Use git_open_noatime when accessing pack data
[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
4865d2b6
SP
38static int git_open_noatime(const char *name);
39
b2cb9425
JH
40int safe_create_leading_directories(char *path)
41{
8385abfd 42 char *pos = path + offset_1st_component(path);
67d42212
JR
43 struct stat st;
44
b2cb9425
JH
45 while (pos) {
46 pos = strchr(pos, '/');
47 if (!pos)
48 break;
5f0bdf50
JH
49 while (*++pos == '/')
50 ;
51 if (!*pos)
52 break;
53 *--pos = '\0';
67d42212
JR
54 if (!stat(path, &st)) {
55 /* path exists */
56 if (!S_ISDIR(st.st_mode)) {
b2cb9425 57 *pos = '/';
67d42212 58 return -3;
b2cb9425 59 }
457f06d6 60 }
67d42212
JR
61 else if (mkdir(path, 0777)) {
62 *pos = '/';
63 return -1;
64 }
457f06d6
JS
65 else if (adjust_shared_perm(path)) {
66 *pos = '/';
67 return -2;
68 }
b2cb9425
JH
69 *pos++ = '/';
70 }
71 return 0;
72}
723c31fe 73
8e21d63b
JK
74int safe_create_leading_directories_const(const char *path)
75{
76 /* path points to cache entries, so xstrdup before messing with it */
77 char *buf = xstrdup(path);
78 int result = safe_create_leading_directories(buf);
79 free(buf);
80 return result;
81}
82
ace1534d
JH
83static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
84{
85 int i;
86 for (i = 0; i < 20; i++) {
87 static char hex[] = "0123456789abcdef";
88 unsigned int val = sha1[i];
89 char *pos = pathbuf + i*2 + (i > 0);
90 *pos++ = hex[val >> 4];
91 *pos = hex[val & 0xf];
92 }
93}
94
0fcfd160
LT
95/*
96 * NOTE! This returns a statically allocated buffer, so you have to be
790296fd 97 * careful about using it. Do an "xstrdup()" if you need to save the
0fcfd160 98 * filename.
ace1534d
JH
99 *
100 * Also note that this returns the location for creating. Reading
101 * SHA1 file can happen from any alternate directory listed in the
d19938ab 102 * DB_ENVIRONMENT environment variable if it is not found in
ace1534d 103 * the primary object database.
0fcfd160
LT
104 */
105char *sha1_file_name(const unsigned char *sha1)
106{
560fb6a1
JK
107 static char buf[PATH_MAX];
108 const char *objdir;
109 int len;
0fcfd160 110
560fb6a1
JK
111 objdir = get_object_directory();
112 len = strlen(objdir);
113
114 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
115 if (len + 43 > PATH_MAX)
116 die("insanely long object directory %s", objdir);
117 memcpy(buf, objdir, len);
118 buf[len] = '/';
119 buf[len+3] = '/';
120 buf[len+42] = '\0';
121 fill_sha1_path(buf + len + 1, sha1);
122 return buf;
0fcfd160
LT
123}
124
633f43e1 125static char *sha1_get_pack_name(const unsigned char *sha1,
6eec46bd 126 char **name, char **base, const char *which)
bf592c50
DB
127{
128 static const char hex[] = "0123456789abcdef";
633f43e1 129 char *buf;
bf592c50
DB
130 int i;
131
633f43e1 132 if (!*base) {
bf592c50
DB
133 const char *sha1_file_directory = get_object_directory();
134 int len = strlen(sha1_file_directory);
633f43e1 135 *base = xmalloc(len + 60);
6eec46bd
JH
136 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
137 sha1_file_directory, which);
633f43e1 138 *name = *base + len + 11;
bf592c50
DB
139 }
140
633f43e1 141 buf = *name;
bf592c50
DB
142
143 for (i = 0; i < 20; i++) {
144 unsigned int val = *sha1++;
145 *buf++ = hex[val >> 4];
146 *buf++ = hex[val & 0xf];
147 }
a6080a0a 148
633f43e1 149 return *base;
bf592c50
DB
150}
151
633f43e1 152char *sha1_pack_name(const unsigned char *sha1)
bf592c50 153{
633f43e1 154 static char *name, *base;
bf592c50 155
6eec46bd 156 return sha1_get_pack_name(sha1, &name, &base, "pack");
633f43e1 157}
bf592c50 158
633f43e1
HO
159char *sha1_pack_index_name(const unsigned char *sha1)
160{
161 static char *name, *base;
a6080a0a 162
6eec46bd 163 return sha1_get_pack_name(sha1, &name, &base, "idx");
bf592c50
DB
164}
165
d5a63b99
JH
166struct alternate_object_database *alt_odb_list;
167static struct alternate_object_database **alt_odb_tail;
ace1534d 168
c2f493a4
MW
169static void read_info_alternates(const char * alternates, int depth);
170
ddd5d056
JH
171/*
172 * Prepare alternate object database registry.
d5a63b99
JH
173 *
174 * The variable alt_odb_list points at the list of struct
175 * alternate_object_database. The elements on this list come from
176 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
177 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
178 * whose contents is similar to that environment variable but can be
179 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
180 * contains "/the/directory/corresponding/to/.git/objects/...", while
181 * its name points just after the slash at the end of ".git/objects/"
182 * in the example above, and has enough space to hold 40-byte hex
183 * SHA1, an extra slash for the first level indirection, and the
184 * terminating NUL.
ddd5d056 185 */
c2f493a4 186static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
ace1534d 187{
1494e038 188 const char *objdir = get_object_directory();
c2f493a4
MW
189 struct alternate_object_database *ent;
190 struct alternate_object_database *alt;
191 /* 43 = 40-byte + 2 '/' + terminating NUL */
192 int pfxlen = len;
193 int entlen = pfxlen + 43;
ccfd3e99 194 int base_len = -1;
d5a63b99 195
85dadc38 196 if (!is_absolute_path(entry) && relative_base) {
c2f493a4
MW
197 /* Relative alt-odb */
198 if (base_len < 0)
199 base_len = strlen(relative_base) + 1;
200 entlen += base_len;
201 pfxlen += base_len;
202 }
203 ent = xmalloc(sizeof(*ent) + entlen);
204
85dadc38 205 if (!is_absolute_path(entry) && relative_base) {
c2f493a4
MW
206 memcpy(ent->base, relative_base, base_len - 1);
207 ent->base[base_len - 1] = '/';
208 memcpy(ent->base + base_len, entry, len);
209 }
210 else
211 memcpy(ent->base, entry, pfxlen);
212
213 ent->name = ent->base + pfxlen + 1;
214 ent->base[pfxlen + 3] = '/';
215 ent->base[pfxlen] = ent->base[entlen-1] = 0;
216
217 /* Detect cases where alternate disappeared */
90b4a71c 218 if (!is_directory(ent->base)) {
c2f493a4
MW
219 error("object directory %s does not exist; "
220 "check .git/objects/info/alternates.",
221 ent->base);
222 free(ent);
223 return -1;
224 }
225
226 /* Prevent the common mistake of listing the same
227 * thing twice, or object directory itself.
228 */
229 for (alt = alt_odb_list; alt; alt = alt->next) {
230 if (!memcmp(ent->base, alt->base, pfxlen)) {
231 free(ent);
232 return -1;
233 }
234 }
235 if (!memcmp(ent->base, objdir, pfxlen)) {
236 free(ent);
237 return -1;
238 }
239
240 /* add the alternate entry */
241 *alt_odb_tail = ent;
242 alt_odb_tail = &(ent->next);
243 ent->next = NULL;
244
245 /* recursively add alternates */
246 read_info_alternates(ent->base, depth + 1);
247
248 ent->base[pfxlen] = '/';
249
250 return 0;
251}
252
253static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
254 const char *relative_base, int depth)
255{
256 const char *cp, *last;
257
258 if (depth > 5) {
259 error("%s: ignoring alternate object stores, nesting too deep.",
260 relative_base);
261 return;
262 }
263
d5a63b99 264 last = alt;
9577e7e3
JH
265 while (last < ep) {
266 cp = last;
267 if (cp < ep && *cp == '#') {
268 while (cp < ep && *cp != sep)
269 cp++;
270 last = cp + 1;
271 continue;
272 }
c2f493a4
MW
273 while (cp < ep && *cp != sep)
274 cp++;
d5a63b99 275 if (last != cp) {
85dadc38 276 if (!is_absolute_path(last) && depth) {
c2f493a4
MW
277 error("%s: ignoring relative alternate object store %s",
278 relative_base, last);
279 } else {
280 link_alt_odb_entry(last, cp - last,
281 relative_base, depth);
1494e038 282 }
d5a63b99 283 }
9577e7e3 284 while (cp < ep && *cp == sep)
d5a63b99
JH
285 cp++;
286 last = cp;
9577e7e3 287 }
d5a63b99
JH
288}
289
c2f493a4 290static void read_info_alternates(const char * relative_base, int depth)
d5a63b99 291{
9577e7e3 292 char *map;
dc49cd76 293 size_t mapsz;
d5a63b99 294 struct stat st;
9cb18f56
JM
295 const char alt_file_name[] = "info/alternates";
296 /* Given that relative_base is no longer than PATH_MAX,
297 ensure that "path" has enough space to append "/", the
298 file name, "info/alternates", and a trailing NUL. */
299 char path[PATH_MAX + 1 + sizeof alt_file_name];
c2f493a4 300 int fd;
d5a63b99 301
9cb18f56 302 sprintf(path, "%s/%s", relative_base, alt_file_name);
4865d2b6 303 fd = git_open_noatime(path);
d5a63b99
JH
304 if (fd < 0)
305 return;
306 if (fstat(fd, &st) || (st.st_size == 0)) {
307 close(fd);
9a217f2a 308 return;
ace1534d 309 }
dc49cd76
SP
310 mapsz = xsize_t(st.st_size);
311 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 312 close(fd);
d5a63b99 313
dc49cd76 314 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
c2f493a4 315
dc49cd76 316 munmap(map, mapsz);
ace1534d
JH
317}
318
bef70b22
DB
319void add_to_alternates_file(const char *reference)
320{
321 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
acd3b9ec 322 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
bef70b22
DB
323 char *alt = mkpath("%s/objects\n", reference);
324 write_or_die(fd, alt, strlen(alt));
325 if (commit_lock_file(lock))
326 die("could not close alternates file");
327 if (alt_odb_tail)
328 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
329}
330
d79796bc
JH
331void foreach_alt_odb(alt_odb_fn fn, void *cb)
332{
333 struct alternate_object_database *ent;
334
335 prepare_alt_odb();
336 for (ent = alt_odb_list; ent; ent = ent->next)
337 if (fn(ent, cb))
338 return;
339}
340
c2f493a4
MW
341void prepare_alt_odb(void)
342{
554fe20d 343 const char *alt;
c2f493a4 344
7dc24aa5
SP
345 if (alt_odb_tail)
346 return;
347
c2f493a4
MW
348 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
349 if (!alt) alt = "";
350
c2f493a4 351 alt_odb_tail = &alt_odb_list;
80ba074f 352 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
353
354 read_info_alternates(get_object_directory(), 0);
355}
356
0f4dc14a 357static int has_loose_object_local(const unsigned char *sha1)
ace1534d 358{
ace1534d 359 char *name = sha1_file_name(sha1);
