doc: consistently use ASCIIDOC_EXTRA
[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
192a6be2 38const signed char hexval_table[256] = {
e49521b5
JH
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
45 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
46 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
47 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
48 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
51 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
69 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
71};
0fcfd160
LT
72
73int get_sha1_hex(const char *hex, unsigned char *sha1)
74{
75 int i;
76 for (i = 0; i < 20; i++) {
77 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
78 if (val & ~0xff)
79 return -1;
80 *sha1++ = val;
81 hex += 2;
82 }
83 return 0;
84}
85
8385abfd
JS
86static inline int offset_1st_component(const char *path)
87{
88 if (has_dos_drive_prefix(path))
89 return 2 + (path[2] == '/');
90 return *path == '/';
91}
92
b2cb9425
JH
93int safe_create_leading_directories(char *path)
94{
8385abfd 95 char *pos = path + offset_1st_component(path);
67d42212
JR
96 struct stat st;
97
b2cb9425
JH
98 while (pos) {
99 pos = strchr(pos, '/');
100 if (!pos)
101 break;
5f0bdf50
JH
102 while (*++pos == '/')
103 ;
104 if (!*pos)
105 break;
106 *--pos = '\0';
67d42212
JR
107 if (!stat(path, &st)) {
108 /* path exists */
109 if (!S_ISDIR(st.st_mode)) {
b2cb9425 110 *pos = '/';
67d42212 111 return -3;
b2cb9425 112 }
457f06d6 113 }
67d42212
JR
114 else if (mkdir(path, 0777)) {
115 *pos = '/';
116 return -1;
117 }
457f06d6
JS
118 else if (adjust_shared_perm(path)) {
119 *pos = '/';
120 return -2;
121 }
b2cb9425
JH
122 *pos++ = '/';
123 }
124 return 0;
125}
723c31fe 126
8e21d63b
JK
127int safe_create_leading_directories_const(const char *path)
128{
129 /* path points to cache entries, so xstrdup before messing with it */
130 char *buf = xstrdup(path);
131 int result = safe_create_leading_directories(buf);
132 free(buf);
133 return result;
134}
135
3bfaf018 136char *sha1_to_hex(const unsigned char *sha1)
0fcfd160 137{
dcb3450f
LT
138 static int bufno;
139 static char hexbuffer[4][50];
0fcfd160 140 static const char hex[] = "0123456789abcdef";
dcb3450f 141 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
0fcfd160
LT
142 int i;
143
144 for (i = 0; i < 20; i++) {
145 unsigned int val = *sha1++;
146 *buf++ = hex[val >> 4];
147 *buf++ = hex[val & 0xf];
148 }
1e80e044
JS
149 *buf = '\0';
150
0fcfd160
LT
151 return buffer;
152}
153
ace1534d
JH
154static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
155{
156 int i;
157 for (i = 0; i < 20; i++) {
158 static char hex[] = "0123456789abcdef";
159 unsigned int val = sha1[i];
160 char *pos = pathbuf + i*2 + (i > 0);
161 *pos++ = hex[val >> 4];
162 *pos = hex[val & 0xf];
163 }
164}
165
0fcfd160
LT
166/*
167 * NOTE! This returns a statically allocated buffer, so you have to be
790296fd 168 * careful about using it. Do an "xstrdup()" if you need to save the
0fcfd160 169 * filename.
ace1534d
JH
170 *
171 * Also note that this returns the location for creating. Reading
172 * SHA1 file can happen from any alternate directory listed in the
d19938ab 173 * DB_ENVIRONMENT environment variable if it is not found in
ace1534d 174 * the primary object database.
0fcfd160
LT
175 */
176char *sha1_file_name(const unsigned char *sha1)
177{
0fcfd160
LT
178 static char *name, *base;
179
180 if (!base) {
d19938ab 181 const char *sha1_file_directory = get_object_directory();
0fcfd160 182 int len = strlen(sha1_file_directory);
812666c8 183 base = xmalloc(len + 60);
0fcfd160
LT
184 memcpy(base, sha1_file_directory, len);
185 memset(base+len, 0, 60);
186 base[len] = '/';
187 base[len+3] = '/';
188 name = base + len + 1;
189 }
ace1534d 190 fill_sha1_path(name, sha1);
0fcfd160
LT
191 return base;
192}
193
633f43e1 194static char *sha1_get_pack_name(const unsigned char *sha1,
6eec46bd 195 char **name, char **base, const char *which)
bf592c50
DB
196{
197 static const char hex[] = "0123456789abcdef";
633f43e1 198 char *buf;
bf592c50
DB
199 int i;
200
633f43e1 201 if (!*base) {
bf592c50
DB
202 const char *sha1_file_directory = get_object_directory();
203 int len = strlen(sha1_file_directory);
633f43e1 204 *base = xmalloc(len + 60);
6eec46bd
JH
205 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
206 sha1_file_directory, which);
633f43e1 207 *name = *base + len + 11;
bf592c50
DB
208 }
209
633f43e1 210 buf = *name;
bf592c50
DB
211
212 for (i = 0; i < 20; i++) {
213 unsigned int val = *sha1++;
214 *buf++ = hex[val >> 4];
215 *buf++ = hex[val & 0xf];
216 }
a6080a0a 217
633f43e1 218 return *base;
bf592c50
DB
219}
220
633f43e1 221char *sha1_pack_name(const unsigned char *sha1)
bf592c50 222{
633f43e1 223 static char *name, *base;
bf592c50 224
6eec46bd 225 return sha1_get_pack_name(sha1, &name, &base, "pack");
633f43e1 226}
bf592c50 227
633f43e1
HO
228char *sha1_pack_index_name(const unsigned char *sha1)
229{
230 static char *name, *base;
a6080a0a 231
6eec46bd 232 return sha1_get_pack_name(sha1, &name, &base, "idx");
bf592c50
DB
233}
234
d5a63b99
JH
235struct alternate_object_database *alt_odb_list;
236static struct alternate_object_database **alt_odb_tail;
ace1534d 237
c2f493a4
MW
238static void read_info_alternates(const char * alternates, int depth);
239
ddd5d056
JH
240/*
241 * Prepare alternate object database registry.
d5a63b99
JH
242 *
243 * The variable alt_odb_list points at the list of struct
244 * alternate_object_database. The elements on this list come from
245 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
246 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
247 * whose contents is similar to that environment variable but can be
248 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
249 * contains "/the/directory/corresponding/to/.git/objects/...", while
250 * its name points just after the slash at the end of ".git/objects/"
251 * in the example above, and has enough space to hold 40-byte hex
252 * SHA1, an extra slash for the first level indirection, and the
253 * terminating NUL.
ddd5d056 254 */
c2f493a4 255static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
ace1534d 256{
1494e038 257 const char *objdir = get_object_directory();
c2f493a4
MW
258 struct alternate_object_database *ent;
259 struct alternate_object_database *alt;
260 /* 43 = 40-byte + 2 '/' + terminating NUL */
261 int pfxlen = len;
262 int entlen = pfxlen + 43;
ccfd3e99 263 int base_len = -1;
d5a63b99 264
85dadc38 265 if (!is_absolute_path(entry) && relative_base) {
c2f493a4
MW
266 /* Relative alt-odb */
267 if (base_len < 0)
268 base_len = strlen(relative_base) + 1;
269 entlen += base_len;
270 pfxlen += base_len;
271 }
272 ent = xmalloc(sizeof(*ent) + entlen);
273
85dadc38 274 if (!is_absolute_path(entry) && relative_base) {
c2f493a4
MW
275 memcpy(ent->base, relative_base, base_len - 1);
276 ent->base[base_len - 1] = '/';
277 memcpy(ent->base + base_len, entry, len);
278 }
279 else
280 memcpy(ent->base, entry, pfxlen);
281
282 ent->name = ent->base + pfxlen + 1;
283 ent->base[pfxlen + 3] = '/';
284 ent->base[pfxlen] = ent->base[entlen-1] = 0;
285
286 /* Detect cases where alternate disappeared */
90b4a71c 287 if (!is_directory(ent->base)) {
c2f493a4
MW
288 error("object directory %s does not exist; "
289 "check .git/objects/info/alternates.",
290 ent->base);
291 free(ent);
292 return -1;
293 }
294
295 /* Prevent the common mistake of listing the same
296 * thing twice, or object directory itself.
