]> git.ipfire.org Git - thirdparty/git.git/blame - sha1_file.c
tests: shell negation portability fix
[thirdparty/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{
c2f493a4 253 struct stat st;
1494e038 254 const char *objdir = get_object_directory();
c2f493a4
MW
255 struct alternate_object_database *ent;
256 struct alternate_object_database *alt;
257 /* 43 = 40-byte + 2 '/' + terminating NUL */
258 int pfxlen = len;
259 int entlen = pfxlen + 43;
ccfd3e99 260 int base_len = -1;
d5a63b99 261
85dadc38 262 if (!is_absolute_path(entry) && relative_base) {
c2f493a4
MW
263 /* Relative alt-odb */
264 if (base_len < 0)
265 base_len = strlen(relative_base) + 1;
266 entlen += base_len;
267 pfxlen += base_len;
268 }
269 ent = xmalloc(sizeof(*ent) + entlen);
270
85dadc38 271 if (!is_absolute_path(entry) && relative_base) {
c2f493a4
MW
272 memcpy(ent->base, relative_base, base_len - 1);
273 ent->base[base_len - 1] = '/';
274 memcpy(ent->base + base_len, entry, len);
275 }
276 else
277 memcpy(ent->base, entry, pfxlen);
278
279 ent->name = ent->base + pfxlen + 1;
280 ent->base[pfxlen + 3] = '/';
281 ent->base[pfxlen] = ent->base[entlen-1] = 0;
282
283 /* Detect cases where alternate disappeared */
284 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
285 error("object directory %s does not exist; "
286 "check .git/objects/info/alternates.",
287 ent->base);
288 free(ent);
289 return -1;
290 }
291
292 /* Prevent the common mistake of listing the same
293 * thing twice, or object directory itself.
294 */
295 for (alt = alt_odb_list; alt; alt = alt->next) {
296 if (!memcmp(ent->base, alt->base, pfxlen)) {
297 free(ent);
298 return -1;
299 }
300 }
301 if (!memcmp(ent->base, objdir, pfxlen)) {
302 free(ent);
303 return -1;
304 }
305
306 /* add the alternate entry */
307 *alt_odb_tail = ent;
308 alt_odb_tail = &(ent->next);
309 ent->next = NULL;
310
311 /* recursively add alternates */
312 read_info_alternates(ent->base, depth + 1);
313
314 ent->base[pfxlen] = '/';
315
316 return 0;
317}
318
319static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
320 const char *relative_base, int depth)
321{
322 const char *cp, *last;
323
324 if (depth > 5) {
325 error("%s: ignoring alternate object stores, nesting too deep.",
326 relative_base);
327 return;
328 }
329
d5a63b99 330 last = alt;
9577e7e3
JH
331 while (last < ep) {
332 cp = last;
333 if (cp < ep && *cp == '#') {
334 while (cp < ep && *cp != sep)
335 cp++;
336 last = cp + 1;
337 continue;
338 }
c2f493a4
MW
339 while (cp < ep && *cp != sep)
340 cp++;
d5a63b99 341 if (last != cp) {
85dadc38 342 if (!is_absolute_path(last) && depth) {
c2f493a4
MW
343 error("%s: ignoring relative alternate object store %s",
344 relative_base, last);
345 } else {
346 link_alt_odb_entry(last, cp - last,
347 relative_base, depth);
1494e038 348 }
d5a63b99 349 }
9577e7e3 350 while (cp < ep && *cp == sep)
d5a63b99
JH
351 cp++;
352 last = cp;
9577e7e3 353 }
d5a63b99
JH
354}
355
c2f493a4 356static void read_info_alternates(const char * relative_base, int depth)
d5a63b99 357{
9577e7e3 358 char *map;
dc49cd76 359 size_t mapsz;
d5a63b99 360 struct stat st;
9cb18f56
JM
361 const char alt_file_name[] = "info/alternates";
362 /* Given that relative_base is no longer than PATH_MAX,
363 ensure that "path" has enough space to append "/", the
364 file name, "info/alternates", and a trailing NUL. */
365 char path[PATH_MAX + 1 + sizeof alt_file_name];
c2f493a4 366 int fd;
d5a63b99 367
9cb18f56 368 sprintf(path, "%s/%s", relative_base, alt_file_name);
d5a63b99
JH
369 fd = open(path, O_RDONLY);
370 if (fd < 0)
371 return;
372 if (fstat(fd, &st) || (st.st_size == 0)) {
373 close(fd);
9a217f2a 374 return;
ace1534d 375 }
dc49cd76
SP
376 mapsz = xsize_t(st.st_size);
377 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 378 close(fd);
d5a63b99 379
dc49cd76 380 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
c2f493a4 381
dc49cd76 382 munmap(map, mapsz);
ace1534d
JH
383}
384
bef70b22
DB
385void add_to_alternates_file(const char *reference)
386{
387 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
388 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), 1);
389 char *alt = mkpath("%s/objects\n", reference);
390 write_or_die(fd, alt, strlen(alt));
391 if (commit_lock_file(lock))
392 die("could not close alternates file");
393 if (alt_odb_tail)
394 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
395}
396
c2f493a4
MW
397void prepare_alt_odb(void)
398{
554fe20d 399 const char *alt;
c2f493a4 400
7dc24aa5
SP
401 if (alt_odb_tail)
402 return;
403
c2f493a4
MW
404 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
405 if (!alt) alt = "";
406
c2f493a4 407 alt_odb_tail = &alt_odb_list;
80ba074f 408 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
409
410 read_info_alternates(get_object_directory(), 0);
411}
412
c529d75a 413static int has_loose_object(const unsigned char *sha1)
ace1534d 414{
ace1534d 415 char *name = sha1_file_name(sha1);
d5a63b99 416 struct alternate_object_database *alt;
ace1534d 417
c529d75a
LT
418 if (!access(name, F_OK))
419 return 1;
9a217f2a 420 prepare_alt_odb();
d5a63b99
JH
421 for (alt = alt_odb_list; alt; alt = alt->next) {
422 name = alt->name;
ace1534d 423 fill_sha1_path(name, sha1);
c529d75a
LT
424 if (!access(alt->base, F_OK))
425 return 1;
ace1534d 426 }
c529d75a 427 return 0;
ace1534d
JH
428}
429
60bb8b14 430static unsigned int pack_used_ctr;
a53128b6
SP
431static unsigned int pack_mmap_calls;
432static unsigned int peak_pack_open_windows;
433static unsigned int pack_open_windows;
434static size_t peak_pack_mapped;
60bb8b14 435static size_t pack_mapped;
9a217f2a 436struct packed_git *packed_git;
1f688557 437
b79d18c9 438void pack_report(void)
a53128b6
SP
439{
440 fprintf(stderr,
e05db0fd
PR
441 "pack_report: getpagesize() = %10" SZ_FMT "\n"
442 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
443 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
444 sz_fmt(getpagesize()),
445 sz_fmt(packed_git_window_size),
446 sz_fmt(packed_git_limit));
a53128b6
SP
447 fprintf(stderr,
448 "pack_report: pack_used_ctr = %10u\n"
449 "pack_report: pack_mmap_calls = %10u\n"
450 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
451 "pack_report: pack_mapped = "
452 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
453 pack_used_ctr,
454 pack_mmap_calls,
455 pack_open_windows, peak_pack_open_windows,
9e42d6a1 456 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
457}
458
42873078 459static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
460{
461 void *idx_map;
42873078 462 struct pack_idx_header *hdr;
dc49cd76 463 size_t idx_size;
74e34e1f 464 uint32_t version, nr, i, *index;
e93ec6f9 465 int fd = open(path, O_RDONLY);
1f688557 466 struct stat st;
42873078 467
1f688557
JH
468 if (fd < 0)
469 return -1;
470 if (fstat(fd, &st)) {
471 close(fd);
472 return -1;
473 }
dc49cd76 474 idx_size = xsize_t(st.st_size);
2d88451b
SP
475 if (idx_size < 4 * 256 + 20 + 20) {
476 close(fd);
477 return error("index file %s is too small", path);
478 }
c4712e45 479 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 480 close(fd);
1f688557 481
42873078
NP
482 hdr = idx_map;
483 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
484 version = ntohl(hdr->idx_version);
485 if (version < 2 || version > 2) {
486 munmap(idx_map, idx_size);
6e1c2344 487 return error("index file %s is version %"PRIu32
74e34e1f
NP
488 " and is not supported by this binary"
489 " (try upgrading GIT to a newer version)",
490 path, version);
491 }
492 } else
493 version = 1;
df1b059d 494
1f688557 495 nr = 0;
42873078 496 index = idx_map;
74e34e1f
NP
497 if (version > 1)
498 index += 2; /* skip index header */
1f688557 499 for (i = 0; i < 256; i++) {
326bf396 500 uint32_t n = ntohl(index[i]);
2d88451b
SP
501 if (n < nr) {
502 munmap(idx_map, idx_size);
df1b059d 503 return error("non-monotonic index %s", path);
2d88451b 504 }
1f688557
JH
505 nr = n;
506 }
507
74e34e1f
NP
508 if (version == 1) {
509 /*
510 * Total size:
511 * - 256 index entries 4 bytes each
512 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
513 * - 20-byte SHA1 of the packfile
514 * - 20-byte SHA1 file checksum
515 */
516 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
517 munmap(idx_map, idx_size);
eef427a0 518 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
519 }
520 } else if (version == 2) {
521 /*
522 * Minimum size:
523 * - 8 bytes of header
524 * - 256 index entries 4 bytes each
525 * - 20-byte sha1 entry * nr
526 * - 4-byte crc entry * nr
527 * - 4-byte offset entry * nr
528 * - 20-byte SHA1 of the packfile
529 * - 20-byte SHA1 file checksum
530 * And after the 4-byte offset table might be a
531 * variable sized table containing 8-byte entries
532 * for offsets larger than 2^31.
