pack-objects: mark file-local variable static
[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
b2cb9425
JH
38int safe_create_leading_directories(char *path)
39{
8385abfd 40 char *pos = path + offset_1st_component(path);
67d42212
JR
41 struct stat st;
42
b2cb9425
JH
43 while (pos) {
44 pos = strchr(pos, '/');
45 if (!pos)
46 break;
5f0bdf50
JH
47 while (*++pos == '/')
48 ;
49 if (!*pos)
50 break;
51 *--pos = '\0';
67d42212
JR
52 if (!stat(path, &st)) {
53 /* path exists */
54 if (!S_ISDIR(st.st_mode)) {
b2cb9425 55 *pos = '/';
67d42212 56 return -3;
b2cb9425 57 }
457f06d6 58 }
67d42212
JR
59 else if (mkdir(path, 0777)) {
60 *pos = '/';
61 return -1;
62 }
457f06d6
JS
63 else if (adjust_shared_perm(path)) {
64 *pos = '/';
65 return -2;
66 }
b2cb9425
JH
67 *pos++ = '/';
68 }
69 return 0;
70}
723c31fe 71
8e21d63b
JK
72int safe_create_leading_directories_const(const char *path)
73{
74 /* path points to cache entries, so xstrdup before messing with it */
75 char *buf = xstrdup(path);
76 int result = safe_create_leading_directories(buf);
77 free(buf);
78 return result;
79}
80
ace1534d
JH
81static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
82{
83 int i;
84 for (i = 0; i < 20; i++) {
85 static char hex[] = "0123456789abcdef";
86 unsigned int val = sha1[i];
87 char *pos = pathbuf + i*2 + (i > 0);
88 *pos++ = hex[val >> 4];
89 *pos = hex[val & 0xf];
90 }
91}
92
0fcfd160
LT
93/*
94 * NOTE! This returns a statically allocated buffer, so you have to be
790296fd 95 * careful about using it. Do an "xstrdup()" if you need to save the
0fcfd160 96 * filename.
ace1534d
JH
97 *
98 * Also note that this returns the location for creating. Reading
99 * SHA1 file can happen from any alternate directory listed in the
d19938ab 100 * DB_ENVIRONMENT environment variable if it is not found in
ace1534d 101 * the primary object database.
0fcfd160
LT
102 */
103char *sha1_file_name(const unsigned char *sha1)
104{
560fb6a1
JK
105 static char buf[PATH_MAX];
106 const char *objdir;
107 int len;
0fcfd160 108
560fb6a1
JK
109 objdir = get_object_directory();
110 len = strlen(objdir);
111
112 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
113 if (len + 43 > PATH_MAX)
114 die("insanely long object directory %s", objdir);
115 memcpy(buf, objdir, len);
116 buf[len] = '/';
117 buf[len+3] = '/';
118 buf[len+42] = '\0';
119 fill_sha1_path(buf + len + 1, sha1);
120 return buf;
0fcfd160
LT
121}
122
633f43e1 123static char *sha1_get_pack_name(const unsigned char *sha1,
6eec46bd 124 char **name, char **base, const char *which)
bf592c50
DB
125{
126 static const char hex[] = "0123456789abcdef";
633f43e1 127 char *buf;
bf592c50
DB
128 int i;
129
633f43e1 130 if (!*base) {
bf592c50
DB
131 const char *sha1_file_directory = get_object_directory();
132 int len = strlen(sha1_file_directory);
633f43e1 133 *base = xmalloc(len + 60);
6eec46bd
JH
134 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
135 sha1_file_directory, which);
633f43e1 136 *name = *base + len + 11;
bf592c50
DB
137 }
138
633f43e1 139 buf = *name;
bf592c50
DB
140
141 for (i = 0; i < 20; i++) {
142 unsigned int val = *sha1++;
143 *buf++ = hex[val >> 4];
144 *buf++ = hex[val & 0xf];
145 }
a6080a0a 146
633f43e1 147 return *base;
bf592c50
DB
148}
149
633f43e1 150char *sha1_pack_name(const unsigned char *sha1)
bf592c50 151{
633f43e1 152 static char *name, *base;
bf592c50 153
6eec46bd 154 return sha1_get_pack_name(sha1, &name, &base, "pack");
633f43e1 155}
bf592c50 156
633f43e1
HO
157char *sha1_pack_index_name(const unsigned char *sha1)
158{
159 static char *name, *base;
a6080a0a 160
6eec46bd 161 return sha1_get_pack_name(sha1, &name, &base, "idx");
bf592c50
DB
162}
163
d5a63b99
JH
164struct alternate_object_database *alt_odb_list;
165static struct alternate_object_database **alt_odb_tail;
ace1534d 166
c2f493a4
MW
167static void read_info_alternates(const char * alternates, int depth);
168
ddd5d056
JH
169/*
170 * Prepare alternate object database registry.
d5a63b99
JH
171 *
172 * The variable alt_odb_list points at the list of struct
173 * alternate_object_database. The elements on this list come from
174 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
175 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
176 * whose contents is similar to that environment variable but can be
177 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
178 * contains "/the/directory/corresponding/to/.git/objects/...", while
179 * its name points just after the slash at the end of ".git/objects/"
180 * in the example above, and has enough space to hold 40-byte hex
181 * SHA1, an extra slash for the first level indirection, and the
182 * terminating NUL.
ddd5d056 183 */
c2f493a4 184static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
ace1534d 185{
1494e038 186 const char *objdir = get_object_directory();
c2f493a4
MW
187 struct alternate_object_database *ent;
188 struct alternate_object_database *alt;
189 /* 43 = 40-byte + 2 '/' + terminating NUL */
190 int pfxlen = len;
191 int entlen = pfxlen + 43;
ccfd3e99 192 int base_len = -1;
d5a63b99 193
85dadc38 194 if (!is_absolute_path(entry) && relative_base) {
c2f493a4
MW
195 /* Relative alt-odb */
196 if (base_len < 0)
197 base_len = strlen(relative_base) + 1;
198 entlen += base_len;
199 pfxlen += base_len;
200 }
201 ent = xmalloc(sizeof(*ent) + entlen);
202
85dadc38 203 if (!is_absolute_path(entry) && relative_base) {
c2f493a4
MW
204 memcpy(ent->base, relative_base, base_len - 1);
205 ent->base[base_len - 1] = '/';
206 memcpy(ent->base + base_len, entry, len);
207 }
208 else
209 memcpy(ent->base, entry, pfxlen);
210
211 ent->name = ent->base + pfxlen + 1;
212 ent->base[pfxlen + 3] = '/';
213 ent->base[pfxlen] = ent->base[entlen-1] = 0;
214
215 /* Detect cases where alternate disappeared */
90b4a71c 216 if (!is_directory(ent->base)) {
c2f493a4
MW
217 error("object directory %s does not exist; "
218 "check .git/objects/info/alternates.",
219 ent->base);
220 free(ent);
221 return -1;
222 }
223
224 /* Prevent the common mistake of listing the same
225 * thing twice, or object directory itself.
226 */
227 for (alt = alt_odb_list; alt; alt = alt->next) {
228 if (!memcmp(ent->base, alt->base, pfxlen)) {
229 free(ent);
230 return -1;
231 }
232 }
233 if (!memcmp(ent->base, objdir, pfxlen)) {
234 free(ent);
235 return -1;
236 }
237
238 /* add the alternate entry */
239 *alt_odb_tail = ent;
240 alt_odb_tail = &(ent->next);
241 ent->next = NULL;
242
243 /* recursively add alternates */
244 read_info_alternates(ent->base, depth + 1);
245
246 ent->base[pfxlen] = '/';
247
248 return 0;
249}
250
251static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
252 const char *relative_base, int depth)
253{
254 const char *cp, *last;
255
256 if (depth > 5) {
257 error("%s: ignoring alternate object stores, nesting too deep.",
258 relative_base);
259 return;
260 }
261
d5a63b99 262 last = alt;
9577e7e3
JH
263 while (last < ep) {
264 cp = last;
265 if (cp < ep && *cp == '#') {
266 while (cp < ep && *cp != sep)
267 cp++;
268 last = cp + 1;
269 continue;
270 }
c2f493a4
MW
271 while (cp < ep && *cp != sep)
272 cp++;
d5a63b99 273 if (last != cp) {
85dadc38 274 if (!is_absolute_path(last) && depth) {
c2f493a4
MW
275 error("%s: ignoring relative alternate object store %s",
276 relative_base, last);
277 } else {
278 link_alt_odb_entry(last, cp - last,
279 relative_base, depth);
1494e038 280 }
d5a63b99 281 }
9577e7e3 282 while (cp < ep && *cp == sep)
d5a63b99
JH
283 cp++;
284 last = cp;
9577e7e3 285 }
d5a63b99
JH
286}
287
c2f493a4 288static void read_info_alternates(const char * relative_base, int depth)
d5a63b99 289{
9577e7e3 290 char *map;
dc49cd76 291 size_t mapsz;
d5a63b99 292 struct stat st;
9cb18f56
JM
293 const char alt_file_name[] = "info/alternates";
294 /* Given that relative_base is no longer than PATH_MAX,
295 ensure that "path" has enough space to append "/", the
296 file name, "info/alternates", and a trailing NUL. */
297 char path[PATH_MAX + 1 + sizeof alt_file_name];
c2f493a4 298 int fd;
d5a63b99 299
9cb18f56 300 sprintf(path, "%s/%s", relative_base, alt_file_name);
d5a63b99
JH
301 fd = open(path, O_RDONLY);
302 if (fd < 0)
303 return;
304 if (fstat(fd, &st) || (st.st_size == 0)) {
305 close(fd);
9a217f2a 306 return;
ace1534d 307 }
dc49cd76
SP
308 mapsz = xsize_t(st.st_size);
309 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 310 close(fd);
d5a63b99 311
dc49cd76 312 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
c2f493a4 313
dc49cd76 314 munmap(map, mapsz);
ace1534d
JH
315}
316
bef70b22
DB
317void add_to_alternates_file(const char *reference)
318{
319 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
acd3b9ec 320 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
bef70b22
DB
321 char *alt = mkpath("%s/objects\n", reference);
322 write_or_die(fd, alt, strlen(alt));
323 if (commit_lock_file(lock))
324 die("could not close alternates file");
325 if (alt_odb_tail)
326 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
327}
328
d79796bc
JH
329void foreach_alt_odb(alt_odb_fn fn, void *cb)
330{
331 struct alternate_object_database *ent;
332
333 prepare_alt_odb();
334 for (ent = alt_odb_list; ent; ent = ent->next)
335 if (fn(ent, cb))
336 return;
337}
338
c2f493a4
MW
339void prepare_alt_odb(void)
340{
554fe20d 341 const char *alt;
c2f493a4 342
7dc24aa5
SP
343 if (alt_odb_tail)
344 return;
345
c2f493a4
MW
346 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
347 if (!alt) alt = "";
348
c2f493a4 349 alt_odb_tail = &alt_odb_list;
80ba074f 350 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
351
352 read_info_alternates(get_object_directory(), 0);
353}
354
0f4dc14a 355static int has_loose_object_local(const unsigned char *sha1)
ace1534d 356{
ace1534d 357 char *name = sha1_file_name(sha1);
0f4dc14a
BC
358 return !access(name, F_OK);
359}
ace1534d 360
0f4dc14a
BC
361int has_loose_object_nonlocal(const unsigned char *sha1)
362{
363 struct alternate_object_database *alt;
9a217f2a 364 prepare_alt_odb();
