sha1_file.c: move find_cached_object up so sha1_object_info can use it
[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"
1f688557 10#include "delta.h"
a733cb60 11#include "pack.h"
8e440259
PE
12#include "blob.h"
13#include "commit.h"
14#include "tag.h"
15#include "tree.h"
f35a6d3b 16#include "refs.h"
70f5d5d3 17#include "pack-revindex.h"
628522ec 18#include "sha1-lookup.h"
0fcfd160 19
144bde78
LT
20#ifndef O_NOATIME
21#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
22#define O_NOATIME 01000000
23#else
24#define O_NOATIME 0
25#endif
26#endif
27
e05db0fd
PR
28#ifdef NO_C99_FORMAT
29#define SZ_FMT "lu"
9e42d6a1 30static unsigned long sz_fmt(size_t s) { return (unsigned long)s; }
e05db0fd
PR
31#else
32#define SZ_FMT "zu"
9e42d6a1 33static size_t sz_fmt(size_t s) { return s; }
e05db0fd
PR
34#endif
35
96f1e58f 36const unsigned char null_sha1[20];
88cd621d 37
f2e872aa 38static int git_open_noatime(const char *name, struct packed_git *p);
4865d2b6 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 = {
55 EMPTY_TREE_SHA1_BIN,
56 OBJ_TREE,
57 "",
58 0
59};
60
61static struct cached_object *find_cached_object(const unsigned char *sha1)
62{
63 int i;
64 struct cached_object *co = cached_objects;
65
66 for (i = 0; i < cached_object_nr; i++, co++) {
67 if (!hashcmp(co->sha1, sha1))
68 return co;
69 }
70 if (!hashcmp(sha1, empty_tree.sha1))
71 return &empty_tree;
72 return NULL;
73}
74
b2cb9425
JH
75int safe_create_leading_directories(char *path)
76{
8385abfd 77 char *pos = path + offset_1st_component(path);
67d42212
JR
78 struct stat st;
79
b2cb9425
JH
80 while (pos) {
81 pos = strchr(pos, '/');
82 if (!pos)
83 break;
5f0bdf50
JH
84 while (*++pos == '/')
85 ;
86 if (!*pos)
87 break;
88 *--pos = '\0';
67d42212
JR
89 if (!stat(path, &st)) {
90 /* path exists */
91 if (!S_ISDIR(st.st_mode)) {
b2cb9425 92 *pos = '/';
67d42212 93 return -3;
b2cb9425 94 }
457f06d6 95 }
67d42212
JR
96 else if (mkdir(path, 0777)) {
97 *pos = '/';
98 return -1;
99 }
457f06d6
JS
100 else if (adjust_shared_perm(path)) {
101 *pos = '/';
102 return -2;
103 }
b2cb9425
JH
104 *pos++ = '/';
105 }
106 return 0;
107}
723c31fe 108
8e21d63b
JK
109int safe_create_leading_directories_const(const char *path)
110{
111 /* path points to cache entries, so xstrdup before messing with it */
112 char *buf = xstrdup(path);
113 int result = safe_create_leading_directories(buf);
114 free(buf);
115 return result;
116}
117
ace1534d
JH
118static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
119{
120 int i;
121 for (i = 0; i < 20; i++) {
122 static char hex[] = "0123456789abcdef";
123 unsigned int val = sha1[i];
124 char *pos = pathbuf + i*2 + (i > 0);
125 *pos++ = hex[val >> 4];
126 *pos = hex[val & 0xf];
127 }
128}
129
0fcfd160
LT
130/*
131 * NOTE! This returns a statically allocated buffer, so you have to be
790296fd 132 * careful about using it. Do an "xstrdup()" if you need to save the
0fcfd160 133 * filename.
ace1534d
JH
134 *
135 * Also note that this returns the location for creating. Reading
136 * SHA1 file can happen from any alternate directory listed in the
d19938ab 137 * DB_ENVIRONMENT environment variable if it is not found in
ace1534d 138 * the primary object database.
0fcfd160
LT
139 */
140char *sha1_file_name(const unsigned char *sha1)
141{
560fb6a1
JK
142 static char buf[PATH_MAX];
143 const char *objdir;
144 int len;
0fcfd160 145
560fb6a1
JK
146 objdir = get_object_directory();
147 len = strlen(objdir);
148
149 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
150 if (len + 43 > PATH_MAX)
151 die("insanely long object directory %s", objdir);
152 memcpy(buf, objdir, len);
153 buf[len] = '/';
154 buf[len+3] = '/';
155 buf[len+42] = '\0';
156 fill_sha1_path(buf + len + 1, sha1);
157 return buf;
0fcfd160
LT
158}
159
633f43e1 160static char *sha1_get_pack_name(const unsigned char *sha1,
6eec46bd 161 char **name, char **base, const char *which)
bf592c50
DB
162{
163 static const char hex[] = "0123456789abcdef";
633f43e1 164 char *buf;
bf592c50
DB
165 int i;
166
633f43e1 167 if (!*base) {
bf592c50
DB
168 const char *sha1_file_directory = get_object_directory();
169 int len = strlen(sha1_file_directory);
633f43e1 170 *base = xmalloc(len + 60);
6eec46bd
JH
171 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
172 sha1_file_directory, which);
633f43e1 173 *name = *base + len + 11;
bf592c50
DB
174 }
175
633f43e1 176 buf = *name;
bf592c50
DB
177
178 for (i = 0; i < 20; i++) {
179 unsigned int val = *sha1++;
180 *buf++ = hex[val >> 4];
181 *buf++ = hex[val & 0xf];
182 }
a6080a0a 183
633f43e1 184 return *base;
bf592c50
DB
185}
186
633f43e1 187char *sha1_pack_name(const unsigned char *sha1)
bf592c50 188{
633f43e1 189 static char *name, *base;
bf592c50 190
6eec46bd 191 return sha1_get_pack_name(sha1, &name, &base, "pack");
633f43e1 192}
bf592c50 193
633f43e1
HO
194char *sha1_pack_index_name(const unsigned char *sha1)
195{
196 static char *name, *base;
a6080a0a 197
6eec46bd 198 return sha1_get_pack_name(sha1, &name, &base, "idx");
bf592c50
DB
199}
200
d5a63b99
JH
201struct alternate_object_database *alt_odb_list;
202static struct alternate_object_database **alt_odb_tail;
ace1534d 203
c2f493a4
MW
204static void read_info_alternates(const char * alternates, int depth);
205
ddd5d056
JH
206/*
207 * Prepare alternate object database registry.
d5a63b99
JH
208 *
209 * The variable alt_odb_list points at the list of struct
210 * alternate_object_database. The elements on this list come from
211 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
212 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
213 * whose contents is similar to that environment variable but can be
214 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
215 * contains "/the/directory/corresponding/to/.git/objects/...", while
216 * its name points just after the slash at the end of ".git/objects/"
217 * in the example above, and has enough space to hold 40-byte hex
218 * SHA1, an extra slash for the first level indirection, and the
219 * terminating NUL.
ddd5d056 220 */
c2f493a4 221static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
ace1534d 222{
1494e038 223 const char *objdir = get_object_directory();
c2f493a4
MW
224 struct alternate_object_database *ent;
225 struct alternate_object_database *alt;
226 /* 43 = 40-byte + 2 '/' + terminating NUL */
227 int pfxlen = len;
228 int entlen = pfxlen + 43;
ccfd3e99 229 int base_len = -1;
d5a63b99 230
85dadc38 231 if (!is_absolute_path(entry) && relative_base) {
c2f493a4
MW
232 /* Relative alt-odb */
233 if (base_len < 0)
234 base_len = strlen(relative_base) + 1;
235 entlen += base_len;
236 pfxlen += base_len;
237 }
238 ent = xmalloc(sizeof(*ent) + entlen);
239
85dadc38 240 if (!is_absolute_path(entry) && relative_base) {
c2f493a4
MW
241 memcpy(ent->base, relative_base, base_len - 1);
242 ent->base[base_len - 1] = '/';
243 memcpy(ent->base + base_len, entry, len);
244 }
245 else
246 memcpy(ent->base, entry, pfxlen);
247
248 ent->name = ent->base + pfxlen + 1;
249 ent->base[pfxlen + 3] = '/';
250 ent->base[pfxlen] = ent->base[entlen-1] = 0;
251
252 /* Detect cases where alternate disappeared */
90b4a71c 253 if (!is_directory(ent->base)) {
c2f493a4
MW
254 error("object directory %s does not exist; "
255 "check .git/objects/info/alternates.",
256 ent->base);
257 free(ent);
258 return -1;
259 }
260
261 /* Prevent the common mistake of listing the same
262 * thing twice, or object directory itself.