533 */
534 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
535 unsigned long max_size = min_size;
536 if (nr)
537 max_size += (nr - 1)*8;
538 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 539 munmap(idx_map, idx_size);
eef427a0 540 return error("wrong index v2 file size in %s", path);
74e34e1f 541 }
7109c889
JH
542 if (idx_size != min_size &&
543 /*
544 * make sure we can deal with large pack offsets.
545 * 31-bit signed offset won't be enough, neither
546 * 32-bit unsigned one will be.
547 */
548 (sizeof(off_t) <= 4)) {
549 munmap(idx_map, idx_size);
550 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 551 }
2d88451b 552 }
1f688557 553
74e34e1f 554 p->index_version = version;
42873078
NP
555 p->index_data = idx_map;
556 p->index_size = idx_size;
57059091 557 p->num_objects = nr;
1f688557
JH
558 return 0;
559}
560
bc8e478a 561int open_pack_index(struct packed_git *p)
d079837e
SP
562{
563 char *idx_name;
564 int ret;
565
566 if (p->index_data)
567 return 0;
568
569 idx_name = xstrdup(p->pack_name);
570 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
571 ret = check_packed_git_idx(idx_name, p);
572 free(idx_name);
573 return ret;
574}
575
11daf39b
SP
576static void scan_windows(struct packed_git *p,
577 struct packed_git **lru_p,
578 struct pack_window **lru_w,
579 struct pack_window **lru_l)
1f688557 580{
11daf39b
SP
581 struct pack_window *w, *w_l;
582
583 for (w_l = NULL, w = p->windows; w; w = w->next) {
584 if (!w->inuse_cnt) {
585 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
586 *lru_p = p;
587 *lru_w = w;
588 *lru_l = w_l;
54044bf8 589 }
54044bf8 590 }
11daf39b 591 w_l = w;
f9253394 592 }
11daf39b
SP
593}
594
d1efefa4 595static int unuse_one_window(struct packed_git *current, int keep_fd)
11daf39b
SP
596{
597 struct packed_git *p, *lru_p = NULL;
598 struct pack_window *lru_w = NULL, *lru_l = NULL;
599
600 if (current)
601 scan_windows(current, &lru_p, &lru_w, &lru_l);
602 for (p = packed_git; p; p = p->next)
603 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
604 if (lru_p) {
605 munmap(lru_w->base, lru_w->len);
606 pack_mapped -= lru_w->len;
607 if (lru_l)
608 lru_l->next = lru_w->next;
609 else {
610 lru_p->windows = lru_w->next;
d1efefa4 611 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
54044bf8
SP
612 close(lru_p->pack_fd);
613 lru_p->pack_fd = -1;
614 }
615 }
616 free(lru_w);
a53128b6 617 pack_open_windows--;
54044bf8
SP
618 return 1;
619 }
620 return 0;
f9253394
JH
621}
622
d1efefa4 623void release_pack_memory(size_t need, int fd)
97bfeb34
SP
624{
625 size_t cur = pack_mapped;
d1efefa4 626 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
97bfeb34
SP
627 ; /* nothing */
628}
629
c9ced051
SP
630void close_pack_windows(struct packed_git *p)
631{
632 while (p->windows) {
633 struct pack_window *w = p->windows;
634
635 if (w->inuse_cnt)
636 die("pack '%s' still has open windows to it",
637 p->pack_name);
638 munmap(w->base, w->len);
639 pack_mapped -= w->len;
640 pack_open_windows--;
641 p->windows = w->next;
642 free(w);
643 }
644}
645
03e79c88 646void unuse_pack(struct pack_window **w_cursor)
f9253394 647{
03e79c88
SP
648 struct pack_window *w = *w_cursor;
649 if (w) {
650 w->inuse_cnt--;
651 *w_cursor = NULL;
652 }
1f688557
JH
653}
654
3cf8b462
SP
655/*
656 * Do not call this directly as this leaks p->pack_fd on error return;
657 * call open_packed_git() instead.
658 */
659static int open_packed_git_1(struct packed_git *p)
1f688557 660{
9bc879c1
SP
661 struct stat st;
662 struct pack_header hdr;
663 unsigned char sha1[20];
664 unsigned char *idx_sha1;
2c039da8 665 long fd_flag;
9bc879c1 666
d079837e
SP
667 if (!p->index_data && open_pack_index(p))
668 return error("packfile %s index unavailable", p->pack_name);
669
9bc879c1
SP
670 p->pack_fd = open(p->pack_name, O_RDONLY);
671 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 672 return -1;
9bc879c1
SP
673
674 /* If we created the struct before we had the pack we lack size. */
bf592c50 675 if (!p->pack_size) {
bf592c50 676 if (!S_ISREG(st.st_mode))
072db278 677 return error("packfile %s not a regular file", p->pack_name);
bf592c50 678 p->pack_size = st.st_size;
9bc879c1 679 } else if (p->pack_size != st.st_size)
072db278 680 return error("packfile %s size changed", p->pack_name);
9bc879c1 681
2c039da8
JH
682 /* We leave these file descriptors open with sliding mmap;
683 * there is no point keeping them open across exec(), though.
684 */
685 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
686 if (fd_flag < 0)
072db278 687 return error("cannot determine file descriptor flags");
2c039da8
JH
688 fd_flag |= FD_CLOEXEC;
689 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 690 return error("cannot set FD_CLOEXEC");
2c039da8 691
9bc879c1 692 /* Verify we recognize this pack file format. */
e6e2bd62 693 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 694 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 695 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 696 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 697 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
698 return error("packfile %s is version %"PRIu32" and not"
699 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
700 p->pack_name, ntohl(hdr.hdr_version));
701
702 /* Verify the pack matches its index. */
57059091 703 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
704 return error("packfile %s claims to have %"PRIu32" objects"
705 " while index indicates %"PRIu32" objects",
57059091
NP
706 p->pack_name, ntohl(hdr.hdr_entries),
707 p->num_objects);
9bc879c1 708 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 709 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 710 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 711 return error("packfile %s signature is unavailable", p->pack_name);
42873078 712 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 713 if (hashcmp(sha1, idx_sha1))
072db278
SP
714 return error("packfile %s does not match index", p->pack_name);
715 return 0;
9bc879c1
SP
716}
717
3cf8b462
SP
718static int open_packed_git(struct packed_git *p)
719{
720 if (!open_packed_git_1(p))
721 return 0;
722 if (p->pack_fd != -1) {
723 close(p->pack_fd);
724 p->pack_fd = -1;
725 }
726 return -1;
727}
728
c4001d92 729static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
730{
731 /* We must promise at least 20 bytes (one hash) after the
732 * offset is available from this window, otherwise the offset
733 * is not actually in this window and a different window (which
734 * has that one hash excess) must be used. This is to support
735 * the object header and delta base parsing routines below.
736 */
737 off_t win_off = win->offset;
738 return win_off <= offset
739 && (offset + 20) <= (win_off + win->len);
740}
741
03e79c88
SP
742unsigned char* use_pack(struct packed_git *p,
743 struct pack_window **w_cursor,
c4001d92 744 off_t offset,
03e79c88 745 unsigned int *left)
9bc879c1 746{
60bb8b14 747 struct pack_window *win = *w_cursor;
03e79c88 748
072db278
SP
749 if (p->pack_fd == -1 && open_packed_git(p))
750 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
751
752 /* Since packfiles end in a hash of their content and its
753 * pointless to ask for an offset into the middle of that
754 * hash, and the in_window function above wouldn't match
755 * don't allow an offset too close to the end of the file.
756 */
757 if (offset > (p->pack_size - 20))
758 die("offset beyond end of packfile (truncated pack?)");
759
760 if (!win || !in_window(win, offset)) {
761 if (win)
762 win->inuse_cnt--;
763 for (win = p->windows; win; win = win->next) {
764 if (in_window(win, offset))
765 break;
766 }
767 if (!win) {
78a28df9 768 size_t window_align = packed_git_window_size / 2;
dc49cd76 769 off_t len;
60bb8b14 770 win = xcalloc(1, sizeof(*win));
78a28df9 771 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
772 len = p->pack_size - win->offset;
773 if (len > packed_git_window_size)
774 len = packed_git_window_size;
775 win->len = (size_t)len;
60bb8b14 776 pack_mapped += win->len;
11daf39b 777 while (packed_git_limit < pack_mapped
d1efefa4 778 && unuse_one_window(p, p->pack_fd))
60bb8b14 779 ; /* nothing */
c4712e45 780 win->base = xmmap(NULL, win->len,
60bb8b14
SP
781 PROT_READ, MAP_PRIVATE,
782 p->pack_fd, win->offset);
783 if (win->base == MAP_FAILED)
73b4e4be
SP
784 die("packfile %s cannot be mapped: %s",
785 p->pack_name,
786 strerror(errno));
a53128b6
SP
787 pack_mmap_calls++;
788 pack_open_windows++;
789 if (pack_mapped > peak_pack_mapped)
790 peak_pack_mapped = pack_mapped;
791 if (pack_open_windows > peak_pack_open_windows)
792 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
793 win->next = p->windows;
794 p->windows = win;
795 }
1f688557 796 }
03e79c88
SP
797 if (win != *w_cursor) {
798 win->last_used = pack_used_ctr++;
799 win->inuse_cnt++;
800 *w_cursor = win;
801 }
60bb8b14 802 offset -= win->offset;
03e79c88 803 if (left)
dc49cd76 804 *left = win->len - xsize_t(offset);
03e79c88 805 return win->base + offset;
1f688557
JH
806}
807
27d69a46
NP
808static struct packed_git *alloc_packed_git(int extra)
809{
810 struct packed_git *p = xmalloc(sizeof(*p) + extra);
811 memset(p, 0, sizeof(*p));
812 p->pack_fd = -1;
813 return p;
814}
815
42873078 816struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557
JH
817{
818 struct stat st;
27d69a46 819 struct packed_git *p = alloc_packed_git(path_len + 2);
1f688557 820
42873078
NP
821 /*
822 * Make sure a corresponding .pack file exists and that
823 * the index looks sane.
