t5700: modernize style
[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];
88cd621d 39
c597ba80
NTND
40/*
41 * This is meant to hold a *small* number of objects that you would
42 * want read_sha1_file() to be able to return, but yet you do not want
43 * to write them into the object store (e.g. a browse-only
44 * application).
45 */
46static struct cached_object {
47 unsigned char sha1[20];
48 enum object_type type;
49 void *buf;
50 unsigned long size;
51} *cached_objects;
52static int cached_object_nr, cached_object_alloc;
53
54static struct cached_object empty_tree = {
dab0d410 55 EMPTY_TREE_SHA1_BIN_LITERAL,
c597ba80
NTND
56 OBJ_TREE,
57 "",
58 0
59};
60
1b1005d1
MH
61/*
62 * A pointer to the last packed_git in which an object was found.
63 * When an object is sought, we look in this packfile first, because
64 * objects that are looked up at similar times are often in the same
65 * packfile as one another.
66 */
c01f51cc
NTND
67static struct packed_git *last_found_pack;
68
c597ba80
NTND
69static struct cached_object *find_cached_object(const unsigned char *sha1)
70{
71 int i;
72 struct cached_object *co = cached_objects;
73
74 for (i = 0; i < cached_object_nr; i++, co++) {
75 if (!hashcmp(co->sha1, sha1))
76 return co;
77 }
78 if (!hashcmp(sha1, empty_tree.sha1))
79 return &empty_tree;
80 return NULL;
81}
82
90a6464b
JH
83int mkdir_in_gitdir(const char *path)
84{
85 if (mkdir(path, 0777)) {
86 int saved_errno = errno;
87 struct stat st;
88 struct strbuf sb = STRBUF_INIT;
89
90 if (errno != EEXIST)
91 return -1;
92 /*
93 * Are we looking at a path in a symlinked worktree
94 * whose original repository does not yet have it?
95 * e.g. .git/rr-cache pointing at its original
96 * repository in which the user hasn't performed any
97 * conflict resolution yet?
98 */
99 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
100 strbuf_readlink(&sb, path, st.st_size) ||
101 !is_absolute_path(sb.buf) ||
102 mkdir(sb.buf, 0777)) {
103 strbuf_release(&sb);
104 errno = saved_errno;
105 return -1;
106 }
107 strbuf_release(&sb);
108 }
109 return adjust_shared_perm(path);
110}
111
0be0521b 112enum scld_error safe_create_leading_directories(char *path)
b2cb9425 113{
26c8ae2a 114 char *next_component = path + offset_1st_component(path);
0be0521b 115 enum scld_error ret = SCLD_OK;
67d42212 116
0be0521b 117 while (ret == SCLD_OK && next_component) {
f0502332 118 struct stat st;
0f527403 119 char *slash = next_component, slash_character;
f0502332 120
0f527403
MH
121 while (*slash && !is_dir_sep(*slash))
122 slash++;
123
124 if (!*slash)
b2cb9425 125 break;
bf10cf70 126
26c8ae2a 127 next_component = slash + 1;
0f527403 128 while (is_dir_sep(*next_component))
bf10cf70 129 next_component++;
26c8ae2a 130 if (!*next_component)
5f0bdf50 131 break;
831651fd 132
0f527403 133 slash_character = *slash;
831651fd 134 *slash = '\0';
67d42212
JR
135 if (!stat(path, &st)) {
136 /* path exists */
9e6f885d 137 if (!S_ISDIR(st.st_mode))
0be0521b 138 ret = SCLD_EXISTS;
53a39721 139 } else if (mkdir(path, 0777)) {
928734d9 140 if (errno == EEXIST &&
9e6f885d 141 !stat(path, &st) && S_ISDIR(st.st_mode))
928734d9 142 ; /* somebody created it since we checked */
18d37e86
MH
143 else if (errno == ENOENT)
144 /*
145 * Either mkdir() failed because
146 * somebody just pruned the containing
147 * directory, or stat() failed because
148 * the file that was in our way was
149 * just removed. Either way, inform
150 * the caller that it might be worth
151 * trying again:
152 */
153 ret = SCLD_VANISHED;
9e6f885d 154 else
0be0521b 155 ret = SCLD_FAILED;
53a39721 156 } else if (adjust_shared_perm(path)) {
0be0521b 157 ret = SCLD_PERMS;
457f06d6 158 }
0f527403 159 *slash = slash_character;
b2cb9425 160 }
9e6f885d 161 return ret;
b2cb9425 162}
723c31fe 163
0be0521b 164enum scld_error safe_create_leading_directories_const(const char *path)
8e21d63b
JK
165{
166 /* path points to cache entries, so xstrdup before messing with it */
167 char *buf = xstrdup(path);
0be0521b 168 enum scld_error result = safe_create_leading_directories(buf);
8e21d63b
JK
169 free(buf);
170 return result;
171}
172
ace1534d
JH
173static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
174{
175 int i;
176 for (i = 0; i < 20; i++) {
177 static char hex[] = "0123456789abcdef";
178 unsigned int val = sha1[i];
179 char *pos = pathbuf + i*2 + (i > 0);
180 *pos++ = hex[val >> 4];
181 *pos = hex[val & 0xf];
182 }
183}
184
30d6c6ea 185const char *sha1_file_name(const unsigned char *sha1)
0fcfd160 186{
560fb6a1
JK
187 static char buf[PATH_MAX];
188 const char *objdir;
189 int len;
0fcfd160 190
560fb6a1
JK
191 objdir = get_object_directory();
192 len = strlen(objdir);
193
194 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
195 if (len + 43 > PATH_MAX)
196 die("insanely long object directory %s", objdir);
197 memcpy(buf, objdir, len);
198 buf[len] = '/';
199 buf[len+3] = '/';
200 buf[len+42] = '\0';
201 fill_sha1_path(buf + len + 1, sha1);
202 return buf;
0fcfd160
LT
203}
204
d40d535b
MH
205/*
206 * Return the name of the pack or index file with the specified sha1
207 * in its filename. *base and *name are scratch space that must be
208 * provided by the caller. which should be "pack" or "idx".
209 */
633f43e1 210static char *sha1_get_pack_name(const unsigned char *sha1,
6eec46bd 211 char **name, char **base, const char *which)
bf592c50
DB
212{
213 static const char hex[] = "0123456789abcdef";
633f43e1 214 char *buf;
bf592c50
DB
215 int i;
216
633f43e1 217 if (!*base) {
bf592c50
DB
218 const char *sha1_file_directory = get_object_directory();
219 int len = strlen(sha1_file_directory);
633f43e1 220 *base = xmalloc(len + 60);
6eec46bd
JH
221 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
222 sha1_file_directory, which);
633f43e1 223 *name = *base + len + 11;
bf592c50
DB
224 }
225
633f43e1 226 buf = *name;
bf592c50
DB
227
228 for (i = 0; i < 20; i++) {
229 unsigned int val = *sha1++;
230 *buf++ = hex[val >> 4];
231 *buf++ = hex[val & 0xf];
232 }
a6080a0a 233
633f43e1 234 return *base;
bf592c50
DB
235}
236
633f43e1 237char *sha1_pack_name(const unsigned char *sha1)
bf592c50 238{
633f43e1 239 static char *name, *base;
bf592c50 240
6eec46bd 241 return sha1_get_pack_name(sha1, &name, &base, "pack");
633f43e1 242}
bf592c50 243
633f43e1
HO
244char *sha1_pack_index_name(const unsigned char *sha1)
245{
246 static char *name, *base;
a6080a0a 247
6eec46bd 248 return sha1_get_pack_name(sha1, &name, &base, "idx");
bf592c50
DB
249}
250
d5a63b99
JH
251struct alternate_object_database *alt_odb_list;
252static struct alternate_object_database **alt_odb_tail;
ace1534d 253
ddd5d056
JH
254/*
255 * Prepare alternate object database registry.
d5a63b99
JH
256 *
257 * The variable alt_odb_list points at the list of struct
258 * alternate_object_database. The elements on this list come from
259 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
260 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
261 * whose contents is similar to that environment variable but can be
262 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
263 * contains "/the/directory/corresponding/to/.git/objects/...", while
264 * its name points just after the slash at the end of ".git/objects/"
265 * in the example above, and has enough space to hold 40-byte hex
266 * SHA1, an extra slash for the first level indirection, and the
267 * terminating NUL.
ddd5d056 268 */
539e7506
EK
269static int link_alt_odb_entry(const char *entry, const char *relative_base,
270 int depth, const char *normalized_objdir)
ace1534d 271{
c2f493a4
MW
272 struct alternate_object_database *ent;
273 struct alternate_object_database *alt;
5bdf0a84
HW
274 int pfxlen, entlen;
275 struct strbuf pathbuf = STRBUF_INIT;
d5a63b99 276
85dadc38 277 if (!is_absolute_path(entry) && relative_base) {
5bdf0a84
HW
278 strbuf_addstr(&pathbuf, real_path(relative_base));
279 strbuf_addch(&pathbuf, '/');
c2f493a4 280 }
6eac50d8 281 strbuf_addstr(&pathbuf, entry);
c2f493a4 282
5bdf0a84
HW
283 normalize_path_copy(pathbuf.buf, pathbuf.buf);
284
285 pfxlen = strlen(pathbuf.buf);
286
287 /*
288 * The trailing slash after the directory name is given by
289 * this function at the end. Remove duplicates.
290 */
291 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
292 pfxlen -= 1;
293
294 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
295 ent = xmalloc(sizeof(*ent) + entlen);
296 memcpy(ent->base, pathbuf.buf, pfxlen);
297 strbuf_release(&pathbuf);
c2f493a4
MW
298
299 ent->name = ent->base + pfxlen + 1;
300 ent->base[pfxlen + 3] = '/';
301 ent->base[pfxlen] = ent->base[entlen-1] = 0;
302
303 /* Detect cases where alternate disappeared */
90b4a71c 304 if (!is_directory(ent->base)) {
c2f493a4
MW
305 error("object directory %s does not exist; "
306 "check .git/objects/info/alternates.",
307 ent->base);
308 free(ent);
309 return -1;
310 }
311
312 /* Prevent the common mistake of listing the same
313 * thing twice, or object directory itself.
314 */
315 for (alt = alt_odb_list; alt; alt = alt->next) {
80b47854
RS
316 if (pfxlen == alt->name - alt->base - 1 &&
317 !memcmp(ent->base, alt->base, pfxlen)) {
c2f493a4
MW
318 free(ent);
319 return -1;
320 }
321 }
539e7506 322 if (!strcmp_icase(ent->base, normalized_objdir)) {
c2f493a4
MW
323 free(ent);
324 return -1;
325 }
326
327 /* add the alternate entry */
328 *alt_odb_tail = ent;
329 alt_odb_tail = &(ent->next);
330 ent->next = NULL;
331
332 /* recursively add alternates */
333 read_info_alternates(ent->base, depth + 1);
334
335 ent->base[pfxlen] = '/';
336
337 return 0;
338}
339
c5950164 340static void link_alt_odb_entries(const char *alt, int len, int sep,
c2f493a4
MW
341 const char *relative_base, int depth)
342{
6eac50d8
MH
343 struct string_list entries = STRING_LIST_INIT_NODUP;
344 char *alt_copy;
345 int i;
539e7506 346 struct strbuf objdirbuf = STRBUF_INIT;
c2f493a4
MW
347
348 if (depth > 5) {
349 error("%s: ignoring alternate object stores, nesting too deep.",
350 relative_base);
351 return;
352 }
353
f655651e 354 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
539e7506
EK
355 normalize_path_copy(objdirbuf.buf, objdirbuf.buf);
356
c5950164 357 alt_copy = xmemdupz(alt, len);
6eac50d8
MH
358 string_list_split_in_place(&entries, alt_copy, sep, -1);
359 for (i = 0; i < entries.nr; i++) {
360 const char *entry = entries.items[i].string;
361 if (entry[0] == '\0' || entry[0] == '#')
9577e7e3 362 continue;
6eac50d8
MH
363 if (!is_absolute_path(entry) && depth) {
364 error("%s: ignoring relative alternate object store %s",
365 relative_base, entry);
366 } else {
539e7506 367 link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
9577e7e3 368 }
9577e7e3 369 }
6eac50d8
MH
370 string_list_clear(&entries, 0);
371 free(alt_copy);
539e7506 372 strbuf_release(&objdirbuf);
d5a63b99
JH
373}
374
5e73633d 375void read_info_alternates(const char * relative_base, int depth)
d5a63b99 376{
9577e7e3 377 char *map;
dc49cd76 378 size_t mapsz;
d5a63b99 379 struct stat st;
9cb18f56
JM
380 const char alt_file_name[] = "info/alternates";
381 /* Given that relative_base is no longer than PATH_MAX,
382 ensure that "path" has enough space to append "/", the
383 file name, "info/alternates", and a trailing NUL. */
384 char path[PATH_MAX + 1 + sizeof alt_file_name];
c2f493a4 385 int fd;
d5a63b99 386
9cb18f56 387 sprintf(path, "%s/%s", relative_base, alt_file_name);
f4e51683 388 fd = git_open_noatime(path);
d5a63b99
JH
389 if (fd < 0)
390 return;
391 if (fstat(fd, &st) || (st.st_size == 0)) {
392 close(fd);
9a217f2a 393 return;
ace1534d 394 }
dc49cd76
SP
395 mapsz = xsize_t(st.st_size);
396 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 397 close(fd);
d5a63b99 398
c5950164 399 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
c2f493a4 400
dc49cd76 401 munmap(map, mapsz);
ace1534d
JH
402}
403
bef70b22
DB
404void add_to_alternates_file(const char *reference)
405{
406 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
acd3b9ec 407 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
dcf69262 408 const char *alt = mkpath("%s\n", reference);
bef70b22
DB
409 write_or_die(fd, alt, strlen(alt));
410 if (commit_lock_file(lock))
411 die("could not close alternates file");
412 if (alt_odb_tail)
c5950164 413 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
bef70b22
DB
414}
415
fe1b2268 416int foreach_alt_odb(alt_odb_fn fn, void *cb)
d79796bc
JH
417{
418 struct alternate_object_database *ent;
fe1b2268 419 int r = 0;
d79796bc
JH
420
421 prepare_alt_odb();
fe1b2268
JK
422 for (ent = alt_odb_list; ent; ent = ent->next) {
423 r = fn(ent, cb);
424 if (r)
425 break;
426 }
427 return r;
d79796bc
JH
428}
429
c2f493a4
MW
430void prepare_alt_odb(void)
431{
554fe20d 432 const char *alt;
c2f493a4 433
7dc24aa5
SP
434 if (alt_odb_tail)
435 return;
436
c2f493a4
MW
437 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
438 if (!alt) alt = "";
439
c2f493a4 440 alt_odb_tail = &alt_odb_list;
c5950164 441 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
442
443 read_info_alternates(get_object_directory(), 0);
444}
445
3096b2ec 446/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
33d4221c 447static int freshen_file(const char *fn)
ace1534d 448{
33d4221c
JK
449 struct utimbuf t;
450 t.actime = t.modtime = time(NULL);
451 return !utime(fn, &t);
0f4dc14a 452}
ace1534d 453
3096b2ec
JK
454/*
455 * All of the check_and_freshen functions return 1 if the file exists and was
456 * freshened (if freshening was requested), 0 otherwise. If they return
457 * 0, you should not assume that it is safe to skip a write of the object (it
458 * either does not exist on disk, or has a stale mtime and may be subject to
459 * pruning).
