make check_object() resilient to pack corruptions
[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
c529d75a 426static int has_loose_object(const unsigned char *sha1)
ace1534d 427{
ace1534d 428 char *name = sha1_file_name(sha1);
d5a63b99 429 struct alternate_object_database *alt;
ace1534d 430
c529d75a
LT
431 if (!access(name, F_OK))
432 return 1;
9a217f2a 433 prepare_alt_odb();
d5a63b99
JH
434 for (alt = alt_odb_list; alt; alt = alt->next) {
435 name = alt->name;
ace1534d 436 fill_sha1_path(name, sha1);
c529d75a
LT
437 if (!access(alt->base, F_OK))
438 return 1;
ace1534d 439 }
c529d75a 440 return 0;
ace1534d
JH
441}
442
60bb8b14 443static unsigned int pack_used_ctr;
a53128b6
SP
444static unsigned int pack_mmap_calls;
445static unsigned int peak_pack_open_windows;
446static unsigned int pack_open_windows;
447static size_t peak_pack_mapped;
60bb8b14 448static size_t pack_mapped;
9a217f2a 449struct packed_git *packed_git;
1f688557 450
b79d18c9 451void pack_report(void)
a53128b6
SP
452{
453 fprintf(stderr,
e05db0fd
PR
454 "pack_report: getpagesize() = %10" SZ_FMT "\n"
455 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
456 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
457 sz_fmt(getpagesize()),
458 sz_fmt(packed_git_window_size),
459 sz_fmt(packed_git_limit));
a53128b6
SP
460 fprintf(stderr,
461 "pack_report: pack_used_ctr = %10u\n"
462 "pack_report: pack_mmap_calls = %10u\n"
463 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
464 "pack_report: pack_mapped = "
465 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
466 pack_used_ctr,
467 pack_mmap_calls,
468 pack_open_windows, peak_pack_open_windows,
9e42d6a1 469 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
470}
471
42873078 472static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
473{
474 void *idx_map;
42873078 475 struct pack_idx_header *hdr;
dc49cd76 476 size_t idx_size;
74e34e1f 477 uint32_t version, nr, i, *index;
e93ec6f9 478 int fd = open(path, O_RDONLY);
1f688557 479 struct stat st;
42873078 480
1f688557
JH
481 if (fd < 0)
482 return -1;
483 if (fstat(fd, &st)) {
484 close(fd);
485 return -1;
486 }
dc49cd76 487 idx_size = xsize_t(st.st_size);
2d88451b
SP
488 if (idx_size < 4 * 256 + 20 + 20) {
489 close(fd);
490 return error("index file %s is too small", path);
491 }
c4712e45 492 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 493 close(fd);
1f688557 494
42873078
NP
495 hdr = idx_map;
496 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
497 version = ntohl(hdr->idx_version);
498 if (version < 2 || version > 2) {
499 munmap(idx_map, idx_size);
6e1c2344 500 return error("index file %s is version %"PRIu32
74e34e1f
NP
501 " and is not supported by this binary"
502 " (try upgrading GIT to a newer version)",
503 path, version);
504 }
505 } else
506 version = 1;
df1b059d 507
1f688557 508 nr = 0;
42873078 509 index = idx_map;
74e34e1f
NP
510 if (version > 1)
511 index += 2; /* skip index header */
1f688557 512 for (i = 0; i < 256; i++) {
326bf396 513 uint32_t n = ntohl(index[i]);
2d88451b
SP
514 if (n < nr) {
515 munmap(idx_map, idx_size);
df1b059d 516 return error("non-monotonic index %s", path);
2d88451b 517 }
1f688557
JH
518 nr = n;
519 }
520
74e34e1f
NP
521 if (version == 1) {
522 /*
523 * Total size:
524 * - 256 index entries 4 bytes each
525 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
526 * - 20-byte SHA1 of the packfile
527 * - 20-byte SHA1 file checksum
528 */
529 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
530 munmap(idx_map, idx_size);
eef427a0 531 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
532 }
533 } else if (version == 2) {
534 /*
535 * Minimum size:
536 * - 8 bytes of header
537 * - 256 index entries 4 bytes each
538 * - 20-byte sha1 entry * nr
539 * - 4-byte crc entry * nr
540 * - 4-byte offset entry * nr
541 * - 20-byte SHA1 of the packfile
542 * - 20-byte SHA1 file checksum
543 * And after the 4-byte offset table might be a
544 * variable sized table containing 8-byte entries
545 * for offsets larger than 2^31.
546 */
547 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
548 unsigned long max_size = min_size;
549 if (nr)
550 max_size += (nr - 1)*8;
551 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 552 munmap(idx_map, idx_size);
eef427a0 553 return error("wrong index v2 file size in %s", path);
74e34e1f 554 }
7109c889
JH
555 if (idx_size != min_size &&
556 /*
557 * make sure we can deal with large pack offsets.
558 * 31-bit signed offset won't be enough, neither
559 * 32-bit unsigned one will be.
560 */
561 (sizeof(off_t) <= 4)) {
562 munmap(idx_map, idx_size);
563 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 564 }
2d88451b 565 }
1f688557 566
74e34e1f 567 p->index_version = version;
42873078
NP
568 p->index_data = idx_map;
569 p->index_size = idx_size;
57059091 570 p->num_objects = nr;
1f688557
JH
571 return 0;
572}
573
bc8e478a 574int open_pack_index(struct packed_git *p)
d079837e
SP
575{
576 char *idx_name;
577 int ret;
578
579 if (p->index_data)
580 return 0;
581
582 idx_name = xstrdup(p->pack_name);
583 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
584 ret = check_packed_git_idx(idx_name, p);
585 free(idx_name);
586 return ret;
587}
588
11daf39b
SP
589static void scan_windows(struct packed_git *p,
590 struct packed_git **lru_p,
591 struct pack_window **lru_w,
592 struct pack_window **lru_l)
1f688557 593{
11daf39b
SP
594 struct pack_window *w, *w_l;
595
596 for (w_l = NULL, w = p->windows; w; w = w->next) {
597 if (!w->inuse_cnt) {
598 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
599 *lru_p = p;
600 *lru_w = w;
601 *lru_l = w_l;
54044bf8 602 }
54044bf8 603 }
11daf39b 604 w_l = w;
f9253394 605 }
11daf39b
SP
606}
607
d1efefa4 608static int unuse_one_window(struct packed_git *current, int keep_fd)
11daf39b
SP
609{
610 struct packed_git *p, *lru_p = NULL;
611 struct pack_window *lru_w = NULL, *lru_l = NULL;
612
613 if (current)
614 scan_windows(current, &lru_p, &lru_w, &lru_l);
615 for (p = packed_git; p; p = p->next)
616 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
617 if (lru_p) {
618 munmap(lru_w->base, lru_w->len);
619 pack_mapped -= lru_w->len;
620 if (lru_l)
621 lru_l->next = lru_w->next;
622 else {
623 lru_p->windows = lru_w->next;
d1efefa4 624 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
54044bf8
SP
625 close(lru_p->pack_fd);
626 lru_p->pack_fd = -1;
627 }
628 }
629 free(lru_w);
a53128b6 630 pack_open_windows--;
54044bf8
SP
631 return 1;
632 }
633 return 0;
f9253394
JH
634}
635
d1efefa4 636void release_pack_memory(size_t need, int fd)
97bfeb34
SP
637{
638 size_t cur = pack_mapped;
d1efefa4 639 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
97bfeb34
SP
640 ; /* nothing */
641}
642
c9ced051
SP
643void close_pack_windows(struct packed_git *p)
644{
645 while (p->windows) {
646 struct pack_window *w = p->windows;
647
648 if (w->inuse_cnt)
649 die("pack '%s' still has open windows to it",
650 p->pack_name);
651 munmap(w->base, w->len);
652 pack_mapped -= w->len;
653 pack_open_windows--;
654 p->windows = w->next;
655 free(w);
656 }
657}
658
03e79c88 659void unuse_pack(struct pack_window **w_cursor)
f9253394 660{
03e79c88
SP
661 struct pack_window *w = *w_cursor;
662 if (w) {
663 w->inuse_cnt--;
664 *w_cursor = NULL;
665 }
1f688557
JH
666}
667
3cf8b462
SP
668/*
669 * Do not call this directly as this leaks p->pack_fd on error return;
670 * call open_packed_git() instead.
671 */
672static int open_packed_git_1(struct packed_git *p)
1f688557 673{
9bc879c1
SP
674 struct stat st;
675 struct pack_header hdr;
676 unsigned char sha1[20];
677 unsigned char *idx_sha1;
2c039da8 678 long fd_flag;
9bc879c1 679
d079837e
SP
680 if (!p->index_data && open_pack_index(p))
681 return error("packfile %s index unavailable", p->pack_name);
682
9bc879c1
SP
683 p->pack_fd = open(p->pack_name, O_RDONLY);
684 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 685 return -1;
9bc879c1
SP
686
687 /* If we created the struct before we had the pack we lack size. */
bf592c50 688 if (!p->pack_size) {
bf592c50 689 if (!S_ISREG(st.st_mode))
072db278 690 return error("packfile %s not a regular file", p->pack_name);
bf592c50 691 p->pack_size = st.st_size;
9bc879c1 692 } else if (p->pack_size != st.st_size)
072db278 693 return error("packfile %s size changed", p->pack_name);
9bc879c1 694
2c039da8
JH
695 /* We leave these file descriptors open with sliding mmap;
696 * there is no point keeping them open across exec(), though.