0f4dc14a
BC
360 return !access(name, F_OK);
361}
ace1534d 362
0f4dc14a
BC
363int has_loose_object_nonlocal(const unsigned char *sha1)
364{
365 struct alternate_object_database *alt;
9a217f2a 366 prepare_alt_odb();
d5a63b99 367 for (alt = alt_odb_list; alt; alt = alt->next) {
0f4dc14a 368 fill_sha1_path(alt->name, sha1);
c529d75a
LT
369 if (!access(alt->base, F_OK))
370 return 1;
ace1534d 371 }
c529d75a 372 return 0;
ace1534d
JH
373}
374
0f4dc14a
BC
375static int has_loose_object(const unsigned char *sha1)
376{
377 return has_loose_object_local(sha1) ||
378 has_loose_object_nonlocal(sha1);
379}
380
60bb8b14 381static unsigned int pack_used_ctr;
a53128b6
SP
382static unsigned int pack_mmap_calls;
383static unsigned int peak_pack_open_windows;
384static unsigned int pack_open_windows;
385static size_t peak_pack_mapped;
60bb8b14 386static size_t pack_mapped;
9a217f2a 387struct packed_git *packed_git;
1f688557 388
b79d18c9 389void pack_report(void)
a53128b6
SP
390{
391 fprintf(stderr,
e05db0fd
PR
392 "pack_report: getpagesize() = %10" SZ_FMT "\n"
393 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
394 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
395 sz_fmt(getpagesize()),
396 sz_fmt(packed_git_window_size),
397 sz_fmt(packed_git_limit));
a53128b6
SP
398 fprintf(stderr,
399 "pack_report: pack_used_ctr = %10u\n"
400 "pack_report: pack_mmap_calls = %10u\n"
401 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
402 "pack_report: pack_mapped = "
403 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
404 pack_used_ctr,
405 pack_mmap_calls,
406 pack_open_windows, peak_pack_open_windows,
9e42d6a1 407 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
408}
409
42873078 410static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
411{
412 void *idx_map;
42873078 413 struct pack_idx_header *hdr;
dc49cd76 414 size_t idx_size;
74e34e1f 415 uint32_t version, nr, i, *index;
4865d2b6 416 int fd = git_open_noatime(path);
1f688557 417 struct stat st;
42873078 418
1f688557
JH
419 if (fd < 0)
420 return -1;
421 if (fstat(fd, &st)) {
422 close(fd);
423 return -1;
424 }
dc49cd76 425 idx_size = xsize_t(st.st_size);
2d88451b
SP
426 if (idx_size < 4 * 256 + 20 + 20) {
427 close(fd);
428 return error("index file %s is too small", path);
429 }
c4712e45 430 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 431 close(fd);
1f688557 432
42873078
NP
433 hdr = idx_map;
434 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
435 version = ntohl(hdr->idx_version);
436 if (version < 2 || version > 2) {
437 munmap(idx_map, idx_size);
6e1c2344 438 return error("index file %s is version %"PRIu32
74e34e1f
NP
439 " and is not supported by this binary"
440 " (try upgrading GIT to a newer version)",
441 path, version);
442 }
443 } else
444 version = 1;
df1b059d 445
1f688557 446 nr = 0;
42873078 447 index = idx_map;
74e34e1f
NP
448 if (version > 1)
449 index += 2; /* skip index header */
1f688557 450 for (i = 0; i < 256; i++) {
326bf396 451 uint32_t n = ntohl(index[i]);
2d88451b
SP
452 if (n < nr) {
453 munmap(idx_map, idx_size);
df1b059d 454 return error("non-monotonic index %s", path);
2d88451b 455 }
1f688557
JH
456 nr = n;
457 }
458
74e34e1f
NP
459 if (version == 1) {
460 /*
461 * Total size:
462 * - 256 index entries 4 bytes each
463 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
464 * - 20-byte SHA1 of the packfile
465 * - 20-byte SHA1 file checksum
466 */
467 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
468 munmap(idx_map, idx_size);
eef427a0 469 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
470 }
471 } else if (version == 2) {
472 /*
473 * Minimum size:
474 * - 8 bytes of header
475 * - 256 index entries 4 bytes each
476 * - 20-byte sha1 entry * nr
477 * - 4-byte crc entry * nr
478 * - 4-byte offset entry * nr
479 * - 20-byte SHA1 of the packfile
480 * - 20-byte SHA1 file checksum
481 * And after the 4-byte offset table might be a
482 * variable sized table containing 8-byte entries
483 * for offsets larger than 2^31.
484 */
485 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
486 unsigned long max_size = min_size;
487 if (nr)
488 max_size += (nr - 1)*8;
489 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 490 munmap(idx_map, idx_size);
eef427a0 491 return error("wrong index v2 file size in %s", path);
74e34e1f 492 }
7109c889
JH
493 if (idx_size != min_size &&
494 /*
495 * make sure we can deal with large pack offsets.
496 * 31-bit signed offset won't be enough, neither
497 * 32-bit unsigned one will be.
498 */
499 (sizeof(off_t) <= 4)) {
500 munmap(idx_map, idx_size);
501 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 502 }
2d88451b 503 }
1f688557 504
74e34e1f 505 p->index_version = version;
42873078
NP
506 p->index_data = idx_map;
507 p->index_size = idx_size;
57059091 508 p->num_objects = nr;
1f688557
JH
509 return 0;
510}
511
bc8e478a 512int open_pack_index(struct packed_git *p)
d079837e
SP
513{
514 char *idx_name;
515 int ret;
516
517 if (p->index_data)
518 return 0;
519
520 idx_name = xstrdup(p->pack_name);
521 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
522 ret = check_packed_git_idx(idx_name, p);
523 free(idx_name);
524 return ret;
525}
526
11daf39b
SP
527static void scan_windows(struct packed_git *p,
528 struct packed_git **lru_p,
529 struct pack_window **lru_w,
530 struct pack_window **lru_l)
1f688557 531{
11daf39b
SP
532 struct pack_window *w, *w_l;
533
534 for (w_l = NULL, w = p->windows; w; w = w->next) {
535 if (!w->inuse_cnt) {
536 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
537 *lru_p = p;
538 *lru_w = w;
539 *lru_l = w_l;
54044bf8 540 }
54044bf8 541 }
11daf39b 542 w_l = w;
f9253394 543 }
11daf39b
SP
544}
545
d1efefa4 546static int unuse_one_window(struct packed_git *current, int keep_fd)
11daf39b
SP
547{
548 struct packed_git *p, *lru_p = NULL;
549 struct pack_window *lru_w = NULL, *lru_l = NULL;
550
551 if (current)
552 scan_windows(current, &lru_p, &lru_w, &lru_l);
553 for (p = packed_git; p; p = p->next)
554 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
555 if (lru_p) {
556 munmap(lru_w->base, lru_w->len);
557 pack_mapped -= lru_w->len;
558 if (lru_l)
559 lru_l->next = lru_w->next;
560 else {
561 lru_p->windows = lru_w->next;
d1efefa4 562 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
54044bf8
SP
563 close(lru_p->pack_fd);
564 lru_p->pack_fd = -1;
565 }
566 }
567 free(lru_w);
a53128b6 568 pack_open_windows--;
54044bf8
SP
569 return 1;
570 }
571 return 0;
f9253394
JH
572}
573
d1efefa4 574void release_pack_memory(size_t need, int fd)
97bfeb34
SP
575{
576 size_t cur = pack_mapped;
d1efefa4 577 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
97bfeb34
SP
578 ; /* nothing */
579}
580
c9ced051
SP
581void close_pack_windows(struct packed_git *p)
582{
583 while (p->windows) {
584 struct pack_window *w = p->windows;
585
586 if (w->inuse_cnt)
587 die("pack '%s' still has open windows to it",
588 p->pack_name);
589 munmap(w->base, w->len);
590 pack_mapped -= w->len;
591 pack_open_windows--;
592 p->windows = w->next;
593 free(w);
594 }
595}
596
03e79c88 597void unuse_pack(struct pack_window **w_cursor)
f9253394 598{
03e79c88
SP
599 struct pack_window *w = *w_cursor;
600 if (w) {
601 w->inuse_cnt--;
602 *w_cursor = NULL;
603 }
1f688557
JH
604}
605
fa5fc15d
SP
606void close_pack_index(struct packed_git *p)
607{
608 if (p->index_data) {
609 munmap((void *)p->index_data, p->index_size);
610 p->index_data = NULL;
611 }
612}
613
c74faea1
NP
614/*
615 * This is used by git-repack in case a newly created pack happens to
616 * contain the same set of objects as an existing one. In that case
617 * the resulting file might be different even if its name would be the
618 * same. It is best to close any reference to the old pack before it is
619 * replaced on disk. Of course no index pointers nor windows for given pack
620 * must subsist at this point. If ever objects from this pack are requested
621 * again, the new version of the pack will be reinitialized through
622 * reprepare_packed_git().
623 */
624void free_pack_by_name(const char *pack_name)
625{
626 struct packed_git *p, **pp = &packed_git;
627
628 while (*pp) {
629 p = *pp;
630 if (strcmp(pack_name, p->pack_name) == 0) {
fa3a0c94 631 clear_delta_base_cache();
c74faea1
NP
632 close_pack_windows(p);
633 if (p->pack_fd != -1)
634 close(p->pack_fd);
fa5fc15d 635 close_pack_index(p);
c74faea1
NP
636 free(p->bad_object_sha1);
637 *pp = p->next;
638 free(p);
639 return;
640 }
641 pp = &p->next;
642 }
643}
644
3cf8b462
SP
645/*
646 * Do not call this directly as this leaks p->pack_fd on error return;
647 * call open_packed_git() instead.
648 */
649static int open_packed_git_1(struct packed_git *p)
1f688557 650{
9bc879c1
SP
651 struct stat st;
652 struct pack_header hdr;
653 unsigned char sha1[20];
654 unsigned char *idx_sha1;
2c039da8 655 long fd_flag;
9bc879c1 656
d079837e
SP
657 if (!p->index_data && open_pack_index(p))
658 return error("packfile %s index unavailable", p->pack_name);
659
4865d2b6 660 p->pack_fd = git_open_noatime(p->pack_name);
fd73ccf2 661 while (p->pack_fd < 0 && errno == EMFILE && unuse_one_window(p, -1))
4865d2b6 662 p->pack_fd = git_open_noatime(p->pack_name);
9bc879c1 663 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 664 return -1;
9bc879c1
SP
665
666 /* If we created the struct before we had the pack we lack size. */
bf592c50 667 if (!p->pack_size) {
bf592c50 668 if (!S_ISREG(st.st_mode))
072db278 669 return error("packfile %s not a regular file", p->pack_name);
bf592c50 670 p->pack_size = st.st_size;
9bc879c1 671 } else if (p->pack_size != st.st_size)
072db278 672 return error("packfile %s size changed", p->pack_name);
9bc879c1 673
2c039da8
JH
674 /* We leave these file descriptors open with sliding mmap;
675 * there is no point keeping them open across exec(), though.