460 */
33d4221c
JK
461static int check_and_freshen_file(const char *fn, int freshen)
462{
463 if (access(fn, F_OK))
464 return 0;
3096b2ec 465 if (freshen && !freshen_file(fn))
33d4221c
JK
466 return 0;
467 return 1;
468}
469
470static int check_and_freshen_local(const unsigned char *sha1, int freshen)
471{
472 return check_and_freshen_file(sha1_file_name(sha1), freshen);
473}
474
475static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
0f4dc14a
BC
476{
477 struct alternate_object_database *alt;
9a217f2a 478 prepare_alt_odb();
d5a63b99 479 for (alt = alt_odb_list; alt; alt = alt->next) {
0f4dc14a 480 fill_sha1_path(alt->name, sha1);
33d4221c 481 if (check_and_freshen_file(alt->base, freshen))
c529d75a 482 return 1;
ace1534d 483 }
c529d75a 484 return 0;
ace1534d
JH
485}
486
33d4221c
JK
487static int check_and_freshen(const unsigned char *sha1, int freshen)
488{
489 return check_and_freshen_local(sha1, freshen) ||
490 check_and_freshen_nonlocal(sha1, freshen);
491}
492
493int has_loose_object_nonlocal(const unsigned char *sha1)
494{
495 return check_and_freshen_nonlocal(sha1, 0);
496}
497
0f4dc14a
BC
498static int has_loose_object(const unsigned char *sha1)
499{
33d4221c 500 return check_and_freshen(sha1, 0);
0f4dc14a
BC
501}
502
60bb8b14 503static unsigned int pack_used_ctr;
a53128b6
SP
504static unsigned int pack_mmap_calls;
505static unsigned int peak_pack_open_windows;
506static unsigned int pack_open_windows;
c7934306
SP
507static unsigned int pack_open_fds;
508static unsigned int pack_max_fds;
a53128b6 509static size_t peak_pack_mapped;
60bb8b14 510static size_t pack_mapped;
9a217f2a 511struct packed_git *packed_git;
1f688557 512
b79d18c9 513void pack_report(void)
a53128b6
SP
514{
515 fprintf(stderr,
e05db0fd
PR
516 "pack_report: getpagesize() = %10" SZ_FMT "\n"
517 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
518 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
519 sz_fmt(getpagesize()),
520 sz_fmt(packed_git_window_size),
521 sz_fmt(packed_git_limit));
a53128b6
SP
522 fprintf(stderr,
523 "pack_report: pack_used_ctr = %10u\n"
524 "pack_report: pack_mmap_calls = %10u\n"
525 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
526 "pack_report: pack_mapped = "
527 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
528 pack_used_ctr,
529 pack_mmap_calls,
530 pack_open_windows, peak_pack_open_windows,
9e42d6a1 531 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
532}
533
d40d535b
MH
534/*
535 * Open and mmap the index file at path, perform a couple of
536 * consistency checks, then record its information to p. Return 0 on
537 * success.
538 */
539static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
540{
541 void *idx_map;
42873078 542 struct pack_idx_header *hdr;
dc49cd76 543 size_t idx_size;
74e34e1f 544 uint32_t version, nr, i, *index;
f4e51683 545 int fd = git_open_noatime(path);
1f688557 546 struct stat st;
42873078 547
1f688557
JH
548 if (fd < 0)
549 return -1;
550 if (fstat(fd, &st)) {
551 close(fd);
552 return -1;
553 }
dc49cd76 554 idx_size = xsize_t(st.st_size);
2d88451b
SP
555 if (idx_size < 4 * 256 + 20 + 20) {
556 close(fd);
557 return error("index file %s is too small", path);
558 }
c4712e45 559 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 560 close(fd);
1f688557 561
42873078
NP
562 hdr = idx_map;
563 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
564 version = ntohl(hdr->idx_version);
565 if (version < 2 || version > 2) {
566 munmap(idx_map, idx_size);
6e1c2344 567 return error("index file %s is version %"PRIu32
74e34e1f
NP
568 " and is not supported by this binary"
569 " (try upgrading GIT to a newer version)",
570 path, version);
571 }
572 } else
573 version = 1;
df1b059d 574
1f688557 575 nr = 0;
42873078 576 index = idx_map;
74e34e1f
NP
577 if (version > 1)
578 index += 2; /* skip index header */
1f688557 579 for (i = 0; i < 256; i++) {
326bf396 580 uint32_t n = ntohl(index[i]);
2d88451b
SP
581 if (n < nr) {
582 munmap(idx_map, idx_size);
df1b059d 583 return error("non-monotonic index %s", path);
2d88451b 584 }
1f688557
JH
585 nr = n;
586 }
587
74e34e1f
NP
588 if (version == 1) {
589 /*
590 * Total size:
591 * - 256 index entries 4 bytes each
592 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
593 * - 20-byte SHA1 of the packfile
594 * - 20-byte SHA1 file checksum
595 */
596 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
597 munmap(idx_map, idx_size);
eef427a0 598 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
599 }
600 } else if (version == 2) {
601 /*
602 * Minimum size:
603 * - 8 bytes of header
604 * - 256 index entries 4 bytes each
605 * - 20-byte sha1 entry * nr
606 * - 4-byte crc entry * nr
607 * - 4-byte offset entry * nr
608 * - 20-byte SHA1 of the packfile
609 * - 20-byte SHA1 file checksum
610 * And after the 4-byte offset table might be a
611 * variable sized table containing 8-byte entries
612 * for offsets larger than 2^31.
613 */
614 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
615 unsigned long max_size = min_size;
616 if (nr)
617 max_size += (nr - 1)*8;
618 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 619 munmap(idx_map, idx_size);
eef427a0 620 return error("wrong index v2 file size in %s", path);
74e34e1f 621 }
7109c889
JH
622 if (idx_size != min_size &&
623 /*
624 * make sure we can deal with large pack offsets.
625 * 31-bit signed offset won't be enough, neither
626 * 32-bit unsigned one will be.
627 */
628 (sizeof(off_t) <= 4)) {
629 munmap(idx_map, idx_size);
630 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 631 }
2d88451b 632 }
1f688557 633
74e34e1f 634 p->index_version = version;
42873078
NP
635 p->index_data = idx_map;
636 p->index_size = idx_size;
57059091 637 p->num_objects = nr;
1f688557
JH
638 return 0;
639}
640
bc8e478a 641int open_pack_index(struct packed_git *p)
d079837e
SP
642{
643 char *idx_name;
644 int ret;
645
646 if (p->index_data)
647 return 0;
648
649 idx_name = xstrdup(p->pack_name);
650 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
651 ret = check_packed_git_idx(idx_name, p);
652 free(idx_name);
653 return ret;
654}
655
11daf39b
SP
656static void scan_windows(struct packed_git *p,
657 struct packed_git **lru_p,
658 struct pack_window **lru_w,
659 struct pack_window **lru_l)
1f688557 660{
11daf39b
SP
661 struct pack_window *w, *w_l;
662
663 for (w_l = NULL, w = p->windows; w; w = w->next) {
664 if (!w->inuse_cnt) {
665 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
666 *lru_p = p;
667 *lru_w = w;
668 *lru_l = w_l;
54044bf8 669 }
54044bf8 670 }
11daf39b 671 w_l = w;
f9253394 672 }
11daf39b
SP
673}
674
7c3ecb32 675static int unuse_one_window(struct packed_git *current)
11daf39b
SP
676{
677 struct packed_git *p, *lru_p = NULL;
678 struct pack_window *lru_w = NULL, *lru_l = NULL;
679
680 if (current)
681 scan_windows(current, &lru_p, &lru_w, &lru_l);
682 for (p = packed_git; p; p = p->next)
683 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
684 if (lru_p) {
685 munmap(lru_w->base, lru_w->len);
686 pack_mapped -= lru_w->len;
687 if (lru_l)
688 lru_l->next = lru_w->next;
7c3ecb32 689 else
54044bf8 690 lru_p->windows = lru_w->next;
54044bf8 691 free(lru_w);
a53128b6 692 pack_open_windows--;
54044bf8
SP
693 return 1;
694 }
695 return 0;
f9253394
JH
696}
697
7c3ecb32 698void release_pack_memory(size_t need)
97bfeb34
SP
699{
700 size_t cur = pack_mapped;
7c3ecb32 701 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
97bfeb34
SP
702 ; /* nothing */
703}
704
02710228
SP
705static void mmap_limit_check(size_t length)
706{
707 static size_t limit = 0;
708 if (!limit) {
709 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
710 if (!limit)
711 limit = SIZE_MAX;
712 }
713 if (length > limit)
714 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
715 (uintmax_t)length, (uintmax_t)limit);
716}
717
1570856b
JK
718void *xmmap_gently(void *start, size_t length,
719 int prot, int flags, int fd, off_t offset)
58ecbd5e 720{
02710228
SP
721 void *ret;
722
723 mmap_limit_check(length);
724 ret = mmap(start, length, prot, flags, fd, offset);
58ecbd5e
JN
725 if (ret == MAP_FAILED) {
726 if (!length)
727 return NULL;
7c3ecb32 728 release_pack_memory(length);
58ecbd5e 729 ret = mmap(start, length, prot, flags, fd, offset);
58ecbd5e
JN
730 }
731 return ret;
732}
733
1570856b
JK
734void *xmmap(void *start, size_t length,
735 int prot, int flags, int fd, off_t offset)
736{
737 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
738 if (ret == MAP_FAILED)
9ca0aaf6 739 die_errno("mmap failed");
1570856b
JK
740 return ret;
741}
742
c9ced051
SP
743void close_pack_windows(struct packed_git *p)
744{
745 while (p->windows) {
746 struct pack_window *w = p->windows;
747
748 if (w->inuse_cnt)
749 die("pack '%s' still has open windows to it",
750 p->pack_name);
751 munmap(w->base, w->len);
752 pack_mapped -= w->len;
753 pack_open_windows--;
754 p->windows = w->next;
755 free(w);
756 }
757}
758
88d0db55
BC
759/*
760 * The LRU pack is the one with the oldest MRU window, preferring packs
761 * with no used windows, or the oldest mtime if it has no windows allocated.
762 */
763static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
764{
765 struct pack_window *w, *this_mru_w;
766 int has_windows_inuse = 0;
767
768 /*
769 * Reject this pack if it has windows and the previously selected
770 * one does not. If this pack does not have windows, reject
771 * it if the pack file is newer than the previously selected one.
772 */
773 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
774 return;
775
776 for (w = this_mru_w = p->windows; w; w = w->next) {
777 /*
778 * Reject this pack if any of its windows are in use,
779 * but the previously selected pack did not have any
780 * inuse windows. Otherwise, record that this pack
781 * has windows in use.
782 */
783 if (w->inuse_cnt) {
784 if (*accept_windows_inuse)
785 has_windows_inuse = 1;
786 else
787 return;
788 }
789
790 if (w->last_used > this_mru_w->last_used)
791 this_mru_w = w;
792
793 /*
794 * Reject this pack if it has windows that have been
795 * used more recently than the previously selected pack.
796 * If the previously selected pack had windows inuse and
797 * we have not encountered a window in this pack that is
798 * inuse, skip this check since we prefer a pack with no
799 * inuse windows to one that has inuse windows.
800 */
801 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
802 this_mru_w->last_used > (*mru_w)->last_used)
803 return;
804 }
805
806 /*
807 * Select this pack.
808 */
809 *mru_w = this_mru_w;
810 *lru_p = p;
811 *accept_windows_inuse = has_windows_inuse;
812}
813
814static int close_one_pack(void)
815{
816 struct packed_git *p, *lru_p = NULL;
817 struct pack_window *mru_w = NULL;
818 int accept_windows_inuse = 1;
819
820 for (p = packed_git; p; p = p->next) {
821 if (p->pack_fd == -1)
822 continue;
823 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
824 }
825
826 if (lru_p) {
827 close(lru_p->pack_fd);
828 pack_open_fds--;
829 lru_p->pack_fd = -1;
830 return 1;
831 }
832
833 return 0;
834}
835
03e79c88 836void unuse_pack(struct pack_window **w_cursor)
f9253394 837{
03e79c88
SP
838 struct pack_window *w = *w_cursor;
839 if (w) {
840 w->inuse_cnt--;
841 *w_cursor = NULL;
842 }
1f688557
JH
843}
844
fa5fc15d
SP
845void close_pack_index(struct packed_git *p)
846{
847 if (p->index_data) {
848 munmap((void *)p->index_data, p->index_size);
849 p->index_data = NULL;
850 }
851}
852
c74faea1
NP
853/*
854 * This is used by git-repack in case a newly created pack happens to
855 * contain the same set of objects as an existing one. In that case
856 * the resulting file might be different even if its name would be the
857 * same. It is best to close any reference to the old pack before it is
01689909 858 * replaced on disk. Of course no index pointers or windows for given pack
c74faea1
NP
859 * must subsist at this point. If ever objects from this pack are requested
860 * again, the new version of the pack will be reinitialized through
861 * reprepare_packed_git().
862 */
863void free_pack_by_name(const char *pack_name)
864{
865 struct packed_git *p, **pp = &packed_git;
866
867 while (*pp) {
868 p = *pp;
869 if (strcmp(pack_name, p->pack_name) == 0) {
fa3a0c94 870 clear_delta_base_cache();
c74faea1 871 close_pack_windows(p);
c7934306 872 if (p->pack_fd != -1) {
c74faea1 873 close(p->pack_fd);
c7934306
SP
874 pack_open_fds--;
875 }
fa5fc15d 876 close_pack_index(p);
c74faea1
NP
877 free(p->bad_object_sha1);
878 *pp = p->next;
c01f51cc
NTND
879 if (last_found_pack == p)
880 last_found_pack = NULL;
c74faea1
NP
881 free(p);
882 return;
883 }
884 pp = &p->next;
885 }
886}
887
a0788266
JS
888static unsigned int get_max_fd_limit(void)
889{
890#ifdef RLIMIT_NOFILE
491a8dec
JH
891 {
892 struct rlimit lim;
a0788266 893
491a8dec
JH
894 if (!getrlimit(RLIMIT_NOFILE, &lim))
895 return lim.rlim_cur;
896 }
897#endif
898
899#ifdef _SC_OPEN_MAX
900 {
901 long open_max = sysconf(_SC_OPEN_MAX);
902 if (0 < open_max)
903 return open_max;
904 /*
905 * Otherwise, we got -1 for one of the two
906 * reasons:
907 *
908 * (1) sysconf() did not understand _SC_OPEN_MAX
909 * and signaled an error with -1; or
910 * (2) sysconf() said there is no limit.
911 *
912 * We _could_ clear errno before calling sysconf() to
913 * tell these two cases apart and return a huge number
914 * in the latter case to let the caller cap it to a
915 * value that is not so selfish, but letting the
916 * fallback OPEN_MAX codepath take care of these cases
917 * is a lot simpler.
918 */
919 }
920#endif
a0788266 921
491a8dec 922#ifdef OPEN_MAX
a0788266
JS
923 return OPEN_MAX;
924#else
925 return 1; /* see the caller ;-) */
926#endif
927}
928
3cf8b462
SP
929/*
930 * Do not call this directly as this leaks p->pack_fd on error return;
931 * call open_packed_git() instead.