824 */
825 path_len -= strlen(".idx");
27d69a46
NP
826 if (path_len < 1) {
827 free(p);
1f688557 828 return NULL;
27d69a46 829 }
42873078
NP
830 memcpy(p->pack_name, path, path_len);
831 strcpy(p->pack_name + path_len, ".pack");
d079837e 832 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 833 free(p);
1f688557
JH
834 return NULL;
835 }
42873078 836
1f688557
JH
837 /* ok, it looks sane as far as we can check without
838 * actually mapping the pack file.
839 */
1f688557 840 p->pack_size = st.st_size;
9d835df2 841 p->pack_local = local;
b867092f 842 p->mtime = st.st_mtime;
42873078
NP
843 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
844 hashclr(p->sha1);
1f688557
JH
845 return p;
846}
847
bf592c50 848struct packed_git *parse_pack_index(unsigned char *sha1)
c508df5e 849{
79c6dca4 850 const char *idx_path = sha1_pack_index_name(sha1);
42873078 851 const char *path = sha1_pack_name(sha1);
27d69a46 852 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
bf592c50 853
27d69a46
NP
854 strcpy(p->pack_name, path);
855 hashcpy(p->sha1, sha1);
42873078
NP
856 if (check_packed_git_idx(idx_path, p)) {
857 free(p);
bf592c50 858 return NULL;
42873078 859 }
bf592c50 860
bf592c50
DB
861 return p;
862}
863
864void install_packed_git(struct packed_git *pack)
865{
866 pack->next = packed_git;
867 packed_git = pack;
868}
869
9d835df2 870static void prepare_packed_git_one(char *objdir, int local)
1f688557 871{
9cb18f56
JM
872 /* Ensure that this buffer is large enough so that we can
873 append "/pack/" without clobbering the stack even if
874 strlen(objdir) were PATH_MAX. */
875 char path[PATH_MAX + 1 + 4 + 1 + 1];
1f688557
JH
876 int len;
877 DIR *dir;
878 struct dirent *de;
879
880 sprintf(path, "%s/pack", objdir);
881 len = strlen(path);
882 dir = opendir(path);
b5b16990 883 if (!dir) {
26125f6b 884 if (errno != ENOENT)
bd2afde8 885 error("unable to open object pack directory: %s: %s",
26125f6b 886 path, strerror(errno));
1f688557 887 return;
b5b16990 888 }
1f688557
JH
889 path[len++] = '/';
890 while ((de = readdir(dir)) != NULL) {
891 int namelen = strlen(de->d_name);
892 struct packed_git *p;
893
5bb1cda5 894 if (!has_extension(de->d_name, ".idx"))
1f688557
JH
895 continue;
896
9cb18f56
JM
897 if (len + namelen + 1 > sizeof(path))
898 continue;
899
54a15a8d 900 /* Don't reopen a pack we already have. */
1f688557 901 strcpy(path + len, de->d_name);
86f7780c
JK
902 for (p = packed_git; p; p = p->next) {
903 if (!memcmp(path, p->pack_name, len + namelen - 4))
904 break;
905 }
906 if (p)
907 continue;
54a15a8d
SP
908 /* See if it really is a valid .idx file with corresponding
909 * .pack file that we can map.
910 */
9d835df2 911 p = add_packed_git(path, len + namelen, local);
1f688557
JH
912 if (!p)
913 continue;
625e9421 914 install_packed_git(p);
1f688557 915 }
5b35bcd5 916 closedir(dir);
1f688557
JH
917}
918
b867092f
JH
919static int sort_pack(const void *a_, const void *b_)
920{
921 struct packed_git *a = *((struct packed_git **)a_);
922 struct packed_git *b = *((struct packed_git **)b_);
923 int st;
924
925 /*
926 * Local packs tend to contain objects specific to our
927 * variant of the project than remote ones. In addition,
928 * remote ones could be on a network mounted filesystem.
929 * Favor local ones for these reasons.
930 */
931 st = a->pack_local - b->pack_local;
932 if (st)
933 return -st;
934
935 /*
936 * Younger packs tend to contain more recent objects,
937 * and more recent objects tend to get accessed more
938 * often.
939 */
940 if (a->mtime < b->mtime)
941 return 1;
942 else if (a->mtime == b->mtime)
943 return 0;
944 return -1;
945}
946
947static void rearrange_packed_git(void)
948{
949 struct packed_git **ary, *p;
950 int i, n;
951
952 for (n = 0, p = packed_git; p; p = p->next)
953 n++;
954 if (n < 2)
955 return;
956
957 /* prepare an array of packed_git for easier sorting */
958 ary = xcalloc(n, sizeof(struct packed_git *));
959 for (n = 0, p = packed_git; p; p = p->next)
960 ary[n++] = p;
961
962 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
963
964 /* link them back again */
965 for (i = 0; i < n - 1; i++)
966 ary[i]->next = ary[i + 1];
967 ary[n - 1]->next = NULL;
968 packed_git = ary[0];
969
970 free(ary);
971}
972
637cdd9d 973static int prepare_packed_git_run_once = 0;
9a217f2a 974void prepare_packed_git(void)
1f688557 975{
d5a63b99 976 struct alternate_object_database *alt;
1f688557 977
637cdd9d 978 if (prepare_packed_git_run_once)
1f688557 979 return;
9d835df2 980 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 981 prepare_alt_odb();
d5a63b99 982 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 983 alt->name[-1] = 0;
9d835df2 984 prepare_packed_git_one(alt->base, 0);
1494e038 985 alt->name[-1] = '/';
1f688557 986 }
b867092f 987 rearrange_packed_git();
637cdd9d
JK
988 prepare_packed_git_run_once = 1;
989}
990
fc04c412 991void reprepare_packed_git(void)
637cdd9d 992{
4b480c67 993 discard_revindex();
637cdd9d
JK
994 prepare_packed_git_run_once = 0;
995 prepare_packed_git();
1f688557
JH
996}
997
8eca0b47
NP
998static void mark_bad_packed_object(struct packed_git *p,
999 const unsigned char *sha1)
1000{
1001 unsigned i;
1002 for (i = 0; i < p->num_bad_objects; i++)
1003 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1004 return;
1005 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1006 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1007 p->num_bad_objects++;
1008}
1009
ac939109
NP
1010static int has_packed_and_bad(const unsigned char *sha1)
1011{
1012 struct packed_git *p;
1013 unsigned i;
1014
1015 for (p = packed_git; p; p = p->next)
1016 for (i = 0; i < p->num_bad_objects; i++)
1017 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1018 return 1;
1019 return 0;
1020}
1021
5d6ccf5c 1022int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
0fcfd160
LT
1023{
1024 unsigned char real_sha1[20];
7cfb5f36 1025 hash_sha1_file(map, size, type, real_sha1);
a89fccd2 1026 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1027}
1028
44d1c19e
LT
1029static int git_open_noatime(const char *name)
1030{
1031 static int sha1_file_open_flag = O_NOATIME;
1032 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1033
1034 /* Might the failure be due to O_NOATIME? */
1035 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1036 fd = open(name, O_RDONLY);
1037 if (fd >= 0)
1038 sha1_file_open_flag = 0;
1039 }
1040 return fd;
1041}
1042
1043static int open_sha1_file(const unsigned char *sha1)
1044{
1045 int fd;
1046 char *name = sha1_file_name(sha1);
1047 struct alternate_object_database *alt;
1048
1049 fd = git_open_noatime(name);
1050 if (fd >= 0)
1051 return fd;
1052
1053 prepare_alt_odb();
1054 errno = ENOENT;
1055 for (alt = alt_odb_list; alt; alt = alt->next) {
1056 name = alt->name;
1057 fill_sha1_path(name, sha1);
1058 fd = git_open_noatime(alt->base);
1059 if (fd >= 0)
1060 return fd;
1061 }
1062 return -1;
1063}
1064
4175e9e3 1065static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1066{
0fcfd160 1067 void *map;
144bde78 1068 int fd;
ace1534d 1069
44d1c19e
LT
1070 fd = open_sha1_file(sha1);
1071 map = NULL;
1072 if (fd >= 0) {
1073 struct stat st;
0fcfd160 1074
44d1c19e
LT
1075 if (!fstat(fd, &st)) {
1076 *size = xsize_t(st.st_size);
1077 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1078 }
44d1c19e 1079 close(fd);
0fcfd160 1080 }
0fcfd160
LT
1081 return map;
1082}
1083
726f852b 1084static int legacy_loose_object(unsigned char *map)
bb6b8e4f
JH
1085{
1086 unsigned int word;
1087
1088 /*
1089 * Is it a zlib-compressed buffer? If so, the first byte
1090 * must be 0x78 (15-bit window size, deflated), and the
1091 * first 16-bit word is evenly divisible by 31
1092 */
1093 word = (map[0] << 8) + map[1];
1094 if (map[0] == 0x78 && !(word % 31))
1095 return 1;
1096 else
1097 return 0;
1098}
1099
72518e9c 1100unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1101{
ad1ed5ee 1102 unsigned shift;
93821bd9 1103 unsigned char c;
93821bd9 1104 unsigned long size;
ad1ed5ee
JH
1105 unsigned long used = 0;
1106
1107 c = buf[used++];
1108 *type = (c >> 4) & 7;
1109 size = c & 15;
1110 shift = 4;
1111 while (c & 0x80) {
1112 if (len <= used)
1113 return 0;
1114 if (sizeof(long) * 8 <= shift)
1115 return 0;
1116 c = buf[used++];
1117 size += (c & 0x7f) << shift;
1118 shift += 7;
1119 }
1120 *sizep = size;
1121 return used;
1122}
1123
1124static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1125{
1126 unsigned long size, used;
1127 static const char valid_loose_object_type[8] = {
1128 0, /* OBJ_EXT */
1129 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1130 0, /* "delta" and others are invalid in a loose object */
93821bd9 1131 };
ad1ed5ee 1132 enum object_type type;
93821bd9 1133
c4483576
LT
1134 /* Get the data stream */
1135 memset(stream, 0, sizeof(*stream));
1136 stream->next_in = map;
1137 stream->avail_in = mapsize;
1138 stream->next_out = buffer;
93821bd9
LT
1139 stream->avail_out = bufsiz;
1140
bb6b8e4f 1141 if (legacy_loose_object(map)) {
93821bd9
LT
1142 inflateInit(stream);
1143 return inflate(stream, 0);
1144 }
1145
726f852b
NP
1146
1147 /*
1148 * There used to be a second loose object header format which
1149 * was meant to mimic the in-pack format, allowing for direct
1150 * copy of the object data. This format turned up not to be
1151 * really worth it and we don't write it any longer. But we
1152 * can still read it.