263 */
264 for (alt = alt_odb_list; alt; alt = alt->next) {
265 if (!memcmp(ent->base, alt->base, pfxlen)) {
266 free(ent);
267 return -1;
268 }
269 }
270 if (!memcmp(ent->base, objdir, pfxlen)) {
271 free(ent);
272 return -1;
273 }
274
275 /* add the alternate entry */
276 *alt_odb_tail = ent;
277 alt_odb_tail = &(ent->next);
278 ent->next = NULL;
279
280 /* recursively add alternates */
281 read_info_alternates(ent->base, depth + 1);
282
283 ent->base[pfxlen] = '/';
284
285 return 0;
286}
287
288static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
289 const char *relative_base, int depth)
290{
291 const char *cp, *last;
292
293 if (depth > 5) {
294 error("%s: ignoring alternate object stores, nesting too deep.",
295 relative_base);
296 return;
297 }
298
d5a63b99 299 last = alt;
9577e7e3
JH
300 while (last < ep) {
301 cp = last;
302 if (cp < ep && *cp == '#') {
303 while (cp < ep && *cp != sep)
304 cp++;
305 last = cp + 1;
306 continue;
307 }
c2f493a4
MW
308 while (cp < ep && *cp != sep)
309 cp++;
d5a63b99 310 if (last != cp) {
85dadc38 311 if (!is_absolute_path(last) && depth) {
c2f493a4
MW
312 error("%s: ignoring relative alternate object store %s",
313 relative_base, last);
314 } else {
315 link_alt_odb_entry(last, cp - last,
316 relative_base, depth);
1494e038 317 }
d5a63b99 318 }
9577e7e3 319 while (cp < ep && *cp == sep)
d5a63b99
JH
320 cp++;
321 last = cp;
9577e7e3 322 }
d5a63b99
JH
323}
324
c2f493a4 325static void read_info_alternates(const char * relative_base, int depth)
d5a63b99 326{
9577e7e3 327 char *map;
dc49cd76 328 size_t mapsz;
d5a63b99 329 struct stat st;
9cb18f56
JM
330 const char alt_file_name[] = "info/alternates";
331 /* Given that relative_base is no longer than PATH_MAX,
332 ensure that "path" has enough space to append "/", the
333 file name, "info/alternates", and a trailing NUL. */
334 char path[PATH_MAX + 1 + sizeof alt_file_name];
c2f493a4 335 int fd;
d5a63b99 336
9cb18f56 337 sprintf(path, "%s/%s", relative_base, alt_file_name);
f2e872aa 338 fd = git_open_noatime(path, NULL);
d5a63b99
JH
339 if (fd < 0)
340 return;
341 if (fstat(fd, &st) || (st.st_size == 0)) {
342 close(fd);
9a217f2a 343 return;
ace1534d 344 }
dc49cd76
SP
345 mapsz = xsize_t(st.st_size);
346 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 347 close(fd);
d5a63b99 348
dc49cd76 349 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
c2f493a4 350
dc49cd76 351 munmap(map, mapsz);
ace1534d
JH
352}
353
bef70b22
DB
354void add_to_alternates_file(const char *reference)
355{
356 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
acd3b9ec 357 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
bef70b22
DB
358 char *alt = mkpath("%s/objects\n", reference);
359 write_or_die(fd, alt, strlen(alt));
360 if (commit_lock_file(lock))
361 die("could not close alternates file");
362 if (alt_odb_tail)
363 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
364}
365
d79796bc
JH
366void foreach_alt_odb(alt_odb_fn fn, void *cb)
367{
368 struct alternate_object_database *ent;
369
370 prepare_alt_odb();
371 for (ent = alt_odb_list; ent; ent = ent->next)
372 if (fn(ent, cb))
373 return;
374}
375
c2f493a4
MW
376void prepare_alt_odb(void)
377{
554fe20d 378 const char *alt;
c2f493a4 379
7dc24aa5
SP
380 if (alt_odb_tail)
381 return;
382
c2f493a4
MW
383 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
384 if (!alt) alt = "";
385
c2f493a4 386 alt_odb_tail = &alt_odb_list;
80ba074f 387 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
388
389 read_info_alternates(get_object_directory(), 0);
390}
391
0f4dc14a 392static int has_loose_object_local(const unsigned char *sha1)
ace1534d 393{
ace1534d 394 char *name = sha1_file_name(sha1);
0f4dc14a
BC
395 return !access(name, F_OK);
396}
ace1534d 397
0f4dc14a
BC
398int has_loose_object_nonlocal(const unsigned char *sha1)
399{
400 struct alternate_object_database *alt;
9a217f2a 401 prepare_alt_odb();
d5a63b99 402 for (alt = alt_odb_list; alt; alt = alt->next) {
0f4dc14a 403 fill_sha1_path(alt->name, sha1);
c529d75a
LT
404 if (!access(alt->base, F_OK))
405 return 1;
ace1534d 406 }
c529d75a 407 return 0;
ace1534d
JH
408}
409
0f4dc14a
BC
410static int has_loose_object(const unsigned char *sha1)
411{
412 return has_loose_object_local(sha1) ||
413 has_loose_object_nonlocal(sha1);
414}
415
60bb8b14 416static unsigned int pack_used_ctr;
a53128b6
SP
417static unsigned int pack_mmap_calls;
418static unsigned int peak_pack_open_windows;
419static unsigned int pack_open_windows;
420static size_t peak_pack_mapped;
60bb8b14 421static size_t pack_mapped;
9a217f2a 422struct packed_git *packed_git;
1f688557 423
b79d18c9 424void pack_report(void)
a53128b6
SP
425{
426 fprintf(stderr,
e05db0fd
PR
427 "pack_report: getpagesize() = %10" SZ_FMT "\n"
428 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
429 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
430 sz_fmt(getpagesize()),
431 sz_fmt(packed_git_window_size),
432 sz_fmt(packed_git_limit));
a53128b6
SP
433 fprintf(stderr,
434 "pack_report: pack_used_ctr = %10u\n"
435 "pack_report: pack_mmap_calls = %10u\n"
436 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
437 "pack_report: pack_mapped = "
438 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
439 pack_used_ctr,
440 pack_mmap_calls,
441 pack_open_windows, peak_pack_open_windows,
9e42d6a1 442 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
443}
444
42873078 445static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
446{
447 void *idx_map;
42873078 448 struct pack_idx_header *hdr;
dc49cd76 449 size_t idx_size;
74e34e1f 450 uint32_t version, nr, i, *index;
f2e872aa 451 int fd = git_open_noatime(path, p);
1f688557 452 struct stat st;
42873078 453
1f688557
JH
454 if (fd < 0)
455 return -1;
456 if (fstat(fd, &st)) {
457 close(fd);
458 return -1;
459 }
dc49cd76 460 idx_size = xsize_t(st.st_size);
2d88451b
SP
461 if (idx_size < 4 * 256 + 20 + 20) {
462 close(fd);
463 return error("index file %s is too small", path);
464 }
c4712e45 465 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 466 close(fd);
1f688557 467
42873078
NP
468 hdr = idx_map;
469 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
470 version = ntohl(hdr->idx_version);
471 if (version < 2 || version > 2) {
472 munmap(idx_map, idx_size);
6e1c2344 473 return error("index file %s is version %"PRIu32
74e34e1f
NP
474 " and is not supported by this binary"
475 " (try upgrading GIT to a newer version)",
476 path, version);
477 }
478 } else
479 version = 1;
df1b059d 480
1f688557 481 nr = 0;
42873078 482 index = idx_map;
74e34e1f
NP
483 if (version > 1)
484 index += 2; /* skip index header */
1f688557 485 for (i = 0; i < 256; i++) {
326bf396 486 uint32_t n = ntohl(index[i]);
2d88451b
SP
487 if (n < nr) {
488 munmap(idx_map, idx_size);
df1b059d 489 return error("non-monotonic index %s", path);
2d88451b 490 }
1f688557
JH
491 nr = n;
492 }
493
74e34e1f
NP
494 if (version == 1) {
495 /*
496 * Total size:
497 * - 256 index entries 4 bytes each
498 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
499 * - 20-byte SHA1 of the packfile
500 * - 20-byte SHA1 file checksum
501 */
502 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
503 munmap(idx_map, idx_size);
eef427a0 504 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
505 }
506 } else if (version == 2) {
507 /*
508 * Minimum size:
509 * - 8 bytes of header
510 * - 256 index entries 4 bytes each
511 * - 20-byte sha1 entry * nr
512 * - 4-byte crc entry * nr
513 * - 4-byte offset entry * nr
514 * - 20-byte SHA1 of the packfile
515 * - 20-byte SHA1 file checksum
516 * And after the 4-byte offset table might be a
517 * variable sized table containing 8-byte entries
518 * for offsets larger than 2^31.
519 */
520 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
521 unsigned long max_size = min_size;
522 if (nr)
523 max_size += (nr - 1)*8;
524 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 525 munmap(idx_map, idx_size);
eef427a0 526 return error("wrong index v2 file size in %s", path);
74e34e1f 527 }
7109c889
JH
528 if (idx_size != min_size &&
529 /*
530 * make sure we can deal with large pack offsets.
531 * 31-bit signed offset won't be enough, neither
532 * 32-bit unsigned one will be.
533 */
534 (sizeof(off_t) <= 4)) {
535 munmap(idx_map, idx_size);
536 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 537 }
2d88451b 538 }
1f688557 539
74e34e1f 540 p->index_version = version;
42873078
NP
541 p->index_data = idx_map;
542 p->index_size = idx_size;
57059091 543 p->num_objects = nr;
1f688557
JH
544 return 0;
545}
546
bc8e478a 547int open_pack_index(struct packed_git *p)
d079837e
SP
548{
549 char *idx_name;
550 int ret;
551
552 if (p->index_data)
553 return 0;
554
555 idx_name = xstrdup(p->pack_name);
556 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
557 ret = check_packed_git_idx(idx_name, p);
558 free(idx_name);
559 return ret;
560}
561
11daf39b
SP
562static void scan_windows(struct packed_git *p,
563 struct packed_git **lru_p,
564 struct pack_window **lru_w,
565 struct pack_window **lru_l)
1f688557 566{
11daf39b
SP
567 struct pack_window *w, *w_l;
568
569 for (w_l = NULL, w = p->windows; w; w = w->next) {
570 if (!w->inuse_cnt) {
571 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
572 *lru_p = p;
573 *lru_w = w;
574 *lru_l = w_l;
54044bf8 575 }
54044bf8 576 }
11daf39b 577 w_l = w;
f9253394 578 }
11daf39b
SP
579}
580
d1efefa4 581static int unuse_one_window(struct packed_git *current, int keep_fd)
11daf39b
SP
582{
583 struct packed_git *p, *lru_p = NULL;
584 struct pack_window *lru_w = NULL, *lru_l = NULL;
585
586 if (current)
587 scan_windows(current, &lru_p, &lru_w, &lru_l);
588 for (p = packed_git; p; p = p->next)
589 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
590 if (lru_p) {
591 munmap(lru_w->base, lru_w->len);
592 pack_mapped -= lru_w->len;
593 if (lru_l)
594 lru_l->next = lru_w->next;
595 else {
596 lru_p->windows = lru_w->next;
d1efefa4 597 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
54044bf8
SP
598 close(lru_p->pack_fd);
599 lru_p->pack_fd = -1;
600 }
601 }
602 free(lru_w);
a53128b6 603 pack_open_windows--;
54044bf8
SP
604 return 1;
605 }
606 return 0;
f9253394
JH
607}
608
d1efefa4 609void release_pack_memory(size_t need, int fd)
97bfeb34
SP
610{
611 size_t cur = pack_mapped;
d1efefa4 612 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
97bfeb34
SP
613 ; /* nothing */
614}
615
58ecbd5e
JN
616void *xmmap(void *start, size_t length,
617 int prot, int flags, int fd, off_t offset)
618{
619 void *ret = mmap(start, length, prot, flags, fd, offset);
620 if (ret == MAP_FAILED) {
621 if (!length)
622 return NULL;
623 release_pack_memory(length, fd);
624 ret = mmap(start, length, prot, flags, fd, offset);
625 if (ret == MAP_FAILED)
626 die_errno("Out of memory? mmap failed");
627 }
628 return ret;
629}
630
c9ced051
SP
631void close_pack_windows(struct packed_git *p)
632{
633 while (p->windows) {
634 struct pack_window *w = p->windows;
635
636 if (w->inuse_cnt)
637 die("pack '%s' still has open windows to it",
638 p->pack_name);
639 munmap(w->base, w->len);
640 pack_mapped -= w->len;
641 pack_open_windows--;
642 p->windows = w->next;
643 free(w);
644 }
645}
646
03e79c88 647void unuse_pack(struct pack_window **w_cursor)
f9253394 648{
03e79c88
SP
649 struct pack_window *w = *w_cursor;
650 if (w) {
651 w->inuse_cnt--;
652 *w_cursor = NULL;
653 }
1f688557
JH
654}
655
fa5fc15d
SP
656void close_pack_index(struct packed_git *p)
657{
658 if (p->index_data) {
659 munmap((void *)p->index_data, p->index_size);
660 p->index_data = NULL;
661 }
662}
663
c74faea1
NP
664/*
665 * This is used by git-repack in case a newly created pack happens to
666 * contain the same set of objects as an existing one. In that case
667 * the resulting file might be different even if its name would be the
668 * same. It is best to close any reference to the old pack before it is
669 * replaced on disk. Of course no index pointers nor windows for given pack
670 * must subsist at this point. If ever objects from this pack are requested
671 * again, the new version of the pack will be reinitialized through
672 * reprepare_packed_git().
673 */
674void free_pack_by_name(const char *pack_name)
675{
676 struct packed_git *p, **pp = &packed_git;
677
678 while (*pp) {
679 p = *pp;
680 if (strcmp(pack_name, p->pack_name) == 0) {
fa3a0c94 681 clear_delta_base_cache();
c74faea1
NP
682 close_pack_windows(p);
683 if (p->pack_fd != -1)
684 close(p->pack_fd);
fa5fc15d 685 close_pack_index(p);
c74faea1
NP
686 free(p->bad_object_sha1);
687 *pp = p->next;
688 free(p);
689 return;
690 }
691 pp = &p->next;
692 }
693}
694
3cf8b462
SP
695/*
696 * Do not call this directly as this leaks p->pack_fd on error return;
697 * call open_packed_git() instead.
