A loose object is not corrupt if it cannot be read due to EMFILE
[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
c9ced051
SP
579void close_pack_windows(struct packed_git *p)
580{
581 while (p->windows) {
582 struct pack_window *w = p->windows;
583
584 if (w->inuse_cnt)
585 die("pack '%s' still has open windows to it",
586 p->pack_name);
587 munmap(w->base, w->len);
588 pack_mapped -= w->len;
589 pack_open_windows--;
590 p->windows = w->next;
591 free(w);
592 }
593}
594
03e79c88 595void unuse_pack(struct pack_window **w_cursor)
f9253394 596{
03e79c88
SP
597 struct pack_window *w = *w_cursor;
598 if (w) {
599 w->inuse_cnt--;
600 *w_cursor = NULL;
601 }
1f688557
JH
602}
603
fa5fc15d
SP
604void close_pack_index(struct packed_git *p)
605{
606 if (p->index_data) {
607 munmap((void *)p->index_data, p->index_size);
608 p->index_data = NULL;
609 }
610}
611
c74faea1
NP
612/*
613 * This is used by git-repack in case a newly created pack happens to
614 * contain the same set of objects as an existing one. In that case
615 * the resulting file might be different even if its name would be the
616 * same. It is best to close any reference to the old pack before it is
617 * replaced on disk. Of course no index pointers nor windows for given pack
618 * must subsist at this point. If ever objects from this pack are requested
619 * again, the new version of the pack will be reinitialized through
620 * reprepare_packed_git().
621 */
622void free_pack_by_name(const char *pack_name)
623{
624 struct packed_git *p, **pp = &packed_git;
625
626 while (*pp) {
627 p = *pp;
628 if (strcmp(pack_name, p->pack_name) == 0) {
fa3a0c94 629 clear_delta_base_cache();
c74faea1
NP
630 close_pack_windows(p);
631 if (p->pack_fd != -1)
632 close(p->pack_fd);
fa5fc15d 633 close_pack_index(p);
c74faea1
NP
634 free(p->bad_object_sha1);
635 *pp = p->next;
636 free(p);
637 return;
638 }
639 pp = &p->next;
640 }
641}
642
3cf8b462
SP
643/*
644 * Do not call this directly as this leaks p->pack_fd on error return;
645 * call open_packed_git() instead.
646 */
647static int open_packed_git_1(struct packed_git *p)
1f688557 648{
9bc879c1
SP
649 struct stat st;
650 struct pack_header hdr;
651 unsigned char sha1[20];
652 unsigned char *idx_sha1;
2c039da8 653 long fd_flag;
9bc879c1 654
d079837e
SP
655 if (!p->index_data && open_pack_index(p))
656 return error("packfile %s index unavailable", p->pack_name);
657
9bc879c1 658 p->pack_fd = open(p->pack_name, O_RDONLY);
fd73ccf2
JS
659 while (p->pack_fd < 0 && errno == EMFILE && unuse_one_window(p, -1))
660 p->pack_fd = open(p->pack_name, O_RDONLY);
9bc879c1 661 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 662 return -1;
9bc879c1
SP
663
664 /* If we created the struct before we had the pack we lack size. */
bf592c50 665 if (!p->pack_size) {
bf592c50 666 if (!S_ISREG(st.st_mode))
072db278 667 return error("packfile %s not a regular file", p->pack_name);
bf592c50 668 p->pack_size = st.st_size;
9bc879c1 669 } else if (p->pack_size != st.st_size)
072db278 670 return error("packfile %s size changed", p->pack_name);
9bc879c1 671
2c039da8
JH
672 /* We leave these file descriptors open with sliding mmap;
673 * there is no point keeping them open across exec(), though.
674 */
675 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
676 if (fd_flag < 0)
072db278 677 return error("cannot determine file descriptor flags");
2c039da8
JH
678 fd_flag |= FD_CLOEXEC;
679 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 680 return error("cannot set FD_CLOEXEC");
2c039da8 681
9bc879c1 682 /* Verify we recognize this pack file format. */
e6e2bd62 683 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 684 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 685 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 686 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 687 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
688 return error("packfile %s is version %"PRIu32" and not"
689 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
690 p->pack_name, ntohl(hdr.hdr_version));
691
692 /* Verify the pack matches its index. */
57059091 693 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
694 return error("packfile %s claims to have %"PRIu32" objects"
695 " while index indicates %"PRIu32" objects",
57059091
NP
696 p->pack_name, ntohl(hdr.hdr_entries),
697 p->num_objects);
9bc879c1 698 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 699 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 700 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 701 return error("packfile %s signature is unavailable", p->pack_name);
42873078 702 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 703 if (hashcmp(sha1, idx_sha1))
072db278
SP
704 return error("packfile %s does not match index", p->pack_name);
705 return 0;
9bc879c1
SP
706}
707
3cf8b462
SP
708static int open_packed_git(struct packed_git *p)
709{
710 if (!open_packed_git_1(p))
711 return 0;
712 if (p->pack_fd != -1) {
713 close(p->pack_fd);
714 p->pack_fd = -1;
715 }
716 return -1;
717}
718
c4001d92 719static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
720{
721 /* We must promise at least 20 bytes (one hash) after the
722 * offset is available from this window, otherwise the offset
723 * is not actually in this window and a different window (which
724 * has that one hash excess) must be used. This is to support
725 * the object header and delta base parsing routines below.
726 */
727 off_t win_off = win->offset;
728 return win_off <= offset
729 && (offset + 20) <= (win_off + win->len);
730}
731
4b25d091 732unsigned char *use_pack(struct packed_git *p,
03e79c88 733 struct pack_window **w_cursor,
c4001d92 734 off_t offset,
03e79c88 735 unsigned int *left)
9bc879c1 736{
60bb8b14 737 struct pack_window *win = *w_cursor;
03e79c88 738
072db278
SP
739 if (p->pack_fd == -1 && open_packed_git(p))
740 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14 741
a9d98a14 742 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
743 * pointless to ask for an offset into the middle of that
744 * hash, and the in_window function above wouldn't match
745 * don't allow an offset too close to the end of the file.