297 */
298 for (alt = alt_odb_list; alt; alt = alt->next) {
299 if (!memcmp(ent->base, alt->base, pfxlen)) {
300 free(ent);
301 return -1;
302 }
303 }
304 if (!memcmp(ent->base, objdir, pfxlen)) {
305 free(ent);
306 return -1;
307 }
308
309 /* add the alternate entry */
310 *alt_odb_tail = ent;
311 alt_odb_tail = &(ent->next);
312 ent->next = NULL;
313
314 /* recursively add alternates */
315 read_info_alternates(ent->base, depth + 1);
316
317 ent->base[pfxlen] = '/';
318
319 return 0;
320}
321
322static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
323 const char *relative_base, int depth)
324{
325 const char *cp, *last;
326
327 if (depth > 5) {
328 error("%s: ignoring alternate object stores, nesting too deep.",
329 relative_base);
330 return;
331 }
332
d5a63b99 333 last = alt;
9577e7e3
JH
334 while (last < ep) {
335 cp = last;
336 if (cp < ep && *cp == '#') {
337 while (cp < ep && *cp != sep)
338 cp++;
339 last = cp + 1;
340 continue;
341 }
c2f493a4
MW
342 while (cp < ep && *cp != sep)
343 cp++;
d5a63b99 344 if (last != cp) {
85dadc38 345 if (!is_absolute_path(last) && depth) {
c2f493a4
MW
346 error("%s: ignoring relative alternate object store %s",
347 relative_base, last);
348 } else {
349 link_alt_odb_entry(last, cp - last,
350 relative_base, depth);
1494e038 351 }
d5a63b99 352 }
9577e7e3 353 while (cp < ep && *cp == sep)
d5a63b99
JH
354 cp++;
355 last = cp;
9577e7e3 356 }
d5a63b99
JH
357}
358
c2f493a4 359static void read_info_alternates(const char * relative_base, int depth)
d5a63b99 360{
9577e7e3 361 char *map;
dc49cd76 362 size_t mapsz;
d5a63b99 363 struct stat st;
9cb18f56
JM
364 const char alt_file_name[] = "info/alternates";
365 /* Given that relative_base is no longer than PATH_MAX,
366 ensure that "path" has enough space to append "/", the
367 file name, "info/alternates", and a trailing NUL. */
368 char path[PATH_MAX + 1 + sizeof alt_file_name];
c2f493a4 369 int fd;
d5a63b99 370
9cb18f56 371 sprintf(path, "%s/%s", relative_base, alt_file_name);
d5a63b99
JH
372 fd = open(path, O_RDONLY);
373 if (fd < 0)
374 return;
375 if (fstat(fd, &st) || (st.st_size == 0)) {
376 close(fd);
9a217f2a 377 return;
ace1534d 378 }
dc49cd76
SP
379 mapsz = xsize_t(st.st_size);
380 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 381 close(fd);
d5a63b99 382
dc49cd76 383 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
c2f493a4 384
dc49cd76 385 munmap(map, mapsz);
ace1534d
JH
386}
387
bef70b22
DB
388void add_to_alternates_file(const char *reference)
389{
390 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
acd3b9ec 391 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
bef70b22
DB
392 char *alt = mkpath("%s/objects\n", reference);
393 write_or_die(fd, alt, strlen(alt));
394 if (commit_lock_file(lock))
395 die("could not close alternates file");
396 if (alt_odb_tail)
397 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
398}
399
d79796bc
JH
400void foreach_alt_odb(alt_odb_fn fn, void *cb)
401{
402 struct alternate_object_database *ent;
403
404 prepare_alt_odb();
405 for (ent = alt_odb_list; ent; ent = ent->next)
406 if (fn(ent, cb))
407 return;
408}
409
c2f493a4
MW
410void prepare_alt_odb(void)
411{
554fe20d 412 const char *alt;
c2f493a4 413
7dc24aa5
SP
414 if (alt_odb_tail)
415 return;
416
c2f493a4
MW
417 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
418 if (!alt) alt = "";
419
c2f493a4 420 alt_odb_tail = &alt_odb_list;
80ba074f 421 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
422
423 read_info_alternates(get_object_directory(), 0);
424}
425
0f4dc14a 426static int has_loose_object_local(const unsigned char *sha1)
ace1534d 427{
ace1534d 428 char *name = sha1_file_name(sha1);
0f4dc14a
BC
429 return !access(name, F_OK);
430}
ace1534d 431
0f4dc14a
BC
432int has_loose_object_nonlocal(const unsigned char *sha1)
433{
434 struct alternate_object_database *alt;
9a217f2a 435 prepare_alt_odb();
d5a63b99 436 for (alt = alt_odb_list; alt; alt = alt->next) {
0f4dc14a 437 fill_sha1_path(alt->name, sha1);
c529d75a
LT
438 if (!access(alt->base, F_OK))
439 return 1;
ace1534d 440 }
c529d75a 441 return 0;
ace1534d
JH
442}
443
0f4dc14a
BC
444static int has_loose_object(const unsigned char *sha1)
445{
446 return has_loose_object_local(sha1) ||
447 has_loose_object_nonlocal(sha1);
448}
449
60bb8b14 450static unsigned int pack_used_ctr;
a53128b6
SP
451static unsigned int pack_mmap_calls;
452static unsigned int peak_pack_open_windows;
453static unsigned int pack_open_windows;
454static size_t peak_pack_mapped;
60bb8b14 455static size_t pack_mapped;
9a217f2a 456struct packed_git *packed_git;
1f688557 457
b79d18c9 458void pack_report(void)
a53128b6
SP
459{
460 fprintf(stderr,
e05db0fd
PR
461 "pack_report: getpagesize() = %10" SZ_FMT "\n"
462 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
463 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
464 sz_fmt(getpagesize()),
465 sz_fmt(packed_git_window_size),
466 sz_fmt(packed_git_limit));
a53128b6
SP
467 fprintf(stderr,
468 "pack_report: pack_used_ctr = %10u\n"
469 "pack_report: pack_mmap_calls = %10u\n"
470 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
471 "pack_report: pack_mapped = "
472 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
473 pack_used_ctr,
474 pack_mmap_calls,
475 pack_open_windows, peak_pack_open_windows,
9e42d6a1 476 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
477}
478
42873078 479static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
480{
481 void *idx_map;
42873078 482 struct pack_idx_header *hdr;
dc49cd76 483 size_t idx_size;
74e34e1f 484 uint32_t version, nr, i, *index;
e93ec6f9 485 int fd = open(path, O_RDONLY);
1f688557 486 struct stat st;
42873078 487
1f688557
JH
488 if (fd < 0)
489 return -1;
490 if (fstat(fd, &st)) {
491 close(fd);
492 return -1;
493 }
dc49cd76 494 idx_size = xsize_t(st.st_size);
2d88451b
SP
495 if (idx_size < 4 * 256 + 20 + 20) {
496 close(fd);
497 return error("index file %s is too small", path);
498 }
c4712e45 499 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 500 close(fd);
1f688557 501
42873078
NP
502 hdr = idx_map;
503 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
504 version = ntohl(hdr->idx_version);
505 if (version < 2 || version > 2) {
506 munmap(idx_map, idx_size);
6e1c2344 507 return error("index file %s is version %"PRIu32
74e34e1f
NP
508 " and is not supported by this binary"
509 " (try upgrading GIT to a newer version)",
510 path, version);
511 }
512 } else
513 version = 1;
df1b059d 514
1f688557 515 nr = 0;
42873078 516 index = idx_map;
74e34e1f
NP
517 if (version > 1)
518 index += 2; /* skip index header */
1f688557 519 for (i = 0; i < 256; i++) {
326bf396 520 uint32_t n = ntohl(index[i]);
2d88451b
SP
521 if (n < nr) {
522 munmap(idx_map, idx_size);
df1b059d 523 return error("non-monotonic index %s", path);
2d88451b 524 }
1f688557
JH
525 nr = n;
526 }
527
74e34e1f
NP
528 if (version == 1) {
529 /*
530 * Total size:
531 * - 256 index entries 4 bytes each
532 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
533 * - 20-byte SHA1 of the packfile
534 * - 20-byte SHA1 file checksum
535 */
536 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
537 munmap(idx_map, idx_size);
eef427a0 538 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
539 }
540 } else if (version == 2) {
541 /*
542 * Minimum size:
543 * - 8 bytes of header
544 * - 256 index entries 4 bytes each
545 * - 20-byte sha1 entry * nr
546 * - 4-byte crc entry * nr
547 * - 4-byte offset entry * nr
548 * - 20-byte SHA1 of the packfile
549 * - 20-byte SHA1 file checksum
550 * And after the 4-byte offset table might be a
551 * variable sized table containing 8-byte entries
552 * for offsets larger than 2^31.
553 */
554 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
555 unsigned long max_size = min_size;
556 if (nr)
557 max_size += (nr - 1)*8;
558 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 559 munmap(idx_map, idx_size);
eef427a0 560 return error("wrong index v2 file size in %s", path);
74e34e1f 561 }
7109c889
JH
562 if (idx_size != min_size &&
563 /*
564 * make sure we can deal with large pack offsets.
565 * 31-bit signed offset won't be enough, neither
566 * 32-bit unsigned one will be.
567 */
568 (sizeof(off_t) <= 4)) {
569 munmap(idx_map, idx_size);
570 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 571 }
2d88451b 572 }
1f688557 573
74e34e1f 574 p->index_version = version;
42873078
NP
575 p->index_data = idx_map;
576 p->index_size = idx_size;
57059091 577 p->num_objects = nr;
1f688557
JH
578 return 0;
579}
580
bc8e478a 581int open_pack_index(struct packed_git *p)
d079837e
SP
582{
583 char *idx_name;
584 int ret;
585
586 if (p->index_data)
587 return 0;
588
589 idx_name = xstrdup(p->pack_name);
590 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
591 ret = check_packed_git_idx(idx_name, p);
592 free(idx_name);
593 return ret;
594}
595
11daf39b
SP
596static void scan_windows(struct packed_git *p,
597 struct packed_git **lru_p,
598 struct pack_window **lru_w,
599 struct pack_window **lru_l)
1f688557 600{
11daf39b
SP
601 struct pack_window *w, *w_l;
602
603 for (w_l = NULL, w = p->windows; w; w = w->next) {
604 if (!w->inuse_cnt) {
605 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
606 *lru_p = p;
607 *lru_w = w;
608 *lru_l = w_l;
54044bf8 609 }
54044bf8 610 }
11daf39b 611 w_l = w;
f9253394 612 }
11daf39b
SP
613}
614
d1efefa4 615static int unuse_one_window(struct packed_git *current, int keep_fd)
11daf39b
SP
616{
617 struct packed_git *p, *lru_p = NULL;
618 struct pack_window *lru_w = NULL, *lru_l = NULL;
619
620 if (current)
621 scan_windows(current, &lru_p, &lru_w, &lru_l);
622 for (p = packed_git; p; p = p->next)
623 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
624 if (lru_p) {
625 munmap(lru_w->base, lru_w->len);
626 pack_mapped -= lru_w->len;
627 if (lru_l)
628 lru_l->next = lru_w->next;
629 else {
630 lru_p->windows = lru_w->next;
d1efefa4 631 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
54044bf8
SP
632 close(lru_p->pack_fd);
633 lru_p->pack_fd = -1;
634 }
635 }
636 free(lru_w);
a53128b6 637 pack_open_windows--;
54044bf8
SP
638 return 1;
639 }
640 return 0;
f9253394
JH
641}
642
d1efefa4 643void release_pack_memory(size_t need, int fd)
97bfeb34
SP
644{
645 size_t cur = pack_mapped;
d1efefa4 646 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
97bfeb34
SP
647 ; /* nothing */
648}
649
c9ced051
SP
650void close_pack_windows(struct packed_git *p)
651{
652 while (p->windows) {
653 struct pack_window *w = p->windows;
654
655 if (w->inuse_cnt)
656 die("pack '%s' still has open windows to it",
657 p->pack_name);
658 munmap(w->base, w->len);
659 pack_mapped -= w->len;
660 pack_open_windows--;
661 p->windows = w->next;
662 free(w);
663 }
664}
665
03e79c88 666void unuse_pack(struct pack_window **w_cursor)
f9253394 667{
03e79c88
SP
668 struct pack_window *w = *w_cursor;
669 if (w) {
670 w->inuse_cnt--;
671 *w_cursor = NULL;
672 }
1f688557
JH
673}
674
c74faea1
NP
675/*
676 * This is used by git-repack in case a newly created pack happens to
677 * contain the same set of objects as an existing one. In that case
678 * the resulting file might be different even if its name would be the
679 * same. It is best to close any reference to the old pack before it is
680 * replaced on disk. Of course no index pointers nor windows for given pack
681 * must subsist at this point. If ever objects from this pack are requested
682 * again, the new version of the pack will be reinitialized through
683 * reprepare_packed_git().