676 */
677 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
678 if (fd_flag < 0)
072db278 679 return error("cannot determine file descriptor flags");
2c039da8
JH
680 fd_flag |= FD_CLOEXEC;
681 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 682 return error("cannot set FD_CLOEXEC");
2c039da8 683
9bc879c1 684 /* Verify we recognize this pack file format. */
e6e2bd62 685 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 686 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 687 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 688 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 689 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
690 return error("packfile %s is version %"PRIu32" and not"
691 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
692 p->pack_name, ntohl(hdr.hdr_version));
693
694 /* Verify the pack matches its index. */
57059091 695 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
696 return error("packfile %s claims to have %"PRIu32" objects"
697 " while index indicates %"PRIu32" objects",
57059091
NP
698 p->pack_name, ntohl(hdr.hdr_entries),
699 p->num_objects);
9bc879c1 700 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 701 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 702 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 703 return error("packfile %s signature is unavailable", p->pack_name);
42873078 704 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 705 if (hashcmp(sha1, idx_sha1))
072db278
SP
706 return error("packfile %s does not match index", p->pack_name);
707 return 0;
9bc879c1
SP
708}
709
3cf8b462
SP
710static int open_packed_git(struct packed_git *p)
711{
712 if (!open_packed_git_1(p))
713 return 0;
714 if (p->pack_fd != -1) {
715 close(p->pack_fd);
716 p->pack_fd = -1;
717 }
718 return -1;
719}
720
c4001d92 721static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
722{
723 /* We must promise at least 20 bytes (one hash) after the
724 * offset is available from this window, otherwise the offset
725 * is not actually in this window and a different window (which
726 * has that one hash excess) must be used. This is to support
727 * the object header and delta base parsing routines below.
728 */
729 off_t win_off = win->offset;
730 return win_off <= offset
731 && (offset + 20) <= (win_off + win->len);
732}
733
4b25d091 734unsigned char *use_pack(struct packed_git *p,
03e79c88 735 struct pack_window **w_cursor,
c4001d92 736 off_t offset,
03e79c88 737 unsigned int *left)
9bc879c1 738{
60bb8b14 739 struct pack_window *win = *w_cursor;
03e79c88 740
072db278
SP
741 if (p->pack_fd == -1 && open_packed_git(p))
742 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14 743
a9d98a14 744 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
745 * pointless to ask for an offset into the middle of that
746 * hash, and the in_window function above wouldn't match
747 * don't allow an offset too close to the end of the file.
748 */
749 if (offset > (p->pack_size - 20))
750 die("offset beyond end of packfile (truncated pack?)");
751
752 if (!win || !in_window(win, offset)) {
753 if (win)
754 win->inuse_cnt--;
755 for (win = p->windows; win; win = win->next) {
756 if (in_window(win, offset))
757 break;
758 }
759 if (!win) {
78a28df9 760 size_t window_align = packed_git_window_size / 2;
dc49cd76 761 off_t len;
60bb8b14 762 win = xcalloc(1, sizeof(*win));
78a28df9 763 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
764 len = p->pack_size - win->offset;
765 if (len > packed_git_window_size)
766 len = packed_git_window_size;
767 win->len = (size_t)len;
60bb8b14 768 pack_mapped += win->len;
11daf39b 769 while (packed_git_limit < pack_mapped
d1efefa4 770 && unuse_one_window(p, p->pack_fd))
60bb8b14 771 ; /* nothing */
c4712e45 772 win->base = xmmap(NULL, win->len,
60bb8b14
SP
773 PROT_READ, MAP_PRIVATE,
774 p->pack_fd, win->offset);
775 if (win->base == MAP_FAILED)
73b4e4be
SP
776 die("packfile %s cannot be mapped: %s",
777 p->pack_name,
778 strerror(errno));
a53128b6
SP
779 pack_mmap_calls++;
780 pack_open_windows++;
781 if (pack_mapped > peak_pack_mapped)
782 peak_pack_mapped = pack_mapped;
783 if (pack_open_windows > peak_pack_open_windows)
784 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
785 win->next = p->windows;
786 p->windows = win;
787 }
1f688557 788 }
03e79c88
SP
789 if (win != *w_cursor) {
790 win->last_used = pack_used_ctr++;
791 win->inuse_cnt++;
792 *w_cursor = win;
793 }
60bb8b14 794 offset -= win->offset;
03e79c88 795 if (left)
dc49cd76 796 *left = win->len - xsize_t(offset);
03e79c88 797 return win->base + offset;
1f688557
JH
798}
799
27d69a46
NP
800static struct packed_git *alloc_packed_git(int extra)
801{
802 struct packed_git *p = xmalloc(sizeof(*p) + extra);
803 memset(p, 0, sizeof(*p));
804 p->pack_fd = -1;
805 return p;
806}
807
42873078 808struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557
JH
809{
810 struct stat st;
27d69a46 811 struct packed_git *p = alloc_packed_git(path_len + 2);
1f688557 812
42873078
NP
813 /*
814 * Make sure a corresponding .pack file exists and that
815 * the index looks sane.
816 */
817 path_len -= strlen(".idx");
27d69a46
NP
818 if (path_len < 1) {
819 free(p);
1f688557 820 return NULL;
27d69a46 821 }
42873078 822 memcpy(p->pack_name, path, path_len);
8d25931d
BC
823
824 strcpy(p->pack_name + path_len, ".keep");
825 if (!access(p->pack_name, F_OK))
826 p->pack_keep = 1;
827
42873078 828 strcpy(p->pack_name + path_len, ".pack");
d079837e 829 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 830 free(p);
1f688557
JH
831 return NULL;
832 }
42873078 833
1f688557
JH
834 /* ok, it looks sane as far as we can check without
835 * actually mapping the pack file.
836 */
1f688557 837 p->pack_size = st.st_size;
9d835df2 838 p->pack_local = local;
b867092f 839 p->mtime = st.st_mtime;
42873078
NP
840 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
841 hashclr(p->sha1);
1f688557
JH
842 return p;
843}
844
7b64469a 845struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 846{
42873078 847 const char *path = sha1_pack_name(sha1);
27d69a46 848 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
bf592c50 849
27d69a46
NP
850 strcpy(p->pack_name, path);
851 hashcpy(p->sha1, sha1);
42873078
NP
852 if (check_packed_git_idx(idx_path, p)) {
853 free(p);
bf592c50 854 return NULL;
42873078 855 }
bf592c50 856
bf592c50
DB
857 return p;
858}
859
860void install_packed_git(struct packed_git *pack)
861{
862 pack->next = packed_git;
863 packed_git = pack;
864}
865
9d835df2 866static void prepare_packed_git_one(char *objdir, int local)
1f688557 867{
9cb18f56
JM
868 /* Ensure that this buffer is large enough so that we can
869 append "/pack/" without clobbering the stack even if
870 strlen(objdir) were PATH_MAX. */
871 char path[PATH_MAX + 1 + 4 + 1 + 1];
1f688557
JH
872 int len;
873 DIR *dir;
874 struct dirent *de;
875
876 sprintf(path, "%s/pack", objdir);
877 len = strlen(path);
878 dir = opendir(path);
fd73ccf2
JS
879 while (!dir && errno == EMFILE && unuse_one_window(packed_git, -1))
880 dir = opendir(path);
b5b16990 881 if (!dir) {
26125f6b 882 if (errno != ENOENT)
bd2afde8 883 error("unable to open object pack directory: %s: %s",
26125f6b 884 path, strerror(errno));
1f688557 885 return;
b5b16990 886 }
1f688557
JH
887 path[len++] = '/';
888 while ((de = readdir(dir)) != NULL) {
889 int namelen = strlen(de->d_name);
890 struct packed_git *p;
891
5bb1cda5 892 if (!has_extension(de->d_name, ".idx"))
1f688557
JH
893 continue;
894
9cb18f56
JM
895 if (len + namelen + 1 > sizeof(path))
896 continue;
897
54a15a8d 898 /* Don't reopen a pack we already have. */
1f688557 899 strcpy(path + len, de->d_name);
86f7780c
JK
900 for (p = packed_git; p; p = p->next) {
901 if (!memcmp(path, p->pack_name, len + namelen - 4))
902 break;
903 }
904 if (p)
905 continue;
54a15a8d
SP
906 /* See if it really is a valid .idx file with corresponding
907 * .pack file that we can map.
908 */
9d835df2 909 p = add_packed_git(path, len + namelen, local);
1f688557
JH
910 if (!p)
911 continue;
625e9421 912 install_packed_git(p);
1f688557 913 }
5b35bcd5 914 closedir(dir);
1f688557
JH
915}
916
b867092f
JH
917static int sort_pack(const void *a_, const void *b_)
918{
919 struct packed_git *a = *((struct packed_git **)a_);
920 struct packed_git *b = *((struct packed_git **)b_);
921 int st;
922
923 /*
924 * Local packs tend to contain objects specific to our
925 * variant of the project than remote ones. In addition,
926 * remote ones could be on a network mounted filesystem.
927 * Favor local ones for these reasons.
928 */
929 st = a->pack_local - b->pack_local;
930 if (st)
931 return -st;
932
933 /*
934 * Younger packs tend to contain more recent objects,
935 * and more recent objects tend to get accessed more
936 * often.