932 */
933static int open_packed_git_1(struct packed_git *p)
1f688557 934{
9bc879c1
SP
935 struct stat st;
936 struct pack_header hdr;
937 unsigned char sha1[20];
938 unsigned char *idx_sha1;
2c039da8 939 long fd_flag;
9bc879c1 940
d079837e
SP
941 if (!p->index_data && open_pack_index(p))
942 return error("packfile %s index unavailable", p->pack_name);
943
c7934306 944 if (!pack_max_fds) {
a0788266 945 unsigned int max_fds = get_max_fd_limit();
c7934306
SP
946
947 /* Save 3 for stdin/stdout/stderr, 22 for work */
948 if (25 < max_fds)
949 pack_max_fds = max_fds - 25;
950 else
951 pack_max_fds = 1;
952 }
953
88d0db55 954 while (pack_max_fds <= pack_open_fds && close_one_pack())
c7934306
SP
955 ; /* nothing */
956
f4e51683 957 p->pack_fd = git_open_noatime(p->pack_name);
9bc879c1 958 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 959 return -1;
c7934306 960 pack_open_fds++;
9bc879c1
SP
961
962 /* If we created the struct before we had the pack we lack size. */
bf592c50 963 if (!p->pack_size) {
bf592c50 964 if (!S_ISREG(st.st_mode))
072db278 965 return error("packfile %s not a regular file", p->pack_name);
bf592c50 966 p->pack_size = st.st_size;
9bc879c1 967 } else if (p->pack_size != st.st_size)
072db278 968 return error("packfile %s size changed", p->pack_name);
9bc879c1 969
2c039da8
JH
970 /* We leave these file descriptors open with sliding mmap;
971 * there is no point keeping them open across exec(), though.
972 */
973 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
974 if (fd_flag < 0)
072db278 975 return error("cannot determine file descriptor flags");
2c039da8
JH
976 fd_flag |= FD_CLOEXEC;
977 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 978 return error("cannot set FD_CLOEXEC");
2c039da8 979
9bc879c1 980 /* Verify we recognize this pack file format. */
e6e2bd62 981 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 982 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 983 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 984 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 985 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
986 return error("packfile %s is version %"PRIu32" and not"
987 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
988 p->pack_name, ntohl(hdr.hdr_version));
989
990 /* Verify the pack matches its index. */
57059091 991 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
992 return error("packfile %s claims to have %"PRIu32" objects"
993 " while index indicates %"PRIu32" objects",
57059091
NP
994 p->pack_name, ntohl(hdr.hdr_entries),
995 p->num_objects);
9bc879c1 996 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 997 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 998 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 999 return error("packfile %s signature is unavailable", p->pack_name);
42873078 1000 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 1001 if (hashcmp(sha1, idx_sha1))
072db278
SP
1002 return error("packfile %s does not match index", p->pack_name);
1003 return 0;
9bc879c1
SP
1004}
1005
3cf8b462
SP
1006static int open_packed_git(struct packed_git *p)
1007{
1008 if (!open_packed_git_1(p))
1009 return 0;
1010 if (p->pack_fd != -1) {
1011 close(p->pack_fd);
c7934306 1012 pack_open_fds--;
3cf8b462
SP
1013 p->pack_fd = -1;
1014 }
1015 return -1;
1016}
1017
c4001d92 1018static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
1019{
1020 /* We must promise at least 20 bytes (one hash) after the
1021 * offset is available from this window, otherwise the offset
1022 * is not actually in this window and a different window (which
1023 * has that one hash excess) must be used. This is to support
1024 * the object header and delta base parsing routines below.
1025 */
1026 off_t win_off = win->offset;
1027 return win_off <= offset
1028 && (offset + 20) <= (win_off + win->len);
1029}
1030
4b25d091 1031unsigned char *use_pack(struct packed_git *p,
03e79c88 1032 struct pack_window **w_cursor,
c4001d92 1033 off_t offset,
ef49a7a0 1034 unsigned long *left)
9bc879c1 1035{
60bb8b14 1036 struct pack_window *win = *w_cursor;
03e79c88 1037
a9d98a14 1038 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
1039 * pointless to ask for an offset into the middle of that
1040 * hash, and the in_window function above wouldn't match
1041 * don't allow an offset too close to the end of the file.
1042 */
d131b7af
SP
1043 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1044 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
1045 if (offset > (p->pack_size - 20))
1046 die("offset beyond end of packfile (truncated pack?)");
1047
1048 if (!win || !in_window(win, offset)) {
1049 if (win)
1050 win->inuse_cnt--;
1051 for (win = p->windows; win; win = win->next) {
1052 if (in_window(win, offset))
1053 break;
1054 }
1055 if (!win) {
78a28df9 1056 size_t window_align = packed_git_window_size / 2;
dc49cd76 1057 off_t len;
d131b7af
SP
1058
1059 if (p->pack_fd == -1 && open_packed_git(p))
1060 die("packfile %s cannot be accessed", p->pack_name);
1061
60bb8b14 1062 win = xcalloc(1, sizeof(*win));
78a28df9 1063 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
1064 len = p->pack_size - win->offset;
1065 if (len > packed_git_window_size)
1066 len = packed_git_window_size;
1067 win->len = (size_t)len;
60bb8b14 1068 pack_mapped += win->len;
11daf39b 1069 while (packed_git_limit < pack_mapped
7c3ecb32 1070 && unuse_one_window(p))
60bb8b14 1071 ; /* nothing */
c4712e45 1072 win->base = xmmap(NULL, win->len,
60bb8b14
SP
1073 PROT_READ, MAP_PRIVATE,
1074 p->pack_fd, win->offset);
1075 if (win->base == MAP_FAILED)
73b4e4be
SP
1076 die("packfile %s cannot be mapped: %s",
1077 p->pack_name,
1078 strerror(errno));
d131b7af
SP
1079 if (!win->offset && win->len == p->pack_size
1080 && !p->do_not_close) {
1081 close(p->pack_fd);
1082 pack_open_fds--;
1083 p->pack_fd = -1;
1084 }
a53128b6
SP
1085 pack_mmap_calls++;
1086 pack_open_windows++;
1087 if (pack_mapped > peak_pack_mapped)
1088 peak_pack_mapped = pack_mapped;
1089 if (pack_open_windows > peak_pack_open_windows)
1090 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
1091 win->next = p->windows;
1092 p->windows = win;
1093 }
1f688557 1094 }
03e79c88
SP
1095 if (win != *w_cursor) {
1096 win->last_used = pack_used_ctr++;
1097 win->inuse_cnt++;
1098 *w_cursor = win;
1099 }
60bb8b14 1100 offset -= win->offset;
03e79c88 1101 if (left)
dc49cd76 1102 *left = win->len - xsize_t(offset);
03e79c88 1103 return win->base + offset;
1f688557
JH
1104}
1105
27d69a46
NP
1106static struct packed_git *alloc_packed_git(int extra)
1107{
1108 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1109 memset(p, 0, sizeof(*p));
1110 p->pack_fd = -1;
1111 return p;
1112}
1113
e0500293
JN
1114static void try_to_free_pack_memory(size_t size)
1115{
7c3ecb32 1116 release_pack_memory(size);
e0500293
JN
1117}
1118
42873078 1119struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557 1120{
e0500293 1121 static int have_set_try_to_free_routine;
1f688557 1122 struct stat st;
27d69a46 1123 struct packed_git *p = alloc_packed_git(path_len + 2);
1f688557 1124
e0500293
JN
1125 if (!have_set_try_to_free_routine) {
1126 have_set_try_to_free_routine = 1;
1127 set_try_to_free_routine(try_to_free_pack_memory);
1128 }
1129
42873078
NP
1130 /*
1131 * Make sure a corresponding .pack file exists and that
1132 * the index looks sane.
1133 */
1134 path_len -= strlen(".idx");
27d69a46
NP
1135 if (path_len < 1) {
1136 free(p);
1f688557 1137 return NULL;
27d69a46 1138 }
42873078 1139 memcpy(p->pack_name, path, path_len);
8d25931d
BC
1140
1141 strcpy(p->pack_name + path_len, ".keep");
1142 if (!access(p->pack_name, F_OK))
1143 p->pack_keep = 1;
1144
42873078 1145 strcpy(p->pack_name + path_len, ".pack");
d079837e 1146 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 1147 free(p);
1f688557
JH
1148 return NULL;
1149 }
42873078 1150
1f688557
JH
1151 /* ok, it looks sane as far as we can check without
1152 * actually mapping the pack file.
1153 */
1f688557 1154 p->pack_size = st.st_size;
9d835df2 1155 p->pack_local = local;
b867092f 1156 p->mtime = st.st_mtime;
42873078
NP
1157 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1158 hashclr(p->sha1);
1f688557
JH
1159 return p;
1160}
1161
7b64469a 1162struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 1163{
42873078 1164 const char *path = sha1_pack_name(sha1);
27d69a46 1165 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
bf592c50 1166
27d69a46
NP
1167 strcpy(p->pack_name, path);
1168 hashcpy(p->sha1, sha1);
42873078
NP
1169 if (check_packed_git_idx(idx_path, p)) {
1170 free(p);
bf592c50 1171 return NULL;
42873078 1172 }
bf592c50 1173
bf592c50
DB
1174 return p;
1175}
1176
1177void install_packed_git(struct packed_git *pack)
1178{
c7934306
SP
1179 if (pack->pack_fd != -1)
1180 pack_open_fds++;
1181
bf592c50
DB
1182 pack->next = packed_git;
1183 packed_git = pack;
1184}
1185
543c5caa
NTND
1186void (*report_garbage)(const char *desc, const char *path);
1187
1188static void report_helper(const struct string_list *list,
1189 int seen_bits, int first, int last)
1190{
1191 const char *msg;
1192 switch (seen_bits) {
1193 case 0:
235e8d59 1194 msg = "no corresponding .idx or .pack";
543c5caa
NTND
1195 break;
1196 case 1:
1197 msg = "no corresponding .idx";
1198 break;
1199 case 2:
1200 msg = "no corresponding .pack";
1201 break;
1202 default:
1203 return;
1204 }
1205 for (; first < last; first++)
1206 report_garbage(msg, list->items[first].string);
1207}
1208
1209static void report_pack_garbage(struct string_list *list)
1210{
1211 int i, baselen = -1, first = 0, seen_bits = 0;
1212
1213 if (!report_garbage)
1214 return;
1215
3383e199 1216 string_list_sort(list);
543c5caa
NTND
1217
1218 for (i = 0; i < list->nr; i++) {
1219 const char *path = list->items[i].string;
1220 if (baselen != -1 &&
1221 strncmp(path, list->items[first].string, baselen)) {
1222 report_helper(list, seen_bits, first, i);
1223 baselen = -1;
1224 seen_bits = 0;
1225 }
1226 if (baselen == -1) {
1227 const char *dot = strrchr(path, '.');
1228 if (!dot) {
1229 report_garbage("garbage found", path);
1230 continue;
1231 }
1232 baselen = dot - path + 1;
1233 first = i;
1234 }
1235 if (!strcmp(path + baselen, "pack"))
1236 seen_bits |= 1;
1237 else if (!strcmp(path + baselen, "idx"))
1238 seen_bits |= 2;
1239 }
1240 report_helper(list, seen_bits, first, list->nr);
1241}
1242
9d835df2 1243static void prepare_packed_git_one(char *objdir, int local)
1f688557 1244{
880fb8de
RS
1245 struct strbuf path = STRBUF_INIT;
1246 size_t dirnamelen;
1f688557
JH
1247 DIR *dir;
1248 struct dirent *de;
543c5caa 1249 struct string_list garbage = STRING_LIST_INIT_DUP;
1f688557 1250
880fb8de
RS
1251 strbuf_addstr(&path, objdir);
1252 strbuf_addstr(&path, "/pack");
1253 dir = opendir(path.buf);
b5b16990 1254 if (!dir) {
26125f6b 1255 if (errno != ENOENT)
bd2afde8 1256 error("unable to open object pack directory: %s: %s",
880fb8de
RS
1257 path.buf, strerror(errno));
1258 strbuf_release(&path);
1f688557 1259 return;
b5b16990 1260 }
880fb8de
RS
1261 strbuf_addch(&path, '/');
1262 dirnamelen = path.len;
1f688557 1263 while ((de = readdir(dir)) != NULL) {
1f688557 1264 struct packed_git *p;
47bf4b0f 1265 size_t base_len;
543c5caa
NTND
1266
1267 if (is_dot_or_dotdot(de->d_name))
9cb18f56
JM
1268 continue;
1269
880fb8de
RS
1270 strbuf_setlen(&path, dirnamelen);
1271 strbuf_addstr(&path, de->d_name);
d90906a9 1272
47bf4b0f
JK
1273 base_len = path.len;
1274 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
d90906a9
NTND
1275 /* Don't reopen a pack we already have. */
1276 for (p = packed_git; p; p = p->next) {
47bf4b0f
JK
1277 size_t len;
1278 if (strip_suffix(p->pack_name, ".pack", &len) &&
1279 len == base_len &&
1280 !memcmp(p->pack_name, path.buf, len))
d90906a9
NTND
1281 break;
1282 }
1283 if (p == NULL &&
1284 /*
1285 * See if it really is a valid .idx file with
1286 * corresponding .pack file that we can map.
1287 */
880fb8de 1288 (p = add_packed_git(path.buf, path.len, local)) != NULL)
d90906a9 1289 install_packed_git(p);
86f7780c 1290 }
543c5caa
NTND
1291
1292 if (!report_garbage)
1293 continue;
1294
2975c770
JK
1295 if (ends_with(de->d_name, ".idx") ||
1296 ends_with(de->d_name, ".pack") ||
1297 ends_with(de->d_name, ".bitmap") ||
1298 ends_with(de->d_name, ".keep"))
880fb8de 1299 string_list_append(&garbage, path.buf);
543c5caa 1300 else
880fb8de 1301 report_garbage("garbage found", path.buf);
1f688557 1302 }
5b35bcd5 1303 closedir(dir);
543c5caa
NTND
1304 report_pack_garbage(&garbage);
1305 string_list_clear(&garbage, 0);
880fb8de 1306 strbuf_release(&path);
1f688557
JH
1307}
1308
b867092f
JH
1309static int sort_pack(const void *a_, const void *b_)
1310{
1311 struct packed_git *a = *((struct packed_git **)a_);
1312 struct packed_git *b = *((struct packed_git **)b_);
1313 int st;
1314
1315 /*
1316 * Local packs tend to contain objects specific to our
1317 * variant of the project than remote ones. In addition,
1318 * remote ones could be on a network mounted filesystem.
1319 * Favor local ones for these reasons.
1320 */
1321 st = a->pack_local - b->pack_local;
1322 if (st)
1323 return -st;
1324
1325 /*
1326 * Younger packs tend to contain more recent objects,
1327 * and more recent objects tend to get accessed more
1328 * often.