746 */
747 if (offset > (p->pack_size - 20))
748 die("offset beyond end of packfile (truncated pack?)");
749
750 if (!win || !in_window(win, offset)) {
751 if (win)
752 win->inuse_cnt--;
753 for (win = p->windows; win; win = win->next) {
754 if (in_window(win, offset))
755 break;
756 }
757 if (!win) {
78a28df9 758 size_t window_align = packed_git_window_size / 2;
dc49cd76 759 off_t len;
60bb8b14 760 win = xcalloc(1, sizeof(*win));
78a28df9 761 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
762 len = p->pack_size - win->offset;
763 if (len > packed_git_window_size)
764 len = packed_git_window_size;
765 win->len = (size_t)len;
60bb8b14 766 pack_mapped += win->len;
11daf39b 767 while (packed_git_limit < pack_mapped
d1efefa4 768 && unuse_one_window(p, p->pack_fd))
60bb8b14 769 ; /* nothing */
c4712e45 770 win->base = xmmap(NULL, win->len,
60bb8b14
SP
771 PROT_READ, MAP_PRIVATE,
772 p->pack_fd, win->offset);
773 if (win->base == MAP_FAILED)
73b4e4be
SP
774 die("packfile %s cannot be mapped: %s",
775 p->pack_name,
776 strerror(errno));
a53128b6
SP
777 pack_mmap_calls++;
778 pack_open_windows++;
779 if (pack_mapped > peak_pack_mapped)
780 peak_pack_mapped = pack_mapped;
781 if (pack_open_windows > peak_pack_open_windows)
782 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
783 win->next = p->windows;
784 p->windows = win;
785 }
1f688557 786 }
03e79c88
SP
787 if (win != *w_cursor) {
788 win->last_used = pack_used_ctr++;
789 win->inuse_cnt++;
790 *w_cursor = win;
791 }
60bb8b14 792 offset -= win->offset;
03e79c88 793 if (left)
dc49cd76 794 *left = win->len - xsize_t(offset);
03e79c88 795 return win->base + offset;
1f688557
JH
796}
797
27d69a46
NP
798static struct packed_git *alloc_packed_git(int extra)
799{
800 struct packed_git *p = xmalloc(sizeof(*p) + extra);
801 memset(p, 0, sizeof(*p));
802 p->pack_fd = -1;
803 return p;
804}
805
42873078 806struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557
JH
807{
808 struct stat st;
27d69a46 809 struct packed_git *p = alloc_packed_git(path_len + 2);
1f688557 810
42873078
NP
811 /*
812 * Make sure a corresponding .pack file exists and that
813 * the index looks sane.
814 */
815 path_len -= strlen(".idx");
27d69a46
NP
816 if (path_len < 1) {
817 free(p);
1f688557 818 return NULL;
27d69a46 819 }
42873078 820 memcpy(p->pack_name, path, path_len);
8d25931d
BC
821
822 strcpy(p->pack_name + path_len, ".keep");
823 if (!access(p->pack_name, F_OK))
824 p->pack_keep = 1;
825
42873078 826 strcpy(p->pack_name + path_len, ".pack");
d079837e 827 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 828 free(p);
1f688557
JH
829 return NULL;
830 }
42873078 831
1f688557
JH
832 /* ok, it looks sane as far as we can check without
833 * actually mapping the pack file.
834 */
1f688557 835 p->pack_size = st.st_size;
9d835df2 836 p->pack_local = local;
b867092f 837 p->mtime = st.st_mtime;
42873078
NP
838 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
839 hashclr(p->sha1);
1f688557
JH
840 return p;
841}
842
7b64469a 843struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 844{
42873078 845 const char *path = sha1_pack_name(sha1);
27d69a46 846 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
bf592c50 847
27d69a46
NP
848 strcpy(p->pack_name, path);
849 hashcpy(p->sha1, sha1);
42873078
NP
850 if (check_packed_git_idx(idx_path, p)) {
851 free(p);
bf592c50 852 return NULL;
42873078 853 }
bf592c50 854
bf592c50
DB
855 return p;
856}
857
858void install_packed_git(struct packed_git *pack)
859{
860 pack->next = packed_git;
861 packed_git = pack;
862}
863
9d835df2 864static void prepare_packed_git_one(char *objdir, int local)
1f688557 865{
9cb18f56
JM
866 /* Ensure that this buffer is large enough so that we can
867 append "/pack/" without clobbering the stack even if
868 strlen(objdir) were PATH_MAX. */
869 char path[PATH_MAX + 1 + 4 + 1 + 1];
1f688557
JH
870 int len;
871 DIR *dir;
872 struct dirent *de;
873
874 sprintf(path, "%s/pack", objdir);
875 len = strlen(path);
876 dir = opendir(path);
fd73ccf2
JS
877 while (!dir && errno == EMFILE && unuse_one_window(packed_git, -1))
878 dir = opendir(path);
b5b16990 879 if (!dir) {
26125f6b 880 if (errno != ENOENT)
bd2afde8 881 error("unable to open object pack directory: %s: %s",
26125f6b 882 path, strerror(errno));
1f688557 883 return;
b5b16990 884 }
1f688557
JH
885 path[len++] = '/';
886 while ((de = readdir(dir)) != NULL) {
887 int namelen = strlen(de->d_name);
888 struct packed_git *p;
889
5bb1cda5 890 if (!has_extension(de->d_name, ".idx"))
1f688557
JH
891 continue;
892
9cb18f56
JM
893 if (len + namelen + 1 > sizeof(path))
894 continue;
895
54a15a8d 896 /* Don't reopen a pack we already have. */
1f688557 897 strcpy(path + len, de->d_name);
86f7780c
JK
898 for (p = packed_git; p; p = p->next) {
899 if (!memcmp(path, p->pack_name, len + namelen - 4))
900 break;
901 }
902 if (p)
903 continue;
54a15a8d
SP
904 /* See if it really is a valid .idx file with corresponding
905 * .pack file that we can map.
906 */
9d835df2 907 p = add_packed_git(path, len + namelen, local);
1f688557
JH
908 if (!p)
909 continue;
625e9421 910 install_packed_git(p);
1f688557 911 }
5b35bcd5 912 closedir(dir);
1f688557
JH
913}
914
b867092f
JH
915static int sort_pack(const void *a_, const void *b_)
916{
917 struct packed_git *a = *((struct packed_git **)a_);
918 struct packed_git *b = *((struct packed_git **)b_);
919 int st;
920
921 /*
922 * Local packs tend to contain objects specific to our
923 * variant of the project than remote ones. In addition,
924 * remote ones could be on a network mounted filesystem.
925 * Favor local ones for these reasons.
926 */
927 st = a->pack_local - b->pack_local;
928 if (st)
929 return -st;
930
931 /*
932 * Younger packs tend to contain more recent objects,
933 * and more recent objects tend to get accessed more
934 * often.