697 */
698 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
699 if (fd_flag < 0)
072db278 700 return error("cannot determine file descriptor flags");
2c039da8
JH
701 fd_flag |= FD_CLOEXEC;
702 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 703 return error("cannot set FD_CLOEXEC");
2c039da8 704
9bc879c1 705 /* Verify we recognize this pack file format. */
e6e2bd62 706 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 707 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 708 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 709 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 710 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
711 return error("packfile %s is version %"PRIu32" and not"
712 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
713 p->pack_name, ntohl(hdr.hdr_version));
714
715 /* Verify the pack matches its index. */
57059091 716 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
717 return error("packfile %s claims to have %"PRIu32" objects"
718 " while index indicates %"PRIu32" objects",
57059091
NP
719 p->pack_name, ntohl(hdr.hdr_entries),
720 p->num_objects);
9bc879c1 721 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 722 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 723 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 724 return error("packfile %s signature is unavailable", p->pack_name);
42873078 725 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 726 if (hashcmp(sha1, idx_sha1))
072db278
SP
727 return error("packfile %s does not match index", p->pack_name);
728 return 0;
9bc879c1
SP
729}
730
3cf8b462
SP
731static int open_packed_git(struct packed_git *p)
732{
733 if (!open_packed_git_1(p))
734 return 0;
735 if (p->pack_fd != -1) {
736 close(p->pack_fd);
737 p->pack_fd = -1;
738 }
739 return -1;
740}
741
c4001d92 742static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
743{
744 /* We must promise at least 20 bytes (one hash) after the
745 * offset is available from this window, otherwise the offset
746 * is not actually in this window and a different window (which
747 * has that one hash excess) must be used. This is to support
748 * the object header and delta base parsing routines below.
749 */
750 off_t win_off = win->offset;
751 return win_off <= offset
752 && (offset + 20) <= (win_off + win->len);
753}
754
03e79c88
SP
755unsigned char* use_pack(struct packed_git *p,
756 struct pack_window **w_cursor,
c4001d92 757 off_t offset,
03e79c88 758 unsigned int *left)
9bc879c1 759{
60bb8b14 760 struct pack_window *win = *w_cursor;
03e79c88 761
072db278
SP
762 if (p->pack_fd == -1 && open_packed_git(p))
763 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
764
765 /* Since packfiles end in a hash of their content and its
766 * pointless to ask for an offset into the middle of that
767 * hash, and the in_window function above wouldn't match
768 * don't allow an offset too close to the end of the file.
769 */
770 if (offset > (p->pack_size - 20))
771 die("offset beyond end of packfile (truncated pack?)");
772
773 if (!win || !in_window(win, offset)) {
774 if (win)
775 win->inuse_cnt--;
776 for (win = p->windows; win; win = win->next) {
777 if (in_window(win, offset))
778 break;
779 }
780 if (!win) {
78a28df9 781 size_t window_align = packed_git_window_size / 2;
dc49cd76 782 off_t len;
60bb8b14 783 win = xcalloc(1, sizeof(*win));
78a28df9 784 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
785 len = p->pack_size - win->offset;
786 if (len > packed_git_window_size)
787 len = packed_git_window_size;
788 win->len = (size_t)len;
60bb8b14 789 pack_mapped += win->len;
11daf39b 790 while (packed_git_limit < pack_mapped
d1efefa4 791 && unuse_one_window(p, p->pack_fd))
60bb8b14 792 ; /* nothing */
c4712e45 793 win->base = xmmap(NULL, win->len,
60bb8b14
SP
794 PROT_READ, MAP_PRIVATE,
795 p->pack_fd, win->offset);
796 if (win->base == MAP_FAILED)
73b4e4be
SP
797 die("packfile %s cannot be mapped: %s",
798 p->pack_name,
799 strerror(errno));
a53128b6
SP
800 pack_mmap_calls++;
801 pack_open_windows++;
802 if (pack_mapped > peak_pack_mapped)
803 peak_pack_mapped = pack_mapped;
804 if (pack_open_windows > peak_pack_open_windows)
805 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
806 win->next = p->windows;
807 p->windows = win;
808 }
1f688557 809 }
03e79c88
SP
810 if (win != *w_cursor) {
811 win->last_used = pack_used_ctr++;
812 win->inuse_cnt++;
813 *w_cursor = win;
814 }
60bb8b14 815 offset -= win->offset;
03e79c88 816 if (left)
dc49cd76 817 *left = win->len - xsize_t(offset);
03e79c88 818 return win->base + offset;
1f688557
JH
819}
820
27d69a46
NP
821static struct packed_git *alloc_packed_git(int extra)
822{
823 struct packed_git *p = xmalloc(sizeof(*p) + extra);
824 memset(p, 0, sizeof(*p));
825 p->pack_fd = -1;
826 return p;
827}
828
42873078 829struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557
JH
830{
831 struct stat st;
27d69a46 832 struct packed_git *p = alloc_packed_git(path_len + 2);
1f688557 833
42873078
NP
834 /*
835 * Make sure a corresponding .pack file exists and that
836 * the index looks sane.
837 */
838 path_len -= strlen(".idx");
27d69a46
NP
839 if (path_len < 1) {
840 free(p);
1f688557 841 return NULL;
27d69a46 842 }
42873078
NP
843 memcpy(p->pack_name, path, path_len);
844 strcpy(p->pack_name + path_len, ".pack");
d079837e 845 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 846 free(p);
1f688557
JH
847 return NULL;
848 }
42873078 849
1f688557
JH
850 /* ok, it looks sane as far as we can check without
851 * actually mapping the pack file.
852 */
1f688557 853 p->pack_size = st.st_size;
9d835df2 854 p->pack_local = local;
b867092f 855 p->mtime = st.st_mtime;
42873078
NP
856 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
857 hashclr(p->sha1);
1f688557
JH
858 return p;
859}
860
bf592c50 861struct packed_git *parse_pack_index(unsigned char *sha1)
c508df5e 862{
79c6dca4 863 const char *idx_path = sha1_pack_index_name(sha1);
42873078 864 const char *path = sha1_pack_name(sha1);
27d69a46 865 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
bf592c50 866
27d69a46
NP
867 strcpy(p->pack_name, path);
868 hashcpy(p->sha1, sha1);
42873078
NP
869 if (check_packed_git_idx(idx_path, p)) {
870 free(p);
bf592c50 871 return NULL;
42873078 872 }
bf592c50 873
bf592c50
DB
874 return p;
875}
876
877void install_packed_git(struct packed_git *pack)
878{
879 pack->next = packed_git;
880 packed_git = pack;
881}
882
9d835df2 883static void prepare_packed_git_one(char *objdir, int local)
1f688557 884{
9cb18f56
JM
885 /* Ensure that this buffer is large enough so that we can
886 append "/pack/" without clobbering the stack even if
887 strlen(objdir) were PATH_MAX. */
888 char path[PATH_MAX + 1 + 4 + 1 + 1];
1f688557
JH
889 int len;
890 DIR *dir;
891 struct dirent *de;
892
893 sprintf(path, "%s/pack", objdir);
894 len = strlen(path);
895 dir = opendir(path);
b5b16990 896 if (!dir) {
26125f6b 897 if (errno != ENOENT)
bd2afde8 898 error("unable to open object pack directory: %s: %s",
26125f6b 899 path, strerror(errno));
1f688557 900 return;
b5b16990 901 }
1f688557
JH
902 path[len++] = '/';
903 while ((de = readdir(dir)) != NULL) {
904 int namelen = strlen(de->d_name);
905 struct packed_git *p;
906
5bb1cda5 907 if (!has_extension(de->d_name, ".idx"))
1f688557
JH
908 continue;
909
9cb18f56
JM
910 if (len + namelen + 1 > sizeof(path))
911 continue;
912
54a15a8d 913 /* Don't reopen a pack we already have. */
1f688557 914 strcpy(path + len, de->d_name);
86f7780c
JK
915 for (p = packed_git; p; p = p->next) {
916 if (!memcmp(path, p->pack_name, len + namelen - 4))
917 break;
918 }
919 if (p)
920 continue;
54a15a8d
SP
921 /* See if it really is a valid .idx file with corresponding
922 * .pack file that we can map.
923 */
9d835df2 924 p = add_packed_git(path, len + namelen, local);
1f688557
JH
925 if (!p)
926 continue;
625e9421 927 install_packed_git(p);
1f688557 928 }
5b35bcd5 929 closedir(dir);
1f688557
JH
930}
931
b867092f
JH
932static int sort_pack(const void *a_, const void *b_)
933{
934 struct packed_git *a = *((struct packed_git **)a_);
935 struct packed_git *b = *((struct packed_git **)b_);
936 int st;
937
938 /*
939 * Local packs tend to contain objects specific to our
940 * variant of the project than remote ones. In addition,
941 * remote ones could be on a network mounted filesystem.
