3313 free(buf);
08bda208 3314 } else {
43df4f86 3315 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
c4ce46fc 3316 ret = index_mem(sha1, buf, size, type, path, flags);
aac17941 3317 munmap(buf, size);
08bda208 3318 }
7b41e1e1
JH
3319 return ret;
3320}
3321
4dd1fbc7 3322/*
568508e7
JH
3323 * This creates one packfile per large blob unless bulk-checkin
3324 * machinery is "plugged".
4dd1fbc7
JH
3325 *
3326 * This also bypasses the usual "convert-to-git" dance, and that is on
3327 * purpose. We could write a streaming version of the converting
3328 * functions and insert that before feeding the data to fast-import
4f22b101
JK
3329 * (or equivalent in-core API described above). However, that is
3330 * somewhat complicated, as we do not know the size of the filter
3331 * result, which we need to know beforehand when writing a git object.
3332 * Since the primary motivation for trying to stream from the working
3333 * tree file and to avoid mmaping it in core is to deal with large
3334 * binary blobs, they generally do not want to get any conversion, and
3335 * callers should avoid this code path when filters are requested.
4dd1fbc7
JH
3336 */
3337static int index_stream(unsigned char *sha1, int fd, size_t size,
3338 enum object_type type, const char *path,
3339 unsigned flags)
3340{
568508e7 3341 return index_bulk_checkin(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
3342}
3343
7b41e1e1
JH
3344int index_fd(unsigned char *sha1, int fd, struct stat *st,
3345 enum object_type type, const char *path, unsigned flags)
3346{
3347 int ret;
7b41e1e1 3348
9079ab7c
SP
3349 /*
3350 * Call xsize_t() only when needed to avoid potentially unnecessary
3351 * die() for large files.
3352 */
9035d75a
SP
3353 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3354 ret = index_stream_convert_blob(sha1, fd, path, flags);
3355 else if (!S_ISREG(st->st_mode))
7b41e1e1 3356 ret = index_pipe(sha1, fd, type, path, flags);
9079ab7c 3357 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
7ce7c760 3358 (path && would_convert_to_git(path)))
9079ab7c
SP
3359 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3360 flags);
4dd1fbc7 3361 else
9079ab7c
SP
3362 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3363 flags);
43df4f86 3364 close(fd);
aac17941 3365 return ret;
74400e71 3366}
ec1fcc16 3367
c4ce46fc 3368int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
ec1fcc16
JH
3369{
3370 int fd;
b760d3aa 3371 struct strbuf sb = STRBUF_INIT;
ec1fcc16
JH
3372
3373 switch (st->st_mode & S_IFMT) {
3374 case S_IFREG:
3375 fd = open(path, O_RDONLY);
3376 if (fd < 0)
7616c6ca 3377 return error_errno("open(\"%s\")", path);
c4ce46fc 3378 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
ec1fcc16
JH
3379 return error("%s: failed to insert into database",
3380 path);
3381 break;
3382 case S_IFLNK:
7616c6ca
NTND
3383 if (strbuf_readlink(&sb, path, st->st_size))
3384 return error_errno("readlink(\"%s\")", path);
c4ce46fc 3385 if (!(flags & HASH_WRITE_OBJECT))
b760d3aa
LT
3386 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3387 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
ec1fcc16
JH
3388 return error("%s: failed to insert into database",
3389 path);
b760d3aa 3390 strbuf_release(&sb);
ec1fcc16 3391 break;
f35a6d3b
LT
3392 case S_IFDIR:
3393 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
3394 default:
3395 return error("%s: unsupported file type", path);
3396 }
3397 return 0;
3398}
a69e5429
JH
3399
3400int read_pack_header(int fd, struct pack_header *header)
3401{
c697ad14
HO
3402 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3403 /* "eof before pack header was fully read" */
3404 return PH_ERROR_EOF;
3405
a69e5429
JH
3406 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3407 /* "protocol error (pack signature mismatch detected)" */
3408 return PH_ERROR_PACK_SIGNATURE;
3409 if (!pack_version_ok(header->hdr_version))
3410 /* "protocol error (pack version unsupported)" */
3411 return PH_ERROR_PROTOCOL;
3412 return 0;
3413}
40d52ff7
JK
3414
3415void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3416{
3417 enum object_type type = sha1_object_info(sha1, NULL);
3418 if (type < 0)
3419 die("%s is not a valid object", sha1_to_hex(sha1));
3420 if (type != expect)
3421 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3422 typename(expect));
3423}
27e1e22d
JK
3424
3425static int for_each_file_in_obj_subdir(int subdir_nr,
3426 struct strbuf *path,
3427 each_loose_object_fn obj_cb,
3428 each_loose_cruft_fn cruft_cb,
3429 each_loose_subdir_fn subdir_cb,
3430 void *data)
3431{
3432 size_t baselen = path->len;
3433 DIR *dir = opendir(path->buf);
3434 struct dirent *de;
3435 int r = 0;
3436
3437 if (!dir) {
3438 if (errno == ENOENT)
3439 return 0;
7616c6ca 3440 return error_errno("unable to open %s", path->buf);
27e1e22d
JK
3441 }
3442
3443 while ((de = readdir(dir))) {
3444 if (is_dot_or_dotdot(de->d_name))
3445 continue;
3446
3447 strbuf_setlen(path, baselen);
3448 strbuf_addf(path, "/%s", de->d_name);
3449
3450 if (strlen(de->d_name) == 38) {
3451 char hex[41];
3452 unsigned char sha1[20];
3453
3454 snprintf(hex, sizeof(hex), "%02x%s",
3455 subdir_nr, de->d_name);
3456 if (!get_sha1_hex(hex, sha1)) {
3457 if (obj_cb) {
3458 r = obj_cb(sha1, path->buf, data);
3459 if (r)
3460 break;
3461 }
3462 continue;
3463 }
3464 }
3465
3466 if (cruft_cb) {
3467 r = cruft_cb(de->d_name, path->buf, data);
3468 if (r)
3469 break;
3470 }
3471 }
094c7e63 3472 closedir(dir);
27e1e22d 3473
094c7e63 3474 strbuf_setlen(path, baselen);
27e1e22d
JK
3475 if (!r && subdir_cb)
3476 r = subdir_cb(subdir_nr, path->buf, data);
3477
27e1e22d
JK
3478 return r;
3479}
3480
e6f875e0 3481int for_each_loose_file_in_objdir_buf(struct strbuf *path,
27e1e22d
JK
3482 each_loose_object_fn obj_cb,
3483 each_loose_cruft_fn cruft_cb,
3484 each_loose_subdir_fn subdir_cb,
3485 void *data)
3486{
e6f875e0 3487 size_t baselen = path->len;
27e1e22d
JK
3488 int r = 0;
3489 int i;
3490
27e1e22d 3491 for (i = 0; i < 256; i++) {
e6f875e0
JK
3492 strbuf_addf(path, "/%02x", i);
3493 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
27e1e22d 3494 subdir_cb, data);
e6f875e0 3495 strbuf_setlen(path, baselen);
27e1e22d
JK
3496 if (r)
3497 break;
3498 }
3499
e6f875e0
JK
3500 return r;
3501}
3502
3503int for_each_loose_file_in_objdir(const char *path,
3504 each_loose_object_fn obj_cb,
3505 each_loose_cruft_fn cruft_cb,
3506 each_loose_subdir_fn subdir_cb,
3507 void *data)
3508{
3509 struct strbuf buf = STRBUF_INIT;
3510 int r;
3511
3512 strbuf_addstr(&buf, path);
3513 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3514 subdir_cb, data);
27e1e22d 3515 strbuf_release(&buf);
e6f875e0 3516
27e1e22d
JK
3517 return r;
3518}
660c889e
JK
3519