1153 */
ad1ed5ee
JH
1154 used = unpack_object_header_gently(map, mapsize, &type, &size);
1155 if (!used || !valid_loose_object_type[type])
93821bd9 1156 return -1;
ad1ed5ee
JH
1157 map += used;
1158 mapsize -= used;
93821bd9
LT
1159
1160 /* Set up the stream for the rest.. */
1161 stream->next_in = map;
1162 stream->avail_in = mapsize;
c4483576 1163 inflateInit(stream);
93821bd9
LT
1164
1165 /* And generate the fake traditional header */
ad1ed5ee 1166 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
df843662 1167 typename(type), size);
93821bd9 1168 return 0;
c4483576
LT
1169}
1170
7efbff75 1171static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1172{
1173 int bytes = strlen(buffer) + 1;
d565b341 1174 unsigned char *buf = xmalloc(1+size);
93821bd9 1175 unsigned long n;
7efbff75 1176 int status = Z_OK;
5180cacc 1177
93821bd9
LT
1178 n = stream->total_out - bytes;
1179 if (n > size)
1180 n = size;
1181 memcpy(buf, (char *) buffer + bytes, n);
1182 bytes = n;
456cdf6e
LT
1183 if (bytes <= size) {
1184 /*
1185 * The above condition must be (bytes <= size), not
1186 * (bytes < size). In other words, even though we
1187 * expect no more output and set avail_out to zer0,
1188 * the input zlib stream may have bytes that express
1189 * "this concludes the stream", and we *do* want to
1190 * eat that input.
1191 *
1192 * Otherwise we would not be able to test that we
1193 * consumed all the input to reach the expected size;
1194 * we also want to check that zlib tells us that all
1195 * went well with status == Z_STREAM_END at the end.
1196 */
5180cacc
LT
1197 stream->next_out = buf + bytes;
1198 stream->avail_out = size - bytes;
7efbff75
JH
1199 while (status == Z_OK)
1200 status = inflate(stream, Z_FINISH);
5180cacc
LT
1201 }
1202 buf[size] = 0;
456cdf6e 1203 if (status == Z_STREAM_END && !stream->avail_in) {
7efbff75
JH
1204 inflateEnd(stream);
1205 return buf;
1206 }
1207
1208 if (status < 0)
1209 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1210 else if (stream->avail_in)
1211 error("garbage at end of loose object '%s'",
1212 sha1_to_hex(sha1));
1213 free(buf);
1214 return NULL;
5180cacc
LT
1215}
1216
1217/*
1218 * We used to just use "sscanf()", but that's actually way
1219 * too permissive for what we want to check. So do an anal
1220 * object header parse by hand.
1221 */
21666f1a 1222static int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1223{
21666f1a 1224 char type[10];
5180cacc
LT
1225 int i;
1226 unsigned long size;
1227
1228 /*
a6080a0a 1229 * The type can be at most ten bytes (including the
5180cacc 1230 * terminating '\0' that we add), and is followed by
21666f1a 1231 * a space.
5180cacc 1232 */
21666f1a 1233 i = 0;
5180cacc
LT
1234 for (;;) {
1235 char c = *hdr++;
1236 if (c == ' ')
1237 break;
21666f1a
NP
1238 type[i++] = c;
1239 if (i >= sizeof(type))
5180cacc 1240 return -1;
5180cacc 1241 }
21666f1a 1242 type[i] = 0;
5180cacc
LT
1243
1244 /*
1245 * The length must follow immediately, and be in canonical
1246 * decimal format (ie "010" is not valid).
1247 */
1248 size = *hdr++ - '0';
1249 if (size > 9)
1250 return -1;
1251 if (size) {
1252 for (;;) {
1253 unsigned long c = *hdr - '0';
1254 if (c > 9)
1255 break;
1256 hdr++;
1257 size = size * 10 + c;
1258 }
1259 }
1260 *sizep = size;
1261
1262 /*
1263 * The length must be followed by a zero byte
1264 */
21666f1a 1265 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1266}
1267
7efbff75 1268static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1269{
5180cacc 1270 int ret;
0fcfd160 1271 z_stream stream;
5180cacc 1272 char hdr[8192];
0fcfd160 1273
5180cacc 1274 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1275 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1276 return NULL;
1277
7efbff75 1278 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1279}
1280
54dab52a
NP
1281unsigned long get_size_from_delta(struct packed_git *p,
1282 struct pack_window **w_curs,
1283 off_t curpos)
1284{
1285 const unsigned char *data;
1286 unsigned char delta_head[20], *in;
1287 z_stream stream;
1288 int st;
1289
1290 memset(&stream, 0, sizeof(stream));
1291 stream.next_out = delta_head;
1292 stream.avail_out = sizeof(delta_head);
1293
1294 inflateInit(&stream);
1295 do {
1296 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1297 stream.next_in = in;
1298 st = inflate(&stream, Z_FINISH);
1299 curpos += stream.next_in - in;
1300 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1301 stream.total_out < sizeof(delta_head));
1302 inflateEnd(&stream);
1303 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1304 die("delta data unpack-initial failed");
1305
1306 /* Examine the initial part of the delta to figure out
1307 * the result size.
1308 */
1309 data = delta_head;
1310
1311 /* ignore base size */
1312 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1313
1314 /* Read the result size */
1315 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1316}
1317
c4001d92 1318static off_t get_delta_base(struct packed_git *p,
03e79c88 1319 struct pack_window **w_curs,
c4001d92 1320 off_t *curpos,
21666f1a 1321 enum object_type type,
c4001d92 1322 off_t delta_obj_offset)
eb32d236 1323{
2b87c45b 1324 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1325 off_t base_offset;
eb32d236 1326
8d8a4ea5
SP
1327 /* use_pack() assured us we have [base_info, base_info + 20)
1328 * as a range that we can look at without walking off the
1329 * end of the mapped window. Its actually the hash size
1330 * that is assured. An OFS_DELTA longer than the hash size
1331 * is stupid, as then a REF_DELTA would be smaller to store.
1332 */
21666f1a 1333 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1334 unsigned used = 0;
1335 unsigned char c = base_info[used++];
1336 base_offset = c & 127;
1337 while (c & 128) {
1338 base_offset += 1;
8723f216 1339 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1340 return 0; /* overflow */
eb32d236
NP
1341 c = base_info[used++];
1342 base_offset = (base_offset << 7) + (c & 127);
1343 }
1344 base_offset = delta_obj_offset - base_offset;
1345 if (base_offset >= delta_obj_offset)
8eca0b47 1346 return 0; /* out of bound */
2b87c45b 1347 *curpos += used;
21666f1a 1348 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1349 /* The base entry _must_ be in the same pack */
1350 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1351 *curpos += 20;
eb32d236
NP
1352 } else
1353 die("I am totally screwed");
2b87c45b 1354 return base_offset;
eb32d236
NP
1355}
1356
f3bf9224 1357/* forward declaration for a mutually recursive function */
c4001d92 1358static int packed_object_info(struct packed_git *p, off_t offset,
21666f1a 1359 unsigned long *sizep);
f3bf9224 1360
43057304 1361static int packed_delta_info(struct packed_git *p,
03e79c88 1362 struct pack_window **w_curs,
c4001d92 1363 off_t curpos,
21666f1a 1364 enum object_type type,
c4001d92 1365 off_t obj_offset,
43057304 1366 unsigned long *sizep)
5db47c2b 1367{
c4001d92 1368 off_t base_offset;
f3bf9224 1369
21666f1a
NP
1370 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1371 type = packed_object_info(p, base_offset, NULL);
f3bf9224 1372
c62266f3
JH
1373 /* We choose to only get the type of the base object and
1374 * ignore potentially corrupt pack file that expects the delta
1375 * based on a base with a wrong size. This saves tons of
1376 * inflate() calls.
1377 */
54dab52a
NP
1378 if (sizep)
1379 *sizep = get_size_from_delta(p, w_curs, curpos);
21666f1a
NP
1380
1381 return type;
5db47c2b
JH
1382}
1383
2b87c45b
NP
1384static int unpack_object_header(struct packed_git *p,
1385 struct pack_window **w_curs,
c4001d92 1386 off_t *curpos,
2b87c45b 1387 unsigned long *sizep)
a733cb60 1388{
03e79c88
SP
1389 unsigned char *base;
1390 unsigned int left;
ad1ed5ee 1391 unsigned long used;
2b87c45b 1392 enum object_type type;
a733cb60 1393
8d8a4ea5
SP
1394 /* use_pack() assures us we have [base, base + 20) available
1395 * as a range that we can look at at. (Its actually the hash
3dff5379 1396 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1397 * the maximum deflated object size is 2^137, which is just
1398 * insane, so we know won't exceed what we have been given.