698 */
699static int open_packed_git_1(struct packed_git *p)
1f688557 700{
9bc879c1
SP
701 struct stat st;
702 struct pack_header hdr;
703 unsigned char sha1[20];
704 unsigned char *idx_sha1;
2c039da8 705 long fd_flag;
9bc879c1 706
d079837e
SP
707 if (!p->index_data && open_pack_index(p))
708 return error("packfile %s index unavailable", p->pack_name);
709
f2e872aa 710 p->pack_fd = git_open_noatime(p->pack_name, p);
9bc879c1 711 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 712 return -1;
9bc879c1
SP
713
714 /* If we created the struct before we had the pack we lack size. */
bf592c50 715 if (!p->pack_size) {
bf592c50 716 if (!S_ISREG(st.st_mode))
072db278 717 return error("packfile %s not a regular file", p->pack_name);
bf592c50 718 p->pack_size = st.st_size;
9bc879c1 719 } else if (p->pack_size != st.st_size)
072db278 720 return error("packfile %s size changed", p->pack_name);
9bc879c1 721
2c039da8
JH
722 /* We leave these file descriptors open with sliding mmap;
723 * there is no point keeping them open across exec(), though.
724 */
725 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
726 if (fd_flag < 0)
072db278 727 return error("cannot determine file descriptor flags");
2c039da8
JH
728 fd_flag |= FD_CLOEXEC;
729 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 730 return error("cannot set FD_CLOEXEC");
2c039da8 731
9bc879c1 732 /* Verify we recognize this pack file format. */
e6e2bd62 733 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 734 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 735 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 736 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 737 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
738 return error("packfile %s is version %"PRIu32" and not"
739 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
740 p->pack_name, ntohl(hdr.hdr_version));
741
742 /* Verify the pack matches its index. */
57059091 743 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
744 return error("packfile %s claims to have %"PRIu32" objects"
745 " while index indicates %"PRIu32" objects",
57059091
NP
746 p->pack_name, ntohl(hdr.hdr_entries),
747 p->num_objects);
9bc879c1 748 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 749 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 750 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 751 return error("packfile %s signature is unavailable", p->pack_name);
42873078 752 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 753 if (hashcmp(sha1, idx_sha1))
072db278
SP
754 return error("packfile %s does not match index", p->pack_name);
755 return 0;
9bc879c1
SP
756}
757
3cf8b462
SP
758static int open_packed_git(struct packed_git *p)
759{
760 if (!open_packed_git_1(p))
761 return 0;
762 if (p->pack_fd != -1) {
763 close(p->pack_fd);
764 p->pack_fd = -1;
765 }
766 return -1;
767}
768
c4001d92 769static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
770{
771 /* We must promise at least 20 bytes (one hash) after the
772 * offset is available from this window, otherwise the offset
773 * is not actually in this window and a different window (which
774 * has that one hash excess) must be used. This is to support
775 * the object header and delta base parsing routines below.
776 */
777 off_t win_off = win->offset;
778 return win_off <= offset
779 && (offset + 20) <= (win_off + win->len);
780}
781
4b25d091 782unsigned char *use_pack(struct packed_git *p,
03e79c88 783 struct pack_window **w_cursor,
c4001d92 784 off_t offset,
03e79c88 785 unsigned int *left)
9bc879c1 786{
60bb8b14 787 struct pack_window *win = *w_cursor;
03e79c88 788
072db278
SP
789 if (p->pack_fd == -1 && open_packed_git(p))
790 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14 791
a9d98a14 792 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
793 * pointless to ask for an offset into the middle of that
794 * hash, and the in_window function above wouldn't match
795 * don't allow an offset too close to the end of the file.
796 */
797 if (offset > (p->pack_size - 20))
798 die("offset beyond end of packfile (truncated pack?)");
799
800 if (!win || !in_window(win, offset)) {
801 if (win)
802 win->inuse_cnt--;
803 for (win = p->windows; win; win = win->next) {
804 if (in_window(win, offset))
805 break;
806 }
807 if (!win) {
78a28df9 808 size_t window_align = packed_git_window_size / 2;
dc49cd76 809 off_t len;
60bb8b14 810 win = xcalloc(1, sizeof(*win));
78a28df9 811 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
812 len = p->pack_size - win->offset;
813 if (len > packed_git_window_size)
814 len = packed_git_window_size;
815 win->len = (size_t)len;
60bb8b14 816 pack_mapped += win->len;
11daf39b 817 while (packed_git_limit < pack_mapped
d1efefa4 818 && unuse_one_window(p, p->pack_fd))
60bb8b14 819 ; /* nothing */
c4712e45 820 win->base = xmmap(NULL, win->len,
60bb8b14
SP
821 PROT_READ, MAP_PRIVATE,
822 p->pack_fd, win->offset);
823 if (win->base == MAP_FAILED)
73b4e4be
SP
824 die("packfile %s cannot be mapped: %s",
825 p->pack_name,
826 strerror(errno));
a53128b6
SP
827 pack_mmap_calls++;
828 pack_open_windows++;
829 if (pack_mapped > peak_pack_mapped)
830 peak_pack_mapped = pack_mapped;
831 if (pack_open_windows > peak_pack_open_windows)
832 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
833 win->next = p->windows;
834 p->windows = win;
835 }
1f688557 836 }
03e79c88
SP
837 if (win != *w_cursor) {
838 win->last_used = pack_used_ctr++;
839 win->inuse_cnt++;
840 *w_cursor = win;
841 }
60bb8b14 842 offset -= win->offset;
03e79c88 843 if (left)
dc49cd76 844 *left = win->len - xsize_t(offset);
03e79c88 845 return win->base + offset;
1f688557
JH
846}
847
27d69a46
NP
848static struct packed_git *alloc_packed_git(int extra)
849{
850 struct packed_git *p = xmalloc(sizeof(*p) + extra);
851 memset(p, 0, sizeof(*p));
852 p->pack_fd = -1;
853 return p;
854}
855
e0500293
JN
856static void try_to_free_pack_memory(size_t size)
857{
858 release_pack_memory(size, -1);
859}
860
42873078 861struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557 862{
e0500293 863 static int have_set_try_to_free_routine;
1f688557 864 struct stat st;
27d69a46 865 struct packed_git *p = alloc_packed_git(path_len + 2);
1f688557 866
e0500293
JN
867 if (!have_set_try_to_free_routine) {
868 have_set_try_to_free_routine = 1;
869 set_try_to_free_routine(try_to_free_pack_memory);
870 }
871
42873078
NP
872 /*
873 * Make sure a corresponding .pack file exists and that
874 * the index looks sane.
875 */
876 path_len -= strlen(".idx");
27d69a46
NP
877 if (path_len < 1) {
878 free(p);
1f688557 879 return NULL;
27d69a46 880 }
42873078 881 memcpy(p->pack_name, path, path_len);
8d25931d
BC
882
883 strcpy(p->pack_name + path_len, ".keep");
884 if (!access(p->pack_name, F_OK))
885 p->pack_keep = 1;
886
42873078 887 strcpy(p->pack_name + path_len, ".pack");
d079837e 888 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 889 free(p);
1f688557
JH
890 return NULL;
891 }
42873078 892
1f688557
JH
893 /* ok, it looks sane as far as we can check without
894 * actually mapping the pack file.
895 */
1f688557 896 p->pack_size = st.st_size;
9d835df2 897 p->pack_local = local;
b867092f 898 p->mtime = st.st_mtime;
42873078
NP
899 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
900 hashclr(p->sha1);
1f688557
JH
901 return p;
902}
903
7b64469a 904struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 905{
42873078 906 const char *path = sha1_pack_name(sha1);
27d69a46 907 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
bf592c50 908
27d69a46
NP
909 strcpy(p->pack_name, path);
910 hashcpy(p->sha1, sha1);
42873078
NP
911 if (check_packed_git_idx(idx_path, p)) {
912 free(p);
bf592c50 913 return NULL;
42873078 914 }
bf592c50 915
bf592c50
DB
916 return p;
917}
918
919void install_packed_git(struct packed_git *pack)
920{
921 pack->next = packed_git;
922 packed_git = pack;
923}
924
9d835df2 925static void prepare_packed_git_one(char *objdir, int local)
1f688557 926{
9cb18f56
JM
927 /* Ensure that this buffer is large enough so that we can
928 append "/pack/" without clobbering the stack even if
929 strlen(objdir) were PATH_MAX. */
930 char path[PATH_MAX + 1 + 4 + 1 + 1];
1f688557
JH
931 int len;
932 DIR *dir;
933 struct dirent *de;
934
935 sprintf(path, "%s/pack", objdir);
936 len = strlen(path);
937 dir = opendir(path);
f2e872aa 938 while (!dir && errno == EMFILE && unuse_one_window(NULL, -1))
fd73ccf2 939 dir = opendir(path);
b5b16990 940 if (!dir) {
26125f6b 941 if (errno != ENOENT)
bd2afde8 942 error("unable to open object pack directory: %s: %s",
26125f6b 943 path, strerror(errno));
1f688557 944 return;
b5b16990 945 }
1f688557
JH
946 path[len++] = '/';
947 while ((de = readdir(dir)) != NULL) {
948 int namelen = strlen(de->d_name);
949 struct packed_git *p;
950
5bb1cda5 951 if (!has_extension(de->d_name, ".idx"))
1f688557
JH
952 continue;
953
9cb18f56
JM
954 if (len + namelen + 1 > sizeof(path))
955 continue;
956
54a15a8d 957 /* Don't reopen a pack we already have. */
1f688557 958 strcpy(path + len, de->d_name);
86f7780c
JK
959 for (p = packed_git; p; p = p->next) {
960 if (!memcmp(path, p->pack_name, len + namelen - 4))
961 break;
962 }
963 if (p)
964 continue;
54a15a8d
SP
965 /* See if it really is a valid .idx file with corresponding
966 * .pack file that we can map.
967 */
9d835df2 968 p = add_packed_git(path, len + namelen, local);
1f688557
JH
969 if (!p)
970 continue;
625e9421 971 install_packed_git(p);
1f688557 972 }
5b35bcd5 973 closedir(dir);
1f688557
JH
974}
975
b867092f
JH
976static int sort_pack(const void *a_, const void *b_)
977{
978 struct packed_git *a = *((struct packed_git **)a_);
979 struct packed_git *b = *((struct packed_git **)b_);
980 int st;
981
982 /*
983 * Local packs tend to contain objects specific to our
984 * variant of the project than remote ones. In addition,
985 * remote ones could be on a network mounted filesystem.
986 * Favor local ones for these reasons.
987 */
988 st = a->pack_local - b->pack_local;
989 if (st)
990 return -st;
991
992 /*
993 * Younger packs tend to contain more recent objects,
994 * and more recent objects tend to get accessed more
995 * often.