684 */
685void free_pack_by_name(const char *pack_name)
686{
687 struct packed_git *p, **pp = &packed_git;
688
689 while (*pp) {
690 p = *pp;
691 if (strcmp(pack_name, p->pack_name) == 0) {
fa3a0c94 692 clear_delta_base_cache();
c74faea1
NP
693 close_pack_windows(p);
694 if (p->pack_fd != -1)
695 close(p->pack_fd);
696 if (p->index_data)
697 munmap((void *)p->index_data, p->index_size);
698 free(p->bad_object_sha1);
699 *pp = p->next;
700 free(p);
701 return;
702 }
703 pp = &p->next;
704 }
705}
706
3cf8b462
SP
707/*
708 * Do not call this directly as this leaks p->pack_fd on error return;
709 * call open_packed_git() instead.
710 */
711static int open_packed_git_1(struct packed_git *p)
1f688557 712{
9bc879c1
SP
713 struct stat st;
714 struct pack_header hdr;
715 unsigned char sha1[20];
716 unsigned char *idx_sha1;
2c039da8 717 long fd_flag;
9bc879c1 718
d079837e
SP
719 if (!p->index_data && open_pack_index(p))
720 return error("packfile %s index unavailable", p->pack_name);
721
9bc879c1
SP
722 p->pack_fd = open(p->pack_name, O_RDONLY);
723 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 724 return -1;
9bc879c1
SP
725
726 /* If we created the struct before we had the pack we lack size. */
bf592c50 727 if (!p->pack_size) {
bf592c50 728 if (!S_ISREG(st.st_mode))
072db278 729 return error("packfile %s not a regular file", p->pack_name);
bf592c50 730 p->pack_size = st.st_size;
9bc879c1 731 } else if (p->pack_size != st.st_size)
072db278 732 return error("packfile %s size changed", p->pack_name);
9bc879c1 733
2c039da8
JH
734 /* We leave these file descriptors open with sliding mmap;
735 * there is no point keeping them open across exec(), though.
736 */
737 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
738 if (fd_flag < 0)
072db278 739 return error("cannot determine file descriptor flags");
2c039da8
JH
740 fd_flag |= FD_CLOEXEC;
741 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 742 return error("cannot set FD_CLOEXEC");
2c039da8 743
9bc879c1 744 /* Verify we recognize this pack file format. */
e6e2bd62 745 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 746 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 747 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 748 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 749 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
750 return error("packfile %s is version %"PRIu32" and not"
751 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
752 p->pack_name, ntohl(hdr.hdr_version));
753
754 /* Verify the pack matches its index. */
57059091 755 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
756 return error("packfile %s claims to have %"PRIu32" objects"
757 " while index indicates %"PRIu32" objects",
57059091
NP
758 p->pack_name, ntohl(hdr.hdr_entries),
759 p->num_objects);
9bc879c1 760 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 761 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 762 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 763 return error("packfile %s signature is unavailable", p->pack_name);
42873078 764 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 765 if (hashcmp(sha1, idx_sha1))
072db278
SP
766 return error("packfile %s does not match index", p->pack_name);
767 return 0;
9bc879c1
SP
768}
769
3cf8b462
SP
770static int open_packed_git(struct packed_git *p)
771{
772 if (!open_packed_git_1(p))
773 return 0;
774 if (p->pack_fd != -1) {
775 close(p->pack_fd);
776 p->pack_fd = -1;
777 }
778 return -1;
779}
780
c4001d92 781static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
782{
783 /* We must promise at least 20 bytes (one hash) after the
784 * offset is available from this window, otherwise the offset
785 * is not actually in this window and a different window (which
786 * has that one hash excess) must be used. This is to support
787 * the object header and delta base parsing routines below.
788 */
789 off_t win_off = win->offset;
790 return win_off <= offset
791 && (offset + 20) <= (win_off + win->len);
792}
793
03e79c88
SP
794unsigned char* use_pack(struct packed_git *p,
795 struct pack_window **w_cursor,
c4001d92 796 off_t offset,
03e79c88 797 unsigned int *left)
9bc879c1 798{
60bb8b14 799 struct pack_window *win = *w_cursor;
03e79c88 800
072db278
SP
801 if (p->pack_fd == -1 && open_packed_git(p))
802 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14 803
a9d98a14 804 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
805 * pointless to ask for an offset into the middle of that
806 * hash, and the in_window function above wouldn't match
807 * don't allow an offset too close to the end of the file.
808 */
809 if (offset > (p->pack_size - 20))
810 die("offset beyond end of packfile (truncated pack?)");
811
812 if (!win || !in_window(win, offset)) {
813 if (win)
814 win->inuse_cnt--;
815 for (win = p->windows; win; win = win->next) {
816 if (in_window(win, offset))
817 break;
818 }
819 if (!win) {
78a28df9 820 size_t window_align = packed_git_window_size / 2;
dc49cd76 821 off_t len;
60bb8b14 822 win = xcalloc(1, sizeof(*win));
78a28df9 823 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
824 len = p->pack_size - win->offset;
825 if (len > packed_git_window_size)
826 len = packed_git_window_size;
827 win->len = (size_t)len;
60bb8b14 828 pack_mapped += win->len;
11daf39b 829 while (packed_git_limit < pack_mapped
d1efefa4 830 && unuse_one_window(p, p->pack_fd))
60bb8b14 831 ; /* nothing */
c4712e45 832 win->base = xmmap(NULL, win->len,
60bb8b14
SP
833 PROT_READ, MAP_PRIVATE,
834 p->pack_fd, win->offset);
835 if (win->base == MAP_FAILED)
73b4e4be
SP
836 die("packfile %s cannot be mapped: %s",
837 p->pack_name,
838 strerror(errno));
a53128b6
SP
839 pack_mmap_calls++;
840 pack_open_windows++;
841 if (pack_mapped > peak_pack_mapped)
842 peak_pack_mapped = pack_mapped;
843 if (pack_open_windows > peak_pack_open_windows)
844 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
845 win->next = p->windows;
846 p->windows = win;
847 }
1f688557 848 }
03e79c88
SP
849 if (win != *w_cursor) {
850 win->last_used = pack_used_ctr++;
851 win->inuse_cnt++;
852 *w_cursor = win;
853 }
60bb8b14 854 offset -= win->offset;
03e79c88 855 if (left)
dc49cd76 856 *left = win->len - xsize_t(offset);
03e79c88 857 return win->base + offset;
1f688557
JH
858}
859
27d69a46
NP
860static struct packed_git *alloc_packed_git(int extra)
861{
862 struct packed_git *p = xmalloc(sizeof(*p) + extra);
863 memset(p, 0, sizeof(*p));
864 p->pack_fd = -1;
865 return p;
866}
867
42873078 868struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557
JH
869{
870 struct stat st;
27d69a46 871 struct packed_git *p = alloc_packed_git(path_len + 2);
1f688557 872
42873078
NP
873 /*
874 * Make sure a corresponding .pack file exists and that
875 * the index looks sane.
876 */
877 path_len -= strlen(".idx");
27d69a46
NP
878 if (path_len < 1) {
879 free(p);
1f688557 880 return NULL;
27d69a46 881 }
42873078 882 memcpy(p->pack_name, path, path_len);
8d25931d
BC
883
884 strcpy(p->pack_name + path_len, ".keep");
885 if (!access(p->pack_name, F_OK))
886 p->pack_keep = 1;
887
42873078 888 strcpy(p->pack_name + path_len, ".pack");
d079837e 889 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 890 free(p);
1f688557
JH
891 return NULL;
892 }
42873078 893
1f688557
JH
894 /* ok, it looks sane as far as we can check without
895 * actually mapping the pack file.
