Prepare for 2.9.4
[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"
0fcfd160 26
144bde78
LT
27#ifndef O_NOATIME
28#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
29#define O_NOATIME 01000000
30#else
31#define O_NOATIME 0
32#endif
33#endif
34
28bd70d8
JN
35#define SZ_FMT PRIuMAX
36static inline uintmax_t sz_fmt(size_t s) { return s; }
e05db0fd 37
96f1e58f 38const unsigned char null_sha1[20];
3e56e724 39const struct object_id null_oid;
88cd621d 40
c597ba80
NTND
41/*
42 * This is meant to hold a *small* number of objects that you would
43 * want read_sha1_file() to be able to return, but yet you do not want
44 * to write them into the object store (e.g. a browse-only
45 * application).
46 */
47static struct cached_object {
48 unsigned char sha1[20];
49 enum object_type type;
50 void *buf;
51 unsigned long size;
52} *cached_objects;
53static int cached_object_nr, cached_object_alloc;
54
55static struct cached_object empty_tree = {
dab0d410 56 EMPTY_TREE_SHA1_BIN_LITERAL,
c597ba80
NTND
57 OBJ_TREE,
58 "",
59 0
60};
61
1b1005d1
MH
62/*
63 * A pointer to the last packed_git in which an object was found.
64 * When an object is sought, we look in this packfile first, because
65 * objects that are looked up at similar times are often in the same
66 * packfile as one another.
67 */
c01f51cc
NTND
68static struct packed_git *last_found_pack;
69
c597ba80
NTND
70static struct cached_object *find_cached_object(const unsigned char *sha1)
71{
72 int i;
73 struct cached_object *co = cached_objects;
74
75 for (i = 0; i < cached_object_nr; i++, co++) {
76 if (!hashcmp(co->sha1, sha1))
77 return co;
78 }
79 if (!hashcmp(sha1, empty_tree.sha1))
80 return &empty_tree;
81 return NULL;
82}
83
90a6464b
JH
84int mkdir_in_gitdir(const char *path)
85{
86 if (mkdir(path, 0777)) {
87 int saved_errno = errno;
88 struct stat st;
89 struct strbuf sb = STRBUF_INIT;
90
91 if (errno != EEXIST)
92 return -1;
93 /*
94 * Are we looking at a path in a symlinked worktree
95 * whose original repository does not yet have it?
96 * e.g. .git/rr-cache pointing at its original
97 * repository in which the user hasn't performed any
98 * conflict resolution yet?
99 */
100 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
101 strbuf_readlink(&sb, path, st.st_size) ||
102 !is_absolute_path(sb.buf) ||
103 mkdir(sb.buf, 0777)) {
104 strbuf_release(&sb);
105 errno = saved_errno;
106 return -1;
107 }
108 strbuf_release(&sb);
109 }
110 return adjust_shared_perm(path);
111}
112
0be0521b 113enum scld_error safe_create_leading_directories(char *path)
b2cb9425 114{
26c8ae2a 115 char *next_component = path + offset_1st_component(path);
0be0521b 116 enum scld_error ret = SCLD_OK;
67d42212 117
0be0521b 118 while (ret == SCLD_OK && next_component) {
f0502332 119 struct stat st;
0f527403 120 char *slash = next_component, slash_character;
f0502332 121
0f527403
MH
122 while (*slash && !is_dir_sep(*slash))
123 slash++;
124
125 if (!*slash)
b2cb9425 126 break;
bf10cf70 127
26c8ae2a 128 next_component = slash + 1;
0f527403 129 while (is_dir_sep(*next_component))
bf10cf70 130 next_component++;
26c8ae2a 131 if (!*next_component)
5f0bdf50 132 break;
831651fd 133
0f527403 134 slash_character = *slash;
831651fd 135 *slash = '\0';
67d42212
JR
136 if (!stat(path, &st)) {
137 /* path exists */
9e6f885d 138 if (!S_ISDIR(st.st_mode))
0be0521b 139 ret = SCLD_EXISTS;
53a39721 140 } else if (mkdir(path, 0777)) {
928734d9 141 if (errno == EEXIST &&
9e6f885d 142 !stat(path, &st) && S_ISDIR(st.st_mode))
928734d9 143 ; /* somebody created it since we checked */
18d37e86
MH
144 else if (errno == ENOENT)
145 /*
146 * Either mkdir() failed because
147 * somebody just pruned the containing
148 * directory, or stat() failed because
149 * the file that was in our way was
150 * just removed. Either way, inform
151 * the caller that it might be worth
152 * trying again:
153 */
154 ret = SCLD_VANISHED;
9e6f885d 155 else
0be0521b 156 ret = SCLD_FAILED;
53a39721 157 } else if (adjust_shared_perm(path)) {
0be0521b 158 ret = SCLD_PERMS;
457f06d6 159 }
0f527403 160 *slash = slash_character;
b2cb9425 161 }
9e6f885d 162 return ret;
b2cb9425 163}
723c31fe 164
0be0521b 165enum scld_error safe_create_leading_directories_const(const char *path)
8e21d63b
JK
166{
167 /* path points to cache entries, so xstrdup before messing with it */
168 char *buf = xstrdup(path);
0be0521b 169 enum scld_error result = safe_create_leading_directories(buf);
8e21d63b
JK
170 free(buf);
171 return result;
172}
173
ace1534d
JH
174static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
175{
176 int i;
177 for (i = 0; i < 20; i++) {
178 static char hex[] = "0123456789abcdef";
179 unsigned int val = sha1[i];
180 char *pos = pathbuf + i*2 + (i > 0);
181 *pos++ = hex[val >> 4];
182 *pos = hex[val & 0xf];
183 }
184}
185
30d6c6ea 186const char *sha1_file_name(const unsigned char *sha1)
0fcfd160 187{
560fb6a1
JK
188 static char buf[PATH_MAX];
189 const char *objdir;
190 int len;
0fcfd160 191
560fb6a1
JK
192 objdir = get_object_directory();
193 len = strlen(objdir);
194
195 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
196 if (len + 43 > PATH_MAX)
197 die("insanely long object directory %s", objdir);
198 memcpy(buf, objdir, len);
199 buf[len] = '/';
200 buf[len+3] = '/';
201 buf[len+42] = '\0';
202 fill_sha1_path(buf + len + 1, sha1);
203 return buf;
0fcfd160
LT
204}
205
d40d535b
MH
206/*
207 * Return the name of the pack or index file with the specified sha1
208 * in its filename. *base and *name are scratch space that must be
209 * provided by the caller. which should be "pack" or "idx".
210 */
633f43e1 211static char *sha1_get_pack_name(const unsigned char *sha1,
ac5190cc
JK
212 struct strbuf *buf,
213 const char *which)
bf592c50 214{
ac5190cc
JK
215 strbuf_reset(buf);
216 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
217 sha1_to_hex(sha1), which);
218 return buf->buf;
bf592c50
DB
219}
220
633f43e1 221char *sha1_pack_name(const unsigned char *sha1)
bf592c50 222{
ac5190cc
JK
223 static struct strbuf buf = STRBUF_INIT;
224 return sha1_get_pack_name(sha1, &buf, "pack");
633f43e1 225}
bf592c50 226
633f43e1
HO
227char *sha1_pack_index_name(const unsigned char *sha1)
228{
ac5190cc
JK
229 static struct strbuf buf = STRBUF_INIT;
230 return sha1_get_pack_name(sha1, &buf, "idx");
bf592c50
DB
231}
232
d5a63b99
JH
233struct alternate_object_database *alt_odb_list;
234static struct alternate_object_database **alt_odb_tail;
ace1534d 235
ddd5d056
JH
236/*
237 * Prepare alternate object database registry.
d5a63b99
JH
238 *
239 * The variable alt_odb_list points at the list of struct
240 * alternate_object_database. The elements on this list come from
241 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
242 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
243 * whose contents is similar to that environment variable but can be
244 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
245 * contains "/the/directory/corresponding/to/.git/objects/...", while
246 * its name points just after the slash at the end of ".git/objects/"
247 * in the example above, and has enough space to hold 40-byte hex
248 * SHA1, an extra slash for the first level indirection, and the
249 * terminating NUL.
ddd5d056 250 */
539e7506
EK
251static int link_alt_odb_entry(const char *entry, const char *relative_base,
252 int depth, const char *normalized_objdir)
ace1534d 253{
c2f493a4
MW
254 struct alternate_object_database *ent;
255 struct alternate_object_database *alt;
50a6c8ef 256 size_t pfxlen, entlen;
5bdf0a84 257 struct strbuf pathbuf = STRBUF_INIT;
d5a63b99 258
85dadc38 259 if (!is_absolute_path(entry) && relative_base) {
5bdf0a84
HW
260 strbuf_addstr(&pathbuf, real_path(relative_base));
261 strbuf_addch(&pathbuf, '/');
c2f493a4 262 }
6eac50d8 263 strbuf_addstr(&pathbuf, entry);
c2f493a4 264
5bdf0a84
HW
265 normalize_path_copy(pathbuf.buf, pathbuf.buf);
266
267 pfxlen = strlen(pathbuf.buf);
268
269 /*
270 * The trailing slash after the directory name is given by
271 * this function at the end. Remove duplicates.
272 */
273 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
274 pfxlen -= 1;
275
50a6c8ef
JK
276 entlen = st_add(pfxlen, 43); /* '/' + 2 hex + '/' + 38 hex + NUL */
277 ent = xmalloc(st_add(sizeof(*ent), entlen));
5bdf0a84
HW
278 memcpy(ent->base, pathbuf.buf, pfxlen);
279 strbuf_release(&pathbuf);
c2f493a4
MW
280
281 ent->name = ent->base + pfxlen + 1;
282 ent->base[pfxlen + 3] = '/';
283 ent->base[pfxlen] = ent->base[entlen-1] = 0;
284
285 /* Detect cases where alternate disappeared */
90b4a71c 286 if (!is_directory(ent->base)) {
c2f493a4
MW
287 error("object directory %s does not exist; "
288 "check .git/objects/info/alternates.",
289 ent->base);
290 free(ent);
291 return -1;
292 }
293
294 /* Prevent the common mistake of listing the same
295 * thing twice, or object directory itself.
296 */
297 for (alt = alt_odb_list; alt; alt = alt->next) {
80b47854
RS
298 if (pfxlen == alt->name - alt->base - 1 &&
299 !memcmp(ent->base, alt->base, pfxlen)) {
c2f493a4
MW
300 free(ent);
301 return -1;
302 }
303 }
ba0897e6 304 if (!fspathcmp(ent->base, normalized_objdir)) {
c2f493a4
MW
305 free(ent);
306 return -1;
307 }
308
309 /* add the alternate entry */
310 *alt_odb_tail = ent;
311 alt_odb_tail = &(ent->next);
312 ent->next = NULL;
313
314 /* recursively add alternates */
315 read_info_alternates(ent->base, depth + 1);
316
317 ent->base[pfxlen] = '/';
318
319 return 0;
320}
321
c5950164 322static void link_alt_odb_entries(const char *alt, int len, int sep,
c2f493a4
MW
323 const char *relative_base, int depth)
324{
6eac50d8
MH
325 struct string_list entries = STRING_LIST_INIT_NODUP;
326 char *alt_copy;
327 int i;
539e7506 328 struct strbuf objdirbuf = STRBUF_INIT;
c2f493a4
MW
329
330 if (depth > 5) {
331 error("%s: ignoring alternate object stores, nesting too deep.",
332 relative_base);
333 return;
334 }
335
f655651e 336 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
539e7506
EK
337 normalize_path_copy(objdirbuf.buf, objdirbuf.buf);
338
c5950164 339 alt_copy = xmemdupz(alt, len);
6eac50d8
MH
340 string_list_split_in_place(&entries, alt_copy, sep, -1);
341 for (i = 0; i < entries.nr; i++) {
342 const char *entry = entries.items[i].string;
343 if (entry[0] == '\0' || entry[0] == '#')
9577e7e3 344 continue;
6eac50d8
MH
345 if (!is_absolute_path(entry) && depth) {
346 error("%s: ignoring relative alternate object store %s",
347 relative_base, entry);
348 } else {
539e7506 349 link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
9577e7e3 350 }
9577e7e3 351 }
6eac50d8
MH
352 string_list_clear(&entries, 0);
353 free(alt_copy);
539e7506 354 strbuf_release(&objdirbuf);
d5a63b99
JH
355}
356
5e73633d 357void read_info_alternates(const char * relative_base, int depth)
d5a63b99 358{
9577e7e3 359 char *map;
dc49cd76 360 size_t mapsz;
d5a63b99 361 struct stat st;
5015f01c 362 char *path;
c2f493a4 363 int fd;
d5a63b99 364
5015f01c 365 path = xstrfmt("%s/info/alternates", relative_base);
f4e51683 366 fd = git_open_noatime(path);
5015f01c 367 free(path);
d5a63b99
JH
368 if (fd < 0)
369 return;
370 if (fstat(fd, &st) || (st.st_size == 0)) {
371 close(fd);
9a217f2a 372 return;
ace1534d 373 }
dc49cd76
SP
374 mapsz = xsize_t(st.st_size);
375 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 376 close(fd);
d5a63b99 377
c5950164 378 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
c2f493a4 379
dc49cd76 380 munmap(map, mapsz);
ace1534d
JH
381}
382
bef70b22
DB
383void add_to_alternates_file(const char *reference)
384{
385 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
77b9b1d1
JK
386 char *alts = git_pathdup("objects/info/alternates");
387 FILE *in, *out;
388
389 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
390 out = fdopen_lock_file(lock, "w");
391 if (!out)
392 die_errno("unable to fdopen alternates lockfile");
393
394 in = fopen(alts, "r");
395 if (in) {
396 struct strbuf line = STRBUF_INIT;
397 int found = 0;
398
3f163962 399 while (strbuf_getline(&line, in) != EOF) {
77b9b1d1
JK
400 if (!strcmp(reference, line.buf)) {
401 found = 1;
402 break;
403 }
404 fprintf_or_die(out, "%s\n", line.buf);
405 }
406
407 strbuf_release(&line);
408 fclose(in);
409
410 if (found) {
411 rollback_lock_file(lock);
412 lock = NULL;
413 }
414 }
415 else if (errno != ENOENT)
416 die_errno("unable to read alternates file");
417
418 if (lock) {
419 fprintf_or_die(out, "%s\n", reference);
420 if (commit_lock_file(lock))
421 die_errno("unable to move new alternates file into place");
422 if (alt_odb_tail)
423 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
424 }
425 free(alts);
bef70b22
DB
426}
427
fe1b2268 428int foreach_alt_odb(alt_odb_fn fn, void *cb)
d79796bc
JH
429{
430 struct alternate_object_database *ent;
fe1b2268 431 int r = 0;
d79796bc
JH
432
433 prepare_alt_odb();
fe1b2268
JK
434 for (ent = alt_odb_list; ent; ent = ent->next) {
435 r = fn(ent, cb);
436 if (r)
437 break;
438 }
439 return r;
d79796bc
JH
440}
441
c2f493a4
MW
442void prepare_alt_odb(void)
443{
554fe20d 444 const char *alt;
c2f493a4 445
7dc24aa5
SP
446 if (alt_odb_tail)
447 return;
448
c2f493a4
MW
449 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
450 if (!alt) alt = "";
451
c2f493a4 452 alt_odb_tail = &alt_odb_list;
c5950164 453 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
454
455 read_info_alternates(get_object_directory(), 0);
456}
457
3096b2ec 458/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
33d4221c 459static int freshen_file(const char *fn)
ace1534d 460{
33d4221c
JK
461 struct utimbuf t;
462 t.actime = t.modtime = time(NULL);
463 return !utime(fn, &t);
0f4dc14a 464}
ace1534d 465
3096b2ec
JK
466/*
467 * All of the check_and_freshen functions return 1 if the file exists and was
468 * freshened (if freshening was requested), 0 otherwise. If they return
469 * 0, you should not assume that it is safe to skip a write of the object (it
470 * either does not exist on disk, or has a stale mtime and may be subject to
471 * pruning).