996 */
997 if (a->mtime < b->mtime)
998 return 1;
999 else if (a->mtime == b->mtime)
1000 return 0;
1001 return -1;
1002}
1003
1004static void rearrange_packed_git(void)
1005{
1006 struct packed_git **ary, *p;
1007 int i, n;
1008
1009 for (n = 0, p = packed_git; p; p = p->next)
1010 n++;
1011 if (n < 2)
1012 return;
1013
1014 /* prepare an array of packed_git for easier sorting */
1015 ary = xcalloc(n, sizeof(struct packed_git *));
1016 for (n = 0, p = packed_git; p; p = p->next)
1017 ary[n++] = p;
1018
1019 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1020
1021 /* link them back again */
1022 for (i = 0; i < n - 1; i++)
1023 ary[i]->next = ary[i + 1];
1024 ary[n - 1]->next = NULL;
1025 packed_git = ary[0];
1026
1027 free(ary);
1028}
1029
637cdd9d 1030static int prepare_packed_git_run_once = 0;
9a217f2a 1031void prepare_packed_git(void)
1f688557 1032{
d5a63b99 1033 struct alternate_object_database *alt;
1f688557 1034
637cdd9d 1035 if (prepare_packed_git_run_once)
1f688557 1036 return;
9d835df2 1037 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 1038 prepare_alt_odb();
d5a63b99 1039 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 1040 alt->name[-1] = 0;
9d835df2 1041 prepare_packed_git_one(alt->base, 0);
1494e038 1042 alt->name[-1] = '/';
1f688557 1043 }
b867092f 1044 rearrange_packed_git();
637cdd9d
JK
1045 prepare_packed_git_run_once = 1;
1046}
1047
fc04c412 1048void reprepare_packed_git(void)
637cdd9d 1049{
4b480c67 1050 discard_revindex();
637cdd9d
JK
1051 prepare_packed_git_run_once = 0;
1052 prepare_packed_git();
1f688557
JH
1053}
1054
8eca0b47
NP
1055static void mark_bad_packed_object(struct packed_git *p,
1056 const unsigned char *sha1)
1057{
1058 unsigned i;
1059 for (i = 0; i < p->num_bad_objects; i++)
1060 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1061 return;
1062 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1063 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1064 p->num_bad_objects++;
1065}
1066
b6c4cecc 1067static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
ac939109
NP
1068{
1069 struct packed_git *p;
1070 unsigned i;
1071
1072 for (p = packed_git; p; p = p->next)
1073 for (i = 0; i < p->num_bad_objects; i++)
1074 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
b6c4cecc
JH
1075 return p;
1076 return NULL;
ac939109
NP
1077}
1078
5d6ccf5c 1079int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
0fcfd160
LT
1080{
1081 unsigned char real_sha1[20];
7cfb5f36 1082 hash_sha1_file(map, size, type, real_sha1);
a89fccd2 1083 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1084}
1085
f2e872aa 1086static int git_open_noatime(const char *name, struct packed_git *p)
44d1c19e
LT
1087{
1088 static int sha1_file_open_flag = O_NOATIME;
44d1c19e 1089
f2e872aa
SP
1090 for (;;) {
1091 int fd = open(name, O_RDONLY | sha1_file_open_flag);
44d1c19e 1092 if (fd >= 0)
f2e872aa
SP
1093 return fd;
1094
1095 /* Might the failure be insufficient file descriptors? */
1096 if (errno == EMFILE) {
1097 if (unuse_one_window(p, -1))
1098 continue;
1099 else
1100 return -1;
1101 }
1102
1103 /* Might the failure be due to O_NOATIME? */
1104 if (errno != ENOENT && sha1_file_open_flag) {
44d1c19e 1105 sha1_file_open_flag = 0;
f2e872aa
SP
1106 continue;
1107 }
1108
1109 return -1;
44d1c19e 1110 }
44d1c19e
LT
1111}
1112
1113static int open_sha1_file(const unsigned char *sha1)
1114{
1115 int fd;
1116 char *name = sha1_file_name(sha1);
1117 struct alternate_object_database *alt;
1118
f2e872aa 1119 fd = git_open_noatime(name, NULL);
44d1c19e
LT
1120 if (fd >= 0)
1121 return fd;
1122
1123 prepare_alt_odb();
1124 errno = ENOENT;
1125 for (alt = alt_odb_list; alt; alt = alt->next) {
1126 name = alt->name;
1127 fill_sha1_path(name, sha1);
f2e872aa 1128 fd = git_open_noatime(alt->base, NULL);
44d1c19e
LT
1129 if (fd >= 0)
1130 return fd;
1131 }
1132 return -1;
1133}
1134
4175e9e3 1135static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1136{
0fcfd160 1137 void *map;
144bde78 1138 int fd;
ace1534d 1139
44d1c19e
LT
1140 fd = open_sha1_file(sha1);
1141 map = NULL;
1142 if (fd >= 0) {
1143 struct stat st;
0fcfd160 1144
44d1c19e
LT
1145 if (!fstat(fd, &st)) {
1146 *size = xsize_t(st.st_size);
1147 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1148 }
44d1c19e 1149 close(fd);
0fcfd160 1150 }
0fcfd160
LT
1151 return map;
1152}
1153
726f852b 1154static int legacy_loose_object(unsigned char *map)
bb6b8e4f
JH
1155{
1156 unsigned int word;
1157
1158 /*
1159 * Is it a zlib-compressed buffer? If so, the first byte
1160 * must be 0x78 (15-bit window size, deflated), and the
1161 * first 16-bit word is evenly divisible by 31
1162 */
1163 word = (map[0] << 8) + map[1];
1164 if (map[0] == 0x78 && !(word % 31))
1165 return 1;
1166 else
1167 return 0;
1168}
1169
09ded04b
NP
1170unsigned long unpack_object_header_buffer(const unsigned char *buf,
1171 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1172{
ad1ed5ee 1173 unsigned shift;
48fb7deb 1174 unsigned long size, c;
ad1ed5ee
JH
1175 unsigned long used = 0;
1176
1177 c = buf[used++];
1178 *type = (c >> 4) & 7;
1179 size = c & 15;
1180 shift = 4;
1181 while (c & 0x80) {
f630cfda 1182 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1183 error("bad object header");
ad1ed5ee 1184 return 0;
09ded04b 1185 }
ad1ed5ee
JH
1186 c = buf[used++];
1187 size += (c & 0x7f) << shift;
1188 shift += 7;
1189 }
1190 *sizep = size;
1191 return used;
1192}
1193
1194static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1195{
1196 unsigned long size, used;
1197 static const char valid_loose_object_type[8] = {
1198 0, /* OBJ_EXT */
1199 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1200 0, /* "delta" and others are invalid in a loose object */
93821bd9 1201 };
ad1ed5ee 1202 enum object_type type;
93821bd9 1203
c4483576
LT
1204 /* Get the data stream */
1205 memset(stream, 0, sizeof(*stream));
1206 stream->next_in = map;
1207 stream->avail_in = mapsize;
1208 stream->next_out = buffer;
93821bd9
LT
1209 stream->avail_out = bufsiz;
1210
bb6b8e4f 1211 if (legacy_loose_object(map)) {
39c68542
LT
1212 git_inflate_init(stream);
1213 return git_inflate(stream, 0);
93821bd9
LT
1214 }
1215
726f852b
NP
1216
1217 /*
1218 * There used to be a second loose object header format which
1219 * was meant to mimic the in-pack format, allowing for direct
1220 * copy of the object data. This format turned up not to be
1221 * really worth it and we don't write it any longer. But we
1222 * can still read it.
1223 */
09ded04b 1224 used = unpack_object_header_buffer(map, mapsize, &type, &size);
ad1ed5ee 1225 if (!used || !valid_loose_object_type[type])
93821bd9 1226 return -1;
ad1ed5ee
JH
1227 map += used;
1228 mapsize -= used;
93821bd9
LT
1229
1230 /* Set up the stream for the rest.. */
1231 stream->next_in = map;
1232 stream->avail_in = mapsize;
39c68542 1233 git_inflate_init(stream);
93821bd9
LT
1234
1235 /* And generate the fake traditional header */
ad1ed5ee 1236 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
df843662 1237 typename(type), size);
93821bd9 1238 return 0;
c4483576
LT
1239}
1240
7efbff75 1241static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1242{
1243 int bytes = strlen(buffer) + 1;
3aee68aa 1244 unsigned char *buf = xmallocz(size);
93821bd9 1245 unsigned long n;
7efbff75 1246 int status = Z_OK;
5180cacc 1247
93821bd9
LT
1248 n = stream->total_out - bytes;
1249 if (n > size)
1250 n = size;
1251 memcpy(buf, (char *) buffer + bytes, n);
1252 bytes = n;
456cdf6e
LT
1253 if (bytes <= size) {
1254 /*
1255 * The above condition must be (bytes <= size), not
1256 * (bytes < size). In other words, even though we
1257 * expect no more output and set avail_out to zer0,
1258 * the input zlib stream may have bytes that express
1259 * "this concludes the stream", and we *do* want to
1260 * eat that input.
1261 *
1262 * Otherwise we would not be able to test that we
1263 * consumed all the input to reach the expected size;
1264 * we also want to check that zlib tells us that all
1265 * went well with status == Z_STREAM_END at the end.
1266 */
5180cacc
LT
1267 stream->next_out = buf + bytes;
1268 stream->avail_out = size - bytes;
7efbff75 1269 while (status == Z_OK)
39c68542 1270 status = git_inflate(stream, Z_FINISH);
5180cacc 1271 }
456cdf6e 1272 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1273 git_inflate_end(stream);
7efbff75
JH
1274 return buf;
1275 }
1276
1277 if (status < 0)
1278 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1279 else if (stream->avail_in)
1280 error("garbage at end of loose object '%s'",
1281 sha1_to_hex(sha1));
1282 free(buf);
1283 return NULL;
5180cacc
LT
1284}
1285
1286/*
1287 * We used to just use "sscanf()", but that's actually way
1288 * too permissive for what we want to check. So do an anal
1289 * object header parse by hand.
1290 */
21666f1a 1291static int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1292{
21666f1a 1293 char type[10];
5180cacc
LT
1294 int i;
1295 unsigned long size;
1296
1297 /*
a6080a0a 1298 * The type can be at most ten bytes (including the
5180cacc 1299 * terminating '\0' that we add), and is followed by
21666f1a 1300 * a space.
5180cacc 1301 */
21666f1a 1302 i = 0;
5180cacc
LT
1303 for (;;) {
1304 char c = *hdr++;
1305 if (c == ' ')
1306 break;
21666f1a
NP
1307 type[i++] = c;
1308 if (i >= sizeof(type))
5180cacc 1309 return -1;
5180cacc 1310 }
21666f1a 1311 type[i] = 0;
5180cacc
LT
1312
1313 /*
1314 * The length must follow immediately, and be in canonical
1315 * decimal format (ie "010" is not valid).
1316 */
1317 size = *hdr++ - '0';
1318 if (size > 9)
1319 return -1;
1320 if (size) {
1321 for (;;) {
1322 unsigned long c = *hdr - '0';
1323 if (c > 9)
1324 break;
1325 hdr++;
1326 size = size * 10 + c;
1327 }
1328 }
1329 *sizep = size;
1330
1331 /*
1332 * The length must be followed by a zero byte
1333 */
21666f1a 1334 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1335}
1336
7efbff75 1337static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1338{
5180cacc 1339 int ret;
0fcfd160 1340 z_stream stream;
5180cacc 1341 char hdr[8192];
0fcfd160 1342
5180cacc 1343 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1344 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1345 return NULL;
1346
7efbff75 1347 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1348}
1349
54dab52a
NP
1350unsigned long get_size_from_delta(struct packed_git *p,
1351 struct pack_window **w_curs,
1352 off_t curpos)
1353{
1354 const unsigned char *data;
1355 unsigned char delta_head[20], *in;
1356 z_stream stream;
1357 int st;
1358
1359 memset(&stream, 0, sizeof(stream));
1360 stream.next_out = delta_head;
1361 stream.avail_out = sizeof(delta_head);
1362
39c68542 1363 git_inflate_init(&stream);
54dab52a
NP
1364 do {
1365 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1366 stream.next_in = in;
39c68542 1367 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1368 curpos += stream.next_in - in;
1369 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1370 stream.total_out < sizeof(delta_head));
39c68542 1371 git_inflate_end(&stream);
3d77d877
NP
1372 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1373 error("delta data unpack-initial failed");
1374 return 0;
1375 }
54dab52a
NP
1376
1377 /* Examine the initial part of the delta to figure out
1378 * the result size.