896 */
1f688557 897 p->pack_size = st.st_size;
9d835df2 898 p->pack_local = local;
b867092f 899 p->mtime = st.st_mtime;
42873078
NP
900 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
901 hashclr(p->sha1);
1f688557
JH
902 return p;
903}
904
bf592c50 905struct packed_git *parse_pack_index(unsigned char *sha1)
c508df5e 906{
79c6dca4 907 const char *idx_path = sha1_pack_index_name(sha1);
42873078 908 const char *path = sha1_pack_name(sha1);
27d69a46 909 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
bf592c50 910
27d69a46
NP
911 strcpy(p->pack_name, path);
912 hashcpy(p->sha1, sha1);
42873078
NP
913 if (check_packed_git_idx(idx_path, p)) {
914 free(p);
bf592c50 915 return NULL;
42873078 916 }
bf592c50 917
bf592c50
DB
918 return p;
919}
920
921void install_packed_git(struct packed_git *pack)
922{
923 pack->next = packed_git;
924 packed_git = pack;
925}
926
9d835df2 927static void prepare_packed_git_one(char *objdir, int local)
1f688557 928{
9cb18f56
JM
929 /* Ensure that this buffer is large enough so that we can
930 append "/pack/" without clobbering the stack even if
931 strlen(objdir) were PATH_MAX. */
932 char path[PATH_MAX + 1 + 4 + 1 + 1];
1f688557
JH
933 int len;
934 DIR *dir;
935 struct dirent *de;
936
937 sprintf(path, "%s/pack", objdir);
938 len = strlen(path);
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
ac939109
NP
1067static int has_packed_and_bad(const unsigned char *sha1)
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))
1075 return 1;
1076 return 0;
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
44d1c19e
LT
1086static int git_open_noatime(const char *name)
1087{
1088 static int sha1_file_open_flag = O_NOATIME;
1089 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1090
1091 /* Might the failure be due to O_NOATIME? */
1092 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1093 fd = open(name, O_RDONLY);
1094 if (fd >= 0)
1095 sha1_file_open_flag = 0;
1096 }
1097 return fd;
1098}
1099
1100static int open_sha1_file(const unsigned char *sha1)
1101{
1102 int fd;
1103 char *name = sha1_file_name(sha1);
1104 struct alternate_object_database *alt;
1105
1106 fd = git_open_noatime(name);
1107 if (fd >= 0)
1108 return fd;
1109
1110 prepare_alt_odb();
1111 errno = ENOENT;
1112 for (alt = alt_odb_list; alt; alt = alt->next) {
1113 name = alt->name;
1114 fill_sha1_path(name, sha1);
1115 fd = git_open_noatime(alt->base);
1116 if (fd >= 0)
1117 return fd;
1118 }
1119 return -1;
1120}
1121
4175e9e3 1122static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1123{
0fcfd160 1124 void *map;
144bde78 1125 int fd;
ace1534d 1126
44d1c19e
LT
1127 fd = open_sha1_file(sha1);
1128 map = NULL;
1129 if (fd >= 0) {
1130 struct stat st;
0fcfd160 1131
44d1c19e
LT
1132 if (!fstat(fd, &st)) {
1133 *size = xsize_t(st.st_size);
1134 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1135 }
44d1c19e 1136 close(fd);
0fcfd160 1137 }
0fcfd160
LT
1138 return map;
1139}
1140
726f852b 1141static int legacy_loose_object(unsigned char *map)
bb6b8e4f
JH
1142{
1143 unsigned int word;
1144
1145 /*
1146 * Is it a zlib-compressed buffer? If so, the first byte
1147 * must be 0x78 (15-bit window size, deflated), and the
1148 * first 16-bit word is evenly divisible by 31
1149 */
1150 word = (map[0] << 8) + map[1];
1151 if (map[0] == 0x78 && !(word % 31))
1152 return 1;
1153 else
1154 return 0;
1155}
1156
09ded04b
NP
1157unsigned long unpack_object_header_buffer(const unsigned char *buf,
1158 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1159{
ad1ed5ee 1160 unsigned shift;
93821bd9 1161 unsigned char c;
93821bd9 1162 unsigned long size;
ad1ed5ee
JH
1163 unsigned long used = 0;
1164
1165 c = buf[used++];
1166 *type = (c >> 4) & 7;
1167 size = c & 15;
1168 shift = 4;
1169 while (c & 0x80) {
09ded04b
NP
1170 if (len <= used || sizeof(long) * 8 <= shift) {
1171 error("bad object header");
ad1ed5ee 1172 return 0;
09ded04b 1173 }
ad1ed5ee
JH
1174 c = buf[used++];
1175 size += (c & 0x7f) << shift;
1176 shift += 7;
1177 }
1178 *sizep = size;
1179 return used;
1180}
1181
1182static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1183{
1184 unsigned long size, used;
1185 static const char valid_loose_object_type[8] = {
1186 0, /* OBJ_EXT */
1187 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1188 0, /* "delta" and others are invalid in a loose object */
93821bd9 1189 };
ad1ed5ee 1190 enum object_type type;
93821bd9 1191
c4483576
LT
1192 /* Get the data stream */
1193 memset(stream, 0, sizeof(*stream));
1194 stream->next_in = map;
1195 stream->avail_in = mapsize;
1196 stream->next_out = buffer;
93821bd9
LT
1197 stream->avail_out = bufsiz;
1198
bb6b8e4f 1199 if (legacy_loose_object(map)) {
39c68542
LT
1200 git_inflate_init(stream);
1201 return git_inflate(stream, 0);
93821bd9
LT
1202 }
1203
726f852b
NP
1204
1205 /*
1206 * There used to be a second loose object header format which
1207 * was meant to mimic the in-pack format, allowing for direct
1208 * copy of the object data. This format turned up not to be
1209 * really worth it and we don't write it any longer. But we
1210 * can still read it.
1211 */
09ded04b 1212 used = unpack_object_header_buffer(map, mapsize, &type, &size);
ad1ed5ee 1213 if (!used || !valid_loose_object_type[type])
93821bd9 1214 return -1;
ad1ed5ee
JH
1215 map += used;
1216 mapsize -= used;
93821bd9
LT
1217
1218 /* Set up the stream for the rest.. */
1219 stream->next_in = map;
1220 stream->avail_in = mapsize;
39c68542 1221 git_inflate_init(stream);
93821bd9
LT
1222
1223 /* And generate the fake traditional header */
ad1ed5ee 1224 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
df843662 1225 typename(type), size);
93821bd9 1226 return 0;
c4483576
LT
1227}
1228
7efbff75 1229static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1230{
1231 int bytes = strlen(buffer) + 1;
d565b341 1232 unsigned char *buf = xmalloc(1+size);
93821bd9 1233 unsigned long n;
7efbff75 1234 int status = Z_OK;
5180cacc 1235
93821bd9
LT
1236 n = stream->total_out - bytes;
1237 if (n > size)
1238 n = size;
1239 memcpy(buf, (char *) buffer + bytes, n);
1240 bytes = n;
456cdf6e
LT
1241 if (bytes <= size) {
1242 /*
1243 * The above condition must be (bytes <= size), not
1244 * (bytes < size). In other words, even though we
1245 * expect no more output and set avail_out to zer0,
1246 * the input zlib stream may have bytes that express
1247 * "this concludes the stream", and we *do* want to
1248 * eat that input.
1249 *
1250 * Otherwise we would not be able to test that we
1251 * consumed all the input to reach the expected size;
1252 * we also want to check that zlib tells us that all
1253 * went well with status == Z_STREAM_END at the end.
1254 */
5180cacc
LT
1255 stream->next_out = buf + bytes;
1256 stream->avail_out = size - bytes;
7efbff75 1257 while (status == Z_OK)
39c68542 1258 status = git_inflate(stream, Z_FINISH);
5180cacc
LT
1259 }
1260 buf[size] = 0;
456cdf6e 1261 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1262 git_inflate_end(stream);
7efbff75
JH
1263 return buf;
1264 }
1265
1266 if (status < 0)
1267 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1268 else if (stream->avail_in)
1269 error("garbage at end of loose object '%s'",
1270 sha1_to_hex(sha1));
1271 free(buf);
1272 return NULL;
5180cacc
LT
1273}
1274
1275/*
1276 * We used to just use "sscanf()", but that's actually way
1277 * too permissive for what we want to check. So do an anal
1278 * object header parse by hand.
1279 */
21666f1a 1280static int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1281{
21666f1a 1282 char type[10];
5180cacc
LT
1283 int i;
1284 unsigned long size;
1285
1286 /*
a6080a0a 1287 * The type can be at most ten bytes (including the
5180cacc 1288 * terminating '\0' that we add), and is followed by
21666f1a 1289 * a space.
5180cacc 1290 */
21666f1a 1291 i = 0;
5180cacc
LT
1292 for (;;) {
1293 char c = *hdr++;
1294 if (c == ' ')
1295 break;
21666f1a
NP
1296 type[i++] = c;
1297 if (i >= sizeof(type))
5180cacc 1298 return -1;
5180cacc 1299 }
21666f1a 1300 type[i] = 0;
5180cacc
LT
1301
1302 /*
1303 * The length must follow immediately, and be in canonical
1304 * decimal format (ie "010" is not valid).
1305 */
1306 size = *hdr++ - '0';
1307 if (size > 9)
1308 return -1;
1309 if (size) {
1310 for (;;) {
1311 unsigned long c = *hdr - '0';
1312 if (c > 9)
1313 break;
1314 hdr++;
1315 size = size * 10 + c;
1316 }
1317 }
1318 *sizep = size;
1319
1320 /*
1321 * The length must be followed by a zero byte
1322 */
21666f1a 1323 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1324}
1325
7efbff75 1326static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1327{
5180cacc 1328 int ret;
0fcfd160 1329 z_stream stream;
5180cacc 1330 char hdr[8192];
0fcfd160 1331
5180cacc 1332 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1333 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1334 return NULL;
1335
7efbff75 1336 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1337}
1338
54dab52a
NP
1339unsigned long get_size_from_delta(struct packed_git *p,
1340 struct pack_window **w_curs,
1341 off_t curpos)
1342{
1343 const unsigned char *data;
1344 unsigned char delta_head[20], *in;
1345 z_stream stream;
1346 int st;
1347
1348 memset(&stream, 0, sizeof(stream));
1349 stream.next_out = delta_head;
1350 stream.avail_out = sizeof(delta_head);
1351
39c68542 1352 git_inflate_init(&stream);
54dab52a
NP
1353 do {
1354 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1355 stream.next_in = in;
39c68542 1356 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1357 curpos += stream.next_in - in;
1358 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1359 stream.total_out < sizeof(delta_head));
39c68542 1360 git_inflate_end(&stream);
3d77d877
NP
1361 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1362 error("delta data unpack-initial failed");
1363 return 0;
1364 }
54dab52a
NP
1365
1366 /* Examine the initial part of the delta to figure out
1367 * the result size.
1368 */
1369 data = delta_head;
1370
1371 /* ignore base size */
1372 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1373
1374 /* Read the result size */
1375 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1376}
1377
c4001d92 1378static off_t get_delta_base(struct packed_git *p,
03e79c88 1379 struct pack_window **w_curs,
c4001d92 1380 off_t *curpos,
21666f1a 1381 enum object_type type,
c4001d92 1382 off_t delta_obj_offset)
eb32d236 1383{
2b87c45b 1384 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1385 off_t base_offset;
eb32d236 1386
8d8a4ea5
SP
1387 /* use_pack() assured us we have [base_info, base_info + 20)
1388 * as a range that we can look at without walking off the
1389 * end of the mapped window. Its actually the hash size
1390 * that is assured. An OFS_DELTA longer than the hash size
1391 * is stupid, as then a REF_DELTA would be smaller to store.
1392 */
21666f1a 1393 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1394 unsigned used = 0;
1395 unsigned char c = base_info[used++];
1396 base_offset = c & 127;
1397 while (c & 128) {
1398 base_offset += 1;
8723f216 1399 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1400 return 0; /* overflow */
eb32d236
NP
1401 c = base_info[used++];
1402 base_offset = (base_offset << 7) + (c & 127);
1403 }
1404 base_offset = delta_obj_offset - base_offset;
d8f32556 1405 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1406 return 0; /* out of bound */
2b87c45b 1407 *curpos += used;
21666f1a 1408 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1409 /* The base entry _must_ be in the same pack */
1410 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1411 *curpos += 20;
eb32d236
NP
1412 } else
1413 die("I am totally screwed");
2b87c45b 1414 return base_offset;
eb32d236
NP
1415}
1416
f3bf9224 1417/* forward declaration for a mutually recursive function */
c4001d92 1418static int packed_object_info(struct packed_git *p, off_t offset,
21666f1a 1419 unsigned long *sizep);
f3bf9224 1420
43057304 1421static int packed_delta_info(struct packed_git *p,
03e79c88 1422 struct pack_window **w_curs,
c4001d92 1423 off_t curpos,
21666f1a 1424 enum object_type type,
c4001d92 1425 off_t obj_offset,
43057304 1426 unsigned long *sizep)
5db47c2b 1427{
c4001d92 1428 off_t base_offset;
f3bf9224 1429
21666f1a 1430 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
3d77d877
NP
1431 if (!base_offset)
1432 return OBJ_BAD;
21666f1a 1433 type = packed_object_info(p, base_offset, NULL);
3d77d877 1434 if (type <= OBJ_NONE) {
08698b1e
NP
1435 struct revindex_entry *revidx;
1436 const unsigned char *base_sha1;
1437 revidx = find_pack_revindex(p, base_offset);
1438 if (!revidx)
1439 return OBJ_BAD;
1440 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
3d77d877
NP
1441 mark_bad_packed_object(p, base_sha1);
1442 type = sha1_object_info(base_sha1, NULL);
1443 if (type <= OBJ_NONE)
1444 return OBJ_BAD;
1445 }
f3bf9224 1446
c62266f3
JH
1447 /* We choose to only get the type of the base object and
1448 * ignore potentially corrupt pack file that expects the delta
1449 * based on a base with a wrong size. This saves tons of
1450 * inflate() calls.
