push: prepare sender to receive extended ref information from the receiver
[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;
102 *pos = 0;
67d42212
JR
103 if (!stat(path, &st)) {
104 /* path exists */
105 if (!S_ISDIR(st.st_mode)) {
b2cb9425 106 *pos = '/';
67d42212 107 return -3;
b2cb9425 108 }
457f06d6 109 }
67d42212
JR
110 else if (mkdir(path, 0777)) {
111 *pos = '/';
112 return -1;
113 }
457f06d6
JS
114 else if (adjust_shared_perm(path)) {
115 *pos = '/';
116 return -2;
117 }
b2cb9425
JH
118 *pos++ = '/';
119 }
120 return 0;
121}
723c31fe 122
8e21d63b
JK
123int safe_create_leading_directories_const(const char *path)
124{
125 /* path points to cache entries, so xstrdup before messing with it */
126 char *buf = xstrdup(path);
127 int result = safe_create_leading_directories(buf);
128 free(buf);
129 return result;
130}
131
3bfaf018 132char *sha1_to_hex(const unsigned char *sha1)
0fcfd160 133{
dcb3450f
LT
134 static int bufno;
135 static char hexbuffer[4][50];
0fcfd160 136 static const char hex[] = "0123456789abcdef";
dcb3450f 137 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
0fcfd160
LT
138 int i;
139
140 for (i = 0; i < 20; i++) {
141 unsigned int val = *sha1++;
142 *buf++ = hex[val >> 4];
143 *buf++ = hex[val & 0xf];
144 }
1e80e044
JS
145 *buf = '\0';
146
0fcfd160
LT
147 return buffer;
148}
149
ace1534d
JH
150static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
151{
152 int i;
153 for (i = 0; i < 20; i++) {
154 static char hex[] = "0123456789abcdef";
155 unsigned int val = sha1[i];
156 char *pos = pathbuf + i*2 + (i > 0);
157 *pos++ = hex[val >> 4];
158 *pos = hex[val & 0xf];
159 }
160}
161
0fcfd160
LT
162/*
163 * NOTE! This returns a statically allocated buffer, so you have to be
790296fd 164 * careful about using it. Do an "xstrdup()" if you need to save the
0fcfd160 165 * filename.
ace1534d
JH
166 *
167 * Also note that this returns the location for creating. Reading
168 * SHA1 file can happen from any alternate directory listed in the
d19938ab 169 * DB_ENVIRONMENT environment variable if it is not found in
ace1534d 170 * the primary object database.
0fcfd160
LT
171 */
172char *sha1_file_name(const unsigned char *sha1)
173{
0fcfd160
LT
174 static char *name, *base;
175
176 if (!base) {
d19938ab 177 const char *sha1_file_directory = get_object_directory();
0fcfd160 178 int len = strlen(sha1_file_directory);
812666c8 179 base = xmalloc(len + 60);
0fcfd160
LT
180 memcpy(base, sha1_file_directory, len);
181 memset(base+len, 0, 60);
182 base[len] = '/';
183 base[len+3] = '/';
184 name = base + len + 1;
185 }
ace1534d 186 fill_sha1_path(name, sha1);
0fcfd160
LT
187 return base;
188}
189
633f43e1 190static char *sha1_get_pack_name(const unsigned char *sha1,
6eec46bd 191 char **name, char **base, const char *which)
bf592c50
DB
192{
193 static const char hex[] = "0123456789abcdef";
633f43e1 194 char *buf;
bf592c50
DB
195 int i;
196
633f43e1 197 if (!*base) {
bf592c50
DB
198 const char *sha1_file_directory = get_object_directory();
199 int len = strlen(sha1_file_directory);
633f43e1 200 *base = xmalloc(len + 60);
6eec46bd
JH
201 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
202 sha1_file_directory, which);
633f43e1 203 *name = *base + len + 11;
bf592c50
DB
204 }
205
633f43e1 206 buf = *name;
bf592c50
DB
207
208 for (i = 0; i < 20; i++) {
209 unsigned int val = *sha1++;
210 *buf++ = hex[val >> 4];
211 *buf++ = hex[val & 0xf];
212 }
a6080a0a 213
633f43e1 214 return *base;
bf592c50
DB
215}
216
633f43e1 217char *sha1_pack_name(const unsigned char *sha1)
bf592c50 218{
633f43e1 219 static char *name, *base;
bf592c50 220
6eec46bd 221 return sha1_get_pack_name(sha1, &name, &base, "pack");
633f43e1 222}
bf592c50 223
633f43e1
HO
224char *sha1_pack_index_name(const unsigned char *sha1)
225{
226 static char *name, *base;
a6080a0a 227
6eec46bd 228 return sha1_get_pack_name(sha1, &name, &base, "idx");
bf592c50
DB
229}
230
d5a63b99
JH
231struct alternate_object_database *alt_odb_list;
232static struct alternate_object_database **alt_odb_tail;
ace1534d 233
c2f493a4
MW
234static void read_info_alternates(const char * alternates, int depth);
235
ddd5d056
JH
236/*
237 * Prepare alternate object database registry.
d5a63b99
JH
238 *
239 * The variable alt_odb_list points at the list of struct
240 * alternate_object_database. The elements on this list come from
241 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
242 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
243 * whose contents is similar to that environment variable but can be
244 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
245 * contains "/the/directory/corresponding/to/.git/objects/...", while
246 * its name points just after the slash at the end of ".git/objects/"
247 * in the example above, and has enough space to hold 40-byte hex
248 * SHA1, an extra slash for the first level indirection, and the
249 * terminating NUL.
ddd5d056 250 */
c2f493a4 251static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
ace1534d 252{
1494e038 253 const char *objdir = get_object_directory();
c2f493a4
MW
254 struct alternate_object_database *ent;
255 struct alternate_object_database *alt;
256 /* 43 = 40-byte + 2 '/' + terminating NUL */
257 int pfxlen = len;
258 int entlen = pfxlen + 43;
ccfd3e99 259 int base_len = -1;
d5a63b99 260
85dadc38 261 if (!is_absolute_path(entry) && relative_base) {
c2f493a4
MW
262 /* Relative alt-odb */
263 if (base_len < 0)
264 base_len = strlen(relative_base) + 1;
265 entlen += base_len;
266 pfxlen += base_len;
267 }
268 ent = xmalloc(sizeof(*ent) + entlen);
269
85dadc38 270 if (!is_absolute_path(entry) && relative_base) {
c2f493a4
MW
271 memcpy(ent->base, relative_base, base_len - 1);
272 ent->base[base_len - 1] = '/';
273 memcpy(ent->base + base_len, entry, len);
274 }
275 else
276 memcpy(ent->base, entry, pfxlen);
277
278 ent->name = ent->base + pfxlen + 1;
279 ent->base[pfxlen + 3] = '/';
280 ent->base[pfxlen] = ent->base[entlen-1] = 0;
281
282 /* Detect cases where alternate disappeared */
90b4a71c 283 if (!is_directory(ent->base)) {
c2f493a4
MW
284 error("object directory %s does not exist; "
285 "check .git/objects/info/alternates.",
286 ent->base);
287 free(ent);
288 return -1;
289 }
290
291 /* Prevent the common mistake of listing the same
292 * thing twice, or object directory itself.
293 */
294 for (alt = alt_odb_list; alt; alt = alt->next) {
295 if (!memcmp(ent->base, alt->base, pfxlen)) {
296 free(ent);
297 return -1;
298 }
299 }
300 if (!memcmp(ent->base, objdir, pfxlen)) {
301 free(ent);
302 return -1;
303 }
304
305 /* add the alternate entry */
306 *alt_odb_tail = ent;
307 alt_odb_tail = &(ent->next);
308 ent->next = NULL;
309
310 /* recursively add alternates */
311 read_info_alternates(ent->base, depth + 1);
312
313 ent->base[pfxlen] = '/';
314
315 return 0;
316}
317
318static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
319 const char *relative_base, int depth)
320{
321 const char *cp, *last;
322
323 if (depth > 5) {
324 error("%s: ignoring alternate object stores, nesting too deep.",
325 relative_base);
326 return;
327 }
328
d5a63b99 329 last = alt;
9577e7e3
JH
330 while (last < ep) {
331 cp = last;
332 if (cp < ep && *cp == '#') {
333 while (cp < ep && *cp != sep)
334 cp++;
335 last = cp + 1;
336 continue;
337 }
c2f493a4
MW
338 while (cp < ep && *cp != sep)
339 cp++;
d5a63b99 340 if (last != cp) {
85dadc38 341 if (!is_absolute_path(last) && depth) {
c2f493a4
MW
342 error("%s: ignoring relative alternate object store %s",
343 relative_base, last);
344 } else {
345 link_alt_odb_entry(last, cp - last,
346 relative_base, depth);
1494e038 347 }
d5a63b99 348 }
9577e7e3 349 while (cp < ep && *cp == sep)
d5a63b99
JH
350 cp++;
351 last = cp;
9577e7e3 352 }
d5a63b99
JH
353}
354
c2f493a4 355static void read_info_alternates(const char * relative_base, int depth)
d5a63b99 356{
9577e7e3 357 char *map;
dc49cd76 358 size_t mapsz;
d5a63b99 359 struct stat st;
9cb18f56
JM
360 const char alt_file_name[] = "info/alternates";
361 /* Given that relative_base is no longer than PATH_MAX,
362 ensure that "path" has enough space to append "/", the
363 file name, "info/alternates", and a trailing NUL. */
364 char path[PATH_MAX + 1 + sizeof alt_file_name];
c2f493a4 365 int fd;
d5a63b99 366
9cb18f56 367 sprintf(path, "%s/%s", relative_base, alt_file_name);
d5a63b99
JH
368 fd = open(path, O_RDONLY);
369 if (fd < 0)
370 return;
371 if (fstat(fd, &st) || (st.st_size == 0)) {
372 close(fd);
9a217f2a 373 return;
ace1534d 374 }
dc49cd76
SP
375 mapsz = xsize_t(st.st_size);
376 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 377 close(fd);
d5a63b99 378
dc49cd76 379 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
c2f493a4 380
dc49cd76 381 munmap(map, mapsz);
ace1534d
JH
382}
383
bef70b22
DB
384void add_to_alternates_file(const char *reference)
385{
386 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
387 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), 1);
388 char *alt = mkpath("%s/objects\n", reference);