942 * Favor local ones for these reasons.
943 */
944 st = a->pack_local - b->pack_local;
945 if (st)
946 return -st;
947
948 /*
949 * Younger packs tend to contain more recent objects,
950 * and more recent objects tend to get accessed more
951 * often.
952 */
953 if (a->mtime < b->mtime)
954 return 1;
955 else if (a->mtime == b->mtime)
956 return 0;
957 return -1;
958}
959
960static void rearrange_packed_git(void)
961{
962 struct packed_git **ary, *p;
963 int i, n;
964
965 for (n = 0, p = packed_git; p; p = p->next)
966 n++;
967 if (n < 2)
968 return;
969
970 /* prepare an array of packed_git for easier sorting */
971 ary = xcalloc(n, sizeof(struct packed_git *));
972 for (n = 0, p = packed_git; p; p = p->next)
973 ary[n++] = p;
974
975 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
976
977 /* link them back again */
978 for (i = 0; i < n - 1; i++)
979 ary[i]->next = ary[i + 1];
980 ary[n - 1]->next = NULL;
981 packed_git = ary[0];
982
983 free(ary);
984}
985
637cdd9d 986static int prepare_packed_git_run_once = 0;
9a217f2a 987void prepare_packed_git(void)
1f688557 988{
d5a63b99 989 struct alternate_object_database *alt;
1f688557 990
637cdd9d 991 if (prepare_packed_git_run_once)
1f688557 992 return;
9d835df2 993 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 994 prepare_alt_odb();
d5a63b99 995 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 996 alt->name[-1] = 0;
9d835df2 997 prepare_packed_git_one(alt->base, 0);
1494e038 998 alt->name[-1] = '/';
1f688557 999 }
b867092f 1000 rearrange_packed_git();
637cdd9d
JK
1001 prepare_packed_git_run_once = 1;
1002}
1003
fc04c412 1004void reprepare_packed_git(void)
637cdd9d 1005{
4b480c67 1006 discard_revindex();
637cdd9d
JK
1007 prepare_packed_git_run_once = 0;
1008 prepare_packed_git();
1f688557
JH
1009}
1010
8eca0b47
NP
1011static void mark_bad_packed_object(struct packed_git *p,
1012 const unsigned char *sha1)
1013{
1014 unsigned i;
1015 for (i = 0; i < p->num_bad_objects; i++)
1016 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1017 return;
1018 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1019 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1020 p->num_bad_objects++;
1021}
1022
ac939109
NP
1023static int has_packed_and_bad(const unsigned char *sha1)
1024{
1025 struct packed_git *p;
1026 unsigned i;
1027
1028 for (p = packed_git; p; p = p->next)
1029 for (i = 0; i < p->num_bad_objects; i++)
1030 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1031 return 1;
1032 return 0;
1033}
1034
5d6ccf5c 1035int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
0fcfd160
LT
1036{
1037 unsigned char real_sha1[20];
7cfb5f36 1038 hash_sha1_file(map, size, type, real_sha1);
a89fccd2 1039 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1040}
1041
44d1c19e
LT
1042static int git_open_noatime(const char *name)
1043{
1044 static int sha1_file_open_flag = O_NOATIME;
1045 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1046
1047 /* Might the failure be due to O_NOATIME? */
1048 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1049 fd = open(name, O_RDONLY);
1050 if (fd >= 0)
1051 sha1_file_open_flag = 0;
1052 }
1053 return fd;
1054}
1055
1056static int open_sha1_file(const unsigned char *sha1)
1057{
1058 int fd;
1059 char *name = sha1_file_name(sha1);
1060 struct alternate_object_database *alt;
1061
1062 fd = git_open_noatime(name);
1063 if (fd >= 0)
1064 return fd;
1065
1066 prepare_alt_odb();
1067 errno = ENOENT;
1068 for (alt = alt_odb_list; alt; alt = alt->next) {
1069 name = alt->name;
1070 fill_sha1_path(name, sha1);
1071 fd = git_open_noatime(alt->base);
1072 if (fd >= 0)
1073 return fd;
1074 }
1075 return -1;
1076}
1077
4175e9e3 1078static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1079{
0fcfd160 1080 void *map;
144bde78 1081 int fd;
ace1534d 1082
44d1c19e
LT
1083 fd = open_sha1_file(sha1);
1084 map = NULL;
1085 if (fd >= 0) {
1086 struct stat st;
0fcfd160 1087
44d1c19e
LT
1088 if (!fstat(fd, &st)) {
1089 *size = xsize_t(st.st_size);
1090 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1091 }
44d1c19e 1092 close(fd);
0fcfd160 1093 }
0fcfd160
LT
1094 return map;
1095}
1096
726f852b 1097static int legacy_loose_object(unsigned char *map)
bb6b8e4f
JH
1098{
1099 unsigned int word;
1100
1101 /*
1102 * Is it a zlib-compressed buffer? If so, the first byte
1103 * must be 0x78 (15-bit window size, deflated), and the
1104 * first 16-bit word is evenly divisible by 31
1105 */
1106 word = (map[0] << 8) + map[1];
1107 if (map[0] == 0x78 && !(word % 31))
1108 return 1;
1109 else
1110 return 0;
1111}
1112
09ded04b
NP
1113unsigned long unpack_object_header_buffer(const unsigned char *buf,
1114 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1115{
ad1ed5ee 1116 unsigned shift;
93821bd9 1117 unsigned char c;
93821bd9 1118 unsigned long size;
ad1ed5ee
JH
1119 unsigned long used = 0;
1120
1121 c = buf[used++];
1122 *type = (c >> 4) & 7;
1123 size = c & 15;
1124 shift = 4;
1125 while (c & 0x80) {
09ded04b
NP
1126 if (len <= used || sizeof(long) * 8 <= shift) {
1127 error("bad object header");
ad1ed5ee 1128 return 0;
09ded04b 1129 }
ad1ed5ee
JH
1130 c = buf[used++];
1131 size += (c & 0x7f) << shift;
1132 shift += 7;
1133 }
1134 *sizep = size;
1135 return used;
1136}
1137
1138static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1139{
1140 unsigned long size, used;
1141 static const char valid_loose_object_type[8] = {
1142 0, /* OBJ_EXT */
1143 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1144 0, /* "delta" and others are invalid in a loose object */
93821bd9 1145 };
ad1ed5ee 1146 enum object_type type;
93821bd9 1147
c4483576
LT
1148 /* Get the data stream */
1149 memset(stream, 0, sizeof(*stream));
1150 stream->next_in = map;
1151 stream->avail_in = mapsize;
1152 stream->next_out = buffer;
93821bd9
LT
1153 stream->avail_out = bufsiz;
1154
bb6b8e4f 1155 if (legacy_loose_object(map)) {
93821bd9
LT
1156 inflateInit(stream);
1157 return inflate(stream, 0);
1158 }
1159
726f852b
NP
1160
1161 /*
1162 * There used to be a second loose object header format which
1163 * was meant to mimic the in-pack format, allowing for direct
1164 * copy of the object data. This format turned up not to be
1165 * really worth it and we don't write it any longer. But we
1166 * can still read it.
1167 */
09ded04b 1168 used = unpack_object_header_buffer(map, mapsize, &type, &size);
ad1ed5ee 1169 if (!used || !valid_loose_object_type[type])
93821bd9 1170 return -1;
ad1ed5ee
JH
1171 map += used;
1172 mapsize -= used;
93821bd9
LT
1173
1174 /* Set up the stream for the rest.. */
1175 stream->next_in = map;
1176 stream->avail_in = mapsize;
c4483576 1177 inflateInit(stream);
93821bd9
LT
1178
1179 /* And generate the fake traditional header */
ad1ed5ee 1180 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
df843662 1181 typename(type), size);
93821bd9 1182 return 0;
c4483576
LT
1183}
1184
7efbff75 1185static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1186{
1187 int bytes = strlen(buffer) + 1;
d565b341 1188 unsigned char *buf = xmalloc(1+size);
93821bd9 1189 unsigned long n;
7efbff75 1190 int status = Z_OK;
5180cacc 1191
93821bd9
LT
1192 n = stream->total_out - bytes;
1193 if (n > size)
1194 n = size;
1195 memcpy(buf, (char *) buffer + bytes, n);
1196 bytes = n;
456cdf6e
LT
1197 if (bytes <= size) {
1198 /*
1199 * The above condition must be (bytes <= size), not
1200 * (bytes < size). In other words, even though we
1201 * expect no more output and set avail_out to zer0,
1202 * the input zlib stream may have bytes that express
1203 * "this concludes the stream", and we *do* want to
1204 * eat that input.
1205 *
1206 * Otherwise we would not be able to test that we
1207 * consumed all the input to reach the expected size;
1208 * we also want to check that zlib tells us that all
1209 * went well with status == Z_STREAM_END at the end.