935 */
936 if (a->mtime < b->mtime)
937 return 1;
938 else if (a->mtime == b->mtime)
939 return 0;
940 return -1;
941}
942
943static void rearrange_packed_git(void)
944{
945 struct packed_git **ary, *p;
946 int i, n;
947
948 for (n = 0, p = packed_git; p; p = p->next)
949 n++;
950 if (n < 2)
951 return;
952
953 /* prepare an array of packed_git for easier sorting */
954 ary = xcalloc(n, sizeof(struct packed_git *));
955 for (n = 0, p = packed_git; p; p = p->next)
956 ary[n++] = p;
957
958 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
959
960 /* link them back again */
961 for (i = 0; i < n - 1; i++)
962 ary[i]->next = ary[i + 1];
963 ary[n - 1]->next = NULL;
964 packed_git = ary[0];
965
966 free(ary);
967}
968
637cdd9d 969static int prepare_packed_git_run_once = 0;
9a217f2a 970void prepare_packed_git(void)
1f688557 971{
d5a63b99 972 struct alternate_object_database *alt;
1f688557 973
637cdd9d 974 if (prepare_packed_git_run_once)
1f688557 975 return;
9d835df2 976 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 977 prepare_alt_odb();
d5a63b99 978 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 979 alt->name[-1] = 0;
9d835df2 980 prepare_packed_git_one(alt->base, 0);
1494e038 981 alt->name[-1] = '/';
1f688557 982 }
b867092f 983 rearrange_packed_git();
637cdd9d
JK
984 prepare_packed_git_run_once = 1;
985}
986
fc04c412 987void reprepare_packed_git(void)
637cdd9d 988{
4b480c67 989 discard_revindex();
637cdd9d
JK
990 prepare_packed_git_run_once = 0;
991 prepare_packed_git();
1f688557
JH
992}
993
8eca0b47
NP
994static void mark_bad_packed_object(struct packed_git *p,
995 const unsigned char *sha1)
996{
997 unsigned i;
998 for (i = 0; i < p->num_bad_objects; i++)
999 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1000 return;
1001 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1002 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1003 p->num_bad_objects++;
1004}
1005
b6c4cecc 1006static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
ac939109
NP
1007{
1008 struct packed_git *p;
1009 unsigned i;
1010
1011 for (p = packed_git; p; p = p->next)
1012 for (i = 0; i < p->num_bad_objects; i++)
1013 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
b6c4cecc
JH
1014 return p;
1015 return NULL;
ac939109
NP
1016}
1017
5d6ccf5c 1018int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
0fcfd160
LT
1019{
1020 unsigned char real_sha1[20];
7cfb5f36 1021 hash_sha1_file(map, size, type, real_sha1);
a89fccd2 1022 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1023}
1024
44d1c19e
LT
1025static int git_open_noatime(const char *name)
1026{
1027 static int sha1_file_open_flag = O_NOATIME;
1028 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1029
1030 /* Might the failure be due to O_NOATIME? */
1031 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1032 fd = open(name, O_RDONLY);
1033 if (fd >= 0)
1034 sha1_file_open_flag = 0;
1035 }
1036 return fd;
1037}
1038
1039static int open_sha1_file(const unsigned char *sha1)
1040{
1041 int fd;
1042 char *name = sha1_file_name(sha1);
1043 struct alternate_object_database *alt;
1044
1045 fd = git_open_noatime(name);
1046 if (fd >= 0)
1047 return fd;
1048
1049 prepare_alt_odb();
1050 errno = ENOENT;
1051 for (alt = alt_odb_list; alt; alt = alt->next) {
1052 name = alt->name;
1053 fill_sha1_path(name, sha1);
1054 fd = git_open_noatime(alt->base);
1055 if (fd >= 0)
1056 return fd;
1057 }
1058 return -1;
1059}
1060
4175e9e3 1061static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1062{
0fcfd160 1063 void *map;
144bde78 1064 int fd;
ace1534d 1065
44d1c19e
LT
1066 fd = open_sha1_file(sha1);
1067 map = NULL;
1068 if (fd >= 0) {
1069 struct stat st;
0fcfd160 1070
44d1c19e
LT
1071 if (!fstat(fd, &st)) {
1072 *size = xsize_t(st.st_size);
1073 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1074 }
44d1c19e 1075 close(fd);
0fcfd160 1076 }
0fcfd160
LT
1077 return map;
1078}
1079
726f852b 1080static int legacy_loose_object(unsigned char *map)
bb6b8e4f
JH
1081{
1082 unsigned int word;
1083
1084 /*
1085 * Is it a zlib-compressed buffer? If so, the first byte
1086 * must be 0x78 (15-bit window size, deflated), and the
1087 * first 16-bit word is evenly divisible by 31
1088 */
1089 word = (map[0] << 8) + map[1];
1090 if (map[0] == 0x78 && !(word % 31))
1091 return 1;
1092 else
1093 return 0;
1094}
1095
09ded04b
NP
1096unsigned long unpack_object_header_buffer(const unsigned char *buf,
1097 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1098{
ad1ed5ee 1099 unsigned shift;
48fb7deb 1100 unsigned long size, c;
ad1ed5ee
JH
1101 unsigned long used = 0;
1102
1103 c = buf[used++];
1104 *type = (c >> 4) & 7;
1105 size = c & 15;
1106 shift = 4;
1107 while (c & 0x80) {
f630cfda 1108 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1109 error("bad object header");
ad1ed5ee 1110 return 0;
09ded04b 1111 }
ad1ed5ee
JH
1112 c = buf[used++];
1113 size += (c & 0x7f) << shift;
1114 shift += 7;
1115 }
1116 *sizep = size;
1117 return used;
1118}
1119
1120static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1121{
1122 unsigned long size, used;
1123 static const char valid_loose_object_type[8] = {
1124 0, /* OBJ_EXT */
1125 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1126 0, /* "delta" and others are invalid in a loose object */
93821bd9 1127 };
ad1ed5ee 1128 enum object_type type;
93821bd9 1129
c4483576
LT
1130 /* Get the data stream */
1131 memset(stream, 0, sizeof(*stream));
1132 stream->next_in = map;
1133 stream->avail_in = mapsize;
1134 stream->next_out = buffer;
93821bd9
LT
1135 stream->avail_out = bufsiz;
1136
bb6b8e4f 1137 if (legacy_loose_object(map)) {
39c68542
LT
1138 git_inflate_init(stream);
1139 return git_inflate(stream, 0);
93821bd9
LT
1140 }
1141
726f852b
NP
1142
1143 /*
1144 * There used to be a second loose object header format which
1145 * was meant to mimic the in-pack format, allowing for direct
1146 * copy of the object data. This format turned up not to be
1147 * really worth it and we don't write it any longer. But we
1148 * can still read it.
1149 */
09ded04b 1150 used = unpack_object_header_buffer(map, mapsize, &type, &size);
ad1ed5ee 1151 if (!used || !valid_loose_object_type[type])
93821bd9 1152 return -1;
ad1ed5ee
JH
1153 map += used;
1154 mapsize -= used;
93821bd9
LT
1155
1156 /* Set up the stream for the rest.. */
1157 stream->next_in = map;
1158 stream->avail_in = mapsize;
39c68542 1159 git_inflate_init(stream);
93821bd9
LT
1160
1161 /* And generate the fake traditional header */
ad1ed5ee 1162 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
df843662 1163 typename(type), size);
93821bd9 1164 return 0;
c4483576
LT
1165}
1166
7efbff75 1167static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1168{
1169 int bytes = strlen(buffer) + 1;
3aee68aa 1170 unsigned char *buf = xmallocz(size);
93821bd9 1171 unsigned long n;
7efbff75 1172 int status = Z_OK;
5180cacc 1173
93821bd9
LT
1174 n = stream->total_out - bytes;
1175 if (n > size)
1176 n = size;
1177 memcpy(buf, (char *) buffer + bytes, n);
1178 bytes = n;
456cdf6e
LT
1179 if (bytes <= size) {
1180 /*
1181 * The above condition must be (bytes <= size), not
1182 * (bytes < size). In other words, even though we
1183 * expect no more output and set avail_out to zer0,
1184 * the input zlib stream may have bytes that express
1185 * "this concludes the stream", and we *do* want to
1186 * eat that input.
1187 *
1188 * Otherwise we would not be able to test that we
1189 * consumed all the input to reach the expected size;
1190 * we also want to check that zlib tells us that all
1191 * went well with status == Z_STREAM_END at the end.