389 write_or_die(fd, alt, strlen(alt));
390 if (commit_lock_file(lock))
391 die("could not close alternates file");
392 if (alt_odb_tail)
393 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
394}
395
c2f493a4
MW
396void prepare_alt_odb(void)
397{
554fe20d 398 const char *alt;
c2f493a4 399
7dc24aa5
SP
400 if (alt_odb_tail)
401 return;
402
c2f493a4
MW
403 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
404 if (!alt) alt = "";
405
c2f493a4 406 alt_odb_tail = &alt_odb_list;
80ba074f 407 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
408
409 read_info_alternates(get_object_directory(), 0);
410}
411
c529d75a 412static int has_loose_object(const unsigned char *sha1)
ace1534d 413{
ace1534d 414 char *name = sha1_file_name(sha1);
d5a63b99 415 struct alternate_object_database *alt;
ace1534d 416
c529d75a
LT
417 if (!access(name, F_OK))
418 return 1;
9a217f2a 419 prepare_alt_odb();
d5a63b99
JH
420 for (alt = alt_odb_list; alt; alt = alt->next) {
421 name = alt->name;
ace1534d 422 fill_sha1_path(name, sha1);
c529d75a
LT
423 if (!access(alt->base, F_OK))
424 return 1;
ace1534d 425 }
c529d75a 426 return 0;
ace1534d
JH
427}
428
60bb8b14 429static unsigned int pack_used_ctr;
a53128b6
SP
430static unsigned int pack_mmap_calls;
431static unsigned int peak_pack_open_windows;
432static unsigned int pack_open_windows;
433static size_t peak_pack_mapped;
60bb8b14 434static size_t pack_mapped;
9a217f2a 435struct packed_git *packed_git;
1f688557 436
b79d18c9 437void pack_report(void)
a53128b6
SP
438{
439 fprintf(stderr,
e05db0fd
PR
440 "pack_report: getpagesize() = %10" SZ_FMT "\n"
441 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
442 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
443 sz_fmt(getpagesize()),
444 sz_fmt(packed_git_window_size),
445 sz_fmt(packed_git_limit));
a53128b6
SP
446 fprintf(stderr,
447 "pack_report: pack_used_ctr = %10u\n"
448 "pack_report: pack_mmap_calls = %10u\n"
449 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
450 "pack_report: pack_mapped = "
451 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
452 pack_used_ctr,
453 pack_mmap_calls,
454 pack_open_windows, peak_pack_open_windows,
9e42d6a1 455 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
456}
457
42873078 458static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
459{
460 void *idx_map;
42873078 461 struct pack_idx_header *hdr;
dc49cd76 462 size_t idx_size;
74e34e1f 463 uint32_t version, nr, i, *index;
e93ec6f9 464 int fd = open(path, O_RDONLY);
1f688557 465 struct stat st;
42873078 466
1f688557
JH
467 if (fd < 0)
468 return -1;
469 if (fstat(fd, &st)) {
470 close(fd);
471 return -1;
472 }
dc49cd76 473 idx_size = xsize_t(st.st_size);
2d88451b
SP
474 if (idx_size < 4 * 256 + 20 + 20) {
475 close(fd);
476 return error("index file %s is too small", path);
477 }
c4712e45 478 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 479 close(fd);
1f688557 480
42873078
NP
481 hdr = idx_map;
482 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
483 version = ntohl(hdr->idx_version);
484 if (version < 2 || version > 2) {
485 munmap(idx_map, idx_size);
6e1c2344 486 return error("index file %s is version %"PRIu32
74e34e1f
NP
487 " and is not supported by this binary"
488 " (try upgrading GIT to a newer version)",
489 path, version);
490 }
491 } else
492 version = 1;
df1b059d 493
1f688557 494 nr = 0;
42873078 495 index = idx_map;
74e34e1f
NP
496 if (version > 1)
497 index += 2; /* skip index header */
1f688557 498 for (i = 0; i < 256; i++) {
326bf396 499 uint32_t n = ntohl(index[i]);
2d88451b
SP
500 if (n < nr) {
501 munmap(idx_map, idx_size);
df1b059d 502 return error("non-monotonic index %s", path);
2d88451b 503 }
1f688557
JH
504 nr = n;
505 }
506
74e34e1f
NP
507 if (version == 1) {
508 /*
509 * Total size:
510 * - 256 index entries 4 bytes each
511 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
512 * - 20-byte SHA1 of the packfile
513 * - 20-byte SHA1 file checksum
514 */
515 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
516 munmap(idx_map, idx_size);
eef427a0 517 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
518 }
519 } else if (version == 2) {
520 /*
521 * Minimum size:
522 * - 8 bytes of header
523 * - 256 index entries 4 bytes each
524 * - 20-byte sha1 entry * nr
525 * - 4-byte crc entry * nr
526 * - 4-byte offset entry * nr
527 * - 20-byte SHA1 of the packfile
528 * - 20-byte SHA1 file checksum
529 * And after the 4-byte offset table might be a
530 * variable sized table containing 8-byte entries
531 * for offsets larger than 2^31.
532 */
533 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
534 unsigned long max_size = min_size;
535 if (nr)
536 max_size += (nr - 1)*8;
537 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 538 munmap(idx_map, idx_size);
eef427a0 539 return error("wrong index v2 file size in %s", path);
74e34e1f 540 }
7109c889
JH
541 if (idx_size != min_size &&
542 /*
543 * make sure we can deal with large pack offsets.
544 * 31-bit signed offset won't be enough, neither
545 * 32-bit unsigned one will be.
546 */
547 (sizeof(off_t) <= 4)) {
548 munmap(idx_map, idx_size);
549 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 550 }
2d88451b 551 }
1f688557 552
74e34e1f 553 p->index_version = version;
42873078
NP
554 p->index_data = idx_map;
555 p->index_size = idx_size;
57059091 556 p->num_objects = nr;
1f688557
JH
557 return 0;
558}
559
bc8e478a 560int open_pack_index(struct packed_git *p)
d079837e
SP
561{
562 char *idx_name;
563 int ret;
564
565 if (p->index_data)
566 return 0;
567
568 idx_name = xstrdup(p->pack_name);
569 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
570 ret = check_packed_git_idx(idx_name, p);
571 free(idx_name);
572 return ret;
573}
574
11daf39b
SP
575static void scan_windows(struct packed_git *p,
576 struct packed_git **lru_p,
577 struct pack_window **lru_w,
578 struct pack_window **lru_l)
1f688557 579{
11daf39b
SP
580 struct pack_window *w, *w_l;
581
582 for (w_l = NULL, w = p->windows; w; w = w->next) {
583 if (!w->inuse_cnt) {
584 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
585 *lru_p = p;
586 *lru_w = w;
587 *lru_l = w_l;
54044bf8 588 }
54044bf8 589 }
11daf39b 590 w_l = w;
f9253394 591 }
11daf39b
SP
592}
593
d1efefa4 594static int unuse_one_window(struct packed_git *current, int keep_fd)
11daf39b
SP
595{
596 struct packed_git *p, *lru_p = NULL;
597 struct pack_window *lru_w = NULL, *lru_l = NULL;
598
599 if (current)
600 scan_windows(current, &lru_p, &lru_w, &lru_l);
601 for (p = packed_git; p; p = p->next)
602 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
603 if (lru_p) {
604 munmap(lru_w->base, lru_w->len);
605 pack_mapped -= lru_w->len;
606 if (lru_l)
607 lru_l->next = lru_w->next;
608 else {
609 lru_p->windows = lru_w->next;
d1efefa4 610 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
54044bf8
SP
611 close(lru_p->pack_fd);
612 lru_p->pack_fd = -1;
613 }
614 }
615 free(lru_w);
a53128b6 616 pack_open_windows--;
54044bf8
SP
617 return 1;
618 }
619 return 0;
f9253394
JH
620}
621
d1efefa4 622void release_pack_memory(size_t need, int fd)
97bfeb34
SP
623{
624 size_t cur = pack_mapped;
d1efefa4 625 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
97bfeb34
SP
626 ; /* nothing */
627}
628
c9ced051
SP
629void close_pack_windows(struct packed_git *p)
630{
631 while (p->windows) {
632 struct pack_window *w = p->windows;
633
634 if (w->inuse_cnt)
635 die("pack '%s' still has open windows to it",
636 p->pack_name);
637 munmap(w->base, w->len);
638 pack_mapped -= w->len;
639 pack_open_windows--;
640 p->windows = w->next;
641 free(w);
642 }
643}
644
03e79c88 645void unuse_pack(struct pack_window **w_cursor)
f9253394 646{
03e79c88
SP
647 struct pack_window *w = *w_cursor;
648 if (w) {
649 w->inuse_cnt--;
650 *w_cursor = NULL;
651 }
1f688557
JH
652}
653
3cf8b462
SP
654/*
655 * Do not call this directly as this leaks p->pack_fd on error return;
656 * call open_packed_git() instead.
657 */
658static int open_packed_git_1(struct packed_git *p)
1f688557 659{
9bc879c1
SP
660 struct stat st;
661 struct pack_header hdr;
662 unsigned char sha1[20];
663 unsigned char *idx_sha1;
2c039da8 664 long fd_flag;
9bc879c1 665
d079837e
SP
666 if (!p->index_data && open_pack_index(p))
667 return error("packfile %s index unavailable", p->pack_name);
668
9bc879c1
SP
669 p->pack_fd = open(p->pack_name, O_RDONLY);
670 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 671 return -1;
9bc879c1
SP
672
673 /* If we created the struct before we had the pack we lack size. */
bf592c50 674 if (!p->pack_size) {
bf592c50 675 if (!S_ISREG(st.st_mode))
072db278 676 return error("packfile %s not a regular file", p->pack_name);
bf592c50 677 p->pack_size = st.st_size;
9bc879c1 678 } else if (p->pack_size != st.st_size)
072db278 679 return error("packfile %s size changed", p->pack_name);
9bc879c1 680
2c039da8
JH
681 /* We leave these file descriptors open with sliding mmap;
682 * there is no point keeping them open across exec(), though.