1210 */
5180cacc
LT
1211 stream->next_out = buf + bytes;
1212 stream->avail_out = size - bytes;
7efbff75
JH
1213 while (status == Z_OK)
1214 status = inflate(stream, Z_FINISH);
5180cacc
LT
1215 }
1216 buf[size] = 0;
456cdf6e 1217 if (status == Z_STREAM_END && !stream->avail_in) {
7efbff75
JH
1218 inflateEnd(stream);
1219 return buf;
1220 }
1221
1222 if (status < 0)
1223 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1224 else if (stream->avail_in)
1225 error("garbage at end of loose object '%s'",
1226 sha1_to_hex(sha1));
1227 free(buf);
1228 return NULL;
5180cacc
LT
1229}
1230
1231/*
1232 * We used to just use "sscanf()", but that's actually way
1233 * too permissive for what we want to check. So do an anal
1234 * object header parse by hand.
1235 */
21666f1a 1236static int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1237{
21666f1a 1238 char type[10];
5180cacc
LT
1239 int i;
1240 unsigned long size;
1241
1242 /*
a6080a0a 1243 * The type can be at most ten bytes (including the
5180cacc 1244 * terminating '\0' that we add), and is followed by
21666f1a 1245 * a space.
5180cacc 1246 */
21666f1a 1247 i = 0;
5180cacc
LT
1248 for (;;) {
1249 char c = *hdr++;
1250 if (c == ' ')
1251 break;
21666f1a
NP
1252 type[i++] = c;
1253 if (i >= sizeof(type))
5180cacc 1254 return -1;
5180cacc 1255 }
21666f1a 1256 type[i] = 0;
5180cacc
LT
1257
1258 /*
1259 * The length must follow immediately, and be in canonical
1260 * decimal format (ie "010" is not valid).
1261 */
1262 size = *hdr++ - '0';
1263 if (size > 9)
1264 return -1;
1265 if (size) {
1266 for (;;) {
1267 unsigned long c = *hdr - '0';
1268 if (c > 9)
1269 break;
1270 hdr++;
1271 size = size * 10 + c;
1272 }
1273 }
1274 *sizep = size;
1275
1276 /*
1277 * The length must be followed by a zero byte
1278 */
21666f1a 1279 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1280}
1281
7efbff75 1282static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1283{
5180cacc 1284 int ret;
0fcfd160 1285 z_stream stream;
5180cacc 1286 char hdr[8192];
0fcfd160 1287
5180cacc 1288 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1289 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1290 return NULL;
1291
7efbff75 1292 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1293}
1294
54dab52a
NP
1295unsigned long get_size_from_delta(struct packed_git *p,
1296 struct pack_window **w_curs,
1297 off_t curpos)
1298{
1299 const unsigned char *data;
1300 unsigned char delta_head[20], *in;
1301 z_stream stream;
1302 int st;
1303
1304 memset(&stream, 0, sizeof(stream));
1305 stream.next_out = delta_head;
1306 stream.avail_out = sizeof(delta_head);
1307
1308 inflateInit(&stream);
1309 do {
1310 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1311 stream.next_in = in;
1312 st = inflate(&stream, Z_FINISH);
1313 curpos += stream.next_in - in;
1314 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1315 stream.total_out < sizeof(delta_head));
1316 inflateEnd(&stream);
3d77d877
NP
1317 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1318 error("delta data unpack-initial failed");
1319 return 0;
1320 }
54dab52a
NP
1321
1322 /* Examine the initial part of the delta to figure out
1323 * the result size.
1324 */
1325 data = delta_head;
1326
1327 /* ignore base size */
1328 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1329
1330 /* Read the result size */
1331 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1332}
1333
c4001d92 1334static off_t get_delta_base(struct packed_git *p,
03e79c88 1335 struct pack_window **w_curs,
c4001d92 1336 off_t *curpos,
21666f1a 1337 enum object_type type,
c4001d92 1338 off_t delta_obj_offset)
eb32d236 1339{
2b87c45b 1340 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1341 off_t base_offset;
eb32d236 1342
8d8a4ea5
SP
1343 /* use_pack() assured us we have [base_info, base_info + 20)
1344 * as a range that we can look at without walking off the
1345 * end of the mapped window. Its actually the hash size
1346 * that is assured. An OFS_DELTA longer than the hash size
1347 * is stupid, as then a REF_DELTA would be smaller to store.
1348 */
21666f1a 1349 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1350 unsigned used = 0;
1351 unsigned char c = base_info[used++];
1352 base_offset = c & 127;
1353 while (c & 128) {
1354 base_offset += 1;
8723f216 1355 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1356 return 0; /* overflow */
eb32d236
NP
1357 c = base_info[used++];
1358 base_offset = (base_offset << 7) + (c & 127);
1359 }
1360 base_offset = delta_obj_offset - base_offset;
d8f32556 1361 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1362 return 0; /* out of bound */
2b87c45b 1363 *curpos += used;
21666f1a 1364 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1365 /* The base entry _must_ be in the same pack */
1366 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1367 *curpos += 20;
eb32d236
NP
1368 } else
1369 die("I am totally screwed");
2b87c45b 1370 return base_offset;
eb32d236
NP
1371}
1372
f3bf9224 1373/* forward declaration for a mutually recursive function */
c4001d92 1374static int packed_object_info(struct packed_git *p, off_t offset,
21666f1a 1375 unsigned long *sizep);
f3bf9224 1376
43057304 1377static int packed_delta_info(struct packed_git *p,
03e79c88 1378 struct pack_window **w_curs,
c4001d92 1379 off_t curpos,
21666f1a 1380 enum object_type type,
c4001d92 1381 off_t obj_offset,
43057304 1382 unsigned long *sizep)
5db47c2b 1383{
c4001d92 1384 off_t base_offset;
f3bf9224 1385
21666f1a 1386 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
3d77d877
NP
1387 if (!base_offset)
1388 return OBJ_BAD;
21666f1a 1389 type = packed_object_info(p, base_offset, NULL);
3d77d877
NP
1390 if (type <= OBJ_NONE) {
1391 struct revindex_entry *revidx = find_pack_revindex(p, base_offset);
1392 const unsigned char *base_sha1 =
1393 nth_packed_object_sha1(p, revidx->nr);
1394 mark_bad_packed_object(p, base_sha1);
1395 type = sha1_object_info(base_sha1, NULL);
1396 if (type <= OBJ_NONE)
1397 return OBJ_BAD;
1398 }
f3bf9224 1399
c62266f3
JH
1400 /* We choose to only get the type of the base object and
1401 * ignore potentially corrupt pack file that expects the delta
1402 * based on a base with a wrong size. This saves tons of
1403 * inflate() calls.
1404 */
3d77d877 1405 if (sizep) {
54dab52a 1406 *sizep = get_size_from_delta(p, w_curs, curpos);
3d77d877
NP
1407 if (*sizep == 0)
1408 type = OBJ_BAD;
1409 }
21666f1a
NP
1410
1411 return type;
5db47c2b
JH
1412}
1413
2b87c45b
NP
1414static int unpack_object_header(struct packed_git *p,
1415 struct pack_window **w_curs,
c4001d92 1416 off_t *curpos,
2b87c45b 1417 unsigned long *sizep)
a733cb60 1418{
03e79c88
SP
1419 unsigned char *base;
1420 unsigned int left;
ad1ed5ee 1421 unsigned long used;
2b87c45b 1422 enum object_type type;
a733cb60 1423
8d8a4ea5
SP
1424 /* use_pack() assures us we have [base, base + 20) available
1425 * as a range that we can look at at. (Its actually the hash
3dff5379 1426 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1427 * the maximum deflated object size is 2^137, which is just
1428 * insane, so we know won't exceed what we have been given.