1192 */
5180cacc
LT
1193 stream->next_out = buf + bytes;
1194 stream->avail_out = size - bytes;
7efbff75 1195 while (status == Z_OK)
39c68542 1196 status = git_inflate(stream, Z_FINISH);
5180cacc 1197 }
456cdf6e 1198 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1199 git_inflate_end(stream);
7efbff75
JH
1200 return buf;
1201 }
1202
1203 if (status < 0)
1204 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1205 else if (stream->avail_in)
1206 error("garbage at end of loose object '%s'",
1207 sha1_to_hex(sha1));
1208 free(buf);
1209 return NULL;
5180cacc
LT
1210}
1211
1212/*
1213 * We used to just use "sscanf()", but that's actually way
1214 * too permissive for what we want to check. So do an anal
1215 * object header parse by hand.
1216 */
21666f1a 1217static int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1218{
21666f1a 1219 char type[10];
5180cacc
LT
1220 int i;
1221 unsigned long size;
1222
1223 /*
a6080a0a 1224 * The type can be at most ten bytes (including the
5180cacc 1225 * terminating '\0' that we add), and is followed by
21666f1a 1226 * a space.
5180cacc 1227 */
21666f1a 1228 i = 0;
5180cacc
LT
1229 for (;;) {
1230 char c = *hdr++;
1231 if (c == ' ')
1232 break;
21666f1a
NP
1233 type[i++] = c;
1234 if (i >= sizeof(type))
5180cacc 1235 return -1;
5180cacc 1236 }
21666f1a 1237 type[i] = 0;
5180cacc
LT
1238
1239 /*
1240 * The length must follow immediately, and be in canonical
1241 * decimal format (ie "010" is not valid).
1242 */
1243 size = *hdr++ - '0';
1244 if (size > 9)
1245 return -1;
1246 if (size) {
1247 for (;;) {
1248 unsigned long c = *hdr - '0';
1249 if (c > 9)
1250 break;
1251 hdr++;
1252 size = size * 10 + c;
1253 }
1254 }
1255 *sizep = size;
1256
1257 /*
1258 * The length must be followed by a zero byte
1259 */
21666f1a 1260 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1261}
1262
7efbff75 1263static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1264{
5180cacc 1265 int ret;
0fcfd160 1266 z_stream stream;
5180cacc 1267 char hdr[8192];
0fcfd160 1268
5180cacc 1269 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1270 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1271 return NULL;
1272
7efbff75 1273 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1274}
1275
54dab52a
NP
1276unsigned long get_size_from_delta(struct packed_git *p,
1277 struct pack_window **w_curs,
1278 off_t curpos)
1279{
1280 const unsigned char *data;
1281 unsigned char delta_head[20], *in;
1282 z_stream stream;
1283 int st;
1284
1285 memset(&stream, 0, sizeof(stream));
1286 stream.next_out = delta_head;
1287 stream.avail_out = sizeof(delta_head);
1288
39c68542 1289 git_inflate_init(&stream);
54dab52a
NP
1290 do {
1291 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1292 stream.next_in = in;
39c68542 1293 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1294 curpos += stream.next_in - in;
1295 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1296 stream.total_out < sizeof(delta_head));
39c68542 1297 git_inflate_end(&stream);
3d77d877
NP
1298 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1299 error("delta data unpack-initial failed");
1300 return 0;
1301 }
54dab52a
NP
1302
1303 /* Examine the initial part of the delta to figure out
1304 * the result size.
1305 */
1306 data = delta_head;
1307
1308 /* ignore base size */
1309 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1310
1311 /* Read the result size */
1312 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1313}
1314
c4001d92 1315static off_t get_delta_base(struct packed_git *p,
03e79c88 1316 struct pack_window **w_curs,
c4001d92 1317 off_t *curpos,
21666f1a 1318 enum object_type type,
c4001d92 1319 off_t delta_obj_offset)
eb32d236 1320{
2b87c45b 1321 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1322 off_t base_offset;
eb32d236 1323
8d8a4ea5
SP
1324 /* use_pack() assured us we have [base_info, base_info + 20)
1325 * as a range that we can look at without walking off the
1326 * end of the mapped window. Its actually the hash size
1327 * that is assured. An OFS_DELTA longer than the hash size
1328 * is stupid, as then a REF_DELTA would be smaller to store.
1329 */
21666f1a 1330 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1331 unsigned used = 0;
1332 unsigned char c = base_info[used++];
1333 base_offset = c & 127;
1334 while (c & 128) {
1335 base_offset += 1;
8723f216 1336 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1337 return 0; /* overflow */
eb32d236
NP
1338 c = base_info[used++];
1339 base_offset = (base_offset << 7) + (c & 127);
1340 }
1341 base_offset = delta_obj_offset - base_offset;
d8f32556 1342 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1343 return 0; /* out of bound */
2b87c45b 1344 *curpos += used;
21666f1a 1345 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1346 /* The base entry _must_ be in the same pack */
1347 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1348 *curpos += 20;
eb32d236
NP
1349 } else
1350 die("I am totally screwed");
2b87c45b 1351 return base_offset;
eb32d236
NP
1352}
1353
f3bf9224 1354/* forward declaration for a mutually recursive function */
c4001d92 1355static int packed_object_info(struct packed_git *p, off_t offset,
21666f1a 1356 unsigned long *sizep);
f3bf9224 1357
43057304 1358static int packed_delta_info(struct packed_git *p,
03e79c88 1359 struct pack_window **w_curs,
c4001d92 1360 off_t curpos,
21666f1a 1361 enum object_type type,
c4001d92 1362 off_t obj_offset,
43057304 1363 unsigned long *sizep)
5db47c2b 1364{
c4001d92 1365 off_t base_offset;
f3bf9224 1366
21666f1a 1367 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
3d77d877
NP
1368 if (!base_offset)
1369 return OBJ_BAD;
21666f1a 1370 type = packed_object_info(p, base_offset, NULL);
3d77d877 1371 if (type <= OBJ_NONE) {
08698b1e
NP
1372 struct revindex_entry *revidx;
1373 const unsigned char *base_sha1;
1374 revidx = find_pack_revindex(p, base_offset);
1375 if (!revidx)
1376 return OBJ_BAD;
1377 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
3d77d877
NP
1378 mark_bad_packed_object(p, base_sha1);
1379 type = sha1_object_info(base_sha1, NULL);
1380 if (type <= OBJ_NONE)
1381 return OBJ_BAD;
1382 }
f3bf9224 1383
c62266f3
JH
1384 /* We choose to only get the type of the base object and
1385 * ignore potentially corrupt pack file that expects the delta
1386 * based on a base with a wrong size. This saves tons of
1387 * inflate() calls.
1388 */
3d77d877 1389 if (sizep) {
54dab52a 1390 *sizep = get_size_from_delta(p, w_curs, curpos);
3d77d877
NP
1391 if (*sizep == 0)
1392 type = OBJ_BAD;
1393 }
21666f1a
NP
1394
1395 return type;
5db47c2b
JH
1396}
1397
2b87c45b
NP
1398static int unpack_object_header(struct packed_git *p,
1399 struct pack_window **w_curs,
c4001d92 1400 off_t *curpos,
2b87c45b 1401 unsigned long *sizep)
a733cb60 1402{
03e79c88
SP
1403 unsigned char *base;
1404 unsigned int left;
ad1ed5ee 1405 unsigned long used;
2b87c45b 1406 enum object_type type;
a733cb60 1407
8d8a4ea5
SP
1408 /* use_pack() assures us we have [base, base + 20) available
1409 * as a range that we can look at at. (Its actually the hash
3dff5379 1410 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1411 * the maximum deflated object size is 2^137, which is just
1412 * insane, so we know won't exceed what we have been given.