937 */
938 if (a->mtime < b->mtime)
939 return 1;
940 else if (a->mtime == b->mtime)
941 return 0;
942 return -1;
943}
944
945static void rearrange_packed_git(void)
946{
947 struct packed_git **ary, *p;
948 int i, n;
949
950 for (n = 0, p = packed_git; p; p = p->next)
951 n++;
952 if (n < 2)
953 return;
954
955 /* prepare an array of packed_git for easier sorting */
956 ary = xcalloc(n, sizeof(struct packed_git *));
957 for (n = 0, p = packed_git; p; p = p->next)
958 ary[n++] = p;
959
960 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
961
962 /* link them back again */
963 for (i = 0; i < n - 1; i++)
964 ary[i]->next = ary[i + 1];
965 ary[n - 1]->next = NULL;
966 packed_git = ary[0];
967
968 free(ary);
969}
970
637cdd9d 971static int prepare_packed_git_run_once = 0;
9a217f2a 972void prepare_packed_git(void)
1f688557 973{
d5a63b99 974 struct alternate_object_database *alt;
1f688557 975
637cdd9d 976 if (prepare_packed_git_run_once)
1f688557 977 return;
9d835df2 978 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 979 prepare_alt_odb();
d5a63b99 980 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 981 alt->name[-1] = 0;
9d835df2 982 prepare_packed_git_one(alt->base, 0);
1494e038 983 alt->name[-1] = '/';
1f688557 984 }
b867092f 985 rearrange_packed_git();
637cdd9d
JK
986 prepare_packed_git_run_once = 1;
987}
988
fc04c412 989void reprepare_packed_git(void)
637cdd9d 990{
4b480c67 991 discard_revindex();
637cdd9d
JK
992 prepare_packed_git_run_once = 0;
993 prepare_packed_git();
1f688557
JH
994}
995
8eca0b47
NP
996static void mark_bad_packed_object(struct packed_git *p,
997 const unsigned char *sha1)
998{
999 unsigned i;
1000 for (i = 0; i < p->num_bad_objects; i++)
1001 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1002 return;
1003 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1004 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1005 p->num_bad_objects++;
1006}
1007
b6c4cecc 1008static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
ac939109
NP
1009{
1010 struct packed_git *p;
1011 unsigned i;
1012
1013 for (p = packed_git; p; p = p->next)
1014 for (i = 0; i < p->num_bad_objects; i++)
1015 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
b6c4cecc
JH
1016 return p;
1017 return NULL;
ac939109
NP
1018}
1019
5d6ccf5c 1020int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
0fcfd160
LT
1021{
1022 unsigned char real_sha1[20];
7cfb5f36 1023 hash_sha1_file(map, size, type, real_sha1);
a89fccd2 1024 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1025}
1026
44d1c19e
LT
1027static int git_open_noatime(const char *name)
1028{
1029 static int sha1_file_open_flag = O_NOATIME;
1030 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1031
1032 /* Might the failure be due to O_NOATIME? */
1033 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1034 fd = open(name, O_RDONLY);
1035 if (fd >= 0)
1036 sha1_file_open_flag = 0;
1037 }
1038 return fd;
1039}
1040
1041static int open_sha1_file(const unsigned char *sha1)
1042{
1043 int fd;
1044 char *name = sha1_file_name(sha1);
1045 struct alternate_object_database *alt;
1046
1047 fd = git_open_noatime(name);
1048 if (fd >= 0)
1049 return fd;
1050
1051 prepare_alt_odb();
1052 errno = ENOENT;
1053 for (alt = alt_odb_list; alt; alt = alt->next) {
1054 name = alt->name;
1055 fill_sha1_path(name, sha1);
1056 fd = git_open_noatime(alt->base);
1057 if (fd >= 0)
1058 return fd;
1059 }
1060 return -1;
1061}
1062
4175e9e3 1063static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1064{
0fcfd160 1065 void *map;
144bde78 1066 int fd;
ace1534d 1067
44d1c19e
LT
1068 fd = open_sha1_file(sha1);
1069 map = NULL;
1070 if (fd >= 0) {
1071 struct stat st;
0fcfd160 1072
44d1c19e
LT
1073 if (!fstat(fd, &st)) {
1074 *size = xsize_t(st.st_size);
1075 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1076 }
44d1c19e 1077 close(fd);
0fcfd160 1078 }
0fcfd160
LT
1079 return map;
1080}
1081
726f852b 1082static int legacy_loose_object(unsigned char *map)
bb6b8e4f
JH
1083{
1084 unsigned int word;
1085
1086 /*
1087 * Is it a zlib-compressed buffer? If so, the first byte
1088 * must be 0x78 (15-bit window size, deflated), and the
1089 * first 16-bit word is evenly divisible by 31
1090 */
1091 word = (map[0] << 8) + map[1];
1092 if (map[0] == 0x78 && !(word % 31))
1093 return 1;
1094 else
1095 return 0;
1096}
1097
09ded04b
NP
1098unsigned long unpack_object_header_buffer(const unsigned char *buf,
1099 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1100{
ad1ed5ee 1101 unsigned shift;
48fb7deb 1102 unsigned long size, c;
ad1ed5ee
JH
1103 unsigned long used = 0;
1104
1105 c = buf[used++];
1106 *type = (c >> 4) & 7;
1107 size = c & 15;
1108 shift = 4;
1109 while (c & 0x80) {
f630cfda 1110 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1111 error("bad object header");
ad1ed5ee 1112 return 0;
09ded04b 1113 }
ad1ed5ee
JH
1114 c = buf[used++];
1115 size += (c & 0x7f) << shift;
1116 shift += 7;
1117 }
1118 *sizep = size;
1119 return used;
1120}
1121
1122static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1123{
1124 unsigned long size, used;
1125 static const char valid_loose_object_type[8] = {
1126 0, /* OBJ_EXT */
1127 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1128 0, /* "delta" and others are invalid in a loose object */
93821bd9 1129 };
ad1ed5ee 1130 enum object_type type;
93821bd9 1131
c4483576
LT
1132 /* Get the data stream */
1133 memset(stream, 0, sizeof(*stream));
1134 stream->next_in = map;
1135 stream->avail_in = mapsize;
1136 stream->next_out = buffer;
93821bd9
LT
1137 stream->avail_out = bufsiz;
1138
bb6b8e4f 1139 if (legacy_loose_object(map)) {
39c68542
LT
1140 git_inflate_init(stream);
1141 return git_inflate(stream, 0);
93821bd9
LT
1142 }
1143
726f852b
NP
1144
1145 /*
1146 * There used to be a second loose object header format which
1147 * was meant to mimic the in-pack format, allowing for direct
1148 * copy of the object data. This format turned up not to be
1149 * really worth it and we don't write it any longer. But we
1150 * can still read it.
1151 */
09ded04b 1152 used = unpack_object_header_buffer(map, mapsize, &type, &size);
ad1ed5ee 1153 if (!used || !valid_loose_object_type[type])
93821bd9 1154 return -1;
ad1ed5ee
JH
1155 map += used;
1156 mapsize -= used;
93821bd9
LT
1157
1158 /* Set up the stream for the rest.. */
1159 stream->next_in = map;
1160 stream->avail_in = mapsize;
39c68542 1161 git_inflate_init(stream);
93821bd9
LT
1162
1163 /* And generate the fake traditional header */
ad1ed5ee 1164 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
df843662 1165 typename(type), size);
93821bd9 1166 return 0;
c4483576
LT
1167}
1168
7efbff75 1169static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1170{
1171 int bytes = strlen(buffer) + 1;
3aee68aa 1172 unsigned char *buf = xmallocz(size);
93821bd9 1173 unsigned long n;
7efbff75 1174 int status = Z_OK;
5180cacc 1175
93821bd9
LT
1176 n = stream->total_out - bytes;
1177 if (n > size)
1178 n = size;
1179 memcpy(buf, (char *) buffer + bytes, n);
1180 bytes = n;
456cdf6e
LT
1181 if (bytes <= size) {
1182 /*
1183 * The above condition must be (bytes <= size), not
1184 * (bytes < size). In other words, even though we
1185 * expect no more output and set avail_out to zer0,
1186 * the input zlib stream may have bytes that express
1187 * "this concludes the stream", and we *do* want to
1188 * eat that input.
1189 *
1190 * Otherwise we would not be able to test that we
1191 * consumed all the input to reach the expected size;
1192 * we also want to check that zlib tells us that all
1193 * went well with status == Z_STREAM_END at the end.
1194 */
5180cacc
LT
1195 stream->next_out = buf + bytes;
1196 stream->avail_out = size - bytes;
7efbff75 1197 while (status == Z_OK)
39c68542 1198 status = git_inflate(stream, Z_FINISH);
5180cacc 1199 }
456cdf6e 1200 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1201 git_inflate_end(stream);
7efbff75
JH
1202 return buf;
1203 }
1204
1205 if (status < 0)
1206 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1207 else if (stream->avail_in)
1208 error("garbage at end of loose object '%s'",
1209 sha1_to_hex(sha1));
1210 free(buf);
1211 return NULL;
5180cacc
LT
1212}
1213
1214/*
1215 * We used to just use "sscanf()", but that's actually way
1216 * too permissive for what we want to check. So do an anal
1217 * object header parse by hand.
1218 */
21666f1a 1219static int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1220{
21666f1a 1221 char type[10];
5180cacc
LT
1222 int i;
1223 unsigned long size;
1224
1225 /*
a6080a0a 1226 * The type can be at most ten bytes (including the
5180cacc 1227 * terminating '\0' that we add), and is followed by
21666f1a 1228 * a space.
5180cacc 1229 */
21666f1a 1230 i = 0;
5180cacc
LT
1231 for (;;) {
1232 char c = *hdr++;
1233 if (c == ' ')
1234 break;
21666f1a
NP
1235 type[i++] = c;
1236 if (i >= sizeof(type))
5180cacc 1237 return -1;
5180cacc 1238 }
21666f1a 1239 type[i] = 0;
5180cacc
LT
1240
1241 /*
1242 * The length must follow immediately, and be in canonical
1243 * decimal format (ie "010" is not valid).
1244 */
1245 size = *hdr++ - '0';
1246 if (size > 9)
1247 return -1;
1248 if (size) {
1249 for (;;) {
1250 unsigned long c = *hdr - '0';
1251 if (c > 9)
1252 break;
1253 hdr++;
1254 size = size * 10 + c;
1255 }
1256 }
1257 *sizep = size;
1258
1259 /*
1260 * The length must be followed by a zero byte
1261 */
21666f1a 1262 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1263}
1264
7efbff75 1265static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1266{
5180cacc 1267 int ret;
0fcfd160 1268 z_stream stream;
5180cacc 1269 char hdr[8192];
0fcfd160 1270
5180cacc 1271 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1272 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1273 return NULL;
1274
7efbff75 1275 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1276}
1277
54dab52a
NP
1278unsigned long get_size_from_delta(struct packed_git *p,
1279 struct pack_window **w_curs,
1280 off_t curpos)
1281{
1282 const unsigned char *data;
1283 unsigned char delta_head[20], *in;
1284 z_stream stream;
1285 int st;
1286
1287 memset(&stream, 0, sizeof(stream));
1288 stream.next_out = delta_head;
1289 stream.avail_out = sizeof(delta_head);
1290
39c68542 1291 git_inflate_init(&stream);
54dab52a
NP
1292 do {
1293 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1294 stream.next_in = in;
39c68542 1295 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1296 curpos += stream.next_in - in;
1297 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1298 stream.total_out < sizeof(delta_head));
39c68542 1299 git_inflate_end(&stream);
3d77d877
NP
1300 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1301 error("delta data unpack-initial failed");
1302 return 0;
1303 }
54dab52a
NP
1304
1305 /* Examine the initial part of the delta to figure out
1306 * the result size.
1307 */
1308 data = delta_head;
1309
1310 /* ignore base size */
1311 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1312
1313 /* Read the result size */
1314 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1315}
1316
c4001d92 1317static off_t get_delta_base(struct packed_git *p,
03e79c88 1318 struct pack_window **w_curs,
c4001d92 1319 off_t *curpos,
21666f1a 1320 enum object_type type,
c4001d92 1321 off_t delta_obj_offset)
eb32d236 1322{
2b87c45b 1323 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1324 off_t base_offset;
eb32d236 1325
8d8a4ea5
SP
1326 /* use_pack() assured us we have [base_info, base_info + 20)
1327 * as a range that we can look at without walking off the
1328 * end of the mapped window. Its actually the hash size
1329 * that is assured. An OFS_DELTA longer than the hash size
1330 * is stupid, as then a REF_DELTA would be smaller to store.