d5a63b99 365 for (alt = alt_odb_list; alt; alt = alt->next) {
0f4dc14a 366 fill_sha1_path(alt->name, sha1);
c529d75a
LT
367 if (!access(alt->base, F_OK))
368 return 1;
ace1534d 369 }
c529d75a 370 return 0;
ace1534d
JH
371}
372
0f4dc14a
BC
373static int has_loose_object(const unsigned char *sha1)
374{
375 return has_loose_object_local(sha1) ||
376 has_loose_object_nonlocal(sha1);
377}
378
60bb8b14 379static unsigned int pack_used_ctr;
a53128b6
SP
380static unsigned int pack_mmap_calls;
381static unsigned int peak_pack_open_windows;
382static unsigned int pack_open_windows;
383static size_t peak_pack_mapped;
60bb8b14 384static size_t pack_mapped;
9a217f2a 385struct packed_git *packed_git;
1f688557 386
b79d18c9 387void pack_report(void)
a53128b6
SP
388{
389 fprintf(stderr,
e05db0fd
PR
390 "pack_report: getpagesize() = %10" SZ_FMT "\n"
391 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
392 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
393 sz_fmt(getpagesize()),
394 sz_fmt(packed_git_window_size),
395 sz_fmt(packed_git_limit));
a53128b6
SP
396 fprintf(stderr,
397 "pack_report: pack_used_ctr = %10u\n"
398 "pack_report: pack_mmap_calls = %10u\n"
399 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
400 "pack_report: pack_mapped = "
401 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
402 pack_used_ctr,
403 pack_mmap_calls,
404 pack_open_windows, peak_pack_open_windows,
9e42d6a1 405 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
406}
407
42873078 408static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
409{
410 void *idx_map;
42873078 411 struct pack_idx_header *hdr;
dc49cd76 412 size_t idx_size;
74e34e1f 413 uint32_t version, nr, i, *index;
e93ec6f9 414 int fd = open(path, O_RDONLY);
1f688557 415 struct stat st;
42873078 416
1f688557
JH
417 if (fd < 0)
418 return -1;
419 if (fstat(fd, &st)) {
420 close(fd);
421 return -1;
422 }
dc49cd76 423 idx_size = xsize_t(st.st_size);
2d88451b
SP
424 if (idx_size < 4 * 256 + 20 + 20) {
425 close(fd);
426 return error("index file %s is too small", path);
427 }
c4712e45 428 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 429 close(fd);
1f688557 430
42873078
NP
431 hdr = idx_map;
432 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
433 version = ntohl(hdr->idx_version);
434 if (version < 2 || version > 2) {
435 munmap(idx_map, idx_size);
6e1c2344 436 return error("index file %s is version %"PRIu32
74e34e1f
NP
437 " and is not supported by this binary"
438 " (try upgrading GIT to a newer version)",
439 path, version);
440 }
441 } else
442 version = 1;
df1b059d 443
1f688557 444 nr = 0;
42873078 445 index = idx_map;
74e34e1f
NP
446 if (version > 1)
447 index += 2; /* skip index header */
1f688557 448 for (i = 0; i < 256; i++) {
326bf396 449 uint32_t n = ntohl(index[i]);
2d88451b
SP
450 if (n < nr) {
451 munmap(idx_map, idx_size);
df1b059d 452 return error("non-monotonic index %s", path);
2d88451b 453 }
1f688557
JH
454 nr = n;
455 }
456
74e34e1f
NP
457 if (version == 1) {
458 /*
459 * Total size:
460 * - 256 index entries 4 bytes each
461 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
462 * - 20-byte SHA1 of the packfile
463 * - 20-byte SHA1 file checksum
464 */
465 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
466 munmap(idx_map, idx_size);
eef427a0 467 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
468 }
469 } else if (version == 2) {
470 /*
471 * Minimum size:
472 * - 8 bytes of header
473 * - 256 index entries 4 bytes each
474 * - 20-byte sha1 entry * nr
475 * - 4-byte crc entry * nr
476 * - 4-byte offset entry * nr
477 * - 20-byte SHA1 of the packfile
478 * - 20-byte SHA1 file checksum
479 * And after the 4-byte offset table might be a
480 * variable sized table containing 8-byte entries
481 * for offsets larger than 2^31.
482 */
483 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
484 unsigned long max_size = min_size;
485 if (nr)
486 max_size += (nr - 1)*8;
487 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 488 munmap(idx_map, idx_size);
eef427a0 489 return error("wrong index v2 file size in %s", path);
74e34e1f 490 }
7109c889
JH
491 if (idx_size != min_size &&
492 /*
493 * make sure we can deal with large pack offsets.
494 * 31-bit signed offset won't be enough, neither
495 * 32-bit unsigned one will be.
496 */
497 (sizeof(off_t) <= 4)) {
498 munmap(idx_map, idx_size);
499 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 500 }
2d88451b 501 }
1f688557 502
74e34e1f 503 p->index_version = version;
42873078
NP
504 p->index_data = idx_map;
505 p->index_size = idx_size;
57059091 506 p->num_objects = nr;
1f688557
JH
507 return 0;
508}
509
bc8e478a 510int open_pack_index(struct packed_git *p)
d079837e
SP
511{
512 char *idx_name;
513 int ret;
514
515 if (p->index_data)
516 return 0;
517
518 idx_name = xstrdup(p->pack_name);
519 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
520 ret = check_packed_git_idx(idx_name, p);
521 free(idx_name);
522 return ret;
523}
524
11daf39b
SP
525static void scan_windows(struct packed_git *p,
526 struct packed_git **lru_p,
527 struct pack_window **lru_w,
528 struct pack_window **lru_l)
1f688557 529{
11daf39b
SP
530 struct pack_window *w, *w_l;
531
532 for (w_l = NULL, w = p->windows; w; w = w->next) {
533 if (!w->inuse_cnt) {
534 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
535 *lru_p = p;
536 *lru_w = w;
537 *lru_l = w_l;
54044bf8 538 }
54044bf8 539 }
11daf39b 540 w_l = w;
f9253394 541 }
11daf39b
SP
542}
543
d1efefa4 544static int unuse_one_window(struct packed_git *current, int keep_fd)
11daf39b
SP
545{
546 struct packed_git *p, *lru_p = NULL;
547 struct pack_window *lru_w = NULL, *lru_l = NULL;
548
549 if (current)
550 scan_windows(current, &lru_p, &lru_w, &lru_l);
551 for (p = packed_git; p; p = p->next)
552 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
553 if (lru_p) {
554 munmap(lru_w->base, lru_w->len);
555 pack_mapped -= lru_w->len;
556 if (lru_l)
557 lru_l->next = lru_w->next;
558 else {
559 lru_p->windows = lru_w->next;
d1efefa4 560 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
54044bf8
SP
561 close(lru_p->pack_fd);
562 lru_p->pack_fd = -1;
563 }
564 }
565 free(lru_w);
a53128b6 566 pack_open_windows--;
54044bf8
SP
567 return 1;
568 }
569 return 0;
f9253394
JH
570}
571
d1efefa4 572void release_pack_memory(size_t need, int fd)
97bfeb34
SP
573{
574 size_t cur = pack_mapped;
d1efefa4 575 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
97bfeb34
SP
576 ; /* nothing */
577}
578
58ecbd5e
JN
579void *xmmap(void *start, size_t length,
580 int prot, int flags, int fd, off_t offset)
581{
582 void *ret = mmap(start, length, prot, flags, fd, offset);
583 if (ret == MAP_FAILED) {
584 if (!length)
585 return NULL;
586 release_pack_memory(length, fd);
587 ret = mmap(start, length, prot, flags, fd, offset);
588 if (ret == MAP_FAILED)
589 die_errno("Out of memory? mmap failed");
590 }
591 return ret;
592}
593
c9ced051
SP
594void close_pack_windows(struct packed_git *p)
595{
596 while (p->windows) {
597 struct pack_window *w = p->windows;
598
599 if (w->inuse_cnt)
600 die("pack '%s' still has open windows to it",
601 p->pack_name);
602 munmap(w->base, w->len);
603 pack_mapped -= w->len;
604 pack_open_windows--;
605 p->windows = w->next;
606 free(w);
607 }
608}
609
03e79c88 610void unuse_pack(struct pack_window **w_cursor)
f9253394 611{
03e79c88
SP
612 struct pack_window *w = *w_cursor;
613 if (w) {
614 w->inuse_cnt--;
615 *w_cursor = NULL;
616 }
1f688557
JH
617}
618
fa5fc15d
SP
619void close_pack_index(struct packed_git *p)
620{
621 if (p->index_data) {
622 munmap((void *)p->index_data, p->index_size);
623 p->index_data = NULL;
624 }
625}
626
c74faea1
NP
627/*
628 * This is used by git-repack in case a newly created pack happens to
629 * contain the same set of objects as an existing one. In that case
630 * the resulting file might be different even if its name would be the
631 * same. It is best to close any reference to the old pack before it is
632 * replaced on disk. Of course no index pointers nor windows for given pack
633 * must subsist at this point. If ever objects from this pack are requested
634 * again, the new version of the pack will be reinitialized through
635 * reprepare_packed_git().
636 */
637void free_pack_by_name(const char *pack_name)
638{
639 struct packed_git *p, **pp = &packed_git;
640
641 while (*pp) {
642 p = *pp;
643 if (strcmp(pack_name, p->pack_name) == 0) {
fa3a0c94 644 clear_delta_base_cache();
c74faea1
NP
645 close_pack_windows(p);
646 if (p->pack_fd != -1)
647 close(p->pack_fd);
fa5fc15d 648 close_pack_index(p);
c74faea1
NP
649 free(p->bad_object_sha1);
650 *pp = p->next;
651 free(p);
652 return;
653 }
654 pp = &p->next;
655 }
656}
657
3cf8b462
SP
658/*
659 * Do not call this directly as this leaks p->pack_fd on error return;
660 * call open_packed_git() instead.
661 */
662static int open_packed_git_1(struct packed_git *p)
1f688557 663{
9bc879c1
SP
664 struct stat st;
665 struct pack_header hdr;
666 unsigned char sha1[20];
667 unsigned char *idx_sha1;
2c039da8 668 long fd_flag;
9bc879c1 669
d079837e
SP
670 if (!p->index_data && open_pack_index(p))
671 return error("packfile %s index unavailable", p->pack_name);
672
9bc879c1 673 p->pack_fd = open(p->pack_name, O_RDONLY);
fd73ccf2
JS
674 while (p->pack_fd < 0 && errno == EMFILE && unuse_one_window(p, -1))
675 p->pack_fd = open(p->pack_name, O_RDONLY);
9bc879c1 676 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 677 return -1;
9bc879c1
SP
678
679 /* If we created the struct before we had the pack we lack size. */
bf592c50 680 if (!p->pack_size) {
bf592c50 681 if (!S_ISREG(st.st_mode))
072db278 682 return error("packfile %s not a regular file", p->pack_name);
bf592c50 683 p->pack_size = st.st_size;
9bc879c1 684 } else if (p->pack_size != st.st_size)
072db278 685 return error("packfile %s size changed", p->pack_name);
9bc879c1 686
2c039da8
JH
687 /* We leave these file descriptors open with sliding mmap;
688 * there is no point keeping them open across exec(), though.