683 */
684 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
685 if (fd_flag < 0)
072db278 686 return error("cannot determine file descriptor flags");
2c039da8
JH
687 fd_flag |= FD_CLOEXEC;
688 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 689 return error("cannot set FD_CLOEXEC");
2c039da8 690
9bc879c1 691 /* Verify we recognize this pack file format. */
e6e2bd62 692 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 693 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 694 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 695 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 696 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
697 return error("packfile %s is version %"PRIu32" and not"
698 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
699 p->pack_name, ntohl(hdr.hdr_version));
700
701 /* Verify the pack matches its index. */
57059091 702 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
703 return error("packfile %s claims to have %"PRIu32" objects"
704 " while index indicates %"PRIu32" objects",
57059091
NP
705 p->pack_name, ntohl(hdr.hdr_entries),
706 p->num_objects);
9bc879c1 707 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 708 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 709 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 710 return error("packfile %s signature is unavailable", p->pack_name);
42873078 711 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 712 if (hashcmp(sha1, idx_sha1))
072db278
SP
713 return error("packfile %s does not match index", p->pack_name);
714 return 0;
9bc879c1
SP
715}
716
3cf8b462
SP
717static int open_packed_git(struct packed_git *p)
718{
719 if (!open_packed_git_1(p))
720 return 0;
721 if (p->pack_fd != -1) {
722 close(p->pack_fd);
723 p->pack_fd = -1;
724 }
725 return -1;
726}
727
c4001d92 728static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
729{
730 /* We must promise at least 20 bytes (one hash) after the
731 * offset is available from this window, otherwise the offset
732 * is not actually in this window and a different window (which
733 * has that one hash excess) must be used. This is to support
734 * the object header and delta base parsing routines below.
735 */
736 off_t win_off = win->offset;
737 return win_off <= offset
738 && (offset + 20) <= (win_off + win->len);
739}
740
03e79c88
SP
741unsigned char* use_pack(struct packed_git *p,
742 struct pack_window **w_cursor,
c4001d92 743 off_t offset,
03e79c88 744 unsigned int *left)
9bc879c1 745{
60bb8b14 746 struct pack_window *win = *w_cursor;
03e79c88 747
072db278
SP
748 if (p->pack_fd == -1 && open_packed_git(p))
749 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
750
751 /* Since packfiles end in a hash of their content and its
752 * pointless to ask for an offset into the middle of that
753 * hash, and the in_window function above wouldn't match
754 * don't allow an offset too close to the end of the file.
755 */
756 if (offset > (p->pack_size - 20))
757 die("offset beyond end of packfile (truncated pack?)");
758
759 if (!win || !in_window(win, offset)) {
760 if (win)
761 win->inuse_cnt--;
762 for (win = p->windows; win; win = win->next) {
763 if (in_window(win, offset))
764 break;
765 }
766 if (!win) {
78a28df9 767 size_t window_align = packed_git_window_size / 2;
dc49cd76 768 off_t len;
60bb8b14 769 win = xcalloc(1, sizeof(*win));
78a28df9 770 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
771 len = p->pack_size - win->offset;
772 if (len > packed_git_window_size)
773 len = packed_git_window_size;
774 win->len = (size_t)len;
60bb8b14 775 pack_mapped += win->len;
11daf39b 776 while (packed_git_limit < pack_mapped
d1efefa4 777 && unuse_one_window(p, p->pack_fd))
60bb8b14 778 ; /* nothing */
c4712e45 779 win->base = xmmap(NULL, win->len,
60bb8b14
SP
780 PROT_READ, MAP_PRIVATE,
781 p->pack_fd, win->offset);
782 if (win->base == MAP_FAILED)
73b4e4be
SP
783 die("packfile %s cannot be mapped: %s",
784 p->pack_name,
785 strerror(errno));
a53128b6
SP
786 pack_mmap_calls++;
787 pack_open_windows++;
788 if (pack_mapped > peak_pack_mapped)
789 peak_pack_mapped = pack_mapped;
790 if (pack_open_windows > peak_pack_open_windows)
791 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
792 win->next = p->windows;
793 p->windows = win;
794 }
1f688557 795 }
03e79c88
SP
796 if (win != *w_cursor) {
797 win->last_used = pack_used_ctr++;
798 win->inuse_cnt++;
799 *w_cursor = win;
800 }
60bb8b14 801 offset -= win->offset;
03e79c88 802 if (left)
dc49cd76 803 *left = win->len - xsize_t(offset);
03e79c88 804 return win->base + offset;
1f688557
JH
805}
806
27d69a46
NP
807static struct packed_git *alloc_packed_git(int extra)
808{
809 struct packed_git *p = xmalloc(sizeof(*p) + extra);
810 memset(p, 0, sizeof(*p));
811 p->pack_fd = -1;
812 return p;
813}
814
42873078 815struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557
JH
816{
817 struct stat st;
27d69a46 818 struct packed_git *p = alloc_packed_git(path_len + 2);
1f688557 819
42873078
NP
820 /*
821 * Make sure a corresponding .pack file exists and that
822 * the index looks sane.
823 */
824 path_len -= strlen(".idx");
27d69a46
NP
825 if (path_len < 1) {
826 free(p);
1f688557 827 return NULL;
27d69a46 828 }
42873078
NP
829 memcpy(p->pack_name, path, path_len);
830 strcpy(p->pack_name + path_len, ".pack");
d079837e 831 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 832 free(p);
1f688557
JH
833 return NULL;
834 }
42873078 835
1f688557
JH
836 /* ok, it looks sane as far as we can check without
837 * actually mapping the pack file.
838 */
1f688557 839 p->pack_size = st.st_size;
9d835df2 840 p->pack_local = local;
b867092f 841 p->mtime = st.st_mtime;
42873078
NP
842 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
843 hashclr(p->sha1);
1f688557
JH
844 return p;
845}
846
bf592c50 847struct packed_git *parse_pack_index(unsigned char *sha1)
c508df5e 848{
79c6dca4 849 const char *idx_path = sha1_pack_index_name(sha1);
42873078 850 const char *path = sha1_pack_name(sha1);
27d69a46 851 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
bf592c50 852
27d69a46
NP
853 strcpy(p->pack_name, path);
854 hashcpy(p->sha1, sha1);
42873078
NP
855 if (check_packed_git_idx(idx_path, p)) {
856 free(p);
bf592c50 857 return NULL;
42873078 858 }
bf592c50 859
bf592c50
DB
860 return p;
861}
862
863void install_packed_git(struct packed_git *pack)
864{
865 pack->next = packed_git;
866 packed_git = pack;
867}
868
9d835df2 869static void prepare_packed_git_one(char *objdir, int local)
1f688557 870{
9cb18f56
JM
871 /* Ensure that this buffer is large enough so that we can
872 append "/pack/" without clobbering the stack even if
873 strlen(objdir) were PATH_MAX. */
874 char path[PATH_MAX + 1 + 4 + 1 + 1];
1f688557
JH
875 int len;
876 DIR *dir;
877 struct dirent *de;
878
879 sprintf(path, "%s/pack", objdir);
880 len = strlen(path);
881 dir = opendir(path);
b5b16990 882 if (!dir) {
26125f6b 883 if (errno != ENOENT)
bd2afde8 884 error("unable to open object pack directory: %s: %s",
26125f6b 885 path, strerror(errno));
1f688557 886 return;
b5b16990 887 }
1f688557
JH
888 path[len++] = '/';
889 while ((de = readdir(dir)) != NULL) {
890 int namelen = strlen(de->d_name);
891 struct packed_git *p;
892
5bb1cda5 893 if (!has_extension(de->d_name, ".idx"))
1f688557
JH
894 continue;
895
9cb18f56
JM
896 if (len + namelen + 1 > sizeof(path))
897 continue;
898
54a15a8d 899 /* Don't reopen a pack we already have. */
1f688557 900 strcpy(path + len, de->d_name);
86f7780c
JK
901 for (p = packed_git; p; p = p->next) {
902 if (!memcmp(path, p->pack_name, len + namelen - 4))
903 break;
904 }
905 if (p)
906 continue;
54a15a8d
SP
907 /* See if it really is a valid .idx file with corresponding
908 * .pack file that we can map.
909 */
9d835df2 910 p = add_packed_git(path, len + namelen, local);
1f688557
JH
911 if (!p)
912 continue;
625e9421 913 install_packed_git(p);
1f688557 914 }
5b35bcd5 915 closedir(dir);
1f688557
JH
916}
917
b867092f
JH
918static int sort_pack(const void *a_, const void *b_)
919{
920 struct packed_git *a = *((struct packed_git **)a_);
921 struct packed_git *b = *((struct packed_git **)b_);
922 int st;
923
924 /*
925 * Local packs tend to contain objects specific to our
926 * variant of the project than remote ones. In addition,
927 * remote ones could be on a network mounted filesystem.
928 * Favor local ones for these reasons.
929 */
930 st = a->pack_local - b->pack_local;
931 if (st)
932 return -st;
933
934 /*
935 * Younger packs tend to contain more recent objects,
936 * and more recent objects tend to get accessed more
937 * often.