472 */
33d4221c
JK
473static int check_and_freshen_file(const char *fn, int freshen)
474{
475 if (access(fn, F_OK))
476 return 0;
3096b2ec 477 if (freshen && !freshen_file(fn))
33d4221c
JK
478 return 0;
479 return 1;
480}
481
482static int check_and_freshen_local(const unsigned char *sha1, int freshen)
483{
484 return check_and_freshen_file(sha1_file_name(sha1), freshen);
485}
486
487static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
0f4dc14a
BC
488{
489 struct alternate_object_database *alt;
9a217f2a 490 prepare_alt_odb();
d5a63b99 491 for (alt = alt_odb_list; alt; alt = alt->next) {
0f4dc14a 492 fill_sha1_path(alt->name, sha1);
33d4221c 493 if (check_and_freshen_file(alt->base, freshen))
c529d75a 494 return 1;
ace1534d 495 }
c529d75a 496 return 0;
ace1534d
JH
497}
498
33d4221c
JK
499static int check_and_freshen(const unsigned char *sha1, int freshen)
500{
501 return check_and_freshen_local(sha1, freshen) ||
502 check_and_freshen_nonlocal(sha1, freshen);
503}
504
505int has_loose_object_nonlocal(const unsigned char *sha1)
506{
507 return check_and_freshen_nonlocal(sha1, 0);
508}
509
0f4dc14a
BC
510static int has_loose_object(const unsigned char *sha1)
511{
33d4221c 512 return check_and_freshen(sha1, 0);
0f4dc14a
BC
513}
514
60bb8b14 515static unsigned int pack_used_ctr;
a53128b6
SP
516static unsigned int pack_mmap_calls;
517static unsigned int peak_pack_open_windows;
518static unsigned int pack_open_windows;
c7934306
SP
519static unsigned int pack_open_fds;
520static unsigned int pack_max_fds;
a53128b6 521static size_t peak_pack_mapped;
60bb8b14 522static size_t pack_mapped;
9a217f2a 523struct packed_git *packed_git;
1f688557 524
b79d18c9 525void pack_report(void)
a53128b6
SP
526{
527 fprintf(stderr,
e05db0fd
PR
528 "pack_report: getpagesize() = %10" SZ_FMT "\n"
529 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
530 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
531 sz_fmt(getpagesize()),
532 sz_fmt(packed_git_window_size),
533 sz_fmt(packed_git_limit));
a53128b6
SP
534 fprintf(stderr,
535 "pack_report: pack_used_ctr = %10u\n"
536 "pack_report: pack_mmap_calls = %10u\n"
537 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
538 "pack_report: pack_mapped = "
539 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
540 pack_used_ctr,
541 pack_mmap_calls,
542 pack_open_windows, peak_pack_open_windows,
9e42d6a1 543 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
544}
545
d40d535b
MH
546/*
547 * Open and mmap the index file at path, perform a couple of
548 * consistency checks, then record its information to p. Return 0 on
549 * success.
550 */
551static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
552{
553 void *idx_map;
42873078 554 struct pack_idx_header *hdr;
dc49cd76 555 size_t idx_size;
74e34e1f 556 uint32_t version, nr, i, *index;
f4e51683 557 int fd = git_open_noatime(path);
1f688557 558 struct stat st;
42873078 559
1f688557
JH
560 if (fd < 0)
561 return -1;
562 if (fstat(fd, &st)) {
563 close(fd);
564 return -1;
565 }
dc49cd76 566 idx_size = xsize_t(st.st_size);
2d88451b
SP
567 if (idx_size < 4 * 256 + 20 + 20) {
568 close(fd);
569 return error("index file %s is too small", path);
570 }
c4712e45 571 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 572 close(fd);
1f688557 573
42873078
NP
574 hdr = idx_map;
575 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
576 version = ntohl(hdr->idx_version);
577 if (version < 2 || version > 2) {
578 munmap(idx_map, idx_size);
6e1c2344 579 return error("index file %s is version %"PRIu32
74e34e1f
NP
580 " and is not supported by this binary"
581 " (try upgrading GIT to a newer version)",
582 path, version);
583 }
584 } else
585 version = 1;
df1b059d 586
1f688557 587 nr = 0;
42873078 588 index = idx_map;
74e34e1f
NP
589 if (version > 1)
590 index += 2; /* skip index header */
1f688557 591 for (i = 0; i < 256; i++) {
326bf396 592 uint32_t n = ntohl(index[i]);
2d88451b
SP
593 if (n < nr) {
594 munmap(idx_map, idx_size);
df1b059d 595 return error("non-monotonic index %s", path);
2d88451b 596 }
1f688557
JH
597 nr = n;
598 }
599
74e34e1f
NP
600 if (version == 1) {
601 /*
602 * Total size:
603 * - 256 index entries 4 bytes each
604 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
605 * - 20-byte SHA1 of the packfile
606 * - 20-byte SHA1 file checksum
607 */
608 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
609 munmap(idx_map, idx_size);
eef427a0 610 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
611 }
612 } else if (version == 2) {
613 /*
614 * Minimum size:
615 * - 8 bytes of header
616 * - 256 index entries 4 bytes each
617 * - 20-byte sha1 entry * nr
618 * - 4-byte crc entry * nr
619 * - 4-byte offset entry * nr
620 * - 20-byte SHA1 of the packfile
621 * - 20-byte SHA1 file checksum
622 * And after the 4-byte offset table might be a
623 * variable sized table containing 8-byte entries
624 * for offsets larger than 2^31.
625 */
626 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
627 unsigned long max_size = min_size;
628 if (nr)
629 max_size += (nr - 1)*8;
630 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 631 munmap(idx_map, idx_size);
eef427a0 632 return error("wrong index v2 file size in %s", path);
74e34e1f 633 }
7109c889
JH
634 if (idx_size != min_size &&
635 /*
636 * make sure we can deal with large pack offsets.
637 * 31-bit signed offset won't be enough, neither
638 * 32-bit unsigned one will be.
639 */
640 (sizeof(off_t) <= 4)) {
641 munmap(idx_map, idx_size);
642 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 643 }
2d88451b 644 }
1f688557 645
74e34e1f 646 p->index_version = version;
42873078
NP
647 p->index_data = idx_map;
648 p->index_size = idx_size;
57059091 649 p->num_objects = nr;
1f688557
JH
650 return 0;
651}
652
bc8e478a 653int open_pack_index(struct packed_git *p)
d079837e
SP
654{
655 char *idx_name;
9ae97018 656 size_t len;
d079837e
SP
657 int ret;
658
659 if (p->index_data)
660 return 0;
661
9ae97018
JK
662 if (!strip_suffix(p->pack_name, ".pack", &len))
663 die("BUG: pack_name does not end in .pack");
664 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
d079837e
SP
665 ret = check_packed_git_idx(idx_name, p);
666 free(idx_name);
667 return ret;
668}
669
11daf39b
SP
670static void scan_windows(struct packed_git *p,
671 struct packed_git **lru_p,
672 struct pack_window **lru_w,
673 struct pack_window **lru_l)
1f688557 674{
11daf39b
SP
675 struct pack_window *w, *w_l;
676
677 for (w_l = NULL, w = p->windows; w; w = w->next) {
678 if (!w->inuse_cnt) {
679 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
680 *lru_p = p;
681 *lru_w = w;
682 *lru_l = w_l;
54044bf8 683 }
54044bf8 684 }
11daf39b 685 w_l = w;
f9253394 686 }
11daf39b
SP
687}
688
7c3ecb32 689static int unuse_one_window(struct packed_git *current)
11daf39b
SP
690{
691 struct packed_git *p, *lru_p = NULL;
692 struct pack_window *lru_w = NULL, *lru_l = NULL;
693
694 if (current)
695 scan_windows(current, &lru_p, &lru_w, &lru_l);
696 for (p = packed_git; p; p = p->next)
697 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
698 if (lru_p) {
699 munmap(lru_w->base, lru_w->len);
700 pack_mapped -= lru_w->len;
701 if (lru_l)
702 lru_l->next = lru_w->next;
7c3ecb32 703 else
54044bf8 704 lru_p->windows = lru_w->next;
54044bf8 705 free(lru_w);
a53128b6 706 pack_open_windows--;
54044bf8
SP
707 return 1;
708 }
709 return 0;
f9253394
JH
710}
711
7c3ecb32 712void release_pack_memory(size_t need)
97bfeb34
SP
713{
714 size_t cur = pack_mapped;
7c3ecb32 715 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
97bfeb34
SP
716 ; /* nothing */
717}
718
02710228
SP
719static void mmap_limit_check(size_t length)
720{
721 static size_t limit = 0;
722 if (!limit) {
723 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
724 if (!limit)
725 limit = SIZE_MAX;
726 }
727 if (length > limit)
728 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
729 (uintmax_t)length, (uintmax_t)limit);
730}
731
1570856b
JK
732void *xmmap_gently(void *start, size_t length,
733 int prot, int flags, int fd, off_t offset)
58ecbd5e 734{
02710228
SP
735 void *ret;
736
737 mmap_limit_check(length);
738 ret = mmap(start, length, prot, flags, fd, offset);
58ecbd5e
JN
739 if (ret == MAP_FAILED) {
740 if (!length)
741 return NULL;
7c3ecb32 742 release_pack_memory(length);
58ecbd5e 743 ret = mmap(start, length, prot, flags, fd, offset);
58ecbd5e
JN
744 }
745 return ret;
746}
747
1570856b
JK
748void *xmmap(void *start, size_t length,
749 int prot, int flags, int fd, off_t offset)
750{
751 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
752 if (ret == MAP_FAILED)
9ca0aaf6 753 die_errno("mmap failed");
1570856b
JK
754 return ret;
755}
756
c9ced051
SP
757void close_pack_windows(struct packed_git *p)
758{
759 while (p->windows) {
760 struct pack_window *w = p->windows;
761
762 if (w->inuse_cnt)
763 die("pack '%s' still has open windows to it",
764 p->pack_name);
765 munmap(w->base, w->len);
766 pack_mapped -= w->len;
767 pack_open_windows--;
768 p->windows = w->next;
769 free(w);
770 }
771}
772
71fe5d7f
JS
773static int close_pack_fd(struct packed_git *p)
774{
775 if (p->pack_fd < 0)
776 return 0;
777
778 close(p->pack_fd);
779 pack_open_fds--;
780 p->pack_fd = -1;
781
782 return 1;
783}
784
38849a81
JS
785static void close_pack(struct packed_git *p)
786{
787 close_pack_windows(p);
788 close_pack_fd(p);
789 close_pack_index(p);
790}
791
792void close_all_packs(void)
793{
794 struct packed_git *p;
795
796 for (p = packed_git; p; p = p->next)
797 if (p->do_not_close)
798 die("BUG! Want to close pack marked 'do-not-close'");
799 else
800 close_pack(p);
801}
802
803
88d0db55
BC
804/*
805 * The LRU pack is the one with the oldest MRU window, preferring packs
806 * with no used windows, or the oldest mtime if it has no windows allocated.
807 */
808static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
809{
810 struct pack_window *w, *this_mru_w;
811 int has_windows_inuse = 0;
812
813 /*
814 * Reject this pack if it has windows and the previously selected
815 * one does not. If this pack does not have windows, reject
816 * it if the pack file is newer than the previously selected one.
817 */
818 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
819 return;
820
821 for (w = this_mru_w = p->windows; w; w = w->next) {
822 /*
823 * Reject this pack if any of its windows are in use,
824 * but the previously selected pack did not have any
825 * inuse windows. Otherwise, record that this pack
826 * has windows in use.
827 */
828 if (w->inuse_cnt) {
829 if (*accept_windows_inuse)
830 has_windows_inuse = 1;
831 else
832 return;
833 }
834
835 if (w->last_used > this_mru_w->last_used)
836 this_mru_w = w;
837
838 /*
839 * Reject this pack if it has windows that have been
840 * used more recently than the previously selected pack.
841 * If the previously selected pack had windows inuse and
842 * we have not encountered a window in this pack that is
843 * inuse, skip this check since we prefer a pack with no
844 * inuse windows to one that has inuse windows.
845 */
846 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
847 this_mru_w->last_used > (*mru_w)->last_used)
848 return;
849 }
850
851 /*
852 * Select this pack.
853 */
854 *mru_w = this_mru_w;
855 *lru_p = p;
856 *accept_windows_inuse = has_windows_inuse;
857}
858
859static int close_one_pack(void)
860{
861 struct packed_git *p, *lru_p = NULL;
862 struct pack_window *mru_w = NULL;
863 int accept_windows_inuse = 1;
864
865 for (p = packed_git; p; p = p->next) {
866 if (p->pack_fd == -1)
867 continue;
868 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
869 }
870
71fe5d7f
JS
871 if (lru_p)
872 return close_pack_fd(lru_p);
88d0db55
BC
873
874 return 0;
875}
876
03e79c88 877void unuse_pack(struct pack_window **w_cursor)
f9253394 878{
03e79c88
SP
879 struct pack_window *w = *w_cursor;
880 if (w) {
881 w->inuse_cnt--;
882 *w_cursor = NULL;
883 }
1f688557
JH
884}
885
fa5fc15d
SP
886void close_pack_index(struct packed_git *p)
887{
888 if (p->index_data) {
889 munmap((void *)p->index_data, p->index_size);
890 p->index_data = NULL;
891 }
892}
893
c74faea1
NP
894/*
895 * This is used by git-repack in case a newly created pack happens to
896 * contain the same set of objects as an existing one. In that case
897 * the resulting file might be different even if its name would be the
898 * same. It is best to close any reference to the old pack before it is
01689909 899 * replaced on disk. Of course no index pointers or windows for given pack
c74faea1
NP
900 * must subsist at this point. If ever objects from this pack are requested
901 * again, the new version of the pack will be reinitialized through
902 * reprepare_packed_git().
903 */
904void free_pack_by_name(const char *pack_name)
905{
906 struct packed_git *p, **pp = &packed_git;
907
908 while (*pp) {
909 p = *pp;
910 if (strcmp(pack_name, p->pack_name) == 0) {
fa3a0c94 911 clear_delta_base_cache();
38849a81 912 close_pack(p);
c74faea1
NP
913 free(p->bad_object_sha1);
914 *pp = p->next;
c01f51cc
NTND
915 if (last_found_pack == p)
916 last_found_pack = NULL;
c74faea1
NP
917 free(p);
918 return;
919 }
920 pp = &p->next;
921 }
922}
923
a0788266
JS
924static unsigned int get_max_fd_limit(void)
925{
926#ifdef RLIMIT_NOFILE
491a8dec
JH
927 {
928 struct rlimit lim;
a0788266 929
491a8dec
JH
930 if (!getrlimit(RLIMIT_NOFILE, &lim))
931 return lim.rlim_cur;
932 }
933#endif
934
935#ifdef _SC_OPEN_MAX
936 {
937 long open_max = sysconf(_SC_OPEN_MAX);
938 if (0 < open_max)
939 return open_max;
940 /*
941 * Otherwise, we got -1 for one of the two
942 * reasons:
943 *
944 * (1) sysconf() did not understand _SC_OPEN_MAX
945 * and signaled an error with -1; or
946 * (2) sysconf() said there is no limit.