1329 */
1330 if (a->mtime < b->mtime)
1331 return 1;
1332 else if (a->mtime == b->mtime)
1333 return 0;
1334 return -1;
1335}
1336
1337static void rearrange_packed_git(void)
1338{
1339 struct packed_git **ary, *p;
1340 int i, n;
1341
1342 for (n = 0, p = packed_git; p; p = p->next)
1343 n++;
1344 if (n < 2)
1345 return;
1346
1347 /* prepare an array of packed_git for easier sorting */
1348 ary = xcalloc(n, sizeof(struct packed_git *));
1349 for (n = 0, p = packed_git; p; p = p->next)
1350 ary[n++] = p;
1351
1352 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1353
1354 /* link them back again */
1355 for (i = 0; i < n - 1; i++)
1356 ary[i]->next = ary[i + 1];
1357 ary[n - 1]->next = NULL;
1358 packed_git = ary[0];
1359
1360 free(ary);
1361}
1362
637cdd9d 1363static int prepare_packed_git_run_once = 0;
9a217f2a 1364void prepare_packed_git(void)
1f688557 1365{
d5a63b99 1366 struct alternate_object_database *alt;
1f688557 1367
637cdd9d 1368 if (prepare_packed_git_run_once)
1f688557 1369 return;
9d835df2 1370 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 1371 prepare_alt_odb();
d5a63b99 1372 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 1373 alt->name[-1] = 0;
9d835df2 1374 prepare_packed_git_one(alt->base, 0);
1494e038 1375 alt->name[-1] = '/';
1f688557 1376 }
b867092f 1377 rearrange_packed_git();
637cdd9d
JK
1378 prepare_packed_git_run_once = 1;
1379}
1380
fc04c412 1381void reprepare_packed_git(void)
637cdd9d
JK
1382{
1383 prepare_packed_git_run_once = 0;
1384 prepare_packed_git();
1f688557
JH
1385}
1386
8eca0b47
NP
1387static void mark_bad_packed_object(struct packed_git *p,
1388 const unsigned char *sha1)
1389{
1390 unsigned i;
1391 for (i = 0; i < p->num_bad_objects; i++)
1392 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1393 return;
1394 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1395 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1396 p->num_bad_objects++;
1397}
1398
b6c4cecc 1399static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
ac939109
NP
1400{
1401 struct packed_git *p;
1402 unsigned i;
1403
1404 for (p = packed_git; p; p = p->next)
1405 for (i = 0; i < p->num_bad_objects; i++)
1406 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
b6c4cecc
JH
1407 return p;
1408 return NULL;
ac939109
NP
1409}
1410
090ea126
NTND
1411/*
1412 * With an in-core object data in "map", rehash it to make sure the
1413 * object name actually matches "sha1" to detect object corruption.
1414 * With "map" == NULL, try reading the object named with "sha1" using
1415 * the streaming interface and rehash it to do the same.
1416 */
1417int check_sha1_signature(const unsigned char *sha1, void *map,
1418 unsigned long size, const char *type)
0fcfd160
LT
1419{
1420 unsigned char real_sha1[20];
090ea126
NTND
1421 enum object_type obj_type;
1422 struct git_istream *st;
1423 git_SHA_CTX c;
1424 char hdr[32];
1425 int hdrlen;
1426
1427 if (map) {
1428 hash_sha1_file(map, size, type, real_sha1);
1429 return hashcmp(sha1, real_sha1) ? -1 : 0;
1430 }
1431
1432 st = open_istream(sha1, &obj_type, &size, NULL);
1433 if (!st)
1434 return -1;
1435
1436 /* Generate the header */
1437 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1438
1439 /* Sha1.. */
1440 git_SHA1_Init(&c);
1441 git_SHA1_Update(&c, hdr, hdrlen);
1442 for (;;) {
1443 char buf[1024 * 16];
1444 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1445
f54fac53
JK
1446 if (readlen < 0) {
1447 close_istream(st);
1448 return -1;
1449 }
090ea126
NTND
1450 if (!readlen)
1451 break;
1452 git_SHA1_Update(&c, buf, readlen);
1453 }
1454 git_SHA1_Final(real_sha1, &c);
1455 close_istream(st);
a89fccd2 1456 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1457}
1458
ec73f580 1459int git_open_noatime(const char *name)
44d1c19e
LT
1460{
1461 static int sha1_file_open_flag = O_NOATIME;
44d1c19e 1462
f2e872aa
SP
1463 for (;;) {
1464 int fd = open(name, O_RDONLY | sha1_file_open_flag);
44d1c19e 1465 if (fd >= 0)
f2e872aa
SP
1466 return fd;
1467
f2e872aa
SP
1468 /* Might the failure be due to O_NOATIME? */
1469 if (errno != ENOENT && sha1_file_open_flag) {
44d1c19e 1470 sha1_file_open_flag = 0;
f2e872aa
SP
1471 continue;
1472 }
1473
1474 return -1;
44d1c19e 1475 }
44d1c19e
LT
1476}
1477
052fe5ea
JK
1478static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1479{
052fe5ea
JK
1480 struct alternate_object_database *alt;
1481
30d6c6ea 1482 if (!lstat(sha1_file_name(sha1), st))
052fe5ea
JK
1483 return 0;
1484
1485 prepare_alt_odb();
1486 errno = ENOENT;
1487 for (alt = alt_odb_list; alt; alt = alt->next) {
30d6c6ea 1488 fill_sha1_path(alt->name, sha1);
052fe5ea
JK
1489 if (!lstat(alt->base, st))
1490 return 0;
1491 }
1492
1493 return -1;
1494}
1495
44d1c19e
LT
1496static int open_sha1_file(const unsigned char *sha1)
1497{
1498 int fd;
44d1c19e 1499 struct alternate_object_database *alt;
d6c8a05b 1500 int most_interesting_errno;
44d1c19e 1501
30d6c6ea 1502 fd = git_open_noatime(sha1_file_name(sha1));
44d1c19e
LT
1503 if (fd >= 0)
1504 return fd;
d6c8a05b 1505 most_interesting_errno = errno;
44d1c19e
LT
1506
1507 prepare_alt_odb();
44d1c19e 1508 for (alt = alt_odb_list; alt; alt = alt->next) {
30d6c6ea 1509 fill_sha1_path(alt->name, sha1);
f4e51683 1510 fd = git_open_noatime(alt->base);
44d1c19e
LT
1511 if (fd >= 0)
1512 return fd;
d6c8a05b
JK
1513 if (most_interesting_errno == ENOENT)
1514 most_interesting_errno = errno;
44d1c19e 1515 }
d6c8a05b 1516 errno = most_interesting_errno;
44d1c19e
LT
1517 return -1;
1518}
1519
f0270efd 1520void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1521{
0fcfd160 1522 void *map;
144bde78 1523 int fd;
ace1534d 1524
44d1c19e
LT
1525 fd = open_sha1_file(sha1);
1526 map = NULL;
1527 if (fd >= 0) {
1528 struct stat st;
0fcfd160 1529
44d1c19e
LT
1530 if (!fstat(fd, &st)) {
1531 *size = xsize_t(st.st_size);
33e42de0
MM
1532 if (!*size) {
1533 /* mmap() is forbidden on empty files */
1534 error("object file %s is empty", sha1_file_name(sha1));
1535 return NULL;
1536 }
44d1c19e 1537 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1538 }
44d1c19e 1539 close(fd);
0fcfd160 1540 }
0fcfd160
LT
1541 return map;
1542}
1543
09ded04b
NP
1544unsigned long unpack_object_header_buffer(const unsigned char *buf,
1545 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1546{
ad1ed5ee 1547 unsigned shift;
48fb7deb 1548 unsigned long size, c;
ad1ed5ee
JH
1549 unsigned long used = 0;
1550
1551 c = buf[used++];
1552 *type = (c >> 4) & 7;
1553 size = c & 15;
1554 shift = 4;
1555 while (c & 0x80) {
f630cfda 1556 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1557 error("bad object header");
ea4f9685
JH
1558 size = used = 0;
1559 break;
09ded04b 1560 }
ad1ed5ee
JH
1561 c = buf[used++];
1562 size += (c & 0x7f) << shift;
1563 shift += 7;
1564 }
1565 *sizep = size;
1566 return used;
1567}
1568
eb4f4076 1569int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
ad1ed5ee 1570{
c4483576
LT
1571 /* Get the data stream */
1572 memset(stream, 0, sizeof(*stream));
1573 stream->next_in = map;
1574 stream->avail_in = mapsize;
1575 stream->next_out = buffer;
93821bd9
LT
1576 stream->avail_out = bufsiz;
1577
39c68542 1578 git_inflate_init(stream);
cc5c54e7 1579 return git_inflate(stream, 0);
c4483576
LT
1580}
1581
46f03448
KN
1582static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1583 unsigned long mapsize, void *buffer,
1584 unsigned long bufsiz, struct strbuf *header)
1585{
1586 int status;
1587
1588 status = unpack_sha1_header(stream, map, mapsize, buffer, bufsiz);
1589
1590 /*
1591 * Check if entire header is unpacked in the first iteration.
1592 */
1593 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1594 return 0;
1595
1596 /*
1597 * buffer[0..bufsiz] was not large enough. Copy the partial
1598 * result out to header, and then append the result of further
1599 * reading the stream.
1600 */
1601 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1602 stream->next_out = buffer;
1603 stream->avail_out = bufsiz;
1604
1605 do {
1606 status = git_inflate(stream, 0);
1607 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1608 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1609 return 0;
1610 stream->next_out = buffer;
1611 stream->avail_out = bufsiz;
1612 } while (status != Z_STREAM_END);
1613 return -1;
1614}
1615
ef49a7a0 1616static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1617{
1618 int bytes = strlen(buffer) + 1;
3aee68aa 1619 unsigned char *buf = xmallocz(size);
93821bd9 1620 unsigned long n;
7efbff75 1621 int status = Z_OK;
5180cacc 1622
93821bd9
LT
1623 n = stream->total_out - bytes;
1624 if (n > size)
1625 n = size;
1626 memcpy(buf, (char *) buffer + bytes, n);
1627 bytes = n;
456cdf6e
LT
1628 if (bytes <= size) {
1629 /*
1630 * The above condition must be (bytes <= size), not
1631 * (bytes < size). In other words, even though we
ccf5ace0 1632 * expect no more output and set avail_out to zero,
456cdf6e
LT
1633 * the input zlib stream may have bytes that express
1634 * "this concludes the stream", and we *do* want to
1635 * eat that input.
1636 *
1637 * Otherwise we would not be able to test that we
1638 * consumed all the input to reach the expected size;
1639 * we also want to check that zlib tells us that all
1640 * went well with status == Z_STREAM_END at the end.
1641 */
5180cacc
LT
1642 stream->next_out = buf + bytes;
1643 stream->avail_out = size - bytes;
7efbff75 1644 while (status == Z_OK)
39c68542 1645 status = git_inflate(stream, Z_FINISH);
5180cacc 1646 }
456cdf6e 1647 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1648 git_inflate_end(stream);
7efbff75
JH
1649 return buf;
1650 }
1651
1652 if (status < 0)
1653 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1654 else if (stream->avail_in)
1655 error("garbage at end of loose object '%s'",
1656 sha1_to_hex(sha1));
1657 free(buf);
1658 return NULL;
5180cacc
LT
1659}
1660
1661/*
1662 * We used to just use "sscanf()", but that's actually way
1663 * too permissive for what we want to check. So do an anal
1664 * object header parse by hand.
1665 */
46f03448
KN
1666static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1667 unsigned int flags)
5180cacc 1668{
46f03448 1669 const char *type_buf = hdr;
5180cacc 1670 unsigned long size;
46f03448 1671 int type, type_len = 0;
5180cacc
LT
1672
1673 /*
46f03448 1674 * The type can be of any size but is followed by
21666f1a 1675 * a space.
5180cacc 1676 */
5180cacc
LT
1677 for (;;) {
1678 char c = *hdr++;
1679 if (c == ' ')
1680 break;
46f03448 1681 type_len++;
5180cacc 1682 }
46f03448
KN
1683
1684 type = type_from_string_gently(type_buf, type_len, 1);
1685 if (oi->typename)
1686 strbuf_add(oi->typename, type_buf, type_len);
1687 /*
1688 * Set type to 0 if its an unknown object and
1689 * we're obtaining the type using '--allow-unkown-type'
1690 * option.
1691 */
1692 if ((flags & LOOKUP_UNKNOWN_OBJECT) && (type < 0))
1693 type = 0;
1694 else if (type < 0)
1695 die("invalid object type");
1696 if (oi->typep)
1697 *oi->typep = type;
5180cacc
LT
1698
1699 /*
1700 * The length must follow immediately, and be in canonical
1701 * decimal format (ie "010" is not valid).
1702 */
1703 size = *hdr++ - '0';
1704 if (size > 9)
1705 return -1;
1706 if (size) {
1707 for (;;) {
1708 unsigned long c = *hdr - '0';
1709 if (c > 9)
1710 break;
1711 hdr++;
1712 size = size * 10 + c;
1713 }
1714 }
46f03448
KN
1715
1716 if (oi->sizep)
1717 *oi->sizep = size;
5180cacc
LT
1718
1719 /*
1720 * The length must be followed by a zero byte
1721 */
46f03448
KN
1722 return *hdr ? -1 : type;
1723}
1724
1725int parse_sha1_header(const char *hdr, unsigned long *sizep)
1726{
1727 struct object_info oi;
1728
1729 oi.sizep = sizep;
1730 oi.typename = NULL;
1731 oi.typep = NULL;
1732 return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
5180cacc
LT
1733}
1734
7efbff75 1735static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1736{
5180cacc 1737 int ret;
ef49a7a0 1738 git_zstream stream;
5180cacc 1739 char hdr[8192];
0fcfd160 1740
5180cacc 1741 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1742 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1743 return NULL;
1744
7efbff75 1745 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1746}
1747
54dab52a
NP
1748unsigned long get_size_from_delta(struct packed_git *p,
1749 struct pack_window **w_curs,
1750 off_t curpos)
1751{
1752 const unsigned char *data;
1753 unsigned char delta_head[20], *in;
ef49a7a0 1754 git_zstream stream;
54dab52a
NP
1755 int st;
1756
1757 memset(&stream, 0, sizeof(stream));
1758 stream.next_out = delta_head;
1759 stream.avail_out = sizeof(delta_head);
1760
39c68542 1761 git_inflate_init(&stream);
54dab52a
NP
1762 do {
1763 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1764 stream.next_in = in;
39c68542 1765 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1766 curpos += stream.next_in - in;
1767 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1768 stream.total_out < sizeof(delta_head));
39c68542 1769 git_inflate_end(&stream);
3d77d877
NP
1770 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1771 error("delta data unpack-initial failed");
1772 return 0;
1773 }
54dab52a
NP
1774
1775 /* Examine the initial part of the delta to figure out
1776 * the result size.
1777 */
1778 data = delta_head;
1779
1780 /* ignore base size */
1781 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1782
1783 /* Read the result size */
1784 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1785}
1786
c4001d92 1787static off_t get_delta_base(struct packed_git *p,
03e79c88 1788 struct pack_window **w_curs,
c4001d92 1789 off_t *curpos,
21666f1a 1790 enum object_type type,
c4001d92 1791 off_t delta_obj_offset)
eb32d236 1792{
2b87c45b 1793 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1794 off_t base_offset;
eb32d236 1795
8d8a4ea5
SP
1796 /* use_pack() assured us we have [base_info, base_info + 20)
1797 * as a range that we can look at without walking off the
1798 * end of the mapped window. Its actually the hash size
1799 * that is assured. An OFS_DELTA longer than the hash size
1800 * is stupid, as then a REF_DELTA would be smaller to store.
1801 */
21666f1a 1802 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1803 unsigned used = 0;
1804 unsigned char c = base_info[used++];
1805 base_offset = c & 127;
1806 while (c & 128) {
1807 base_offset += 1;
8723f216 1808 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1809 return 0; /* overflow */
eb32d236
NP
1810 c = base_info[used++];
1811 base_offset = (base_offset << 7) + (c & 127);
1812 }
1813 base_offset = delta_obj_offset - base_offset;
d8f32556 1814 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1815 return 0; /* out of bound */
2b87c45b 1816 *curpos += used;
21666f1a 1817 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1818 /* The base entry _must_ be in the same pack */
1819 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1820 *curpos += 20;
eb32d236
NP
1821 } else
1822 die("I am totally screwed");
2b87c45b 1823 return base_offset;
eb32d236
NP
1824}
1825
5d642e75
JK
1826/*
1827 * Like get_delta_base above, but we return the sha1 instead of the pack
1828 * offset. This means it is cheaper for REF deltas (we do not have to do
1829 * the final object lookup), but more expensive for OFS deltas (we
1830 * have to load the revidx to convert the offset back into a sha1).