1379 */
1380 data = delta_head;
1381
1382 /* ignore base size */
1383 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1384
1385 /* Read the result size */
1386 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1387}
1388
c4001d92 1389static off_t get_delta_base(struct packed_git *p,
03e79c88 1390 struct pack_window **w_curs,
c4001d92 1391 off_t *curpos,
21666f1a 1392 enum object_type type,
c4001d92 1393 off_t delta_obj_offset)
eb32d236 1394{
2b87c45b 1395 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1396 off_t base_offset;
eb32d236 1397
8d8a4ea5
SP
1398 /* use_pack() assured us we have [base_info, base_info + 20)
1399 * as a range that we can look at without walking off the
1400 * end of the mapped window. Its actually the hash size
1401 * that is assured. An OFS_DELTA longer than the hash size
1402 * is stupid, as then a REF_DELTA would be smaller to store.
1403 */
21666f1a 1404 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1405 unsigned used = 0;
1406 unsigned char c = base_info[used++];
1407 base_offset = c & 127;
1408 while (c & 128) {
1409 base_offset += 1;
8723f216 1410 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1411 return 0; /* overflow */
eb32d236
NP
1412 c = base_info[used++];
1413 base_offset = (base_offset << 7) + (c & 127);
1414 }
1415 base_offset = delta_obj_offset - base_offset;
d8f32556 1416 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1417 return 0; /* out of bound */
2b87c45b 1418 *curpos += used;
21666f1a 1419 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1420 /* The base entry _must_ be in the same pack */
1421 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1422 *curpos += 20;
eb32d236
NP
1423 } else
1424 die("I am totally screwed");
2b87c45b 1425 return base_offset;
eb32d236
NP
1426}
1427
f3bf9224 1428/* forward declaration for a mutually recursive function */
c4001d92 1429static int packed_object_info(struct packed_git *p, off_t offset,
21666f1a 1430 unsigned long *sizep);
f3bf9224 1431
43057304 1432static int packed_delta_info(struct packed_git *p,
03e79c88 1433 struct pack_window **w_curs,
c4001d92 1434 off_t curpos,
21666f1a 1435 enum object_type type,
c4001d92 1436 off_t obj_offset,
43057304 1437 unsigned long *sizep)
5db47c2b 1438{
c4001d92 1439 off_t base_offset;
f3bf9224 1440
21666f1a 1441 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
3d77d877
NP
1442 if (!base_offset)
1443 return OBJ_BAD;
21666f1a 1444 type = packed_object_info(p, base_offset, NULL);
3d77d877 1445 if (type <= OBJ_NONE) {
08698b1e
NP
1446 struct revindex_entry *revidx;
1447 const unsigned char *base_sha1;
1448 revidx = find_pack_revindex(p, base_offset);
1449 if (!revidx)
1450 return OBJ_BAD;
1451 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
3d77d877
NP
1452 mark_bad_packed_object(p, base_sha1);
1453 type = sha1_object_info(base_sha1, NULL);
1454 if (type <= OBJ_NONE)
1455 return OBJ_BAD;
1456 }
f3bf9224 1457
c62266f3
JH
1458 /* We choose to only get the type of the base object and
1459 * ignore potentially corrupt pack file that expects the delta
1460 * based on a base with a wrong size. This saves tons of
1461 * inflate() calls.
1462 */
3d77d877 1463 if (sizep) {
54dab52a 1464 *sizep = get_size_from_delta(p, w_curs, curpos);
3d77d877
NP
1465 if (*sizep == 0)
1466 type = OBJ_BAD;
1467 }
21666f1a
NP
1468
1469 return type;
5db47c2b
JH
1470}
1471
2b87c45b
NP
1472static int unpack_object_header(struct packed_git *p,
1473 struct pack_window **w_curs,
c4001d92 1474 off_t *curpos,
2b87c45b 1475 unsigned long *sizep)
a733cb60 1476{
03e79c88
SP
1477 unsigned char *base;
1478 unsigned int left;
ad1ed5ee 1479 unsigned long used;
2b87c45b 1480 enum object_type type;
a733cb60 1481
8d8a4ea5
SP
1482 /* use_pack() assures us we have [base, base + 20) available
1483 * as a range that we can look at at. (Its actually the hash
3dff5379 1484 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1485 * the maximum deflated object size is 2^137, which is just
1486 * insane, so we know won't exceed what we have been given.
1487 */
2b87c45b 1488 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1489 used = unpack_object_header_buffer(base, left, &type, sizep);
1490 if (!used) {
1491 type = OBJ_BAD;
1492 } else
1493 *curpos += used;
ad1ed5ee 1494
2b87c45b 1495 return type;
a733cb60
LT
1496}
1497
21666f1a 1498const char *packed_object_info_detail(struct packed_git *p,
c4001d92 1499 off_t obj_offset,
21666f1a
NP
1500 unsigned long *size,
1501 unsigned long *store_size,
1502 unsigned int *delta_chain_length,
1503 unsigned char *base_sha1)
ad8c80a5 1504{
03e79c88 1505 struct pack_window *w_curs = NULL;
c4001d92
SP
1506 off_t curpos;
1507 unsigned long dummy;
43057304 1508 unsigned char *next_sha1;
21666f1a 1509 enum object_type type;
70f5d5d3 1510 struct revindex_entry *revidx;
ad8c80a5 1511
43057304 1512 *delta_chain_length = 0;
2b87c45b 1513 curpos = obj_offset;
21666f1a 1514 type = unpack_object_header(p, &w_curs, &curpos, size);
43057304 1515
70f5d5d3
NP
1516 revidx = find_pack_revindex(p, obj_offset);
1517 *store_size = revidx[1].offset - obj_offset;
1518
43057304 1519 for (;;) {
21666f1a 1520 switch (type) {
43057304 1521 default:
08a19d87 1522 die("pack %s contains unknown object type %d",
21666f1a 1523 p->pack_name, type);
43057304
NP
1524 case OBJ_COMMIT:
1525 case OBJ_TREE:
1526 case OBJ_BLOB:
1527 case OBJ_TAG:
03e79c88 1528 unuse_pack(&w_curs);
21666f1a 1529 return typename(type);
eb32d236 1530 case OBJ_OFS_DELTA:
21666f1a 1531 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
8eca0b47
NP
1532 if (!obj_offset)
1533 die("pack %s contains bad delta base reference of type %s",
1534 p->pack_name, typename(type));
eb32d236 1535 if (*delta_chain_length == 0) {
70f5d5d3
NP
1536 revidx = find_pack_revindex(p, obj_offset);
1537 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
eb32d236
NP
1538 }
1539 break;
1540 case OBJ_REF_DELTA:
2b87c45b 1541 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
43057304
NP
1542 if (*delta_chain_length == 0)
1543 hashcpy(base_sha1, next_sha1);
2b87c45b 1544 obj_offset = find_pack_entry_one(next_sha1, p);
43057304
NP
1545 break;
1546 }
43057304 1547 (*delta_chain_length)++;
2b87c45b 1548 curpos = obj_offset;
21666f1a 1549 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
ad8c80a5 1550 }
ad8c80a5
JH
1551}
1552
c4001d92 1553static int packed_object_info(struct packed_git *p, off_t obj_offset,
21666f1a 1554 unsigned long *sizep)
1f688557 1555{
03e79c88 1556 struct pack_window *w_curs = NULL;
c4001d92
SP
1557 unsigned long size;
1558 off_t curpos = obj_offset;
21666f1a 1559 enum object_type type;
5db47c2b 1560
21666f1a 1561 type = unpack_object_header(p, &w_curs, &curpos, &size);
5db47c2b 1562
21666f1a 1563 switch (type) {
eb32d236
NP
1564 case OBJ_OFS_DELTA:
1565 case OBJ_REF_DELTA:
21666f1a
NP
1566 type = packed_delta_info(p, &w_curs, curpos,
1567 type, obj_offset, sizep);
2b87c45b 1568 break;
a733cb60 1569 case OBJ_COMMIT:
a733cb60 1570 case OBJ_TREE:
a733cb60 1571 case OBJ_BLOB:
a733cb60 1572 case OBJ_TAG:
2b87c45b
NP
1573 if (sizep)
1574 *sizep = size;
a69d0943 1575 break;
1f688557 1576 default:
3d77d877
NP
1577 error("unknown object type %i at offset %"PRIuMAX" in %s",
1578 type, (uintmax_t)obj_offset, p->pack_name);
1579 type = OBJ_BAD;
1f688557 1580 }
2b87c45b 1581 unuse_pack(&w_curs);
21666f1a 1582 return type;
1f688557
JH
1583}
1584
eb950c19 1585static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1586 struct pack_window **w_curs,
c4001d92 1587 off_t curpos,
eb950c19 1588 unsigned long size)
de530aaa
SP
1589{
1590 int st;
1591 z_stream stream;
079afb18 1592 unsigned char *buffer, *in;
de530aaa 1593
4ab07e4d 1594 buffer = xmallocz(size);
de530aaa 1595 memset(&stream, 0, sizeof(stream));
de530aaa 1596 stream.next_out = buffer;
39eea7bd 1597 stream.avail_out = size + 1;
de530aaa 1598
39c68542 1599 git_inflate_init(&stream);
079afb18 1600 do {
2b87c45b 1601 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 1602 stream.next_in = in;
39c68542 1603 st = git_inflate(&stream, Z_FINISH);
39eea7bd
JH
1604 if (!stream.avail_out)
1605 break; /* the payload is larger than it should be */
2b87c45b 1606 curpos += stream.next_in - in;
079afb18 1607 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 1608 git_inflate_end(&stream);
de530aaa
SP
1609 if ((st != Z_STREAM_END) || stream.total_out != size) {
1610 free(buffer);
1611 return NULL;
1612 }
1613
1614 return buffer;
1615}
1616
e5e01619
LT
1617#define MAX_DELTA_CACHE (256)
1618
18bdec11 1619static size_t delta_base_cached;
5e08ecbf
NP
1620
1621static struct delta_base_cache_lru_list {
1622 struct delta_base_cache_lru_list *prev;
1623 struct delta_base_cache_lru_list *next;
1624} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1625
e5e01619 1626static struct delta_base_cache_entry {
5e08ecbf
NP
1627 struct delta_base_cache_lru_list lru;
1628 void *data;
e5e01619
LT
1629 struct packed_git *p;
1630 off_t base_offset;
1631 unsigned long size;
e5e01619
LT
1632 enum object_type type;
1633} delta_base_cache[MAX_DELTA_CACHE];
1634
1635static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1636{
1637 unsigned long hash;
1638
1639 hash = (unsigned long)p + (unsigned long)base_offset;
1640 hash += (hash >> 8) + (hash >> 16);
3358004a 1641 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1642}
1643
62f255ad 1644static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 1645 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 1646{
e5e01619
LT
1647 void *ret;
1648 unsigned long hash = pack_entry_hash(p, base_offset);
1649 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1650
1651 ret = ent->data;
749bc58c
MV
1652 if (!ret || ent->p != p || ent->base_offset != base_offset)
1653 return unpack_entry(p, base_offset, type, base_size);
e5e01619 1654
18bdec11 1655 if (!keep_cache) {
a0cba108 1656 ent->data = NULL;
5e08ecbf
NP
1657 ent->lru.next->prev = ent->lru.prev;