1413 */
2b87c45b 1414 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1415 used = unpack_object_header_buffer(base, left, &type, sizep);
1416 if (!used) {
1417 type = OBJ_BAD;
1418 } else
1419 *curpos += used;
ad1ed5ee 1420
2b87c45b 1421 return type;
a733cb60
LT
1422}
1423
21666f1a 1424const char *packed_object_info_detail(struct packed_git *p,
c4001d92 1425 off_t obj_offset,
21666f1a
NP
1426 unsigned long *size,
1427 unsigned long *store_size,
1428 unsigned int *delta_chain_length,
1429 unsigned char *base_sha1)
ad8c80a5 1430{
03e79c88 1431 struct pack_window *w_curs = NULL;
c4001d92
SP
1432 off_t curpos;
1433 unsigned long dummy;
43057304 1434 unsigned char *next_sha1;
21666f1a 1435 enum object_type type;
70f5d5d3 1436 struct revindex_entry *revidx;
ad8c80a5 1437
43057304 1438 *delta_chain_length = 0;
2b87c45b 1439 curpos = obj_offset;
21666f1a 1440 type = unpack_object_header(p, &w_curs, &curpos, size);
43057304 1441
70f5d5d3
NP
1442 revidx = find_pack_revindex(p, obj_offset);
1443 *store_size = revidx[1].offset - obj_offset;
1444
43057304 1445 for (;;) {
21666f1a 1446 switch (type) {
43057304 1447 default:
08a19d87 1448 die("pack %s contains unknown object type %d",
21666f1a 1449 p->pack_name, type);
43057304
NP
1450 case OBJ_COMMIT:
1451 case OBJ_TREE:
1452 case OBJ_BLOB:
1453 case OBJ_TAG:
03e79c88 1454 unuse_pack(&w_curs);
21666f1a 1455 return typename(type);
eb32d236 1456 case OBJ_OFS_DELTA:
21666f1a 1457 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
8eca0b47
NP
1458 if (!obj_offset)
1459 die("pack %s contains bad delta base reference of type %s",
1460 p->pack_name, typename(type));
eb32d236 1461 if (*delta_chain_length == 0) {
70f5d5d3
NP
1462 revidx = find_pack_revindex(p, obj_offset);
1463 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
eb32d236
NP
1464 }
1465 break;
1466 case OBJ_REF_DELTA:
2b87c45b 1467 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
43057304
NP
1468 if (*delta_chain_length == 0)
1469 hashcpy(base_sha1, next_sha1);
2b87c45b 1470 obj_offset = find_pack_entry_one(next_sha1, p);
43057304
NP
1471 break;
1472 }
43057304 1473 (*delta_chain_length)++;
2b87c45b 1474 curpos = obj_offset;
21666f1a 1475 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
ad8c80a5 1476 }
ad8c80a5
JH
1477}
1478
c4001d92 1479static int packed_object_info(struct packed_git *p, off_t obj_offset,
21666f1a 1480 unsigned long *sizep)
1f688557 1481{
03e79c88 1482 struct pack_window *w_curs = NULL;
c4001d92
SP
1483 unsigned long size;
1484 off_t curpos = obj_offset;
21666f1a 1485 enum object_type type;
5db47c2b 1486
21666f1a 1487 type = unpack_object_header(p, &w_curs, &curpos, &size);
5db47c2b 1488
21666f1a 1489 switch (type) {
eb32d236
NP
1490 case OBJ_OFS_DELTA:
1491 case OBJ_REF_DELTA:
21666f1a
NP
1492 type = packed_delta_info(p, &w_curs, curpos,
1493 type, obj_offset, sizep);
2b87c45b 1494 break;
a733cb60 1495 case OBJ_COMMIT:
a733cb60 1496 case OBJ_TREE:
a733cb60 1497 case OBJ_BLOB:
a733cb60 1498 case OBJ_TAG:
2b87c45b
NP
1499 if (sizep)
1500 *sizep = size;
a69d0943 1501 break;
1f688557 1502 default:
3d77d877
NP
1503 error("unknown object type %i at offset %"PRIuMAX" in %s",
1504 type, (uintmax_t)obj_offset, p->pack_name);
1505 type = OBJ_BAD;
1f688557 1506 }
2b87c45b 1507 unuse_pack(&w_curs);
21666f1a 1508 return type;
1f688557
JH
1509}
1510
eb950c19 1511static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1512 struct pack_window **w_curs,
c4001d92 1513 off_t curpos,
eb950c19 1514 unsigned long size)
de530aaa
SP
1515{
1516 int st;
1517 z_stream stream;
079afb18 1518 unsigned char *buffer, *in;
de530aaa 1519
4ab07e4d 1520 buffer = xmallocz(size);
de530aaa 1521 memset(&stream, 0, sizeof(stream));
de530aaa 1522 stream.next_out = buffer;
39eea7bd 1523 stream.avail_out = size + 1;
de530aaa 1524
39c68542 1525 git_inflate_init(&stream);
079afb18 1526 do {
2b87c45b 1527 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 1528 stream.next_in = in;
39c68542 1529 st = git_inflate(&stream, Z_FINISH);
39eea7bd
JH
1530 if (!stream.avail_out)
1531 break; /* the payload is larger than it should be */
2b87c45b 1532 curpos += stream.next_in - in;
079afb18 1533 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 1534 git_inflate_end(&stream);
de530aaa
SP
1535 if ((st != Z_STREAM_END) || stream.total_out != size) {
1536 free(buffer);
1537 return NULL;
1538 }
1539
1540 return buffer;
1541}
1542
e5e01619
LT
1543#define MAX_DELTA_CACHE (256)
1544
18bdec11 1545static size_t delta_base_cached;
5e08ecbf
NP
1546
1547static struct delta_base_cache_lru_list {
1548 struct delta_base_cache_lru_list *prev;
1549 struct delta_base_cache_lru_list *next;
1550} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1551
e5e01619 1552static struct delta_base_cache_entry {
5e08ecbf
NP
1553 struct delta_base_cache_lru_list lru;
1554 void *data;
e5e01619
LT
1555 struct packed_git *p;
1556 off_t base_offset;
1557 unsigned long size;
e5e01619
LT
1558 enum object_type type;
1559} delta_base_cache[MAX_DELTA_CACHE];
1560
1561static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1562{
1563 unsigned long hash;
1564
1565 hash = (unsigned long)p + (unsigned long)base_offset;
1566 hash += (hash >> 8) + (hash >> 16);
3358004a 1567 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1568}
1569
62f255ad 1570static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 1571 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 1572{
e5e01619
LT
1573 void *ret;
1574 unsigned long hash = pack_entry_hash(p, base_offset);
1575 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1576
1577 ret = ent->data;
749bc58c
MV
1578 if (!ret || ent->p != p || ent->base_offset != base_offset)
1579 return unpack_entry(p, base_offset, type, base_size);
e5e01619 1580
18bdec11 1581 if (!keep_cache) {
a0cba108 1582 ent->data = NULL;
5e08ecbf
NP
1583 ent->lru.next->prev = ent->lru.prev;
1584 ent->lru.prev->next = ent->lru.next;
18bdec11 1585 delta_base_cached -= ent->size;
182af834
PH
1586 } else {
1587 ret = xmemdupz(ent->data, ent->size);
a0cba108 1588 }
e5e01619
LT
1589 *type = ent->type;
1590 *base_size = ent->size;
1591 return ret;
62f255ad
LT
1592}
1593
18bdec11
SP
1594static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1595{
1596 if (ent->data) {