1831 */
1832static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1833 struct pack_window **w_curs,
1834 off_t curpos,
1835 enum object_type type,
1836 off_t delta_obj_offset)
1837{
1838 if (type == OBJ_REF_DELTA) {
1839 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1840 return base;
1841 } else if (type == OBJ_OFS_DELTA) {
1842 struct revindex_entry *revidx;
1843 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1844 type, delta_obj_offset);
1845
1846 if (!base_offset)
1847 return NULL;
1848
1849 revidx = find_pack_revindex(p, base_offset);
1850 if (!revidx)
1851 return NULL;
1852
1853 return nth_packed_object_sha1(p, revidx->nr);
1854 } else
1855 return NULL;
1856}
1857
f8c8abc5
JH
1858int unpack_object_header(struct packed_git *p,
1859 struct pack_window **w_curs,
1860 off_t *curpos,
1861 unsigned long *sizep)
a733cb60 1862{
03e79c88 1863 unsigned char *base;
ef49a7a0 1864 unsigned long left;
ad1ed5ee 1865 unsigned long used;
2b87c45b 1866 enum object_type type;
a733cb60 1867
8d8a4ea5 1868 /* use_pack() assures us we have [base, base + 20) available
0353a0c4 1869 * as a range that we can look at. (Its actually the hash
3dff5379 1870 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1871 * the maximum deflated object size is 2^137, which is just
1872 * insane, so we know won't exceed what we have been given.
1873 */
2b87c45b 1874 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1875 used = unpack_object_header_buffer(base, left, &type, sizep);
1876 if (!used) {
1877 type = OBJ_BAD;
1878 } else
1879 *curpos += used;
ad1ed5ee 1880
2b87c45b 1881 return type;
a733cb60
LT
1882}
1883
790d96c0
TR
1884static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1885{
1886 int type;
1887 struct revindex_entry *revidx;
1888 const unsigned char *sha1;
1889 revidx = find_pack_revindex(p, obj_offset);
1890 if (!revidx)
1891 return OBJ_BAD;
1892 sha1 = nth_packed_object_sha1(p, revidx->nr);
1893 mark_bad_packed_object(p, sha1);
1894 type = sha1_object_info(sha1, NULL);
1895 if (type <= OBJ_NONE)
1896 return OBJ_BAD;
1897 return type;
1898}
1899
790d96c0
TR
1900#define POI_STACK_PREALLOC 64
1901
90191d37
JK
1902static enum object_type packed_to_object_type(struct packed_git *p,
1903 off_t obj_offset,
1904 enum object_type type,
1905 struct pack_window **w_curs,
1906 off_t curpos)
1f688557 1907{
790d96c0
TR
1908 off_t small_poi_stack[POI_STACK_PREALLOC];
1909 off_t *poi_stack = small_poi_stack;
1910 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
5db47c2b 1911
790d96c0
TR
1912 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1913 off_t base_offset;
90191d37 1914 unsigned long size;
790d96c0
TR
1915 /* Push the object we're going to leave behind */
1916 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1917 poi_stack_alloc = alloc_nr(poi_stack_nr);
1918 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1919 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1920 } else {
1921 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1922 }
1923 poi_stack[poi_stack_nr++] = obj_offset;
1924 /* If parsing the base offset fails, just unwind */
90191d37 1925 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
790d96c0
TR
1926 if (!base_offset)
1927 goto unwind;
1928 curpos = obj_offset = base_offset;
90191d37 1929 type = unpack_object_header(p, w_curs, &curpos, &size);
790d96c0
TR
1930 if (type <= OBJ_NONE) {
1931 /* If getting the base itself fails, we first
1932 * retry the base, otherwise unwind */
1933 type = retry_bad_packed_offset(p, base_offset);
1934 if (type > OBJ_NONE)
1935 goto out;
1936 goto unwind;
1937 }
1938 }
1939
21666f1a 1940 switch (type) {
790d96c0 1941 case OBJ_BAD:
a733cb60 1942 case OBJ_COMMIT:
a733cb60 1943 case OBJ_TREE:
a733cb60 1944 case OBJ_BLOB:
a733cb60 1945 case OBJ_TAG:
a69d0943 1946 break;
1f688557 1947 default:
3d77d877
NP
1948 error("unknown object type %i at offset %"PRIuMAX" in %s",
1949 type, (uintmax_t)obj_offset, p->pack_name);
1950 type = OBJ_BAD;
1f688557 1951 }
790d96c0
TR
1952
1953out:
1954 if (poi_stack != small_poi_stack)
1955 free(poi_stack);
21666f1a 1956 return type;
790d96c0
TR
1957
1958unwind:
1959 while (poi_stack_nr) {
1960 obj_offset = poi_stack[--poi_stack_nr];
1961 type = retry_bad_packed_offset(p, obj_offset);
1962 if (type > OBJ_NONE)
1963 goto out;
1964 }
1965 type = OBJ_BAD;
1966 goto out;
1f688557
JH
1967}
1968
90191d37 1969static int packed_object_info(struct packed_git *p, off_t obj_offset,
23c339c0 1970 struct object_info *oi)
90191d37
JK
1971{
1972 struct pack_window *w_curs = NULL;
1973 unsigned long size;
1974 off_t curpos = obj_offset;
1975 enum object_type type;
1976
412916ee
JK
1977 /*
1978 * We always get the representation type, but only convert it to
1979 * a "real" type later if the caller is interested.
1980 */
90191d37
JK
1981 type = unpack_object_header(p, &w_curs, &curpos, &size);
1982
23c339c0 1983 if (oi->sizep) {
90191d37
JK
1984 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1985 off_t tmp_pos = curpos;
1986 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1987 type, obj_offset);
1988 if (!base_offset) {
1989 type = OBJ_BAD;
1990 goto out;
1991 }
23c339c0
JK
1992 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1993 if (*oi->sizep == 0) {
90191d37
JK
1994 type = OBJ_BAD;
1995 goto out;
1996 }
1997 } else {
23c339c0 1998 *oi->sizep = size;
90191d37
JK
1999 }
2000 }
2001
23c339c0 2002 if (oi->disk_sizep) {
90191d37 2003 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
23c339c0 2004 *oi->disk_sizep = revidx[1].offset - obj_offset;
90191d37
JK
2005 }
2006
23c339c0
JK
2007 if (oi->typep) {
2008 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
2009 if (*oi->typep < 0) {
412916ee
JK
2010 type = OBJ_BAD;
2011 goto out;
2012 }
2013 }
90191d37 2014
5d642e75
JK
2015 if (oi->delta_base_sha1) {
2016 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2017 const unsigned char *base;
2018
2019 base = get_delta_base_sha1(p, &w_curs, curpos,
2020 type, obj_offset);
2021 if (!base) {
2022 type = OBJ_BAD;
2023 goto out;
2024 }
2025
2026 hashcpy(oi->delta_base_sha1, base);
2027 } else
2028 hashclr(oi->delta_base_sha1);
2029 }
2030
90191d37
JK
2031out:
2032 unuse_pack(&w_curs);
2033 return type;
2034}
2035
eb950c19 2036static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 2037 struct pack_window **w_curs,
c4001d92 2038 off_t curpos,
eb950c19 2039 unsigned long size)
de530aaa
SP
2040{
2041 int st;
ef49a7a0 2042 git_zstream stream;
079afb18 2043 unsigned char *buffer, *in;
de530aaa 2044
735efde8
NTND
2045 buffer = xmallocz_gently(size);
2046 if (!buffer)
2047 return NULL;
de530aaa 2048 memset(&stream, 0, sizeof(stream));
de530aaa 2049 stream.next_out = buffer;
39eea7bd 2050 stream.avail_out = size + 1;
de530aaa 2051
39c68542 2052 git_inflate_init(&stream);
079afb18 2053 do {
2b87c45b 2054 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 2055 stream.next_in = in;
39c68542 2056 st = git_inflate(&stream, Z_FINISH);
39eea7bd
JH
2057 if (!stream.avail_out)
2058 break; /* the payload is larger than it should be */
2b87c45b 2059 curpos += stream.next_in - in;
079afb18 2060 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 2061 git_inflate_end(&stream);
de530aaa
SP
2062 if ((st != Z_STREAM_END) || stream.total_out != size) {
2063 free(buffer);
2064 return NULL;
2065 }
2066
2067 return buffer;
2068}
2069
e5e01619
LT
2070#define MAX_DELTA_CACHE (256)
2071
18bdec11 2072static size_t delta_base_cached;
5e08ecbf
NP
2073
2074static struct delta_base_cache_lru_list {
2075 struct delta_base_cache_lru_list *prev;
2076 struct delta_base_cache_lru_list *next;
2077} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
2078
e5e01619 2079static struct delta_base_cache_entry {
5e08ecbf
NP
2080 struct delta_base_cache_lru_list lru;
2081 void *data;
e5e01619
LT
2082 struct packed_git *p;
2083 off_t base_offset;
2084 unsigned long size;
e5e01619
LT
2085 enum object_type type;
2086} delta_base_cache[MAX_DELTA_CACHE];
2087
2088static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
2089{
2090 unsigned long hash;
2091
2092 hash = (unsigned long)p + (unsigned long)base_offset;
2093 hash += (hash >> 8) + (hash >> 16);
3358004a 2094 return hash % MAX_DELTA_CACHE;
e5e01619
LT
2095}
2096
84dd81c1
TR
2097static struct delta_base_cache_entry *
2098get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
5266d369
JH
2099{
2100 unsigned long hash = pack_entry_hash(p, base_offset);
84dd81c1
TR
2101 return delta_base_cache + hash;
2102}
2103
2104static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
2105 struct packed_git *p, off_t base_offset)
2106{
5266d369
JH
2107 return (ent->data && ent->p == p && ent->base_offset == base_offset);
2108}
2109
84dd81c1
TR
2110static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2111{
2112 struct delta_base_cache_entry *ent;
2113 ent = get_delta_base_cache_entry(p, base_offset);
2114 return eq_delta_base_cache_entry(ent, p, base_offset);
2115}
2116
2117static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2118{
2119 ent->data = NULL;
2120 ent->lru.next->prev = ent->lru.prev;
2121 ent->lru.prev->next = ent->lru.next;
2122 delta_base_cached -= ent->size;
2123}
2124
62f255ad 2125static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 2126 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 2127{
84dd81c1 2128 struct delta_base_cache_entry *ent;
e5e01619 2129 void *ret;
e5e01619 2130
84dd81c1
TR
2131 ent = get_delta_base_cache_entry(p, base_offset);
2132
2133 if (!eq_delta_base_cache_entry(ent, p, base_offset))
749bc58c 2134 return unpack_entry(p, base_offset, type, base_size);
e5e01619 2135
84dd81c1
TR
2136 ret = ent->data;
2137
2138 if (!keep_cache)
2139 clear_delta_base_cache_entry(ent);
2140 else
182af834 2141 ret = xmemdupz(ent->data, ent->size);
e5e01619
LT
2142 *type = ent->type;
2143 *base_size = ent->size;
2144 return ret;
62f255ad
LT
2145}
2146
18bdec11
SP
2147static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2148{
2149 if (ent->data) {
2150 free(ent->data);
2151 ent->data = NULL;
5e08ecbf
NP
2152 ent->lru.next->prev = ent->lru.prev;
2153 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
2154 delta_base_cached -= ent->size;
2155 }
2156}
2157
3d20c636
SP
2158void clear_delta_base_cache(void)
2159{
2160 unsigned long p;
2161 for (p = 0; p < MAX_DELTA_CACHE; p++)
2162 release_delta_base_cache(&delta_base_cache[p]);
2163}
2164
62f255ad
LT
2165static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2166 void *base, unsigned long base_size, enum object_type type)
2167{
5e08ecbf 2168 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 2169 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 2170 struct delta_base_cache_lru_list *lru;
e5e01619 2171
18bdec11
SP
2172 release_delta_base_cache(ent);
2173 delta_base_cached += base_size;
5e08ecbf
NP
2174
2175 for (lru = delta_base_cache_lru.next;
2176 delta_base_cached > delta_base_cache_limit
2177 && lru != &delta_base_cache_lru;
2178 lru = lru->next) {
2179 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
2180 if (f->type == OBJ_BLOB)
2181 release_delta_base_cache(f);
2182 }
5e08ecbf
NP
2183 for (lru = delta_base_cache_lru.next;
2184 delta_base_cached > delta_base_cache_limit
2185 && lru != &delta_base_cache_lru;
2186 lru = lru->next) {
2187 struct delta_base_cache_entry *f = (void *)lru;
2188 release_delta_base_cache(f);
2189 }
18bdec11 2190
e5e01619
LT
2191 ent->p = p;
2192 ent->base_offset = base_offset;
2193 ent->type = type;
2194 ent->data = base;
2195 ent->size = base_size;
5e08ecbf
NP
2196 ent->lru.next = &delta_base_cache_lru;
2197 ent->lru.prev = delta_base_cache_lru.prev;
2198 delta_base_cache_lru.prev->next = &ent->lru;
2199 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
2200}
2201
c2c5b270
CC
2202static void *read_object(const unsigned char *sha1, enum object_type *type,
2203 unsigned long *size);
2204
5f44324d
JH
2205static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2206{
67dc598e
KB
2207 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2208 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2209 p->pack_name, (uintmax_t)obj_offset);
5f44324d
JH
2210}
2211
0e8189e2
NP
2212int do_check_packed_object_crc;
2213
abe601bb
TR
2214#define UNPACK_ENTRY_STACK_PREALLOC 64
2215struct unpack_entry_stack_ent {
2216 off_t obj_offset;
2217 off_t curpos;
2218 unsigned long size;
2219};
2220
c4001d92 2221void *unpack_entry(struct packed_git *p, off_t obj_offset,
abe601bb 2222 enum object_type *final_type, unsigned long *final_size)
1f688557 2223{
03e79c88 2224 struct pack_window *w_curs = NULL;
c4001d92 2225 off_t curpos = obj_offset;
abe601bb
TR
2226 void *data = NULL;
2227 unsigned long size;
2228 enum object_type type;
2229 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2230 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2231 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2232 int base_from_cache = 0;
1f688557 2233
67dc598e 2234 write_pack_access_log(p, obj_offset);
5f44324d 2235
abe601bb
TR
2236 /* PHASE 1: drill down to the innermost base object */
2237 for (;;) {
2238 off_t base_offset;
2239 int i;
2240 struct delta_base_cache_entry *ent;
2241
77965f8b
NTND
2242 ent = get_delta_base_cache_entry(p, curpos);
2243 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2244 type = ent->type;
2245 data = ent->data;
2246 size = ent->size;
2247 clear_delta_base_cache_entry(ent);
2248 base_from_cache = 1;
2249 break;
2250 }
2251
abe601bb
TR
2252 if (do_check_packed_object_crc && p->index_version > 1) {
2253 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2254 unsigned long len = revidx[1].offset - obj_offset;
2255 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2256 const unsigned char *sha1 =
2257 nth_packed_object_sha1(p, revidx->nr);
2258 error("bad packed object CRC for %s",
2259 sha1_to_hex(sha1));
2260 mark_bad_packed_object(p, sha1);
2261 unuse_pack(&w_curs);
2262 return NULL;
2263 }
2264 }
2265
abe601bb
TR
2266 type = unpack_object_header(p, &w_curs, &curpos, &size);
2267 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2268 break;
2269
2270 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2271 if (!base_offset) {
2272 error("failed to validate delta base reference "
2273 "at offset %"PRIuMAX" from %s",
2274 (uintmax_t)curpos, p->pack_name);
2275 /* bail to phase 2, in hopes of recovery */
2276 data = NULL;
2277 break;
2278 }
2279
2280 /* push object, proceed to base */
2281 if (delta_stack_nr >= delta_stack_alloc
2282 && delta_stack == small_delta_stack) {
2283 delta_stack_alloc = alloc_nr(delta_stack_nr);
2284 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2285 memcpy(delta_stack, small_delta_stack,
2286 sizeof(*delta_stack)*delta_stack_nr);
2287 } else {
2288 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2289 }
2290 i = delta_stack_nr++;
2291 delta_stack[i].obj_offset = obj_offset;
2292 delta_stack[i].curpos = curpos;
2293 delta_stack[i].size = size;
2294
2295 curpos = obj_offset = base_offset;
0e8189e2
NP
2296 }
2297
abe601bb
TR
2298 /* PHASE 2: handle the base */
2299 switch (type) {
eb32d236
NP
2300 case OBJ_OFS_DELTA:
2301 case OBJ_REF_DELTA:
abe601bb
TR
2302 if (data)
2303 die("BUG in unpack_entry: left loop at a valid delta");
4d703a1a 2304 break;
a733cb60 2305 case OBJ_COMMIT:
a733cb60 2306 case OBJ_TREE:
a733cb60 2307 case OBJ_BLOB:
a733cb60 2308 case OBJ_TAG:
abe601bb
TR
2309 if (!base_from_cache)
2310 data = unpack_compressed_entry(p, &w_curs, curpos, size);
4d703a1a 2311 break;
1f688557 2312 default:
8eca0b47
NP
2313 data = NULL;
2314 error("unknown object type %i at offset %"PRIuMAX" in %s",
abe601bb
TR
2315 type, (uintmax_t)obj_offset, p->pack_name);
2316 }
2317
2318 /* PHASE 3: apply deltas in order */
2319
2320 /* invariants:
2321 * 'data' holds the base data, or NULL if there was corruption
2322 */
2323 while (delta_stack_nr) {
2324 void *delta_data;
2325 void *base = data;
2326 unsigned long delta_size, base_size = size;
2327 int i;
2328
2329 data = NULL;
2330
2331 if (base)
2332 add_delta_base_cache(p, obj_offset, base, base_size, type);
2333
2334 if (!base) {
2335 /*
2336 * We're probably in deep shit, but let's try to fetch
2337 * the required base anyway from another pack or loose.