689 */
690 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
691 if (fd_flag < 0)
072db278 692 return error("cannot determine file descriptor flags");
2c039da8
JH
693 fd_flag |= FD_CLOEXEC;
694 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 695 return error("cannot set FD_CLOEXEC");
2c039da8 696
9bc879c1 697 /* Verify we recognize this pack file format. */
e6e2bd62 698 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 699 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 700 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 701 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 702 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
703 return error("packfile %s is version %"PRIu32" and not"
704 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
705 p->pack_name, ntohl(hdr.hdr_version));
706
707 /* Verify the pack matches its index. */
57059091 708 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
709 return error("packfile %s claims to have %"PRIu32" objects"
710 " while index indicates %"PRIu32" objects",
57059091
NP
711 p->pack_name, ntohl(hdr.hdr_entries),
712 p->num_objects);
9bc879c1 713 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 714 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 715 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 716 return error("packfile %s signature is unavailable", p->pack_name);
42873078 717 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 718 if (hashcmp(sha1, idx_sha1))
072db278
SP
719 return error("packfile %s does not match index", p->pack_name);
720 return 0;
9bc879c1
SP
721}
722
3cf8b462
SP
723static int open_packed_git(struct packed_git *p)
724{
725 if (!open_packed_git_1(p))
726 return 0;
727 if (p->pack_fd != -1) {
728 close(p->pack_fd);
729 p->pack_fd = -1;
730 }
731 return -1;
732}
733
c4001d92 734static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
735{
736 /* We must promise at least 20 bytes (one hash) after the
737 * offset is available from this window, otherwise the offset
738 * is not actually in this window and a different window (which
739 * has that one hash excess) must be used. This is to support
740 * the object header and delta base parsing routines below.
741 */
742 off_t win_off = win->offset;
743 return win_off <= offset
744 && (offset + 20) <= (win_off + win->len);
745}
746
4b25d091 747unsigned char *use_pack(struct packed_git *p,
03e79c88 748 struct pack_window **w_cursor,
c4001d92 749 off_t offset,
03e79c88 750 unsigned int *left)
9bc879c1 751{
60bb8b14 752 struct pack_window *win = *w_cursor;
03e79c88 753
072db278
SP
754 if (p->pack_fd == -1 && open_packed_git(p))
755 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14 756
a9d98a14 757 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
758 * pointless to ask for an offset into the middle of that
759 * hash, and the in_window function above wouldn't match
760 * don't allow an offset too close to the end of the file.
761 */
762 if (offset > (p->pack_size - 20))
763 die("offset beyond end of packfile (truncated pack?)");
764
765 if (!win || !in_window(win, offset)) {
766 if (win)
767 win->inuse_cnt--;
768 for (win = p->windows; win; win = win->next) {
769 if (in_window(win, offset))
770 break;
771 }
772 if (!win) {
78a28df9 773 size_t window_align = packed_git_window_size / 2;
dc49cd76 774 off_t len;
60bb8b14 775 win = xcalloc(1, sizeof(*win));
78a28df9 776 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
777 len = p->pack_size - win->offset;
778 if (len > packed_git_window_size)
779 len = packed_git_window_size;
780 win->len = (size_t)len;
60bb8b14 781 pack_mapped += win->len;
11daf39b 782 while (packed_git_limit < pack_mapped
d1efefa4 783 && unuse_one_window(p, p->pack_fd))
60bb8b14 784 ; /* nothing */
c4712e45 785 win->base = xmmap(NULL, win->len,
60bb8b14
SP
786 PROT_READ, MAP_PRIVATE,
787 p->pack_fd, win->offset);
788 if (win->base == MAP_FAILED)
73b4e4be
SP
789 die("packfile %s cannot be mapped: %s",
790 p->pack_name,
791 strerror(errno));
a53128b6
SP
792 pack_mmap_calls++;
793 pack_open_windows++;
794 if (pack_mapped > peak_pack_mapped)
795 peak_pack_mapped = pack_mapped;
796 if (pack_open_windows > peak_pack_open_windows)
797 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
798 win->next = p->windows;
799 p->windows = win;
800 }
1f688557 801 }
03e79c88
SP
802 if (win != *w_cursor) {
803 win->last_used = pack_used_ctr++;
804 win->inuse_cnt++;
805 *w_cursor = win;
806 }
60bb8b14 807 offset -= win->offset;
03e79c88 808 if (left)
dc49cd76 809 *left = win->len - xsize_t(offset);
03e79c88 810 return win->base + offset;
1f688557
JH
811}
812
27d69a46
NP
813static struct packed_git *alloc_packed_git(int extra)
814{
815 struct packed_git *p = xmalloc(sizeof(*p) + extra);
816 memset(p, 0, sizeof(*p));
817 p->pack_fd = -1;
818 return p;
819}
820
42873078 821struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557
JH
822{
823 struct stat st;
27d69a46 824 struct packed_git *p = alloc_packed_git(path_len + 2);
1f688557 825
42873078
NP
826 /*
827 * Make sure a corresponding .pack file exists and that
828 * the index looks sane.
829 */
830 path_len -= strlen(".idx");
27d69a46
NP
831 if (path_len < 1) {
832 free(p);
1f688557 833 return NULL;
27d69a46 834 }
42873078 835 memcpy(p->pack_name, path, path_len);
8d25931d
BC
836
837 strcpy(p->pack_name + path_len, ".keep");
838 if (!access(p->pack_name, F_OK))
839 p->pack_keep = 1;
840
42873078 841 strcpy(p->pack_name + path_len, ".pack");
d079837e 842 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 843 free(p);
1f688557
JH
844 return NULL;
845 }
42873078 846
1f688557
JH
847 /* ok, it looks sane as far as we can check without
848 * actually mapping the pack file.
849 */
1f688557 850 p->pack_size = st.st_size;
9d835df2 851 p->pack_local = local;
b867092f 852 p->mtime = st.st_mtime;
42873078
NP
853 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
854 hashclr(p->sha1);
1f688557
JH
855 return p;
856}
857
7b64469a 858struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 859{
42873078 860 const char *path = sha1_pack_name(sha1);
27d69a46 861 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
bf592c50 862
27d69a46
NP
863 strcpy(p->pack_name, path);
864 hashcpy(p->sha1, sha1);
42873078
NP
865 if (check_packed_git_idx(idx_path, p)) {
866 free(p);
bf592c50 867 return NULL;
42873078 868 }
bf592c50 869
bf592c50
DB
870 return p;
871}
872
873void install_packed_git(struct packed_git *pack)
874{
875 pack->next = packed_git;
876 packed_git = pack;
877}
878
9d835df2 879static void prepare_packed_git_one(char *objdir, int local)
1f688557 880{
9cb18f56
JM
881 /* Ensure that this buffer is large enough so that we can
882 append "/pack/" without clobbering the stack even if
883 strlen(objdir) were PATH_MAX. */
884 char path[PATH_MAX + 1 + 4 + 1 + 1];
1f688557
JH
885 int len;
886 DIR *dir;
887 struct dirent *de;
888
889 sprintf(path, "%s/pack", objdir);
890 len = strlen(path);
891 dir = opendir(path);
fd73ccf2
JS
892 while (!dir && errno == EMFILE && unuse_one_window(packed_git, -1))
893 dir = opendir(path);
b5b16990 894 if (!dir) {
26125f6b 895 if (errno != ENOENT)
bd2afde8 896 error("unable to open object pack directory: %s: %s",
26125f6b 897 path, strerror(errno));
1f688557 898 return;
b5b16990 899 }
1f688557
JH
900 path[len++] = '/';
901 while ((de = readdir(dir)) != NULL) {
902 int namelen = strlen(de->d_name);
903 struct packed_git *p;
904
5bb1cda5 905 if (!has_extension(de->d_name, ".idx"))
1f688557
JH
906 continue;
907
9cb18f56
JM
908 if (len + namelen + 1 > sizeof(path))
909 continue;
910
54a15a8d 911 /* Don't reopen a pack we already have. */
1f688557 912 strcpy(path + len, de->d_name);
86f7780c
JK
913 for (p = packed_git; p; p = p->next) {
914 if (!memcmp(path, p->pack_name, len + namelen - 4))
915 break;
916 }
917 if (p)
918 continue;
54a15a8d
SP
919 /* See if it really is a valid .idx file with corresponding
920 * .pack file that we can map.
921 */
9d835df2 922 p = add_packed_git(path, len + namelen, local);
1f688557
JH
923 if (!p)
924 continue;
625e9421 925 install_packed_git(p);
1f688557 926 }
5b35bcd5 927 closedir(dir);
1f688557
JH
928}
929
b867092f
JH
930static int sort_pack(const void *a_, const void *b_)
931{
932 struct packed_git *a = *((struct packed_git **)a_);
933 struct packed_git *b = *((struct packed_git **)b_);
934 int st;
935
936 /*
937 * Local packs tend to contain objects specific to our
938 * variant of the project than remote ones. In addition,
939 * remote ones could be on a network mounted filesystem.
940 * Favor local ones for these reasons.
941 */
942 st = a->pack_local - b->pack_local;
943 if (st)
944 return -st;
945
946 /*
947 * Younger packs tend to contain more recent objects,
948 * and more recent objects tend to get accessed more
949 * often.