1399 */
2b87c45b
NP
1400 base = use_pack(p, w_curs, *curpos, &left);
1401 used = unpack_object_header_gently(base, left, &type, sizep);
ad1ed5ee
JH
1402 if (!used)
1403 die("object offset outside of pack file");
2b87c45b 1404 *curpos += used;
ad1ed5ee 1405
2b87c45b 1406 return type;
a733cb60
LT
1407}
1408
21666f1a 1409const char *packed_object_info_detail(struct packed_git *p,
c4001d92 1410 off_t obj_offset,
21666f1a
NP
1411 unsigned long *size,
1412 unsigned long *store_size,
1413 unsigned int *delta_chain_length,
1414 unsigned char *base_sha1)
ad8c80a5 1415{
03e79c88 1416 struct pack_window *w_curs = NULL;
c4001d92
SP
1417 off_t curpos;
1418 unsigned long dummy;
43057304 1419 unsigned char *next_sha1;
21666f1a 1420 enum object_type type;
70f5d5d3 1421 struct revindex_entry *revidx;
ad8c80a5 1422
43057304 1423 *delta_chain_length = 0;
2b87c45b 1424 curpos = obj_offset;
21666f1a 1425 type = unpack_object_header(p, &w_curs, &curpos, size);
43057304 1426
70f5d5d3
NP
1427 revidx = find_pack_revindex(p, obj_offset);
1428 *store_size = revidx[1].offset - obj_offset;
1429
43057304 1430 for (;;) {
21666f1a 1431 switch (type) {
43057304 1432 default:
08a19d87 1433 die("pack %s contains unknown object type %d",
21666f1a 1434 p->pack_name, type);
43057304
NP
1435 case OBJ_COMMIT:
1436 case OBJ_TREE:
1437 case OBJ_BLOB:
1438 case OBJ_TAG:
03e79c88 1439 unuse_pack(&w_curs);
21666f1a 1440 return typename(type);
eb32d236 1441 case OBJ_OFS_DELTA:
21666f1a 1442 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
8eca0b47
NP
1443 if (!obj_offset)
1444 die("pack %s contains bad delta base reference of type %s",
1445 p->pack_name, typename(type));
eb32d236 1446 if (*delta_chain_length == 0) {
70f5d5d3
NP
1447 revidx = find_pack_revindex(p, obj_offset);
1448 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
eb32d236
NP
1449 }
1450 break;
1451 case OBJ_REF_DELTA:
2b87c45b 1452 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
43057304
NP
1453 if (*delta_chain_length == 0)
1454 hashcpy(base_sha1, next_sha1);
2b87c45b 1455 obj_offset = find_pack_entry_one(next_sha1, p);
43057304
NP
1456 break;
1457 }
43057304 1458 (*delta_chain_length)++;
2b87c45b 1459 curpos = obj_offset;
21666f1a 1460 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
ad8c80a5 1461 }
ad8c80a5
JH
1462}
1463
c4001d92 1464static int packed_object_info(struct packed_git *p, off_t obj_offset,
21666f1a 1465 unsigned long *sizep)
1f688557 1466{
03e79c88 1467 struct pack_window *w_curs = NULL;
c4001d92
SP
1468 unsigned long size;
1469 off_t curpos = obj_offset;
21666f1a 1470 enum object_type type;
5db47c2b 1471
21666f1a 1472 type = unpack_object_header(p, &w_curs, &curpos, &size);
5db47c2b 1473
21666f1a 1474 switch (type) {
eb32d236
NP
1475 case OBJ_OFS_DELTA:
1476 case OBJ_REF_DELTA:
21666f1a
NP
1477 type = packed_delta_info(p, &w_curs, curpos,
1478 type, obj_offset, sizep);
2b87c45b 1479 break;
a733cb60 1480 case OBJ_COMMIT:
a733cb60 1481 case OBJ_TREE:
a733cb60 1482 case OBJ_BLOB:
a733cb60 1483 case OBJ_TAG:
2b87c45b
NP
1484 if (sizep)
1485 *sizep = size;
a69d0943 1486 break;
1f688557 1487 default:
08a19d87 1488 die("pack %s contains unknown object type %d",
21666f1a 1489 p->pack_name, type);
1f688557 1490 }
2b87c45b 1491 unuse_pack(&w_curs);
21666f1a 1492 return type;
1f688557
JH
1493}
1494
eb950c19 1495static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1496 struct pack_window **w_curs,
c4001d92 1497 off_t curpos,
eb950c19 1498 unsigned long size)
de530aaa
SP
1499{
1500 int st;
1501 z_stream stream;
079afb18 1502 unsigned char *buffer, *in;
de530aaa
SP
1503
1504 buffer = xmalloc(size + 1);
1505 buffer[size] = 0;
1506 memset(&stream, 0, sizeof(stream));
de530aaa
SP
1507 stream.next_out = buffer;
1508 stream.avail_out = size;
1509
1510 inflateInit(&stream);
079afb18 1511 do {
2b87c45b 1512 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18
SP
1513 stream.next_in = in;
1514 st = inflate(&stream, Z_FINISH);
2b87c45b 1515 curpos += stream.next_in - in;
079afb18 1516 } while (st == Z_OK || st == Z_BUF_ERROR);
de530aaa
SP
1517 inflateEnd(&stream);
1518 if ((st != Z_STREAM_END) || stream.total_out != size) {
1519 free(buffer);
1520 return NULL;
1521 }
1522
1523 return buffer;
1524}
1525
e5e01619
LT
1526#define MAX_DELTA_CACHE (256)
1527
18bdec11 1528static size_t delta_base_cached;
5e08ecbf
NP
1529
1530static struct delta_base_cache_lru_list {
1531 struct delta_base_cache_lru_list *prev;
1532 struct delta_base_cache_lru_list *next;
1533} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1534
e5e01619 1535static struct delta_base_cache_entry {
5e08ecbf
NP
1536 struct delta_base_cache_lru_list lru;
1537 void *data;
e5e01619
LT
1538 struct packed_git *p;
1539 off_t base_offset;
1540 unsigned long size;
e5e01619
LT
1541 enum object_type type;
1542} delta_base_cache[MAX_DELTA_CACHE];
1543
1544static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1545{
1546 unsigned long hash;
1547
1548 hash = (unsigned long)p + (unsigned long)base_offset;
1549 hash += (hash >> 8) + (hash >> 16);
3358004a 1550 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1551}
1552
62f255ad 1553static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 1554 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 1555{
e5e01619
LT
1556 void *ret;
1557 unsigned long hash = pack_entry_hash(p, base_offset);
1558 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1559
1560 ret = ent->data;
1561 if (ret && ent->p == p && ent->base_offset == base_offset)
1562 goto found_cache_entry;
62f255ad 1563 return unpack_entry(p, base_offset, type, base_size);
e5e01619
LT
1564
1565found_cache_entry:
18bdec11 1566 if (!keep_cache) {
a0cba108 1567 ent->data = NULL;
5e08ecbf
NP
1568 ent->lru.next->prev = ent->lru.prev;
1569 ent->lru.prev->next = ent->lru.next;
18bdec11 1570 delta_base_cached -= ent->size;
182af834
PH
1571 } else {
1572 ret = xmemdupz(ent->data, ent->size);
a0cba108 1573 }
e5e01619
LT
1574 *type = ent->type;
1575 *base_size = ent->size;
1576 return ret;
62f255ad
LT
1577}
1578
18bdec11
SP
1579static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1580{
1581 if (ent->data) {
1582 free(ent->data);
1583 ent->data = NULL;
5e08ecbf
NP
1584 ent->lru.next->prev = ent->lru.prev;
1585 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
1586 delta_base_cached -= ent->size;
1587 }
1588}
1589
62f255ad
LT
1590static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1591 void *base, unsigned long base_size, enum object_type type)
1592{
5e08ecbf 1593 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 1594 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 1595 struct delta_base_cache_lru_list *lru;
e5e01619 1596
18bdec11
SP
1597 release_delta_base_cache(ent);
1598 delta_base_cached += base_size;
5e08ecbf
NP
1599
1600 for (lru = delta_base_cache_lru.next;
1601 delta_base_cached > delta_base_cache_limit
1602 && lru != &delta_base_cache_lru;
1603 lru = lru->next) {
1604 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
1605 if (f->type == OBJ_BLOB)
1606 release_delta_base_cache(f);
1607 }
5e08ecbf
NP
1608 for (lru = delta_base_cache_lru.next;
1609 delta_base_cached > delta_base_cache_limit
1610 && lru != &delta_base_cache_lru;
1611 lru = lru->next) {
1612 struct delta_base_cache_entry *f = (void *)lru;
1613 release_delta_base_cache(f);
1614 }
18bdec11 1615
e5e01619
LT
1616 ent->p = p;
1617 ent->base_offset = base_offset;
1618 ent->type = type;
1619 ent->data = base;
1620 ent->size = base_size;
5e08ecbf
NP
1621 ent->lru.next = &delta_base_cache_lru;
1622 ent->lru.prev = delta_base_cache_lru.prev;
1623 delta_base_cache_lru.prev->next = &ent->lru;
1624 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
1625}
1626
eb950c19 1627static void *unpack_delta_entry(struct packed_git *p,
03e79c88 1628 struct pack_window **w_curs,
c4001d92 1629 off_t curpos,
1f688557 1630 unsigned long delta_size,
c4001d92 1631 off_t obj_offset,
21666f1a 1632 enum object_type *type,
eb950c19 1633 unsigned long *sizep)
1f688557 1634{
7c3e8be3 1635 void *delta_data, *result, *base;
c4001d92
SP
1636 unsigned long base_size;
1637 off_t base_offset;
43057304 1638
21666f1a 1639 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
8eca0b47
NP
1640 if (!base_offset) {
1641 error("failed to validate delta base reference "
1642 "at offset %"PRIuMAX" from %s",
1643 (uintmax_t)curpos, p->pack_name);
1644 return NULL;
1645 }
eac12e2d 1646 unuse_pack(w_curs);
a0cba108 1647 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
8eca0b47
NP
1648 if (!base) {
1649 /*
1650 * We're probably in deep shit, but let's try to fetch
1651 * the required base anyway from another pack or loose.