1658 ent->lru.prev->next = ent->lru.next;
18bdec11 1659 delta_base_cached -= ent->size;
182af834
PH
1660 } else {
1661 ret = xmemdupz(ent->data, ent->size);
a0cba108 1662 }
e5e01619
LT
1663 *type = ent->type;
1664 *base_size = ent->size;
1665 return ret;
62f255ad
LT
1666}
1667
18bdec11
SP
1668static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1669{
1670 if (ent->data) {
1671 free(ent->data);
1672 ent->data = NULL;
5e08ecbf
NP
1673 ent->lru.next->prev = ent->lru.prev;
1674 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
1675 delta_base_cached -= ent->size;
1676 }
1677}
1678
3d20c636
SP
1679void clear_delta_base_cache(void)
1680{
1681 unsigned long p;
1682 for (p = 0; p < MAX_DELTA_CACHE; p++)
1683 release_delta_base_cache(&delta_base_cache[p]);
1684}
1685
62f255ad
LT
1686static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1687 void *base, unsigned long base_size, enum object_type type)
1688{
5e08ecbf 1689 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 1690 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 1691 struct delta_base_cache_lru_list *lru;
e5e01619 1692
18bdec11
SP
1693 release_delta_base_cache(ent);
1694 delta_base_cached += base_size;
5e08ecbf
NP
1695
1696 for (lru = delta_base_cache_lru.next;
1697 delta_base_cached > delta_base_cache_limit
1698 && lru != &delta_base_cache_lru;
1699 lru = lru->next) {
1700 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
1701 if (f->type == OBJ_BLOB)
1702 release_delta_base_cache(f);
1703 }
5e08ecbf
NP
1704 for (lru = delta_base_cache_lru.next;
1705 delta_base_cached > delta_base_cache_limit
1706 && lru != &delta_base_cache_lru;
1707 lru = lru->next) {
1708 struct delta_base_cache_entry *f = (void *)lru;
1709 release_delta_base_cache(f);
1710 }
18bdec11 1711
e5e01619
LT
1712 ent->p = p;
1713 ent->base_offset = base_offset;
1714 ent->type = type;
1715 ent->data = base;
1716 ent->size = base_size;
5e08ecbf
NP
1717 ent->lru.next = &delta_base_cache_lru;
1718 ent->lru.prev = delta_base_cache_lru.prev;
1719 delta_base_cache_lru.prev->next = &ent->lru;
1720 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
1721}
1722
c2c5b270
CC
1723static void *read_object(const unsigned char *sha1, enum object_type *type,
1724 unsigned long *size);
1725
eb950c19 1726static void *unpack_delta_entry(struct packed_git *p,
03e79c88 1727 struct pack_window **w_curs,
c4001d92 1728 off_t curpos,
1f688557 1729 unsigned long delta_size,
c4001d92 1730 off_t obj_offset,
21666f1a 1731 enum object_type *type,
eb950c19 1732 unsigned long *sizep)
1f688557 1733{
7c3e8be3 1734 void *delta_data, *result, *base;
c4001d92
SP
1735 unsigned long base_size;
1736 off_t base_offset;
43057304 1737
21666f1a 1738 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
8eca0b47
NP
1739 if (!base_offset) {
1740 error("failed to validate delta base reference "
1741 "at offset %"PRIuMAX" from %s",
1742 (uintmax_t)curpos, p->pack_name);
1743 return NULL;
1744 }
eac12e2d 1745 unuse_pack(w_curs);
a0cba108 1746 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
8eca0b47
NP
1747 if (!base) {
1748 /*
1749 * We're probably in deep shit, but let's try to fetch
1750 * the required base anyway from another pack or loose.
1751 * This is costly but should happen only in the presence
1752 * of a corrupted pack, and is better than failing outright.
1753 */
08698b1e
NP
1754 struct revindex_entry *revidx;
1755 const unsigned char *base_sha1;
1756 revidx = find_pack_revindex(p, base_offset);
1757 if (!revidx)
1758 return NULL;
1759 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
8eca0b47
NP
1760 error("failed to read delta base object %s"
1761 " at offset %"PRIuMAX" from %s",
1762 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1763 p->pack_name);
1764 mark_bad_packed_object(p, base_sha1);
ac939109 1765 base = read_object(base_sha1, type, &base_size);
8eca0b47
NP
1766 if (!base)
1767 return NULL;
1768 }
67686d95 1769
2b87c45b 1770 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
8eca0b47
NP
1771 if (!delta_data) {
1772 error("failed to unpack compressed delta "
1773 "at offset %"PRIuMAX" from %s",
1774 (uintmax_t)curpos, p->pack_name);
1775 free(base);
1776 return NULL;
1777 }
1f688557
JH
1778 result = patch_delta(base, base_size,
1779 delta_data, delta_size,
21666f1a 1780 sizep);
1f688557
JH
1781 if (!result)
1782 die("failed to apply delta");
1783 free(delta_data);
62f255ad 1784 add_delta_base_cache(p, base_offset, base, base_size, *type);
1f688557
JH
1785 return result;
1786}
1787
0e8189e2
NP
1788int do_check_packed_object_crc;
1789
c4001d92 1790void *unpack_entry(struct packed_git *p, off_t obj_offset,
21666f1a 1791 enum object_type *type, unsigned long *sizep)
1f688557 1792{
03e79c88 1793 struct pack_window *w_curs = NULL;
c4001d92 1794 off_t curpos = obj_offset;
21666f1a 1795 void *data;
1f688557 1796
0e8189e2
NP
1797 if (do_check_packed_object_crc && p->index_version > 1) {
1798 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1799 unsigned long len = revidx[1].offset - obj_offset;
1800 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1801 const unsigned char *sha1 =
1802 nth_packed_object_sha1(p, revidx->nr);
1803 error("bad packed object CRC for %s",
1804 sha1_to_hex(sha1));
1805 mark_bad_packed_object(p, sha1);
f755bb99 1806 unuse_pack(&w_curs);
0e8189e2
NP
1807 return NULL;
1808 }
1809 }
1810
21666f1a
NP
1811 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1812 switch (*type) {
eb32d236
NP
1813 case OBJ_OFS_DELTA:
1814 case OBJ_REF_DELTA:
21666f1a
NP
1815 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1816 obj_offset, type, sizep);
4d703a1a 1817 break;
a733cb60 1818 case OBJ_COMMIT:
a733cb60 1819 case OBJ_TREE:
a733cb60 1820 case OBJ_BLOB:
a733cb60 1821 case OBJ_TAG:
21666f1a 1822 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
4d703a1a 1823 break;
1f688557 1824 default:
8eca0b47
NP
1825 data = NULL;
1826 error("unknown object type %i at offset %"PRIuMAX" in %s",
1827 *type, (uintmax_t)obj_offset, p->pack_name);
1f688557 1828 }
03e79c88 1829 unuse_pack(&w_curs);
21666f1a 1830 return data;
1f688557
JH
1831}
1832
d079837e 1833const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 1834 uint32_t n)
9a217f2a 1835{
42873078 1836 const unsigned char *index = p->index_data;
d079837e
SP
1837 if (!index) {
1838 if (open_pack_index(p))
1839 return NULL;
1840 index = p->index_data;
1841 }
57059091 1842 if (n >= p->num_objects)
d72308e0 1843 return NULL;
74e34e1f
NP
1844 index += 4 * 256;
1845 if (p->index_version == 1) {
1846 return index + 24 * n + 4;
1847 } else {
1848 index += 8;
1849 return index + 20 * n;
1850 }
1851}
1852
99093238 1853off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
1854{
1855 const unsigned char *index = p->index_data;
1856 index += 4 * 256;
1857 if (p->index_version == 1) {
1858 return ntohl(*((uint32_t *)(index + 24 * n)));
1859 } else {
1860 uint32_t off;
1861 index += 8 + p->num_objects * (20 + 4);
1862 off = ntohl(*((uint32_t *)(index + 4 * n)));
1863 if (!(off & 0x80000000))
1864 return off;
1865 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1866 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1867 ntohl(*((uint32_t *)(index + 4)));
1868 }
9a217f2a
JH
1869}
1870
c4001d92 1871off_t find_pack_entry_one(const unsigned char *sha1,
43057304 1872 struct packed_git *p)
1f688557 1873{
42873078 1874 const uint32_t *level1_ofs = p->index_data;
42873078 1875 const unsigned char *index = p->index_data;
628522ec
JH
1876 unsigned hi, lo, stride;
1877 static int use_lookup = -1;
1878 static int debug_lookup = -1;
1879
1880 if (debug_lookup < 0)
1881 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 1882
d079837e
SP
1883 if (!index) {
1884 if (open_pack_index(p))
1885 return 0;
1886 level1_ofs = p->index_data;
1887 index = p->index_data;
1888 }
74e34e1f
NP
1889 if (p->index_version > 1) {
1890 level1_ofs += 2;
1891 index += 8;
1892 }
42873078 1893 index += 4 * 256;
74e34e1f
NP
1894 hi = ntohl(level1_ofs[*sha1]);
1895 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
1896 if (p->index_version > 1) {
1897 stride = 20;
1898 } else {
1899 stride = 24;
1900 index += 4;
1901 }
1902
1903 if (debug_lookup)
6e1c2344 1904 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
1905 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1906
1907 if (use_lookup < 0)
1908 use_lookup = !!getenv("GIT_USE_LOOKUP");
1909 if (use_lookup) {
1910 int pos = sha1_entry_pos(index, stride, 0,
1911 lo, hi, p->num_objects, sha1);
1912 if (pos < 0)
1913 return 0;
1914 return nth_packed_object_offset(p, pos);
1915 }
1f688557
JH
1916
1917 do {
74e34e1f 1918 unsigned mi = (lo + hi) / 2;
628522ec
JH
1919 int cmp = hashcmp(index + mi * stride, sha1);
1920
1921 if (debug_lookup)
1922 printf("lo %u hi %u rg %u mi %u\n",
1923 lo, hi, hi - lo, mi);
43057304 1924 if (!cmp)
74e34e1f 1925 return nth_packed_object_offset(p, mi);
1f688557
JH
1926 if (cmp > 0)
1927 hi = mi;
1928 else
1929 lo = mi+1;
1930 } while (lo < hi);
1931 return 0;
1932}
1933
4d6acb70 1934static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557 1935{
f7c22cc6 1936 static struct packed_git *last_found = (void *)1;
1f688557 1937 struct packed_git *p;
c4001d92 1938 off_t offset;
43057304 1939
1f688557 1940 prepare_packed_git();
f7c22cc6
NP
1941 if (!packed_git)
1942 return 0;
1943 p = (last_found == (void *)1) ? packed_git : last_found;
1f688557 1944
f7c22cc6 1945 do {
8eca0b47
NP
1946 if (p->num_bad_objects) {
1947 unsigned i;
1948 for (i = 0; i < p->num_bad_objects; i++)
1949 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1950 goto next;
1951 }
1952
43057304
NP
1953 offset = find_pack_entry_one(sha1, p);
1954 if (offset) {
c715f783
SP
1955 /*
1956 * We are about to tell the caller where they can
1957 * locate the requested object. We better make
1958 * sure the packfile is still here and can be
1959 * accessed before supplying that answer, as
1960 * it may have been deleted since the index
1961 * was loaded!