950 */
951 if (a->mtime < b->mtime)
952 return 1;
953 else if (a->mtime == b->mtime)
954 return 0;
955 return -1;
956}
957
958static void rearrange_packed_git(void)
959{
960 struct packed_git **ary, *p;
961 int i, n;
962
963 for (n = 0, p = packed_git; p; p = p->next)
964 n++;
965 if (n < 2)
966 return;
967
968 /* prepare an array of packed_git for easier sorting */
969 ary = xcalloc(n, sizeof(struct packed_git *));
970 for (n = 0, p = packed_git; p; p = p->next)
971 ary[n++] = p;
972
973 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
974
975 /* link them back again */
976 for (i = 0; i < n - 1; i++)
977 ary[i]->next = ary[i + 1];
978 ary[n - 1]->next = NULL;
979 packed_git = ary[0];
980
981 free(ary);
982}
983
637cdd9d 984static int prepare_packed_git_run_once = 0;
9a217f2a 985void prepare_packed_git(void)
1f688557 986{
d5a63b99 987 struct alternate_object_database *alt;
1f688557 988
637cdd9d 989 if (prepare_packed_git_run_once)
1f688557 990 return;
9d835df2 991 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 992 prepare_alt_odb();
d5a63b99 993 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 994 alt->name[-1] = 0;
9d835df2 995 prepare_packed_git_one(alt->base, 0);
1494e038 996 alt->name[-1] = '/';
1f688557 997 }
b867092f 998 rearrange_packed_git();
637cdd9d
JK
999 prepare_packed_git_run_once = 1;
1000}
1001
fc04c412 1002void reprepare_packed_git(void)
637cdd9d 1003{
4b480c67 1004 discard_revindex();
637cdd9d
JK
1005 prepare_packed_git_run_once = 0;
1006 prepare_packed_git();
1f688557
JH
1007}
1008
8eca0b47
NP
1009static void mark_bad_packed_object(struct packed_git *p,
1010 const unsigned char *sha1)
1011{
1012 unsigned i;
1013 for (i = 0; i < p->num_bad_objects; i++)
1014 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1015 return;
1016 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1017 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1018 p->num_bad_objects++;
1019}
1020
ac939109
NP
1021static int has_packed_and_bad(const unsigned char *sha1)
1022{
1023 struct packed_git *p;
1024 unsigned i;
1025
1026 for (p = packed_git; p; p = p->next)
1027 for (i = 0; i < p->num_bad_objects; i++)
1028 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1029 return 1;
1030 return 0;
1031}
1032
5d6ccf5c 1033int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
0fcfd160
LT
1034{
1035 unsigned char real_sha1[20];
7cfb5f36 1036 hash_sha1_file(map, size, type, real_sha1);
a89fccd2 1037 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1038}
1039
44d1c19e
LT
1040static int git_open_noatime(const char *name)
1041{
1042 static int sha1_file_open_flag = O_NOATIME;
1043 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1044
1045 /* Might the failure be due to O_NOATIME? */
1046 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1047 fd = open(name, O_RDONLY);
1048 if (fd >= 0)
1049 sha1_file_open_flag = 0;
1050 }
1051 return fd;
1052}
1053
1054static int open_sha1_file(const unsigned char *sha1)
1055{
1056 int fd;
1057 char *name = sha1_file_name(sha1);
1058 struct alternate_object_database *alt;
1059
1060 fd = git_open_noatime(name);
1061 if (fd >= 0)
1062 return fd;
1063
1064 prepare_alt_odb();
1065 errno = ENOENT;
1066 for (alt = alt_odb_list; alt; alt = alt->next) {
1067 name = alt->name;
1068 fill_sha1_path(name, sha1);
1069 fd = git_open_noatime(alt->base);
1070 if (fd >= 0)
1071 return fd;
1072 }
1073 return -1;
1074}
1075
4175e9e3 1076static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1077{
0fcfd160 1078 void *map;
144bde78 1079 int fd;
ace1534d 1080
44d1c19e
LT
1081 fd = open_sha1_file(sha1);
1082 map = NULL;
1083 if (fd >= 0) {
1084 struct stat st;
0fcfd160 1085
44d1c19e
LT
1086 if (!fstat(fd, &st)) {
1087 *size = xsize_t(st.st_size);
1088 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1089 }
44d1c19e 1090 close(fd);
0fcfd160 1091 }
0fcfd160
LT
1092 return map;
1093}
1094
726f852b 1095static int legacy_loose_object(unsigned char *map)
bb6b8e4f
JH
1096{
1097 unsigned int word;
1098
1099 /*
1100 * Is it a zlib-compressed buffer? If so, the first byte
1101 * must be 0x78 (15-bit window size, deflated), and the
1102 * first 16-bit word is evenly divisible by 31
1103 */
1104 word = (map[0] << 8) + map[1];
1105 if (map[0] == 0x78 && !(word % 31))
1106 return 1;
1107 else
1108 return 0;
1109}
1110
09ded04b
NP
1111unsigned long unpack_object_header_buffer(const unsigned char *buf,
1112 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1113{
ad1ed5ee 1114 unsigned shift;
48fb7deb 1115 unsigned long size, c;
ad1ed5ee
JH
1116 unsigned long used = 0;
1117
1118 c = buf[used++];
1119 *type = (c >> 4) & 7;
1120 size = c & 15;
1121 shift = 4;
1122 while (c & 0x80) {
f630cfda 1123 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1124 error("bad object header");
ad1ed5ee 1125 return 0;
09ded04b 1126 }
ad1ed5ee
JH
1127 c = buf[used++];
1128 size += (c & 0x7f) << shift;
1129 shift += 7;
1130 }
1131 *sizep = size;
1132 return used;
1133}
1134
1135static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1136{
1137 unsigned long size, used;
1138 static const char valid_loose_object_type[8] = {
1139 0, /* OBJ_EXT */
1140 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1141 0, /* "delta" and others are invalid in a loose object */
93821bd9 1142 };
ad1ed5ee 1143 enum object_type type;
93821bd9 1144
c4483576
LT
1145 /* Get the data stream */
1146 memset(stream, 0, sizeof(*stream));
1147 stream->next_in = map;
1148 stream->avail_in = mapsize;
1149 stream->next_out = buffer;
93821bd9
LT
1150 stream->avail_out = bufsiz;
1151
bb6b8e4f 1152 if (legacy_loose_object(map)) {
39c68542
LT
1153 git_inflate_init(stream);
1154 return git_inflate(stream, 0);
93821bd9
LT
1155 }
1156
726f852b
NP
1157
1158 /*
1159 * There used to be a second loose object header format which
1160 * was meant to mimic the in-pack format, allowing for direct
1161 * copy of the object data. This format turned up not to be
1162 * really worth it and we don't write it any longer. But we
1163 * can still read it.
1164 */
09ded04b 1165 used = unpack_object_header_buffer(map, mapsize, &type, &size);
ad1ed5ee 1166 if (!used || !valid_loose_object_type[type])
93821bd9 1167 return -1;
ad1ed5ee
JH
1168 map += used;
1169 mapsize -= used;
93821bd9
LT
1170
1171 /* Set up the stream for the rest.. */
1172 stream->next_in = map;
1173 stream->avail_in = mapsize;
39c68542 1174 git_inflate_init(stream);
93821bd9
LT
1175
1176 /* And generate the fake traditional header */
ad1ed5ee 1177 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
df843662 1178 typename(type), size);
93821bd9 1179 return 0;
c4483576
LT
1180}
1181
7efbff75 1182static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1183{
1184 int bytes = strlen(buffer) + 1;
3aee68aa 1185 unsigned char *buf = xmallocz(size);
93821bd9 1186 unsigned long n;
7efbff75 1187 int status = Z_OK;
5180cacc 1188
93821bd9
LT
1189 n = stream->total_out - bytes;
1190 if (n > size)
1191 n = size;
1192 memcpy(buf, (char *) buffer + bytes, n);
1193 bytes = n;
456cdf6e
LT
1194 if (bytes <= size) {
1195 /*
1196 * The above condition must be (bytes <= size), not
1197 * (bytes < size). In other words, even though we
1198 * expect no more output and set avail_out to zer0,
1199 * the input zlib stream may have bytes that express
1200 * "this concludes the stream", and we *do* want to
1201 * eat that input.
1202 *
1203 * Otherwise we would not be able to test that we
1204 * consumed all the input to reach the expected size;
1205 * we also want to check that zlib tells us that all
1206 * went well with status == Z_STREAM_END at the end.
1207 */
5180cacc
LT
1208 stream->next_out = buf + bytes;
1209 stream->avail_out = size - bytes;
7efbff75 1210 while (status == Z_OK)
39c68542 1211 status = git_inflate(stream, Z_FINISH);
5180cacc 1212 }
456cdf6e 1213 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1214 git_inflate_end(stream);
7efbff75
JH
1215 return buf;
1216 }
1217
1218 if (status < 0)
1219 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1220 else if (stream->avail_in)
1221 error("garbage at end of loose object '%s'",
1222 sha1_to_hex(sha1));
1223 free(buf);
1224 return NULL;
5180cacc
LT
1225}
1226
1227/*
1228 * We used to just use "sscanf()", but that's actually way
1229 * too permissive for what we want to check. So do an anal
1230 * object header parse by hand.
1231 */
21666f1a 1232static int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1233{
21666f1a 1234 char type[10];
5180cacc
LT
1235 int i;
1236 unsigned long size;
1237
1238 /*
a6080a0a 1239 * The type can be at most ten bytes (including the
5180cacc 1240 * terminating '\0' that we add), and is followed by
21666f1a 1241 * a space.
5180cacc 1242 */
21666f1a 1243 i = 0;
5180cacc
LT
1244 for (;;) {
1245 char c = *hdr++;
1246 if (c == ' ')
1247 break;
21666f1a
NP
1248 type[i++] = c;
1249 if (i >= sizeof(type))
5180cacc 1250 return -1;
5180cacc 1251 }
21666f1a 1252 type[i] = 0;
5180cacc
LT
1253
1254 /*
1255 * The length must follow immediately, and be in canonical
1256 * decimal format (ie "010" is not valid).
1257 */
1258 size = *hdr++ - '0';
1259 if (size > 9)
1260 return -1;
1261 if (size) {
1262 for (;;) {
1263 unsigned long c = *hdr - '0';
1264 if (c > 9)
1265 break;
1266 hdr++;
1267 size = size * 10 + c;
1268 }
1269 }
1270 *sizep = size;
1271
1272 /*
1273 * The length must be followed by a zero byte
1274 */
21666f1a 1275 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1276}
1277
7efbff75 1278static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1279{
5180cacc 1280 int ret;
0fcfd160 1281 z_stream stream;
5180cacc 1282 char hdr[8192];
0fcfd160 1283
5180cacc 1284 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1285 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1286 return NULL;
1287
7efbff75 1288 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1289}
1290
54dab52a
NP
1291unsigned long get_size_from_delta(struct packed_git *p,
1292 struct pack_window **w_curs,
1293 off_t curpos)
1294{
1295 const unsigned char *data;
1296 unsigned char delta_head[20], *in;
1297 z_stream stream;
1298 int st;
1299
1300 memset(&stream, 0, sizeof(stream));
1301 stream.next_out = delta_head;
1302 stream.avail_out = sizeof(delta_head);
1303
39c68542 1304 git_inflate_init(&stream);
54dab52a
NP
1305 do {
1306 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1307 stream.next_in = in;
39c68542 1308 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1309 curpos += stream.next_in - in;
1310 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1311 stream.total_out < sizeof(delta_head));
39c68542 1312 git_inflate_end(&stream);
3d77d877
NP
1313 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1314 error("delta data unpack-initial failed");
1315 return 0;
1316 }
54dab52a
NP
1317
1318 /* Examine the initial part of the delta to figure out
1319 * the result size.
1320 */
1321 data = delta_head;
1322
1323 /* ignore base size */
1324 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1325
1326 /* Read the result size */
1327 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1328}
1329
c4001d92 1330static off_t get_delta_base(struct packed_git *p,
03e79c88 1331 struct pack_window **w_curs,
c4001d92 1332 off_t *curpos,
21666f1a 1333 enum object_type type,
c4001d92 1334 off_t delta_obj_offset)
eb32d236 1335{
2b87c45b 1336 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1337 off_t base_offset;
eb32d236 1338
8d8a4ea5
SP
1339 /* use_pack() assured us we have [base_info, base_info + 20)
1340 * as a range that we can look at without walking off the
1341 * end of the mapped window. Its actually the hash size
1342 * that is assured. An OFS_DELTA longer than the hash size
1343 * is stupid, as then a REF_DELTA would be smaller to store.