947 *
948 * We _could_ clear errno before calling sysconf() to
949 * tell these two cases apart and return a huge number
950 * in the latter case to let the caller cap it to a
951 * value that is not so selfish, but letting the
952 * fallback OPEN_MAX codepath take care of these cases
953 * is a lot simpler.
954 */
955 }
956#endif
a0788266 957
491a8dec 958#ifdef OPEN_MAX
a0788266
JS
959 return OPEN_MAX;
960#else
961 return 1; /* see the caller ;-) */
962#endif
963}
964
3cf8b462
SP
965/*
966 * Do not call this directly as this leaks p->pack_fd on error return;
967 * call open_packed_git() instead.
968 */
969static int open_packed_git_1(struct packed_git *p)
1f688557 970{
9bc879c1
SP
971 struct stat st;
972 struct pack_header hdr;
973 unsigned char sha1[20];
974 unsigned char *idx_sha1;
2c039da8 975 long fd_flag;
9bc879c1 976
d079837e
SP
977 if (!p->index_data && open_pack_index(p))
978 return error("packfile %s index unavailable", p->pack_name);
979
c7934306 980 if (!pack_max_fds) {
a0788266 981 unsigned int max_fds = get_max_fd_limit();
c7934306
SP
982
983 /* Save 3 for stdin/stdout/stderr, 22 for work */
984 if (25 < max_fds)
985 pack_max_fds = max_fds - 25;
986 else
987 pack_max_fds = 1;
988 }
989
88d0db55 990 while (pack_max_fds <= pack_open_fds && close_one_pack())
c7934306
SP
991 ; /* nothing */
992
f4e51683 993 p->pack_fd = git_open_noatime(p->pack_name);
9bc879c1 994 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 995 return -1;
c7934306 996 pack_open_fds++;
9bc879c1
SP
997
998 /* If we created the struct before we had the pack we lack size. */
bf592c50 999 if (!p->pack_size) {
bf592c50 1000 if (!S_ISREG(st.st_mode))
072db278 1001 return error("packfile %s not a regular file", p->pack_name);
bf592c50 1002 p->pack_size = st.st_size;
9bc879c1 1003 } else if (p->pack_size != st.st_size)
072db278 1004 return error("packfile %s size changed", p->pack_name);
9bc879c1 1005
2c039da8
JH
1006 /* We leave these file descriptors open with sliding mmap;
1007 * there is no point keeping them open across exec(), though.
1008 */
1009 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
1010 if (fd_flag < 0)
072db278 1011 return error("cannot determine file descriptor flags");
2c039da8
JH
1012 fd_flag |= FD_CLOEXEC;
1013 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 1014 return error("cannot set FD_CLOEXEC");
2c039da8 1015
9bc879c1 1016 /* Verify we recognize this pack file format. */
e6e2bd62 1017 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 1018 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 1019 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 1020 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 1021 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
1022 return error("packfile %s is version %"PRIu32" and not"
1023 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
1024 p->pack_name, ntohl(hdr.hdr_version));
1025
1026 /* Verify the pack matches its index. */
57059091 1027 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
1028 return error("packfile %s claims to have %"PRIu32" objects"
1029 " while index indicates %"PRIu32" objects",
57059091
NP
1030 p->pack_name, ntohl(hdr.hdr_entries),
1031 p->num_objects);
9bc879c1 1032 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 1033 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 1034 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 1035 return error("packfile %s signature is unavailable", p->pack_name);
42873078 1036 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 1037 if (hashcmp(sha1, idx_sha1))
072db278
SP
1038 return error("packfile %s does not match index", p->pack_name);
1039 return 0;
9bc879c1
SP
1040}
1041
3cf8b462
SP
1042static int open_packed_git(struct packed_git *p)
1043{
1044 if (!open_packed_git_1(p))
1045 return 0;
71fe5d7f 1046 close_pack_fd(p);
3cf8b462
SP
1047 return -1;
1048}
1049
c4001d92 1050static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
1051{
1052 /* We must promise at least 20 bytes (one hash) after the
1053 * offset is available from this window, otherwise the offset
1054 * is not actually in this window and a different window (which
1055 * has that one hash excess) must be used. This is to support
1056 * the object header and delta base parsing routines below.
1057 */
1058 off_t win_off = win->offset;
1059 return win_off <= offset
1060 && (offset + 20) <= (win_off + win->len);
1061}
1062
4b25d091 1063unsigned char *use_pack(struct packed_git *p,
03e79c88 1064 struct pack_window **w_cursor,
c4001d92 1065 off_t offset,
ef49a7a0 1066 unsigned long *left)
9bc879c1 1067{
60bb8b14 1068 struct pack_window *win = *w_cursor;
03e79c88 1069
a9d98a14 1070 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
1071 * pointless to ask for an offset into the middle of that
1072 * hash, and the in_window function above wouldn't match
1073 * don't allow an offset too close to the end of the file.
1074 */
d131b7af
SP
1075 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1076 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
1077 if (offset > (p->pack_size - 20))
1078 die("offset beyond end of packfile (truncated pack?)");
13e0b0d3 1079 if (offset < 0)
7465feba 1080 die(_("offset before end of packfile (broken .idx?)"));
60bb8b14
SP
1081
1082 if (!win || !in_window(win, offset)) {
1083 if (win)
1084 win->inuse_cnt--;
1085 for (win = p->windows; win; win = win->next) {
1086 if (in_window(win, offset))
1087 break;
1088 }
1089 if (!win) {
78a28df9 1090 size_t window_align = packed_git_window_size / 2;
dc49cd76 1091 off_t len;
d131b7af
SP
1092
1093 if (p->pack_fd == -1 && open_packed_git(p))
1094 die("packfile %s cannot be accessed", p->pack_name);
1095
60bb8b14 1096 win = xcalloc(1, sizeof(*win));
78a28df9 1097 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
1098 len = p->pack_size - win->offset;
1099 if (len > packed_git_window_size)
1100 len = packed_git_window_size;
1101 win->len = (size_t)len;
60bb8b14 1102 pack_mapped += win->len;
11daf39b 1103 while (packed_git_limit < pack_mapped
7c3ecb32 1104 && unuse_one_window(p))
60bb8b14 1105 ; /* nothing */
c4712e45 1106 win->base = xmmap(NULL, win->len,
60bb8b14
SP
1107 PROT_READ, MAP_PRIVATE,
1108 p->pack_fd, win->offset);
1109 if (win->base == MAP_FAILED)
7616c6ca
NTND
1110 die_errno("packfile %s cannot be mapped",
1111 p->pack_name);
d131b7af 1112 if (!win->offset && win->len == p->pack_size
71fe5d7f
JS
1113 && !p->do_not_close)
1114 close_pack_fd(p);
a53128b6
SP
1115 pack_mmap_calls++;
1116 pack_open_windows++;
1117 if (pack_mapped > peak_pack_mapped)
1118 peak_pack_mapped = pack_mapped;
1119 if (pack_open_windows > peak_pack_open_windows)
1120 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
1121 win->next = p->windows;
1122 p->windows = win;
1123 }
1f688557 1124 }
03e79c88
SP
1125 if (win != *w_cursor) {
1126 win->last_used = pack_used_ctr++;
1127 win->inuse_cnt++;
1128 *w_cursor = win;
1129 }
60bb8b14 1130 offset -= win->offset;
03e79c88 1131 if (left)
dc49cd76 1132 *left = win->len - xsize_t(offset);
03e79c88 1133 return win->base + offset;
1f688557
JH
1134}
1135
27d69a46
NP
1136static struct packed_git *alloc_packed_git(int extra)
1137{
50a6c8ef 1138 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
27d69a46
NP
1139 memset(p, 0, sizeof(*p));
1140 p->pack_fd = -1;
1141 return p;
1142}
1143
e0500293
JN
1144static void try_to_free_pack_memory(size_t size)
1145{
7c3ecb32 1146 release_pack_memory(size);
e0500293
JN
1147}
1148
48bcc1c3 1149struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
1f688557 1150{
e0500293 1151 static int have_set_try_to_free_routine;
1f688557 1152 struct stat st;
48bcc1c3
JK
1153 size_t alloc;
1154 struct packed_git *p;
1f688557 1155
e0500293
JN
1156 if (!have_set_try_to_free_routine) {
1157 have_set_try_to_free_routine = 1;
1158 set_try_to_free_routine(try_to_free_pack_memory);
1159 }
1160
42873078
NP
1161 /*
1162 * Make sure a corresponding .pack file exists and that
1163 * the index looks sane.
1164 */
48bcc1c3 1165 if (!strip_suffix_mem(path, &path_len, ".idx"))
1f688557 1166 return NULL;
48bcc1c3
JK
1167
1168 /*
1169 * ".pack" is long enough to hold any suffix we're adding (and
1170 * the use xsnprintf double-checks that)
1171 */
50a6c8ef 1172 alloc = st_add3(path_len, strlen(".pack"), 1);
48bcc1c3 1173 p = alloc_packed_git(alloc);
42873078 1174 memcpy(p->pack_name, path, path_len);
8d25931d 1175
48bcc1c3 1176 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
8d25931d
BC
1177 if (!access(p->pack_name, F_OK))
1178 p->pack_keep = 1;
1179
48bcc1c3 1180 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
d079837e 1181 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 1182 free(p);
1f688557
JH
1183 return NULL;
1184 }
42873078 1185
1f688557
JH
1186 /* ok, it looks sane as far as we can check without
1187 * actually mapping the pack file.
1188 */
1f688557 1189 p->pack_size = st.st_size;
9d835df2 1190 p->pack_local = local;
b867092f 1191 p->mtime = st.st_mtime;
42873078
NP
1192 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1193 hashclr(p->sha1);
1f688557
JH
1194 return p;
1195}
1196
7b64469a 1197struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 1198{
42873078 1199 const char *path = sha1_pack_name(sha1);
50a6c8ef 1200 size_t alloc = st_add(strlen(path), 1);
c7ab0ba3 1201 struct packed_git *p = alloc_packed_git(alloc);
bf592c50 1202
c7ab0ba3 1203 memcpy(p->pack_name, path, alloc); /* includes NUL */
27d69a46 1204 hashcpy(p->sha1, sha1);
42873078
NP
1205 if (check_packed_git_idx(idx_path, p)) {
1206 free(p);
bf592c50 1207 return NULL;
42873078 1208 }
bf592c50 1209
bf592c50
DB
1210 return p;
1211}
1212
1213void install_packed_git(struct packed_git *pack)
1214{
c7934306
SP
1215 if (pack->pack_fd != -1)
1216 pack_open_fds++;
1217
bf592c50
DB
1218 pack->next = packed_git;
1219 packed_git = pack;
1220}
1221
0a489b06 1222void (*report_garbage)(unsigned seen_bits, const char *path);
543c5caa
NTND
1223
1224static void report_helper(const struct string_list *list,
1225 int seen_bits, int first, int last)
1226{
0a489b06 1227 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
543c5caa 1228 return;
0a489b06 1229
543c5caa 1230 for (; first < last; first++)
0a489b06 1231 report_garbage(seen_bits, list->items[first].string);
543c5caa
NTND
1232}
1233
1234static void report_pack_garbage(struct string_list *list)
1235{
1236 int i, baselen = -1, first = 0, seen_bits = 0;
1237
1238 if (!report_garbage)
1239 return;
1240
3383e199 1241 string_list_sort(list);
543c5caa
NTND
1242
1243 for (i = 0; i < list->nr; i++) {
1244 const char *path = list->items[i].string;
1245 if (baselen != -1 &&
1246 strncmp(path, list->items[first].string, baselen)) {
1247 report_helper(list, seen_bits, first, i);
1248 baselen = -1;
1249 seen_bits = 0;
1250 }
1251 if (baselen == -1) {
1252 const char *dot = strrchr(path, '.');
1253 if (!dot) {
0a489b06 1254 report_garbage(PACKDIR_FILE_GARBAGE, path);
543c5caa
NTND
1255 continue;
1256 }
1257 baselen = dot - path + 1;
1258 first = i;
1259 }
1260 if (!strcmp(path + baselen, "pack"))
1261 seen_bits |= 1;
1262 else if (!strcmp(path + baselen, "idx"))
1263 seen_bits |= 2;
1264 }
1265 report_helper(list, seen_bits, first, list->nr);
1266}
1267
9d835df2 1268static void prepare_packed_git_one(char *objdir, int local)
1f688557 1269{
880fb8de
RS
1270 struct strbuf path = STRBUF_INIT;
1271 size_t dirnamelen;
1f688557
JH
1272 DIR *dir;
1273 struct dirent *de;
543c5caa 1274 struct string_list garbage = STRING_LIST_INIT_DUP;
1f688557 1275
880fb8de
RS
1276 strbuf_addstr(&path, objdir);
1277 strbuf_addstr(&path, "/pack");
1278 dir = opendir(path.buf);
b5b16990 1279 if (!dir) {
26125f6b 1280 if (errno != ENOENT)
7616c6ca
NTND
1281 error_errno("unable to open object pack directory: %s",
1282 path.buf);
880fb8de 1283 strbuf_release(&path);
1f688557 1284 return;
b5b16990 1285 }
880fb8de
RS
1286 strbuf_addch(&path, '/');
1287 dirnamelen = path.len;
1f688557 1288 while ((de = readdir(dir)) != NULL) {
1f688557 1289 struct packed_git *p;
47bf4b0f 1290 size_t base_len;
543c5caa
NTND
1291
1292 if (is_dot_or_dotdot(de->d_name))
9cb18f56
JM
1293 continue;
1294
880fb8de
RS
1295 strbuf_setlen(&path, dirnamelen);
1296 strbuf_addstr(&path, de->d_name);
d90906a9 1297
47bf4b0f
JK
1298 base_len = path.len;
1299 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
d90906a9
NTND
1300 /* Don't reopen a pack we already have. */
1301 for (p = packed_git; p; p = p->next) {
47bf4b0f
JK
1302 size_t len;
1303 if (strip_suffix(p->pack_name, ".pack", &len) &&
1304 len == base_len &&
1305 !memcmp(p->pack_name, path.buf, len))
d90906a9
NTND
1306 break;
1307 }
1308 if (p == NULL &&
1309 /*
1310 * See if it really is a valid .idx file with
1311 * corresponding .pack file that we can map.
1312 */
880fb8de 1313 (p = add_packed_git(path.buf, path.len, local)) != NULL)
d90906a9 1314 install_packed_git(p);
86f7780c 1315 }
543c5caa
NTND
1316
1317 if (!report_garbage)
1318 continue;
1319
2975c770
JK
1320 if (ends_with(de->d_name, ".idx") ||
1321 ends_with(de->d_name, ".pack") ||
1322 ends_with(de->d_name, ".bitmap") ||
1323 ends_with(de->d_name, ".keep"))
880fb8de 1324 string_list_append(&garbage, path.buf);
543c5caa 1325 else
0a489b06 1326 report_garbage(PACKDIR_FILE_GARBAGE, path.buf);
1f688557 1327 }
5b35bcd5 1328 closedir(dir);
543c5caa
NTND
1329 report_pack_garbage(&garbage);
1330 string_list_clear(&garbage, 0);
880fb8de 1331 strbuf_release(&path);
1f688557
JH
1332}
1333
b867092f
JH
1334static int sort_pack(const void *a_, const void *b_)
1335{
1336 struct packed_git *a = *((struct packed_git **)a_);
1337 struct packed_git *b = *((struct packed_git **)b_);
1338 int st;
1339
1340 /*
1341 * Local packs tend to contain objects specific to our
1342 * variant of the project than remote ones. In addition,
1343 * remote ones could be on a network mounted filesystem.