1451 */
3d77d877 1452 if (sizep) {
54dab52a 1453 *sizep = get_size_from_delta(p, w_curs, curpos);
3d77d877
NP
1454 if (*sizep == 0)
1455 type = OBJ_BAD;
1456 }
21666f1a
NP
1457
1458 return type;
5db47c2b
JH
1459}
1460
2b87c45b
NP
1461static int unpack_object_header(struct packed_git *p,
1462 struct pack_window **w_curs,
c4001d92 1463 off_t *curpos,
2b87c45b 1464 unsigned long *sizep)
a733cb60 1465{
03e79c88
SP
1466 unsigned char *base;
1467 unsigned int left;
ad1ed5ee 1468 unsigned long used;
2b87c45b 1469 enum object_type type;
a733cb60 1470
8d8a4ea5
SP
1471 /* use_pack() assures us we have [base, base + 20) available
1472 * as a range that we can look at at. (Its actually the hash
3dff5379 1473 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1474 * the maximum deflated object size is 2^137, which is just
1475 * insane, so we know won't exceed what we have been given.
1476 */
2b87c45b 1477 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1478 used = unpack_object_header_buffer(base, left, &type, sizep);
1479 if (!used) {
1480 type = OBJ_BAD;
1481 } else
1482 *curpos += used;
ad1ed5ee 1483
2b87c45b 1484 return type;
a733cb60
LT
1485}
1486
21666f1a 1487const char *packed_object_info_detail(struct packed_git *p,
c4001d92 1488 off_t obj_offset,
21666f1a
NP
1489 unsigned long *size,
1490 unsigned long *store_size,
1491 unsigned int *delta_chain_length,
1492 unsigned char *base_sha1)
ad8c80a5 1493{
03e79c88 1494 struct pack_window *w_curs = NULL;
c4001d92
SP
1495 off_t curpos;
1496 unsigned long dummy;
43057304 1497 unsigned char *next_sha1;
21666f1a 1498 enum object_type type;
70f5d5d3 1499 struct revindex_entry *revidx;
ad8c80a5 1500
43057304 1501 *delta_chain_length = 0;
2b87c45b 1502 curpos = obj_offset;
21666f1a 1503 type = unpack_object_header(p, &w_curs, &curpos, size);
43057304 1504
70f5d5d3
NP
1505 revidx = find_pack_revindex(p, obj_offset);
1506 *store_size = revidx[1].offset - obj_offset;
1507
43057304 1508 for (;;) {
21666f1a 1509 switch (type) {
43057304 1510 default:
08a19d87 1511 die("pack %s contains unknown object type %d",
21666f1a 1512 p->pack_name, type);
43057304
NP
1513 case OBJ_COMMIT:
1514 case OBJ_TREE:
1515 case OBJ_BLOB:
1516 case OBJ_TAG:
03e79c88 1517 unuse_pack(&w_curs);
21666f1a 1518 return typename(type);
eb32d236 1519 case OBJ_OFS_DELTA:
21666f1a 1520 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
8eca0b47
NP
1521 if (!obj_offset)
1522 die("pack %s contains bad delta base reference of type %s",
1523 p->pack_name, typename(type));
eb32d236 1524 if (*delta_chain_length == 0) {
70f5d5d3
NP
1525 revidx = find_pack_revindex(p, obj_offset);
1526 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
eb32d236
NP
1527 }
1528 break;
1529 case OBJ_REF_DELTA:
2b87c45b 1530 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
43057304
NP
1531 if (*delta_chain_length == 0)
1532 hashcpy(base_sha1, next_sha1);
2b87c45b 1533 obj_offset = find_pack_entry_one(next_sha1, p);
43057304
NP
1534 break;
1535 }
43057304 1536 (*delta_chain_length)++;
2b87c45b 1537 curpos = obj_offset;
21666f1a 1538 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
ad8c80a5 1539 }
ad8c80a5
JH
1540}
1541
c4001d92 1542static int packed_object_info(struct packed_git *p, off_t obj_offset,
21666f1a 1543 unsigned long *sizep)
1f688557 1544{
03e79c88 1545 struct pack_window *w_curs = NULL;
c4001d92
SP
1546 unsigned long size;
1547 off_t curpos = obj_offset;
21666f1a 1548 enum object_type type;
5db47c2b 1549
21666f1a 1550 type = unpack_object_header(p, &w_curs, &curpos, &size);
5db47c2b 1551
21666f1a 1552 switch (type) {
eb32d236
NP
1553 case OBJ_OFS_DELTA:
1554 case OBJ_REF_DELTA:
21666f1a
NP
1555 type = packed_delta_info(p, &w_curs, curpos,
1556 type, obj_offset, sizep);
2b87c45b 1557 break;
a733cb60 1558 case OBJ_COMMIT:
a733cb60 1559 case OBJ_TREE:
a733cb60 1560 case OBJ_BLOB:
a733cb60 1561 case OBJ_TAG:
2b87c45b
NP
1562 if (sizep)
1563 *sizep = size;
a69d0943 1564 break;
1f688557 1565 default:
3d77d877
NP
1566 error("unknown object type %i at offset %"PRIuMAX" in %s",
1567 type, (uintmax_t)obj_offset, p->pack_name);
1568 type = OBJ_BAD;
1f688557 1569 }
2b87c45b 1570 unuse_pack(&w_curs);
21666f1a 1571 return type;
1f688557
JH
1572}
1573
eb950c19 1574static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1575 struct pack_window **w_curs,
c4001d92 1576 off_t curpos,
eb950c19 1577 unsigned long size)
de530aaa
SP
1578{
1579 int st;
1580 z_stream stream;
079afb18 1581 unsigned char *buffer, *in;
de530aaa
SP
1582
1583 buffer = xmalloc(size + 1);
1584 buffer[size] = 0;
1585 memset(&stream, 0, sizeof(stream));
de530aaa
SP
1586 stream.next_out = buffer;
1587 stream.avail_out = size;
1588
39c68542 1589 git_inflate_init(&stream);
079afb18 1590 do {
2b87c45b 1591 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 1592 stream.next_in = in;
39c68542 1593 st = git_inflate(&stream, Z_FINISH);
2b87c45b 1594 curpos += stream.next_in - in;
079afb18 1595 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 1596 git_inflate_end(&stream);
de530aaa
SP
1597 if ((st != Z_STREAM_END) || stream.total_out != size) {
1598 free(buffer);
1599 return NULL;
1600 }
1601
1602 return buffer;
1603}
1604
e5e01619
LT
1605#define MAX_DELTA_CACHE (256)
1606
18bdec11 1607static size_t delta_base_cached;
5e08ecbf
NP
1608
1609static struct delta_base_cache_lru_list {
1610 struct delta_base_cache_lru_list *prev;
1611 struct delta_base_cache_lru_list *next;
1612} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1613
e5e01619 1614static struct delta_base_cache_entry {
5e08ecbf
NP
1615 struct delta_base_cache_lru_list lru;
1616 void *data;
e5e01619
LT
1617 struct packed_git *p;
1618 off_t base_offset;
1619 unsigned long size;
e5e01619
LT
1620 enum object_type type;
1621} delta_base_cache[MAX_DELTA_CACHE];
1622
1623static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1624{
1625 unsigned long hash;
1626
1627 hash = (unsigned long)p + (unsigned long)base_offset;
1628 hash += (hash >> 8) + (hash >> 16);
3358004a 1629 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1630}
1631
62f255ad 1632static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 1633 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 1634{
e5e01619
LT
1635 void *ret;
1636 unsigned long hash = pack_entry_hash(p, base_offset);
1637 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1638
1639 ret = ent->data;
749bc58c
MV
1640 if (!ret || ent->p != p || ent->base_offset != base_offset)
1641 return unpack_entry(p, base_offset, type, base_size);
e5e01619 1642
18bdec11 1643 if (!keep_cache) {
a0cba108 1644 ent->data = NULL;
5e08ecbf
NP
1645 ent->lru.next->prev = ent->lru.prev;
1646 ent->lru.prev->next = ent->lru.next;
18bdec11 1647 delta_base_cached -= ent->size;
182af834
PH
1648 } else {
1649 ret = xmemdupz(ent->data, ent->size);
a0cba108 1650 }
e5e01619
LT
1651 *type = ent->type;
1652 *base_size = ent->size;
1653 return ret;
62f255ad
LT
1654}
1655
18bdec11
SP
1656static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1657{
1658 if (ent->data) {
1659 free(ent->data);
1660 ent->data = NULL;
5e08ecbf
NP
1661 ent->lru.next->prev = ent->lru.prev;
1662 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
1663 delta_base_cached -= ent->size;
1664 }
1665}
1666
3d20c636
SP
1667void clear_delta_base_cache(void)
1668{
1669 unsigned long p;
1670 for (p = 0; p < MAX_DELTA_CACHE; p++)
1671 release_delta_base_cache(&delta_base_cache[p]);
1672}
1673
62f255ad
LT
1674static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1675 void *base, unsigned long base_size, enum object_type type)
1676{
5e08ecbf 1677 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 1678 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 1679 struct delta_base_cache_lru_list *lru;
e5e01619 1680
18bdec11
SP
1681 release_delta_base_cache(ent);
1682 delta_base_cached += base_size;
5e08ecbf
NP
1683
1684 for (lru = delta_base_cache_lru.next;
1685 delta_base_cached > delta_base_cache_limit
1686 && lru != &delta_base_cache_lru;
1687 lru = lru->next) {
1688 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
1689 if (f->type == OBJ_BLOB)
1690 release_delta_base_cache(f);
1691 }
5e08ecbf
NP
1692 for (lru = delta_base_cache_lru.next;
1693 delta_base_cached > delta_base_cache_limit
1694 && lru != &delta_base_cache_lru;
1695 lru = lru->next) {
1696 struct delta_base_cache_entry *f = (void *)lru;
1697 release_delta_base_cache(f);
1698 }
18bdec11 1699
e5e01619
LT
1700 ent->p = p;
1701 ent->base_offset = base_offset;
1702 ent->type = type;
1703 ent->data = base;
1704 ent->size = base_size;
5e08ecbf
NP
1705 ent->lru.next = &delta_base_cache_lru;
1706 ent->lru.prev = delta_base_cache_lru.prev;
1707 delta_base_cache_lru.prev->next = &ent->lru;
1708 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
1709}
1710
c2c5b270
CC
1711static void *read_object(const unsigned char *sha1, enum object_type *type,
1712 unsigned long *size);
1713
eb950c19 1714static void *unpack_delta_entry(struct packed_git *p,
03e79c88 1715 struct pack_window **w_curs,
c4001d92 1716 off_t curpos,
1f688557 1717 unsigned long delta_size,
c4001d92 1718 off_t obj_offset,
21666f1a 1719 enum object_type *type,
eb950c19 1720 unsigned long *sizep)
1f688557 1721{
7c3e8be3 1722 void *delta_data, *result, *base;
c4001d92
SP
1723 unsigned long base_size;
1724 off_t base_offset;
43057304 1725
21666f1a 1726 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
8eca0b47
NP
1727 if (!base_offset) {
1728 error("failed to validate delta base reference "
1729 "at offset %"PRIuMAX" from %s",
1730 (uintmax_t)curpos, p->pack_name);
1731 return NULL;
1732 }
eac12e2d 1733 unuse_pack(w_curs);
a0cba108 1734 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
8eca0b47
NP
1735 if (!base) {
1736 /*
1737 * We're probably in deep shit, but let's try to fetch
1738 * the required base anyway from another pack or loose.