1962 */
1963 if (p->pack_fd == -1 && open_packed_git(p)) {
1964 error("packfile %s cannot be accessed", p->pack_name);
f7c22cc6 1965 goto next;
c715f783 1966 }
43057304
NP
1967 e->offset = offset;
1968 e->p = p;
1969 hashcpy(e->sha1, sha1);
f7c22cc6 1970 last_found = p;
1f688557 1971 return 1;
43057304 1972 }
f7c22cc6
NP
1973
1974 next:
1975 if (p == last_found)
1976 p = packed_git;
1977 else
1978 p = p->next;
1979 if (p == last_found)
1980 p = p->next;
1981 } while (p);
1f688557
JH
1982 return 0;
1983}
1984
a6080a0a 1985struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
1986 struct packed_git *packs)
1987{
1988 struct packed_git *p;
bf592c50
DB
1989
1990 for (p = packs; p; p = p->next) {
43057304 1991 if (find_pack_entry_one(sha1, p))
bf592c50
DB
1992 return p;
1993 }
1994 return NULL;
21666f1a 1995
bf592c50
DB
1996}
1997
21666f1a 1998static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
65c2e0c3 1999{
36e4d74a 2000 int status;
65c2e0c3
JH
2001 unsigned long mapsize, size;
2002 void *map;
2003 z_stream stream;
d65a16f6 2004 char hdr[32];
65c2e0c3 2005
bb6b8e4f 2006 map = map_sha1_file(sha1, &mapsize);
f0df4ed5
JS
2007 if (!map)
2008 return error("unable to find %s", sha1_to_hex(sha1));
36e4d74a
JH
2009 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2010 status = error("unable to unpack %s header",
2011 sha1_to_hex(sha1));
21666f1a 2012 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 2013 status = error("unable to parse %s header", sha1_to_hex(sha1));
21666f1a
NP
2014 else if (sizep)
2015 *sizep = size;
39c68542 2016 git_inflate_end(&stream);
65c2e0c3
JH
2017 munmap(map, mapsize);
2018 return status;
2019}
2020
21666f1a 2021int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
f0df4ed5 2022{
f0df4ed5 2023 struct pack_entry e;
ddd63e64 2024 int status;
f0df4ed5 2025
cd673c1f 2026 if (!find_pack_entry(sha1, &e)) {
ddd63e64
SG
2027 /* Most likely it's a loose object. */
2028 status = sha1_loose_object_info(sha1, sizep);
2029 if (status >= 0)
2030 return status;
2031
2032 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 2033 reprepare_packed_git();
cd673c1f 2034 if (!find_pack_entry(sha1, &e))
ddd63e64 2035 return status;
f0df4ed5 2036 }
3d77d877
NP
2037
2038 status = packed_object_info(e.p, e.offset, sizep);
2039 if (status < 0) {
2040 mark_bad_packed_object(e.p, sha1);
2041 status = sha1_object_info(sha1, sizep);
2042 }
2043
2044 return status;
f0df4ed5
JS
2045}
2046
21666f1a
NP
2047static void *read_packed_sha1(const unsigned char *sha1,
2048 enum object_type *type, unsigned long *size)
1f688557
JH
2049{
2050 struct pack_entry e;
8eca0b47 2051 void *data;
1f688557 2052
cd673c1f 2053 if (!find_pack_entry(sha1, &e))
1f688557 2054 return NULL;
8eca0b47
NP
2055 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2056 if (!data) {
2057 /*
2058 * We're probably in deep shit, but let's try to fetch
2059 * the required object anyway from another pack or loose.
2060 * This should happen only in the presence of a corrupted
2061 * pack, and is better than failing outright.
2062 */
2063 error("failed to read object %s at offset %"PRIuMAX" from %s",
2064 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2065 mark_bad_packed_object(e.p, sha1);
ac939109 2066 data = read_object(sha1, type, size);
8eca0b47
NP
2067 }
2068 return data;
1f688557
JH
2069}
2070
21666f1a
NP
2071int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2072 unsigned char *sha1)
d66b37bb
JH
2073{
2074 struct cached_object *co;
2075
21666f1a 2076 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2077 if (has_sha1_file(sha1) || find_cached_object(sha1))
2078 return 0;
2079 if (cached_object_alloc <= cached_object_nr) {
2080 cached_object_alloc = alloc_nr(cached_object_alloc);
2081 cached_objects = xrealloc(cached_objects,
2082 sizeof(*cached_objects) *
2083 cached_object_alloc);
2084 }
2085 co = &cached_objects[cached_object_nr++];
2086 co->size = len;
21666f1a 2087 co->type = type;
efa13f7b
JH
2088 co->buf = xmalloc(len);
2089 memcpy(co->buf, buf, len);
d66b37bb
JH
2090 hashcpy(co->sha1, sha1);
2091 return 0;
2092}
2093
c2c5b270
CC
2094static void *read_object(const unsigned char *sha1, enum object_type *type,
2095 unsigned long *size)
0fcfd160
LT
2096{
2097 unsigned long mapsize;
2098 void *map, *buf;
d66b37bb
JH
2099 struct cached_object *co;
2100
2101 co = find_cached_object(sha1);
2102 if (co) {
21666f1a 2103 *type = co->type;
d66b37bb 2104 *size = co->size;
182af834 2105 return xmemdupz(co->buf, co->size);
d66b37bb 2106 }
0fcfd160 2107
8276c007
PE
2108 buf = read_packed_sha1(sha1, type, size);
2109 if (buf)
2110 return buf;
bb6b8e4f 2111 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2112 if (map) {
7efbff75 2113 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2114 munmap(map, mapsize);
2115 return buf;
2116 }
637cdd9d 2117 reprepare_packed_git();
8276c007 2118 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2119}
2120
b6c4cecc
JH
2121/*
2122 * This function dies on corrupt objects; the callers who want to
2123 * deal with them should arrange to call read_object() and give error
2124 * messages themselves.
2125 */
f5552aee
CC
2126void *read_sha1_file_repl(const unsigned char *sha1,
2127 enum object_type *type,
2128 unsigned long *size,
2129 const unsigned char **replacement)
ac939109 2130{
68095570 2131 const unsigned char *repl = lookup_replace_object(sha1);
3ba7a065 2132 void *data;
e8b15e61 2133 char *path;
b6c4cecc
JH
2134 const struct packed_git *p;
2135
3ba7a065
JH
2136 errno = 0;
2137 data = read_object(repl, type, size);
b6c4cecc
JH
2138 if (data) {
2139 if (replacement)
2140 *replacement = repl;
2141 return data;
2142 }
68095570 2143
25f3af3f 2144 if (errno && errno != ENOENT)
3ba7a065
JH
2145 die_errno("failed to read object %s", sha1_to_hex(sha1));
2146
68095570 2147 /* die if we replaced an object with one that does not exist */
b6c4cecc 2148 if (repl != sha1)
68095570
CC
2149 die("replacement %s not found for %s",
2150 sha1_to_hex(repl), sha1_to_hex(sha1));
2151
b6c4cecc
JH
2152 if (has_loose_object(repl)) {
2153 path = sha1_file_name(sha1);
2154 die("loose object %s (stored in %s) is corrupt",
2155 sha1_to_hex(repl), path);
e8b15e61 2156 }
68095570 2157
b6c4cecc
JH
2158 if ((p = has_packed_and_bad(repl)) != NULL)
2159 die("packed object %s (stored in %s) is corrupt",
2160 sha1_to_hex(repl), p->pack_name);
f5552aee 2161
b6c4cecc 2162 return NULL;
ac939109
NP
2163}
2164
40469ee9 2165void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2166 const char *required_type_name,
40469ee9
JH
2167 unsigned long *size,
2168 unsigned char *actual_sha1_return)
f4913f91 2169{
21666f1a 2170 enum object_type type, required_type;
f4913f91
JH
2171 void *buffer;
2172 unsigned long isize;
40469ee9 2173 unsigned char actual_sha1[20];
f4913f91 2174
21666f1a 2175 required_type = type_from_string(required_type_name);
e702496e 2176 hashcpy(actual_sha1, sha1);
40469ee9
JH
2177 while (1) {
2178 int ref_length = -1;
2179 const char *ref_type = NULL;
f4913f91 2180
21666f1a 2181 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2182 if (!buffer)
2183 return NULL;
21666f1a 2184 if (type == required_type) {
40469ee9
JH
2185 *size = isize;
2186 if (actual_sha1_return)
e702496e 2187 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2188 return buffer;
2189 }
2190 /* Handle references */
21666f1a 2191 else if (type == OBJ_COMMIT)
40469ee9 2192 ref_type = "tree ";
21666f1a 2193 else if (type == OBJ_TAG)
40469ee9
JH
2194 ref_type = "object ";
2195 else {
2196 free(buffer);
2197 return NULL;
2198 }
2199 ref_length = strlen(ref_type);
f4913f91 2200
50974ec9
MK
2201 if (ref_length + 40 > isize ||
2202 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2203 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2204 free(buffer);
2205 return NULL;
2206 }
1cf58e72 2207 free(buffer);
40469ee9
JH
2208 /* Now we have the ID of the referred-to object in
2209 * actual_sha1. Check again. */
f4913f91 2210 }
f4913f91
JH
2211}
2212
ce9fbf16 2213static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2214 const char *type, unsigned char *sha1,
d65a16f6 2215 char *hdr, int *hdrlen)
d410c0f5 2216{
9126f009 2217 git_SHA_CTX c;
d410c0f5
JH
2218
2219 /* Generate the header */
d65a16f6 2220 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
2221
2222 /* Sha1.. */
9126f009
NP
2223 git_SHA1_Init(&c);
2224 git_SHA1_Update(&c, hdr, *hdrlen);
2225 git_SHA1_Update(&c, buf, len);
2226 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2227}
2228
230f1322 2229/*
5a688fe4
JH
2230 * Move the just written object into its final resting place.
2231 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2232 * "moving" is only a part of what it does, when no patch between
2233 * master to pu changes the call sites of this function.
230f1322 2234 */
839837b9 2235int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322 2236{
e32c0a9c 2237 int ret = 0;
5a688fe4 2238
348df166 2239 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
2240 goto try_rename;
2241 else if (link(tmpfile, filename))
e32c0a9c 2242 ret = errno;
7ebb6fca
LT
2243
2244 /*
2245 * Coda hack - coda doesn't like cross-directory links,
2246 * so we fall back to a rename, which will mean that it
2247 * won't be able to check collisions, but that's not a
2248 * big deal.
2249 *
2250 * The same holds for FAT formatted media.
2251 *
3be1f18e 2252 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
2253 * left to unlink.