1344 * Favor local ones for these reasons.
1345 */
1346 st = a->pack_local - b->pack_local;
1347 if (st)
1348 return -st;
1349
1350 /*
1351 * Younger packs tend to contain more recent objects,
1352 * and more recent objects tend to get accessed more
1353 * often.
1354 */
1355 if (a->mtime < b->mtime)
1356 return 1;
1357 else if (a->mtime == b->mtime)
1358 return 0;
1359 return -1;
1360}
1361
1362static void rearrange_packed_git(void)
1363{
1364 struct packed_git **ary, *p;
1365 int i, n;
1366
1367 for (n = 0, p = packed_git; p; p = p->next)
1368 n++;
1369 if (n < 2)
1370 return;
1371
1372 /* prepare an array of packed_git for easier sorting */
1373 ary = xcalloc(n, sizeof(struct packed_git *));
1374 for (n = 0, p = packed_git; p; p = p->next)
1375 ary[n++] = p;
1376
1377 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1378
1379 /* link them back again */
1380 for (i = 0; i < n - 1; i++)
1381 ary[i]->next = ary[i + 1];
1382 ary[n - 1]->next = NULL;
1383 packed_git = ary[0];
1384
1385 free(ary);
1386}
1387
637cdd9d 1388static int prepare_packed_git_run_once = 0;
9a217f2a 1389void prepare_packed_git(void)
1f688557 1390{
d5a63b99 1391 struct alternate_object_database *alt;
1f688557 1392
637cdd9d 1393 if (prepare_packed_git_run_once)
1f688557 1394 return;
9d835df2 1395 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 1396 prepare_alt_odb();
d5a63b99 1397 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 1398 alt->name[-1] = 0;
9d835df2 1399 prepare_packed_git_one(alt->base, 0);
1494e038 1400 alt->name[-1] = '/';
1f688557 1401 }
b867092f 1402 rearrange_packed_git();
637cdd9d
JK
1403 prepare_packed_git_run_once = 1;
1404}
1405
fc04c412 1406void reprepare_packed_git(void)
637cdd9d
JK
1407{
1408 prepare_packed_git_run_once = 0;
1409 prepare_packed_git();
1f688557
JH
1410}
1411
8eca0b47
NP
1412static void mark_bad_packed_object(struct packed_git *p,
1413 const unsigned char *sha1)
1414{
1415 unsigned i;
1416 for (i = 0; i < p->num_bad_objects; i++)
50a6c8ef 1417 if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
8eca0b47 1418 return;
50a6c8ef
JK
1419 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1420 st_mult(GIT_SHA1_RAWSZ,
1421 st_add(p->num_bad_objects, 1)));
1422 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
8eca0b47
NP
1423 p->num_bad_objects++;
1424}
1425
b6c4cecc 1426static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
ac939109
NP
1427{
1428 struct packed_git *p;
1429 unsigned i;
1430
1431 for (p = packed_git; p; p = p->next)
1432 for (i = 0; i < p->num_bad_objects; i++)
1433 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
b6c4cecc
JH
1434 return p;
1435 return NULL;
ac939109
NP
1436}
1437
090ea126
NTND
1438/*
1439 * With an in-core object data in "map", rehash it to make sure the
1440 * object name actually matches "sha1" to detect object corruption.
1441 * With "map" == NULL, try reading the object named with "sha1" using
1442 * the streaming interface and rehash it to do the same.
1443 */
1444int check_sha1_signature(const unsigned char *sha1, void *map,
1445 unsigned long size, const char *type)
0fcfd160
LT
1446{
1447 unsigned char real_sha1[20];
090ea126
NTND
1448 enum object_type obj_type;
1449 struct git_istream *st;
1450 git_SHA_CTX c;
1451 char hdr[32];
1452 int hdrlen;
1453
1454 if (map) {
1455 hash_sha1_file(map, size, type, real_sha1);
1456 return hashcmp(sha1, real_sha1) ? -1 : 0;
1457 }
1458
1459 st = open_istream(sha1, &obj_type, &size, NULL);
1460 if (!st)
1461 return -1;
1462
1463 /* Generate the header */
ef1286d3 1464 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
090ea126
NTND
1465
1466 /* Sha1.. */
1467 git_SHA1_Init(&c);
1468 git_SHA1_Update(&c, hdr, hdrlen);
1469 for (;;) {
1470 char buf[1024 * 16];
1471 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1472
f54fac53
JK
1473 if (readlen < 0) {
1474 close_istream(st);
1475 return -1;
1476 }
090ea126
NTND
1477 if (!readlen)
1478 break;
1479 git_SHA1_Update(&c, buf, readlen);
1480 }
1481 git_SHA1_Final(real_sha1, &c);
1482 close_istream(st);
a89fccd2 1483 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1484}
1485
ec73f580 1486int git_open_noatime(const char *name)
44d1c19e
LT
1487{
1488 static int sha1_file_open_flag = O_NOATIME;
44d1c19e 1489
f2e872aa 1490 for (;;) {
dff6f280
CB
1491 int fd;
1492
1493 errno = 0;
1494 fd = open(name, O_RDONLY | sha1_file_open_flag);
44d1c19e 1495 if (fd >= 0)
f2e872aa
SP
1496 return fd;
1497
f2e872aa
SP
1498 /* Might the failure be due to O_NOATIME? */
1499 if (errno != ENOENT && sha1_file_open_flag) {
44d1c19e 1500 sha1_file_open_flag = 0;
f2e872aa
SP
1501 continue;
1502 }
1503
1504 return -1;
44d1c19e 1505 }
44d1c19e
LT
1506}
1507
052fe5ea
JK
1508static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1509{
052fe5ea
JK
1510 struct alternate_object_database *alt;
1511
30d6c6ea 1512 if (!lstat(sha1_file_name(sha1), st))
052fe5ea
JK
1513 return 0;
1514
1515 prepare_alt_odb();
1516 errno = ENOENT;
1517 for (alt = alt_odb_list; alt; alt = alt->next) {
30d6c6ea 1518 fill_sha1_path(alt->name, sha1);
052fe5ea
JK
1519 if (!lstat(alt->base, st))
1520 return 0;
1521 }
1522
1523 return -1;
1524}
1525
44d1c19e
LT
1526static int open_sha1_file(const unsigned char *sha1)
1527{
1528 int fd;
44d1c19e 1529 struct alternate_object_database *alt;
d6c8a05b 1530 int most_interesting_errno;
44d1c19e 1531
30d6c6ea 1532 fd = git_open_noatime(sha1_file_name(sha1));
44d1c19e
LT
1533 if (fd >= 0)
1534 return fd;
d6c8a05b 1535 most_interesting_errno = errno;
44d1c19e
LT
1536
1537 prepare_alt_odb();
44d1c19e 1538 for (alt = alt_odb_list; alt; alt = alt->next) {
30d6c6ea 1539 fill_sha1_path(alt->name, sha1);
f4e51683 1540 fd = git_open_noatime(alt->base);
44d1c19e
LT
1541 if (fd >= 0)
1542 return fd;
d6c8a05b
JK
1543 if (most_interesting_errno == ENOENT)
1544 most_interesting_errno = errno;
44d1c19e 1545 }
d6c8a05b 1546 errno = most_interesting_errno;
44d1c19e
LT
1547 return -1;
1548}
1549
f0270efd 1550void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1551{
0fcfd160 1552 void *map;
144bde78 1553 int fd;
ace1534d 1554
44d1c19e
LT
1555 fd = open_sha1_file(sha1);
1556 map = NULL;
1557 if (fd >= 0) {
1558 struct stat st;
0fcfd160 1559
44d1c19e
LT
1560 if (!fstat(fd, &st)) {
1561 *size = xsize_t(st.st_size);
33e42de0
MM
1562 if (!*size) {
1563 /* mmap() is forbidden on empty files */
1564 error("object file %s is empty", sha1_file_name(sha1));
1565 return NULL;
1566 }
44d1c19e 1567 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1568 }
44d1c19e 1569 close(fd);
0fcfd160 1570 }
0fcfd160
LT
1571 return map;
1572}
1573
09ded04b
NP
1574unsigned long unpack_object_header_buffer(const unsigned char *buf,
1575 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1576{
ad1ed5ee 1577 unsigned shift;
48fb7deb 1578 unsigned long size, c;
ad1ed5ee
JH
1579 unsigned long used = 0;
1580
1581 c = buf[used++];
1582 *type = (c >> 4) & 7;
1583 size = c & 15;
1584 shift = 4;
1585 while (c & 0x80) {
f630cfda 1586 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1587 error("bad object header");
ea4f9685
JH
1588 size = used = 0;
1589 break;
09ded04b 1590 }
ad1ed5ee
JH
1591 c = buf[used++];
1592 size += (c & 0x7f) << shift;
1593 shift += 7;
1594 }
1595 *sizep = size;
1596 return used;
1597}
1598
eb4f4076 1599int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
ad1ed5ee 1600{
c4483576
LT
1601 /* Get the data stream */
1602 memset(stream, 0, sizeof(*stream));
1603 stream->next_in = map;
1604 stream->avail_in = mapsize;
1605 stream->next_out = buffer;
93821bd9
LT
1606 stream->avail_out = bufsiz;
1607
39c68542 1608 git_inflate_init(stream);
cc5c54e7 1609 return git_inflate(stream, 0);
c4483576
LT
1610}
1611
46f03448
KN
1612static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1613 unsigned long mapsize, void *buffer,
1614 unsigned long bufsiz, struct strbuf *header)
1615{
1616 int status;
1617
1618 status = unpack_sha1_header(stream, map, mapsize, buffer, bufsiz);
1619
1620 /*
1621 * Check if entire header is unpacked in the first iteration.
1622 */
1623 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1624 return 0;
1625
1626 /*
1627 * buffer[0..bufsiz] was not large enough. Copy the partial
1628 * result out to header, and then append the result of further
1629 * reading the stream.
1630 */
1631 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1632 stream->next_out = buffer;
1633 stream->avail_out = bufsiz;
1634
1635 do {
1636 status = git_inflate(stream, 0);
1637 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1638 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1639 return 0;
1640 stream->next_out = buffer;
1641 stream->avail_out = bufsiz;
1642 } while (status != Z_STREAM_END);
1643 return -1;
1644}
1645
ef49a7a0 1646static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1647{
1648 int bytes = strlen(buffer) + 1;
3aee68aa 1649 unsigned char *buf = xmallocz(size);
93821bd9 1650 unsigned long n;
7efbff75 1651 int status = Z_OK;
5180cacc 1652
93821bd9
LT
1653 n = stream->total_out - bytes;
1654 if (n > size)
1655 n = size;
1656 memcpy(buf, (char *) buffer + bytes, n);
1657 bytes = n;
456cdf6e
LT
1658 if (bytes <= size) {
1659 /*
1660 * The above condition must be (bytes <= size), not
1661 * (bytes < size). In other words, even though we
ccf5ace0 1662 * expect no more output and set avail_out to zero,
456cdf6e
LT
1663 * the input zlib stream may have bytes that express
1664 * "this concludes the stream", and we *do* want to
1665 * eat that input.
1666 *
1667 * Otherwise we would not be able to test that we
1668 * consumed all the input to reach the expected size;
1669 * we also want to check that zlib tells us that all
1670 * went well with status == Z_STREAM_END at the end.
1671 */
5180cacc
LT
1672 stream->next_out = buf + bytes;
1673 stream->avail_out = size - bytes;
7efbff75 1674 while (status == Z_OK)
39c68542 1675 status = git_inflate(stream, Z_FINISH);
5180cacc 1676 }
456cdf6e 1677 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1678 git_inflate_end(stream);
7efbff75
JH
1679 return buf;
1680 }
1681
1682 if (status < 0)
1683 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1684 else if (stream->avail_in)
1685 error("garbage at end of loose object '%s'",
1686 sha1_to_hex(sha1));
1687 free(buf);
1688 return NULL;
5180cacc
LT
1689}
1690
1691/*
1692 * We used to just use "sscanf()", but that's actually way
1693 * too permissive for what we want to check. So do an anal
1694 * object header parse by hand.
1695 */
46f03448
KN
1696static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1697 unsigned int flags)
5180cacc 1698{
46f03448 1699 const char *type_buf = hdr;
5180cacc 1700 unsigned long size;
46f03448 1701 int type, type_len = 0;
5180cacc
LT
1702
1703 /*
46f03448 1704 * The type can be of any size but is followed by
21666f1a 1705 * a space.
5180cacc 1706 */
5180cacc
LT
1707 for (;;) {
1708 char c = *hdr++;
1709 if (c == ' ')
1710 break;
46f03448 1711 type_len++;
5180cacc 1712 }
46f03448
KN
1713
1714 type = type_from_string_gently(type_buf, type_len, 1);
1715 if (oi->typename)
1716 strbuf_add(oi->typename, type_buf, type_len);
1717 /*
1718 * Set type to 0 if its an unknown object and
1719 * we're obtaining the type using '--allow-unkown-type'
1720 * option.
1721 */
1722 if ((flags & LOOKUP_UNKNOWN_OBJECT) && (type < 0))
1723 type = 0;
1724 else if (type < 0)
1725 die("invalid object type");
1726 if (oi->typep)
1727 *oi->typep = type;
5180cacc
LT
1728
1729 /*
1730 * The length must follow immediately, and be in canonical
1731 * decimal format (ie "010" is not valid).
1732 */
1733 size = *hdr++ - '0';
1734 if (size > 9)
1735 return -1;
1736 if (size) {
1737 for (;;) {
1738 unsigned long c = *hdr - '0';
1739 if (c > 9)
1740 break;
1741 hdr++;
1742 size = size * 10 + c;
1743 }
1744 }
46f03448
KN
1745
1746 if (oi->sizep)
1747 *oi->sizep = size;
5180cacc
LT
1748
1749 /*
1750 * The length must be followed by a zero byte
1751 */
46f03448
KN
1752 return *hdr ? -1 : type;
1753}
1754
1755int parse_sha1_header(const char *hdr, unsigned long *sizep)
1756{
1757 struct object_info oi;
1758
1759 oi.sizep = sizep;
1760 oi.typename = NULL;
1761 oi.typep = NULL;
1762 return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
5180cacc
LT
1763}
1764
7efbff75 1765static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1766{
5180cacc 1767 int ret;
ef49a7a0 1768 git_zstream stream;
5180cacc 1769 char hdr[8192];
0fcfd160 1770
5180cacc 1771 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1772 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1773 return NULL;
1774
7efbff75 1775 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1776}
1777
54dab52a
NP
1778unsigned long get_size_from_delta(struct packed_git *p,
1779 struct pack_window **w_curs,
1780 off_t curpos)
1781{
1782 const unsigned char *data;
1783 unsigned char delta_head[20], *in;
ef49a7a0 1784 git_zstream stream;
54dab52a
NP
1785 int st;
1786
1787 memset(&stream, 0, sizeof(stream));
1788 stream.next_out = delta_head;
1789 stream.avail_out = sizeof(delta_head);
1790
39c68542 1791 git_inflate_init(&stream);
54dab52a
NP
1792 do {
1793 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1794 stream.next_in = in;
39c68542 1795 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1796 curpos += stream.next_in - in;
1797 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1798 stream.total_out < sizeof(delta_head));
39c68542 1799 git_inflate_end(&stream);
3d77d877
NP
1800 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1801 error("delta data unpack-initial failed");
1802 return 0;
1803 }
54dab52a
NP
1804
1805 /* Examine the initial part of the delta to figure out
1806 * the result size.