1331 */
21666f1a 1332 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1333 unsigned used = 0;
1334 unsigned char c = base_info[used++];
1335 base_offset = c & 127;
1336 while (c & 128) {
1337 base_offset += 1;
8723f216 1338 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1339 return 0; /* overflow */
eb32d236
NP
1340 c = base_info[used++];
1341 base_offset = (base_offset << 7) + (c & 127);
1342 }
1343 base_offset = delta_obj_offset - base_offset;
d8f32556 1344 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1345 return 0; /* out of bound */
2b87c45b 1346 *curpos += used;
21666f1a 1347 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1348 /* The base entry _must_ be in the same pack */
1349 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1350 *curpos += 20;
eb32d236
NP
1351 } else
1352 die("I am totally screwed");
2b87c45b 1353 return base_offset;
eb32d236
NP
1354}
1355
f3bf9224 1356/* forward declaration for a mutually recursive function */
c4001d92 1357static int packed_object_info(struct packed_git *p, off_t offset,
21666f1a 1358 unsigned long *sizep);
f3bf9224 1359
43057304 1360static int packed_delta_info(struct packed_git *p,
03e79c88 1361 struct pack_window **w_curs,
c4001d92 1362 off_t curpos,
21666f1a 1363 enum object_type type,
c4001d92 1364 off_t obj_offset,
43057304 1365 unsigned long *sizep)
5db47c2b 1366{
c4001d92 1367 off_t base_offset;
f3bf9224 1368
21666f1a 1369 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
3d77d877
NP
1370 if (!base_offset)
1371 return OBJ_BAD;
21666f1a 1372 type = packed_object_info(p, base_offset, NULL);
3d77d877 1373 if (type <= OBJ_NONE) {
08698b1e
NP
1374 struct revindex_entry *revidx;
1375 const unsigned char *base_sha1;
1376 revidx = find_pack_revindex(p, base_offset);
1377 if (!revidx)
1378 return OBJ_BAD;
1379 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
3d77d877
NP
1380 mark_bad_packed_object(p, base_sha1);
1381 type = sha1_object_info(base_sha1, NULL);
1382 if (type <= OBJ_NONE)
1383 return OBJ_BAD;
1384 }
f3bf9224 1385
c62266f3
JH
1386 /* We choose to only get the type of the base object and
1387 * ignore potentially corrupt pack file that expects the delta
1388 * based on a base with a wrong size. This saves tons of
1389 * inflate() calls.
1390 */
3d77d877 1391 if (sizep) {
54dab52a 1392 *sizep = get_size_from_delta(p, w_curs, curpos);
3d77d877
NP
1393 if (*sizep == 0)
1394 type = OBJ_BAD;
1395 }
21666f1a
NP
1396
1397 return type;
5db47c2b
JH
1398}
1399
2b87c45b
NP
1400static int unpack_object_header(struct packed_git *p,
1401 struct pack_window **w_curs,
c4001d92 1402 off_t *curpos,
2b87c45b 1403 unsigned long *sizep)
a733cb60 1404{
03e79c88
SP
1405 unsigned char *base;
1406 unsigned int left;
ad1ed5ee 1407 unsigned long used;
2b87c45b 1408 enum object_type type;
a733cb60 1409
8d8a4ea5
SP
1410 /* use_pack() assures us we have [base, base + 20) available
1411 * as a range that we can look at at. (Its actually the hash
3dff5379 1412 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1413 * the maximum deflated object size is 2^137, which is just
1414 * insane, so we know won't exceed what we have been given.
1415 */
2b87c45b 1416 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1417 used = unpack_object_header_buffer(base, left, &type, sizep);
1418 if (!used) {
1419 type = OBJ_BAD;
1420 } else
1421 *curpos += used;
ad1ed5ee 1422
2b87c45b 1423 return type;
a733cb60
LT
1424}
1425
21666f1a 1426const char *packed_object_info_detail(struct packed_git *p,
c4001d92 1427 off_t obj_offset,
21666f1a
NP
1428 unsigned long *size,
1429 unsigned long *store_size,
1430 unsigned int *delta_chain_length,
1431 unsigned char *base_sha1)
ad8c80a5 1432{
03e79c88 1433 struct pack_window *w_curs = NULL;
c4001d92
SP
1434 off_t curpos;
1435 unsigned long dummy;
43057304 1436 unsigned char *next_sha1;
21666f1a 1437 enum object_type type;
70f5d5d3 1438 struct revindex_entry *revidx;
ad8c80a5 1439
43057304 1440 *delta_chain_length = 0;
2b87c45b 1441 curpos = obj_offset;
21666f1a 1442 type = unpack_object_header(p, &w_curs, &curpos, size);
43057304 1443
70f5d5d3
NP
1444 revidx = find_pack_revindex(p, obj_offset);
1445 *store_size = revidx[1].offset - obj_offset;
1446
43057304 1447 for (;;) {
21666f1a 1448 switch (type) {
43057304 1449 default:
08a19d87 1450 die("pack %s contains unknown object type %d",
21666f1a 1451 p->pack_name, type);
43057304
NP
1452 case OBJ_COMMIT:
1453 case OBJ_TREE:
1454 case OBJ_BLOB:
1455 case OBJ_TAG:
03e79c88 1456 unuse_pack(&w_curs);
21666f1a 1457 return typename(type);
eb32d236 1458 case OBJ_OFS_DELTA:
21666f1a 1459 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
8eca0b47
NP
1460 if (!obj_offset)
1461 die("pack %s contains bad delta base reference of type %s",
1462 p->pack_name, typename(type));
eb32d236 1463 if (*delta_chain_length == 0) {
70f5d5d3
NP
1464 revidx = find_pack_revindex(p, obj_offset);
1465 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
eb32d236
NP
1466 }
1467 break;
1468 case OBJ_REF_DELTA:
2b87c45b 1469 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
43057304
NP
1470 if (*delta_chain_length == 0)
1471 hashcpy(base_sha1, next_sha1);
2b87c45b 1472 obj_offset = find_pack_entry_one(next_sha1, p);
43057304
NP
1473 break;
1474 }
43057304 1475 (*delta_chain_length)++;
2b87c45b 1476 curpos = obj_offset;
21666f1a 1477 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
ad8c80a5 1478 }
ad8c80a5
JH
1479}
1480
c4001d92 1481static int packed_object_info(struct packed_git *p, off_t obj_offset,
21666f1a 1482 unsigned long *sizep)
1f688557 1483{
03e79c88 1484 struct pack_window *w_curs = NULL;
c4001d92
SP
1485 unsigned long size;
1486 off_t curpos = obj_offset;
21666f1a 1487 enum object_type type;
5db47c2b 1488
21666f1a 1489 type = unpack_object_header(p, &w_curs, &curpos, &size);
5db47c2b 1490
21666f1a 1491 switch (type) {
eb32d236
NP
1492 case OBJ_OFS_DELTA:
1493 case OBJ_REF_DELTA:
21666f1a
NP
1494 type = packed_delta_info(p, &w_curs, curpos,
1495 type, obj_offset, sizep);
2b87c45b 1496 break;
a733cb60 1497 case OBJ_COMMIT:
a733cb60 1498 case OBJ_TREE:
a733cb60 1499 case OBJ_BLOB:
a733cb60 1500 case OBJ_TAG:
2b87c45b
NP
1501 if (sizep)
1502 *sizep = size;
a69d0943 1503 break;
1f688557 1504 default:
3d77d877
NP
1505 error("unknown object type %i at offset %"PRIuMAX" in %s",
1506 type, (uintmax_t)obj_offset, p->pack_name);
1507 type = OBJ_BAD;
1f688557 1508 }
2b87c45b 1509 unuse_pack(&w_curs);
21666f1a 1510 return type;
1f688557
JH
1511}
1512
eb950c19 1513static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1514 struct pack_window **w_curs,
c4001d92 1515 off_t curpos,
eb950c19 1516 unsigned long size)
de530aaa
SP
1517{
1518 int st;
1519 z_stream stream;
079afb18 1520 unsigned char *buffer, *in;
de530aaa 1521
4ab07e4d 1522 buffer = xmallocz(size);
de530aaa 1523 memset(&stream, 0, sizeof(stream));
de530aaa 1524 stream.next_out = buffer;
39eea7bd 1525 stream.avail_out = size + 1;
de530aaa 1526
39c68542 1527 git_inflate_init(&stream);
079afb18 1528 do {
2b87c45b 1529 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 1530 stream.next_in = in;
39c68542 1531 st = git_inflate(&stream, Z_FINISH);
39eea7bd
JH
1532 if (!stream.avail_out)
1533 break; /* the payload is larger than it should be */
2b87c45b 1534 curpos += stream.next_in - in;
079afb18 1535 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 1536 git_inflate_end(&stream);
de530aaa
SP
1537 if ((st != Z_STREAM_END) || stream.total_out != size) {
1538 free(buffer);
1539 return NULL;
1540 }
1541
1542 return buffer;
1543}
1544
e5e01619
LT
1545#define MAX_DELTA_CACHE (256)
1546
18bdec11 1547static size_t delta_base_cached;
5e08ecbf
NP
1548
1549static struct delta_base_cache_lru_list {
1550 struct delta_base_cache_lru_list *prev;
1551 struct delta_base_cache_lru_list *next;
1552} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1553
e5e01619 1554static struct delta_base_cache_entry {
5e08ecbf
NP
1555 struct delta_base_cache_lru_list lru;
1556 void *data;
e5e01619
LT
1557 struct packed_git *p;
1558 off_t base_offset;
1559 unsigned long size;
e5e01619
LT
1560 enum object_type type;
1561} delta_base_cache[MAX_DELTA_CACHE];
1562
1563static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1564{
1565 unsigned long hash;
1566
1567 hash = (unsigned long)p + (unsigned long)base_offset;
1568 hash += (hash >> 8) + (hash >> 16);
3358004a 1569 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1570}
1571
62f255ad 1572static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 1573 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 1574{
e5e01619
LT
1575 void *ret;
1576 unsigned long hash = pack_entry_hash(p, base_offset);
1577 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1578
1579 ret = ent->data;
749bc58c
MV
1580 if (!ret || ent->p != p || ent->base_offset != base_offset)
1581 return unpack_entry(p, base_offset, type, base_size);
e5e01619 1582
18bdec11 1583 if (!keep_cache) {
a0cba108 1584 ent->data = NULL;
5e08ecbf
NP
1585 ent->lru.next->prev = ent->lru.prev;
1586 ent->lru.prev->next = ent->lru.next;
18bdec11 1587 delta_base_cached -= ent->size;
182af834
PH
1588 } else {
1589 ret = xmemdupz(ent->data, ent->size);
a0cba108 1590 }
e5e01619
LT
1591 *type = ent->type;
1592 *base_size = ent->size;
1593 return ret;
62f255ad
LT
1594}
1595
18bdec11
SP
1596static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1597{
1598 if (ent->data) {
1599 free(ent->data);
1600 ent->data = NULL;
5e08ecbf
NP
1601 ent->lru.next->prev = ent->lru.prev;
1602 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
1603 delta_base_cached -= ent->size;
1604 }
1605}
1606
3d20c636
SP
1607void clear_delta_base_cache(void)
1608{
1609 unsigned long p;
1610 for (p = 0; p < MAX_DELTA_CACHE; p++)
1611 release_delta_base_cache(&delta_base_cache[p]);
1612}
1613
62f255ad
LT
1614static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1615 void *base, unsigned long base_size, enum object_type type)
1616{
5e08ecbf 1617 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 1618 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 1619 struct delta_base_cache_lru_list *lru;
e5e01619 1620
18bdec11
SP
1621 release_delta_base_cache(ent);
1622 delta_base_cached += base_size;
5e08ecbf
NP
1623
1624 for (lru = delta_base_cache_lru.next;
1625 delta_base_cached > delta_base_cache_limit
1626 && lru != &delta_base_cache_lru;
1627 lru = lru->next) {
1628 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
1629 if (f->type == OBJ_BLOB)
1630 release_delta_base_cache(f);
1631 }
5e08ecbf
NP
1632 for (lru = delta_base_cache_lru.next;
1633 delta_base_cached > delta_base_cache_limit
1634 && lru != &delta_base_cache_lru;
1635 lru = lru->next) {
1636 struct delta_base_cache_entry *f = (void *)lru;
1637 release_delta_base_cache(f);
1638 }
18bdec11 1639
e5e01619
LT
1640 ent->p = p;
1641 ent->base_offset = base_offset;
1642 ent->type = type;
1643 ent->data = base;
1644 ent->size = base_size;
5e08ecbf
NP
1645 ent->lru.next = &delta_base_cache_lru;
1646 ent->lru.prev = delta_base_cache_lru.prev;
1647 delta_base_cache_lru.prev->next = &ent->lru;
1648 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
1649}
1650
c2c5b270
CC
1651static void *read_object(const unsigned char *sha1, enum object_type *type,
1652 unsigned long *size);
1653
eb950c19 1654static void *unpack_delta_entry(struct packed_git *p,
03e79c88 1655 struct pack_window **w_curs,
c4001d92 1656 off_t curpos,
1f688557 1657 unsigned long delta_size,
c4001d92 1658 off_t obj_offset,
21666f1a 1659 enum object_type *type,
eb950c19 1660 unsigned long *sizep)
1f688557 1661{
7c3e8be3 1662 void *delta_data, *result, *base;
c4001d92
SP
1663 unsigned long base_size;
1664 off_t base_offset;
43057304 1665
21666f1a 1666 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
8eca0b47
NP
1667 if (!base_offset) {
1668 error("failed to validate delta base reference "
1669 "at offset %"PRIuMAX" from %s",
1670 (uintmax_t)curpos, p->pack_name);
1671 return NULL;
1672 }
eac12e2d 1673 unuse_pack(w_curs);
a0cba108 1674 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
8eca0b47
NP
1675 if (!base) {
1676 /*
1677 * We're probably in deep shit, but let's try to fetch
1678 * the required base anyway from another pack or loose.