1344 */
21666f1a 1345 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1346 unsigned used = 0;
1347 unsigned char c = base_info[used++];
1348 base_offset = c & 127;
1349 while (c & 128) {
1350 base_offset += 1;
8723f216 1351 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1352 return 0; /* overflow */
eb32d236
NP
1353 c = base_info[used++];
1354 base_offset = (base_offset << 7) + (c & 127);
1355 }
1356 base_offset = delta_obj_offset - base_offset;
d8f32556 1357 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1358 return 0; /* out of bound */
2b87c45b 1359 *curpos += used;
21666f1a 1360 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1361 /* The base entry _must_ be in the same pack */
1362 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1363 *curpos += 20;
eb32d236
NP
1364 } else
1365 die("I am totally screwed");
2b87c45b 1366 return base_offset;
eb32d236
NP
1367}
1368
f3bf9224 1369/* forward declaration for a mutually recursive function */
c4001d92 1370static int packed_object_info(struct packed_git *p, off_t offset,
21666f1a 1371 unsigned long *sizep);
f3bf9224 1372
43057304 1373static int packed_delta_info(struct packed_git *p,
03e79c88 1374 struct pack_window **w_curs,
c4001d92 1375 off_t curpos,
21666f1a 1376 enum object_type type,
c4001d92 1377 off_t obj_offset,
43057304 1378 unsigned long *sizep)
5db47c2b 1379{
c4001d92 1380 off_t base_offset;
f3bf9224 1381
21666f1a 1382 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
3d77d877
NP
1383 if (!base_offset)
1384 return OBJ_BAD;
21666f1a 1385 type = packed_object_info(p, base_offset, NULL);
3d77d877 1386 if (type <= OBJ_NONE) {
08698b1e
NP
1387 struct revindex_entry *revidx;
1388 const unsigned char *base_sha1;
1389 revidx = find_pack_revindex(p, base_offset);
1390 if (!revidx)
1391 return OBJ_BAD;
1392 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
3d77d877
NP
1393 mark_bad_packed_object(p, base_sha1);
1394 type = sha1_object_info(base_sha1, NULL);
1395 if (type <= OBJ_NONE)
1396 return OBJ_BAD;
1397 }
f3bf9224 1398
c62266f3
JH
1399 /* We choose to only get the type of the base object and
1400 * ignore potentially corrupt pack file that expects the delta
1401 * based on a base with a wrong size. This saves tons of
1402 * inflate() calls.
1403 */
3d77d877 1404 if (sizep) {
54dab52a 1405 *sizep = get_size_from_delta(p, w_curs, curpos);
3d77d877
NP
1406 if (*sizep == 0)
1407 type = OBJ_BAD;
1408 }
21666f1a
NP
1409
1410 return type;
5db47c2b
JH
1411}
1412
2b87c45b
NP
1413static int unpack_object_header(struct packed_git *p,
1414 struct pack_window **w_curs,
c4001d92 1415 off_t *curpos,
2b87c45b 1416 unsigned long *sizep)
a733cb60 1417{
03e79c88
SP
1418 unsigned char *base;
1419 unsigned int left;
ad1ed5ee 1420 unsigned long used;
2b87c45b 1421 enum object_type type;
a733cb60 1422
8d8a4ea5
SP
1423 /* use_pack() assures us we have [base, base + 20) available
1424 * as a range that we can look at at. (Its actually the hash
3dff5379 1425 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1426 * the maximum deflated object size is 2^137, which is just
1427 * insane, so we know won't exceed what we have been given.
1428 */
2b87c45b 1429 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1430 used = unpack_object_header_buffer(base, left, &type, sizep);
1431 if (!used) {
1432 type = OBJ_BAD;
1433 } else
1434 *curpos += used;
ad1ed5ee 1435
2b87c45b 1436 return type;
a733cb60
LT
1437}
1438
21666f1a 1439const char *packed_object_info_detail(struct packed_git *p,
c4001d92 1440 off_t obj_offset,
21666f1a
NP
1441 unsigned long *size,
1442 unsigned long *store_size,
1443 unsigned int *delta_chain_length,
1444 unsigned char *base_sha1)
ad8c80a5 1445{
03e79c88 1446 struct pack_window *w_curs = NULL;
c4001d92
SP
1447 off_t curpos;
1448 unsigned long dummy;
43057304 1449 unsigned char *next_sha1;
21666f1a 1450 enum object_type type;
70f5d5d3 1451 struct revindex_entry *revidx;
ad8c80a5 1452
43057304 1453 *delta_chain_length = 0;
2b87c45b 1454 curpos = obj_offset;
21666f1a 1455 type = unpack_object_header(p, &w_curs, &curpos, size);
43057304 1456
70f5d5d3
NP
1457 revidx = find_pack_revindex(p, obj_offset);
1458 *store_size = revidx[1].offset - obj_offset;
1459
43057304 1460 for (;;) {
21666f1a 1461 switch (type) {
43057304 1462 default:
08a19d87 1463 die("pack %s contains unknown object type %d",
21666f1a 1464 p->pack_name, type);
43057304
NP
1465 case OBJ_COMMIT:
1466 case OBJ_TREE:
1467 case OBJ_BLOB:
1468 case OBJ_TAG:
03e79c88 1469 unuse_pack(&w_curs);
21666f1a 1470 return typename(type);
eb32d236 1471 case OBJ_OFS_DELTA:
21666f1a 1472 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
8eca0b47
NP
1473 if (!obj_offset)
1474 die("pack %s contains bad delta base reference of type %s",
1475 p->pack_name, typename(type));
eb32d236 1476 if (*delta_chain_length == 0) {
70f5d5d3
NP
1477 revidx = find_pack_revindex(p, obj_offset);
1478 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
eb32d236
NP
1479 }
1480 break;
1481 case OBJ_REF_DELTA:
2b87c45b 1482 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
43057304
NP
1483 if (*delta_chain_length == 0)
1484 hashcpy(base_sha1, next_sha1);
2b87c45b 1485 obj_offset = find_pack_entry_one(next_sha1, p);
43057304
NP
1486 break;
1487 }
43057304 1488 (*delta_chain_length)++;
2b87c45b 1489 curpos = obj_offset;
21666f1a 1490 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
ad8c80a5 1491 }
ad8c80a5
JH
1492}
1493
c4001d92 1494static int packed_object_info(struct packed_git *p, off_t obj_offset,
21666f1a 1495 unsigned long *sizep)
1f688557 1496{
03e79c88 1497 struct pack_window *w_curs = NULL;
c4001d92
SP
1498 unsigned long size;
1499 off_t curpos = obj_offset;
21666f1a 1500 enum object_type type;
5db47c2b 1501
21666f1a 1502 type = unpack_object_header(p, &w_curs, &curpos, &size);
5db47c2b 1503
21666f1a 1504 switch (type) {
eb32d236
NP
1505 case OBJ_OFS_DELTA:
1506 case OBJ_REF_DELTA:
21666f1a
NP
1507 type = packed_delta_info(p, &w_curs, curpos,
1508 type, obj_offset, sizep);
2b87c45b 1509 break;
a733cb60 1510 case OBJ_COMMIT:
a733cb60 1511 case OBJ_TREE:
a733cb60 1512 case OBJ_BLOB:
a733cb60 1513 case OBJ_TAG:
2b87c45b
NP
1514 if (sizep)
1515 *sizep = size;
a69d0943 1516 break;
1f688557 1517 default:
3d77d877
NP
1518 error("unknown object type %i at offset %"PRIuMAX" in %s",
1519 type, (uintmax_t)obj_offset, p->pack_name);
1520 type = OBJ_BAD;
1f688557 1521 }
2b87c45b 1522 unuse_pack(&w_curs);
21666f1a 1523 return type;
1f688557
JH
1524}
1525
eb950c19 1526static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1527 struct pack_window **w_curs,
c4001d92 1528 off_t curpos,
eb950c19 1529 unsigned long size)
de530aaa
SP
1530{
1531 int st;
1532 z_stream stream;
079afb18 1533 unsigned char *buffer, *in;
de530aaa 1534
4ab07e4d 1535 buffer = xmallocz(size);
de530aaa 1536 memset(&stream, 0, sizeof(stream));
de530aaa 1537 stream.next_out = buffer;
39eea7bd 1538 stream.avail_out = size + 1;
de530aaa 1539
39c68542 1540 git_inflate_init(&stream);
079afb18 1541 do {
2b87c45b 1542 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 1543 stream.next_in = in;
39c68542 1544 st = git_inflate(&stream, Z_FINISH);
39eea7bd
JH
1545 if (!stream.avail_out)
1546 break; /* the payload is larger than it should be */
2b87c45b 1547 curpos += stream.next_in - in;
079afb18 1548 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 1549 git_inflate_end(&stream);
de530aaa
SP
1550 if ((st != Z_STREAM_END) || stream.total_out != size) {
1551 free(buffer);
1552 return NULL;
1553 }
1554
1555 return buffer;
1556}
1557
e5e01619
LT
1558#define MAX_DELTA_CACHE (256)
1559
18bdec11 1560static size_t delta_base_cached;
5e08ecbf
NP
1561
1562static struct delta_base_cache_lru_list {
1563 struct delta_base_cache_lru_list *prev;
1564 struct delta_base_cache_lru_list *next;
1565} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1566
e5e01619 1567static struct delta_base_cache_entry {
5e08ecbf
NP
1568 struct delta_base_cache_lru_list lru;
1569 void *data;
e5e01619
LT
1570 struct packed_git *p;
1571 off_t base_offset;
1572 unsigned long size;
e5e01619
LT
1573 enum object_type type;
1574} delta_base_cache[MAX_DELTA_CACHE];
1575
1576static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1577{
1578 unsigned long hash;
1579
1580 hash = (unsigned long)p + (unsigned long)base_offset;
1581 hash += (hash >> 8) + (hash >> 16);
3358004a 1582 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1583}
1584
62f255ad 1585static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 1586 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 1587{
e5e01619
LT
1588 void *ret;
1589 unsigned long hash = pack_entry_hash(p, base_offset);
1590 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1591
1592 ret = ent->data;
749bc58c
MV
1593 if (!ret || ent->p != p || ent->base_offset != base_offset)
1594 return unpack_entry(p, base_offset, type, base_size);
e5e01619 1595
18bdec11 1596 if (!keep_cache) {
a0cba108 1597 ent->data = NULL;
5e08ecbf
NP
1598 ent->lru.next->prev = ent->lru.prev;
1599 ent->lru.prev->next = ent->lru.next;
18bdec11 1600 delta_base_cached -= ent->size;
182af834
PH
1601 } else {
1602 ret = xmemdupz(ent->data, ent->size);
a0cba108 1603 }
e5e01619
LT
1604 *type = ent->type;
1605 *base_size = ent->size;
1606 return ret;
62f255ad
LT
1607}
1608
18bdec11
SP
1609static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1610{
1611 if (ent->data) {
1612 free(ent->data);
1613 ent->data = NULL;
5e08ecbf
NP
1614 ent->lru.next->prev = ent->lru.prev;
1615 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
1616 delta_base_cached -= ent->size;
1617 }
1618}
1619
3d20c636
SP
1620void clear_delta_base_cache(void)
1621{
1622 unsigned long p;
1623 for (p = 0; p < MAX_DELTA_CACHE; p++)
1624 release_delta_base_cache(&delta_base_cache[p]);
1625}
1626
62f255ad
LT
1627static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1628 void *base, unsigned long base_size, enum object_type type)
1629{
5e08ecbf 1630 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 1631 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 1632 struct delta_base_cache_lru_list *lru;
e5e01619 1633
18bdec11
SP
1634 release_delta_base_cache(ent);
1635 delta_base_cached += base_size;
5e08ecbf
NP
1636
1637 for (lru = delta_base_cache_lru.next;
1638 delta_base_cached > delta_base_cache_limit
1639 && lru != &delta_base_cache_lru;
1640 lru = lru->next) {
1641 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
1642 if (f->type == OBJ_BLOB)
1643 release_delta_base_cache(f);
1644 }
5e08ecbf
NP
1645 for (lru = delta_base_cache_lru.next;
1646 delta_base_cached > delta_base_cache_limit
1647 && lru != &delta_base_cache_lru;
1648 lru = lru->next) {
1649 struct delta_base_cache_entry *f = (void *)lru;
1650 release_delta_base_cache(f);
1651 }
18bdec11 1652
e5e01619
LT
1653 ent->p = p;
1654 ent->base_offset = base_offset;
1655 ent->type = type;
1656 ent->data = base;
1657 ent->size = base_size;
5e08ecbf
NP
1658 ent->lru.next = &delta_base_cache_lru;
1659 ent->lru.prev = delta_base_cache_lru.prev;
1660 delta_base_cache_lru.prev->next = &ent->lru;
1661 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
1662}
1663
c2c5b270
CC
1664static void *read_object(const unsigned char *sha1, enum object_type *type,
1665 unsigned long *size);
1666
eb950c19 1667static void *unpack_delta_entry(struct packed_git *p,
03e79c88 1668 struct pack_window **w_curs,
c4001d92 1669 off_t curpos,
1f688557 1670 unsigned long delta_size,
c4001d92 1671 off_t obj_offset,
21666f1a 1672 enum object_type *type,
eb950c19 1673 unsigned long *sizep)
1f688557 1674{
7c3e8be3 1675 void *delta_data, *result, *base;
c4001d92
SP
1676 unsigned long base_size;
1677 off_t base_offset;
43057304 1678
21666f1a 1679 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
8eca0b47
NP
1680 if (!base_offset) {
1681 error("failed to validate delta base reference "
1682 "at offset %"PRIuMAX" from %s",
1683 (uintmax_t)curpos, p->pack_name);
1684 return NULL;
1685 }
eac12e2d 1686 unuse_pack(w_curs);
a0cba108 1687 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
8eca0b47
NP
1688 if (!base) {
1689 /*
1690 * We're probably in deep shit, but let's try to fetch
1691 * the required base anyway from another pack or loose.