1807 */
1808 data = delta_head;
1809
1810 /* ignore base size */
1811 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1812
1813 /* Read the result size */
1814 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1815}
1816
c4001d92 1817static off_t get_delta_base(struct packed_git *p,
03e79c88 1818 struct pack_window **w_curs,
c4001d92 1819 off_t *curpos,
21666f1a 1820 enum object_type type,
c4001d92 1821 off_t delta_obj_offset)
eb32d236 1822{
2b87c45b 1823 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1824 off_t base_offset;
eb32d236 1825
8d8a4ea5
SP
1826 /* use_pack() assured us we have [base_info, base_info + 20)
1827 * as a range that we can look at without walking off the
1828 * end of the mapped window. Its actually the hash size
1829 * that is assured. An OFS_DELTA longer than the hash size
1830 * is stupid, as then a REF_DELTA would be smaller to store.
1831 */
21666f1a 1832 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1833 unsigned used = 0;
1834 unsigned char c = base_info[used++];
1835 base_offset = c & 127;
1836 while (c & 128) {
1837 base_offset += 1;
8723f216 1838 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1839 return 0; /* overflow */
eb32d236
NP
1840 c = base_info[used++];
1841 base_offset = (base_offset << 7) + (c & 127);
1842 }
1843 base_offset = delta_obj_offset - base_offset;
d8f32556 1844 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1845 return 0; /* out of bound */
2b87c45b 1846 *curpos += used;
21666f1a 1847 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1848 /* The base entry _must_ be in the same pack */
1849 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1850 *curpos += 20;
eb32d236
NP
1851 } else
1852 die("I am totally screwed");
2b87c45b 1853 return base_offset;
eb32d236
NP
1854}
1855
5d642e75
JK
1856/*
1857 * Like get_delta_base above, but we return the sha1 instead of the pack
1858 * offset. This means it is cheaper for REF deltas (we do not have to do
1859 * the final object lookup), but more expensive for OFS deltas (we
1860 * have to load the revidx to convert the offset back into a sha1).
1861 */
1862static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1863 struct pack_window **w_curs,
1864 off_t curpos,
1865 enum object_type type,
1866 off_t delta_obj_offset)
1867{
1868 if (type == OBJ_REF_DELTA) {
1869 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1870 return base;
1871 } else if (type == OBJ_OFS_DELTA) {
1872 struct revindex_entry *revidx;
1873 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1874 type, delta_obj_offset);
1875
1876 if (!base_offset)
1877 return NULL;
1878
1879 revidx = find_pack_revindex(p, base_offset);
1880 if (!revidx)
1881 return NULL;
1882
1883 return nth_packed_object_sha1(p, revidx->nr);
1884 } else
1885 return NULL;
1886}
1887
f8c8abc5
JH
1888int unpack_object_header(struct packed_git *p,
1889 struct pack_window **w_curs,
1890 off_t *curpos,
1891 unsigned long *sizep)
a733cb60 1892{
03e79c88 1893 unsigned char *base;
ef49a7a0 1894 unsigned long left;
ad1ed5ee 1895 unsigned long used;
2b87c45b 1896 enum object_type type;
a733cb60 1897
8d8a4ea5 1898 /* use_pack() assures us we have [base, base + 20) available
0353a0c4 1899 * as a range that we can look at. (Its actually the hash
3dff5379 1900 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1901 * the maximum deflated object size is 2^137, which is just
1902 * insane, so we know won't exceed what we have been given.
1903 */
2b87c45b 1904 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1905 used = unpack_object_header_buffer(base, left, &type, sizep);
1906 if (!used) {
1907 type = OBJ_BAD;
1908 } else
1909 *curpos += used;
ad1ed5ee 1910
2b87c45b 1911 return type;
a733cb60
LT
1912}
1913
790d96c0
TR
1914static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1915{
1916 int type;
1917 struct revindex_entry *revidx;
1918 const unsigned char *sha1;
1919 revidx = find_pack_revindex(p, obj_offset);
1920 if (!revidx)
1921 return OBJ_BAD;
1922 sha1 = nth_packed_object_sha1(p, revidx->nr);
1923 mark_bad_packed_object(p, sha1);
1924 type = sha1_object_info(sha1, NULL);
1925 if (type <= OBJ_NONE)
1926 return OBJ_BAD;
1927 return type;
1928}
1929
790d96c0
TR
1930#define POI_STACK_PREALLOC 64
1931
90191d37
JK
1932static enum object_type packed_to_object_type(struct packed_git *p,
1933 off_t obj_offset,
1934 enum object_type type,
1935 struct pack_window **w_curs,
1936 off_t curpos)
1f688557 1937{
790d96c0
TR
1938 off_t small_poi_stack[POI_STACK_PREALLOC];
1939 off_t *poi_stack = small_poi_stack;
1940 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
5db47c2b 1941
790d96c0
TR
1942 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1943 off_t base_offset;
90191d37 1944 unsigned long size;
790d96c0
TR
1945 /* Push the object we're going to leave behind */
1946 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1947 poi_stack_alloc = alloc_nr(poi_stack_nr);
b32fa95f 1948 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
790d96c0
TR
1949 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1950 } else {
1951 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1952 }
1953 poi_stack[poi_stack_nr++] = obj_offset;
1954 /* If parsing the base offset fails, just unwind */
90191d37 1955 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
790d96c0
TR
1956 if (!base_offset)
1957 goto unwind;
1958 curpos = obj_offset = base_offset;
90191d37 1959 type = unpack_object_header(p, w_curs, &curpos, &size);
790d96c0
TR
1960 if (type <= OBJ_NONE) {
1961 /* If getting the base itself fails, we first
1962 * retry the base, otherwise unwind */
1963 type = retry_bad_packed_offset(p, base_offset);
1964 if (type > OBJ_NONE)
1965 goto out;
1966 goto unwind;
1967 }
1968 }
1969
21666f1a 1970 switch (type) {
790d96c0 1971 case OBJ_BAD:
a733cb60 1972 case OBJ_COMMIT:
a733cb60 1973 case OBJ_TREE:
a733cb60 1974 case OBJ_BLOB:
a733cb60 1975 case OBJ_TAG:
a69d0943 1976 break;
1f688557 1977 default:
3d77d877
NP
1978 error("unknown object type %i at offset %"PRIuMAX" in %s",
1979 type, (uintmax_t)obj_offset, p->pack_name);
1980 type = OBJ_BAD;
1f688557 1981 }
790d96c0
TR
1982
1983out:
1984 if (poi_stack != small_poi_stack)
1985 free(poi_stack);
21666f1a 1986 return type;
790d96c0
TR
1987
1988unwind:
1989 while (poi_stack_nr) {
1990 obj_offset = poi_stack[--poi_stack_nr];
1991 type = retry_bad_packed_offset(p, obj_offset);
1992 if (type > OBJ_NONE)
1993 goto out;
1994 }
1995 type = OBJ_BAD;
1996 goto out;
1f688557
JH
1997}
1998
90191d37 1999static int packed_object_info(struct packed_git *p, off_t obj_offset,
23c339c0 2000 struct object_info *oi)
90191d37
JK
2001{
2002 struct pack_window *w_curs = NULL;
2003 unsigned long size;
2004 off_t curpos = obj_offset;
2005 enum object_type type;
2006
412916ee
JK
2007 /*
2008 * We always get the representation type, but only convert it to
2009 * a "real" type later if the caller is interested.
2010 */
90191d37
JK
2011 type = unpack_object_header(p, &w_curs, &curpos, &size);
2012
23c339c0 2013 if (oi->sizep) {
90191d37
JK
2014 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2015 off_t tmp_pos = curpos;
2016 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
2017 type, obj_offset);
2018 if (!base_offset) {
2019 type = OBJ_BAD;
2020 goto out;
2021 }
23c339c0
JK
2022 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2023 if (*oi->sizep == 0) {
90191d37
JK
2024 type = OBJ_BAD;
2025 goto out;
2026 }
2027 } else {
23c339c0 2028 *oi->sizep = size;
90191d37
JK
2029 }
2030 }
2031
23c339c0 2032 if (oi->disk_sizep) {
90191d37 2033 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
23c339c0 2034 *oi->disk_sizep = revidx[1].offset - obj_offset;
90191d37
JK
2035 }
2036
23c339c0
JK
2037 if (oi->typep) {
2038 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
2039 if (*oi->typep < 0) {
412916ee
JK
2040 type = OBJ_BAD;
2041 goto out;
2042 }
2043 }
90191d37 2044
5d642e75
JK
2045 if (oi->delta_base_sha1) {
2046 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2047 const unsigned char *base;
2048
2049 base = get_delta_base_sha1(p, &w_curs, curpos,
2050 type, obj_offset);
2051 if (!base) {
2052 type = OBJ_BAD;
2053 goto out;
2054 }
2055
2056 hashcpy(oi->delta_base_sha1, base);
2057 } else
2058 hashclr(oi->delta_base_sha1);
2059 }
2060
90191d37
JK
2061out:
2062 unuse_pack(&w_curs);
2063 return type;
2064}
2065
eb950c19 2066static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 2067 struct pack_window **w_curs,
c4001d92 2068 off_t curpos,
eb950c19 2069 unsigned long size)
de530aaa
SP
2070{
2071 int st;
ef49a7a0 2072 git_zstream stream;
079afb18 2073 unsigned char *buffer, *in;
de530aaa 2074
735efde8
NTND
2075 buffer = xmallocz_gently(size);
2076 if (!buffer)
2077 return NULL;
de530aaa 2078 memset(&stream, 0, sizeof(stream));
de530aaa 2079 stream.next_out = buffer;
39eea7bd 2080 stream.avail_out = size + 1;
de530aaa 2081
39c68542 2082 git_inflate_init(&stream);
079afb18 2083 do {
2b87c45b 2084 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 2085 stream.next_in = in;
39c68542 2086 st = git_inflate(&stream, Z_FINISH);
39eea7bd
JH
2087 if (!stream.avail_out)
2088 break; /* the payload is larger than it should be */
2b87c45b 2089 curpos += stream.next_in - in;
079afb18 2090 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 2091 git_inflate_end(&stream);
de530aaa
SP
2092 if ((st != Z_STREAM_END) || stream.total_out != size) {
2093 free(buffer);
2094 return NULL;
2095 }
2096
2097 return buffer;
2098}
2099
e5e01619
LT
2100#define MAX_DELTA_CACHE (256)
2101
18bdec11 2102static size_t delta_base_cached;
5e08ecbf
NP
2103
2104static struct delta_base_cache_lru_list {
2105 struct delta_base_cache_lru_list *prev;
2106 struct delta_base_cache_lru_list *next;
2107} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
2108
e5e01619 2109static struct delta_base_cache_entry {
5e08ecbf
NP
2110 struct delta_base_cache_lru_list lru;
2111 void *data;
e5e01619
LT
2112 struct packed_git *p;
2113 off_t base_offset;
2114 unsigned long size;
e5e01619
LT
2115 enum object_type type;
2116} delta_base_cache[MAX_DELTA_CACHE];
2117
2118static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
2119{
2120 unsigned long hash;
2121
56a1a3ab 2122 hash = (unsigned long)(intptr_t)p + (unsigned long)base_offset;
e5e01619 2123 hash += (hash >> 8) + (hash >> 16);
3358004a 2124 return hash % MAX_DELTA_CACHE;
e5e01619
LT
2125}
2126
84dd81c1
TR
2127static struct delta_base_cache_entry *
2128get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
5266d369
JH
2129{
2130 unsigned long hash = pack_entry_hash(p, base_offset);
84dd81c1
TR
2131 return delta_base_cache + hash;
2132}
2133
2134static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
2135 struct packed_git *p, off_t base_offset)
2136{
5266d369
JH
2137 return (ent->data && ent->p == p && ent->base_offset == base_offset);
2138}
2139
84dd81c1
TR
2140static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2141{
2142 struct delta_base_cache_entry *ent;
2143 ent = get_delta_base_cache_entry(p, base_offset);
2144 return eq_delta_base_cache_entry(ent, p, base_offset);
2145}
2146
2147static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2148{
2149 ent->data = NULL;
2150 ent->lru.next->prev = ent->lru.prev;
2151 ent->lru.prev->next = ent->lru.next;
2152 delta_base_cached -= ent->size;
2153}
2154
62f255ad 2155static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 2156 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 2157{
84dd81c1 2158 struct delta_base_cache_entry *ent;
e5e01619 2159 void *ret;
e5e01619 2160
84dd81c1
TR
2161 ent = get_delta_base_cache_entry(p, base_offset);
2162
2163 if (!eq_delta_base_cache_entry(ent, p, base_offset))
749bc58c 2164 return unpack_entry(p, base_offset, type, base_size);
e5e01619 2165
84dd81c1
TR
2166 ret = ent->data;
2167
2168 if (!keep_cache)
2169 clear_delta_base_cache_entry(ent);
2170 else
182af834 2171 ret = xmemdupz(ent->data, ent->size);
e5e01619
LT
2172 *type = ent->type;
2173 *base_size = ent->size;
2174 return ret;
62f255ad
LT
2175}
2176
18bdec11
SP
2177static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2178{
2179 if (ent->data) {
2180 free(ent->data);
2181 ent->data = NULL;
5e08ecbf
NP
2182 ent->lru.next->prev = ent->lru.prev;
2183 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
2184 delta_base_cached -= ent->size;
2185 }
2186}
2187
3d20c636
SP
2188void clear_delta_base_cache(void)
2189{
2190 unsigned long p;
2191 for (p = 0; p < MAX_DELTA_CACHE; p++)
2192 release_delta_base_cache(&delta_base_cache[p]);
2193}
2194
62f255ad
LT
2195static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2196 void *base, unsigned long base_size, enum object_type type)
2197{
5e08ecbf 2198 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 2199 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 2200 struct delta_base_cache_lru_list *lru;
e5e01619 2201
18bdec11
SP
2202 release_delta_base_cache(ent);
2203 delta_base_cached += base_size;
5e08ecbf
NP
2204
2205 for (lru = delta_base_cache_lru.next;
2206 delta_base_cached > delta_base_cache_limit
2207 && lru != &delta_base_cache_lru;
2208 lru = lru->next) {
2209 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
2210 if (f->type == OBJ_BLOB)
2211 release_delta_base_cache(f);
2212 }
5e08ecbf
NP
2213 for (lru = delta_base_cache_lru.next;
2214 delta_base_cached > delta_base_cache_limit
2215 && lru != &delta_base_cache_lru;
2216 lru = lru->next) {
2217 struct delta_base_cache_entry *f = (void *)lru;
2218 release_delta_base_cache(f);
2219 }
18bdec11 2220
e5e01619
LT
2221 ent->p = p;
2222 ent->base_offset = base_offset;
2223 ent->type = type;
2224 ent->data = base;
2225 ent->size = base_size;
5e08ecbf
NP
2226 ent->lru.next = &delta_base_cache_lru;
2227 ent->lru.prev = delta_base_cache_lru.prev;
2228 delta_base_cache_lru.prev->next = &ent->lru;
2229 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
2230}
2231
c2c5b270
CC
2232static void *read_object(const unsigned char *sha1, enum object_type *type,
2233 unsigned long *size);
2234
5f44324d
JH
2235static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2236{
67dc598e
KB
2237 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2238 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2239 p->pack_name, (uintmax_t)obj_offset);
5f44324d
JH
2240}
2241
0e8189e2
NP
2242int do_check_packed_object_crc;
2243
abe601bb
TR
2244#define UNPACK_ENTRY_STACK_PREALLOC 64
2245struct unpack_entry_stack_ent {
2246 off_t obj_offset;
2247 off_t curpos;
2248 unsigned long size;
2249};
2250
c4001d92 2251void *unpack_entry(struct packed_git *p, off_t obj_offset,
abe601bb 2252 enum object_type *final_type, unsigned long *final_size)
1f688557 2253{
03e79c88 2254 struct pack_window *w_curs = NULL;
c4001d92 2255 off_t curpos = obj_offset;
abe601bb
TR
2256 void *data = NULL;
2257 unsigned long size;
2258 enum object_type type;
2259 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2260 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2261 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2262 int base_from_cache = 0;
1f688557 2263
67dc598e 2264 write_pack_access_log(p, obj_offset);
5f44324d 2265
abe601bb
TR
2266 /* PHASE 1: drill down to the innermost base object */
2267 for (;;) {
2268 off_t base_offset;
2269 int i;
2270 struct delta_base_cache_entry *ent;
2271
77965f8b
NTND
2272 ent = get_delta_base_cache_entry(p, curpos);
2273 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2274 type = ent->type;
2275 data = ent->data;
2276 size = ent->size;
2277 clear_delta_base_cache_entry(ent);
2278 base_from_cache = 1;
2279 break;
2280 }
2281
abe601bb
TR
2282 if (do_check_packed_object_crc && p->index_version > 1) {
2283 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
211c61c6 2284 off_t len = revidx[1].offset - obj_offset;
abe601bb
TR
2285 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2286 const unsigned char *sha1 =
2287 nth_packed_object_sha1(p, revidx->nr);
2288 error("bad packed object CRC for %s",
2289 sha1_to_hex(sha1));
2290 mark_bad_packed_object(p, sha1);
2291 unuse_pack(&w_curs);
2292 return NULL;
2293 }
2294 }
2295
abe601bb
TR
2296 type = unpack_object_header(p, &w_curs, &curpos, &size);
2297 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2298 break;
2299
2300 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2301 if (!base_offset) {
2302 error("failed to validate delta base reference "
2303 "at offset %"PRIuMAX" from %s",
2304 (uintmax_t)curpos, p->pack_name);
2305 /* bail to phase 2, in hopes of recovery */
2306 data = NULL;
2307 break;
2308 }
2309
2310 /* push object, proceed to base */
2311 if (delta_stack_nr >= delta_stack_alloc
2312 && delta_stack == small_delta_stack) {
2313 delta_stack_alloc = alloc_nr(delta_stack_nr);
b32fa95f 2314 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
abe601bb
TR
2315 memcpy(delta_stack, small_delta_stack,
2316 sizeof(*delta_stack)*delta_stack_nr);
2317 } else {
2318 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2319 }
2320 i = delta_stack_nr++;
2321 delta_stack[i].obj_offset = obj_offset;
2322 delta_stack[i].curpos = curpos;
2323 delta_stack[i].size = size;
2324
2325 curpos = obj_offset = base_offset;
0e8189e2
NP
2326 }
2327
abe601bb
TR
2328 /* PHASE 2: handle the base */
2329 switch (type) {
eb32d236
NP
2330 case OBJ_OFS_DELTA:
2331 case OBJ_REF_DELTA:
abe601bb
TR
2332 if (data)
2333 die("BUG in unpack_entry: left loop at a valid delta");
4d703a1a 2334 break;
a733cb60 2335 case OBJ_COMMIT:
a733cb60 2336 case OBJ_TREE:
a733cb60 2337 case OBJ_BLOB:
a733cb60 2338 case OBJ_TAG:
abe601bb
TR
2339 if (!base_from_cache)
2340 data = unpack_compressed_entry(p, &w_curs, curpos, size);
4d703a1a 2341 break;
1f688557 2342 default:
8eca0b47
NP
2343 data = NULL;
2344 error("unknown object type %i at offset %"PRIuMAX" in %s",
abe601bb
TR
2345 type, (uintmax_t)obj_offset, p->pack_name);
2346 }
2347
2348 /* PHASE 3: apply deltas in order */
2349
2350 /* invariants:
2351 * 'data' holds the base data, or NULL if there was corruption
2352 */
2353 while (delta_stack_nr) {
2354 void *delta_data;
2355 void *base = data;
2356 unsigned long delta_size, base_size = size;
2357 int i;
2358
2359 data = NULL;
2360
2361 if (base)
2362 add_delta_base_cache(p, obj_offset, base, base_size, type);
2363
2364 if (!base) {
2365 /*
2366 * We're probably in deep shit, but let's try to fetch
2367 * the required base anyway from another pack or loose.