1739 * This is costly but should happen only in the presence
1740 * of a corrupted pack, and is better than failing outright.
1741 */
08698b1e
NP
1742 struct revindex_entry *revidx;
1743 const unsigned char *base_sha1;
1744 revidx = find_pack_revindex(p, base_offset);
1745 if (!revidx)
1746 return NULL;
1747 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
8eca0b47
NP
1748 error("failed to read delta base object %s"
1749 " at offset %"PRIuMAX" from %s",
1750 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1751 p->pack_name);
1752 mark_bad_packed_object(p, base_sha1);
ac939109 1753 base = read_object(base_sha1, type, &base_size);
8eca0b47
NP
1754 if (!base)
1755 return NULL;
1756 }
67686d95 1757
2b87c45b 1758 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
8eca0b47
NP
1759 if (!delta_data) {
1760 error("failed to unpack compressed delta "
1761 "at offset %"PRIuMAX" from %s",
1762 (uintmax_t)curpos, p->pack_name);
1763 free(base);
1764 return NULL;
1765 }
1f688557
JH
1766 result = patch_delta(base, base_size,
1767 delta_data, delta_size,
21666f1a 1768 sizep);
1f688557
JH
1769 if (!result)
1770 die("failed to apply delta");
1771 free(delta_data);
62f255ad 1772 add_delta_base_cache(p, base_offset, base, base_size, *type);
1f688557
JH
1773 return result;
1774}
1775
0e8189e2
NP
1776int do_check_packed_object_crc;
1777
c4001d92 1778void *unpack_entry(struct packed_git *p, off_t obj_offset,
21666f1a 1779 enum object_type *type, unsigned long *sizep)
1f688557 1780{
03e79c88 1781 struct pack_window *w_curs = NULL;
c4001d92 1782 off_t curpos = obj_offset;
21666f1a 1783 void *data;
1f688557 1784
0e8189e2
NP
1785 if (do_check_packed_object_crc && p->index_version > 1) {
1786 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1787 unsigned long len = revidx[1].offset - obj_offset;
1788 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1789 const unsigned char *sha1 =
1790 nth_packed_object_sha1(p, revidx->nr);
1791 error("bad packed object CRC for %s",
1792 sha1_to_hex(sha1));
1793 mark_bad_packed_object(p, sha1);
f755bb99 1794 unuse_pack(&w_curs);
0e8189e2
NP
1795 return NULL;
1796 }
1797 }
1798
21666f1a
NP
1799 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1800 switch (*type) {
eb32d236
NP
1801 case OBJ_OFS_DELTA:
1802 case OBJ_REF_DELTA:
21666f1a
NP
1803 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1804 obj_offset, type, sizep);
4d703a1a 1805 break;
a733cb60 1806 case OBJ_COMMIT:
a733cb60 1807 case OBJ_TREE:
a733cb60 1808 case OBJ_BLOB:
a733cb60 1809 case OBJ_TAG:
21666f1a 1810 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
4d703a1a 1811 break;
1f688557 1812 default:
8eca0b47
NP
1813 data = NULL;
1814 error("unknown object type %i at offset %"PRIuMAX" in %s",
1815 *type, (uintmax_t)obj_offset, p->pack_name);
1f688557 1816 }
03e79c88 1817 unuse_pack(&w_curs);
21666f1a 1818 return data;
1f688557
JH
1819}
1820
d079837e 1821const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 1822 uint32_t n)
9a217f2a 1823{
42873078 1824 const unsigned char *index = p->index_data;
d079837e
SP
1825 if (!index) {
1826 if (open_pack_index(p))
1827 return NULL;
1828 index = p->index_data;
1829 }
57059091 1830 if (n >= p->num_objects)
d72308e0 1831 return NULL;
74e34e1f
NP
1832 index += 4 * 256;
1833 if (p->index_version == 1) {
1834 return index + 24 * n + 4;
1835 } else {
1836 index += 8;
1837 return index + 20 * n;
1838 }
1839}
1840
99093238 1841off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
1842{
1843 const unsigned char *index = p->index_data;
1844 index += 4 * 256;
1845 if (p->index_version == 1) {
1846 return ntohl(*((uint32_t *)(index + 24 * n)));
1847 } else {
1848 uint32_t off;
1849 index += 8 + p->num_objects * (20 + 4);
1850 off = ntohl(*((uint32_t *)(index + 4 * n)));
1851 if (!(off & 0x80000000))
1852 return off;
1853 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1854 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1855 ntohl(*((uint32_t *)(index + 4)));
1856 }
9a217f2a
JH
1857}
1858
c4001d92 1859off_t find_pack_entry_one(const unsigned char *sha1,
43057304 1860 struct packed_git *p)
1f688557 1861{
42873078 1862 const uint32_t *level1_ofs = p->index_data;
42873078 1863 const unsigned char *index = p->index_data;
628522ec
JH
1864 unsigned hi, lo, stride;
1865 static int use_lookup = -1;
1866 static int debug_lookup = -1;
1867
1868 if (debug_lookup < 0)
1869 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 1870
d079837e
SP
1871 if (!index) {
1872 if (open_pack_index(p))
1873 return 0;
1874 level1_ofs = p->index_data;
1875 index = p->index_data;
1876 }
74e34e1f
NP
1877 if (p->index_version > 1) {
1878 level1_ofs += 2;
1879 index += 8;
1880 }
42873078 1881 index += 4 * 256;
74e34e1f
NP
1882 hi = ntohl(level1_ofs[*sha1]);
1883 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
1884 if (p->index_version > 1) {
1885 stride = 20;
1886 } else {
1887 stride = 24;
1888 index += 4;
1889 }
1890
1891 if (debug_lookup)
6e1c2344 1892 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
1893 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1894
1895 if (use_lookup < 0)
1896 use_lookup = !!getenv("GIT_USE_LOOKUP");
1897 if (use_lookup) {
1898 int pos = sha1_entry_pos(index, stride, 0,
1899 lo, hi, p->num_objects, sha1);
1900 if (pos < 0)
1901 return 0;
1902 return nth_packed_object_offset(p, pos);
1903 }
1f688557
JH
1904
1905 do {
74e34e1f 1906 unsigned mi = (lo + hi) / 2;
628522ec
JH
1907 int cmp = hashcmp(index + mi * stride, sha1);
1908
1909 if (debug_lookup)
1910 printf("lo %u hi %u rg %u mi %u\n",
1911 lo, hi, hi - lo, mi);
43057304 1912 if (!cmp)
74e34e1f 1913 return nth_packed_object_offset(p, mi);
1f688557
JH
1914 if (cmp > 0)
1915 hi = mi;
1916 else
1917 lo = mi+1;
1918 } while (lo < hi);
1919 return 0;
1920}
1921
4d6acb70 1922static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557 1923{
f7c22cc6 1924 static struct packed_git *last_found = (void *)1;
1f688557 1925 struct packed_git *p;
c4001d92 1926 off_t offset;
43057304 1927
1f688557 1928 prepare_packed_git();
f7c22cc6
NP
1929 if (!packed_git)
1930 return 0;
1931 p = (last_found == (void *)1) ? packed_git : last_found;
1f688557 1932
f7c22cc6 1933 do {
8eca0b47
NP
1934 if (p->num_bad_objects) {
1935 unsigned i;
1936 for (i = 0; i < p->num_bad_objects; i++)
1937 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1938 goto next;
1939 }
1940
43057304
NP
1941 offset = find_pack_entry_one(sha1, p);
1942 if (offset) {
c715f783
SP
1943 /*
1944 * We are about to tell the caller where they can
1945 * locate the requested object. We better make
1946 * sure the packfile is still here and can be
1947 * accessed before supplying that answer, as
1948 * it may have been deleted since the index
1949 * was loaded!