2338 * This is costly but should happen only in the presence
2339 * of a corrupted pack, and is better than failing outright.
2340 */
2341 struct revindex_entry *revidx;
2342 const unsigned char *base_sha1;
2343 revidx = find_pack_revindex(p, obj_offset);
2344 if (revidx) {
2345 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2346 error("failed to read delta base object %s"
2347 " at offset %"PRIuMAX" from %s",
2348 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2349 p->pack_name);
2350 mark_bad_packed_object(p, base_sha1);
2351 base = read_object(base_sha1, &type, &base_size);
2352 }
2353 }
2354
2355 i = --delta_stack_nr;
2356 obj_offset = delta_stack[i].obj_offset;
2357 curpos = delta_stack[i].curpos;
2358 delta_size = delta_stack[i].size;
2359
2360 if (!base)
2361 continue;
2362
2363 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2364
2365 if (!delta_data) {
2366 error("failed to unpack compressed delta "
2367 "at offset %"PRIuMAX" from %s",
2368 (uintmax_t)curpos, p->pack_name);
abe601bb
TR
2369 data = NULL;
2370 continue;
2371 }
2372
2373 data = patch_delta(base, base_size,
2374 delta_data, delta_size,
2375 &size);
1ee886c1
JK
2376
2377 /*
2378 * We could not apply the delta; warn the user, but keep going.
2379 * Our failure will be noticed either in the next iteration of
2380 * the loop, or if this is the final delta, in the caller when
2381 * we return NULL. Those code paths will take care of making
2382 * a more explicit warning and retrying with another copy of
2383 * the object.
2384 */
abe601bb 2385 if (!data)
1ee886c1 2386 error("failed to apply delta");
abe601bb 2387
4b8f772c 2388 free(delta_data);
1f688557 2389 }
abe601bb
TR
2390
2391 *final_type = type;
2392 *final_size = size;
2393
03e79c88 2394 unuse_pack(&w_curs);
019d1e65
NTND
2395
2396 if (delta_stack != small_delta_stack)
2397 free(delta_stack);
2398
21666f1a 2399 return data;
1f688557
JH
2400}
2401
d079837e 2402const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 2403 uint32_t n)
9a217f2a 2404{
42873078 2405 const unsigned char *index = p->index_data;
d079837e
SP
2406 if (!index) {
2407 if (open_pack_index(p))
2408 return NULL;
2409 index = p->index_data;
2410 }
57059091 2411 if (n >= p->num_objects)
d72308e0 2412 return NULL;
74e34e1f
NP
2413 index += 4 * 256;
2414 if (p->index_version == 1) {
2415 return index + 24 * n + 4;
2416 } else {
2417 index += 8;
2418 return index + 20 * n;
2419 }
2420}
2421
99093238 2422off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
2423{
2424 const unsigned char *index = p->index_data;
2425 index += 4 * 256;
2426 if (p->index_version == 1) {
2427 return ntohl(*((uint32_t *)(index + 24 * n)));
2428 } else {
2429 uint32_t off;
2430 index += 8 + p->num_objects * (20 + 4);
2431 off = ntohl(*((uint32_t *)(index + 4 * n)));
2432 if (!(off & 0x80000000))
2433 return off;
2434 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2435 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2436 ntohl(*((uint32_t *)(index + 4)));
2437 }
9a217f2a
JH
2438}
2439
c4001d92 2440off_t find_pack_entry_one(const unsigned char *sha1,
43057304 2441 struct packed_git *p)
1f688557 2442{
42873078 2443 const uint32_t *level1_ofs = p->index_data;
42873078 2444 const unsigned char *index = p->index_data;
628522ec
JH
2445 unsigned hi, lo, stride;
2446 static int use_lookup = -1;
2447 static int debug_lookup = -1;
2448
2449 if (debug_lookup < 0)
2450 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 2451
d079837e
SP
2452 if (!index) {
2453 if (open_pack_index(p))
2454 return 0;
2455 level1_ofs = p->index_data;
2456 index = p->index_data;
2457 }
74e34e1f
NP
2458 if (p->index_version > 1) {
2459 level1_ofs += 2;
2460 index += 8;
2461 }
42873078 2462 index += 4 * 256;
74e34e1f
NP
2463 hi = ntohl(level1_ofs[*sha1]);
2464 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
2465 if (p->index_version > 1) {
2466 stride = 20;
2467 } else {
2468 stride = 24;
2469 index += 4;
2470 }
2471
2472 if (debug_lookup)
6e1c2344 2473 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
2474 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2475
2476 if (use_lookup < 0)
2477 use_lookup = !!getenv("GIT_USE_LOOKUP");
2478 if (use_lookup) {
2479 int pos = sha1_entry_pos(index, stride, 0,
2480 lo, hi, p->num_objects, sha1);
2481 if (pos < 0)
2482 return 0;
2483 return nth_packed_object_offset(p, pos);
2484 }
1f688557
JH
2485
2486 do {
74e34e1f 2487 unsigned mi = (lo + hi) / 2;
628522ec
JH
2488 int cmp = hashcmp(index + mi * stride, sha1);
2489
2490 if (debug_lookup)
2491 printf("lo %u hi %u rg %u mi %u\n",
2492 lo, hi, hi - lo, mi);
43057304 2493 if (!cmp)
74e34e1f 2494 return nth_packed_object_offset(p, mi);
1f688557
JH
2495 if (cmp > 0)
2496 hi = mi;
2497 else
2498 lo = mi+1;
2499 } while (lo < hi);
2500 return 0;
2501}
2502
4c080182 2503int is_pack_valid(struct packed_git *p)
d131b7af
SP
2504{
2505 /* An already open pack is known to be valid. */
2506 if (p->pack_fd != -1)
2507 return 1;
2508
2509 /* If the pack has one window completely covering the
2510 * file size, the pack is known to be valid even if
2511 * the descriptor is not currently open.
2512 */
2513 if (p->windows) {
2514 struct pack_window *w = p->windows;
2515
2516 if (!w->offset && w->len == p->pack_size)
2517 return 1;
2518 }
2519
2520 /* Force the pack to open to prove its valid. */
2521 return !open_packed_git(p);
2522}
2523
95099731
NTND
2524static int fill_pack_entry(const unsigned char *sha1,
2525 struct pack_entry *e,
2526 struct packed_git *p)
2527{
2528 off_t offset;
2529
2530 if (p->num_bad_objects) {
2531 unsigned i;
2532 for (i = 0; i < p->num_bad_objects; i++)
2533 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2534 return 0;
2535 }
2536
2537 offset = find_pack_entry_one(sha1, p);
2538 if (!offset)
2539 return 0;
2540
2541 /*
2542 * We are about to tell the caller where they can locate the
2543 * requested object. We better make sure the packfile is
2544 * still here and can be accessed before supplying that
2545 * answer, as it may have been deleted since the index was
2546 * loaded!
2547 */
319b678a 2548 if (!is_pack_valid(p))
95099731 2549 return 0;
95099731
NTND
2550 e->offset = offset;
2551 e->p = p;
2552 hashcpy(e->sha1, sha1);
2553 return 1;
2554}
2555
d40d535b
MH
2556/*
2557 * Iff a pack file contains the object named by sha1, return true and
2558 * store its location to e.
2559 */
4d6acb70 2560static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557
JH
2561{
2562 struct packed_git *p;
43057304 2563
1f688557 2564 prepare_packed_git();
f7c22cc6
NP
2565 if (!packed_git)
2566 return 0;
1f688557 2567
c01f51cc
NTND
2568 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2569 return 1;
8eca0b47 2570
c01f51cc 2571 for (p = packed_git; p; p = p->next) {
1b1005d1
MH
2572 if (p == last_found_pack)
2573 continue; /* we already checked this one */
f7c22cc6 2574
1b1005d1
MH
2575 if (fill_pack_entry(sha1, e, p)) {
2576 last_found_pack = p;
2577 return 1;
2578 }
c01f51cc 2579 }
1f688557
JH
2580 return 0;
2581}
2582
a6080a0a 2583struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
2584 struct packed_git *packs)
2585{
2586 struct packed_git *p;
bf592c50
DB
2587
2588 for (p = packs; p; p = p->next) {
43057304 2589 if (find_pack_entry_one(sha1, p))
bf592c50
DB
2590 return p;
2591 }
2592 return NULL;
21666f1a 2593
bf592c50
DB
2594}
2595
052fe5ea 2596static int sha1_loose_object_info(const unsigned char *sha1,
46f03448
KN
2597 struct object_info *oi,
2598 int flags)
65c2e0c3 2599{
46f03448
KN
2600 int status = 0;
2601 unsigned long mapsize;
65c2e0c3 2602 void *map;
ef49a7a0 2603 git_zstream stream;
d65a16f6 2604 char hdr[32];
46f03448 2605 struct strbuf hdrbuf = STRBUF_INIT;
65c2e0c3 2606
5d642e75
JK
2607 if (oi->delta_base_sha1)
2608 hashclr(oi->delta_base_sha1);
2609
052fe5ea
JK
2610 /*
2611 * If we don't care about type or size, then we don't
4ef8d1dd
JH
2612 * need to look inside the object at all. Note that we
2613 * do not optimize out the stat call, even if the
2614 * caller doesn't care about the disk-size, since our
2615 * return value implicitly indicates whether the
2616 * object even exists.
052fe5ea 2617 */
46f03448 2618 if (!oi->typep && !oi->typename && !oi->sizep) {
4ef8d1dd
JH
2619 struct stat st;
2620 if (stat_sha1_file(sha1, &st) < 0)
2621 return -1;
2622 if (oi->disk_sizep)
23c339c0 2623 *oi->disk_sizep = st.st_size;
052fe5ea
JK
2624 return 0;
2625 }
2626
bb6b8e4f 2627 map = map_sha1_file(sha1, &mapsize);
f0df4ed5 2628 if (!map)
dbea72a8 2629 return -1;
23c339c0
JK
2630 if (oi->disk_sizep)
2631 *oi->disk_sizep = mapsize;
46f03448
KN
2632 if ((flags & LOOKUP_UNKNOWN_OBJECT)) {
2633 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2634 status = error("unable to unpack %s header with --allow-unknown-type",
2635 sha1_to_hex(sha1));
2636 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
36e4d74a
JH
2637 status = error("unable to unpack %s header",
2638 sha1_to_hex(sha1));
46f03448
KN
2639 if (status < 0)
2640 ; /* Do nothing */
2641 else if (hdrbuf.len) {
2642 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2643 status = error("unable to parse %s header with --allow-unknown-type",
2644 sha1_to_hex(sha1));
2645 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
36e4d74a 2646 status = error("unable to parse %s header", sha1_to_hex(sha1));
39c68542 2647 git_inflate_end(&stream);
65c2e0c3 2648 munmap(map, mapsize);
46f03448 2649 if (status && oi->typep)
23c339c0 2650 *oi->typep = status;
46f03448 2651 strbuf_release(&hdrbuf);
052fe5ea 2652 return 0;
65c2e0c3
JH
2653}
2654
de7b5d62 2655int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
f0df4ed5 2656{
c4d9986f 2657 struct cached_object *co;
f0df4ed5 2658 struct pack_entry e;
5b086407 2659 int rtype;
46f03448 2660 enum object_type real_type;
1f7117ef 2661 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
f0df4ed5 2662
1f7117ef 2663 co = find_cached_object(real);
c4d9986f 2664 if (co) {
5b086407
JK
2665 if (oi->typep)
2666 *(oi->typep) = co->type;
9a490590
JH
2667 if (oi->sizep)
2668 *(oi->sizep) = co->size;
161f00e7
JK
2669 if (oi->disk_sizep)
2670 *(oi->disk_sizep) = 0;
5d642e75
JK
2671 if (oi->delta_base_sha1)
2672 hashclr(oi->delta_base_sha1);
46f03448
KN
2673 if (oi->typename)
2674 strbuf_addstr(oi->typename, typename(co->type));
9a490590 2675 oi->whence = OI_CACHED;
5b086407 2676 return 0;
c4d9986f
NTND
2677 }
2678
1f7117ef 2679 if (!find_pack_entry(real, &e)) {
ddd63e64 2680 /* Most likely it's a loose object. */
46f03448 2681 if (!sha1_loose_object_info(real, oi, flags)) {
9a490590 2682 oi->whence = OI_LOOSE;
5b086407 2683 return 0;
9a490590 2684 }
ddd63e64
SG
2685
2686 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 2687 reprepare_packed_git();
1f7117ef 2688 if (!find_pack_entry(real, &e))
052fe5ea 2689 return -1;
f0df4ed5 2690 }
3d77d877 2691
46f03448
KN
2692 /*
2693 * packed_object_info() does not follow the delta chain to
2694 * find out the real type, unless it is given oi->typep.