1652 * This is costly but should happen only in the presence
1653 * of a corrupted pack, and is better than failing outright.
1654 */
1655 struct revindex_entry *revidx = find_pack_revindex(p, base_offset);
1656 const unsigned char *base_sha1 =
1657 nth_packed_object_sha1(p, revidx->nr);
1658 error("failed to read delta base object %s"
1659 " at offset %"PRIuMAX" from %s",
1660 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1661 p->pack_name);
1662 mark_bad_packed_object(p, base_sha1);
ac939109 1663 base = read_object(base_sha1, type, &base_size);
8eca0b47
NP
1664 if (!base)
1665 return NULL;
1666 }
67686d95 1667
2b87c45b 1668 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
8eca0b47
NP
1669 if (!delta_data) {
1670 error("failed to unpack compressed delta "
1671 "at offset %"PRIuMAX" from %s",
1672 (uintmax_t)curpos, p->pack_name);
1673 free(base);
1674 return NULL;
1675 }
1f688557
JH
1676 result = patch_delta(base, base_size,
1677 delta_data, delta_size,
21666f1a 1678 sizep);
1f688557
JH
1679 if (!result)
1680 die("failed to apply delta");
1681 free(delta_data);
62f255ad 1682 add_delta_base_cache(p, base_offset, base, base_size, *type);
1f688557
JH
1683 return result;
1684}
1685
c4001d92 1686void *unpack_entry(struct packed_git *p, off_t obj_offset,
21666f1a 1687 enum object_type *type, unsigned long *sizep)
1f688557 1688{
03e79c88 1689 struct pack_window *w_curs = NULL;
c4001d92 1690 off_t curpos = obj_offset;
21666f1a 1691 void *data;
1f688557 1692
21666f1a
NP
1693 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1694 switch (*type) {
eb32d236
NP
1695 case OBJ_OFS_DELTA:
1696 case OBJ_REF_DELTA:
21666f1a
NP
1697 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1698 obj_offset, type, sizep);
4d703a1a 1699 break;
a733cb60 1700 case OBJ_COMMIT:
a733cb60 1701 case OBJ_TREE:
a733cb60 1702 case OBJ_BLOB:
a733cb60 1703 case OBJ_TAG:
21666f1a 1704 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
4d703a1a 1705 break;
1f688557 1706 default:
8eca0b47
NP
1707 data = NULL;
1708 error("unknown object type %i at offset %"PRIuMAX" in %s",
1709 *type, (uintmax_t)obj_offset, p->pack_name);
1f688557 1710 }
03e79c88 1711 unuse_pack(&w_curs);
21666f1a 1712 return data;
1f688557
JH
1713}
1714
d079837e 1715const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 1716 uint32_t n)
9a217f2a 1717{
42873078 1718 const unsigned char *index = p->index_data;
d079837e
SP
1719 if (!index) {
1720 if (open_pack_index(p))
1721 return NULL;
1722 index = p->index_data;
1723 }
57059091 1724 if (n >= p->num_objects)
d72308e0 1725 return NULL;
74e34e1f
NP
1726 index += 4 * 256;
1727 if (p->index_version == 1) {
1728 return index + 24 * n + 4;
1729 } else {
1730 index += 8;
1731 return index + 20 * n;
1732 }
1733}
1734
99093238 1735off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
1736{
1737 const unsigned char *index = p->index_data;
1738 index += 4 * 256;
1739 if (p->index_version == 1) {
1740 return ntohl(*((uint32_t *)(index + 24 * n)));
1741 } else {
1742 uint32_t off;
1743 index += 8 + p->num_objects * (20 + 4);
1744 off = ntohl(*((uint32_t *)(index + 4 * n)));
1745 if (!(off & 0x80000000))
1746 return off;
1747 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1748 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1749 ntohl(*((uint32_t *)(index + 4)));
1750 }
9a217f2a
JH
1751}
1752
c4001d92 1753off_t find_pack_entry_one(const unsigned char *sha1,
43057304 1754 struct packed_git *p)
1f688557 1755{
42873078 1756 const uint32_t *level1_ofs = p->index_data;
42873078 1757 const unsigned char *index = p->index_data;
628522ec
JH
1758 unsigned hi, lo, stride;
1759 static int use_lookup = -1;
1760 static int debug_lookup = -1;
1761
1762 if (debug_lookup < 0)
1763 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 1764
d079837e
SP
1765 if (!index) {
1766 if (open_pack_index(p))
1767 return 0;
1768 level1_ofs = p->index_data;
1769 index = p->index_data;
1770 }
74e34e1f
NP
1771 if (p->index_version > 1) {
1772 level1_ofs += 2;
1773 index += 8;
1774 }
42873078 1775 index += 4 * 256;
74e34e1f
NP
1776 hi = ntohl(level1_ofs[*sha1]);
1777 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
1778 if (p->index_version > 1) {
1779 stride = 20;
1780 } else {
1781 stride = 24;
1782 index += 4;
1783 }
1784
1785 if (debug_lookup)
6e1c2344 1786 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
1787 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1788
1789 if (use_lookup < 0)
1790 use_lookup = !!getenv("GIT_USE_LOOKUP");
1791 if (use_lookup) {
1792 int pos = sha1_entry_pos(index, stride, 0,
1793 lo, hi, p->num_objects, sha1);
1794 if (pos < 0)
1795 return 0;
1796 return nth_packed_object_offset(p, pos);
1797 }
1f688557
JH
1798
1799 do {
74e34e1f 1800 unsigned mi = (lo + hi) / 2;
628522ec
JH
1801 int cmp = hashcmp(index + mi * stride, sha1);
1802
1803 if (debug_lookup)
1804 printf("lo %u hi %u rg %u mi %u\n",
1805 lo, hi, hi - lo, mi);
43057304 1806 if (!cmp)
74e34e1f 1807 return nth_packed_object_offset(p, mi);
1f688557
JH
1808 if (cmp > 0)
1809 hi = mi;
1810 else
1811 lo = mi+1;
1812 } while (lo < hi);
1813 return 0;
1814}
1815
000dfd3f 1816int matches_pack_name(struct packed_git *p, const char *name)
d4ff6d92
SP
1817{
1818 const char *last_c, *c;
1819
000dfd3f
JH
1820 if (!strcmp(p->pack_name, name))
1821 return 1;
d4ff6d92
SP
1822
1823 for (c = p->pack_name, last_c = c; *c;)
1824 if (*c == '/')
1825 last_c = ++c;
1826 else
1827 ++c;
000dfd3f
JH
1828 if (!strcmp(last_c, name))
1829 return 1;
d4ff6d92 1830
000dfd3f 1831 return 0;
d4ff6d92
SP
1832}
1833
106d710b 1834static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1f688557 1835{
f7c22cc6 1836 static struct packed_git *last_found = (void *)1;
1f688557 1837 struct packed_git *p;
c4001d92 1838 off_t offset;
43057304 1839
1f688557 1840 prepare_packed_git();
f7c22cc6
NP
1841 if (!packed_git)
1842 return 0;
1843 p = (last_found == (void *)1) ? packed_git : last_found;
1f688557 1844
f7c22cc6 1845 do {
106d710b
JH
1846 if (ignore_packed) {
1847 const char **ig;
1848 for (ig = ignore_packed; *ig; ig++)
000dfd3f 1849 if (matches_pack_name(p, *ig))
106d710b
JH
1850 break;
1851 if (*ig)
f7c22cc6 1852 goto next;
106d710b 1853 }
f7c22cc6 1854
8eca0b47
NP
1855 if (p->num_bad_objects) {
1856 unsigned i;
1857 for (i = 0; i < p->num_bad_objects; i++)
1858 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1859 goto next;
1860 }
1861
43057304
NP
1862 offset = find_pack_entry_one(sha1, p);
1863 if (offset) {
c715f783
SP
1864 /*
1865 * We are about to tell the caller where they can
1866 * locate the requested object. We better make
1867 * sure the packfile is still here and can be
1868 * accessed before supplying that answer, as
1869 * it may have been deleted since the index
1870 * was loaded!