1679 * This is costly but should happen only in the presence
1680 * of a corrupted pack, and is better than failing outright.
1681 */
08698b1e
NP
1682 struct revindex_entry *revidx;
1683 const unsigned char *base_sha1;
1684 revidx = find_pack_revindex(p, base_offset);
1685 if (!revidx)
1686 return NULL;
1687 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
8eca0b47
NP
1688 error("failed to read delta base object %s"
1689 " at offset %"PRIuMAX" from %s",
1690 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1691 p->pack_name);
1692 mark_bad_packed_object(p, base_sha1);
ac939109 1693 base = read_object(base_sha1, type, &base_size);
8eca0b47
NP
1694 if (!base)
1695 return NULL;
1696 }
67686d95 1697
2b87c45b 1698 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
8eca0b47
NP
1699 if (!delta_data) {
1700 error("failed to unpack compressed delta "
1701 "at offset %"PRIuMAX" from %s",
1702 (uintmax_t)curpos, p->pack_name);
1703 free(base);
1704 return NULL;
1705 }
1f688557
JH
1706 result = patch_delta(base, base_size,
1707 delta_data, delta_size,
21666f1a 1708 sizep);
1f688557
JH
1709 if (!result)
1710 die("failed to apply delta");
1711 free(delta_data);
62f255ad 1712 add_delta_base_cache(p, base_offset, base, base_size, *type);
1f688557
JH
1713 return result;
1714}
1715
0e8189e2
NP
1716int do_check_packed_object_crc;
1717
c4001d92 1718void *unpack_entry(struct packed_git *p, off_t obj_offset,
21666f1a 1719 enum object_type *type, unsigned long *sizep)
1f688557 1720{
03e79c88 1721 struct pack_window *w_curs = NULL;
c4001d92 1722 off_t curpos = obj_offset;
21666f1a 1723 void *data;
1f688557 1724
0e8189e2
NP
1725 if (do_check_packed_object_crc && p->index_version > 1) {
1726 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1727 unsigned long len = revidx[1].offset - obj_offset;
1728 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1729 const unsigned char *sha1 =
1730 nth_packed_object_sha1(p, revidx->nr);
1731 error("bad packed object CRC for %s",
1732 sha1_to_hex(sha1));
1733 mark_bad_packed_object(p, sha1);
f755bb99 1734 unuse_pack(&w_curs);
0e8189e2
NP
1735 return NULL;
1736 }
1737 }
1738
21666f1a
NP
1739 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1740 switch (*type) {
eb32d236
NP
1741 case OBJ_OFS_DELTA:
1742 case OBJ_REF_DELTA:
21666f1a
NP
1743 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1744 obj_offset, type, sizep);
4d703a1a 1745 break;
a733cb60 1746 case OBJ_COMMIT:
a733cb60 1747 case OBJ_TREE:
a733cb60 1748 case OBJ_BLOB:
a733cb60 1749 case OBJ_TAG:
21666f1a 1750 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
4d703a1a 1751 break;
1f688557 1752 default:
8eca0b47
NP
1753 data = NULL;
1754 error("unknown object type %i at offset %"PRIuMAX" in %s",
1755 *type, (uintmax_t)obj_offset, p->pack_name);
1f688557 1756 }
03e79c88 1757 unuse_pack(&w_curs);
21666f1a 1758 return data;
1f688557
JH
1759}
1760
d079837e 1761const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 1762 uint32_t n)
9a217f2a 1763{
42873078 1764 const unsigned char *index = p->index_data;
d079837e
SP
1765 if (!index) {
1766 if (open_pack_index(p))
1767 return NULL;
1768 index = p->index_data;
1769 }
57059091 1770 if (n >= p->num_objects)
d72308e0 1771 return NULL;
74e34e1f
NP
1772 index += 4 * 256;
1773 if (p->index_version == 1) {
1774 return index + 24 * n + 4;
1775 } else {
1776 index += 8;
1777 return index + 20 * n;
1778 }
1779}
1780
99093238 1781off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
1782{
1783 const unsigned char *index = p->index_data;
1784 index += 4 * 256;
1785 if (p->index_version == 1) {
1786 return ntohl(*((uint32_t *)(index + 24 * n)));
1787 } else {
1788 uint32_t off;
1789 index += 8 + p->num_objects * (20 + 4);
1790 off = ntohl(*((uint32_t *)(index + 4 * n)));
1791 if (!(off & 0x80000000))
1792 return off;
1793 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1794 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1795 ntohl(*((uint32_t *)(index + 4)));
1796 }
9a217f2a
JH
1797}
1798
c4001d92 1799off_t find_pack_entry_one(const unsigned char *sha1,
43057304 1800 struct packed_git *p)
1f688557 1801{
42873078 1802 const uint32_t *level1_ofs = p->index_data;
42873078 1803 const unsigned char *index = p->index_data;
628522ec
JH
1804 unsigned hi, lo, stride;
1805 static int use_lookup = -1;
1806 static int debug_lookup = -1;
1807
1808 if (debug_lookup < 0)
1809 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 1810
d079837e
SP
1811 if (!index) {
1812 if (open_pack_index(p))
1813 return 0;
1814 level1_ofs = p->index_data;
1815 index = p->index_data;
1816 }
74e34e1f
NP
1817 if (p->index_version > 1) {
1818 level1_ofs += 2;
1819 index += 8;
1820 }
42873078 1821 index += 4 * 256;
74e34e1f
NP
1822 hi = ntohl(level1_ofs[*sha1]);
1823 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
1824 if (p->index_version > 1) {
1825 stride = 20;
1826 } else {
1827 stride = 24;
1828 index += 4;
1829 }
1830
1831 if (debug_lookup)
6e1c2344 1832 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
1833 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1834
1835 if (use_lookup < 0)
1836 use_lookup = !!getenv("GIT_USE_LOOKUP");
1837 if (use_lookup) {
1838 int pos = sha1_entry_pos(index, stride, 0,
1839 lo, hi, p->num_objects, sha1);
1840 if (pos < 0)
1841 return 0;
1842 return nth_packed_object_offset(p, pos);
1843 }
1f688557
JH
1844
1845 do {
74e34e1f 1846 unsigned mi = (lo + hi) / 2;
628522ec
JH
1847 int cmp = hashcmp(index + mi * stride, sha1);
1848
1849 if (debug_lookup)
1850 printf("lo %u hi %u rg %u mi %u\n",
1851 lo, hi, hi - lo, mi);
43057304 1852 if (!cmp)
74e34e1f 1853 return nth_packed_object_offset(p, mi);
1f688557
JH
1854 if (cmp > 0)
1855 hi = mi;
1856 else
1857 lo = mi+1;
1858 } while (lo < hi);
1859 return 0;
1860}
1861
4d6acb70 1862static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557 1863{
f7c22cc6 1864 static struct packed_git *last_found = (void *)1;
1f688557 1865 struct packed_git *p;
c4001d92 1866 off_t offset;
43057304 1867
1f688557 1868 prepare_packed_git();
f7c22cc6
NP
1869 if (!packed_git)
1870 return 0;
1871 p = (last_found == (void *)1) ? packed_git : last_found;
1f688557 1872
f7c22cc6 1873 do {
8eca0b47
NP
1874 if (p->num_bad_objects) {
1875 unsigned i;
1876 for (i = 0; i < p->num_bad_objects; i++)
1877 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1878 goto next;
1879 }
1880
43057304
NP
1881 offset = find_pack_entry_one(sha1, p);
1882 if (offset) {
c715f783
SP
1883 /*
1884 * We are about to tell the caller where they can
1885 * locate the requested object. We better make
1886 * sure the packfile is still here and can be
1887 * accessed before supplying that answer, as
1888 * it may have been deleted since the index
1889 * was loaded!