1597 free(ent->data);
1598 ent->data = NULL;
5e08ecbf
NP
1599 ent->lru.next->prev = ent->lru.prev;
1600 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
1601 delta_base_cached -= ent->size;
1602 }
1603}
1604
3d20c636
SP
1605void clear_delta_base_cache(void)
1606{
1607 unsigned long p;
1608 for (p = 0; p < MAX_DELTA_CACHE; p++)
1609 release_delta_base_cache(&delta_base_cache[p]);
1610}
1611
62f255ad
LT
1612static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1613 void *base, unsigned long base_size, enum object_type type)
1614{
5e08ecbf 1615 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 1616 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 1617 struct delta_base_cache_lru_list *lru;
e5e01619 1618
18bdec11
SP
1619 release_delta_base_cache(ent);
1620 delta_base_cached += base_size;
5e08ecbf
NP
1621
1622 for (lru = delta_base_cache_lru.next;
1623 delta_base_cached > delta_base_cache_limit
1624 && lru != &delta_base_cache_lru;
1625 lru = lru->next) {
1626 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
1627 if (f->type == OBJ_BLOB)
1628 release_delta_base_cache(f);
1629 }
5e08ecbf
NP
1630 for (lru = delta_base_cache_lru.next;
1631 delta_base_cached > delta_base_cache_limit
1632 && lru != &delta_base_cache_lru;
1633 lru = lru->next) {
1634 struct delta_base_cache_entry *f = (void *)lru;
1635 release_delta_base_cache(f);
1636 }
18bdec11 1637
e5e01619
LT
1638 ent->p = p;
1639 ent->base_offset = base_offset;
1640 ent->type = type;
1641 ent->data = base;
1642 ent->size = base_size;
5e08ecbf
NP
1643 ent->lru.next = &delta_base_cache_lru;
1644 ent->lru.prev = delta_base_cache_lru.prev;
1645 delta_base_cache_lru.prev->next = &ent->lru;
1646 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
1647}
1648
c2c5b270
CC
1649static void *read_object(const unsigned char *sha1, enum object_type *type,
1650 unsigned long *size);
1651
eb950c19 1652static void *unpack_delta_entry(struct packed_git *p,
03e79c88 1653 struct pack_window **w_curs,
c4001d92 1654 off_t curpos,
1f688557 1655 unsigned long delta_size,
c4001d92 1656 off_t obj_offset,
21666f1a 1657 enum object_type *type,
eb950c19 1658 unsigned long *sizep)
1f688557 1659{
7c3e8be3 1660 void *delta_data, *result, *base;
c4001d92
SP
1661 unsigned long base_size;
1662 off_t base_offset;
43057304 1663
21666f1a 1664 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
8eca0b47
NP
1665 if (!base_offset) {
1666 error("failed to validate delta base reference "
1667 "at offset %"PRIuMAX" from %s",
1668 (uintmax_t)curpos, p->pack_name);
1669 return NULL;
1670 }
eac12e2d 1671 unuse_pack(w_curs);
a0cba108 1672 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
8eca0b47
NP
1673 if (!base) {
1674 /*
1675 * We're probably in deep shit, but let's try to fetch
1676 * the required base anyway from another pack or loose.
1677 * This is costly but should happen only in the presence
1678 * of a corrupted pack, and is better than failing outright.
1679 */
08698b1e
NP
1680 struct revindex_entry *revidx;
1681 const unsigned char *base_sha1;
1682 revidx = find_pack_revindex(p, base_offset);
1683 if (!revidx)
1684 return NULL;
1685 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
8eca0b47
NP
1686 error("failed to read delta base object %s"
1687 " at offset %"PRIuMAX" from %s",
1688 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1689 p->pack_name);
1690 mark_bad_packed_object(p, base_sha1);
ac939109 1691 base = read_object(base_sha1, type, &base_size);
8eca0b47
NP
1692 if (!base)
1693 return NULL;
1694 }
67686d95 1695
2b87c45b 1696 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
8eca0b47
NP
1697 if (!delta_data) {
1698 error("failed to unpack compressed delta "
1699 "at offset %"PRIuMAX" from %s",
1700 (uintmax_t)curpos, p->pack_name);
1701 free(base);
1702 return NULL;
1703 }
1f688557
JH
1704 result = patch_delta(base, base_size,
1705 delta_data, delta_size,
21666f1a 1706 sizep);
1f688557
JH
1707 if (!result)
1708 die("failed to apply delta");
1709 free(delta_data);
62f255ad 1710 add_delta_base_cache(p, base_offset, base, base_size, *type);
1f688557
JH
1711 return result;
1712}
1713
0e8189e2
NP
1714int do_check_packed_object_crc;
1715
c4001d92 1716void *unpack_entry(struct packed_git *p, off_t obj_offset,
21666f1a 1717 enum object_type *type, unsigned long *sizep)
1f688557 1718{
03e79c88 1719 struct pack_window *w_curs = NULL;
c4001d92 1720 off_t curpos = obj_offset;
21666f1a 1721 void *data;
1f688557 1722
0e8189e2
NP
1723 if (do_check_packed_object_crc && p->index_version > 1) {
1724 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1725 unsigned long len = revidx[1].offset - obj_offset;
1726 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1727 const unsigned char *sha1 =
1728 nth_packed_object_sha1(p, revidx->nr);
1729 error("bad packed object CRC for %s",
1730 sha1_to_hex(sha1));
1731 mark_bad_packed_object(p, sha1);
f755bb99 1732 unuse_pack(&w_curs);
0e8189e2
NP
1733 return NULL;
1734 }
1735 }
1736
21666f1a
NP
1737 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1738 switch (*type) {
eb32d236
NP
1739 case OBJ_OFS_DELTA:
1740 case OBJ_REF_DELTA:
21666f1a
NP
1741 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1742 obj_offset, type, sizep);
4d703a1a 1743 break;
a733cb60 1744 case OBJ_COMMIT:
a733cb60 1745 case OBJ_TREE:
a733cb60 1746 case OBJ_BLOB:
a733cb60 1747 case OBJ_TAG:
21666f1a 1748 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
4d703a1a 1749 break;
1f688557 1750 default:
8eca0b47
NP
1751 data = NULL;
1752 error("unknown object type %i at offset %"PRIuMAX" in %s",
1753 *type, (uintmax_t)obj_offset, p->pack_name);
1f688557 1754 }
03e79c88 1755 unuse_pack(&w_curs);
21666f1a 1756 return data;
1f688557
JH
1757}
1758
d079837e 1759const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 1760 uint32_t n)
9a217f2a 1761{
42873078 1762 const unsigned char *index = p->index_data;
d079837e
SP
1763 if (!index) {
1764 if (open_pack_index(p))
1765 return NULL;
1766 index = p->index_data;
1767 }
57059091 1768 if (n >= p->num_objects)
d72308e0 1769 return NULL;
74e34e1f
NP
1770 index += 4 * 256;
1771 if (p->index_version == 1) {
1772 return index + 24 * n + 4;
1773 } else {
1774 index += 8;
1775 return index + 20 * n;
1776 }
1777}
1778