1871 */
1872 if (p->pack_fd == -1 && open_packed_git(p)) {
1873 error("packfile %s cannot be accessed", p->pack_name);
f7c22cc6 1874 goto next;
c715f783 1875 }
43057304
NP
1876 e->offset = offset;
1877 e->p = p;
1878 hashcpy(e->sha1, sha1);
f7c22cc6 1879 last_found = p;
1f688557 1880 return 1;
43057304 1881 }
f7c22cc6
NP
1882
1883 next:
1884 if (p == last_found)
1885 p = packed_git;
1886 else
1887 p = p->next;
1888 if (p == last_found)
1889 p = p->next;
1890 } while (p);
1f688557
JH
1891 return 0;
1892}
1893
a6080a0a 1894struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
1895 struct packed_git *packs)
1896{
1897 struct packed_git *p;
bf592c50
DB
1898
1899 for (p = packs; p; p = p->next) {
43057304 1900 if (find_pack_entry_one(sha1, p))
bf592c50
DB
1901 return p;
1902 }
1903 return NULL;
21666f1a 1904
bf592c50
DB
1905}
1906
21666f1a 1907static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
65c2e0c3 1908{
36e4d74a 1909 int status;
65c2e0c3
JH
1910 unsigned long mapsize, size;
1911 void *map;
1912 z_stream stream;
d65a16f6 1913 char hdr[32];
65c2e0c3 1914
bb6b8e4f 1915 map = map_sha1_file(sha1, &mapsize);
f0df4ed5
JS
1916 if (!map)
1917 return error("unable to find %s", sha1_to_hex(sha1));
36e4d74a
JH
1918 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1919 status = error("unable to unpack %s header",
1920 sha1_to_hex(sha1));
21666f1a 1921 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 1922 status = error("unable to parse %s header", sha1_to_hex(sha1));
21666f1a
NP
1923 else if (sizep)
1924 *sizep = size;
65c2e0c3
JH
1925 inflateEnd(&stream);
1926 munmap(map, mapsize);
1927 return status;
1928}
1929
21666f1a 1930int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
f0df4ed5 1931{
f0df4ed5 1932 struct pack_entry e;
ddd63e64 1933 int status;
f0df4ed5
JS
1934
1935 if (!find_pack_entry(sha1, &e, NULL)) {
ddd63e64
SG
1936 /* Most likely it's a loose object. */
1937 status = sha1_loose_object_info(sha1, sizep);
1938 if (status >= 0)
1939 return status;
1940
1941 /* Not a loose object; someone else may have just packed it. */
f0df4ed5
JS
1942 reprepare_packed_git();
1943 if (!find_pack_entry(sha1, &e, NULL))
ddd63e64 1944 return status;
f0df4ed5 1945 }
21666f1a 1946 return packed_object_info(e.p, e.offset, sizep);
f0df4ed5
JS
1947}
1948
21666f1a
NP
1949static void *read_packed_sha1(const unsigned char *sha1,
1950 enum object_type *type, unsigned long *size)
1f688557
JH
1951{
1952 struct pack_entry e;
8eca0b47 1953 void *data;
1f688557 1954
8276c007 1955 if (!find_pack_entry(sha1, &e, NULL))
1f688557 1956 return NULL;
8eca0b47
NP
1957 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1958 if (!data) {
1959 /*
1960 * We're probably in deep shit, but let's try to fetch
1961 * the required object anyway from another pack or loose.
1962 * This should happen only in the presence of a corrupted
1963 * pack, and is better than failing outright.
1964 */
1965 error("failed to read object %s at offset %"PRIuMAX" from %s",
1966 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
1967 mark_bad_packed_object(e.p, sha1);
ac939109 1968 data = read_object(sha1, type, size);
8eca0b47
NP
1969 }
1970 return data;
1f688557
JH
1971}
1972
d66b37bb
JH
1973/*
1974 * This is meant to hold a *small* number of objects that you would
1975 * want read_sha1_file() to be able to return, but yet you do not want
1976 * to write them into the object store (e.g. a browse-only
1977 * application).
1978 */
1979static struct cached_object {
1980 unsigned char sha1[20];
21666f1a 1981 enum object_type type;
d66b37bb
JH
1982 void *buf;
1983 unsigned long size;
1984} *cached_objects;
1985static int cached_object_nr, cached_object_alloc;
1986
346245a1
JK
1987static struct cached_object empty_tree = {
1988 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
1989 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
1990 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
1991 OBJ_TREE,
1992 "",
1993 0
1994};
1995
d66b37bb
JH
1996static struct cached_object *find_cached_object(const unsigned char *sha1)
1997{
1998 int i;
1999 struct cached_object *co = cached_objects;
2000
2001 for (i = 0; i < cached_object_nr; i++, co++) {
2002 if (!hashcmp(co->sha1, sha1))
2003 return co;
2004 }
346245a1
JK
2005 if (!hashcmp(sha1, empty_tree.sha1))
2006 return &empty_tree;
d66b37bb
JH
2007 return NULL;
2008}
2009
21666f1a
NP
2010int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2011 unsigned char *sha1)
d66b37bb
JH
2012{
2013 struct cached_object *co;
2014
21666f1a 2015 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2016 if (has_sha1_file(sha1) || find_cached_object(sha1))
2017 return 0;
2018 if (cached_object_alloc <= cached_object_nr) {
2019 cached_object_alloc = alloc_nr(cached_object_alloc);
2020 cached_objects = xrealloc(cached_objects,
2021 sizeof(*cached_objects) *
2022 cached_object_alloc);
2023 }
2024 co = &cached_objects[cached_object_nr++];
2025 co->size = len;
21666f1a 2026 co->type = type;
efa13f7b
JH
2027 co->buf = xmalloc(len);
2028 memcpy(co->buf, buf, len);
d66b37bb
JH
2029 hashcpy(co->sha1, sha1);
2030 return 0;
2031}
2032
ac939109
NP
2033void *read_object(const unsigned char *sha1, enum object_type *type,
2034 unsigned long *size)
0fcfd160
LT
2035{
2036 unsigned long mapsize;
2037 void *map, *buf;
d66b37bb
JH
2038 struct cached_object *co;
2039
2040 co = find_cached_object(sha1);
2041 if (co) {
21666f1a 2042 *type = co->type;
d66b37bb 2043 *size = co->size;
182af834 2044 return xmemdupz(co->buf, co->size);
d66b37bb 2045 }
0fcfd160 2046
8276c007
PE
2047 buf = read_packed_sha1(sha1, type, size);
2048 if (buf)
2049 return buf;
bb6b8e4f 2050 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2051 if (map) {
7efbff75 2052 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2053 munmap(map, mapsize);
2054 return buf;
2055 }
637cdd9d 2056 reprepare_packed_git();
8276c007 2057 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2058}
2059
ac939109
NP
2060void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
2061 unsigned long *size)
2062{
2063 void *data = read_object(sha1, type, size);
2064 /* legacy behavior is to die on corrupted objects */
2065 if (!data && (has_loose_object(sha1) || has_packed_and_bad(sha1)))
2066 die("object %s is corrupted", sha1_to_hex(sha1));
2067 return data;
2068}
2069
40469ee9 2070void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2071 const char *required_type_name,
40469ee9
JH
2072 unsigned long *size,
2073 unsigned char *actual_sha1_return)
f4913f91 2074{
21666f1a 2075 enum object_type type, required_type;
f4913f91
JH
2076 void *buffer;
2077 unsigned long isize;
40469ee9 2078 unsigned char actual_sha1[20];
f4913f91 2079
21666f1a 2080 required_type = type_from_string(required_type_name);
e702496e 2081 hashcpy(actual_sha1, sha1);
40469ee9
JH
2082 while (1) {
2083 int ref_length = -1;
2084 const char *ref_type = NULL;
f4913f91 2085
21666f1a 2086 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2087 if (!buffer)
2088 return NULL;
21666f1a 2089 if (type == required_type) {
40469ee9
JH
2090 *size = isize;
2091 if (actual_sha1_return)
e702496e 2092 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2093 return buffer;
2094 }
2095 /* Handle references */
21666f1a 2096 else if (type == OBJ_COMMIT)
40469ee9 2097 ref_type = "tree ";
21666f1a 2098 else if (type == OBJ_TAG)
40469ee9
JH
2099 ref_type = "object ";
2100 else {
2101 free(buffer);
2102 return NULL;
2103 }
2104 ref_length = strlen(ref_type);
f4913f91 2105
50974ec9
MK
2106 if (ref_length + 40 > isize ||
2107 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2108 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2109 free(buffer);
2110 return NULL;
2111 }
1cf58e72 2112 free(buffer);
40469ee9
JH
2113 /* Now we have the ID of the referred-to object in
2114 * actual_sha1. Check again. */
f4913f91 2115 }
f4913f91
JH
2116}
2117
ce9fbf16 2118static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2119 const char *type, unsigned char *sha1,
d65a16f6 2120 char *hdr, int *hdrlen)
d410c0f5
JH
2121{
2122 SHA_CTX c;
2123
2124 /* Generate the header */
d65a16f6 2125 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
2126
2127 /* Sha1.. */
2128 SHA1_Init(&c);
2129 SHA1_Update(&c, hdr, *hdrlen);
2130 SHA1_Update(&c, buf, len);
2131 SHA1_Final(sha1, &c);
d410c0f5
JH
2132}
2133
230f1322
LT
2134/*
2135 * Move the just written object into its final resting place
2136 */
839837b9 2137int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322 2138{
e32c0a9c
TR
2139 int ret = 0;
2140 if (link(tmpfile, filename))
2141 ret = errno;
7ebb6fca
LT
2142
2143 /*
2144 * Coda hack - coda doesn't like cross-directory links,
2145 * so we fall back to a rename, which will mean that it
2146 * won't be able to check collisions, but that's not a
2147 * big deal.
2148 *
2149 * The same holds for FAT formatted media.
2150 *
2151 * When this succeeds, we just return 0. We have nothing
2152 * left to unlink.
2153 */
2154 if (ret && ret != EEXIST) {
2155 if (!rename(tmpfile, filename))
230f1322 2156 return 0;
9e48b389 2157 ret = errno;
230f1322
LT
2158 }
2159 unlink(tmpfile);
2160 if (ret) {
2161 if (ret != EEXIST) {
916d081b 2162 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
230f1322
LT
2163 }
2164 /* FIXME!!! Collision check here ? */
2165 }
2166
2167 return 0;
2168}
2169
4d548150
LT
2170static int write_buffer(int fd, const void *buf, size_t len)
2171{
d34cf19b 2172 if (write_in_full(fd, buf, len) < 0)
93822c22 2173 return error("file write error (%s)", strerror(errno));
4d548150
LT
2174 return 0;
2175}
2176
ce9fbf16 2177int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2178 unsigned char *sha1)
2179{
d65a16f6 2180 char hdr[32];
abdc3fc8
RS
2181 int hdrlen;
2182 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2183 return 0;
2184}
2185
e9039dd3
LT
2186/* Finalize a file on disk, and close it. */
2187static void close_sha1_file(int fd)
2188{
aafe9fba
LT
2189 if (fsync_object_files)
2190 fsync_or_die(fd, "sha1 file");
e9039dd3
LT
2191 fchmod(fd, 0444);
2192 if (close(fd) != 0)
2193 die("unable to write sha1 file");
2194}
2195
5723fe7e
LT
2196/* Size of directory component, including the ending '/' */
2197static inline int directory_size(const char *filename)
2198{
2199 const char *s = strrchr(filename, '/');
2200 if (!s)
2201 return 0;
2202 return s - filename + 1;
2203}
2204
2205/*
2206 * This creates a temporary file in the same directory as the final
2207 * 'filename'
2208 *
2209 * We want to avoid cross-directory filename renames, because those
2210 * can have problems on various filesystems (FAT, NFS, Coda).