1950 */
1951 if (p->pack_fd == -1 && open_packed_git(p)) {
1952 error("packfile %s cannot be accessed", p->pack_name);
f7c22cc6 1953 goto next;
c715f783 1954 }
43057304
NP
1955 e->offset = offset;
1956 e->p = p;
1957 hashcpy(e->sha1, sha1);
f7c22cc6 1958 last_found = p;
1f688557 1959 return 1;
43057304 1960 }
f7c22cc6
NP
1961
1962 next:
1963 if (p == last_found)
1964 p = packed_git;
1965 else
1966 p = p->next;
1967 if (p == last_found)
1968 p = p->next;
1969 } while (p);
1f688557
JH
1970 return 0;
1971}
1972
a6080a0a 1973struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
1974 struct packed_git *packs)
1975{
1976 struct packed_git *p;
bf592c50
DB
1977
1978 for (p = packs; p; p = p->next) {
43057304 1979 if (find_pack_entry_one(sha1, p))
bf592c50
DB
1980 return p;
1981 }
1982 return NULL;
21666f1a 1983
bf592c50
DB
1984}
1985
21666f1a 1986static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
65c2e0c3 1987{
36e4d74a 1988 int status;
65c2e0c3
JH
1989 unsigned long mapsize, size;
1990 void *map;
1991 z_stream stream;
d65a16f6 1992 char hdr[32];
65c2e0c3 1993
bb6b8e4f 1994 map = map_sha1_file(sha1, &mapsize);
f0df4ed5
JS
1995 if (!map)
1996 return error("unable to find %s", sha1_to_hex(sha1));
36e4d74a
JH
1997 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1998 status = error("unable to unpack %s header",
1999 sha1_to_hex(sha1));
21666f1a 2000 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 2001 status = error("unable to parse %s header", sha1_to_hex(sha1));
21666f1a
NP
2002 else if (sizep)
2003 *sizep = size;
39c68542 2004 git_inflate_end(&stream);
65c2e0c3
JH
2005 munmap(map, mapsize);
2006 return status;
2007}
2008
21666f1a 2009int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
f0df4ed5 2010{
f0df4ed5 2011 struct pack_entry e;
ddd63e64 2012 int status;
f0df4ed5 2013
cd673c1f 2014 if (!find_pack_entry(sha1, &e)) {
ddd63e64
SG
2015 /* Most likely it's a loose object. */
2016 status = sha1_loose_object_info(sha1, sizep);
2017 if (status >= 0)
2018 return status;
2019
2020 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 2021 reprepare_packed_git();
cd673c1f 2022 if (!find_pack_entry(sha1, &e))
ddd63e64 2023 return status;
f0df4ed5 2024 }
3d77d877
NP
2025
2026 status = packed_object_info(e.p, e.offset, sizep);
2027 if (status < 0) {
2028 mark_bad_packed_object(e.p, sha1);
2029 status = sha1_object_info(sha1, sizep);
2030 }
2031
2032 return status;
f0df4ed5
JS
2033}
2034
21666f1a
NP
2035static void *read_packed_sha1(const unsigned char *sha1,
2036 enum object_type *type, unsigned long *size)
1f688557
JH
2037{
2038 struct pack_entry e;
8eca0b47 2039 void *data;
1f688557 2040
cd673c1f 2041 if (!find_pack_entry(sha1, &e))
1f688557 2042 return NULL;
8eca0b47
NP
2043 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2044 if (!data) {
2045 /*
2046 * We're probably in deep shit, but let's try to fetch
2047 * the required object anyway from another pack or loose.
2048 * This should happen only in the presence of a corrupted
2049 * pack, and is better than failing outright.
2050 */
2051 error("failed to read object %s at offset %"PRIuMAX" from %s",
2052 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2053 mark_bad_packed_object(e.p, sha1);
ac939109 2054 data = read_object(sha1, type, size);
8eca0b47
NP
2055 }
2056 return data;
1f688557
JH
2057}
2058
d66b37bb
JH
2059/*
2060 * This is meant to hold a *small* number of objects that you would
2061 * want read_sha1_file() to be able to return, but yet you do not want
2062 * to write them into the object store (e.g. a browse-only
2063 * application).
2064 */
2065static struct cached_object {
2066 unsigned char sha1[20];
21666f1a 2067 enum object_type type;
d66b37bb
JH
2068 void *buf;
2069 unsigned long size;
2070} *cached_objects;
2071static int cached_object_nr, cached_object_alloc;
2072
346245a1 2073static struct cached_object empty_tree = {
14d9c578 2074 EMPTY_TREE_SHA1_BIN,
346245a1
JK
2075 OBJ_TREE,
2076 "",
2077 0
2078};
2079
d66b37bb
JH
2080static struct cached_object *find_cached_object(const unsigned char *sha1)
2081{
2082 int i;
2083 struct cached_object *co = cached_objects;
2084
2085 for (i = 0; i < cached_object_nr; i++, co++) {
2086 if (!hashcmp(co->sha1, sha1))
2087 return co;
2088 }
346245a1
JK
2089 if (!hashcmp(sha1, empty_tree.sha1))
2090 return &empty_tree;
d66b37bb
JH
2091 return NULL;
2092}
2093
21666f1a
NP
2094int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2095 unsigned char *sha1)
d66b37bb
JH
2096{
2097 struct cached_object *co;
2098
21666f1a 2099 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2100 if (has_sha1_file(sha1) || find_cached_object(sha1))
2101 return 0;
2102 if (cached_object_alloc <= cached_object_nr) {
2103 cached_object_alloc = alloc_nr(cached_object_alloc);
2104 cached_objects = xrealloc(cached_objects,
2105 sizeof(*cached_objects) *
2106 cached_object_alloc);
2107 }
2108 co = &cached_objects[cached_object_nr++];
2109 co->size = len;
21666f1a 2110 co->type = type;
efa13f7b
JH
2111 co->buf = xmalloc(len);
2112 memcpy(co->buf, buf, len);
d66b37bb
JH
2113 hashcpy(co->sha1, sha1);
2114 return 0;
2115}
2116
c2c5b270
CC
2117static void *read_object(const unsigned char *sha1, enum object_type *type,
2118 unsigned long *size)
0fcfd160
LT
2119{
2120 unsigned long mapsize;
2121 void *map, *buf;
d66b37bb
JH
2122 struct cached_object *co;
2123
2124 co = find_cached_object(sha1);
2125 if (co) {
21666f1a 2126 *type = co->type;
d66b37bb 2127 *size = co->size;
182af834 2128 return xmemdupz(co->buf, co->size);
d66b37bb 2129 }
0fcfd160 2130
8276c007
PE
2131 buf = read_packed_sha1(sha1, type, size);
2132 if (buf)
2133 return buf;
bb6b8e4f 2134 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2135 if (map) {
7efbff75 2136 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2137 munmap(map, mapsize);
2138 return buf;
2139 }
637cdd9d 2140 reprepare_packed_git();
8276c007 2141 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2142}
2143
ac939109
NP
2144void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
2145 unsigned long *size)
2146{
2147 void *data = read_object(sha1, type, size);
2148 /* legacy behavior is to die on corrupted objects */
2149 if (!data && (has_loose_object(sha1) || has_packed_and_bad(sha1)))
2150 die("object %s is corrupted", sha1_to_hex(sha1));
2151 return data;
2152}
2153
40469ee9 2154void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2155 const char *required_type_name,
40469ee9
JH
2156 unsigned long *size,
2157 unsigned char *actual_sha1_return)
f4913f91 2158{
21666f1a 2159 enum object_type type, required_type;
f4913f91
JH
2160 void *buffer;
2161 unsigned long isize;
40469ee9 2162 unsigned char actual_sha1[20];
f4913f91 2163
21666f1a 2164 required_type = type_from_string(required_type_name);
e702496e 2165 hashcpy(actual_sha1, sha1);
40469ee9
JH
2166 while (1) {
2167 int ref_length = -1;
2168 const char *ref_type = NULL;
f4913f91 2169
21666f1a 2170 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2171 if (!buffer)
2172 return NULL;
21666f1a 2173 if (type == required_type) {
40469ee9
JH
2174 *size = isize;
2175 if (actual_sha1_return)
e702496e 2176 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2177 return buffer;
2178 }
2179 /* Handle references */
21666f1a 2180 else if (type == OBJ_COMMIT)
40469ee9 2181 ref_type = "tree ";
21666f1a 2182 else if (type == OBJ_TAG)
40469ee9
JH
2183 ref_type = "object ";
2184 else {
2185 free(buffer);
2186 return NULL;
2187 }
2188 ref_length = strlen(ref_type);
f4913f91 2189
50974ec9
MK
2190 if (ref_length + 40 > isize ||
2191 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2192 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2193 free(buffer);
2194 return NULL;
2195 }
1cf58e72 2196 free(buffer);
40469ee9
JH
2197 /* Now we have the ID of the referred-to object in
2198 * actual_sha1. Check again. */
f4913f91 2199 }
f4913f91
JH
2200}
2201
ce9fbf16 2202static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2203 const char *type, unsigned char *sha1,
d65a16f6 2204 char *hdr, int *hdrlen)
d410c0f5 2205{
9126f009 2206 git_SHA_CTX c;
d410c0f5
JH
2207
2208 /* Generate the header */
d65a16f6 2209 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
2210
2211 /* Sha1.. */
9126f009
NP
2212 git_SHA1_Init(&c);
2213 git_SHA1_Update(&c, hdr, *hdrlen);
2214 git_SHA1_Update(&c, buf, len);
2215 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2216}
2217
230f1322 2218/*
5a688fe4
JH
2219 * Move the just written object into its final resting place.
2220 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2221 * "moving" is only a part of what it does, when no patch between
2222 * master to pu changes the call sites of this function.
230f1322 2223 */
839837b9 2224int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322 2225{
e32c0a9c 2226 int ret = 0;
5a688fe4 2227
be66a6c4
JS
2228 if (unreliable_hardlinks)
2229 goto try_rename;
2230 else if (link(tmpfile, filename))
e32c0a9c 2231 ret = errno;
7ebb6fca
LT
2232
2233 /*
2234 * Coda hack - coda doesn't like cross-directory links,
2235 * so we fall back to a rename, which will mean that it
2236 * won't be able to check collisions, but that's not a
2237 * big deal.
2238 *
2239 * The same holds for FAT formatted media.
2240 *
3be1f18e 2241 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
2242 * left to unlink.