1429 */
2b87c45b 1430 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1431 used = unpack_object_header_buffer(base, left, &type, sizep);
1432 if (!used) {
1433 type = OBJ_BAD;
1434 } else
1435 *curpos += used;
ad1ed5ee 1436
2b87c45b 1437 return type;
a733cb60
LT
1438}
1439
21666f1a 1440const char *packed_object_info_detail(struct packed_git *p,
c4001d92 1441 off_t obj_offset,
21666f1a
NP
1442 unsigned long *size,
1443 unsigned long *store_size,
1444 unsigned int *delta_chain_length,
1445 unsigned char *base_sha1)
ad8c80a5 1446{
03e79c88 1447 struct pack_window *w_curs = NULL;
c4001d92
SP
1448 off_t curpos;
1449 unsigned long dummy;
43057304 1450 unsigned char *next_sha1;
21666f1a 1451 enum object_type type;
70f5d5d3 1452 struct revindex_entry *revidx;
ad8c80a5 1453
43057304 1454 *delta_chain_length = 0;
2b87c45b 1455 curpos = obj_offset;
21666f1a 1456 type = unpack_object_header(p, &w_curs, &curpos, size);
43057304 1457
70f5d5d3
NP
1458 revidx = find_pack_revindex(p, obj_offset);
1459 *store_size = revidx[1].offset - obj_offset;
1460
43057304 1461 for (;;) {
21666f1a 1462 switch (type) {
43057304 1463 default:
08a19d87 1464 die("pack %s contains unknown object type %d",
21666f1a 1465 p->pack_name, type);
43057304
NP
1466 case OBJ_COMMIT:
1467 case OBJ_TREE:
1468 case OBJ_BLOB:
1469 case OBJ_TAG:
03e79c88 1470 unuse_pack(&w_curs);
21666f1a 1471 return typename(type);
eb32d236 1472 case OBJ_OFS_DELTA:
21666f1a 1473 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
8eca0b47
NP
1474 if (!obj_offset)
1475 die("pack %s contains bad delta base reference of type %s",
1476 p->pack_name, typename(type));
eb32d236 1477 if (*delta_chain_length == 0) {
70f5d5d3
NP
1478 revidx = find_pack_revindex(p, obj_offset);
1479 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
eb32d236
NP
1480 }
1481 break;
1482 case OBJ_REF_DELTA:
2b87c45b 1483 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
43057304
NP
1484 if (*delta_chain_length == 0)
1485 hashcpy(base_sha1, next_sha1);
2b87c45b 1486 obj_offset = find_pack_entry_one(next_sha1, p);
43057304
NP
1487 break;
1488 }
43057304 1489 (*delta_chain_length)++;
2b87c45b 1490 curpos = obj_offset;
21666f1a 1491 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
ad8c80a5 1492 }
ad8c80a5
JH
1493}
1494
c4001d92 1495static int packed_object_info(struct packed_git *p, off_t obj_offset,
21666f1a 1496 unsigned long *sizep)
1f688557 1497{
03e79c88 1498 struct pack_window *w_curs = NULL;
c4001d92
SP
1499 unsigned long size;
1500 off_t curpos = obj_offset;
21666f1a 1501 enum object_type type;
5db47c2b 1502
21666f1a 1503 type = unpack_object_header(p, &w_curs, &curpos, &size);
5db47c2b 1504
21666f1a 1505 switch (type) {
eb32d236
NP
1506 case OBJ_OFS_DELTA:
1507 case OBJ_REF_DELTA:
21666f1a
NP
1508 type = packed_delta_info(p, &w_curs, curpos,
1509 type, obj_offset, sizep);
2b87c45b 1510 break;
a733cb60 1511 case OBJ_COMMIT:
a733cb60 1512 case OBJ_TREE:
a733cb60 1513 case OBJ_BLOB:
a733cb60 1514 case OBJ_TAG:
2b87c45b
NP
1515 if (sizep)
1516 *sizep = size;
a69d0943 1517 break;
1f688557 1518 default:
3d77d877
NP
1519 error("unknown object type %i at offset %"PRIuMAX" in %s",
1520 type, (uintmax_t)obj_offset, p->pack_name);
1521 type = OBJ_BAD;
1f688557 1522 }
2b87c45b 1523 unuse_pack(&w_curs);
21666f1a 1524 return type;
1f688557
JH
1525}
1526
eb950c19 1527static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1528 struct pack_window **w_curs,
c4001d92 1529 off_t curpos,
eb950c19 1530 unsigned long size)
de530aaa
SP
1531{
1532 int st;
1533 z_stream stream;
079afb18 1534 unsigned char *buffer, *in;
de530aaa
SP
1535
1536 buffer = xmalloc(size + 1);
1537 buffer[size] = 0;
1538 memset(&stream, 0, sizeof(stream));
de530aaa
SP
1539 stream.next_out = buffer;
1540 stream.avail_out = size;
1541
1542 inflateInit(&stream);
079afb18 1543 do {
2b87c45b 1544 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18
SP
1545 stream.next_in = in;
1546 st = inflate(&stream, Z_FINISH);
2b87c45b 1547 curpos += stream.next_in - in;
079afb18 1548 } while (st == Z_OK || st == Z_BUF_ERROR);
de530aaa
SP
1549 inflateEnd(&stream);
1550 if ((st != Z_STREAM_END) || stream.total_out != size) {
1551 free(buffer);
1552 return NULL;
1553 }
1554
1555 return buffer;
1556}
1557
e5e01619
LT
1558#define MAX_DELTA_CACHE (256)
1559
18bdec11 1560static size_t delta_base_cached;
5e08ecbf
NP
1561
1562static struct delta_base_cache_lru_list {
1563 struct delta_base_cache_lru_list *prev;
1564 struct delta_base_cache_lru_list *next;
1565} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1566
e5e01619 1567static struct delta_base_cache_entry {
5e08ecbf
NP
1568 struct delta_base_cache_lru_list lru;
1569 void *data;
e5e01619
LT
1570 struct packed_git *p;
1571 off_t base_offset;
1572 unsigned long size;
e5e01619
LT
1573 enum object_type type;
1574} delta_base_cache[MAX_DELTA_CACHE];
1575
1576static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1577{
1578 unsigned long hash;
1579
1580 hash = (unsigned long)p + (unsigned long)base_offset;
1581 hash += (hash >> 8) + (hash >> 16);
3358004a 1582 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1583}
1584
62f255ad 1585static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 1586 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 1587{
e5e01619
LT
1588 void *ret;
1589 unsigned long hash = pack_entry_hash(p, base_offset);
1590 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1591
1592 ret = ent->data;
749bc58c
MV
1593 if (!ret || ent->p != p || ent->base_offset != base_offset)
1594 return unpack_entry(p, base_offset, type, base_size);
e5e01619 1595
18bdec11 1596 if (!keep_cache) {
a0cba108 1597 ent->data = NULL;
5e08ecbf
NP
1598 ent->lru.next->prev = ent->lru.prev;
1599 ent->lru.prev->next = ent->lru.next;
18bdec11 1600 delta_base_cached -= ent->size;
182af834
PH
1601 } else {
1602 ret = xmemdupz(ent->data, ent->size);
a0cba108 1603 }
e5e01619
LT
1604 *type = ent->type;
1605 *base_size = ent->size;
1606 return ret;
62f255ad
LT
1607}
1608
18bdec11
SP
1609static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1610{
1611 if (ent->data) {
1612 free(ent->data);
1613 ent->data = NULL;
5e08ecbf
NP
1614 ent->lru.next->prev = ent->lru.prev;
1615 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
1616 delta_base_cached -= ent->size;
1617 }
1618}
1619
62f255ad
LT
1620static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1621 void *base, unsigned long base_size, enum object_type type)
1622{
5e08ecbf 1623 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 1624 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 1625 struct delta_base_cache_lru_list *lru;
e5e01619 1626
18bdec11
SP
1627 release_delta_base_cache(ent);
1628 delta_base_cached += base_size;
5e08ecbf
NP
1629
1630 for (lru = delta_base_cache_lru.next;
1631 delta_base_cached > delta_base_cache_limit
1632 && lru != &delta_base_cache_lru;
1633 lru = lru->next) {
1634 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
1635 if (f->type == OBJ_BLOB)
1636 release_delta_base_cache(f);
1637 }
5e08ecbf
NP
1638 for (lru = delta_base_cache_lru.next;
1639 delta_base_cached > delta_base_cache_limit
1640 && lru != &delta_base_cache_lru;
1641 lru = lru->next) {
1642 struct delta_base_cache_entry *f = (void *)lru;
1643 release_delta_base_cache(f);
1644 }
18bdec11 1645
e5e01619
LT
1646 ent->p = p;
1647 ent->base_offset = base_offset;
1648 ent->type = type;
1649 ent->data = base;
1650 ent->size = base_size;
5e08ecbf
NP
1651 ent->lru.next = &delta_base_cache_lru;
1652 ent->lru.prev = delta_base_cache_lru.prev;
1653 delta_base_cache_lru.prev->next = &ent->lru;
1654 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
1655}
1656
eb950c19 1657static void *unpack_delta_entry(struct packed_git *p,
03e79c88 1658 struct pack_window **w_curs,
c4001d92 1659 off_t curpos,
1f688557 1660 unsigned long delta_size,
c4001d92 1661 off_t obj_offset,
21666f1a 1662 enum object_type *type,
eb950c19 1663 unsigned long *sizep)
1f688557 1664{
7c3e8be3 1665 void *delta_data, *result, *base;
c4001d92
SP
1666 unsigned long base_size;
1667 off_t base_offset;
43057304 1668
21666f1a 1669 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
8eca0b47
NP
1670 if (!base_offset) {
1671 error("failed to validate delta base reference "
1672 "at offset %"PRIuMAX" from %s",
1673 (uintmax_t)curpos, p->pack_name);
1674 return NULL;
1675 }
eac12e2d 1676 unuse_pack(w_curs);
a0cba108 1677 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
8eca0b47
NP
1678 if (!base) {
1679 /*
1680 * We're probably in deep shit, but let's try to fetch
1681 * the required base anyway from another pack or loose.
1682 * This is costly but should happen only in the presence
1683 * of a corrupted pack, and is better than failing outright.