2211 */
2212static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2213{
2214 int fd, dirlen = directory_size(filename);
2215
2216 if (dirlen + 20 > bufsiz) {
2217 errno = ENAMETOOLONG;
2218 return -1;
2219 }
2220 memcpy(buffer, filename, dirlen);
2221 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2222 fd = mkstemp(buffer);
2223 if (fd < 0 && dirlen) {
2224 /* Make sure the directory exists */
6ff6af62 2225 memcpy(buffer, filename, dirlen);
5723fe7e 2226 buffer[dirlen-1] = 0;
3bfaf018 2227 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
5723fe7e
LT
2228 return -1;
2229
2230 /* Try again */
2231 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2232 fd = mkstemp(buffer);
2233 }
2234 return fd;
2235}
2236
bbac7311
NP
2237static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2238 void *buf, unsigned long len, time_t mtime)
0fcfd160 2239{
bbac7311 2240 int fd, size, ret;
bf0f910d 2241 unsigned char *compressed;
0fcfd160 2242 z_stream stream;
706bc531 2243 char *filename;
aac17941 2244 static char tmpfile[PATH_MAX];
a44c9a5e 2245
972a9155 2246 filename = sha1_file_name(sha1);
5723fe7e 2247 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
aac17941 2248 if (fd < 0) {
916d081b
PB
2249 if (errno == EPERM)
2250 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2251 else
2252 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
aac17941
LT
2253 }
2254
0fcfd160
LT
2255 /* Set it up */
2256 memset(&stream, 0, sizeof(stream));
12f6c308 2257 deflateInit(&stream, zlib_compression_level);
93821bd9 2258 size = 8 + deflateBound(&stream, len+hdrlen);
812666c8 2259 compressed = xmalloc(size);
0fcfd160
LT
2260
2261 /* Compress it */
0fcfd160
LT
2262 stream.next_out = compressed;
2263 stream.avail_out = size;
a44c9a5e
LT
2264
2265 /* First header.. */
d65a16f6 2266 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2267 stream.avail_in = hdrlen;
726f852b
NP
2268 while (deflate(&stream, 0) == Z_OK)
2269 /* nothing */;
a44c9a5e
LT
2270
2271 /* Then the data itself.. */
2272 stream.next_in = buf;
2273 stream.avail_in = len;
ac54c277
LT
2274 ret = deflate(&stream, Z_FINISH);
2275 if (ret != Z_STREAM_END)
2276 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2277
2278 ret = deflateEnd(&stream);
2279 if (ret != Z_OK)
2280 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2281
0fcfd160
LT
2282 size = stream.total_out;
2283
4d548150
LT
2284 if (write_buffer(fd, compressed, size) < 0)
2285 die("unable to write sha1 file");
e9039dd3 2286 close_sha1_file(fd);
383f85b7 2287 free(compressed);
0fcfd160 2288
bbac7311
NP
2289 if (mtime) {
2290 struct utimbuf utb;
2291 utb.actime = mtime;
2292 utb.modtime = mtime;
2293 if (utime(tmpfile, &utb) < 0)
2294 warning("failed utime() on %s: %s",
2295 tmpfile, strerror(errno));
2296 }
2297
230f1322 2298 return move_temp_to_file(tmpfile, filename);
0fcfd160 2299}
8237b185 2300
bbac7311
NP
2301int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2302{
2303 unsigned char sha1[20];
2304 char hdr[32];
2305 int hdrlen;
2306
2307 /* Normally if we have it in the pack then we do not bother writing
2308 * it out into .git/objects/??/?{38} file.
2309 */
2310 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2311 if (returnsha1)
2312 hashcpy(returnsha1, sha1);
2313 if (has_sha1_file(sha1))
2314 return 0;
2315 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2316}
2317
2318int force_object_loose(const unsigned char *sha1, time_t mtime)
2319{
bbac7311
NP
2320 void *buf;
2321 unsigned long len;
2322 enum object_type type;
2323 char hdr[32];
2324 int hdrlen;
2325
c529d75a 2326 if (has_loose_object(sha1))
bbac7311
NP
2327 return 0;
2328 buf = read_packed_sha1(sha1, &type, &len);
2329 if (!buf)
2330 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2331 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2332 return write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2333}
2334
bf592c50
DB
2335int has_pack_index(const unsigned char *sha1)
2336{
2337 struct stat st;
2338 if (stat(sha1_pack_index_name(sha1), &st))
2339 return 0;
2340 return 1;
2341}
2342
2343int has_pack_file(const unsigned char *sha1)
2344{
2345 struct stat st;
2346 if (stat(sha1_pack_name(sha1), &st))
2347 return 0;
2348 return 1;
2349}
2350
106d710b 2351int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
dade09c2
LT
2352{
2353 struct pack_entry e;
106d710b 2354 return find_pack_entry(sha1, &e, ignore_packed);
dade09c2
LT
2355}
2356
8237b185
DB
2357int has_sha1_file(const unsigned char *sha1)
2358{
1f688557
JH
2359 struct pack_entry e;
2360
106d710b 2361 if (find_pack_entry(sha1, &e, NULL))
1f688557 2362 return 1;
c529d75a 2363 return has_loose_object(sha1);
8237b185 2364}
74400e71 2365
e7332f96
BE
2366int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2367{
fd17f5b5 2368 struct strbuf buf;
e7332f96 2369 int ret;
e7332f96 2370
fd17f5b5
PH
2371 strbuf_init(&buf, 0);
2372 if (strbuf_read(&buf, fd, 4096) < 0) {
2373 strbuf_release(&buf);
024510c8
DB
2374 return -1;
2375 }
e7332f96 2376
024510c8 2377 if (!type)
8e440259 2378 type = blob_type;
024510c8 2379 if (write_object)
fd17f5b5 2380 ret = write_sha1_file(buf.buf, buf.len, type, sha1);
abdc3fc8 2381 else
fd17f5b5
PH
2382 ret = hash_sha1_file(buf.buf, buf.len, type, sha1);
2383 strbuf_release(&buf);
2384
024510c8
DB
2385 return ret;
2386}
2387
edaec3fb 2388int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
53bca91a 2389 enum object_type type, const char *path)
74400e71 2390{
dc49cd76 2391 size_t size = xsize_t(st->st_size);
3a55602e 2392 void *buf = NULL;
6c510bee 2393 int ret, re_allocated = 0;
74400e71 2394
74400e71 2395 if (size)
c4712e45 2396 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
74400e71 2397 close(fd);
74400e71 2398
7672db20 2399 if (!type)
edaec3fb 2400 type = OBJ_BLOB;
6c510bee
LT
2401
2402 /*
2403 * Convert blobs to git internal format
2404 */
d0d8e14d 2405 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
5ecd293d
PH
2406 struct strbuf nbuf;
2407 strbuf_init(&nbuf, 0);
21e5ad50
SP
2408 if (convert_to_git(path, buf, size, &nbuf,
2409 write_object ? safe_crlf : 0)) {
ac78e548 2410 munmap(buf, size);
b315c5c0 2411 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
2412 re_allocated = 1;
2413 }
2414 }
2415
7672db20 2416 if (write_object)
edaec3fb 2417 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 2418 else
edaec3fb 2419 ret = hash_sha1_file(buf, size, typename(type), sha1);
6c510bee
LT
2420 if (re_allocated) {
2421 free(buf);
2422 return ret;
2423 }
aac17941
LT
2424 if (size)
2425 munmap(buf, size);
2426 return ret;
74400e71 2427}
ec1fcc16
JH
2428
2429int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2430{
2431 int fd;
2432 char *target;
dc49cd76 2433 size_t len;
ec1fcc16
JH
2434
2435 switch (st->st_mode & S_IFMT) {
2436 case S_IFREG:
2437 fd = open(path, O_RDONLY);
2438 if (fd < 0)
2439 return error("open(\"%s\"): %s", path,
2440 strerror(errno));
53bca91a 2441 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
ec1fcc16
JH
2442 return error("%s: failed to insert into database",
2443 path);
2444 break;
2445 case S_IFLNK:
dc49cd76
SP
2446 len = xsize_t(st->st_size);
2447 target = xmalloc(len + 1);
2448 if (readlink(path, target, len + 1) != st->st_size) {
ec1fcc16
JH
2449 char *errstr = strerror(errno);
2450 free(target);
2451 return error("readlink(\"%s\"): %s", path,
2452 errstr);
2453 }
abdc3fc8 2454 if (!write_object)
dc49cd76
SP
2455 hash_sha1_file(target, len, blob_type, sha1);
2456 else if (write_sha1_file(target, len, blob_type, sha1))
ec1fcc16
JH
2457 return error("%s: failed to insert into database",
2458 path);
2459 free(target);
2460 break;
f35a6d3b
LT
2461 case S_IFDIR:
2462 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
2463 default:
2464 return error("%s: unsupported file type", path);
2465 }
2466 return 0;
2467}
a69e5429
JH
2468
2469int read_pack_header(int fd, struct pack_header *header)
2470{
c697ad14
HO
2471 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2472 /* "eof before pack header was fully read" */
2473 return PH_ERROR_EOF;
2474
a69e5429
JH
2475 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2476 /* "protocol error (pack signature mismatch detected)" */
2477 return PH_ERROR_PACK_SIGNATURE;
2478 if (!pack_version_ok(header->hdr_version))
2479 /* "protocol error (pack version unsupported)" */
2480 return PH_ERROR_PROTOCOL;
2481 return 0;
2482}