2243 */
2244 if (ret && ret != EEXIST) {
be66a6c4 2245 try_rename:
7ebb6fca 2246 if (!rename(tmpfile, filename))
3be1f18e 2247 goto out;
9e48b389 2248 ret = errno;
230f1322
LT
2249 }
2250 unlink(tmpfile);
2251 if (ret) {
2252 if (ret != EEXIST) {
916d081b 2253 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
230f1322
LT
2254 }
2255 /* FIXME!!! Collision check here ? */
2256 }
2257
3be1f18e 2258out:
17e61b82 2259 if (set_shared_perm(filename, (S_IFREG|0444)))
5a688fe4 2260 return error("unable to set permission to '%s'", filename);
230f1322
LT
2261 return 0;
2262}
2263
4d548150
LT
2264static int write_buffer(int fd, const void *buf, size_t len)
2265{
d34cf19b 2266 if (write_in_full(fd, buf, len) < 0)
93822c22 2267 return error("file write error (%s)", strerror(errno));
4d548150
LT
2268 return 0;
2269}
2270
ce9fbf16 2271int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2272 unsigned char *sha1)
2273{
d65a16f6 2274 char hdr[32];
abdc3fc8
RS
2275 int hdrlen;
2276 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2277 return 0;
2278}
2279
e9039dd3
LT
2280/* Finalize a file on disk, and close it. */
2281static void close_sha1_file(int fd)
2282{
aafe9fba
LT
2283 if (fsync_object_files)
2284 fsync_or_die(fd, "sha1 file");
e9039dd3 2285 if (close(fd) != 0)
e8bd78c3 2286 die("error when closing sha1 file (%s)", strerror(errno));
e9039dd3
LT
2287}
2288
5723fe7e
LT
2289/* Size of directory component, including the ending '/' */
2290static inline int directory_size(const char *filename)
2291{
2292 const char *s = strrchr(filename, '/');
2293 if (!s)
2294 return 0;
2295 return s - filename + 1;
2296}
2297
2298/*
2299 * This creates a temporary file in the same directory as the final
2300 * 'filename'
2301 *
2302 * We want to avoid cross-directory filename renames, because those
2303 * can have problems on various filesystems (FAT, NFS, Coda).
2304 */
2305static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2306{
2307 int fd, dirlen = directory_size(filename);
2308
2309 if (dirlen + 20 > bufsiz) {
2310 errno = ENAMETOOLONG;
2311 return -1;
2312 }
2313 memcpy(buffer, filename, dirlen);
2314 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2315 fd = mkstemp(buffer);
cbacbf4e 2316 if (fd < 0 && dirlen && errno == ENOENT) {
5723fe7e 2317 /* Make sure the directory exists */
6ff6af62 2318 memcpy(buffer, filename, dirlen);
5723fe7e 2319 buffer[dirlen-1] = 0;
3bfaf018 2320 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
5723fe7e
LT
2321 return -1;
2322
2323 /* Try again */
2324 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2325 fd = mkstemp(buffer);
2326 }
2327 return fd;
2328}
2329
bbac7311
NP
2330static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2331 void *buf, unsigned long len, time_t mtime)
0fcfd160 2332{
915308b1
JK
2333 int fd, ret;
2334 size_t size;
bf0f910d 2335 unsigned char *compressed;
0fcfd160 2336 z_stream stream;
706bc531 2337 char *filename;
aac17941 2338 static char tmpfile[PATH_MAX];
a44c9a5e 2339
972a9155 2340 filename = sha1_file_name(sha1);
5723fe7e 2341 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
aac17941 2342 if (fd < 0) {
35243577 2343 if (errno == EACCES)
916d081b
PB
2344 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2345 else
2346 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
aac17941
LT
2347 }
2348
0fcfd160
LT
2349 /* Set it up */
2350 memset(&stream, 0, sizeof(stream));
12f6c308 2351 deflateInit(&stream, zlib_compression_level);
93821bd9 2352 size = 8 + deflateBound(&stream, len+hdrlen);
812666c8 2353 compressed = xmalloc(size);
0fcfd160
LT
2354
2355 /* Compress it */
0fcfd160
LT
2356 stream.next_out = compressed;
2357 stream.avail_out = size;
a44c9a5e
LT
2358
2359 /* First header.. */
d65a16f6 2360 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2361 stream.avail_in = hdrlen;
726f852b
NP
2362 while (deflate(&stream, 0) == Z_OK)
2363 /* nothing */;
a44c9a5e
LT
2364
2365 /* Then the data itself.. */
2366 stream.next_in = buf;
2367 stream.avail_in = len;
ac54c277
LT
2368 ret = deflate(&stream, Z_FINISH);
2369 if (ret != Z_STREAM_END)
2370 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2371
2372 ret = deflateEnd(&stream);
2373 if (ret != Z_OK)
2374 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2375
0fcfd160
LT
2376 size = stream.total_out;
2377
4d548150
LT
2378 if (write_buffer(fd, compressed, size) < 0)
2379 die("unable to write sha1 file");
e9039dd3 2380 close_sha1_file(fd);
383f85b7 2381 free(compressed);
0fcfd160 2382
bbac7311
NP
2383 if (mtime) {
2384 struct utimbuf utb;
2385 utb.actime = mtime;
2386 utb.modtime = mtime;
2387 if (utime(tmpfile, &utb) < 0)
2388 warning("failed utime() on %s: %s",
2389 tmpfile, strerror(errno));
2390 }
2391
230f1322 2392 return move_temp_to_file(tmpfile, filename);
0fcfd160 2393}
8237b185 2394
bbac7311
NP
2395int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2396{
2397 unsigned char sha1[20];
2398 char hdr[32];
2399 int hdrlen;
2400
2401 /* Normally if we have it in the pack then we do not bother writing
2402 * it out into .git/objects/??/?{38} file.
2403 */
2404 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2405 if (returnsha1)
2406 hashcpy(returnsha1, sha1);
2407 if (has_sha1_file(sha1))
2408 return 0;
2409 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2410}
2411
2412int force_object_loose(const unsigned char *sha1, time_t mtime)
2413{
bbac7311
NP
2414 void *buf;
2415 unsigned long len;
2416 enum object_type type;
2417 char hdr[32];
2418 int hdrlen;
1fb23e65 2419 int ret;
bbac7311 2420
c529d75a 2421 if (has_loose_object(sha1))
bbac7311
NP
2422 return 0;
2423 buf = read_packed_sha1(sha1, &type, &len);
2424 if (!buf)
2425 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2426 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1fb23e65
BS
2427 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2428 free(buf);
2429
2430 return ret;
bbac7311
NP
2431}
2432
bf592c50
DB
2433int has_pack_index(const unsigned char *sha1)
2434{
2435 struct stat st;
2436 if (stat(sha1_pack_index_name(sha1), &st))
2437 return 0;
2438 return 1;
2439}
2440
2441int has_pack_file(const unsigned char *sha1)
2442{
2443 struct stat st;
2444 if (stat(sha1_pack_name(sha1), &st))
2445 return 0;
2446 return 1;
2447}
2448
cd673c1f
JH
2449int has_sha1_pack(const unsigned char *sha1)
2450{
2451 struct pack_entry e;
2452 return find_pack_entry(sha1, &e);
2453}
2454
8237b185
DB
2455int has_sha1_file(const unsigned char *sha1)
2456{
1f688557
JH
2457 struct pack_entry e;
2458
cd673c1f 2459 if (find_pack_entry(sha1, &e))
1f688557 2460 return 1;
c529d75a 2461 return has_loose_object(sha1);
8237b185 2462}
74400e71 2463
43df4f86
DP
2464static int index_mem(unsigned char *sha1, void *buf, size_t size,
2465 int write_object, enum object_type type, const char *path)
e7332f96 2466{
6c510bee 2467 int ret, re_allocated = 0;
74400e71 2468
7672db20 2469 if (!type)
edaec3fb 2470 type = OBJ_BLOB;
6c510bee
LT
2471
2472 /*
2473 * Convert blobs to git internal format
2474 */
43df4f86 2475 if ((type == OBJ_BLOB) && path) {
f285a2d7 2476 struct strbuf nbuf = STRBUF_INIT;
21e5ad50
SP
2477 if (convert_to_git(path, buf, size, &nbuf,
2478 write_object ? safe_crlf : 0)) {
b315c5c0 2479 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
2480 re_allocated = 1;
2481 }
2482 }
2483
7672db20 2484 if (write_object)
edaec3fb 2485 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 2486 else
edaec3fb 2487 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 2488 if (re_allocated)
6c510bee 2489 free(buf);
43df4f86
DP
2490 return ret;
2491}
2492
2493int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2494 enum object_type type, const char *path)
2495{
2496 int ret;
2497 size_t size = xsize_t(st->st_size);
2498
2499 if (!S_ISREG(st->st_mode)) {
f285a2d7 2500 struct strbuf sbuf = STRBUF_INIT;
43df4f86
DP
2501 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2502 ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2503 type, path);
2504 else
2505 ret = -1;
2506 strbuf_release(&sbuf);
2507 } else if (size) {
2508 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2509 ret = index_mem(sha1, buf, size, write_object, type, path);
aac17941 2510 munmap(buf, size);
43df4f86
DP
2511 } else
2512 ret = index_mem(sha1, NULL, size, write_object, type, path);
2513 close(fd);
aac17941 2514 return ret;
74400e71 2515}
ec1fcc16
JH
2516
2517int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2518{
2519 int fd;
b760d3aa 2520 struct strbuf sb = STRBUF_INIT;
ec1fcc16
JH
2521
2522 switch (st->st_mode & S_IFMT) {
2523 case S_IFREG:
2524 fd = open(path, O_RDONLY);
2525 if (fd < 0)
2526 return error("open(\"%s\"): %s", path,
2527 strerror(errno));
53bca91a 2528 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
ec1fcc16
JH
2529 return error("%s: failed to insert into database",
2530 path);
2531 break;
2532 case S_IFLNK:
b760d3aa 2533 if (strbuf_readlink(&sb, path, st->st_size)) {
ec1fcc16 2534 char *errstr = strerror(errno);
ec1fcc16
JH
2535 return error("readlink(\"%s\"): %s", path,
2536 errstr);
2537 }
abdc3fc8 2538 if (!write_object)
b760d3aa
LT
2539 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2540 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
ec1fcc16
JH
2541 return error("%s: failed to insert into database",
2542 path);
b760d3aa 2543 strbuf_release(&sb);
ec1fcc16 2544 break;
f35a6d3b
LT
2545 case S_IFDIR:
2546 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
2547 default:
2548 return error("%s: unsupported file type", path);
2549 }
2550 return 0;
2551}
a69e5429
JH
2552
2553int read_pack_header(int fd, struct pack_header *header)
2554{
c697ad14
HO
2555 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2556 /* "eof before pack header was fully read" */
2557 return PH_ERROR_EOF;
2558
a69e5429
JH
2559 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2560 /* "protocol error (pack signature mismatch detected)" */
2561 return PH_ERROR_PACK_SIGNATURE;
2562 if (!pack_version_ok(header->hdr_version))
2563 /* "protocol error (pack version unsupported)" */
2564 return PH_ERROR_PROTOCOL;
2565 return 0;
2566}