938 */
939 if (a->mtime < b->mtime)
940 return 1;
941 else if (a->mtime == b->mtime)
942 return 0;
943 return -1;
944}
945
946static void rearrange_packed_git(void)
947{
948 struct packed_git **ary, *p;
949 int i, n;
950
951 for (n = 0, p = packed_git; p; p = p->next)
952 n++;
953 if (n < 2)
954 return;
955
956 /* prepare an array of packed_git for easier sorting */
957 ary = xcalloc(n, sizeof(struct packed_git *));
958 for (n = 0, p = packed_git; p; p = p->next)
959 ary[n++] = p;
960
961 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
962
963 /* link them back again */
964 for (i = 0; i < n - 1; i++)
965 ary[i]->next = ary[i + 1];
966 ary[n - 1]->next = NULL;
967 packed_git = ary[0];
968
969 free(ary);
970}
971
637cdd9d 972static int prepare_packed_git_run_once = 0;
9a217f2a 973void prepare_packed_git(void)
1f688557 974{
d5a63b99 975 struct alternate_object_database *alt;
1f688557 976
637cdd9d 977 if (prepare_packed_git_run_once)
1f688557 978 return;
9d835df2 979 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 980 prepare_alt_odb();
d5a63b99 981 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 982 alt->name[-1] = 0;
9d835df2 983 prepare_packed_git_one(alt->base, 0);
1494e038 984 alt->name[-1] = '/';
1f688557 985 }
b867092f 986 rearrange_packed_git();
637cdd9d
JK
987 prepare_packed_git_run_once = 1;
988}
989
fc04c412 990void reprepare_packed_git(void)
637cdd9d 991{
4b480c67 992 discard_revindex();
637cdd9d
JK
993 prepare_packed_git_run_once = 0;
994 prepare_packed_git();
1f688557
JH
995}
996
8eca0b47
NP
997static void mark_bad_packed_object(struct packed_git *p,
998 const unsigned char *sha1)
999{
1000 unsigned i;
1001 for (i = 0; i < p->num_bad_objects; i++)
1002 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1003 return;
1004 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1005 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1006 p->num_bad_objects++;
1007}
1008
ac939109
NP
1009static int has_packed_and_bad(const unsigned char *sha1)
1010{
1011 struct packed_git *p;
1012 unsigned i;
1013
1014 for (p = packed_git; p; p = p->next)
1015 for (i = 0; i < p->num_bad_objects; i++)
1016 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1017 return 1;
1018 return 0;
1019}
1020
5d6ccf5c 1021int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
0fcfd160
LT
1022{
1023 unsigned char real_sha1[20];
7cfb5f36 1024 hash_sha1_file(map, size, type, real_sha1);
a89fccd2 1025 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1026}
1027
44d1c19e
LT
1028static int git_open_noatime(const char *name)
1029{
1030 static int sha1_file_open_flag = O_NOATIME;
1031 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1032
1033 /* Might the failure be due to O_NOATIME? */
1034 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1035 fd = open(name, O_RDONLY);
1036 if (fd >= 0)
1037 sha1_file_open_flag = 0;
1038 }
1039 return fd;
1040}
1041
1042static int open_sha1_file(const unsigned char *sha1)
1043{
1044 int fd;
1045 char *name = sha1_file_name(sha1);
1046 struct alternate_object_database *alt;
1047
1048 fd = git_open_noatime(name);
1049 if (fd >= 0)
1050 return fd;
1051
1052 prepare_alt_odb();
1053 errno = ENOENT;
1054 for (alt = alt_odb_list; alt; alt = alt->next) {
1055 name = alt->name;
1056 fill_sha1_path(name, sha1);
1057 fd = git_open_noatime(alt->base);
1058 if (fd >= 0)
1059 return fd;
1060 }
1061 return -1;
1062}
1063
4175e9e3 1064static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1065{
0fcfd160 1066 void *map;
144bde78 1067 int fd;
ace1534d 1068
44d1c19e
LT
1069 fd = open_sha1_file(sha1);
1070 map = NULL;
1071 if (fd >= 0) {
1072 struct stat st;
0fcfd160 1073
44d1c19e
LT
1074 if (!fstat(fd, &st)) {
1075 *size = xsize_t(st.st_size);
1076 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1077 }
44d1c19e 1078 close(fd);
0fcfd160 1079 }
0fcfd160
LT
1080 return map;
1081}
1082
726f852b 1083static int legacy_loose_object(unsigned char *map)
bb6b8e4f
JH
1084{
1085 unsigned int word;
1086
1087 /*
1088 * Is it a zlib-compressed buffer? If so, the first byte
1089 * must be 0x78 (15-bit window size, deflated), and the
1090 * first 16-bit word is evenly divisible by 31
1091 */
1092 word = (map[0] << 8) + map[1];
1093 if (map[0] == 0x78 && !(word % 31))
1094 return 1;
1095 else
1096 return 0;
1097}
1098
72518e9c 1099unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1100{
ad1ed5ee 1101 unsigned shift;
93821bd9 1102 unsigned char c;
93821bd9 1103 unsigned long size;
ad1ed5ee
JH
1104 unsigned long used = 0;
1105
1106 c = buf[used++];
1107 *type = (c >> 4) & 7;
1108 size = c & 15;
1109 shift = 4;
1110 while (c & 0x80) {
1111 if (len <= used)
1112 return 0;
1113 if (sizeof(long) * 8 <= shift)
1114 return 0;
1115 c = buf[used++];
1116 size += (c & 0x7f) << shift;
1117 shift += 7;
1118 }
1119 *sizep = size;
1120 return used;
1121}
1122
1123static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1124{
1125 unsigned long size, used;
1126 static const char valid_loose_object_type[8] = {
1127 0, /* OBJ_EXT */
1128 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1129 0, /* "delta" and others are invalid in a loose object */
93821bd9 1130 };
ad1ed5ee 1131 enum object_type type;
93821bd9 1132
c4483576
LT
1133 /* Get the data stream */
1134 memset(stream, 0, sizeof(*stream));
1135 stream->next_in = map;
1136 stream->avail_in = mapsize;
1137 stream->next_out = buffer;
93821bd9
LT
1138 stream->avail_out = bufsiz;
1139
bb6b8e4f 1140 if (legacy_loose_object(map)) {
93821bd9
LT
1141 inflateInit(stream);
1142 return inflate(stream, 0);
1143 }
1144
726f852b
NP
1145
1146 /*
1147 * There used to be a second loose object header format which
1148 * was meant to mimic the in-pack format, allowing for direct
1149 * copy of the object data. This format turned up not to be
1150 * really worth it and we don't write it any longer. But we
1151 * can still read it.
1152 */
ad1ed5ee
JH
1153 used = unpack_object_header_gently(map, mapsize, &type, &size);
1154 if (!used || !valid_loose_object_type[type])
93821bd9 1155 return -1;
ad1ed5ee
JH
1156 map += used;
1157 mapsize -= used;
93821bd9
LT
1158
1159 /* Set up the stream for the rest.. */
1160 stream->next_in = map;
1161 stream->avail_in = mapsize;
c4483576 1162 inflateInit(stream);
93821bd9
LT
1163
1164 /* And generate the fake traditional header */
ad1ed5ee 1165 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
df843662 1166 typename(type), size);
93821bd9 1167 return 0;
c4483576
LT
1168}
1169
7efbff75 1170static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1171{
1172 int bytes = strlen(buffer) + 1;
d565b341 1173 unsigned char *buf = xmalloc(1+size);
93821bd9 1174 unsigned long n;
7efbff75 1175 int status = Z_OK;
5180cacc 1176
93821bd9
LT
1177 n = stream->total_out - bytes;
1178 if (n > size)
1179 n = size;
1180 memcpy(buf, (char *) buffer + bytes, n);
1181 bytes = n;
456cdf6e
LT
1182 if (bytes <= size) {
1183 /*
1184 * The above condition must be (bytes <= size), not
1185 * (bytes < size). In other words, even though we
1186 * expect no more output and set avail_out to zer0,
1187 * the input zlib stream may have bytes that express
1188 * "this concludes the stream", and we *do* want to
1189 * eat that input.
1190 *
1191 * Otherwise we would not be able to test that we
1192 * consumed all the input to reach the expected size;
1193 * we also want to check that zlib tells us that all
1194 * went well with status == Z_STREAM_END at the end.
1195 */
5180cacc
LT
1196 stream->next_out = buf + bytes;
1197 stream->avail_out = size - bytes;
7efbff75
JH
1198 while (status == Z_OK)
1199 status = inflate(stream, Z_FINISH);
5180cacc
LT
1200 }
1201 buf[size] = 0;
456cdf6e 1202 if (status == Z_STREAM_END && !stream->avail_in) {
7efbff75
JH
1203 inflateEnd(stream);
1204 return buf;
1205 }
1206
1207 if (status < 0)
1208 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1209 else if (stream->avail_in)
1210 error("garbage at end of loose object '%s'",
1211 sha1_to_hex(sha1));
1212 free(buf);
1213 return NULL;
5180cacc
LT
1214}
1215
1216/*
1217 * We used to just use "sscanf()", but that's actually way
1218 * too permissive for what we want to check. So do an anal
1219 * object header parse by hand.
1220 */
21666f1a 1221static int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1222{
21666f1a 1223 char type[10];
5180cacc
LT
1224 int i;
1225 unsigned long size;
1226
1227 /*
a6080a0a 1228 * The type can be at most ten bytes (including the
5180cacc 1229 * terminating '\0' that we add), and is followed by
21666f1a 1230 * a space.
5180cacc 1231 */
21666f1a 1232 i = 0;
5180cacc
LT
1233 for (;;) {
1234 char c = *hdr++;
1235 if (c == ' ')
1236 break;
21666f1a
NP
1237 type[i++] = c;
1238 if (i >= sizeof(type))
5180cacc 1239 return -1;
5180cacc 1240 }
21666f1a 1241 type[i] = 0;
5180cacc
LT
1242
1243 /*
1244 * The length must follow immediately, and be in canonical
1245 * decimal format (ie "010" is not valid).
1246 */
1247 size = *hdr++ - '0';
1248 if (size > 9)
1249 return -1;
1250 if (size) {
1251 for (;;) {
1252 unsigned long c = *hdr - '0';
1253 if (c > 9)
1254 break;
1255 hdr++;
1256 size = size * 10 + c;
1257 }
1258 }
1259 *sizep = size;
1260
1261 /*
1262 * The length must be followed by a zero byte
1263 */
21666f1a 1264 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1265}
1266
7efbff75 1267static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1268{
5180cacc 1269 int ret;
0fcfd160 1270 z_stream stream;
5180cacc 1271 char hdr[8192];
0fcfd160 1272
5180cacc 1273 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1274 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1275 return NULL;
1276
7efbff75 1277 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1278}
1279
54dab52a
NP
1280unsigned long get_size_from_delta(struct packed_git *p,
1281 struct pack_window **w_curs,
1282 off_t curpos)
1283{
1284 const unsigned char *data;
1285 unsigned char delta_head[20], *in;
1286 z_stream stream;
1287 int st;
1288
1289 memset(&stream, 0, sizeof(stream));
1290 stream.next_out = delta_head;
1291 stream.avail_out = sizeof(delta_head);
1292
1293 inflateInit(&stream);
1294 do {
1295 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1296 stream.next_in = in;
1297 st = inflate(&stream, Z_FINISH);
1298 curpos += stream.next_in - in;
1299 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1300 stream.total_out < sizeof(delta_head));
1301 inflateEnd(&stream);
1302 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1303 die("delta data unpack-initial failed");
1304
1305 /* Examine the initial part of the delta to figure out
1306 * the result size.
1307 */
1308 data = delta_head;
1309
1310 /* ignore base size */
1311 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1312
1313 /* Read the result size */
1314 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1315}
1316
c4001d92 1317static off_t get_delta_base(struct packed_git *p,
03e79c88 1318 struct pack_window **w_curs,
c4001d92 1319 off_t *curpos,
21666f1a 1320 enum object_type type,
c4001d92 1321 off_t delta_obj_offset)
eb32d236 1322{
2b87c45b 1323 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1324 off_t base_offset;
eb32d236 1325
8d8a4ea5
SP
1326 /* use_pack() assured us we have [base_info, base_info + 20)
1327 * as a range that we can look at without walking off the
1328 * end of the mapped window. Its actually the hash size
1329 * that is assured. An OFS_DELTA longer than the hash size
1330 * is stupid, as then a REF_DELTA would be smaller to store.