1890 */
1891 if (p->pack_fd == -1 && open_packed_git(p)) {
1892 error("packfile %s cannot be accessed", p->pack_name);
f7c22cc6 1893 goto next;
c715f783 1894 }
43057304
NP
1895 e->offset = offset;
1896 e->p = p;
1897 hashcpy(e->sha1, sha1);
f7c22cc6 1898 last_found = p;
1f688557 1899 return 1;
43057304 1900 }
f7c22cc6
NP
1901
1902 next:
1903 if (p == last_found)
1904 p = packed_git;
1905 else
1906 p = p->next;
1907 if (p == last_found)
1908 p = p->next;
1909 } while (p);
1f688557
JH
1910 return 0;
1911}
1912
a6080a0a 1913struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
1914 struct packed_git *packs)
1915{
1916 struct packed_git *p;
bf592c50
DB
1917
1918 for (p = packs; p; p = p->next) {
43057304 1919 if (find_pack_entry_one(sha1, p))
bf592c50
DB
1920 return p;
1921 }
1922 return NULL;
21666f1a 1923
bf592c50
DB
1924}
1925
21666f1a 1926static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
65c2e0c3 1927{
36e4d74a 1928 int status;
65c2e0c3
JH
1929 unsigned long mapsize, size;
1930 void *map;
1931 z_stream stream;
d65a16f6 1932 char hdr[32];
65c2e0c3 1933
bb6b8e4f 1934 map = map_sha1_file(sha1, &mapsize);
f0df4ed5
JS
1935 if (!map)
1936 return error("unable to find %s", sha1_to_hex(sha1));
36e4d74a
JH
1937 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1938 status = error("unable to unpack %s header",
1939 sha1_to_hex(sha1));
21666f1a 1940 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 1941 status = error("unable to parse %s header", sha1_to_hex(sha1));
21666f1a
NP
1942 else if (sizep)
1943 *sizep = size;
39c68542 1944 git_inflate_end(&stream);
65c2e0c3
JH
1945 munmap(map, mapsize);
1946 return status;
1947}
1948
21666f1a 1949int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
f0df4ed5 1950{
f0df4ed5 1951 struct pack_entry e;
ddd63e64 1952 int status;
f0df4ed5 1953
cd673c1f 1954 if (!find_pack_entry(sha1, &e)) {
ddd63e64
SG
1955 /* Most likely it's a loose object. */
1956 status = sha1_loose_object_info(sha1, sizep);
1957 if (status >= 0)
1958 return status;
1959
1960 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 1961 reprepare_packed_git();
cd673c1f 1962 if (!find_pack_entry(sha1, &e))
ddd63e64 1963 return status;
f0df4ed5 1964 }
3d77d877
NP
1965
1966 status = packed_object_info(e.p, e.offset, sizep);
1967 if (status < 0) {
1968 mark_bad_packed_object(e.p, sha1);
1969 status = sha1_object_info(sha1, sizep);
1970 }
1971
1972 return status;
f0df4ed5
JS
1973}
1974
21666f1a
NP
1975static void *read_packed_sha1(const unsigned char *sha1,
1976 enum object_type *type, unsigned long *size)
1f688557
JH
1977{
1978 struct pack_entry e;
8eca0b47 1979 void *data;
1f688557 1980
cd673c1f 1981 if (!find_pack_entry(sha1, &e))
1f688557 1982 return NULL;
8eca0b47
NP
1983 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1984 if (!data) {
1985 /*
1986 * We're probably in deep shit, but let's try to fetch
1987 * the required object anyway from another pack or loose.
1988 * This should happen only in the presence of a corrupted
1989 * pack, and is better than failing outright.
1990 */
1991 error("failed to read object %s at offset %"PRIuMAX" from %s",
1992 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
1993 mark_bad_packed_object(e.p, sha1);
ac939109 1994 data = read_object(sha1, type, size);
8eca0b47
NP
1995 }
1996 return data;
1f688557
JH
1997}
1998
d66b37bb
JH
1999/*
2000 * This is meant to hold a *small* number of objects that you would
2001 * want read_sha1_file() to be able to return, but yet you do not want
2002 * to write them into the object store (e.g. a browse-only
2003 * application).
2004 */
2005static struct cached_object {
2006 unsigned char sha1[20];
21666f1a 2007 enum object_type type;
d66b37bb
JH
2008 void *buf;
2009 unsigned long size;
2010} *cached_objects;
2011static int cached_object_nr, cached_object_alloc;
2012
346245a1 2013static struct cached_object empty_tree = {
14d9c578 2014 EMPTY_TREE_SHA1_BIN,
346245a1
JK
2015 OBJ_TREE,
2016 "",
2017 0
2018};
2019
d66b37bb
JH
2020static struct cached_object *find_cached_object(const unsigned char *sha1)
2021{
2022 int i;
2023 struct cached_object *co = cached_objects;
2024
2025 for (i = 0; i < cached_object_nr; i++, co++) {
2026 if (!hashcmp(co->sha1, sha1))
2027 return co;
2028 }
346245a1
JK
2029 if (!hashcmp(sha1, empty_tree.sha1))
2030 return &empty_tree;
d66b37bb
JH
2031 return NULL;
2032}
2033
21666f1a
NP
2034int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2035 unsigned char *sha1)
d66b37bb
JH
2036{
2037 struct cached_object *co;
2038
21666f1a 2039 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2040 if (has_sha1_file(sha1) || find_cached_object(sha1))
2041 return 0;
2042 if (cached_object_alloc <= cached_object_nr) {
2043 cached_object_alloc = alloc_nr(cached_object_alloc);
2044 cached_objects = xrealloc(cached_objects,
2045 sizeof(*cached_objects) *
2046 cached_object_alloc);
2047 }
2048 co = &cached_objects[cached_object_nr++];
2049 co->size = len;
21666f1a 2050 co->type = type;
efa13f7b
JH
2051 co->buf = xmalloc(len);
2052 memcpy(co->buf, buf, len);
d66b37bb
JH
2053 hashcpy(co->sha1, sha1);
2054 return 0;
2055}
2056
c2c5b270
CC
2057static void *read_object(const unsigned char *sha1, enum object_type *type,
2058 unsigned long *size)
0fcfd160
LT
2059{
2060 unsigned long mapsize;
2061 void *map, *buf;
d66b37bb
JH
2062 struct cached_object *co;
2063
2064 co = find_cached_object(sha1);
2065 if (co) {
21666f1a 2066 *type = co->type;
d66b37bb 2067 *size = co->size;
182af834 2068 return xmemdupz(co->buf, co->size);
d66b37bb 2069 }
0fcfd160 2070
8276c007
PE
2071 buf = read_packed_sha1(sha1, type, size);
2072 if (buf)
2073 return buf;
bb6b8e4f 2074 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2075 if (map) {
7efbff75 2076 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2077 munmap(map, mapsize);
2078 return buf;
2079 }
637cdd9d 2080 reprepare_packed_git();
8276c007 2081 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2082}
2083
b6c4cecc
JH
2084/*
2085 * This function dies on corrupt objects; the callers who want to
2086 * deal with them should arrange to call read_object() and give error
2087 * messages themselves.
2088 */
f5552aee
CC
2089void *read_sha1_file_repl(const unsigned char *sha1,
2090 enum object_type *type,
2091 unsigned long *size,
2092 const unsigned char **replacement)
ac939109 2093{
68095570 2094 const unsigned char *repl = lookup_replace_object(sha1);
3ba7a065 2095 void *data;
e8b15e61 2096 char *path;
b6c4cecc
JH
2097 const struct packed_git *p;
2098
3ba7a065
JH
2099 errno = 0;
2100 data = read_object(repl, type, size);
b6c4cecc
JH
2101 if (data) {
2102 if (replacement)
2103 *replacement = repl;
2104 return data;
2105 }
68095570 2106
3ba7a065
JH
2107 if (errno != ENOENT)
2108 die_errno("failed to read object %s", sha1_to_hex(sha1));
2109
68095570 2110 /* die if we replaced an object with one that does not exist */
b6c4cecc 2111 if (repl != sha1)
68095570
CC
2112 die("replacement %s not found for %s",
2113 sha1_to_hex(repl), sha1_to_hex(sha1));
2114
b6c4cecc
JH
2115 if (has_loose_object(repl)) {
2116 path = sha1_file_name(sha1);
2117 die("loose object %s (stored in %s) is corrupt",
2118 sha1_to_hex(repl), path);
e8b15e61 2119 }
68095570 2120
b6c4cecc
JH
2121 if ((p = has_packed_and_bad(repl)) != NULL)
2122 die("packed object %s (stored in %s) is corrupt",
2123 sha1_to_hex(repl), p->pack_name);
f5552aee 2124
b6c4cecc 2125 return NULL;
ac939109
NP
2126}
2127
40469ee9 2128void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2129 const char *required_type_name,
40469ee9
JH
2130 unsigned long *size,
2131 unsigned char *actual_sha1_return)
f4913f91 2132{
21666f1a 2133 enum object_type type, required_type;
f4913f91
JH
2134 void *buffer;
2135 unsigned long isize;
40469ee9 2136 unsigned char actual_sha1[20];
f4913f91 2137
21666f1a 2138 required_type = type_from_string(required_type_name);
e702496e 2139 hashcpy(actual_sha1, sha1);
40469ee9
JH
2140 while (1) {
2141 int ref_length = -1;
2142 const char *ref_type = NULL;
f4913f91 2143
21666f1a 2144 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2145 if (!buffer)
2146 return NULL;
21666f1a 2147 if (type == required_type) {
40469ee9
JH
2148 *size = isize;
2149 if (actual_sha1_return)
e702496e 2150 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2151 return buffer;
2152 }
2153 /* Handle references */
21666f1a 2154 else if (type == OBJ_COMMIT)
40469ee9 2155 ref_type = "tree ";
21666f1a 2156 else if (type == OBJ_TAG)
40469ee9
JH
2157 ref_type = "object ";
2158 else {
2159 free(buffer);
2160 return NULL;
2161 }
2162 ref_length = strlen(ref_type);
f4913f91 2163
50974ec9
MK
2164 if (ref_length + 40 > isize ||
2165 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2166 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2167 free(buffer);
2168 return NULL;
2169 }
1cf58e72 2170 free(buffer);
40469ee9
JH
2171 /* Now we have the ID of the referred-to object in
2172 * actual_sha1. Check again. */
f4913f91 2173 }
f4913f91
JH
2174}
2175
ce9fbf16 2176static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2177 const char *type, unsigned char *sha1,
d65a16f6 2178 char *hdr, int *hdrlen)
d410c0f5 2179{
9126f009 2180 git_SHA_CTX c;
d410c0f5
JH
2181
2182 /* Generate the header */
d65a16f6 2183 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
2184
2185 /* Sha1.. */
9126f009
NP
2186 git_SHA1_Init(&c);
2187 git_SHA1_Update(&c, hdr, *hdrlen);
2188 git_SHA1_Update(&c, buf, len);
2189 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2190}
2191
230f1322 2192/*
5a688fe4
JH
2193 * Move the just written object into its final resting place.
2194 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2195 * "moving" is only a part of what it does, when no patch between
2196 * master to pu changes the call sites of this function.
230f1322 2197 */
839837b9 2198int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322 2199{
e32c0a9c 2200 int ret = 0;
5a688fe4 2201
348df166 2202 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
2203 goto try_rename;
2204 else if (link(tmpfile, filename))
e32c0a9c 2205 ret = errno;
7ebb6fca
LT
2206
2207 /*
2208 * Coda hack - coda doesn't like cross-directory links,
2209 * so we fall back to a rename, which will mean that it
2210 * won't be able to check collisions, but that's not a
2211 * big deal.
2212 *
2213 * The same holds for FAT formatted media.
2214 *
3be1f18e 2215 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
2216 * left to unlink.