2695 */
2696 if (oi->typename && !oi->typep)
2697 oi->typep = &real_type;
2698
23c339c0 2699 rtype = packed_object_info(e.p, e.offset, oi);
412916ee 2700 if (rtype < 0) {
1f7117ef 2701 mark_bad_packed_object(e.p, real);
46f03448
KN
2702 if (oi->typep == &real_type)
2703 oi->typep = NULL;
1f7117ef 2704 return sha1_object_info_extended(real, oi, 0);
5266d369
JH
2705 } else if (in_delta_base_cache(e.p, e.offset)) {
2706 oi->whence = OI_DBCACHED;
9a490590
JH
2707 } else {
2708 oi->whence = OI_PACKED;
2709 oi->u.packed.offset = e.offset;
2710 oi->u.packed.pack = e.p;
2711 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2712 rtype == OBJ_OFS_DELTA);
3d77d877 2713 }
46f03448
KN
2714 if (oi->typename)
2715 strbuf_addstr(oi->typename, typename(*oi->typep));
2716 if (oi->typep == &real_type)
2717 oi->typep = NULL;
3d77d877 2718
5b086407 2719 return 0;
f0df4ed5
JS
2720}
2721
3fc0dca9 2722/* returns enum object_type or negative */
9a490590
JH
2723int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2724{
5b086407 2725 enum object_type type;
d099b717 2726 struct object_info oi = {NULL};
9a490590 2727
5b086407 2728 oi.typep = &type;
9a490590 2729 oi.sizep = sizep;
de7b5d62 2730 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
5b086407
JK
2731 return -1;
2732 return type;
9a490590
JH
2733}
2734
21666f1a
NP
2735static void *read_packed_sha1(const unsigned char *sha1,
2736 enum object_type *type, unsigned long *size)
1f688557
JH
2737{
2738 struct pack_entry e;
8eca0b47 2739 void *data;
1f688557 2740
cd673c1f 2741 if (!find_pack_entry(sha1, &e))
1f688557 2742 return NULL;
8eca0b47
NP
2743 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2744 if (!data) {
2745 /*
2746 * We're probably in deep shit, but let's try to fetch
2747 * the required object anyway from another pack or loose.
2748 * This should happen only in the presence of a corrupted
2749 * pack, and is better than failing outright.
2750 */
2751 error("failed to read object %s at offset %"PRIuMAX" from %s",
2752 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2753 mark_bad_packed_object(e.p, sha1);
ac939109 2754 data = read_object(sha1, type, size);
8eca0b47
NP
2755 }
2756 return data;
1f688557
JH
2757}
2758
21666f1a
NP
2759int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2760 unsigned char *sha1)
d66b37bb
JH
2761{
2762 struct cached_object *co;
2763
21666f1a 2764 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2765 if (has_sha1_file(sha1) || find_cached_object(sha1))
2766 return 0;
c7353967 2767 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
d66b37bb
JH
2768 co = &cached_objects[cached_object_nr++];
2769 co->size = len;
21666f1a 2770 co->type = type;
efa13f7b
JH
2771 co->buf = xmalloc(len);
2772 memcpy(co->buf, buf, len);
d66b37bb
JH
2773 hashcpy(co->sha1, sha1);
2774 return 0;
2775}
2776
c2c5b270
CC
2777static void *read_object(const unsigned char *sha1, enum object_type *type,
2778 unsigned long *size)
0fcfd160
LT
2779{
2780 unsigned long mapsize;
2781 void *map, *buf;
d66b37bb
JH
2782 struct cached_object *co;
2783
2784 co = find_cached_object(sha1);
2785 if (co) {
21666f1a 2786 *type = co->type;
d66b37bb 2787 *size = co->size;
182af834 2788 return xmemdupz(co->buf, co->size);
d66b37bb 2789 }
0fcfd160 2790
8276c007
PE
2791 buf = read_packed_sha1(sha1, type, size);
2792 if (buf)
2793 return buf;
bb6b8e4f 2794 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2795 if (map) {
7efbff75 2796 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2797 munmap(map, mapsize);
2798 return buf;
2799 }
637cdd9d 2800 reprepare_packed_git();
8276c007 2801 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2802}
2803
b6c4cecc
JH
2804/*
2805 * This function dies on corrupt objects; the callers who want to
2806 * deal with them should arrange to call read_object() and give error
2807 * messages themselves.
2808 */
5bf29b95
JH
2809void *read_sha1_file_extended(const unsigned char *sha1,
2810 enum object_type *type,
2811 unsigned long *size,
2812 unsigned flag)
ac939109 2813{
3ba7a065 2814 void *data;
b6c4cecc 2815 const struct packed_git *p;
bf93eea0 2816 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
b6c4cecc 2817
3ba7a065
JH
2818 errno = 0;
2819 data = read_object(repl, type, size);
4bbf5a26 2820 if (data)
b6c4cecc 2821 return data;
68095570 2822
25f3af3f 2823 if (errno && errno != ENOENT)
3ba7a065
JH
2824 die_errno("failed to read object %s", sha1_to_hex(sha1));
2825
68095570 2826 /* die if we replaced an object with one that does not exist */
b6c4cecc 2827 if (repl != sha1)
68095570
CC
2828 die("replacement %s not found for %s",
2829 sha1_to_hex(repl), sha1_to_hex(sha1));
2830
b6c4cecc 2831 if (has_loose_object(repl)) {
30d6c6ea
MH
2832 const char *path = sha1_file_name(sha1);
2833
b6c4cecc
JH
2834 die("loose object %s (stored in %s) is corrupt",
2835 sha1_to_hex(repl), path);
e8b15e61 2836 }
68095570 2837
b6c4cecc
JH
2838 if ((p = has_packed_and_bad(repl)) != NULL)
2839 die("packed object %s (stored in %s) is corrupt",
2840 sha1_to_hex(repl), p->pack_name);
f5552aee 2841
b6c4cecc 2842 return NULL;
ac939109
NP
2843}
2844
40469ee9 2845void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2846 const char *required_type_name,
40469ee9
JH
2847 unsigned long *size,
2848 unsigned char *actual_sha1_return)
f4913f91 2849{
21666f1a 2850 enum object_type type, required_type;
f4913f91
JH
2851 void *buffer;
2852 unsigned long isize;
40469ee9 2853 unsigned char actual_sha1[20];
f4913f91 2854
21666f1a 2855 required_type = type_from_string(required_type_name);
e702496e 2856 hashcpy(actual_sha1, sha1);
40469ee9
JH
2857 while (1) {
2858 int ref_length = -1;
2859 const char *ref_type = NULL;
f4913f91 2860
21666f1a 2861 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2862 if (!buffer)
2863 return NULL;
21666f1a 2864 if (type == required_type) {
40469ee9
JH
2865 *size = isize;
2866 if (actual_sha1_return)
e702496e 2867 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2868 return buffer;
2869 }
2870 /* Handle references */
21666f1a 2871 else if (type == OBJ_COMMIT)
40469ee9 2872 ref_type = "tree ";
21666f1a 2873 else if (type == OBJ_TAG)
40469ee9
JH
2874 ref_type = "object ";
2875 else {
2876 free(buffer);
2877 return NULL;
2878 }
2879 ref_length = strlen(ref_type);
f4913f91 2880
50974ec9
MK
2881 if (ref_length + 40 > isize ||
2882 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2883 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2884 free(buffer);
2885 return NULL;
2886 }
1cf58e72 2887 free(buffer);
40469ee9
JH
2888 /* Now we have the ID of the referred-to object in
2889 * actual_sha1. Check again. */
f4913f91 2890 }
f4913f91
JH
2891}
2892
ce9fbf16 2893static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2894 const char *type, unsigned char *sha1,
d65a16f6 2895 char *hdr, int *hdrlen)
d410c0f5 2896{
9126f009 2897 git_SHA_CTX c;
d410c0f5
JH
2898
2899 /* Generate the header */
d65a16f6 2900 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
2901
2902 /* Sha1.. */
9126f009
NP
2903 git_SHA1_Init(&c);
2904 git_SHA1_Update(&c, hdr, *hdrlen);
2905 git_SHA1_Update(&c, buf, len);
2906 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2907}
2908
230f1322 2909/*
5a688fe4
JH
2910 * Move the just written object into its final resting place.
2911 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2912 * "moving" is only a part of what it does, when no patch between
2913 * master to pu changes the call sites of this function.
230f1322 2914 */
839837b9 2915int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322 2916{
e32c0a9c 2917 int ret = 0;
5a688fe4 2918
348df166 2919 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
2920 goto try_rename;
2921 else if (link(tmpfile, filename))
e32c0a9c 2922 ret = errno;
7ebb6fca
LT
2923
2924 /*
2925 * Coda hack - coda doesn't like cross-directory links,
2926 * so we fall back to a rename, which will mean that it
2927 * won't be able to check collisions, but that's not a
2928 * big deal.
2929 *
2930 * The same holds for FAT formatted media.
2931 *
3be1f18e 2932 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
2933 * left to unlink.
2934 */
2935 if (ret && ret != EEXIST) {
be66a6c4 2936 try_rename:
7ebb6fca 2937 if (!rename(tmpfile, filename))
3be1f18e 2938 goto out;
9e48b389 2939 ret = errno;
230f1322 2940 }
691f1a28 2941 unlink_or_warn(tmpfile);
230f1322
LT
2942 if (ret) {
2943 if (ret != EEXIST) {
82247e9b 2944 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
230f1322
LT
2945 }
2946 /* FIXME!!! Collision check here ? */
2947 }
2948
3be1f18e 2949out:
5256b006 2950 if (adjust_shared_perm(filename))
5a688fe4 2951 return error("unable to set permission to '%s'", filename);
230f1322
LT
2952 return 0;
2953}
2954
4d548150
LT
2955static int write_buffer(int fd, const void *buf, size_t len)
2956{
d34cf19b 2957 if (write_in_full(fd, buf, len) < 0)
93822c22 2958 return error("file write error (%s)", strerror(errno));
4d548150
LT
2959 return 0;
2960}
2961
ce9fbf16 2962int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2963 unsigned char *sha1)
2964{
d65a16f6 2965 char hdr[32];
abdc3fc8
RS
2966 int hdrlen;
2967 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2968 return 0;
2969}
2970
e9039dd3
LT
2971/* Finalize a file on disk, and close it. */
2972static void close_sha1_file(int fd)
2973{
aafe9fba
LT
2974 if (fsync_object_files)
2975 fsync_or_die(fd, "sha1 file");
e9039dd3 2976 if (close(fd) != 0)
d824cbba 2977 die_errno("error when closing sha1 file");
e9039dd3
LT
2978}
2979
5723fe7e
LT
2980/* Size of directory component, including the ending '/' */
2981static inline int directory_size(const char *filename)
2982{
2983 const char *s = strrchr(filename, '/');
2984 if (!s)
2985 return 0;
2986 return s - filename + 1;
2987}
2988
2989/*
2990 * This creates a temporary file in the same directory as the final
2991 * 'filename'
2992 *
2993 * We want to avoid cross-directory filename renames, because those
2994 * can have problems on various filesystems (FAT, NFS, Coda).
2995 */
2996static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2997{
2998 int fd, dirlen = directory_size(filename);
2999
3000 if (dirlen + 20 > bufsiz) {
3001 errno = ENAMETOOLONG;
3002 return -1;
3003 }
3004 memcpy(buffer, filename, dirlen);
3005 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
5256b006 3006 fd = git_mkstemp_mode(buffer, 0444);
cbacbf4e 3007 if (fd < 0 && dirlen && errno == ENOENT) {
5723fe7e 3008 /* Make sure the directory exists */
6ff6af62 3009 memcpy(buffer, filename, dirlen);
5723fe7e 3010 buffer[dirlen-1] = 0;
b2476a60
JH
3011 if (mkdir(buffer, 0777) && errno != EEXIST)
3012 return -1;
3013 if (adjust_shared_perm(buffer))
5723fe7e
LT
3014 return -1;
3015
3016 /* Try again */
3017 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
5256b006 3018 fd = git_mkstemp_mode(buffer, 0444);
5723fe7e
LT
3019 }
3020 return fd;
3021}
3022
bbac7311 3023static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 3024 const void *buf, unsigned long len, time_t mtime)
0fcfd160 3025{
915308b1 3026 int fd, ret;
9892beba 3027 unsigned char compressed[4096];
ef49a7a0 3028 git_zstream stream;
748af44c
NP
3029 git_SHA_CTX c;
3030 unsigned char parano_sha1[20];
ab1900a3 3031 static char tmp_file[PATH_MAX];
30d6c6ea 3032 const char *filename = sha1_file_name(sha1);
a44c9a5e 3033
ab1900a3 3034 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
aac17941 3035 if (fd < 0) {
35243577 3036 if (errno == EACCES)
82247e9b 3037 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
916d081b 3038 else
5eaeda70 3039 return error("unable to create temporary file: %s", strerror(errno));
aac17941
LT
3040 }
3041
0fcfd160 3042 /* Set it up */
55bb5c91 3043 git_deflate_init(&stream, zlib_compression_level);
0fcfd160 3044 stream.next_out = compressed;
9892beba 3045 stream.avail_out = sizeof(compressed);
748af44c 3046 git_SHA1_Init(&c);
a44c9a5e
LT
3047
3048 /* First header.. */
d65a16f6 3049 stream.next_in = (unsigned char *)hdr;
a44c9a5e 3050 stream.avail_in = hdrlen;
55bb5c91
JH
3051 while (git_deflate(&stream, 0) == Z_OK)
3052 ; /* nothing */
748af44c 3053 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
3054
3055 /* Then the data itself.. */
c00e657d 3056 stream.next_in = (void *)buf;
a44c9a5e 3057 stream.avail_in = len;
9892beba 3058 do {
748af44c 3059 unsigned char *in0 = stream.next_in;
55bb5c91 3060 ret = git_deflate(&stream, Z_FINISH);
748af44c 3061 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
3062 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3063 die("unable to write sha1 file");
3064 stream.next_out = compressed;
3065 stream.avail_out = sizeof(compressed);
3066 } while (ret == Z_OK);
3067
ac54c277
LT
3068 if (ret != Z_STREAM_END)
3069 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
55bb5c91 3070 ret = git_deflate_end_gently(&stream);
ac54c277
LT
3071 if (ret != Z_OK)
3072 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
3073 git_SHA1_Final(parano_sha1, &c);
3074 if (hashcmp(sha1, parano_sha1) != 0)
3075 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 3076
e9039dd3 3077 close_sha1_file(fd);
0fcfd160 3078
bbac7311
NP
3079 if (mtime) {
3080 struct utimbuf utb;
3081 utb.actime = mtime;
3082 utb.modtime = mtime;
ab1900a3 3083 if (utime(tmp_file, &utb) < 0)
bbac7311 3084 warning("failed utime() on %s: %s",
ab1900a3 3085 tmp_file, strerror(errno));
bbac7311
NP
3086 }
3087
ab1900a3 3088 return move_temp_to_file(tmp_file, filename);
0fcfd160 3089}
8237b185 3090
33d4221c
JK
3091static int freshen_loose_object(const unsigned char *sha1)
3092{
3093 return check_and_freshen(sha1, 1);
3094}
3095
3096static int freshen_packed_object(const unsigned char *sha1)
3097{
3098 struct pack_entry e;
ee1c6c34
JK
3099 if (!find_pack_entry(sha1, &e))
3100 return 0;
3101 if (e.p->freshened)
3102 return 1;
3103 if (!freshen_file(e.p->pack_name))
3104 return 0;
3105 e.p->freshened = 1;
3106 return 1;
33d4221c
JK
3107}
3108
1427a7ff 3109int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
bbac7311 3110{
bbac7311
NP
3111 char hdr[32];
3112 int hdrlen;
3113
3114 /* Normally if we have it in the pack then we do not bother writing
3115 * it out into .git/objects/??/?{38} file.