99093238 1779off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
1780{
1781 const unsigned char *index = p->index_data;
1782 index += 4 * 256;
1783 if (p->index_version == 1) {
1784 return ntohl(*((uint32_t *)(index + 24 * n)));
1785 } else {
1786 uint32_t off;
1787 index += 8 + p->num_objects * (20 + 4);
1788 off = ntohl(*((uint32_t *)(index + 4 * n)));
1789 if (!(off & 0x80000000))
1790 return off;
1791 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1792 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1793 ntohl(*((uint32_t *)(index + 4)));
1794 }
9a217f2a
JH
1795}
1796
c4001d92 1797off_t find_pack_entry_one(const unsigned char *sha1,
43057304 1798 struct packed_git *p)
1f688557 1799{
42873078 1800 const uint32_t *level1_ofs = p->index_data;
42873078 1801 const unsigned char *index = p->index_data;
628522ec
JH
1802 unsigned hi, lo, stride;
1803 static int use_lookup = -1;
1804 static int debug_lookup = -1;
1805
1806 if (debug_lookup < 0)
1807 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 1808
d079837e
SP
1809 if (!index) {
1810 if (open_pack_index(p))
1811 return 0;
1812 level1_ofs = p->index_data;
1813 index = p->index_data;
1814 }
74e34e1f
NP
1815 if (p->index_version > 1) {
1816 level1_ofs += 2;
1817 index += 8;
1818 }
42873078 1819 index += 4 * 256;
74e34e1f
NP
1820 hi = ntohl(level1_ofs[*sha1]);
1821 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
1822 if (p->index_version > 1) {
1823 stride = 20;
1824 } else {
1825 stride = 24;
1826 index += 4;
1827 }
1828
1829 if (debug_lookup)
6e1c2344 1830 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
1831 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1832
1833 if (use_lookup < 0)
1834 use_lookup = !!getenv("GIT_USE_LOOKUP");
1835 if (use_lookup) {
1836 int pos = sha1_entry_pos(index, stride, 0,
1837 lo, hi, p->num_objects, sha1);
1838 if (pos < 0)
1839 return 0;
1840 return nth_packed_object_offset(p, pos);
1841 }
1f688557
JH
1842
1843 do {
74e34e1f 1844 unsigned mi = (lo + hi) / 2;
628522ec
JH
1845 int cmp = hashcmp(index + mi * stride, sha1);
1846
1847 if (debug_lookup)
1848 printf("lo %u hi %u rg %u mi %u\n",
1849 lo, hi, hi - lo, mi);
43057304 1850 if (!cmp)
74e34e1f 1851 return nth_packed_object_offset(p, mi);
1f688557
JH
1852 if (cmp > 0)
1853 hi = mi;
1854 else
1855 lo = mi+1;
1856 } while (lo < hi);
1857 return 0;
1858}
1859
4d6acb70 1860static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557 1861{
f7c22cc6 1862 static struct packed_git *last_found = (void *)1;
1f688557 1863 struct packed_git *p;
c4001d92 1864 off_t offset;
43057304 1865
1f688557 1866 prepare_packed_git();
f7c22cc6
NP
1867 if (!packed_git)
1868 return 0;
1869 p = (last_found == (void *)1) ? packed_git : last_found;
1f688557 1870
f7c22cc6 1871 do {
8eca0b47
NP
1872 if (p->num_bad_objects) {
1873 unsigned i;
1874 for (i = 0; i < p->num_bad_objects; i++)
1875 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1876 goto next;
1877 }
1878
43057304
NP
1879 offset = find_pack_entry_one(sha1, p);
1880 if (offset) {
c715f783
SP
1881 /*
1882 * We are about to tell the caller where they can
1883 * locate the requested object. We better make
1884 * sure the packfile is still here and can be
1885 * accessed before supplying that answer, as
1886 * it may have been deleted since the index
1887 * was loaded!
1888 */
1889 if (p->pack_fd == -1 && open_packed_git(p)) {
1890 error("packfile %s cannot be accessed", p->pack_name);
f7c22cc6 1891 goto next;
c715f783 1892 }
43057304
NP
1893 e->offset = offset;
1894 e->p = p;
1895 hashcpy(e->sha1, sha1);
f7c22cc6 1896 last_found = p;
1f688557 1897 return 1;
43057304 1898 }
f7c22cc6
NP
1899
1900 next:
1901 if (p == last_found)
1902 p = packed_git;
1903 else
1904 p = p->next;
1905 if (p == last_found)
1906 p = p->next;
1907 } while (p);
1f688557
JH
1908 return 0;
1909}
1910
a6080a0a 1911struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
1912 struct packed_git *packs)
1913{
1914 struct packed_git *p;
bf592c50
DB
1915
1916 for (p = packs; p; p = p->next) {
43057304 1917 if (find_pack_entry_one(sha1, p))
bf592c50
DB
1918 return p;
1919 }
1920 return NULL;
21666f1a 1921
bf592c50
DB
1922}
1923
21666f1a 1924static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
65c2e0c3 1925{
36e4d74a 1926 int status;
65c2e0c3
JH
1927 unsigned long mapsize, size;
1928 void *map;
1929 z_stream stream;
d65a16f6 1930 char hdr[32];
65c2e0c3 1931
bb6b8e4f 1932 map = map_sha1_file(sha1, &mapsize);
f0df4ed5
JS
1933 if (!map)
1934 return error("unable to find %s", sha1_to_hex(sha1));
36e4d74a
JH
1935 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1936 status = error("unable to unpack %s header",
1937 sha1_to_hex(sha1));
21666f1a 1938 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 1939 status = error("unable to parse %s header", sha1_to_hex(sha1));
21666f1a
NP
1940 else if (sizep)
1941 *sizep = size;
39c68542 1942 git_inflate_end(&stream);
65c2e0c3
JH
1943 munmap(map, mapsize);
1944 return status;
1945}
1946
21666f1a 1947int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
f0df4ed5 1948{
f0df4ed5 1949 struct pack_entry e;
ddd63e64 1950 int status;
f0df4ed5 1951
cd673c1f 1952 if (!find_pack_entry(sha1, &e)) {
ddd63e64
SG
1953 /* Most likely it's a loose object. */
1954 status = sha1_loose_object_info(sha1, sizep);
1955 if (status >= 0)
1956 return status;
1957
1958 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 1959 reprepare_packed_git();
cd673c1f 1960 if (!find_pack_entry(sha1, &e))
ddd63e64 1961 return status;
f0df4ed5 1962 }
3d77d877
NP
1963
1964 status = packed_object_info(e.p, e.offset, sizep);
1965 if (status < 0) {
1966 mark_bad_packed_object(e.p, sha1);
1967 status = sha1_object_info(sha1, sizep);
1968 }
1969
1970 return status;
f0df4ed5
JS
1971}
1972
21666f1a
NP
1973static void *read_packed_sha1(const unsigned char *sha1,
1974 enum object_type *type, unsigned long *size)
1f688557
JH
1975{
1976 struct pack_entry e;
8eca0b47 1977 void *data;
1f688557 1978
cd673c1f 1979 if (!find_pack_entry(sha1, &e))
1f688557 1980 return NULL;
8eca0b47
NP
1981 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1982 if (!data) {
1983 /*
1984 * We're probably in deep shit, but let's try to fetch
1985 * the required object anyway from another pack or loose.