2368 * This is costly but should happen only in the presence
2369 * of a corrupted pack, and is better than failing outright.
2370 */
2371 struct revindex_entry *revidx;
2372 const unsigned char *base_sha1;
2373 revidx = find_pack_revindex(p, obj_offset);
2374 if (revidx) {
2375 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2376 error("failed to read delta base object %s"
2377 " at offset %"PRIuMAX" from %s",
2378 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2379 p->pack_name);
2380 mark_bad_packed_object(p, base_sha1);
2381 base = read_object(base_sha1, &type, &base_size);
2382 }
2383 }
2384
2385 i = --delta_stack_nr;
2386 obj_offset = delta_stack[i].obj_offset;
2387 curpos = delta_stack[i].curpos;
2388 delta_size = delta_stack[i].size;
2389
2390 if (!base)
2391 continue;
2392
2393 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2394
2395 if (!delta_data) {
2396 error("failed to unpack compressed delta "
2397 "at offset %"PRIuMAX" from %s",
2398 (uintmax_t)curpos, p->pack_name);
abe601bb
TR
2399 data = NULL;
2400 continue;
2401 }
2402
2403 data = patch_delta(base, base_size,
2404 delta_data, delta_size,
2405 &size);
1ee886c1
JK
2406
2407 /*
2408 * We could not apply the delta; warn the user, but keep going.
2409 * Our failure will be noticed either in the next iteration of
2410 * the loop, or if this is the final delta, in the caller when
2411 * we return NULL. Those code paths will take care of making
2412 * a more explicit warning and retrying with another copy of
2413 * the object.
2414 */
abe601bb 2415 if (!data)
1ee886c1 2416 error("failed to apply delta");
abe601bb 2417
4b8f772c 2418 free(delta_data);
1f688557 2419 }
abe601bb
TR
2420
2421 *final_type = type;
2422 *final_size = size;
2423
03e79c88 2424 unuse_pack(&w_curs);
019d1e65
NTND
2425
2426 if (delta_stack != small_delta_stack)
2427 free(delta_stack);
2428
21666f1a 2429 return data;
1f688557
JH
2430}
2431
d079837e 2432const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 2433 uint32_t n)
9a217f2a 2434{
42873078 2435 const unsigned char *index = p->index_data;
d079837e
SP
2436 if (!index) {
2437 if (open_pack_index(p))
2438 return NULL;
2439 index = p->index_data;
2440 }
57059091 2441 if (n >= p->num_objects)
d72308e0 2442 return NULL;
74e34e1f
NP
2443 index += 4 * 256;
2444 if (p->index_version == 1) {
2445 return index + 24 * n + 4;
2446 } else {
2447 index += 8;
2448 return index + 20 * n;
2449 }
2450}
2451
47fe3f6e
JK
2452void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
2453{
2454 const unsigned char *ptr = vptr;
2455 const unsigned char *start = p->index_data;
2456 const unsigned char *end = start + p->index_size;
2457 if (ptr < start)
7465feba 2458 die(_("offset before start of pack index for %s (corrupt index?)"),
47fe3f6e
JK
2459 p->pack_name);
2460 /* No need to check for underflow; .idx files must be at least 8 bytes */
2461 if (ptr >= end - 8)
7465feba 2462 die(_("offset beyond end of pack index for %s (truncated index?)"),
47fe3f6e
JK
2463 p->pack_name);
2464}
2465
99093238 2466off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
2467{
2468 const unsigned char *index = p->index_data;
2469 index += 4 * 256;
2470 if (p->index_version == 1) {
2471 return ntohl(*((uint32_t *)(index + 24 * n)));
2472 } else {
2473 uint32_t off;
2474 index += 8 + p->num_objects * (20 + 4);
2475 off = ntohl(*((uint32_t *)(index + 4 * n)));
2476 if (!(off & 0x80000000))
2477 return off;
2478 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
47fe3f6e 2479 check_pack_index_ptr(p, index);
74e34e1f
NP
2480 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2481 ntohl(*((uint32_t *)(index + 4)));
2482 }
9a217f2a
JH
2483}
2484
c4001d92 2485off_t find_pack_entry_one(const unsigned char *sha1,
43057304 2486 struct packed_git *p)
1f688557 2487{
42873078 2488 const uint32_t *level1_ofs = p->index_data;
42873078 2489 const unsigned char *index = p->index_data;
628522ec
JH
2490 unsigned hi, lo, stride;
2491 static int use_lookup = -1;
2492 static int debug_lookup = -1;
2493
2494 if (debug_lookup < 0)
2495 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 2496
d079837e
SP
2497 if (!index) {
2498 if (open_pack_index(p))
2499 return 0;
2500 level1_ofs = p->index_data;
2501 index = p->index_data;
2502 }
74e34e1f
NP
2503 if (p->index_version > 1) {
2504 level1_ofs += 2;
2505 index += 8;
2506 }
42873078 2507 index += 4 * 256;
74e34e1f
NP
2508 hi = ntohl(level1_ofs[*sha1]);
2509 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
2510 if (p->index_version > 1) {
2511 stride = 20;
2512 } else {
2513 stride = 24;
2514 index += 4;
2515 }
2516
2517 if (debug_lookup)
6e1c2344 2518 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
2519 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2520
2521 if (use_lookup < 0)
2522 use_lookup = !!getenv("GIT_USE_LOOKUP");
2523 if (use_lookup) {
2524 int pos = sha1_entry_pos(index, stride, 0,
2525 lo, hi, p->num_objects, sha1);
2526 if (pos < 0)
2527 return 0;
2528 return nth_packed_object_offset(p, pos);
2529 }
1f688557
JH
2530
2531 do {
74e34e1f 2532 unsigned mi = (lo + hi) / 2;
628522ec
JH
2533 int cmp = hashcmp(index + mi * stride, sha1);
2534
2535 if (debug_lookup)
2536 printf("lo %u hi %u rg %u mi %u\n",
2537 lo, hi, hi - lo, mi);
43057304 2538 if (!cmp)
74e34e1f 2539 return nth_packed_object_offset(p, mi);
1f688557
JH
2540 if (cmp > 0)
2541 hi = mi;
2542 else
2543 lo = mi+1;
2544 } while (lo < hi);
2545 return 0;
2546}
2547
4c080182 2548int is_pack_valid(struct packed_git *p)
d131b7af
SP
2549{
2550 /* An already open pack is known to be valid. */
2551 if (p->pack_fd != -1)
2552 return 1;
2553
2554 /* If the pack has one window completely covering the
2555 * file size, the pack is known to be valid even if
2556 * the descriptor is not currently open.
2557 */
2558 if (p->windows) {
2559 struct pack_window *w = p->windows;
2560
2561 if (!w->offset && w->len == p->pack_size)
2562 return 1;
2563 }
2564
2565 /* Force the pack to open to prove its valid. */
2566 return !open_packed_git(p);
2567}
2568
95099731
NTND
2569static int fill_pack_entry(const unsigned char *sha1,
2570 struct pack_entry *e,
2571 struct packed_git *p)
2572{
2573 off_t offset;
2574
2575 if (p->num_bad_objects) {
2576 unsigned i;
2577 for (i = 0; i < p->num_bad_objects; i++)
2578 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2579 return 0;
2580 }
2581
2582 offset = find_pack_entry_one(sha1, p);
2583 if (!offset)
2584 return 0;
2585
2586 /*
2587 * We are about to tell the caller where they can locate the
2588 * requested object. We better make sure the packfile is
2589 * still here and can be accessed before supplying that
2590 * answer, as it may have been deleted since the index was
2591 * loaded!
2592 */
319b678a 2593 if (!is_pack_valid(p))
95099731 2594 return 0;
95099731
NTND
2595 e->offset = offset;
2596 e->p = p;
2597 hashcpy(e->sha1, sha1);
2598 return 1;
2599}
2600
d40d535b
MH
2601/*
2602 * Iff a pack file contains the object named by sha1, return true and
2603 * store its location to e.
2604 */
4d6acb70 2605static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557
JH
2606{
2607 struct packed_git *p;
43057304 2608
1f688557 2609 prepare_packed_git();
f7c22cc6
NP
2610 if (!packed_git)
2611 return 0;
1f688557 2612
c01f51cc
NTND
2613 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2614 return 1;
8eca0b47 2615
c01f51cc 2616 for (p = packed_git; p; p = p->next) {
1b1005d1
MH
2617 if (p == last_found_pack)
2618 continue; /* we already checked this one */
f7c22cc6 2619
1b1005d1
MH
2620 if (fill_pack_entry(sha1, e, p)) {
2621 last_found_pack = p;
2622 return 1;
2623 }
c01f51cc 2624 }
1f688557
JH
2625 return 0;
2626}
2627
a6080a0a 2628struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
2629 struct packed_git *packs)
2630{
2631 struct packed_git *p;
bf592c50
DB
2632
2633 for (p = packs; p; p = p->next) {
43057304 2634 if (find_pack_entry_one(sha1, p))
bf592c50
DB
2635 return p;
2636 }
2637 return NULL;
21666f1a 2638
bf592c50
DB
2639}
2640
052fe5ea 2641static int sha1_loose_object_info(const unsigned char *sha1,
46f03448
KN
2642 struct object_info *oi,
2643 int flags)
65c2e0c3 2644{
46f03448
KN
2645 int status = 0;
2646 unsigned long mapsize;
65c2e0c3 2647 void *map;
ef49a7a0 2648 git_zstream stream;
d65a16f6 2649 char hdr[32];
46f03448 2650 struct strbuf hdrbuf = STRBUF_INIT;
65c2e0c3 2651
5d642e75
JK
2652 if (oi->delta_base_sha1)
2653 hashclr(oi->delta_base_sha1);
2654
052fe5ea
JK
2655 /*
2656 * If we don't care about type or size, then we don't
4ef8d1dd
JH
2657 * need to look inside the object at all. Note that we
2658 * do not optimize out the stat call, even if the
2659 * caller doesn't care about the disk-size, since our
2660 * return value implicitly indicates whether the
2661 * object even exists.