1684 */
1685 struct revindex_entry *revidx = find_pack_revindex(p, base_offset);
1686 const unsigned char *base_sha1 =
1687 nth_packed_object_sha1(p, revidx->nr);
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);
1734 return NULL;
1735 }
1736 }
1737
21666f1a
NP
1738 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1739 switch (*type) {
eb32d236
NP
1740 case OBJ_OFS_DELTA:
1741 case OBJ_REF_DELTA:
21666f1a
NP
1742 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1743 obj_offset, type, sizep);
4d703a1a 1744 break;
a733cb60 1745 case OBJ_COMMIT:
a733cb60 1746 case OBJ_TREE:
a733cb60 1747 case OBJ_BLOB:
a733cb60 1748 case OBJ_TAG:
21666f1a 1749 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
4d703a1a 1750 break;
1f688557 1751 default:
8eca0b47
NP
1752 data = NULL;
1753 error("unknown object type %i at offset %"PRIuMAX" in %s",
1754 *type, (uintmax_t)obj_offset, p->pack_name);
1f688557 1755 }
03e79c88 1756 unuse_pack(&w_curs);
21666f1a 1757 return data;
1f688557
JH
1758}
1759
d079837e 1760const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 1761 uint32_t n)
9a217f2a 1762{
42873078 1763 const unsigned char *index = p->index_data;
d079837e
SP
1764 if (!index) {
1765 if (open_pack_index(p))
1766 return NULL;
1767 index = p->index_data;
1768 }
57059091 1769 if (n >= p->num_objects)
d72308e0 1770 return NULL;
74e34e1f
NP
1771 index += 4 * 256;
1772 if (p->index_version == 1) {
1773 return index + 24 * n + 4;
1774 } else {
1775 index += 8;
1776 return index + 20 * n;
1777 }
1778}
1779
99093238 1780off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
1781{
1782 const unsigned char *index = p->index_data;
1783 index += 4 * 256;
1784 if (p->index_version == 1) {
1785 return ntohl(*((uint32_t *)(index + 24 * n)));
1786 } else {
1787 uint32_t off;
1788 index += 8 + p->num_objects * (20 + 4);
1789 off = ntohl(*((uint32_t *)(index + 4 * n)));
1790 if (!(off & 0x80000000))
1791 return off;
1792 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1793 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1794 ntohl(*((uint32_t *)(index + 4)));
1795 }
9a217f2a
JH
1796}
1797
c4001d92 1798off_t find_pack_entry_one(const unsigned char *sha1,
43057304 1799 struct packed_git *p)
1f688557 1800{
42873078 1801 const uint32_t *level1_ofs = p->index_data;
42873078 1802 const unsigned char *index = p->index_data;
628522ec
JH
1803 unsigned hi, lo, stride;
1804 static int use_lookup = -1;
1805 static int debug_lookup = -1;
1806
1807 if (debug_lookup < 0)
1808 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 1809
d079837e
SP
1810 if (!index) {
1811 if (open_pack_index(p))
1812 return 0;
1813 level1_ofs = p->index_data;
1814 index = p->index_data;
1815 }
74e34e1f
NP
1816 if (p->index_version > 1) {
1817 level1_ofs += 2;
1818 index += 8;
1819 }
42873078 1820 index += 4 * 256;
74e34e1f
NP
1821 hi = ntohl(level1_ofs[*sha1]);
1822 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
1823 if (p->index_version > 1) {
1824 stride = 20;
1825 } else {
1826 stride = 24;
1827 index += 4;
1828 }
1829
1830 if (debug_lookup)
6e1c2344 1831 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
1832 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1833
1834 if (use_lookup < 0)
1835 use_lookup = !!getenv("GIT_USE_LOOKUP");
1836 if (use_lookup) {
1837 int pos = sha1_entry_pos(index, stride, 0,
1838 lo, hi, p->num_objects, sha1);
1839 if (pos < 0)
1840 return 0;
1841 return nth_packed_object_offset(p, pos);
1842 }
1f688557
JH
1843
1844 do {
74e34e1f 1845 unsigned mi = (lo + hi) / 2;
628522ec
JH
1846 int cmp = hashcmp(index + mi * stride, sha1);
1847
1848 if (debug_lookup)
1849 printf("lo %u hi %u rg %u mi %u\n",
1850 lo, hi, hi - lo, mi);
43057304 1851 if (!cmp)
74e34e1f 1852 return nth_packed_object_offset(p, mi);
1f688557
JH
1853 if (cmp > 0)
1854 hi = mi;
1855 else
1856 lo = mi+1;
1857 } while (lo < hi);
1858 return 0;
1859}
1860
000dfd3f 1861int matches_pack_name(struct packed_git *p, const char *name)
d4ff6d92
SP
1862{
1863 const char *last_c, *c;
1864
000dfd3f
JH
1865 if (!strcmp(p->pack_name, name))
1866 return 1;
d4ff6d92
SP
1867
1868 for (c = p->pack_name, last_c = c; *c;)
1869 if (*c == '/')
1870 last_c = ++c;
1871 else
1872 ++c;
000dfd3f
JH
1873 if (!strcmp(last_c, name))
1874 return 1;
d4ff6d92 1875
000dfd3f 1876 return 0;
d4ff6d92
SP
1877}
1878
106d710b 1879static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1f688557 1880{
f7c22cc6 1881 static struct packed_git *last_found = (void *)1;
1f688557 1882 struct packed_git *p;
c4001d92 1883 off_t offset;
43057304 1884
1f688557 1885 prepare_packed_git();
f7c22cc6
NP
1886 if (!packed_git)
1887 return 0;
1888 p = (last_found == (void *)1) ? packed_git : last_found;
1f688557 1889
f7c22cc6 1890 do {
106d710b
JH
1891 if (ignore_packed) {
1892 const char **ig;
1893 for (ig = ignore_packed; *ig; ig++)
000dfd3f 1894 if (matches_pack_name(p, *ig))
106d710b
JH
1895 break;
1896 if (*ig)
f7c22cc6 1897 goto next;
106d710b 1898 }
f7c22cc6 1899
8eca0b47
NP
1900 if (p->num_bad_objects) {
1901 unsigned i;
1902 for (i = 0; i < p->num_bad_objects; i++)
1903 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1904 goto next;
1905 }
1906
43057304
NP
1907 offset = find_pack_entry_one(sha1, p);
1908 if (offset) {
c715f783
SP
1909 /*
1910 * We are about to tell the caller where they can
1911 * locate the requested object. We better make
1912 * sure the packfile is still here and can be
1913 * accessed before supplying that answer, as
1914 * it may have been deleted since the index
1915 * was loaded!
1916 */
1917 if (p->pack_fd == -1 && open_packed_git(p)) {
1918 error("packfile %s cannot be accessed", p->pack_name);
f7c22cc6 1919 goto next;
c715f783 1920 }
43057304
NP
1921 e->offset = offset;
1922 e->p = p;
1923 hashcpy(e->sha1, sha1);
f7c22cc6 1924 last_found = p;
1f688557 1925 return 1;
43057304 1926 }
f7c22cc6
NP
1927
1928 next:
1929 if (p == last_found)
1930 p = packed_git;
1931 else
1932 p = p->next;
1933 if (p == last_found)
1934 p = p->next;
1935 } while (p);
1f688557
JH
1936 return 0;
1937}
1938
a6080a0a 1939struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
1940 struct packed_git *packs)
1941{
1942 struct packed_git *p;
bf592c50
DB
1943
1944 for (p = packs; p; p = p->next) {
43057304 1945 if (find_pack_entry_one(sha1, p))
bf592c50
DB
1946 return p;
1947 }
1948 return NULL;
21666f1a 1949
bf592c50
DB
1950}
1951
21666f1a 1952static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
65c2e0c3 1953{
36e4d74a 1954 int status;
65c2e0c3
JH
1955 unsigned long mapsize, size;
1956 void *map;
1957 z_stream stream;
d65a16f6 1958 char hdr[32];
65c2e0c3 1959
bb6b8e4f 1960 map = map_sha1_file(sha1, &mapsize);
f0df4ed5
JS
1961 if (!map)
1962 return error("unable to find %s", sha1_to_hex(sha1));
36e4d74a
JH
1963 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1964 status = error("unable to unpack %s header",
1965 sha1_to_hex(sha1));
21666f1a 1966 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 1967 status = error("unable to parse %s header", sha1_to_hex(sha1));
21666f1a
NP
1968 else if (sizep)
1969 *sizep = size;
65c2e0c3
JH
1970 inflateEnd(&stream);
1971 munmap(map, mapsize);
1972 return status;
1973}
1974
21666f1a 1975int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
f0df4ed5 1976{
f0df4ed5 1977 struct pack_entry e;
ddd63e64 1978 int status;
f0df4ed5
JS
1979
1980 if (!find_pack_entry(sha1, &e, NULL)) {
ddd63e64
SG
1981 /* Most likely it's a loose object. */
1982 status = sha1_loose_object_info(sha1, sizep);
1983 if (status >= 0)
1984 return status;
1985
1986 /* Not a loose object; someone else may have just packed it. */
f0df4ed5
JS
1987 reprepare_packed_git();
1988 if (!find_pack_entry(sha1, &e, NULL))
ddd63e64 1989 return status;
f0df4ed5 1990 }
3d77d877
NP
1991
1992 status = packed_object_info(e.p, e.offset, sizep);
1993 if (status < 0) {
1994 mark_bad_packed_object(e.p, sha1);
1995 status = sha1_object_info(sha1, sizep);
1996 }
1997
1998 return status;
f0df4ed5
JS
1999}
2000
21666f1a
NP
2001static void *read_packed_sha1(const unsigned char *sha1,
2002 enum object_type *type, unsigned long *size)
1f688557
JH
2003{
2004 struct pack_entry e;
8eca0b47 2005 void *data;
1f688557 2006
8276c007 2007 if (!find_pack_entry(sha1, &e, NULL))
1f688557 2008 return NULL;
8eca0b47
NP
2009 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2010 if (!data) {
2011 /*
2012 * We're probably in deep shit, but let's try to fetch
2013 * the required object anyway from another pack or loose.