2254 */
2255 if (ret && ret != EEXIST) {
be66a6c4 2256 try_rename:
7ebb6fca 2257 if (!rename(tmpfile, filename))
3be1f18e 2258 goto out;
9e48b389 2259 ret = errno;
230f1322 2260 }
691f1a28 2261 unlink_or_warn(tmpfile);
230f1322
LT
2262 if (ret) {
2263 if (ret != EEXIST) {
916d081b 2264 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
230f1322
LT
2265 }
2266 /* FIXME!!! Collision check here ? */
2267 }
2268
3be1f18e 2269out:
5256b006 2270 if (adjust_shared_perm(filename))
5a688fe4 2271 return error("unable to set permission to '%s'", filename);
230f1322
LT
2272 return 0;
2273}
2274
4d548150
LT
2275static int write_buffer(int fd, const void *buf, size_t len)
2276{
d34cf19b 2277 if (write_in_full(fd, buf, len) < 0)
93822c22 2278 return error("file write error (%s)", strerror(errno));
4d548150
LT
2279 return 0;
2280}
2281
ce9fbf16 2282int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2283 unsigned char *sha1)
2284{
d65a16f6 2285 char hdr[32];
abdc3fc8
RS
2286 int hdrlen;
2287 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2288 return 0;
2289}
2290
e9039dd3
LT
2291/* Finalize a file on disk, and close it. */
2292static void close_sha1_file(int fd)
2293{
aafe9fba
LT
2294 if (fsync_object_files)
2295 fsync_or_die(fd, "sha1 file");
e9039dd3 2296 if (close(fd) != 0)
d824cbba 2297 die_errno("error when closing sha1 file");
e9039dd3
LT
2298}
2299
5723fe7e
LT
2300/* Size of directory component, including the ending '/' */
2301static inline int directory_size(const char *filename)
2302{
2303 const char *s = strrchr(filename, '/');
2304 if (!s)
2305 return 0;
2306 return s - filename + 1;
2307}
2308
2309/*
2310 * This creates a temporary file in the same directory as the final
2311 * 'filename'
2312 *
2313 * We want to avoid cross-directory filename renames, because those
2314 * can have problems on various filesystems (FAT, NFS, Coda).
2315 */
2316static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2317{
2318 int fd, dirlen = directory_size(filename);
2319
2320 if (dirlen + 20 > bufsiz) {
2321 errno = ENAMETOOLONG;
2322 return -1;
2323 }
2324 memcpy(buffer, filename, dirlen);
2325 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
5256b006 2326 fd = git_mkstemp_mode(buffer, 0444);
cbacbf4e 2327 if (fd < 0 && dirlen && errno == ENOENT) {
5723fe7e 2328 /* Make sure the directory exists */
6ff6af62 2329 memcpy(buffer, filename, dirlen);
5723fe7e 2330 buffer[dirlen-1] = 0;
3bfaf018 2331 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
5723fe7e
LT
2332 return -1;
2333
2334 /* Try again */
2335 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
5256b006 2336 fd = git_mkstemp_mode(buffer, 0444);
5723fe7e
LT
2337 }
2338 return fd;
2339}
2340
bbac7311 2341static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 2342 const void *buf, unsigned long len, time_t mtime)
0fcfd160 2343{
915308b1 2344 int fd, ret;
9892beba 2345 unsigned char compressed[4096];
0fcfd160 2346 z_stream stream;
748af44c
NP
2347 git_SHA_CTX c;
2348 unsigned char parano_sha1[20];
706bc531 2349 char *filename;
aac17941 2350 static char tmpfile[PATH_MAX];
a44c9a5e 2351
972a9155 2352 filename = sha1_file_name(sha1);
5723fe7e 2353 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
f2e872aa 2354 while (fd < 0 && errno == EMFILE && unuse_one_window(NULL, -1))
fd73ccf2 2355 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
aac17941 2356 if (fd < 0) {
35243577 2357 if (errno == EACCES)
916d081b
PB
2358 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2359 else
2360 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
aac17941
LT
2361 }
2362
0fcfd160
LT
2363 /* Set it up */
2364 memset(&stream, 0, sizeof(stream));
12f6c308 2365 deflateInit(&stream, zlib_compression_level);
0fcfd160 2366 stream.next_out = compressed;
9892beba 2367 stream.avail_out = sizeof(compressed);
748af44c 2368 git_SHA1_Init(&c);
a44c9a5e
LT
2369
2370 /* First header.. */
d65a16f6 2371 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2372 stream.avail_in = hdrlen;
726f852b
NP
2373 while (deflate(&stream, 0) == Z_OK)
2374 /* nothing */;
748af44c 2375 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
2376
2377 /* Then the data itself.. */
c00e657d 2378 stream.next_in = (void *)buf;
a44c9a5e 2379 stream.avail_in = len;
9892beba 2380 do {
748af44c 2381 unsigned char *in0 = stream.next_in;
9892beba 2382 ret = deflate(&stream, Z_FINISH);
748af44c 2383 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
2384 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2385 die("unable to write sha1 file");
2386 stream.next_out = compressed;
2387 stream.avail_out = sizeof(compressed);
2388 } while (ret == Z_OK);
2389
ac54c277
LT
2390 if (ret != Z_STREAM_END)
2391 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
ac54c277
LT
2392 ret = deflateEnd(&stream);
2393 if (ret != Z_OK)
2394 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
2395 git_SHA1_Final(parano_sha1, &c);
2396 if (hashcmp(sha1, parano_sha1) != 0)
2397 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 2398
e9039dd3 2399 close_sha1_file(fd);
0fcfd160 2400
bbac7311
NP
2401 if (mtime) {
2402 struct utimbuf utb;
2403 utb.actime = mtime;
2404 utb.modtime = mtime;
2405 if (utime(tmpfile, &utb) < 0)
2406 warning("failed utime() on %s: %s",
2407 tmpfile, strerror(errno));
2408 }
2409
230f1322 2410 return move_temp_to_file(tmpfile, filename);
0fcfd160 2411}
8237b185 2412
c00e657d 2413int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
bbac7311
NP
2414{
2415 unsigned char sha1[20];
2416 char hdr[32];
2417 int hdrlen;
2418
2419 /* Normally if we have it in the pack then we do not bother writing
2420 * it out into .git/objects/??/?{38} file.
2421 */
2422 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2423 if (returnsha1)
2424 hashcpy(returnsha1, sha1);
2425 if (has_sha1_file(sha1))
2426 return 0;
2427 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2428}
2429
2430int force_object_loose(const unsigned char *sha1, time_t mtime)
2431{
bbac7311
NP
2432 void *buf;
2433 unsigned long len;
2434 enum object_type type;
2435 char hdr[32];
2436 int hdrlen;
1fb23e65 2437 int ret;
bbac7311 2438
c529d75a 2439 if (has_loose_object(sha1))
bbac7311
NP
2440 return 0;
2441 buf = read_packed_sha1(sha1, &type, &len);
2442 if (!buf)
2443 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2444 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1fb23e65
BS
2445 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2446 free(buf);
2447
2448 return ret;
bbac7311
NP
2449}
2450
bf592c50
DB
2451int has_pack_index(const unsigned char *sha1)
2452{
2453 struct stat st;
2454 if (stat(sha1_pack_index_name(sha1), &st))
2455 return 0;
2456 return 1;
bf592c50
DB
2457}
2458
cd673c1f
JH
2459int has_sha1_pack(const unsigned char *sha1)
2460{
2461 struct pack_entry e;
2462 return find_pack_entry(sha1, &e);
2463}
2464
8237b185
DB
2465int has_sha1_file(const unsigned char *sha1)
2466{
1f688557
JH
2467 struct pack_entry e;
2468
cd673c1f 2469 if (find_pack_entry(sha1, &e))
1f688557 2470 return 1;
c529d75a 2471 return has_loose_object(sha1);
8237b185 2472}
74400e71 2473
43df4f86
DP
2474static int index_mem(unsigned char *sha1, void *buf, size_t size,
2475 int write_object, enum object_type type, const char *path)
e7332f96 2476{
6c510bee 2477 int ret, re_allocated = 0;
74400e71 2478
7672db20 2479 if (!type)
edaec3fb 2480 type = OBJ_BLOB;
6c510bee
LT
2481
2482 /*
2483 * Convert blobs to git internal format
2484 */
43df4f86 2485 if ((type == OBJ_BLOB) && path) {
f285a2d7 2486 struct strbuf nbuf = STRBUF_INIT;
21e5ad50
SP
2487 if (convert_to_git(path, buf, size, &nbuf,
2488 write_object ? safe_crlf : 0)) {
b315c5c0 2489 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
2490 re_allocated = 1;
2491 }
2492 }
2493
7672db20 2494 if (write_object)
edaec3fb 2495 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 2496 else
edaec3fb 2497 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 2498 if (re_allocated)
6c510bee 2499 free(buf);
43df4f86
DP
2500 return ret;
2501}
2502
ea68b0ce
DP
2503#define SMALL_FILE_SIZE (32*1024)
2504
43df4f86
DP
2505int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2506 enum object_type type, const char *path)
2507{
2508 int ret;
2509 size_t size = xsize_t(st->st_size);
2510
2511 if (!S_ISREG(st->st_mode)) {
f285a2d7 2512 struct strbuf sbuf = STRBUF_INIT;
43df4f86
DP
2513 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2514 ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2515 type, path);
2516 else
2517 ret = -1;
2518 strbuf_release(&sbuf);
08bda208
DP
2519 } else if (!size) {
2520 ret = index_mem(sha1, NULL, size, write_object, type, path);
ea68b0ce
DP
2521 } else if (size <= SMALL_FILE_SIZE) {
2522 char *buf = xmalloc(size);
2523 if (size == read_in_full(fd, buf, size))
2524 ret = index_mem(sha1, buf, size, write_object, type,
2525 path);
2526 else
2527 ret = error("short read %s", strerror(errno));
2528 free(buf);
08bda208 2529 } else {
43df4f86
DP
2530 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2531 ret = index_mem(sha1, buf, size, write_object, type, path);
aac17941 2532 munmap(buf, size);
08bda208 2533 }
43df4f86 2534 close(fd);
aac17941 2535 return ret;
74400e71 2536}
ec1fcc16
JH
2537
2538int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2539{
2540 int fd;
b760d3aa 2541 struct strbuf sb = STRBUF_INIT;
ec1fcc16
JH
2542
2543 switch (st->st_mode & S_IFMT) {
2544 case S_IFREG:
2545 fd = open(path, O_RDONLY);
2546 if (fd < 0)
2547 return error("open(\"%s\"): %s", path,
2548 strerror(errno));
53bca91a 2549 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
ec1fcc16
JH
2550 return error("%s: failed to insert into database",
2551 path);
2552 break;
2553 case S_IFLNK:
b760d3aa 2554 if (strbuf_readlink(&sb, path, st->st_size)) {
ec1fcc16 2555 char *errstr = strerror(errno);
ec1fcc16
JH
2556 return error("readlink(\"%s\"): %s", path,
2557 errstr);
2558 }
abdc3fc8 2559 if (!write_object)
b760d3aa
LT
2560 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2561 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
ec1fcc16
JH
2562 return error("%s: failed to insert into database",
2563 path);
b760d3aa 2564 strbuf_release(&sb);
ec1fcc16 2565 break;
f35a6d3b
LT
2566 case S_IFDIR:
2567 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
2568 default:
2569 return error("%s: unsupported file type", path);
2570 }
2571 return 0;
2572}
a69e5429
JH
2573
2574int read_pack_header(int fd, struct pack_header *header)
2575{
c697ad14
HO
2576 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2577 /* "eof before pack header was fully read" */
2578 return PH_ERROR_EOF;
2579
a69e5429
JH
2580 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2581 /* "protocol error (pack signature mismatch detected)" */
2582 return PH_ERROR_PACK_SIGNATURE;
2583 if (!pack_version_ok(header->hdr_version))
2584 /* "protocol error (pack version unsupported)" */
2585 return PH_ERROR_PROTOCOL;
2586 return 0;
2587}
40d52ff7
JK
2588
2589void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2590{
2591 enum object_type type = sha1_object_info(sha1, NULL);
2592 if (type < 0)
2593 die("%s is not a valid object", sha1_to_hex(sha1));
2594 if (type != expect)
2595 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2596 typename(expect));
2597}