1331 */
21666f1a 1332 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1333 unsigned used = 0;
1334 unsigned char c = base_info[used++];
1335 base_offset = c & 127;
1336 while (c & 128) {
1337 base_offset += 1;
8723f216 1338 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1339 return 0; /* overflow */
eb32d236
NP
1340 c = base_info[used++];
1341 base_offset = (base_offset << 7) + (c & 127);
1342 }
1343 base_offset = delta_obj_offset - base_offset;
1344 if (base_offset >= delta_obj_offset)
8eca0b47 1345 return 0; /* out of bound */
2b87c45b 1346 *curpos += used;
21666f1a 1347 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1348 /* The base entry _must_ be in the same pack */
1349 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1350 *curpos += 20;
eb32d236
NP
1351 } else
1352 die("I am totally screwed");
2b87c45b 1353 return base_offset;
eb32d236
NP
1354}
1355
f3bf9224 1356/* forward declaration for a mutually recursive function */
c4001d92 1357static int packed_object_info(struct packed_git *p, off_t offset,
21666f1a 1358 unsigned long *sizep);
f3bf9224 1359
43057304 1360static int packed_delta_info(struct packed_git *p,
03e79c88 1361 struct pack_window **w_curs,
c4001d92 1362 off_t curpos,
21666f1a 1363 enum object_type type,
c4001d92 1364 off_t obj_offset,
43057304 1365 unsigned long *sizep)
5db47c2b 1366{
c4001d92 1367 off_t base_offset;
f3bf9224 1368
21666f1a
NP
1369 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1370 type = packed_object_info(p, base_offset, NULL);
f3bf9224 1371
c62266f3
JH
1372 /* We choose to only get the type of the base object and
1373 * ignore potentially corrupt pack file that expects the delta
1374 * based on a base with a wrong size. This saves tons of
1375 * inflate() calls.
1376 */
54dab52a
NP
1377 if (sizep)
1378 *sizep = get_size_from_delta(p, w_curs, curpos);
21666f1a
NP
1379
1380 return type;
5db47c2b
JH
1381}
1382
2b87c45b
NP
1383static int unpack_object_header(struct packed_git *p,
1384 struct pack_window **w_curs,
c4001d92 1385 off_t *curpos,
2b87c45b 1386 unsigned long *sizep)
a733cb60 1387{
03e79c88
SP
1388 unsigned char *base;
1389 unsigned int left;
ad1ed5ee 1390 unsigned long used;
2b87c45b 1391 enum object_type type;
a733cb60 1392
8d8a4ea5
SP
1393 /* use_pack() assures us we have [base, base + 20) available
1394 * as a range that we can look at at. (Its actually the hash
3dff5379 1395 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1396 * the maximum deflated object size is 2^137, which is just
1397 * insane, so we know won't exceed what we have been given.
1398 */
2b87c45b
NP
1399 base = use_pack(p, w_curs, *curpos, &left);
1400 used = unpack_object_header_gently(base, left, &type, sizep);
ad1ed5ee
JH
1401 if (!used)
1402 die("object offset outside of pack file");
2b87c45b 1403 *curpos += used;
ad1ed5ee 1404
2b87c45b 1405 return type;
a733cb60
LT
1406}
1407
21666f1a 1408const char *packed_object_info_detail(struct packed_git *p,
c4001d92 1409 off_t obj_offset,
21666f1a
NP
1410 unsigned long *size,
1411 unsigned long *store_size,
1412 unsigned int *delta_chain_length,
1413 unsigned char *base_sha1)
ad8c80a5 1414{
03e79c88 1415 struct pack_window *w_curs = NULL;
c4001d92
SP
1416 off_t curpos;
1417 unsigned long dummy;
43057304 1418 unsigned char *next_sha1;
21666f1a 1419 enum object_type type;
70f5d5d3 1420 struct revindex_entry *revidx;
ad8c80a5 1421
43057304 1422 *delta_chain_length = 0;
2b87c45b 1423 curpos = obj_offset;
21666f1a 1424 type = unpack_object_header(p, &w_curs, &curpos, size);
43057304 1425
70f5d5d3
NP
1426 revidx = find_pack_revindex(p, obj_offset);
1427 *store_size = revidx[1].offset - obj_offset;
1428
43057304 1429 for (;;) {
21666f1a 1430 switch (type) {
43057304 1431 default:
08a19d87 1432 die("pack %s contains unknown object type %d",
21666f1a 1433 p->pack_name, type);
43057304
NP
1434 case OBJ_COMMIT:
1435 case OBJ_TREE:
1436 case OBJ_BLOB:
1437 case OBJ_TAG:
03e79c88 1438 unuse_pack(&w_curs);
21666f1a 1439 return typename(type);
eb32d236 1440 case OBJ_OFS_DELTA:
21666f1a 1441 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
8eca0b47
NP
1442 if (!obj_offset)
1443 die("pack %s contains bad delta base reference of type %s",
1444 p->pack_name, typename(type));
eb32d236 1445 if (*delta_chain_length == 0) {
70f5d5d3
NP
1446 revidx = find_pack_revindex(p, obj_offset);
1447 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
eb32d236
NP
1448 }
1449 break;
1450 case OBJ_REF_DELTA:
2b87c45b 1451 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
43057304
NP
1452 if (*delta_chain_length == 0)
1453 hashcpy(base_sha1, next_sha1);
2b87c45b 1454 obj_offset = find_pack_entry_one(next_sha1, p);
43057304
NP
1455 break;
1456 }
43057304 1457 (*delta_chain_length)++;
2b87c45b 1458 curpos = obj_offset;
21666f1a 1459 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
ad8c80a5 1460 }
ad8c80a5
JH
1461}
1462
c4001d92 1463static int packed_object_info(struct packed_git *p, off_t obj_offset,
21666f1a 1464 unsigned long *sizep)
1f688557 1465{
03e79c88 1466 struct pack_window *w_curs = NULL;
c4001d92
SP
1467 unsigned long size;
1468 off_t curpos = obj_offset;
21666f1a 1469 enum object_type type;
5db47c2b 1470
21666f1a 1471 type = unpack_object_header(p, &w_curs, &curpos, &size);
5db47c2b 1472
21666f1a 1473 switch (type) {
eb32d236
NP
1474 case OBJ_OFS_DELTA:
1475 case OBJ_REF_DELTA:
21666f1a
NP
1476 type = packed_delta_info(p, &w_curs, curpos,
1477 type, obj_offset, sizep);
2b87c45b 1478 break;
a733cb60 1479 case OBJ_COMMIT:
a733cb60 1480 case OBJ_TREE:
a733cb60 1481 case OBJ_BLOB:
a733cb60 1482 case OBJ_TAG:
2b87c45b
NP
1483 if (sizep)
1484 *sizep = size;
a69d0943 1485 break;
1f688557 1486 default:
08a19d87 1487 die("pack %s contains unknown object type %d",
21666f1a 1488 p->pack_name, type);
1f688557 1489 }
2b87c45b 1490 unuse_pack(&w_curs);
21666f1a 1491 return type;
1f688557
JH
1492}
1493
eb950c19 1494static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1495 struct pack_window **w_curs,
c4001d92 1496 off_t curpos,
eb950c19 1497 unsigned long size)
de530aaa
SP
1498{
1499 int st;
1500 z_stream stream;
079afb18 1501 unsigned char *buffer, *in;
de530aaa
SP
1502
1503 buffer = xmalloc(size + 1);
1504 buffer[size] = 0;
1505 memset(&stream, 0, sizeof(stream));
de530aaa
SP
1506 stream.next_out = buffer;
1507 stream.avail_out = size;
1508
1509 inflateInit(&stream);
079afb18 1510 do {
2b87c45b 1511 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18
SP
1512 stream.next_in = in;
1513 st = inflate(&stream, Z_FINISH);
2b87c45b 1514 curpos += stream.next_in - in;
079afb18 1515 } while (st == Z_OK || st == Z_BUF_ERROR);
de530aaa
SP
1516 inflateEnd(&stream);
1517 if ((st != Z_STREAM_END) || stream.total_out != size) {
1518 free(buffer);
1519 return NULL;
1520 }
1521
1522 return buffer;
1523}
1524
e5e01619
LT
1525#define MAX_DELTA_CACHE (256)
1526
18bdec11 1527static size_t delta_base_cached;
5e08ecbf
NP
1528
1529static struct delta_base_cache_lru_list {
1530 struct delta_base_cache_lru_list *prev;
1531 struct delta_base_cache_lru_list *next;
1532} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1533
e5e01619 1534static struct delta_base_cache_entry {
5e08ecbf
NP
1535 struct delta_base_cache_lru_list lru;
1536 void *data;
e5e01619
LT
1537 struct packed_git *p;
1538 off_t base_offset;
1539 unsigned long size;
e5e01619
LT
1540 enum object_type type;
1541} delta_base_cache[MAX_DELTA_CACHE];
1542
1543static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1544{
1545 unsigned long hash;
1546
1547 hash = (unsigned long)p + (unsigned long)base_offset;
1548 hash += (hash >> 8) + (hash >> 16);
3358004a 1549 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1550}
1551
62f255ad 1552static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 1553 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 1554{
e5e01619
LT
1555 void *ret;
1556 unsigned long hash = pack_entry_hash(p, base_offset);
1557 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1558
1559 ret = ent->data;
1560 if (ret && ent->p == p && ent->base_offset == base_offset)
1561 goto found_cache_entry;
62f255ad 1562 return unpack_entry(p, base_offset, type, base_size);
e5e01619
LT
1563
1564found_cache_entry:
18bdec11 1565 if (!keep_cache) {
a0cba108 1566 ent->data = NULL;
5e08ecbf
NP
1567 ent->lru.next->prev = ent->lru.prev;
1568 ent->lru.prev->next = ent->lru.next;
18bdec11 1569 delta_base_cached -= ent->size;
182af834
PH
1570 } else {
1571 ret = xmemdupz(ent->data, ent->size);
a0cba108 1572 }
e5e01619
LT
1573 *type = ent->type;
1574 *base_size = ent->size;
1575 return ret;
62f255ad
LT
1576}
1577
18bdec11
SP
1578static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1579{
1580 if (ent->data) {
1581 free(ent->data);
1582 ent->data = NULL;
5e08ecbf
NP
1583 ent->lru.next->prev = ent->lru.prev;
1584 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
1585 delta_base_cached -= ent->size;
1586 }
1587}
1588
62f255ad
LT
1589static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1590 void *base, unsigned long base_size, enum object_type type)
1591{
5e08ecbf 1592 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 1593 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 1594 struct delta_base_cache_lru_list *lru;
e5e01619 1595
18bdec11
SP
1596 release_delta_base_cache(ent);
1597 delta_base_cached += base_size;
5e08ecbf
NP
1598
1599 for (lru = delta_base_cache_lru.next;
1600 delta_base_cached > delta_base_cache_limit
1601 && lru != &delta_base_cache_lru;
1602 lru = lru->next) {
1603 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
1604 if (f->type == OBJ_BLOB)
1605 release_delta_base_cache(f);
1606 }
5e08ecbf
NP
1607 for (lru = delta_base_cache_lru.next;
1608 delta_base_cached > delta_base_cache_limit
1609 && lru != &delta_base_cache_lru;
1610 lru = lru->next) {
1611 struct delta_base_cache_entry *f = (void *)lru;
1612 release_delta_base_cache(f);
1613 }
18bdec11 1614
e5e01619
LT
1615 ent->p = p;
1616 ent->base_offset = base_offset;
1617 ent->type = type;
1618 ent->data = base;
1619 ent->size = base_size;
5e08ecbf
NP
1620 ent->lru.next = &delta_base_cache_lru;
1621 ent->lru.prev = delta_base_cache_lru.prev;
1622 delta_base_cache_lru.prev->next = &ent->lru;
1623 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
1624}
1625
eb950c19 1626static void *unpack_delta_entry(struct packed_git *p,
03e79c88 1627 struct pack_window **w_curs,
c4001d92 1628 off_t curpos,
1f688557 1629 unsigned long delta_size,
c4001d92 1630 off_t obj_offset,
21666f1a 1631 enum object_type *type,
eb950c19 1632 unsigned long *sizep)
1f688557 1633{
7c3e8be3 1634 void *delta_data, *result, *base;
c4001d92
SP
1635 unsigned long base_size;
1636 off_t base_offset;
43057304 1637
21666f1a 1638 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
8eca0b47
NP
1639 if (!base_offset) {
1640 error("failed to validate delta base reference "
1641 "at offset %"PRIuMAX" from %s",
1642 (uintmax_t)curpos, p->pack_name);
1643 return NULL;
1644 }
eac12e2d 1645 unuse_pack(w_curs);
a0cba108 1646 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
8eca0b47
NP
1647 if (!base) {
1648 /*
1649 * We're probably in deep shit, but let's try to fetch
1650 * the required base anyway from another pack or loose.