2014 * This should happen only in the presence of a corrupted
2015 * pack, and is better than failing outright.
2016 */
2017 error("failed to read object %s at offset %"PRIuMAX" from %s",
2018 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2019 mark_bad_packed_object(e.p, sha1);
ac939109 2020 data = read_object(sha1, type, size);
8eca0b47
NP
2021 }
2022 return data;
1f688557
JH
2023}
2024
d66b37bb
JH
2025/*
2026 * This is meant to hold a *small* number of objects that you would
2027 * want read_sha1_file() to be able to return, but yet you do not want
2028 * to write them into the object store (e.g. a browse-only
2029 * application).
2030 */
2031static struct cached_object {
2032 unsigned char sha1[20];
21666f1a 2033 enum object_type type;
d66b37bb
JH
2034 void *buf;
2035 unsigned long size;
2036} *cached_objects;
2037static int cached_object_nr, cached_object_alloc;
2038
346245a1
JK
2039static struct cached_object empty_tree = {
2040 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
2041 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
2042 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
2043 OBJ_TREE,
2044 "",
2045 0
2046};
2047
d66b37bb
JH
2048static struct cached_object *find_cached_object(const unsigned char *sha1)
2049{
2050 int i;
2051 struct cached_object *co = cached_objects;
2052
2053 for (i = 0; i < cached_object_nr; i++, co++) {
2054 if (!hashcmp(co->sha1, sha1))
2055 return co;
2056 }
346245a1
JK
2057 if (!hashcmp(sha1, empty_tree.sha1))
2058 return &empty_tree;
d66b37bb
JH
2059 return NULL;
2060}
2061
21666f1a
NP
2062int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2063 unsigned char *sha1)
d66b37bb
JH
2064{
2065 struct cached_object *co;
2066
21666f1a 2067 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2068 if (has_sha1_file(sha1) || find_cached_object(sha1))
2069 return 0;
2070 if (cached_object_alloc <= cached_object_nr) {
2071 cached_object_alloc = alloc_nr(cached_object_alloc);
2072 cached_objects = xrealloc(cached_objects,
2073 sizeof(*cached_objects) *
2074 cached_object_alloc);
2075 }
2076 co = &cached_objects[cached_object_nr++];
2077 co->size = len;
21666f1a 2078 co->type = type;
efa13f7b
JH
2079 co->buf = xmalloc(len);
2080 memcpy(co->buf, buf, len);
d66b37bb
JH
2081 hashcpy(co->sha1, sha1);
2082 return 0;
2083}
2084
ac939109
NP
2085void *read_object(const unsigned char *sha1, enum object_type *type,
2086 unsigned long *size)
0fcfd160
LT
2087{
2088 unsigned long mapsize;
2089 void *map, *buf;
d66b37bb
JH
2090 struct cached_object *co;
2091
2092 co = find_cached_object(sha1);
2093 if (co) {
21666f1a 2094 *type = co->type;
d66b37bb 2095 *size = co->size;
182af834 2096 return xmemdupz(co->buf, co->size);
d66b37bb 2097 }
0fcfd160 2098
8276c007
PE
2099 buf = read_packed_sha1(sha1, type, size);
2100 if (buf)
2101 return buf;
bb6b8e4f 2102 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2103 if (map) {
7efbff75 2104 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2105 munmap(map, mapsize);
2106 return buf;
2107 }
637cdd9d 2108 reprepare_packed_git();
8276c007 2109 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2110}
2111
ac939109
NP
2112void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
2113 unsigned long *size)
2114{
2115 void *data = read_object(sha1, type, size);
2116 /* legacy behavior is to die on corrupted objects */
2117 if (!data && (has_loose_object(sha1) || has_packed_and_bad(sha1)))
2118 die("object %s is corrupted", sha1_to_hex(sha1));
2119 return data;
2120}
2121
40469ee9 2122void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2123 const char *required_type_name,
40469ee9
JH
2124 unsigned long *size,
2125 unsigned char *actual_sha1_return)
f4913f91 2126{
21666f1a 2127 enum object_type type, required_type;
f4913f91
JH
2128 void *buffer;
2129 unsigned long isize;
40469ee9 2130 unsigned char actual_sha1[20];
f4913f91 2131
21666f1a 2132 required_type = type_from_string(required_type_name);
e702496e 2133 hashcpy(actual_sha1, sha1);
40469ee9
JH
2134 while (1) {
2135 int ref_length = -1;
2136 const char *ref_type = NULL;
f4913f91 2137
21666f1a 2138 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2139 if (!buffer)
2140 return NULL;
21666f1a 2141 if (type == required_type) {
40469ee9
JH
2142 *size = isize;
2143 if (actual_sha1_return)
e702496e 2144 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2145 return buffer;
2146 }
2147 /* Handle references */
21666f1a 2148 else if (type == OBJ_COMMIT)
40469ee9 2149 ref_type = "tree ";
21666f1a 2150 else if (type == OBJ_TAG)
40469ee9
JH
2151 ref_type = "object ";
2152 else {
2153 free(buffer);
2154 return NULL;
2155 }
2156 ref_length = strlen(ref_type);
f4913f91 2157
50974ec9
MK
2158 if (ref_length + 40 > isize ||
2159 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2160 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2161 free(buffer);
2162 return NULL;
2163 }
1cf58e72 2164 free(buffer);
40469ee9
JH
2165 /* Now we have the ID of the referred-to object in
2166 * actual_sha1. Check again. */
f4913f91 2167 }
f4913f91
JH
2168}
2169
ce9fbf16 2170static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2171 const char *type, unsigned char *sha1,
d65a16f6 2172 char *hdr, int *hdrlen)
d410c0f5 2173{
9126f009 2174 git_SHA_CTX c;
d410c0f5
JH
2175
2176 /* Generate the header */
d65a16f6 2177 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
2178
2179 /* Sha1.. */
9126f009
NP
2180 git_SHA1_Init(&c);
2181 git_SHA1_Update(&c, hdr, *hdrlen);
2182 git_SHA1_Update(&c, buf, len);
2183 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2184}
2185
230f1322
LT
2186/*
2187 * Move the just written object into its final resting place
2188 */
839837b9 2189int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322 2190{
e32c0a9c
TR
2191 int ret = 0;
2192 if (link(tmpfile, filename))
2193 ret = errno;
7ebb6fca
LT
2194
2195 /*
2196 * Coda hack - coda doesn't like cross-directory links,
2197 * so we fall back to a rename, which will mean that it
2198 * won't be able to check collisions, but that's not a
2199 * big deal.
2200 *
2201 * The same holds for FAT formatted media.
2202 *
2203 * When this succeeds, we just return 0. We have nothing
2204 * left to unlink.
2205 */
2206 if (ret && ret != EEXIST) {
2207 if (!rename(tmpfile, filename))
230f1322 2208 return 0;
9e48b389 2209 ret = errno;
230f1322
LT
2210 }
2211 unlink(tmpfile);
2212 if (ret) {
2213 if (ret != EEXIST) {
916d081b 2214 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
230f1322
LT
2215 }
2216 /* FIXME!!! Collision check here ? */
2217 }
2218
2219 return 0;
2220}
2221
4d548150
LT
2222static int write_buffer(int fd, const void *buf, size_t len)
2223{
d34cf19b 2224 if (write_in_full(fd, buf, len) < 0)
93822c22 2225 return error("file write error (%s)", strerror(errno));
4d548150
LT
2226 return 0;
2227}
2228
ce9fbf16 2229int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2230 unsigned char *sha1)
2231{
d65a16f6 2232 char hdr[32];
abdc3fc8
RS
2233 int hdrlen;
2234 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2235 return 0;
2236}
2237
e9039dd3
LT
2238/* Finalize a file on disk, and close it. */
2239static void close_sha1_file(int fd)
2240{
aafe9fba
LT
2241 if (fsync_object_files)
2242 fsync_or_die(fd, "sha1 file");
e9039dd3
LT
2243 fchmod(fd, 0444);
2244 if (close(fd) != 0)
2245 die("unable to write sha1 file");
2246}
2247
5723fe7e
LT
2248/* Size of directory component, including the ending '/' */
2249static inline int directory_size(const char *filename)
2250{
2251 const char *s = strrchr(filename, '/');
2252 if (!s)
2253 return 0;
2254 return s - filename + 1;
2255}
2256
2257/*
2258 * This creates a temporary file in the same directory as the final
2259 * 'filename'
2260 *
2261 * We want to avoid cross-directory filename renames, because those
2262 * can have problems on various filesystems (FAT, NFS, Coda).