3116 */
3117 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
b5f52f37 3118 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
bbac7311
NP
3119 return 0;
3120 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3121}
3122
0c3db67c
ES
3123int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3124 unsigned char *sha1, unsigned flags)
3125{
3126 char *header;
3127 int hdrlen, status = 0;
3128
3129 /* type string, SP, %lu of the length plus NUL must fit this */
3130 header = xmalloc(strlen(type) + 32);
3131 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3132
3133 if (!(flags & HASH_WRITE_OBJECT))
3134 goto cleanup;
051086b9 3135 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
0c3db67c
ES
3136 goto cleanup;
3137 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3138
3139cleanup:
3140 free(header);
3141 return status;
3142}
3143
bbac7311
NP
3144int force_object_loose(const unsigned char *sha1, time_t mtime)
3145{
bbac7311
NP
3146 void *buf;
3147 unsigned long len;
3148 enum object_type type;
3149 char hdr[32];
3150 int hdrlen;
1fb23e65 3151 int ret;
bbac7311 3152
c529d75a 3153 if (has_loose_object(sha1))
bbac7311
NP
3154 return 0;
3155 buf = read_packed_sha1(sha1, &type, &len);
3156 if (!buf)
3157 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3158 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1fb23e65
BS
3159 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3160 free(buf);
3161
3162 return ret;
bbac7311
NP
3163}
3164
bf592c50
DB
3165int has_pack_index(const unsigned char *sha1)
3166{
3167 struct stat st;
3168 if (stat(sha1_pack_index_name(sha1), &st))
3169 return 0;
3170 return 1;
bf592c50
DB
3171}
3172
cd673c1f
JH
3173int has_sha1_pack(const unsigned char *sha1)
3174{
3175 struct pack_entry e;
3176 return find_pack_entry(sha1, &e);
3177}
3178
0eeb077b 3179int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
8237b185 3180{
1f688557
JH
3181 struct pack_entry e;
3182
cd673c1f 3183 if (find_pack_entry(sha1, &e))
1f688557 3184 return 1;
45e8a748
JK
3185 if (has_loose_object(sha1))
3186 return 1;
0eeb077b
JK
3187 if (flags & HAS_SHA1_QUICK)
3188 return 0;
45e8a748
JK
3189 reprepare_packed_git();
3190 return find_pack_entry(sha1, &e);
8237b185 3191}
74400e71 3192
c879daa2
NTND
3193static void check_tree(const void *buf, size_t size)
3194{
3195 struct tree_desc desc;
3196 struct name_entry entry;
3197
3198 init_tree_desc(&desc, buf, size);
3199 while (tree_entry(&desc, &entry))
3200 /* do nothing
3201 * tree_entry() will die() on malformed entries */
3202 ;
3203}
3204
3205static void check_commit(const void *buf, size_t size)
3206{
3207 struct commit c;
3208 memset(&c, 0, sizeof(c));
3209 if (parse_commit_buffer(&c, buf, size))
3210 die("corrupt commit");
3211}
3212
3213static void check_tag(const void *buf, size_t size)
3214{
3215 struct tag t;
3216 memset(&t, 0, sizeof(t));
3217 if (parse_tag_buffer(&t, buf, size))
3218 die("corrupt tag");
3219}
3220
43df4f86 3221static int index_mem(unsigned char *sha1, void *buf, size_t size,
c4ce46fc
JH
3222 enum object_type type,
3223 const char *path, unsigned flags)
e7332f96 3224{
6c510bee 3225 int ret, re_allocated = 0;
c4ce46fc 3226 int write_object = flags & HASH_WRITE_OBJECT;
74400e71 3227
7672db20 3228 if (!type)
edaec3fb 3229 type = OBJ_BLOB;
6c510bee
LT
3230
3231 /*
3232 * Convert blobs to git internal format
3233 */
43df4f86 3234 if ((type == OBJ_BLOB) && path) {
f285a2d7 3235 struct strbuf nbuf = STRBUF_INIT;
21e5ad50 3236 if (convert_to_git(path, buf, size, &nbuf,
5e12e78e 3237 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
b315c5c0 3238 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
3239 re_allocated = 1;
3240 }
3241 }
c4ce46fc 3242 if (flags & HASH_FORMAT_CHECK) {
c879daa2
NTND
3243 if (type == OBJ_TREE)
3244 check_tree(buf, size);
3245 if (type == OBJ_COMMIT)
3246 check_commit(buf, size);
3247 if (type == OBJ_TAG)
3248 check_tag(buf, size);
3249 }
6c510bee 3250
7672db20 3251 if (write_object)
edaec3fb 3252 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 3253 else
edaec3fb 3254 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 3255 if (re_allocated)
6c510bee 3256 free(buf);
43df4f86
DP
3257 return ret;
3258}
3259
9035d75a
SP
3260static int index_stream_convert_blob(unsigned char *sha1, int fd,
3261 const char *path, unsigned flags)
3262{
3263 int ret;
3264 const int write_object = flags & HASH_WRITE_OBJECT;
3265 struct strbuf sbuf = STRBUF_INIT;
3266
3267 assert(path);
3268 assert(would_convert_to_git_filter_fd(path));
3269
3270 convert_to_git_filter_fd(path, fd, &sbuf,
3271 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3272
3273 if (write_object)
3274 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3275 sha1);
3276 else
3277 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3278 sha1);
3279 strbuf_release(&sbuf);
3280 return ret;
3281}
3282
7b41e1e1
JH
3283static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3284 const char *path, unsigned flags)
3285{
3286 struct strbuf sbuf = STRBUF_INIT;
3287 int ret;
3288
3289 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3290 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3291 else
3292 ret = -1;
3293 strbuf_release(&sbuf);
3294 return ret;
3295}
3296
ea68b0ce
DP
3297#define SMALL_FILE_SIZE (32*1024)
3298
7b41e1e1
JH
3299static int index_core(unsigned char *sha1, int fd, size_t size,
3300 enum object_type type, const char *path,
3301 unsigned flags)
43df4f86
DP
3302{
3303 int ret;
43df4f86 3304
7b41e1e1 3305 if (!size) {
f6a1e1e2 3306 ret = index_mem(sha1, "", size, type, path, flags);
ea68b0ce
DP
3307 } else if (size <= SMALL_FILE_SIZE) {
3308 char *buf = xmalloc(size);
3309 if (size == read_in_full(fd, buf, size))
c4ce46fc 3310 ret = index_mem(sha1, buf, size, type, path, flags);
ea68b0ce
DP
3311 else
3312 ret = error("short read %s", strerror(errno));
3313 free(buf);
08bda208 3314 } else {
43df4f86 3315 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
c4ce46fc 3316 ret = index_mem(sha1, buf, size, type, path, flags);
aac17941 3317 munmap(buf, size);
08bda208 3318 }
7b41e1e1
JH
3319 return ret;
3320}
3321
4dd1fbc7 3322/*
568508e7
JH
3323 * This creates one packfile per large blob unless bulk-checkin
3324 * machinery is "plugged".
4dd1fbc7
JH
3325 *
3326 * This also bypasses the usual "convert-to-git" dance, and that is on
3327 * purpose. We could write a streaming version of the converting
3328 * functions and insert that before feeding the data to fast-import
4f22b101
JK
3329 * (or equivalent in-core API described above). However, that is
3330 * somewhat complicated, as we do not know the size of the filter
3331 * result, which we need to know beforehand when writing a git object.
3332 * Since the primary motivation for trying to stream from the working
3333 * tree file and to avoid mmaping it in core is to deal with large
3334 * binary blobs, they generally do not want to get any conversion, and
3335 * callers should avoid this code path when filters are requested.
4dd1fbc7
JH
3336 */
3337static int index_stream(unsigned char *sha1, int fd, size_t size,
3338 enum object_type type, const char *path,
3339 unsigned flags)
3340{
568508e7 3341 return index_bulk_checkin(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
3342}
3343
7b41e1e1
JH
3344int index_fd(unsigned char *sha1, int fd, struct stat *st,
3345 enum object_type type, const char *path, unsigned flags)
3346{
3347 int ret;
7b41e1e1 3348
9079ab7c
SP
3349 /*
3350 * Call xsize_t() only when needed to avoid potentially unnecessary
3351 * die() for large files.
3352 */
9035d75a
SP
3353 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3354 ret = index_stream_convert_blob(sha1, fd, path, flags);
3355 else if (!S_ISREG(st->st_mode))
7b41e1e1 3356 ret = index_pipe(sha1, fd, type, path, flags);
9079ab7c 3357 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
7ce7c760 3358 (path && would_convert_to_git(path)))
9079ab7c
SP
3359 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3360 flags);
4dd1fbc7 3361 else
9079ab7c
SP
3362 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3363 flags);
43df4f86 3364 close(fd);
aac17941 3365 return ret;
74400e71 3366}
ec1fcc16 3367
c4ce46fc 3368int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
ec1fcc16
JH
3369{
3370 int fd;
b760d3aa 3371 struct strbuf sb = STRBUF_INIT;
ec1fcc16
JH
3372
3373 switch (st->st_mode & S_IFMT) {
3374 case S_IFREG:
3375 fd = open(path, O_RDONLY);
3376 if (fd < 0)
3377 return error("open(\"%s\"): %s", path,
3378 strerror(errno));
c4ce46fc 3379 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
ec1fcc16
JH
3380 return error("%s: failed to insert into database",
3381 path);
3382 break;
3383 case S_IFLNK:
b760d3aa 3384 if (strbuf_readlink(&sb, path, st->st_size)) {
ec1fcc16 3385 char *errstr = strerror(errno);
ec1fcc16
JH
3386 return error("readlink(\"%s\"): %s", path,
3387 errstr);
3388 }
c4ce46fc 3389 if (!(flags & HASH_WRITE_OBJECT))
b760d3aa
LT
3390 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3391 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
ec1fcc16
JH
3392 return error("%s: failed to insert into database",
3393 path);
b760d3aa 3394 strbuf_release(&sb);
ec1fcc16 3395 break;
f35a6d3b
LT
3396 case S_IFDIR:
3397 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
3398 default:
3399 return error("%s: unsupported file type", path);
3400 }
3401 return 0;
3402}
a69e5429
JH
3403
3404int read_pack_header(int fd, struct pack_header *header)
3405{
c697ad14
HO
3406 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3407 /* "eof before pack header was fully read" */
3408 return PH_ERROR_EOF;
3409
a69e5429
JH
3410 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3411 /* "protocol error (pack signature mismatch detected)" */
3412 return PH_ERROR_PACK_SIGNATURE;
3413 if (!pack_version_ok(header->hdr_version))
3414 /* "protocol error (pack version unsupported)" */
3415 return PH_ERROR_PROTOCOL;
3416 return 0;
3417}
40d52ff7
JK
3418
3419void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3420{
3421 enum object_type type = sha1_object_info(sha1, NULL);
3422 if (type < 0)
3423 die("%s is not a valid object", sha1_to_hex(sha1));
3424 if (type != expect)
3425 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3426 typename(expect));
3427}
27e1e22d
JK
3428
3429static int for_each_file_in_obj_subdir(int subdir_nr,
3430 struct strbuf *path,
3431 each_loose_object_fn obj_cb,
3432 each_loose_cruft_fn cruft_cb,
3433 each_loose_subdir_fn subdir_cb,
3434 void *data)
3435{
3436 size_t baselen = path->len;
3437 DIR *dir = opendir(path->buf);
3438 struct dirent *de;
3439 int r = 0;
3440
3441 if (!dir) {
3442 if (errno == ENOENT)
3443 return 0;
3444 return error("unable to open %s: %s", path->buf, strerror(errno));
3445 }
3446
3447 while ((de = readdir(dir))) {
3448 if (is_dot_or_dotdot(de->d_name))
3449 continue;
3450
3451 strbuf_setlen(path, baselen);
3452 strbuf_addf(path, "/%s", de->d_name);
3453
3454 if (strlen(de->d_name) == 38) {
3455 char hex[41];
3456 unsigned char sha1[20];
3457
3458 snprintf(hex, sizeof(hex), "%02x%s",
3459 subdir_nr, de->d_name);
3460 if (!get_sha1_hex(hex, sha1)) {
3461 if (obj_cb) {
3462 r = obj_cb(sha1, path->buf, data);
3463 if (r)
3464 break;
3465 }
3466 continue;
3467 }
3468 }
3469
3470 if (cruft_cb) {
3471 r = cruft_cb(de->d_name, path->buf, data);
3472 if (r)
3473 break;
3474 }
3475 }
3476 strbuf_setlen(path, baselen);
3477
3478 if (!r && subdir_cb)
3479 r = subdir_cb(subdir_nr, path->buf, data);
3480
3481 closedir(dir);
3482 return r;
3483}
3484
e6f875e0 3485int for_each_loose_file_in_objdir_buf(struct strbuf *path,
27e1e22d
JK
3486 each_loose_object_fn obj_cb,
3487 each_loose_cruft_fn cruft_cb,
3488 each_loose_subdir_fn subdir_cb,
3489 void *data)
3490{
e6f875e0 3491 size_t baselen = path->len;
27e1e22d
JK
3492 int r = 0;
3493 int i;
3494
27e1e22d 3495 for (i = 0; i < 256; i++) {
e6f875e0
JK
3496 strbuf_addf(path, "/%02x", i);
3497 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
27e1e22d 3498 subdir_cb, data);
e6f875e0 3499 strbuf_setlen(path, baselen);
27e1e22d
JK
3500 if (r)
3501 break;
3502 }
3503
e6f875e0
JK
3504 return r;
3505}
3506
3507int for_each_loose_file_in_objdir(const char *path,
3508 each_loose_object_fn obj_cb,
3509 each_loose_cruft_fn cruft_cb,
3510 each_loose_subdir_fn subdir_cb,
3511 void *data)
3512{
3513 struct strbuf buf = STRBUF_INIT;
3514 int r;
3515
3516 strbuf_addstr(&buf, path);
3517 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3518 subdir_cb, data);
27e1e22d 3519 strbuf_release(&buf);
e6f875e0 3520
27e1e22d
JK
3521 return r;
3522}
660c889e
JK
3523
3524struct loose_alt_odb_data {
3525 each_loose_object_fn *cb;
3526 void *data;
3527};
3528
3529static int loose_from_alt_odb(struct alternate_object_database *alt,
3530 void *vdata)
3531{
3532 struct loose_alt_odb_data *data = vdata;
b0a42642
JM
3533 struct strbuf buf = STRBUF_INIT;
3534 int r;
3535
3536 /* copy base not including trailing '/' */
3537 strbuf_add(&buf, alt->base, alt->name - alt->base - 1);
3538 r = for_each_loose_file_in_objdir_buf(&buf,
3539 data->cb, NULL, NULL,
3540 data->data);
3541 strbuf_release(&buf);
3542 return r;
660c889e
JK
3543}
3544
1385bb7b 3545int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
660c889e
JK
3546{
3547 struct loose_alt_odb_data alt;
3548 int r;
3549
3550 r = for_each_loose_file_in_objdir(get_object_directory(),
3551 cb, NULL, NULL, data);
3552 if (r)
3553 return r;
3554
1385bb7b
JK
3555 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)