1651 * This is costly but should happen only in the presence
1652 * of a corrupted pack, and is better than failing outright.
1653 */
1654 struct revindex_entry *revidx = find_pack_revindex(p, base_offset);
1655 const unsigned char *base_sha1 =
1656 nth_packed_object_sha1(p, revidx->nr);
1657 error("failed to read delta base object %s"
1658 " at offset %"PRIuMAX" from %s",
1659 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1660 p->pack_name);
1661 mark_bad_packed_object(p, base_sha1);
ac939109 1662 base = read_object(base_sha1, type, &base_size);
8eca0b47
NP
1663 if (!base)
1664 return NULL;
1665 }
67686d95 1666
2b87c45b 1667 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
8eca0b47
NP
1668 if (!delta_data) {
1669 error("failed to unpack compressed delta "
1670 "at offset %"PRIuMAX" from %s",
1671 (uintmax_t)curpos, p->pack_name);
1672 free(base);
1673 return NULL;
1674 }
1f688557
JH
1675 result = patch_delta(base, base_size,
1676 delta_data, delta_size,
21666f1a 1677 sizep);
1f688557
JH
1678 if (!result)
1679 die("failed to apply delta");
1680 free(delta_data);
62f255ad 1681 add_delta_base_cache(p, base_offset, base, base_size, *type);
1f688557
JH
1682 return result;
1683}
1684
c4001d92 1685void *unpack_entry(struct packed_git *p, off_t obj_offset,
21666f1a 1686 enum object_type *type, unsigned long *sizep)
1f688557 1687{
03e79c88 1688 struct pack_window *w_curs = NULL;
c4001d92 1689 off_t curpos = obj_offset;
21666f1a 1690 void *data;
1f688557 1691
21666f1a
NP
1692 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1693 switch (*type) {
eb32d236
NP
1694 case OBJ_OFS_DELTA:
1695 case OBJ_REF_DELTA:
21666f1a
NP
1696 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1697 obj_offset, type, sizep);
4d703a1a 1698 break;
a733cb60 1699 case OBJ_COMMIT:
a733cb60 1700 case OBJ_TREE:
a733cb60 1701 case OBJ_BLOB:
a733cb60 1702 case OBJ_TAG:
21666f1a 1703 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
4d703a1a 1704 break;
1f688557 1705 default:
8eca0b47
NP
1706 data = NULL;
1707 error("unknown object type %i at offset %"PRIuMAX" in %s",
1708 *type, (uintmax_t)obj_offset, p->pack_name);
1f688557 1709 }
03e79c88 1710 unuse_pack(&w_curs);
21666f1a 1711 return data;
1f688557
JH
1712}
1713
d079837e 1714const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 1715 uint32_t n)
9a217f2a 1716{
42873078 1717 const unsigned char *index = p->index_data;
d079837e
SP
1718 if (!index) {
1719 if (open_pack_index(p))
1720 return NULL;
1721 index = p->index_data;
1722 }
57059091 1723 if (n >= p->num_objects)
d72308e0 1724 return NULL;
74e34e1f
NP
1725 index += 4 * 256;
1726 if (p->index_version == 1) {
1727 return index + 24 * n + 4;
1728 } else {
1729 index += 8;
1730 return index + 20 * n;
1731 }
1732}
1733
99093238 1734off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
1735{
1736 const unsigned char *index = p->index_data;
1737 index += 4 * 256;
1738 if (p->index_version == 1) {
1739 return ntohl(*((uint32_t *)(index + 24 * n)));
1740 } else {
1741 uint32_t off;
1742 index += 8 + p->num_objects * (20 + 4);
1743 off = ntohl(*((uint32_t *)(index + 4 * n)));
1744 if (!(off & 0x80000000))
1745 return off;
1746 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1747 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1748 ntohl(*((uint32_t *)(index + 4)));
1749 }
9a217f2a
JH
1750}
1751
c4001d92 1752off_t find_pack_entry_one(const unsigned char *sha1,
43057304 1753 struct packed_git *p)
1f688557 1754{
42873078 1755 const uint32_t *level1_ofs = p->index_data;
42873078 1756 const unsigned char *index = p->index_data;
628522ec
JH
1757 unsigned hi, lo, stride;
1758 static int use_lookup = -1;
1759 static int debug_lookup = -1;
1760
1761 if (debug_lookup < 0)
1762 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 1763
d079837e
SP
1764 if (!index) {
1765 if (open_pack_index(p))
1766 return 0;
1767 level1_ofs = p->index_data;
1768 index = p->index_data;
1769 }
74e34e1f
NP
1770 if (p->index_version > 1) {
1771 level1_ofs += 2;
1772 index += 8;
1773 }
42873078 1774 index += 4 * 256;
74e34e1f
NP
1775 hi = ntohl(level1_ofs[*sha1]);
1776 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
1777 if (p->index_version > 1) {
1778 stride = 20;
1779 } else {
1780 stride = 24;
1781 index += 4;
1782 }
1783
1784 if (debug_lookup)
6e1c2344 1785 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
1786 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1787
1788 if (use_lookup < 0)
1789 use_lookup = !!getenv("GIT_USE_LOOKUP");
1790 if (use_lookup) {
1791 int pos = sha1_entry_pos(index, stride, 0,
1792 lo, hi, p->num_objects, sha1);
1793 if (pos < 0)
1794 return 0;
1795 return nth_packed_object_offset(p, pos);
1796 }
1f688557
JH
1797
1798 do {
74e34e1f 1799 unsigned mi = (lo + hi) / 2;
628522ec
JH
1800 int cmp = hashcmp(index + mi * stride, sha1);
1801
1802 if (debug_lookup)
1803 printf("lo %u hi %u rg %u mi %u\n",
1804 lo, hi, hi - lo, mi);
43057304 1805 if (!cmp)
74e34e1f 1806 return nth_packed_object_offset(p, mi);
1f688557
JH
1807 if (cmp > 0)
1808 hi = mi;
1809 else
1810 lo = mi+1;
1811 } while (lo < hi);
1812 return 0;
1813}
1814
000dfd3f 1815int matches_pack_name(struct packed_git *p, const char *name)
d4ff6d92
SP
1816{
1817 const char *last_c, *c;
1818
000dfd3f
JH
1819 if (!strcmp(p->pack_name, name))
1820 return 1;
d4ff6d92
SP
1821
1822 for (c = p->pack_name, last_c = c; *c;)
1823 if (*c == '/')
1824 last_c = ++c;
1825 else
1826 ++c;
000dfd3f
JH
1827 if (!strcmp(last_c, name))
1828 return 1;
d4ff6d92 1829
000dfd3f 1830 return 0;
d4ff6d92
SP
1831}
1832
106d710b 1833static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1f688557 1834{
f7c22cc6 1835 static struct packed_git *last_found = (void *)1;
1f688557 1836 struct packed_git *p;
c4001d92 1837 off_t offset;
43057304 1838
1f688557 1839 prepare_packed_git();
f7c22cc6
NP
1840 if (!packed_git)
1841 return 0;
1842 p = (last_found == (void *)1) ? packed_git : last_found;
1f688557 1843
f7c22cc6 1844 do {
106d710b
JH
1845 if (ignore_packed) {
1846 const char **ig;
1847 for (ig = ignore_packed; *ig; ig++)
000dfd3f 1848 if (matches_pack_name(p, *ig))
106d710b
JH
1849 break;
1850 if (*ig)
f7c22cc6 1851 goto next;
106d710b 1852 }
f7c22cc6 1853
8eca0b47
NP
1854 if (p->num_bad_objects) {
1855 unsigned i;
1856 for (i = 0; i < p->num_bad_objects; i++)
1857 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1858 goto next;
1859 }
1860
43057304
NP
1861 offset = find_pack_entry_one(sha1, p);
1862 if (offset) {
c715f783
SP
1863 /*
1864 * We are about to tell the caller where they can
1865 * locate the requested object. We better make
1866 * sure the packfile is still here and can be
1867 * accessed before supplying that answer, as
1868 * it may have been deleted since the index
1869 * was loaded!