052fe5ea 2662 */
46f03448 2663 if (!oi->typep && !oi->typename && !oi->sizep) {
4ef8d1dd
JH
2664 struct stat st;
2665 if (stat_sha1_file(sha1, &st) < 0)
2666 return -1;
2667 if (oi->disk_sizep)
23c339c0 2668 *oi->disk_sizep = st.st_size;
052fe5ea
JK
2669 return 0;
2670 }
2671
bb6b8e4f 2672 map = map_sha1_file(sha1, &mapsize);
f0df4ed5 2673 if (!map)
dbea72a8 2674 return -1;
23c339c0
JK
2675 if (oi->disk_sizep)
2676 *oi->disk_sizep = mapsize;
46f03448
KN
2677 if ((flags & LOOKUP_UNKNOWN_OBJECT)) {
2678 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2679 status = error("unable to unpack %s header with --allow-unknown-type",
2680 sha1_to_hex(sha1));
2681 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
36e4d74a
JH
2682 status = error("unable to unpack %s header",
2683 sha1_to_hex(sha1));
46f03448
KN
2684 if (status < 0)
2685 ; /* Do nothing */
2686 else if (hdrbuf.len) {
2687 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2688 status = error("unable to parse %s header with --allow-unknown-type",
2689 sha1_to_hex(sha1));
2690 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
36e4d74a 2691 status = error("unable to parse %s header", sha1_to_hex(sha1));
39c68542 2692 git_inflate_end(&stream);
65c2e0c3 2693 munmap(map, mapsize);
46f03448 2694 if (status && oi->typep)
23c339c0 2695 *oi->typep = status;
46f03448 2696 strbuf_release(&hdrbuf);
052fe5ea 2697 return 0;
65c2e0c3
JH
2698}
2699
de7b5d62 2700int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
f0df4ed5 2701{
c4d9986f 2702 struct cached_object *co;
f0df4ed5 2703 struct pack_entry e;
5b086407 2704 int rtype;
46f03448 2705 enum object_type real_type;
1f7117ef 2706 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
f0df4ed5 2707
1f7117ef 2708 co = find_cached_object(real);
c4d9986f 2709 if (co) {
5b086407
JK
2710 if (oi->typep)
2711 *(oi->typep) = co->type;
9a490590
JH
2712 if (oi->sizep)
2713 *(oi->sizep) = co->size;
161f00e7
JK
2714 if (oi->disk_sizep)
2715 *(oi->disk_sizep) = 0;
5d642e75
JK
2716 if (oi->delta_base_sha1)
2717 hashclr(oi->delta_base_sha1);
46f03448
KN
2718 if (oi->typename)
2719 strbuf_addstr(oi->typename, typename(co->type));
9a490590 2720 oi->whence = OI_CACHED;
5b086407 2721 return 0;
c4d9986f
NTND
2722 }
2723
1f7117ef 2724 if (!find_pack_entry(real, &e)) {
ddd63e64 2725 /* Most likely it's a loose object. */
46f03448 2726 if (!sha1_loose_object_info(real, oi, flags)) {
9a490590 2727 oi->whence = OI_LOOSE;
5b086407 2728 return 0;
9a490590 2729 }
ddd63e64
SG
2730
2731 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 2732 reprepare_packed_git();
1f7117ef 2733 if (!find_pack_entry(real, &e))
052fe5ea 2734 return -1;
f0df4ed5 2735 }
3d77d877 2736
46f03448
KN
2737 /*
2738 * packed_object_info() does not follow the delta chain to
2739 * find out the real type, unless it is given oi->typep.
2740 */
2741 if (oi->typename && !oi->typep)
2742 oi->typep = &real_type;
2743
23c339c0 2744 rtype = packed_object_info(e.p, e.offset, oi);
412916ee 2745 if (rtype < 0) {
1f7117ef 2746 mark_bad_packed_object(e.p, real);
46f03448
KN
2747 if (oi->typep == &real_type)
2748 oi->typep = NULL;
1f7117ef 2749 return sha1_object_info_extended(real, oi, 0);
5266d369
JH
2750 } else if (in_delta_base_cache(e.p, e.offset)) {
2751 oi->whence = OI_DBCACHED;
9a490590
JH
2752 } else {
2753 oi->whence = OI_PACKED;
2754 oi->u.packed.offset = e.offset;
2755 oi->u.packed.pack = e.p;
2756 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2757 rtype == OBJ_OFS_DELTA);
3d77d877 2758 }
46f03448
KN
2759 if (oi->typename)
2760 strbuf_addstr(oi->typename, typename(*oi->typep));
2761 if (oi->typep == &real_type)
2762 oi->typep = NULL;
3d77d877 2763
5b086407 2764 return 0;
f0df4ed5
JS
2765}
2766
3fc0dca9 2767/* returns enum object_type or negative */
9a490590
JH
2768int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2769{
5b086407 2770 enum object_type type;
d099b717 2771 struct object_info oi = {NULL};
9a490590 2772
5b086407 2773 oi.typep = &type;
9a490590 2774 oi.sizep = sizep;
de7b5d62 2775 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
5b086407
JK
2776 return -1;
2777 return type;
9a490590
JH
2778}
2779
21666f1a
NP
2780static void *read_packed_sha1(const unsigned char *sha1,
2781 enum object_type *type, unsigned long *size)
1f688557
JH
2782{
2783 struct pack_entry e;
8eca0b47 2784 void *data;
1f688557 2785
cd673c1f 2786 if (!find_pack_entry(sha1, &e))
1f688557 2787 return NULL;
8eca0b47
NP
2788 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2789 if (!data) {
2790 /*
2791 * We're probably in deep shit, but let's try to fetch
2792 * the required object anyway from another pack or loose.
2793 * This should happen only in the presence of a corrupted
2794 * pack, and is better than failing outright.
2795 */
2796 error("failed to read object %s at offset %"PRIuMAX" from %s",
2797 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2798 mark_bad_packed_object(e.p, sha1);
ac939109 2799 data = read_object(sha1, type, size);
8eca0b47
NP
2800 }
2801 return data;
1f688557
JH
2802}
2803
21666f1a
NP
2804int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2805 unsigned char *sha1)
d66b37bb
JH
2806{
2807 struct cached_object *co;
2808
21666f1a 2809 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2810 if (has_sha1_file(sha1) || find_cached_object(sha1))
2811 return 0;
c7353967 2812 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
d66b37bb
JH
2813 co = &cached_objects[cached_object_nr++];
2814 co->size = len;
21666f1a 2815 co->type = type;
efa13f7b
JH
2816 co->buf = xmalloc(len);
2817 memcpy(co->buf, buf, len);
d66b37bb
JH
2818 hashcpy(co->sha1, sha1);
2819 return 0;
2820}
2821
c2c5b270
CC
2822static void *read_object(const unsigned char *sha1, enum object_type *type,
2823 unsigned long *size)
0fcfd160
LT
2824{
2825 unsigned long mapsize;
2826 void *map, *buf;
d66b37bb
JH
2827 struct cached_object *co;
2828
2829 co = find_cached_object(sha1);
2830 if (co) {
21666f1a 2831 *type = co->type;
d66b37bb 2832 *size = co->size;
182af834 2833 return xmemdupz(co->buf, co->size);
d66b37bb 2834 }
0fcfd160 2835
8276c007
PE
2836 buf = read_packed_sha1(sha1, type, size);
2837 if (buf)
2838 return buf;
bb6b8e4f 2839 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2840 if (map) {
7efbff75 2841 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2842 munmap(map, mapsize);
2843 return buf;
2844 }
637cdd9d 2845 reprepare_packed_git();
8276c007 2846 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2847}
2848
b6c4cecc
JH
2849/*
2850 * This function dies on corrupt objects; the callers who want to
2851 * deal with them should arrange to call read_object() and give error
2852 * messages themselves.
2853 */
5bf29b95
JH
2854void *read_sha1_file_extended(const unsigned char *sha1,
2855 enum object_type *type,
2856 unsigned long *size,
2857 unsigned flag)
ac939109 2858{
3ba7a065 2859 void *data;
b6c4cecc 2860 const struct packed_git *p;
bf93eea0 2861 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
b6c4cecc 2862
3ba7a065
JH
2863 errno = 0;
2864 data = read_object(repl, type, size);
4bbf5a26 2865 if (data)
b6c4cecc 2866 return data;
68095570 2867
25f3af3f 2868 if (errno && errno != ENOENT)
3ba7a065
JH
2869 die_errno("failed to read object %s", sha1_to_hex(sha1));
2870
68095570 2871 /* die if we replaced an object with one that does not exist */
b6c4cecc 2872 if (repl != sha1)
68095570
CC
2873 die("replacement %s not found for %s",
2874 sha1_to_hex(repl), sha1_to_hex(sha1));
2875
b6c4cecc 2876 if (has_loose_object(repl)) {
30d6c6ea
MH
2877 const char *path = sha1_file_name(sha1);
2878
b6c4cecc
JH
2879 die("loose object %s (stored in %s) is corrupt",
2880 sha1_to_hex(repl), path);
e8b15e61 2881 }
68095570 2882
b6c4cecc
JH
2883 if ((p = has_packed_and_bad(repl)) != NULL)
2884 die("packed object %s (stored in %s) is corrupt",
2885 sha1_to_hex(repl), p->pack_name);
f5552aee 2886
b6c4cecc 2887 return NULL;
ac939109
NP
2888}
2889
40469ee9 2890void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2891 const char *required_type_name,
40469ee9
JH
2892 unsigned long *size,
2893 unsigned char *actual_sha1_return)
f4913f91 2894{
21666f1a 2895 enum object_type type, required_type;
f4913f91
JH
2896 void *buffer;
2897 unsigned long isize;
40469ee9 2898 unsigned char actual_sha1[20];
f4913f91 2899
21666f1a 2900 required_type = type_from_string(required_type_name);
e702496e 2901 hashcpy(actual_sha1, sha1);
40469ee9
JH
2902 while (1) {
2903 int ref_length = -1;
2904 const char *ref_type = NULL;
f4913f91 2905
21666f1a 2906 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2907 if (!buffer)
2908 return NULL;
21666f1a 2909 if (type == required_type) {
40469ee9
JH
2910 *size = isize;
2911 if (actual_sha1_return)
e702496e 2912 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2913 return buffer;
2914 }
2915 /* Handle references */
21666f1a 2916 else if (type == OBJ_COMMIT)
40469ee9 2917 ref_type = "tree ";
21666f1a 2918 else if (type == OBJ_TAG)
40469ee9
JH
2919 ref_type = "object ";
2920 else {
2921 free(buffer);
2922 return NULL;
2923 }
2924 ref_length = strlen(ref_type);
f4913f91 2925
50974ec9
MK
2926 if (ref_length + 40 > isize ||
2927 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2928 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2929 free(buffer);
2930 return NULL;
2931 }
1cf58e72 2932 free(buffer);
40469ee9
JH
2933 /* Now we have the ID of the referred-to object in
2934 * actual_sha1. Check again. */
f4913f91 2935 }
f4913f91
JH
2936}
2937
ce9fbf16 2938static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2939 const char *type, unsigned char *sha1,
d65a16f6 2940 char *hdr, int *hdrlen)
d410c0f5 2941{
9126f009 2942 git_SHA_CTX c;
d410c0f5
JH
2943
2944 /* Generate the header */
ef1286d3 2945 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
d410c0f5
JH
2946
2947 /* Sha1.. */
9126f009
NP
2948 git_SHA1_Init(&c);
2949 git_SHA1_Update(&c, hdr, *hdrlen);
2950 git_SHA1_Update(&c, buf, len);
2951 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2952}
2953
230f1322 2954/*
5a688fe4 2955 * Move the just written object into its final resting place.
230f1322 2956 */
cb5add58 2957int finalize_object_file(const char *tmpfile, const char *filename)
230f1322 2958{
e32c0a9c 2959 int ret = 0;
5a688fe4 2960
348df166 2961 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
2962 goto try_rename;
2963 else if (link(tmpfile, filename))
e32c0a9c 2964 ret = errno;
7ebb6fca
LT
2965
2966 /*
2967 * Coda hack - coda doesn't like cross-directory links,
2968 * so we fall back to a rename, which will mean that it
2969 * won't be able to check collisions, but that's not a
2970 * big deal.
2971 *
2972 * The same holds for FAT formatted media.
2973 *
3be1f18e 2974 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
2975 * left to unlink.
2976 */
2977 if (ret && ret != EEXIST) {
be66a6c4 2978 try_rename:
7ebb6fca 2979 if (!rename(tmpfile, filename))
3be1f18e 2980 goto out;
9e48b389 2981 ret = errno;
230f1322 2982 }
691f1a28 2983 unlink_or_warn(tmpfile);
230f1322
LT
2984 if (ret) {
2985 if (ret != EEXIST) {
7616c6ca 2986 return error_errno("unable to write sha1 filename %s", filename);
230f1322
LT
2987 }
2988 /* FIXME!!! Collision check here ? */
2989 }
2990
3be1f18e 2991out:
5256b006 2992 if (adjust_shared_perm(filename))
5a688fe4 2993 return error("unable to set permission to '%s'", filename);
230f1322
LT
2994 return 0;
2995}
2996
4d548150
LT
2997static int write_buffer(int fd, const void *buf, size_t len)
2998{
d34cf19b 2999 if (write_in_full(fd, buf, len) < 0)
7616c6ca 3000 return error_errno("file write error");
4d548150
LT
3001 return 0;
3002}
3003
ce9fbf16 3004int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
3005 unsigned char *sha1)
3006{
d65a16f6 3007 char hdr[32];
ef1286d3 3008 int hdrlen = sizeof(hdr);
abdc3fc8
RS
3009 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3010 return 0;
3011}
3012
e9039dd3
LT
3013/* Finalize a file on disk, and close it. */
3014static void close_sha1_file(int fd)
3015{
aafe9fba
LT
3016 if (fsync_object_files)
3017 fsync_or_die(fd, "sha1 file");
e9039dd3 3018 if (close(fd) != 0)
d824cbba 3019 die_errno("error when closing sha1 file");
e9039dd3
LT
3020}
3021
5723fe7e
LT
3022/* Size of directory component, including the ending '/' */
3023static inline int directory_size(const char *filename)
3024{
3025 const char *s = strrchr(filename, '/');
3026 if (!s)
3027 return 0;
3028 return s - filename + 1;
3029}
3030
3031/*
3032 * This creates a temporary file in the same directory as the final
3033 * 'filename'
3034 *
3035 * We want to avoid cross-directory filename renames, because those
3036 * can have problems on various filesystems (FAT, NFS, Coda).
3037 */
d4b3d11a 3038static int create_tmpfile(struct strbuf *tmp, const char *filename)
5723fe7e
LT
3039{
3040 int fd, dirlen = directory_size(filename);
3041
d4b3d11a
JK
3042 strbuf_reset(tmp);
3043 strbuf_add(tmp, filename, dirlen);
3044 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
3045 fd = git_mkstemp_mode(tmp->buf, 0444);
cbacbf4e 3046 if (fd < 0 && dirlen && errno == ENOENT) {
d4b3d11a
JK
3047 /*
3048 * Make sure the directory exists; note that the contents
3049 * of the buffer are undefined after mkstemp returns an
3050 * error, so we have to rewrite the whole buffer from
3051 * scratch.