1692 * This is costly but should happen only in the presence
1693 * of a corrupted pack, and is better than failing outright.
1694 */
08698b1e
NP
1695 struct revindex_entry *revidx;
1696 const unsigned char *base_sha1;
1697 revidx = find_pack_revindex(p, base_offset);
1698 if (!revidx)
1699 return NULL;
1700 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
8eca0b47
NP
1701 error("failed to read delta base object %s"
1702 " at offset %"PRIuMAX" from %s",
1703 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1704 p->pack_name);
1705 mark_bad_packed_object(p, base_sha1);
ac939109 1706 base = read_object(base_sha1, type, &base_size);
8eca0b47
NP
1707 if (!base)
1708 return NULL;
1709 }
67686d95 1710
2b87c45b 1711 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
8eca0b47
NP
1712 if (!delta_data) {
1713 error("failed to unpack compressed delta "
1714 "at offset %"PRIuMAX" from %s",
1715 (uintmax_t)curpos, p->pack_name);
1716 free(base);
1717 return NULL;
1718 }
1f688557
JH
1719 result = patch_delta(base, base_size,
1720 delta_data, delta_size,
21666f1a 1721 sizep);
1f688557
JH
1722 if (!result)
1723 die("failed to apply delta");
1724 free(delta_data);
62f255ad 1725 add_delta_base_cache(p, base_offset, base, base_size, *type);
1f688557
JH
1726 return result;
1727}
1728
0e8189e2
NP
1729int do_check_packed_object_crc;
1730
c4001d92 1731void *unpack_entry(struct packed_git *p, off_t obj_offset,
21666f1a 1732 enum object_type *type, unsigned long *sizep)
1f688557 1733{
03e79c88 1734 struct pack_window *w_curs = NULL;
c4001d92 1735 off_t curpos = obj_offset;
21666f1a 1736 void *data;
1f688557 1737
0e8189e2
NP
1738 if (do_check_packed_object_crc && p->index_version > 1) {
1739 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1740 unsigned long len = revidx[1].offset - obj_offset;
1741 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1742 const unsigned char *sha1 =
1743 nth_packed_object_sha1(p, revidx->nr);
1744 error("bad packed object CRC for %s",
1745 sha1_to_hex(sha1));
1746 mark_bad_packed_object(p, sha1);
f755bb99 1747 unuse_pack(&w_curs);
0e8189e2
NP
1748 return NULL;
1749 }
1750 }
1751
21666f1a
NP
1752 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1753 switch (*type) {
eb32d236
NP
1754 case OBJ_OFS_DELTA:
1755 case OBJ_REF_DELTA:
21666f1a
NP
1756 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1757 obj_offset, type, sizep);
4d703a1a 1758 break;
a733cb60 1759 case OBJ_COMMIT:
a733cb60 1760 case OBJ_TREE:
a733cb60 1761 case OBJ_BLOB:
a733cb60 1762 case OBJ_TAG:
21666f1a 1763 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
4d703a1a 1764 break;
1f688557 1765 default:
8eca0b47
NP
1766 data = NULL;
1767 error("unknown object type %i at offset %"PRIuMAX" in %s",
1768 *type, (uintmax_t)obj_offset, p->pack_name);
1f688557 1769 }
03e79c88 1770 unuse_pack(&w_curs);
21666f1a 1771 return data;
1f688557
JH
1772}
1773
d079837e 1774const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 1775 uint32_t n)
9a217f2a 1776{
42873078 1777 const unsigned char *index = p->index_data;
d079837e
SP
1778 if (!index) {
1779 if (open_pack_index(p))
1780 return NULL;
1781 index = p->index_data;
1782 }
57059091 1783 if (n >= p->num_objects)
d72308e0 1784 return NULL;
74e34e1f
NP
1785 index += 4 * 256;
1786 if (p->index_version == 1) {
1787 return index + 24 * n + 4;
1788 } else {
1789 index += 8;
1790 return index + 20 * n;
1791 }
1792}
1793
99093238 1794off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
1795{
1796 const unsigned char *index = p->index_data;
1797 index += 4 * 256;
1798 if (p->index_version == 1) {
1799 return ntohl(*((uint32_t *)(index + 24 * n)));
1800 } else {
1801 uint32_t off;
1802 index += 8 + p->num_objects * (20 + 4);
1803 off = ntohl(*((uint32_t *)(index + 4 * n)));
1804 if (!(off & 0x80000000))
1805 return off;
1806 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1807 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1808 ntohl(*((uint32_t *)(index + 4)));
1809 }
9a217f2a
JH
1810}
1811
c4001d92 1812off_t find_pack_entry_one(const unsigned char *sha1,
43057304 1813 struct packed_git *p)
1f688557 1814{
42873078 1815 const uint32_t *level1_ofs = p->index_data;
42873078 1816 const unsigned char *index = p->index_data;
628522ec
JH
1817 unsigned hi, lo, stride;
1818 static int use_lookup = -1;
1819 static int debug_lookup = -1;
1820
1821 if (debug_lookup < 0)
1822 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 1823
d079837e
SP
1824 if (!index) {
1825 if (open_pack_index(p))
1826 return 0;
1827 level1_ofs = p->index_data;
1828 index = p->index_data;
1829 }
74e34e1f
NP
1830 if (p->index_version > 1) {
1831 level1_ofs += 2;
1832 index += 8;
1833 }
42873078 1834 index += 4 * 256;
74e34e1f
NP
1835 hi = ntohl(level1_ofs[*sha1]);
1836 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
1837 if (p->index_version > 1) {
1838 stride = 20;
1839 } else {
1840 stride = 24;
1841 index += 4;
1842 }
1843
1844 if (debug_lookup)
6e1c2344 1845 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
1846 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1847
1848 if (use_lookup < 0)
1849 use_lookup = !!getenv("GIT_USE_LOOKUP");
1850 if (use_lookup) {
1851 int pos = sha1_entry_pos(index, stride, 0,
1852 lo, hi, p->num_objects, sha1);
1853 if (pos < 0)
1854 return 0;
1855 return nth_packed_object_offset(p, pos);
1856 }
1f688557
JH
1857
1858 do {
74e34e1f 1859 unsigned mi = (lo + hi) / 2;
628522ec
JH
1860 int cmp = hashcmp(index + mi * stride, sha1);
1861
1862 if (debug_lookup)
1863 printf("lo %u hi %u rg %u mi %u\n",
1864 lo, hi, hi - lo, mi);
43057304 1865 if (!cmp)
74e34e1f 1866 return nth_packed_object_offset(p, mi);
1f688557
JH
1867 if (cmp > 0)
1868 hi = mi;
1869 else
1870 lo = mi+1;
1871 } while (lo < hi);
1872 return 0;
1873}
1874
4d6acb70 1875static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557 1876{
f7c22cc6 1877 static struct packed_git *last_found = (void *)1;
1f688557 1878 struct packed_git *p;
c4001d92 1879 off_t offset;
43057304 1880
1f688557 1881 prepare_packed_git();
f7c22cc6
NP
1882 if (!packed_git)
1883 return 0;
1884 p = (last_found == (void *)1) ? packed_git : last_found;
1f688557 1885
f7c22cc6 1886 do {
8eca0b47
NP
1887 if (p->num_bad_objects) {
1888 unsigned i;
1889 for (i = 0; i < p->num_bad_objects; i++)
1890 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1891 goto next;
1892 }
1893
43057304
NP
1894 offset = find_pack_entry_one(sha1, p);
1895 if (offset) {
c715f783
SP
1896 /*
1897 * We are about to tell the caller where they can
1898 * locate the requested object. We better make
1899 * sure the packfile is still here and can be
1900 * accessed before supplying that answer, as
1901 * it may have been deleted since the index
1902 * was loaded!