1870 */
1871 if (p->pack_fd == -1 && open_packed_git(p)) {
1872 error("packfile %s cannot be accessed", p->pack_name);
f7c22cc6 1873 goto next;
c715f783 1874 }
43057304
NP
1875 e->offset = offset;
1876 e->p = p;
1877 hashcpy(e->sha1, sha1);
f7c22cc6 1878 last_found = p;
1f688557 1879 return 1;
43057304 1880 }
f7c22cc6
NP
1881
1882 next:
1883 if (p == last_found)
1884 p = packed_git;
1885 else
1886 p = p->next;
1887 if (p == last_found)
1888 p = p->next;
1889 } while (p);
1f688557
JH
1890 return 0;
1891}
1892
a6080a0a 1893struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
1894 struct packed_git *packs)
1895{
1896 struct packed_git *p;
bf592c50
DB
1897
1898 for (p = packs; p; p = p->next) {
43057304 1899 if (find_pack_entry_one(sha1, p))
bf592c50
DB
1900 return p;
1901 }
1902 return NULL;
21666f1a 1903
bf592c50
DB
1904}
1905
21666f1a 1906static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
65c2e0c3 1907{
36e4d74a 1908 int status;
65c2e0c3
JH
1909 unsigned long mapsize, size;
1910 void *map;
1911 z_stream stream;
d65a16f6 1912 char hdr[32];
65c2e0c3 1913
bb6b8e4f 1914 map = map_sha1_file(sha1, &mapsize);
f0df4ed5
JS
1915 if (!map)
1916 return error("unable to find %s", sha1_to_hex(sha1));
36e4d74a
JH
1917 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1918 status = error("unable to unpack %s header",
1919 sha1_to_hex(sha1));
21666f1a 1920 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 1921 status = error("unable to parse %s header", sha1_to_hex(sha1));
21666f1a
NP
1922 else if (sizep)
1923 *sizep = size;
65c2e0c3
JH
1924 inflateEnd(&stream);
1925 munmap(map, mapsize);
1926 return status;
1927}
1928
21666f1a 1929int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
f0df4ed5 1930{
f0df4ed5 1931 struct pack_entry e;
ddd63e64 1932 int status;
f0df4ed5
JS
1933
1934 if (!find_pack_entry(sha1, &e, NULL)) {
ddd63e64
SG
1935 /* Most likely it's a loose object. */
1936 status = sha1_loose_object_info(sha1, sizep);
1937 if (status >= 0)
1938 return status;
1939
1940 /* Not a loose object; someone else may have just packed it. */
f0df4ed5
JS
1941 reprepare_packed_git();
1942 if (!find_pack_entry(sha1, &e, NULL))
ddd63e64 1943 return status;
f0df4ed5 1944 }
21666f1a 1945 return packed_object_info(e.p, e.offset, sizep);
f0df4ed5
JS
1946}
1947
21666f1a
NP
1948static void *read_packed_sha1(const unsigned char *sha1,
1949 enum object_type *type, unsigned long *size)
1f688557
JH
1950{
1951 struct pack_entry e;
8eca0b47 1952 void *data;
1f688557 1953
8276c007 1954 if (!find_pack_entry(sha1, &e, NULL))
1f688557 1955 return NULL;
8eca0b47
NP
1956 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1957 if (!data) {
1958 /*
1959 * We're probably in deep shit, but let's try to fetch
1960 * the required object anyway from another pack or loose.
1961 * This should happen only in the presence of a corrupted
1962 * pack, and is better than failing outright.
1963 */
1964 error("failed to read object %s at offset %"PRIuMAX" from %s",
1965 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
1966 mark_bad_packed_object(e.p, sha1);
ac939109 1967 data = read_object(sha1, type, size);
8eca0b47
NP
1968 }
1969 return data;
1f688557
JH
1970}
1971
d66b37bb
JH
1972/*
1973 * This is meant to hold a *small* number of objects that you would
1974 * want read_sha1_file() to be able to return, but yet you do not want
1975 * to write them into the object store (e.g. a browse-only
1976 * application).
1977 */
1978static struct cached_object {
1979 unsigned char sha1[20];
21666f1a 1980 enum object_type type;
d66b37bb
JH
1981 void *buf;
1982 unsigned long size;
1983} *cached_objects;
1984static int cached_object_nr, cached_object_alloc;
1985
346245a1
JK
1986static struct cached_object empty_tree = {
1987 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
1988 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
1989 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
1990 OBJ_TREE,
1991 "",
1992 0
1993};
1994
d66b37bb
JH
1995static struct cached_object *find_cached_object(const unsigned char *sha1)
1996{
1997 int i;
1998 struct cached_object *co = cached_objects;
1999
2000 for (i = 0; i < cached_object_nr; i++, co++) {
2001 if (!hashcmp(co->sha1, sha1))
2002 return co;
2003 }
346245a1
JK
2004 if (!hashcmp(sha1, empty_tree.sha1))
2005 return &empty_tree;
d66b37bb
JH
2006 return NULL;
2007}
2008
21666f1a
NP
2009int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2010 unsigned char *sha1)
d66b37bb
JH
2011{
2012 struct cached_object *co;
2013
21666f1a 2014 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2015 if (has_sha1_file(sha1) || find_cached_object(sha1))
2016 return 0;
2017 if (cached_object_alloc <= cached_object_nr) {
2018 cached_object_alloc = alloc_nr(cached_object_alloc);
2019 cached_objects = xrealloc(cached_objects,
2020 sizeof(*cached_objects) *
2021 cached_object_alloc);
2022 }
2023 co = &cached_objects[cached_object_nr++];
2024 co->size = len;
21666f1a 2025 co->type = type;
efa13f7b
JH
2026 co->buf = xmalloc(len);
2027 memcpy(co->buf, buf, len);
d66b37bb
JH
2028 hashcpy(co->sha1, sha1);
2029 return 0;
2030}
2031
ac939109
NP
2032void *read_object(const unsigned char *sha1, enum object_type *type,
2033 unsigned long *size)
0fcfd160
LT
2034{
2035 unsigned long mapsize;
2036 void *map, *buf;
d66b37bb
JH
2037 struct cached_object *co;
2038
2039 co = find_cached_object(sha1);
2040 if (co) {
21666f1a 2041 *type = co->type;
d66b37bb 2042 *size = co->size;
182af834 2043 return xmemdupz(co->buf, co->size);
d66b37bb 2044 }
0fcfd160 2045
8276c007
PE
2046 buf = read_packed_sha1(sha1, type, size);
2047 if (buf)
2048 return buf;
bb6b8e4f 2049 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2050 if (map) {
7efbff75 2051 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2052 munmap(map, mapsize);
2053 return buf;
2054 }
637cdd9d 2055 reprepare_packed_git();
8276c007 2056 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2057}
2058
ac939109
NP
2059void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
2060 unsigned long *size)
2061{
2062 void *data = read_object(sha1, type, size);
2063 /* legacy behavior is to die on corrupted objects */
2064 if (!data && (has_loose_object(sha1) || has_packed_and_bad(sha1)))
2065 die("object %s is corrupted", sha1_to_hex(sha1));
2066 return data;
2067}
2068
40469ee9 2069void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2070 const char *required_type_name,
40469ee9
JH
2071 unsigned long *size,
2072 unsigned char *actual_sha1_return)
f4913f91 2073{
21666f1a 2074 enum object_type type, required_type;
f4913f91
JH
2075 void *buffer;
2076 unsigned long isize;
40469ee9 2077 unsigned char actual_sha1[20];
f4913f91 2078
21666f1a 2079 required_type = type_from_string(required_type_name);
e702496e 2080 hashcpy(actual_sha1, sha1);
40469ee9
JH
2081 while (1) {
2082 int ref_length = -1;
2083 const char *ref_type = NULL;
f4913f91 2084
21666f1a 2085 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2086 if (!buffer)
2087 return NULL;
21666f1a 2088 if (type == required_type) {
40469ee9
JH
2089 *size = isize;
2090 if (actual_sha1_return)
e702496e 2091 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2092 return buffer;
2093 }
2094 /* Handle references */
21666f1a 2095 else if (type == OBJ_COMMIT)
40469ee9 2096 ref_type = "tree ";
21666f1a 2097 else if (type == OBJ_TAG)
40469ee9
JH
2098 ref_type = "object ";
2099 else {
2100 free(buffer);
2101 return NULL;
2102 }
2103 ref_length = strlen(ref_type);
f4913f91 2104
50974ec9
MK
2105 if (ref_length + 40 > isize ||
2106 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2107 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2108 free(buffer);
2109 return NULL;
2110 }
1cf58e72 2111 free(buffer);
40469ee9
JH
2112 /* Now we have the ID of the referred-to object in
2113 * actual_sha1. Check again. */
f4913f91 2114 }
f4913f91
JH
2115}
2116
ce9fbf16 2117static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2118 const char *type, unsigned char *sha1,
d65a16f6 2119 char *hdr, int *hdrlen)
d410c0f5
JH
2120{
2121 SHA_CTX c;
2122
2123 /* Generate the header */
d65a16f6 2124 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
2125
2126 /* Sha1.. */
2127 SHA1_Init(&c);
2128 SHA1_Update(&c, hdr, *hdrlen);
2129 SHA1_Update(&c, buf, len);
2130 SHA1_Final(sha1, &c);
d410c0f5
JH
2131}
2132
230f1322
LT
2133/*
2134 * Move the just written object into its final resting place
2135 */
839837b9 2136int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322 2137{
5723fe7e 2138 int ret = link(tmpfile, filename);
7ebb6fca
LT
2139
2140 /*
2141 * Coda hack - coda doesn't like cross-directory links,
2142 * so we fall back to a rename, which will mean that it
2143 * won't be able to check collisions, but that's not a
2144 * big deal.
2145 *
2146 * The same holds for FAT formatted media.
2147 *
2148 * When this succeeds, we just return 0. We have nothing
2149 * left to unlink.