2217 */
2218 if (ret && ret != EEXIST) {
be66a6c4 2219 try_rename:
7ebb6fca 2220 if (!rename(tmpfile, filename))
3be1f18e 2221 goto out;
9e48b389 2222 ret = errno;
230f1322 2223 }
691f1a28 2224 unlink_or_warn(tmpfile);
230f1322
LT
2225 if (ret) {
2226 if (ret != EEXIST) {
916d081b 2227 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
230f1322
LT
2228 }
2229 /* FIXME!!! Collision check here ? */
2230 }
2231
3be1f18e 2232out:
5256b006 2233 if (adjust_shared_perm(filename))
5a688fe4 2234 return error("unable to set permission to '%s'", filename);
230f1322
LT
2235 return 0;
2236}
2237
4d548150
LT
2238static int write_buffer(int fd, const void *buf, size_t len)
2239{
d34cf19b 2240 if (write_in_full(fd, buf, len) < 0)
93822c22 2241 return error("file write error (%s)", strerror(errno));
4d548150
LT
2242 return 0;
2243}
2244
ce9fbf16 2245int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2246 unsigned char *sha1)
2247{
d65a16f6 2248 char hdr[32];
abdc3fc8
RS
2249 int hdrlen;
2250 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2251 return 0;
2252}
2253
e9039dd3
LT
2254/* Finalize a file on disk, and close it. */
2255static void close_sha1_file(int fd)
2256{
aafe9fba
LT
2257 if (fsync_object_files)
2258 fsync_or_die(fd, "sha1 file");
e9039dd3 2259 if (close(fd) != 0)
d824cbba 2260 die_errno("error when closing sha1 file");
e9039dd3
LT
2261}
2262
5723fe7e
LT
2263/* Size of directory component, including the ending '/' */
2264static inline int directory_size(const char *filename)
2265{
2266 const char *s = strrchr(filename, '/');
2267 if (!s)
2268 return 0;
2269 return s - filename + 1;
2270}
2271
2272/*
2273 * This creates a temporary file in the same directory as the final
2274 * 'filename'
2275 *
2276 * We want to avoid cross-directory filename renames, because those
2277 * can have problems on various filesystems (FAT, NFS, Coda).
2278 */
2279static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2280{
2281 int fd, dirlen = directory_size(filename);
2282
2283 if (dirlen + 20 > bufsiz) {
2284 errno = ENAMETOOLONG;
2285 return -1;
2286 }
2287 memcpy(buffer, filename, dirlen);
2288 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
5256b006 2289 fd = git_mkstemp_mode(buffer, 0444);
cbacbf4e 2290 if (fd < 0 && dirlen && errno == ENOENT) {
5723fe7e 2291 /* Make sure the directory exists */
6ff6af62 2292 memcpy(buffer, filename, dirlen);
5723fe7e 2293 buffer[dirlen-1] = 0;
3bfaf018 2294 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
5723fe7e
LT
2295 return -1;
2296
2297 /* Try again */
2298 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
5256b006 2299 fd = git_mkstemp_mode(buffer, 0444);
5723fe7e
LT
2300 }
2301 return fd;
2302}
2303
bbac7311 2304static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 2305 const void *buf, unsigned long len, time_t mtime)
0fcfd160 2306{
915308b1 2307 int fd, ret;
9892beba 2308 unsigned char compressed[4096];
0fcfd160 2309 z_stream stream;
748af44c
NP
2310 git_SHA_CTX c;
2311 unsigned char parano_sha1[20];
706bc531 2312 char *filename;
aac17941 2313 static char tmpfile[PATH_MAX];
a44c9a5e 2314
972a9155 2315 filename = sha1_file_name(sha1);
5723fe7e 2316 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
fd73ccf2
JS
2317 while (fd < 0 && errno == EMFILE && unuse_one_window(packed_git, -1))
2318 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
aac17941 2319 if (fd < 0) {
35243577 2320 if (errno == EACCES)
916d081b
PB
2321 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2322 else
2323 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
aac17941
LT
2324 }
2325
0fcfd160
LT
2326 /* Set it up */
2327 memset(&stream, 0, sizeof(stream));
12f6c308 2328 deflateInit(&stream, zlib_compression_level);
0fcfd160 2329 stream.next_out = compressed;
9892beba 2330 stream.avail_out = sizeof(compressed);
748af44c 2331 git_SHA1_Init(&c);
a44c9a5e
LT
2332
2333 /* First header.. */
d65a16f6 2334 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2335 stream.avail_in = hdrlen;
726f852b
NP
2336 while (deflate(&stream, 0) == Z_OK)
2337 /* nothing */;
748af44c 2338 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
2339
2340 /* Then the data itself.. */
c00e657d 2341 stream.next_in = (void *)buf;
a44c9a5e 2342 stream.avail_in = len;
9892beba 2343 do {
748af44c 2344 unsigned char *in0 = stream.next_in;
9892beba 2345 ret = deflate(&stream, Z_FINISH);
748af44c 2346 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
2347 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2348 die("unable to write sha1 file");
2349 stream.next_out = compressed;
2350 stream.avail_out = sizeof(compressed);
2351 } while (ret == Z_OK);
2352
ac54c277
LT
2353 if (ret != Z_STREAM_END)
2354 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
ac54c277
LT
2355 ret = deflateEnd(&stream);
2356 if (ret != Z_OK)
2357 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
2358 git_SHA1_Final(parano_sha1, &c);
2359 if (hashcmp(sha1, parano_sha1) != 0)
2360 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 2361
e9039dd3 2362 close_sha1_file(fd);
0fcfd160 2363
bbac7311
NP
2364 if (mtime) {
2365 struct utimbuf utb;
2366 utb.actime = mtime;
2367 utb.modtime = mtime;
2368 if (utime(tmpfile, &utb) < 0)
2369 warning("failed utime() on %s: %s",
2370 tmpfile, strerror(errno));
2371 }
2372
230f1322 2373 return move_temp_to_file(tmpfile, filename);
0fcfd160 2374}
8237b185 2375
c00e657d 2376int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
bbac7311
NP
2377{
2378 unsigned char sha1[20];
2379 char hdr[32];
2380 int hdrlen;
2381
2382 /* Normally if we have it in the pack then we do not bother writing
2383 * it out into .git/objects/??/?{38} file.
2384 */
2385 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2386 if (returnsha1)
2387 hashcpy(returnsha1, sha1);
2388 if (has_sha1_file(sha1))
2389 return 0;
2390 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2391}
2392
2393int force_object_loose(const unsigned char *sha1, time_t mtime)
2394{
bbac7311
NP
2395 void *buf;
2396 unsigned long len;
2397 enum object_type type;
2398 char hdr[32];
2399 int hdrlen;
1fb23e65 2400 int ret;
bbac7311 2401
c529d75a 2402 if (has_loose_object(sha1))
bbac7311
NP
2403 return 0;
2404 buf = read_packed_sha1(sha1, &type, &len);
2405 if (!buf)
2406 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2407 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1fb23e65
BS
2408 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2409 free(buf);
2410
2411 return ret;
bbac7311
NP
2412}
2413
bf592c50
DB
2414int has_pack_index(const unsigned char *sha1)
2415{
2416 struct stat st;
2417 if (stat(sha1_pack_index_name(sha1), &st))
2418 return 0;
2419 return 1;
bf592c50
DB
2420}
2421
cd673c1f
JH
2422int has_sha1_pack(const unsigned char *sha1)
2423{
2424 struct pack_entry e;
2425 return find_pack_entry(sha1, &e);
2426}
2427
8237b185
DB
2428int has_sha1_file(const unsigned char *sha1)
2429{
1f688557
JH
2430 struct pack_entry e;
2431
cd673c1f 2432 if (find_pack_entry(sha1, &e))
1f688557 2433 return 1;
c529d75a 2434 return has_loose_object(sha1);
8237b185 2435}
74400e71 2436
43df4f86
DP
2437static int index_mem(unsigned char *sha1, void *buf, size_t size,
2438 int write_object, enum object_type type, const char *path)
e7332f96 2439{
6c510bee 2440 int ret, re_allocated = 0;
74400e71 2441
7672db20 2442 if (!type)
edaec3fb 2443 type = OBJ_BLOB;
6c510bee
LT
2444
2445 /*
2446 * Convert blobs to git internal format
2447 */
43df4f86 2448 if ((type == OBJ_BLOB) && path) {
f285a2d7 2449 struct strbuf nbuf = STRBUF_INIT;
21e5ad50
SP
2450 if (convert_to_git(path, buf, size, &nbuf,
2451 write_object ? safe_crlf : 0)) {
b315c5c0 2452 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
2453 re_allocated = 1;
2454 }
2455 }
2456
7672db20 2457 if (write_object)
edaec3fb 2458 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 2459 else
edaec3fb 2460 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 2461 if (re_allocated)
6c510bee 2462 free(buf);
43df4f86
DP
2463 return ret;
2464}
2465
ea68b0ce
DP
2466#define SMALL_FILE_SIZE (32*1024)
2467
43df4f86
DP
2468int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2469 enum object_type type, const char *path)
2470{
2471 int ret;
2472 size_t size = xsize_t(st->st_size);
2473
2474 if (!S_ISREG(st->st_mode)) {
f285a2d7 2475 struct strbuf sbuf = STRBUF_INIT;
43df4f86
DP
2476 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2477 ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2478 type, path);
2479 else
2480 ret = -1;
2481 strbuf_release(&sbuf);
08bda208
DP
2482 } else if (!size) {
2483 ret = index_mem(sha1, NULL, size, write_object, type, path);
ea68b0ce
DP
2484 } else if (size <= SMALL_FILE_SIZE) {
2485 char *buf = xmalloc(size);
2486 if (size == read_in_full(fd, buf, size))
2487 ret = index_mem(sha1, buf, size, write_object, type,
2488 path);
2489 else
2490 ret = error("short read %s", strerror(errno));
2491 free(buf);
08bda208 2492 } else {
43df4f86
DP
2493 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2494 ret = index_mem(sha1, buf, size, write_object, type, path);
aac17941 2495 munmap(buf, size);
08bda208 2496 }
43df4f86 2497 close(fd);
aac17941 2498 return ret;
74400e71 2499}
ec1fcc16
JH
2500
2501int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2502{
2503 int fd;
b760d3aa 2504 struct strbuf sb = STRBUF_INIT;
ec1fcc16
JH
2505
2506 switch (st->st_mode & S_IFMT) {
2507 case S_IFREG:
2508 fd = open(path, O_RDONLY);
2509 if (fd < 0)
2510 return error("open(\"%s\"): %s", path,
2511 strerror(errno));
53bca91a 2512 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
ec1fcc16
JH
2513 return error("%s: failed to insert into database",
2514 path);
2515 break;
2516 case S_IFLNK:
b760d3aa 2517 if (strbuf_readlink(&sb, path, st->st_size)) {
ec1fcc16 2518 char *errstr = strerror(errno);
ec1fcc16
JH
2519 return error("readlink(\"%s\"): %s", path,
2520 errstr);
2521 }
abdc3fc8 2522 if (!write_object)
b760d3aa
LT
2523 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2524 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
ec1fcc16
JH
2525 return error("%s: failed to insert into database",
2526 path);
b760d3aa 2527 strbuf_release(&sb);
ec1fcc16 2528 break;
f35a6d3b
LT
2529 case S_IFDIR:
2530 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
2531 default:
2532 return error("%s: unsupported file type", path);
2533 }
2534 return 0;
2535}
a69e5429
JH
2536
2537int read_pack_header(int fd, struct pack_header *header)
2538{
c697ad14
HO
2539 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2540 /* "eof before pack header was fully read" */
2541 return PH_ERROR_EOF;
2542
a69e5429
JH
2543 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2544 /* "protocol error (pack signature mismatch detected)" */
2545 return PH_ERROR_PACK_SIGNATURE;
2546 if (!pack_version_ok(header->hdr_version))
2547 /* "protocol error (pack version unsupported)" */
2548 return PH_ERROR_PROTOCOL;
2549 return 0;
2550}
40d52ff7
JK
2551
2552void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2553{
2554 enum object_type type = sha1_object_info(sha1, NULL);
2555 if (type < 0)
2556 die("%s is not a valid object", sha1_to_hex(sha1));
2557 if (type != expect)
2558 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2559 typename(expect));
2560}