1903 */
1904 if (p->pack_fd == -1 && open_packed_git(p)) {
1905 error("packfile %s cannot be accessed", p->pack_name);
f7c22cc6 1906 goto next;
c715f783 1907 }
43057304
NP
1908 e->offset = offset;
1909 e->p = p;
1910 hashcpy(e->sha1, sha1);
f7c22cc6 1911 last_found = p;
1f688557 1912 return 1;
43057304 1913 }
f7c22cc6
NP
1914
1915 next:
1916 if (p == last_found)
1917 p = packed_git;
1918 else
1919 p = p->next;
1920 if (p == last_found)
1921 p = p->next;
1922 } while (p);
1f688557
JH
1923 return 0;
1924}
1925
a6080a0a 1926struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
1927 struct packed_git *packs)
1928{
1929 struct packed_git *p;
bf592c50
DB
1930
1931 for (p = packs; p; p = p->next) {
43057304 1932 if (find_pack_entry_one(sha1, p))
bf592c50
DB
1933 return p;
1934 }
1935 return NULL;
21666f1a 1936
bf592c50
DB
1937}
1938
21666f1a 1939static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
65c2e0c3 1940{
36e4d74a 1941 int status;
65c2e0c3
JH
1942 unsigned long mapsize, size;
1943 void *map;
1944 z_stream stream;
d65a16f6 1945 char hdr[32];
65c2e0c3 1946
bb6b8e4f 1947 map = map_sha1_file(sha1, &mapsize);
f0df4ed5
JS
1948 if (!map)
1949 return error("unable to find %s", sha1_to_hex(sha1));
36e4d74a
JH
1950 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1951 status = error("unable to unpack %s header",
1952 sha1_to_hex(sha1));
21666f1a 1953 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 1954 status = error("unable to parse %s header", sha1_to_hex(sha1));
21666f1a
NP
1955 else if (sizep)
1956 *sizep = size;
39c68542 1957 git_inflate_end(&stream);
65c2e0c3
JH
1958 munmap(map, mapsize);
1959 return status;
1960}
1961
21666f1a 1962int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
f0df4ed5 1963{
f0df4ed5 1964 struct pack_entry e;
ddd63e64 1965 int status;
f0df4ed5 1966
cd673c1f 1967 if (!find_pack_entry(sha1, &e)) {
ddd63e64
SG
1968 /* Most likely it's a loose object. */
1969 status = sha1_loose_object_info(sha1, sizep);
1970 if (status >= 0)
1971 return status;
1972
1973 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 1974 reprepare_packed_git();
cd673c1f 1975 if (!find_pack_entry(sha1, &e))
ddd63e64 1976 return status;
f0df4ed5 1977 }
3d77d877
NP
1978
1979 status = packed_object_info(e.p, e.offset, sizep);
1980 if (status < 0) {
1981 mark_bad_packed_object(e.p, sha1);
1982 status = sha1_object_info(sha1, sizep);
1983 }
1984
1985 return status;
f0df4ed5
JS
1986}
1987
21666f1a
NP
1988static void *read_packed_sha1(const unsigned char *sha1,
1989 enum object_type *type, unsigned long *size)
1f688557
JH
1990{
1991 struct pack_entry e;
8eca0b47 1992 void *data;
1f688557 1993
cd673c1f 1994 if (!find_pack_entry(sha1, &e))
1f688557 1995 return NULL;
8eca0b47
NP
1996 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1997 if (!data) {
1998 /*
1999 * We're probably in deep shit, but let's try to fetch
2000 * the required object anyway from another pack or loose.
2001 * This should happen only in the presence of a corrupted
2002 * pack, and is better than failing outright.
2003 */
2004 error("failed to read object %s at offset %"PRIuMAX" from %s",
2005 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2006 mark_bad_packed_object(e.p, sha1);
ac939109 2007 data = read_object(sha1, type, size);
8eca0b47
NP
2008 }
2009 return data;
1f688557
JH
2010}
2011
d66b37bb
JH
2012/*
2013 * This is meant to hold a *small* number of objects that you would
2014 * want read_sha1_file() to be able to return, but yet you do not want
2015 * to write them into the object store (e.g. a browse-only
2016 * application).
2017 */
2018static struct cached_object {
2019 unsigned char sha1[20];
21666f1a 2020 enum object_type type;
d66b37bb
JH
2021 void *buf;
2022 unsigned long size;
2023} *cached_objects;
2024static int cached_object_nr, cached_object_alloc;
2025
346245a1 2026static struct cached_object empty_tree = {
14d9c578 2027 EMPTY_TREE_SHA1_BIN,
346245a1
JK
2028 OBJ_TREE,
2029 "",
2030 0
2031};
2032
d66b37bb
JH
2033static struct cached_object *find_cached_object(const unsigned char *sha1)
2034{
2035 int i;
2036 struct cached_object *co = cached_objects;
2037
2038 for (i = 0; i < cached_object_nr; i++, co++) {
2039 if (!hashcmp(co->sha1, sha1))
2040 return co;
2041 }
346245a1
JK
2042 if (!hashcmp(sha1, empty_tree.sha1))
2043 return &empty_tree;
d66b37bb
JH
2044 return NULL;
2045}
2046
21666f1a
NP
2047int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2048 unsigned char *sha1)
d66b37bb
JH
2049{
2050 struct cached_object *co;
2051
21666f1a 2052 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2053 if (has_sha1_file(sha1) || find_cached_object(sha1))
2054 return 0;
2055 if (cached_object_alloc <= cached_object_nr) {
2056 cached_object_alloc = alloc_nr(cached_object_alloc);
2057 cached_objects = xrealloc(cached_objects,
2058 sizeof(*cached_objects) *
2059 cached_object_alloc);
2060 }
2061 co = &cached_objects[cached_object_nr++];
2062 co->size = len;
21666f1a 2063 co->type = type;
efa13f7b
JH
2064 co->buf = xmalloc(len);
2065 memcpy(co->buf, buf, len);
d66b37bb
JH
2066 hashcpy(co->sha1, sha1);
2067 return 0;
2068}
2069
c2c5b270
CC
2070static void *read_object(const unsigned char *sha1, enum object_type *type,
2071 unsigned long *size)
0fcfd160
LT
2072{
2073 unsigned long mapsize;
2074 void *map, *buf;
d66b37bb
JH
2075 struct cached_object *co;
2076
2077 co = find_cached_object(sha1);
2078 if (co) {
21666f1a 2079 *type = co->type;
d66b37bb 2080 *size = co->size;
182af834 2081 return xmemdupz(co->buf, co->size);
d66b37bb 2082 }
0fcfd160 2083
8276c007
PE
2084 buf = read_packed_sha1(sha1, type, size);
2085 if (buf)
2086 return buf;
bb6b8e4f 2087 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2088 if (map) {
7efbff75 2089 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2090 munmap(map, mapsize);
2091 return buf;
2092 }
637cdd9d 2093 reprepare_packed_git();
8276c007 2094 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2095}
2096
f5552aee
CC
2097void *read_sha1_file_repl(const unsigned char *sha1,
2098 enum object_type *type,
2099 unsigned long *size,
2100 const unsigned char **replacement)
ac939109 2101{
68095570
CC
2102 const unsigned char *repl = lookup_replace_object(sha1);
2103 void *data = read_object(repl, type, size);
e8b15e61 2104 char *path;
68095570
CC
2105
2106 /* die if we replaced an object with one that does not exist */
2107 if (!data && repl != sha1)
2108 die("replacement %s not found for %s",
2109 sha1_to_hex(repl), sha1_to_hex(sha1));
2110
ac939109 2111 /* legacy behavior is to die on corrupted objects */
e8b15e61
ÆAB
2112 if (!data) {
2113 if (has_loose_object(repl)) {
2114 path = sha1_file_name(sha1);
2115 die("loose object %s (stored in %s) is corrupted", sha1_to_hex(repl), path);
2116 }
2117 if (has_packed_and_bad(repl)) {
2118 path = sha1_pack_name(sha1);
2119 die("packed object %s (stored in %s) is corrupted", sha1_to_hex(repl), path);
2120 }
2121 }
68095570 2122
f5552aee
CC
2123 if (replacement)
2124 *replacement = repl;
2125
ac939109
NP
2126 return data;
2127}
2128
40469ee9 2129void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2130 const char *required_type_name,
40469ee9
JH
2131 unsigned long *size,
2132 unsigned char *actual_sha1_return)
f4913f91 2133{
21666f1a 2134 enum object_type type, required_type;
f4913f91
JH
2135 void *buffer;
2136 unsigned long isize;
40469ee9 2137 unsigned char actual_sha1[20];
f4913f91 2138
21666f1a 2139 required_type = type_from_string(required_type_name);
e702496e 2140 hashcpy(actual_sha1, sha1);
40469ee9
JH
2141 while (1) {
2142 int ref_length = -1;
2143 const char *ref_type = NULL;
f4913f91 2144
21666f1a 2145 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2146 if (!buffer)
2147 return NULL;
21666f1a 2148 if (type == required_type) {
40469ee9
JH
2149 *size = isize;
2150 if (actual_sha1_return)
e702496e 2151 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2152 return buffer;
2153 }
2154 /* Handle references */
21666f1a 2155 else if (type == OBJ_COMMIT)
40469ee9 2156 ref_type = "tree ";
21666f1a 2157 else if (type == OBJ_TAG)
40469ee9
JH
2158 ref_type = "object ";
2159 else {
2160 free(buffer);
2161 return NULL;
2162 }
2163 ref_length = strlen(ref_type);
f4913f91 2164
50974ec9
MK
2165 if (ref_length + 40 > isize ||
2166 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2167 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2168 free(buffer);
2169 return NULL;
2170 }
1cf58e72 2171 free(buffer);
40469ee9
JH
2172 /* Now we have the ID of the referred-to object in
2173 * actual_sha1. Check again. */
f4913f91 2174 }
f4913f91
JH
2175}
2176
ce9fbf16 2177static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2178 const char *type, unsigned char *sha1,
d65a16f6 2179 char *hdr, int *hdrlen)
d410c0f5 2180{
9126f009 2181 git_SHA_CTX c;
d410c0f5
JH
2182
2183 /* Generate the header */
d65a16f6 2184 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
2185
2186 /* Sha1.. */
9126f009
NP
2187 git_SHA1_Init(&c);
2188 git_SHA1_Update(&c, hdr, *hdrlen);
2189 git_SHA1_Update(&c, buf, len);
2190 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2191}
2192
230f1322 2193/*
5a688fe4
JH
2194 * Move the just written object into its final resting place.
2195 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2196 * "moving" is only a part of what it does, when no patch between
2197 * master to pu changes the call sites of this function.
230f1322 2198 */
839837b9 2199int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322 2200{
e32c0a9c 2201 int ret = 0;
5a688fe4 2202
348df166 2203 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
2204 goto try_rename;
2205 else if (link(tmpfile, filename))
e32c0a9c 2206 ret = errno;
7ebb6fca
LT
2207
2208 /*
2209 * Coda hack - coda doesn't like cross-directory links,
2210 * so we fall back to a rename, which will mean that it
2211 * won't be able to check collisions, but that's not a
2212 * big deal.
2213 *
2214 * The same holds for FAT formatted media.
2215 *
3be1f18e 2216 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
2217 * left to unlink.
2218 */
2219 if (ret && ret != EEXIST) {
be66a6c4 2220 try_rename:
7ebb6fca 2221 if (!rename(tmpfile, filename))
3be1f18e 2222 goto out;
9e48b389 2223 ret = errno;
230f1322 2224 }
691f1a28 2225 unlink_or_warn(tmpfile);
230f1322
LT
2226 if (ret) {
2227 if (ret != EEXIST) {
916d081b 2228 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
230f1322
LT
2229 }
2230 /* FIXME!!! Collision check here ? */
2231 }
2232
3be1f18e 2233out:
5256b006 2234 if (adjust_shared_perm(filename))
5a688fe4 2235 return error("unable to set permission to '%s'", filename);
230f1322
LT
2236 return 0;
2237}
2238
4d548150
LT
2239static int write_buffer(int fd, const void *buf, size_t len)
2240{
d34cf19b 2241 if (write_in_full(fd, buf, len) < 0)
93822c22 2242 return error("file write error (%s)", strerror(errno));
4d548150
LT
2243 return 0;
2244}
2245
ce9fbf16 2246int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2247 unsigned char *sha1)
2248{
d65a16f6 2249 char hdr[32];
abdc3fc8
RS
2250 int hdrlen;
2251 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2252 return 0;
2253}
2254
e9039dd3
LT
2255/* Finalize a file on disk, and close it. */
2256static void close_sha1_file(int fd)
2257{
aafe9fba
LT
2258 if (fsync_object_files)
2259 fsync_or_die(fd, "sha1 file");
e9039dd3 2260 if (close(fd) != 0)
d824cbba 2261 die_errno("error when closing sha1 file");
e9039dd3
LT
2262}
2263
5723fe7e
LT
2264/* Size of directory component, including the ending '/' */
2265static inline int directory_size(const char *filename)
2266{
2267 const char *s = strrchr(filename, '/');
2268 if (!s)
2269 return 0;
2270 return s - filename + 1;
2271}
2272
2273/*
2274 * This creates a temporary file in the same directory as the final
2275 * 'filename'
2276 *
2277 * We want to avoid cross-directory filename renames, because those
2278 * can have problems on various filesystems (FAT, NFS, Coda).