2150 */
2151 if (ret && ret != EEXIST) {
2152 if (!rename(tmpfile, filename))
230f1322 2153 return 0;
9e48b389 2154 ret = errno;
230f1322
LT
2155 }
2156 unlink(tmpfile);
2157 if (ret) {
2158 if (ret != EEXIST) {
916d081b 2159 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
230f1322
LT
2160 }
2161 /* FIXME!!! Collision check here ? */
2162 }
2163
2164 return 0;
2165}
2166
4d548150
LT
2167static int write_buffer(int fd, const void *buf, size_t len)
2168{
d34cf19b 2169 if (write_in_full(fd, buf, len) < 0)
93822c22 2170 return error("file write error (%s)", strerror(errno));
4d548150
LT
2171 return 0;
2172}
2173
ce9fbf16 2174int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2175 unsigned char *sha1)
2176{
d65a16f6 2177 char hdr[32];
abdc3fc8
RS
2178 int hdrlen;
2179 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2180 return 0;
2181}
2182
e9039dd3
LT
2183/* Finalize a file on disk, and close it. */
2184static void close_sha1_file(int fd)
2185{
aafe9fba
LT
2186 if (fsync_object_files)
2187 fsync_or_die(fd, "sha1 file");
e9039dd3
LT
2188 fchmod(fd, 0444);
2189 if (close(fd) != 0)
2190 die("unable to write sha1 file");
2191}
2192
5723fe7e
LT
2193/* Size of directory component, including the ending '/' */
2194static inline int directory_size(const char *filename)
2195{
2196 const char *s = strrchr(filename, '/');
2197 if (!s)
2198 return 0;
2199 return s - filename + 1;
2200}
2201
2202/*
2203 * This creates a temporary file in the same directory as the final
2204 * 'filename'
2205 *
2206 * We want to avoid cross-directory filename renames, because those
2207 * can have problems on various filesystems (FAT, NFS, Coda).
2208 */
2209static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2210{
2211 int fd, dirlen = directory_size(filename);
2212
2213 if (dirlen + 20 > bufsiz) {
2214 errno = ENAMETOOLONG;
2215 return -1;
2216 }
2217 memcpy(buffer, filename, dirlen);
2218 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2219 fd = mkstemp(buffer);
2220 if (fd < 0 && dirlen) {
2221 /* Make sure the directory exists */
6ff6af62 2222 memcpy(buffer, filename, dirlen);
5723fe7e 2223 buffer[dirlen-1] = 0;
3bfaf018 2224 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
5723fe7e
LT
2225 return -1;
2226
2227 /* Try again */
2228 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2229 fd = mkstemp(buffer);
2230 }
2231 return fd;
2232}
2233
bbac7311
NP
2234static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2235 void *buf, unsigned long len, time_t mtime)
0fcfd160 2236{
bbac7311 2237 int fd, size, ret;
bf0f910d 2238 unsigned char *compressed;
0fcfd160 2239 z_stream stream;
706bc531 2240 char *filename;
aac17941 2241 static char tmpfile[PATH_MAX];
a44c9a5e 2242
972a9155 2243 filename = sha1_file_name(sha1);
5723fe7e 2244 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
aac17941 2245 if (fd < 0) {
916d081b
PB
2246 if (errno == EPERM)
2247 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2248 else
2249 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
aac17941
LT
2250 }
2251
0fcfd160
LT
2252 /* Set it up */
2253 memset(&stream, 0, sizeof(stream));
12f6c308 2254 deflateInit(&stream, zlib_compression_level);
93821bd9 2255 size = 8 + deflateBound(&stream, len+hdrlen);
812666c8 2256 compressed = xmalloc(size);
0fcfd160
LT
2257
2258 /* Compress it */
0fcfd160
LT
2259 stream.next_out = compressed;
2260 stream.avail_out = size;
a44c9a5e
LT
2261
2262 /* First header.. */
d65a16f6 2263 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2264 stream.avail_in = hdrlen;
726f852b
NP
2265 while (deflate(&stream, 0) == Z_OK)
2266 /* nothing */;
a44c9a5e
LT
2267
2268 /* Then the data itself.. */
2269 stream.next_in = buf;
2270 stream.avail_in = len;
ac54c277
LT
2271 ret = deflate(&stream, Z_FINISH);
2272 if (ret != Z_STREAM_END)
2273 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2274
2275 ret = deflateEnd(&stream);
2276 if (ret != Z_OK)
2277 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2278
0fcfd160
LT
2279 size = stream.total_out;
2280
4d548150
LT
2281 if (write_buffer(fd, compressed, size) < 0)
2282 die("unable to write sha1 file");
e9039dd3 2283 close_sha1_file(fd);
383f85b7 2284 free(compressed);
0fcfd160 2285
bbac7311
NP
2286 if (mtime) {
2287 struct utimbuf utb;
2288 utb.actime = mtime;
2289 utb.modtime = mtime;
2290 if (utime(tmpfile, &utb) < 0)
2291 warning("failed utime() on %s: %s",
2292 tmpfile, strerror(errno));
2293 }
2294
230f1322 2295 return move_temp_to_file(tmpfile, filename);
0fcfd160 2296}
8237b185 2297
bbac7311
NP
2298int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2299{
2300 unsigned char sha1[20];
2301 char hdr[32];
2302 int hdrlen;
2303
2304 /* Normally if we have it in the pack then we do not bother writing
2305 * it out into .git/objects/??/?{38} file.
2306 */
2307 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2308 if (returnsha1)
2309 hashcpy(returnsha1, sha1);
2310 if (has_sha1_file(sha1))
2311 return 0;
2312 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2313}
2314
2315int force_object_loose(const unsigned char *sha1, time_t mtime)
2316{
bbac7311
NP
2317 void *buf;
2318 unsigned long len;
2319 enum object_type type;
2320 char hdr[32];
2321 int hdrlen;
2322
c529d75a 2323 if (has_loose_object(sha1))
bbac7311
NP
2324 return 0;
2325 buf = read_packed_sha1(sha1, &type, &len);
2326 if (!buf)
2327 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2328 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2329 return write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2330}
2331
bf592c50
DB
2332int has_pack_index(const unsigned char *sha1)
2333{
2334 struct stat st;
2335 if (stat(sha1_pack_index_name(sha1), &st))
2336 return 0;
2337 return 1;
2338}
2339
2340int has_pack_file(const unsigned char *sha1)
2341{
2342 struct stat st;
2343 if (stat(sha1_pack_name(sha1), &st))
2344 return 0;
2345 return 1;
2346}
2347
106d710b 2348int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
dade09c2
LT
2349{
2350 struct pack_entry e;
106d710b 2351 return find_pack_entry(sha1, &e, ignore_packed);
dade09c2
LT
2352}
2353
8237b185
DB
2354int has_sha1_file(const unsigned char *sha1)
2355{
1f688557
JH
2356 struct pack_entry e;
2357
106d710b 2358 if (find_pack_entry(sha1, &e, NULL))
1f688557 2359 return 1;
c529d75a 2360 return has_loose_object(sha1);
8237b185 2361}
74400e71 2362
43df4f86
DP
2363static int index_mem(unsigned char *sha1, void *buf, size_t size,
2364 int write_object, enum object_type type, const char *path)
e7332f96 2365{
6c510bee 2366 int ret, re_allocated = 0;
74400e71 2367
7672db20 2368 if (!type)
edaec3fb 2369 type = OBJ_BLOB;
6c510bee
LT
2370
2371 /*
2372 * Convert blobs to git internal format
2373 */
43df4f86 2374 if ((type == OBJ_BLOB) && path) {
5ecd293d
PH
2375 struct strbuf nbuf;
2376 strbuf_init(&nbuf, 0);
21e5ad50
SP
2377 if (convert_to_git(path, buf, size, &nbuf,
2378 write_object ? safe_crlf : 0)) {
b315c5c0 2379 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
2380 re_allocated = 1;
2381 }
2382 }
2383
7672db20 2384 if (write_object)
edaec3fb 2385 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 2386 else
edaec3fb 2387 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 2388 if (re_allocated)
6c510bee 2389 free(buf);
43df4f86
DP
2390 return ret;
2391}
2392
2393int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2394 enum object_type type, const char *path)
2395{
2396 int ret;
2397 size_t size = xsize_t(st->st_size);
2398
2399 if (!S_ISREG(st->st_mode)) {
2400 struct strbuf sbuf;
2401 strbuf_init(&sbuf, 0);
2402 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2403 ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2404 type, path);
2405 else
2406 ret = -1;
2407 strbuf_release(&sbuf);
2408 } else if (size) {
2409 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2410 ret = index_mem(sha1, buf, size, write_object, type, path);
aac17941 2411 munmap(buf, size);
43df4f86
DP
2412 } else
2413 ret = index_mem(sha1, NULL, size, write_object, type, path);
2414 close(fd);
aac17941 2415 return ret;
74400e71 2416}
ec1fcc16
JH
2417
2418int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2419{
2420 int fd;
2421 char *target;
dc49cd76 2422 size_t len;
ec1fcc16
JH
2423
2424 switch (st->st_mode & S_IFMT) {
2425 case S_IFREG:
2426 fd = open(path, O_RDONLY);
2427 if (fd < 0)
2428 return error("open(\"%s\"): %s", path,
2429 strerror(errno));
53bca91a 2430 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
ec1fcc16
JH
2431 return error("%s: failed to insert into database",
2432 path);
2433 break;
2434 case S_IFLNK:
dc49cd76
SP
2435 len = xsize_t(st->st_size);
2436 target = xmalloc(len + 1);
2437 if (readlink(path, target, len + 1) != st->st_size) {
ec1fcc16
JH
2438 char *errstr = strerror(errno);
2439 free(target);
2440 return error("readlink(\"%s\"): %s", path,
2441 errstr);
2442 }
abdc3fc8 2443 if (!write_object)
dc49cd76
SP
2444 hash_sha1_file(target, len, blob_type, sha1);
2445 else if (write_sha1_file(target, len, blob_type, sha1))
ec1fcc16
JH
2446 return error("%s: failed to insert into database",
2447 path);
2448 free(target);
2449 break;
f35a6d3b
LT
2450 case S_IFDIR:
2451 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
2452 default:
2453 return error("%s: unsupported file type", path);
2454 }
2455 return 0;
2456}
a69e5429
JH
2457
2458int read_pack_header(int fd, struct pack_header *header)
2459{
c697ad14
HO
2460 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2461 /* "eof before pack header was fully read" */
2462 return PH_ERROR_EOF;
2463
a69e5429
JH
2464 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2465 /* "protocol error (pack signature mismatch detected)" */
2466 return PH_ERROR_PACK_SIGNATURE;
2467 if (!pack_version_ok(header->hdr_version))
2468 /* "protocol error (pack version unsupported)" */
2469 return PH_ERROR_PROTOCOL;
2470 return 0;
2471}