3052 */
3053 strbuf_reset(tmp);
3054 strbuf_add(tmp, filename, dirlen - 1);
3055 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
b2476a60 3056 return -1;
d4b3d11a 3057 if (adjust_shared_perm(tmp->buf))
5723fe7e
LT
3058 return -1;
3059
3060 /* Try again */
d4b3d11a
JK
3061 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
3062 fd = git_mkstemp_mode(tmp->buf, 0444);
5723fe7e
LT
3063 }
3064 return fd;
3065}
3066
bbac7311 3067static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 3068 const void *buf, unsigned long len, time_t mtime)
0fcfd160 3069{
915308b1 3070 int fd, ret;
9892beba 3071 unsigned char compressed[4096];
ef49a7a0 3072 git_zstream stream;
748af44c
NP
3073 git_SHA_CTX c;
3074 unsigned char parano_sha1[20];
d4b3d11a 3075 static struct strbuf tmp_file = STRBUF_INIT;
30d6c6ea 3076 const char *filename = sha1_file_name(sha1);
a44c9a5e 3077
d4b3d11a 3078 fd = create_tmpfile(&tmp_file, filename);
aac17941 3079 if (fd < 0) {
35243577 3080 if (errno == EACCES)
82247e9b 3081 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
916d081b 3082 else
7616c6ca 3083 return error_errno("unable to create temporary file");
aac17941
LT
3084 }
3085
0fcfd160 3086 /* Set it up */
55bb5c91 3087 git_deflate_init(&stream, zlib_compression_level);
0fcfd160 3088 stream.next_out = compressed;
9892beba 3089 stream.avail_out = sizeof(compressed);
748af44c 3090 git_SHA1_Init(&c);
a44c9a5e
LT
3091
3092 /* First header.. */
d65a16f6 3093 stream.next_in = (unsigned char *)hdr;
a44c9a5e 3094 stream.avail_in = hdrlen;
55bb5c91
JH
3095 while (git_deflate(&stream, 0) == Z_OK)
3096 ; /* nothing */
748af44c 3097 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
3098
3099 /* Then the data itself.. */
c00e657d 3100 stream.next_in = (void *)buf;
a44c9a5e 3101 stream.avail_in = len;
9892beba 3102 do {
748af44c 3103 unsigned char *in0 = stream.next_in;
55bb5c91 3104 ret = git_deflate(&stream, Z_FINISH);
748af44c 3105 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
3106 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3107 die("unable to write sha1 file");
3108 stream.next_out = compressed;
3109 stream.avail_out = sizeof(compressed);
3110 } while (ret == Z_OK);
3111
ac54c277
LT
3112 if (ret != Z_STREAM_END)
3113 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
55bb5c91 3114 ret = git_deflate_end_gently(&stream);
ac54c277
LT
3115 if (ret != Z_OK)
3116 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
3117 git_SHA1_Final(parano_sha1, &c);
3118 if (hashcmp(sha1, parano_sha1) != 0)
3119 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 3120
e9039dd3 3121 close_sha1_file(fd);
0fcfd160 3122
bbac7311
NP
3123 if (mtime) {
3124 struct utimbuf utb;
3125 utb.actime = mtime;
3126 utb.modtime = mtime;
d4b3d11a 3127 if (utime(tmp_file.buf, &utb) < 0)
7616c6ca 3128 warning_errno("failed utime() on %s", tmp_file.buf);
bbac7311
NP
3129 }
3130
d4b3d11a 3131 return finalize_object_file(tmp_file.buf, filename);
0fcfd160 3132}
8237b185 3133
33d4221c
JK
3134static int freshen_loose_object(const unsigned char *sha1)
3135{
3136 return check_and_freshen(sha1, 1);
3137}
3138
3139static int freshen_packed_object(const unsigned char *sha1)
3140{
3141 struct pack_entry e;
ee1c6c34
JK
3142 if (!find_pack_entry(sha1, &e))
3143 return 0;
3144 if (e.p->freshened)
3145 return 1;
3146 if (!freshen_file(e.p->pack_name))
3147 return 0;
3148 e.p->freshened = 1;
3149 return 1;
33d4221c
JK
3150}
3151
1427a7ff 3152int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
bbac7311 3153{
bbac7311 3154 char hdr[32];
ef1286d3 3155 int hdrlen = sizeof(hdr);
bbac7311
NP
3156
3157 /* Normally if we have it in the pack then we do not bother writing
3158 * it out into .git/objects/??/?{38} file.
3159 */
3160 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
b5f52f37 3161 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
bbac7311
NP
3162 return 0;
3163 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3164}
3165
0c3db67c
ES
3166int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3167 unsigned char *sha1, unsigned flags)
3168{
3169 char *header;
3170 int hdrlen, status = 0;
3171
3172 /* type string, SP, %lu of the length plus NUL must fit this */
ef1286d3
JK
3173 hdrlen = strlen(type) + 32;
3174 header = xmalloc(hdrlen);
0c3db67c
ES
3175 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3176
3177 if (!(flags & HASH_WRITE_OBJECT))
3178 goto cleanup;
051086b9 3179 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
0c3db67c
ES
3180 goto cleanup;
3181 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3182
3183cleanup:
3184 free(header);
3185 return status;
3186}
3187
bbac7311
NP
3188int force_object_loose(const unsigned char *sha1, time_t mtime)
3189{
bbac7311
NP
3190 void *buf;
3191 unsigned long len;
3192 enum object_type type;
3193 char hdr[32];
3194 int hdrlen;
1fb23e65 3195 int ret;
bbac7311 3196
c529d75a 3197 if (has_loose_object(sha1))
bbac7311
NP
3198 return 0;
3199 buf = read_packed_sha1(sha1, &type, &len);
3200 if (!buf)
3201 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
ef1286d3 3202 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1fb23e65
BS
3203 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3204 free(buf);
3205
3206 return ret;
bbac7311
NP
3207}
3208
bf592c50
DB
3209int has_pack_index(const unsigned char *sha1)
3210{
3211 struct stat st;
3212 if (stat(sha1_pack_index_name(sha1), &st))
3213 return 0;
3214 return 1;
bf592c50
DB
3215}
3216
cd673c1f
JH
3217int has_sha1_pack(const unsigned char *sha1)
3218{
3219 struct pack_entry e;
3220 return find_pack_entry(sha1, &e);
3221}
3222
0eeb077b 3223int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
8237b185 3224{
1f688557
JH
3225 struct pack_entry e;
3226
cd673c1f 3227 if (find_pack_entry(sha1, &e))
1f688557 3228 return 1;
45e8a748
JK
3229 if (has_loose_object(sha1))
3230 return 1;
0eeb077b
JK
3231 if (flags & HAS_SHA1_QUICK)
3232 return 0;
45e8a748
JK
3233 reprepare_packed_git();
3234 return find_pack_entry(sha1, &e);
b419aa25 3235}
3236
3237int has_object_file(const struct object_id *oid)
3238{
3239 return has_sha1_file(oid->hash);
8237b185 3240}
74400e71 3241
c879daa2
NTND
3242static void check_tree(const void *buf, size_t size)
3243{
3244 struct tree_desc desc;
3245 struct name_entry entry;
3246
3247 init_tree_desc(&desc, buf, size);
3248 while (tree_entry(&desc, &entry))
3249 /* do nothing
3250 * tree_entry() will die() on malformed entries */
3251 ;
3252}
3253
3254static void check_commit(const void *buf, size_t size)
3255{
3256 struct commit c;
3257 memset(&c, 0, sizeof(c));
3258 if (parse_commit_buffer(&c, buf, size))
3259 die("corrupt commit");
3260}
3261
3262static void check_tag(const void *buf, size_t size)
3263{
3264 struct tag t;
3265 memset(&t, 0, sizeof(t));
3266 if (parse_tag_buffer(&t, buf, size))
3267 die("corrupt tag");
3268}
3269
43df4f86 3270static int index_mem(unsigned char *sha1, void *buf, size_t size,
c4ce46fc
JH
3271 enum object_type type,
3272 const char *path, unsigned flags)
e7332f96 3273{
6c510bee 3274 int ret, re_allocated = 0;
c4ce46fc 3275 int write_object = flags & HASH_WRITE_OBJECT;
74400e71 3276
7672db20 3277 if (!type)
edaec3fb 3278 type = OBJ_BLOB;
6c510bee
LT
3279
3280 /*
3281 * Convert blobs to git internal format
3282 */
43df4f86 3283 if ((type == OBJ_BLOB) && path) {
f285a2d7 3284 struct strbuf nbuf = STRBUF_INIT;
21e5ad50 3285 if (convert_to_git(path, buf, size, &nbuf,
5e12e78e 3286 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
b315c5c0 3287 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
3288 re_allocated = 1;
3289 }
3290 }
c4ce46fc 3291 if (flags & HASH_FORMAT_CHECK) {
c879daa2
NTND
3292 if (type == OBJ_TREE)
3293 check_tree(buf, size);
3294 if (type == OBJ_COMMIT)
3295 check_commit(buf, size);
3296 if (type == OBJ_TAG)
3297 check_tag(buf, size);
3298 }
6c510bee 3299
7672db20 3300 if (write_object)
edaec3fb 3301 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 3302 else
edaec3fb 3303 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 3304 if (re_allocated)
6c510bee 3305 free(buf);
43df4f86
DP
3306 return ret;
3307}
3308
9035d75a
SP
3309static int index_stream_convert_blob(unsigned char *sha1, int fd,
3310 const char *path, unsigned flags)
3311{
3312 int ret;
3313 const int write_object = flags & HASH_WRITE_OBJECT;
3314 struct strbuf sbuf = STRBUF_INIT;
3315
3316 assert(path);
3317 assert(would_convert_to_git_filter_fd(path));
3318
3319 convert_to_git_filter_fd(path, fd, &sbuf,
3320 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3321
3322 if (write_object)
3323 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3324 sha1);
3325 else
3326 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3327 sha1);
3328 strbuf_release(&sbuf);
3329 return ret;
3330}
3331
7b41e1e1
JH
3332static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3333 const char *path, unsigned flags)
3334{
3335 struct strbuf sbuf = STRBUF_INIT;
3336 int ret;
3337
3338 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3339 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3340 else
3341 ret = -1;
3342 strbuf_release(&sbuf);
3343 return ret;
3344}
3345
ea68b0ce
DP
3346#define SMALL_FILE_SIZE (32*1024)
3347
7b41e1e1
JH
3348static int index_core(unsigned char *sha1, int fd, size_t size,
3349 enum object_type type, const char *path,
3350 unsigned flags)
43df4f86
DP
3351{
3352 int ret;
43df4f86 3353
7b41e1e1 3354 if (!size) {
f6a1e1e2 3355 ret = index_mem(sha1, "", size, type, path, flags);
ea68b0ce
DP
3356 } else if (size <= SMALL_FILE_SIZE) {
3357 char *buf = xmalloc(size);
3358 if (size == read_in_full(fd, buf, size))
c4ce46fc 3359 ret = index_mem(sha1, buf, size, type, path, flags);
ea68b0ce 3360 else
7616c6ca 3361 ret = error_errno("short read");
ea68b0ce 3362 free(buf);
08bda208 3363 } else {
43df4f86 3364 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
c4ce46fc 3365 ret = index_mem(sha1, buf, size, type, path, flags);
aac17941 3366 munmap(buf, size);
08bda208 3367 }
7b41e1e1
JH
3368 return ret;
3369}
3370
4dd1fbc7 3371/*
568508e7
JH
3372 * This creates one packfile per large blob unless bulk-checkin
3373 * machinery is "plugged".
4dd1fbc7
JH
3374 *
3375 * This also bypasses the usual "convert-to-git" dance, and that is on
3376 * purpose. We could write a streaming version of the converting
3377 * functions and insert that before feeding the data to fast-import
4f22b101
JK
3378 * (or equivalent in-core API described above). However, that is
3379 * somewhat complicated, as we do not know the size of the filter
3380 * result, which we need to know beforehand when writing a git object.
3381 * Since the primary motivation for trying to stream from the working
3382 * tree file and to avoid mmaping it in core is to deal with large
3383 * binary blobs, they generally do not want to get any conversion, and
3384 * callers should avoid this code path when filters are requested.
4dd1fbc7
JH
3385 */
3386static int index_stream(unsigned char *sha1, int fd, size_t size,
3387 enum object_type type, const char *path,
3388 unsigned flags)
3389{
568508e7 3390 return index_bulk_checkin(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
3391}
3392
7b41e1e1
JH
3393int index_fd(unsigned char *sha1, int fd, struct stat *st,
3394 enum object_type type, const char *path, unsigned flags)
3395{
3396 int ret;
7b41e1e1 3397
9079ab7c
SP
3398 /*
3399 * Call xsize_t() only when needed to avoid potentially unnecessary
3400 * die() for large files.
3401 */
9035d75a
SP
3402 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3403 ret = index_stream_convert_blob(sha1, fd, path, flags);
3404 else if (!S_ISREG(st->st_mode))
7b41e1e1 3405 ret = index_pipe(sha1, fd, type, path, flags);
9079ab7c 3406 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
7ce7c760 3407 (path && would_convert_to_git(path)))
9079ab7c
SP
3408 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3409 flags);
4dd1fbc7 3410 else
9079ab7c
SP
3411 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3412 flags);
43df4f86 3413 close(fd);
aac17941 3414 return ret;
74400e71 3415}
ec1fcc16 3416
c4ce46fc 3417int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
ec1fcc16
JH
3418{
3419 int fd;
b760d3aa 3420 struct strbuf sb = STRBUF_INIT;
ec1fcc16
JH
3421
3422 switch (st->st_mode & S_IFMT) {
3423 case S_IFREG:
3424 fd = open(path, O_RDONLY);
3425 if (fd < 0)
7616c6ca 3426 return error_errno("open(\"%s\")", path);
c4ce46fc 3427 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
ec1fcc16
JH
3428 return error("%s: failed to insert into database",
3429 path);
3430 break;
3431 case S_IFLNK:
7616c6ca
NTND
3432 if (strbuf_readlink(&sb, path, st->st_size))
3433 return error_errno("readlink(\"%s\")", path);
c4ce46fc 3434 if (!(flags & HASH_WRITE_OBJECT))
b760d3aa
LT
3435 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3436 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
ec1fcc16
JH
3437 return error("%s: failed to insert into database",
3438 path);
b760d3aa 3439 strbuf_release(&sb);
ec1fcc16 3440 break;
f35a6d3b
LT
3441 case S_IFDIR:
3442 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
3443 default:
3444 return error("%s: unsupported file type", path);
3445 }
3446 return 0;
3447}
a69e5429
JH
3448
3449int read_pack_header(int fd, struct pack_header *header)
3450{
c697ad14
HO
3451 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3452 /* "eof before pack header was fully read" */
3453 return PH_ERROR_EOF;
3454
a69e5429
JH
3455 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3456 /* "protocol error (pack signature mismatch detected)" */
3457 return PH_ERROR_PACK_SIGNATURE;
3458 if (!pack_version_ok(header->hdr_version))
3459 /* "protocol error (pack version unsupported)" */
3460 return PH_ERROR_PROTOCOL;
3461 return 0;
3462}
40d52ff7
JK
3463
3464void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3465{
3466 enum object_type type = sha1_object_info(sha1, NULL);
3467 if (type < 0)
3468 die("%s is not a valid object", sha1_to_hex(sha1));
3469 if (type != expect)
3470 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3471 typename(expect));
3472}
27e1e22d
JK
3473
3474static int for_each_file_in_obj_subdir(int subdir_nr,
3475 struct strbuf *path,
3476 each_loose_object_fn obj_cb,
3477 each_loose_cruft_fn cruft_cb,
3478 each_loose_subdir_fn subdir_cb,
3479 void *data)
3480{
3481 size_t baselen = path->len;
3482 DIR *dir = opendir(path->buf);
3483 struct dirent *de;
3484 int r = 0;
3485
3486 if (!dir) {
3487 if (errno == ENOENT)
3488 return 0;
7616c6ca 3489 return error_errno("unable to open %s", path->buf);
27e1e22d
JK
3490 }
3491
3492 while ((de = readdir(dir))) {
3493 if (is_dot_or_dotdot(de->d_name))
3494 continue;
3495
3496 strbuf_setlen(path, baselen);
3497 strbuf_addf(path, "/%s", de->d_name);
3498
3499 if (strlen(de->d_name) == 38) {
3500 char hex[41];
3501 unsigned char sha1[20];
3502
3503 snprintf(hex, sizeof(hex), "%02x%s",
3504 subdir_nr, de->d_name);
3505 if (!get_sha1_hex(hex, sha1)) {
3506 if (obj_cb) {
3507 r = obj_cb(sha1, path->buf, data);
3508 if (r)
3509 break;
3510 }
3511 continue;
3512 }
3513 }
3514
3515 if (cruft_cb) {
3516 r = cruft_cb(de->d_name, path->buf, data);
3517 if (r)
3518 break;
3519 }
3520 }
094c7e63 3521 closedir(dir);
27e1e22d 3522
094c7e63 3523 strbuf_setlen(path, baselen);
27e1e22d
JK
3524 if (!r && subdir_cb)
3525 r = subdir_cb(subdir_nr, path->buf, data);
3526
27e1e22d
JK
3527 return r;
3528}
3529
e6f875e0