2279 */
2280static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2281{
2282 int fd, dirlen = directory_size(filename);
2283
2284 if (dirlen + 20 > bufsiz) {
2285 errno = ENAMETOOLONG;
2286 return -1;
2287 }
2288 memcpy(buffer, filename, dirlen);
2289 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
5256b006 2290 fd = git_mkstemp_mode(buffer, 0444);
cbacbf4e 2291 if (fd < 0 && dirlen && errno == ENOENT) {
5723fe7e 2292 /* Make sure the directory exists */
6ff6af62 2293 memcpy(buffer, filename, dirlen);
5723fe7e 2294 buffer[dirlen-1] = 0;
3bfaf018 2295 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
5723fe7e
LT
2296 return -1;
2297
2298 /* Try again */
2299 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
5256b006 2300 fd = git_mkstemp_mode(buffer, 0444);
5723fe7e
LT
2301 }
2302 return fd;
2303}
2304
bbac7311 2305static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 2306 const void *buf, unsigned long len, time_t mtime)
0fcfd160 2307{
915308b1 2308 int fd, ret;
9892beba 2309 unsigned char compressed[4096];
0fcfd160 2310 z_stream stream;
748af44c
NP
2311 git_SHA_CTX c;
2312 unsigned char parano_sha1[20];
706bc531 2313 char *filename;
aac17941 2314 static char tmpfile[PATH_MAX];
a44c9a5e 2315
972a9155 2316 filename = sha1_file_name(sha1);
5723fe7e 2317 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
fd73ccf2
JS
2318 while (fd < 0 && errno == EMFILE && unuse_one_window(packed_git, -1))
2319 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
aac17941 2320 if (fd < 0) {
35243577 2321 if (errno == EACCES)
916d081b
PB
2322 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2323 else
2324 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
aac17941
LT
2325 }
2326
0fcfd160
LT
2327 /* Set it up */
2328 memset(&stream, 0, sizeof(stream));
12f6c308 2329 deflateInit(&stream, zlib_compression_level);
0fcfd160 2330 stream.next_out = compressed;
9892beba 2331 stream.avail_out = sizeof(compressed);
748af44c 2332 git_SHA1_Init(&c);
a44c9a5e
LT
2333
2334 /* First header.. */
d65a16f6 2335 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2336 stream.avail_in = hdrlen;
726f852b
NP
2337 while (deflate(&stream, 0) == Z_OK)
2338 /* nothing */;
748af44c 2339 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
2340
2341 /* Then the data itself.. */
c00e657d 2342 stream.next_in = (void *)buf;
a44c9a5e 2343 stream.avail_in = len;
9892beba 2344 do {
748af44c 2345 unsigned char *in0 = stream.next_in;
9892beba 2346 ret = deflate(&stream, Z_FINISH);
748af44c 2347 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
2348 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2349 die("unable to write sha1 file");
2350 stream.next_out = compressed;
2351 stream.avail_out = sizeof(compressed);
2352 } while (ret == Z_OK);
2353
ac54c277
LT
2354 if (ret != Z_STREAM_END)
2355 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
ac54c277
LT
2356 ret = deflateEnd(&stream);
2357 if (ret != Z_OK)
2358 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
2359 git_SHA1_Final(parano_sha1, &c);
2360 if (hashcmp(sha1, parano_sha1) != 0)
2361 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 2362
e9039dd3 2363 close_sha1_file(fd);
0fcfd160 2364
bbac7311
NP
2365 if (mtime) {
2366 struct utimbuf utb;
2367 utb.actime = mtime;
2368 utb.modtime = mtime;
2369 if (utime(tmpfile, &utb) < 0)
2370 warning("failed utime() on %s: %s",
2371 tmpfile, strerror(errno));
2372 }
2373
230f1322 2374 return move_temp_to_file(tmpfile, filename);
0fcfd160 2375}
8237b185 2376
c00e657d 2377int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
bbac7311
NP
2378{
2379 unsigned char sha1[20];
2380 char hdr[32];
2381 int hdrlen;
2382
2383 /* Normally if we have it in the pack then we do not bother writing
2384 * it out into .git/objects/??/?{38} file.
2385 */
2386 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2387 if (returnsha1)
2388 hashcpy(returnsha1, sha1);
2389 if (has_sha1_file(sha1))
2390 return 0;
2391 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2392}
2393
2394int force_object_loose(const unsigned char *sha1, time_t mtime)
2395{
bbac7311
NP
2396 void *buf;
2397 unsigned long len;
2398 enum object_type type;
2399 char hdr[32];
2400 int hdrlen;
1fb23e65 2401 int ret;
bbac7311 2402
c529d75a 2403 if (has_loose_object(sha1))
bbac7311
NP
2404 return 0;
2405 buf = read_packed_sha1(sha1, &type, &len);
2406 if (!buf)
2407 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2408 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1fb23e65
BS
2409 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2410 free(buf);
2411
2412 return ret;
bbac7311
NP
2413}
2414
bf592c50
DB
2415int has_pack_index(const unsigned char *sha1)
2416{
2417 struct stat st;
2418 if (stat(sha1_pack_index_name(sha1), &st))
2419 return 0;
2420 return 1;
bf592c50
DB
2421}
2422
cd673c1f
JH
2423int has_sha1_pack(const unsigned char *sha1)
2424{
2425 struct pack_entry e;
2426 return find_pack_entry(sha1, &e);
2427}
2428
8237b185
DB
2429int has_sha1_file(const unsigned char *sha1)
2430{
1f688557
JH
2431 struct pack_entry e;
2432
cd673c1f 2433 if (find_pack_entry(sha1, &e))
1f688557 2434 return 1;
c529d75a 2435 return has_loose_object(sha1);
8237b185 2436}
74400e71 2437
43df4f86
DP
2438static int index_mem(unsigned char *sha1, void *buf, size_t size,
2439 int write_object, enum object_type type, const char *path)
e7332f96 2440{
6c510bee 2441 int ret, re_allocated = 0;
74400e71 2442
7672db20 2443 if (!type)
edaec3fb 2444 type = OBJ_BLOB;
6c510bee
LT
2445
2446 /*
2447 * Convert blobs to git internal format
2448 */
43df4f86 2449 if ((type == OBJ_BLOB) && path) {
f285a2d7 2450 struct strbuf nbuf = STRBUF_INIT;
21e5ad50
SP
2451 if (convert_to_git(path, buf, size, &nbuf,
2452 write_object ? safe_crlf : 0)) {
b315c5c0 2453 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
2454 re_allocated = 1;
2455 }
2456 }
2457
7672db20 2458 if (write_object)
edaec3fb 2459 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 2460 else
edaec3fb 2461 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 2462 if (re_allocated)
6c510bee 2463 free(buf);
43df4f86
DP
2464 return ret;
2465}
2466
ea68b0ce
DP
2467#define SMALL_FILE_SIZE (32*1024)
2468
43df4f86
DP
2469int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2470 enum object_type type, const char *path)
2471{
2472 int ret;
2473 size_t size = xsize_t(st->st_size);
2474
2475 if (!S_ISREG(st->st_mode)) {
f285a2d7 2476 struct strbuf sbuf = STRBUF_INIT;
43df4f86
DP
2477 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2478 ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2479 type, path);
2480 else
2481 ret = -1;
2482 strbuf_release(&sbuf);
08bda208
DP
2483 } else if (!size) {
2484 ret = index_mem(sha1, NULL, size, write_object, type, path);
ea68b0ce
DP
2485 } else if (size <= SMALL_FILE_SIZE) {
2486 char *buf = xmalloc(size);
2487 if (size == read_in_full(fd, buf, size))
2488 ret = index_mem(sha1, buf, size, write_object, type,
2489 path);
2490 else
2491 ret = error("short read %s", strerror(errno));
2492 free(buf);
08bda208 2493 } else {
43df4f86
DP
2494 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2495 ret = index_mem(sha1, buf, size, write_object, type, path);
aac17941 2496 munmap(buf, size);
08bda208 2497 }
43df4f86 2498 close(fd);
aac17941 2499 return ret;
74400e71 2500}
ec1fcc16
JH
2501
2502int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2503{
2504 int fd;
b760d3aa 2505 struct strbuf sb = STRBUF_INIT;
ec1fcc16
JH
2506
2507 switch (st->st_mode & S_IFMT) {
2508 case S_IFREG:
2509 fd = open(path, O_RDONLY);
2510 if (fd < 0)
2511 return error("open(\"%s\"): %s", path,
2512 strerror(errno));
53bca91a 2513 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
ec1fcc16
JH
2514 return error("%s: failed to insert into database",
2515 path);
2516 break;
2517 case S_IFLNK:
b760d3aa 2518 if (strbuf_readlink(&sb, path, st->st_size)) {
ec1fcc16 2519 char *errstr = strerror(errno);
ec1fcc16
JH
2520 return error("readlink(\"%s\"): %s", path,
2521 errstr);
2522 }
abdc3fc8 2523 if (!write_object)
b760d3aa
LT
2524 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2525 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
ec1fcc16
JH
2526 return error("%s: failed to insert into database",
2527 path);
b760d3aa 2528 strbuf_release(&sb);
ec1fcc16 2529 break;
f35a6d3b
LT
2530 case S_IFDIR:
2531 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
2532 default:
2533 return error("%s: unsupported file type", path);
2534 }
2535 return 0;
2536}
a69e5429
JH
2537
2538int read_pack_header(int fd, struct pack_header *header)
2539{
c697ad14
HO
2540 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2541 /* "eof before pack header was fully read" */
2542 return PH_ERROR_EOF;
2543
a69e5429
JH
2544 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2545 /* "protocol error (pack signature mismatch detected)" */
2546 return PH_ERROR_PACK_SIGNATURE;
2547 if (!pack_version_ok(header->hdr_version))
2548 /* "protocol error (pack version unsupported)" */
2549 return PH_ERROR_PROTOCOL;
2550 return 0;
2551}
40d52ff7
JK
2552
2553void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2554{
2555 enum object_type type = sha1_object_info(sha1, NULL);
2556 if (type < 0)
2557 die("%s is not a valid object", sha1_to_hex(sha1));
2558 if (type != expect)
2559 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2560 typename(expect));
2561}