2263 */
2264static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2265{
2266 int fd, dirlen = directory_size(filename);
2267
2268 if (dirlen + 20 > bufsiz) {
2269 errno = ENAMETOOLONG;
2270 return -1;
2271 }
2272 memcpy(buffer, filename, dirlen);
2273 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2274 fd = mkstemp(buffer);
2275 if (fd < 0 && dirlen) {
2276 /* Make sure the directory exists */
6ff6af62 2277 memcpy(buffer, filename, dirlen);
5723fe7e 2278 buffer[dirlen-1] = 0;
3bfaf018 2279 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
5723fe7e
LT
2280 return -1;
2281
2282 /* Try again */
2283 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2284 fd = mkstemp(buffer);
2285 }
2286 return fd;
2287}
2288
bbac7311
NP
2289static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2290 void *buf, unsigned long len, time_t mtime)
0fcfd160 2291{
bbac7311 2292 int fd, size, ret;
bf0f910d 2293 unsigned char *compressed;
0fcfd160 2294 z_stream stream;
706bc531 2295 char *filename;
aac17941 2296 static char tmpfile[PATH_MAX];
a44c9a5e 2297
972a9155 2298 filename = sha1_file_name(sha1);
5723fe7e 2299 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
aac17941 2300 if (fd < 0) {
916d081b
PB
2301 if (errno == EPERM)
2302 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2303 else
2304 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
aac17941
LT
2305 }
2306
0fcfd160
LT
2307 /* Set it up */
2308 memset(&stream, 0, sizeof(stream));
12f6c308 2309 deflateInit(&stream, zlib_compression_level);
93821bd9 2310 size = 8 + deflateBound(&stream, len+hdrlen);
812666c8 2311 compressed = xmalloc(size);
0fcfd160
LT
2312
2313 /* Compress it */
0fcfd160
LT
2314 stream.next_out = compressed;
2315 stream.avail_out = size;
a44c9a5e
LT
2316
2317 /* First header.. */
d65a16f6 2318 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2319 stream.avail_in = hdrlen;
726f852b
NP
2320 while (deflate(&stream, 0) == Z_OK)
2321 /* nothing */;
a44c9a5e
LT
2322
2323 /* Then the data itself.. */
2324 stream.next_in = buf;
2325 stream.avail_in = len;
ac54c277
LT
2326 ret = deflate(&stream, Z_FINISH);
2327 if (ret != Z_STREAM_END)
2328 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2329
2330 ret = deflateEnd(&stream);
2331 if (ret != Z_OK)
2332 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2333
0fcfd160
LT
2334 size = stream.total_out;
2335
4d548150
LT
2336 if (write_buffer(fd, compressed, size) < 0)
2337 die("unable to write sha1 file");
e9039dd3 2338 close_sha1_file(fd);
383f85b7 2339 free(compressed);
0fcfd160 2340
bbac7311
NP
2341 if (mtime) {
2342 struct utimbuf utb;
2343 utb.actime = mtime;
2344 utb.modtime = mtime;
2345 if (utime(tmpfile, &utb) < 0)
2346 warning("failed utime() on %s: %s",
2347 tmpfile, strerror(errno));
2348 }
2349
230f1322 2350 return move_temp_to_file(tmpfile, filename);
0fcfd160 2351}
8237b185 2352
bbac7311
NP
2353int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2354{
2355 unsigned char sha1[20];
2356 char hdr[32];
2357 int hdrlen;
2358
2359 /* Normally if we have it in the pack then we do not bother writing
2360 * it out into .git/objects/??/?{38} file.
2361 */
2362 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2363 if (returnsha1)
2364 hashcpy(returnsha1, sha1);
2365 if (has_sha1_file(sha1))
2366 return 0;
2367 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2368}
2369
2370int force_object_loose(const unsigned char *sha1, time_t mtime)
2371{
bbac7311
NP
2372 void *buf;
2373 unsigned long len;
2374 enum object_type type;
2375 char hdr[32];
2376 int hdrlen;
1fb23e65 2377 int ret;
bbac7311 2378
c529d75a 2379 if (has_loose_object(sha1))
bbac7311
NP
2380 return 0;
2381 buf = read_packed_sha1(sha1, &type, &len);
2382 if (!buf)
2383 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2384 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1fb23e65
BS
2385 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2386 free(buf);
2387
2388 return ret;
bbac7311
NP
2389}
2390
bf592c50
DB
2391int has_pack_index(const unsigned char *sha1)
2392{
2393 struct stat st;
2394 if (stat(sha1_pack_index_name(sha1), &st))
2395 return 0;
2396 return 1;
2397}
2398
2399int has_pack_file(const unsigned char *sha1)
2400{
2401 struct stat st;
2402 if (stat(sha1_pack_name(sha1), &st))
2403 return 0;
2404 return 1;
2405}
2406
106d710b 2407int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
dade09c2
LT
2408{
2409 struct pack_entry e;
106d710b 2410 return find_pack_entry(sha1, &e, ignore_packed);
dade09c2
LT
2411}
2412
8237b185
DB
2413int has_sha1_file(const unsigned char *sha1)
2414{
1f688557
JH
2415 struct pack_entry e;
2416
106d710b 2417 if (find_pack_entry(sha1, &e, NULL))
1f688557 2418 return 1;
c529d75a 2419 return has_loose_object(sha1);
8237b185 2420}
74400e71 2421
43df4f86
DP
2422static int index_mem(unsigned char *sha1, void *buf, size_t size,
2423 int write_object, enum object_type type, const char *path)
e7332f96 2424{
6c510bee 2425 int ret, re_allocated = 0;
74400e71 2426
7672db20 2427 if (!type)
edaec3fb 2428 type = OBJ_BLOB;
6c510bee
LT
2429
2430 /*
2431 * Convert blobs to git internal format
2432 */
43df4f86 2433 if ((type == OBJ_BLOB) && path) {
f285a2d7 2434 struct strbuf nbuf = STRBUF_INIT;
21e5ad50
SP
2435 if (convert_to_git(path, buf, size, &nbuf,
2436 write_object ? safe_crlf : 0)) {
b315c5c0 2437 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
2438 re_allocated = 1;
2439 }
2440 }
2441
7672db20 2442 if (write_object)
edaec3fb 2443 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 2444 else
edaec3fb 2445 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 2446 if (re_allocated)
6c510bee 2447 free(buf);
43df4f86
DP
2448 return ret;
2449}
2450
2451int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2452 enum object_type type, const char *path)
2453{
2454 int ret;
2455 size_t size = xsize_t(st->st_size);
2456
2457 if (!S_ISREG(st->st_mode)) {
f285a2d7 2458 struct strbuf sbuf = STRBUF_INIT;
43df4f86
DP
2459 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2460 ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2461 type, path);
2462 else
2463 ret = -1;
2464 strbuf_release(&sbuf);
2465 } else if (size) {
2466 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2467 ret = index_mem(sha1, buf, size, write_object, type, path);
aac17941 2468 munmap(buf, size);
43df4f86
DP
2469 } else
2470 ret = index_mem(sha1, NULL, size, write_object, type, path);
2471 close(fd);
aac17941 2472 return ret;
74400e71 2473}
ec1fcc16
JH
2474
2475int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2476{
2477 int fd;
2478 char *target;
dc49cd76 2479 size_t len;
ec1fcc16
JH
2480
2481 switch (st->st_mode & S_IFMT) {
2482 case S_IFREG:
2483 fd = open(path, O_RDONLY);
2484 if (fd < 0)
2485 return error("open(\"%s\"): %s", path,
2486 strerror(errno));
53bca91a 2487 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
ec1fcc16
JH
2488 return error("%s: failed to insert into database",
2489 path);
2490 break;
2491 case S_IFLNK:
dc49cd76
SP
2492 len = xsize_t(st->st_size);
2493 target = xmalloc(len + 1);
2494 if (readlink(path, target, len + 1) != st->st_size) {
ec1fcc16
JH
2495 char *errstr = strerror(errno);
2496 free(target);
2497 return error("readlink(\"%s\"): %s", path,
2498 errstr);
2499 }
abdc3fc8 2500 if (!write_object)
dc49cd76
SP
2501 hash_sha1_file(target, len, blob_type, sha1);
2502 else if (write_sha1_file(target, len, blob_type, sha1))
ec1fcc16
JH
2503 return error("%s: failed to insert into database",
2504 path);
2505 free(target);
2506 break;
f35a6d3b
LT
2507 case S_IFDIR:
2508 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
2509 default:
2510 return error("%s: unsupported file type", path);
2511 }
2512 return 0;
2513}
a69e5429
JH
2514
2515int read_pack_header(int fd, struct pack_header *header)
2516{
c697ad14
HO
2517 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2518 /* "eof before pack header was fully read" */
2519 return PH_ERROR_EOF;
2520
a69e5429
JH
2521 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2522 /* "protocol error (pack signature mismatch detected)" */
2523 return PH_ERROR_PACK_SIGNATURE;
2524 if (!pack_version_ok(header->hdr_version))
2525 /* "protocol error (pack version unsupported)" */
2526 return PH_ERROR_PROTOCOL;
2527 return 0;
2528}