1986 * This should happen only in the presence of a corrupted
1987 * pack, and is better than failing outright.
1988 */
1989 error("failed to read object %s at offset %"PRIuMAX" from %s",
1990 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
1991 mark_bad_packed_object(e.p, sha1);
ac939109 1992 data = read_object(sha1, type, size);
8eca0b47
NP
1993 }
1994 return data;
1f688557
JH
1995}
1996
d66b37bb
JH
1997/*
1998 * This is meant to hold a *small* number of objects that you would
1999 * want read_sha1_file() to be able to return, but yet you do not want
2000 * to write them into the object store (e.g. a browse-only
2001 * application).
2002 */
2003static struct cached_object {
2004 unsigned char sha1[20];
21666f1a 2005 enum object_type type;
d66b37bb
JH
2006 void *buf;
2007 unsigned long size;
2008} *cached_objects;
2009static int cached_object_nr, cached_object_alloc;
2010
346245a1 2011static struct cached_object empty_tree = {
14d9c578 2012 EMPTY_TREE_SHA1_BIN,
346245a1
JK
2013 OBJ_TREE,
2014 "",
2015 0
2016};
2017
d66b37bb
JH
2018static struct cached_object *find_cached_object(const unsigned char *sha1)
2019{
2020 int i;
2021 struct cached_object *co = cached_objects;
2022
2023 for (i = 0; i < cached_object_nr; i++, co++) {
2024 if (!hashcmp(co->sha1, sha1))
2025 return co;
2026 }
346245a1
JK
2027 if (!hashcmp(sha1, empty_tree.sha1))
2028 return &empty_tree;
d66b37bb
JH
2029 return NULL;
2030}
2031
21666f1a
NP
2032int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2033 unsigned char *sha1)
d66b37bb
JH
2034{
2035 struct cached_object *co;
2036
21666f1a 2037 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2038 if (has_sha1_file(sha1) || find_cached_object(sha1))
2039 return 0;
2040 if (cached_object_alloc <= cached_object_nr) {
2041 cached_object_alloc = alloc_nr(cached_object_alloc);
2042 cached_objects = xrealloc(cached_objects,
2043 sizeof(*cached_objects) *
2044 cached_object_alloc);
2045 }
2046 co = &cached_objects[cached_object_nr++];
2047 co->size = len;
21666f1a 2048 co->type = type;
efa13f7b
JH
2049 co->buf = xmalloc(len);
2050 memcpy(co->buf, buf, len);
d66b37bb
JH
2051 hashcpy(co->sha1, sha1);
2052 return 0;
2053}
2054
c2c5b270
CC
2055static void *read_object(const unsigned char *sha1, enum object_type *type,
2056 unsigned long *size)
0fcfd160
LT
2057{
2058 unsigned long mapsize;
2059 void *map, *buf;
d66b37bb
JH
2060 struct cached_object *co;
2061
2062 co = find_cached_object(sha1);
2063 if (co) {
21666f1a 2064 *type = co->type;
d66b37bb 2065 *size = co->size;
182af834 2066 return xmemdupz(co->buf, co->size);
d66b37bb 2067 }
0fcfd160 2068
8276c007
PE
2069 buf = read_packed_sha1(sha1, type, size);
2070 if (buf)
2071 return buf;
bb6b8e4f 2072 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2073 if (map) {
7efbff75 2074 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2075 munmap(map, mapsize);
2076 return buf;
2077 }
637cdd9d 2078 reprepare_packed_git();
8276c007 2079 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2080}
2081
b6c4cecc
JH
2082/*
2083 * This function dies on corrupt objects; the callers who want to
2084 * deal with them should arrange to call read_object() and give error
2085 * messages themselves.
2086 */
f5552aee
CC
2087void *read_sha1_file_repl(const unsigned char *sha1,
2088 enum object_type *type,
2089 unsigned long *size,
2090 const unsigned char **replacement)
ac939109 2091{
68095570 2092 const unsigned char *repl = lookup_replace_object(sha1);
3ba7a065 2093 void *data;
e8b15e61 2094 char *path;
b6c4cecc
JH
2095 const struct packed_git *p;
2096
3ba7a065
JH
2097 errno = 0;
2098 data = read_object(repl, type, size);
b6c4cecc
JH
2099 if (data) {
2100 if (replacement)
2101 *replacement = repl;
2102 return data;
2103 }
68095570 2104
3ba7a065
JH
2105 if (errno != ENOENT)
2106 die_errno("failed to read object %s", sha1_to_hex(sha1));
2107
68095570 2108 /* die if we replaced an object with one that does not exist */
b6c4cecc 2109 if (repl != sha1)
68095570
CC
2110 die("replacement %s not found for %s",
2111 sha1_to_hex(repl), sha1_to_hex(sha1));
2112
b6c4cecc
JH
2113 if (has_loose_object(repl)) {
2114 path = sha1_file_name(sha1);
2115 die("loose object %s (stored in %s) is corrupt",
2116 sha1_to_hex(repl), path);
e8b15e61 2117 }
68095570 2118
b6c4cecc
JH
2119 if ((p = has_packed_and_bad(repl)) != NULL)
2120 die("packed object %s (stored in %s) is corrupt",
2121 sha1_to_hex(repl), p->pack_name);
f5552aee 2122
b6c4cecc 2123 return NULL;
ac939109
NP
2124}
2125
40469ee9 2126void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2127 const char *required_type_name,
40469ee9
JH
2128 unsigned long *size,
2129 unsigned char *actual_sha1_return)
f4913f91 2130{
21666f1a 2131 enum object_type type, required_type;
f4913f91
JH
2132 void *buffer;
2133 unsigned long isize;
40469ee9 2134 unsigned char actual_sha1[20];
f4913f91 2135
21666f1a 2136 required_type = type_from_string(required_type_name);
e702496e 2137 hashcpy(actual_sha1, sha1);
40469ee9
JH
2138 while (1) {
2139 int ref_length = -1;
2140 const char *ref_type = NULL;
f4913f91 2141
21666f1a 2142 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2143 if (!buffer)
2144 return NULL;
21666f1a 2145 if (type == required_type) {
40469ee9
JH
2146 *size = isize;
2147 if (actual_sha1_return)
e702496e 2148 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2149 return buffer;
2150 }
2151 /* Handle references */
21666f1a 2152 else if (type == OBJ_COMMIT)
40469ee9 2153 ref_type = "tree ";
21666f1a 2154 else if (type == OBJ_TAG)
40469ee9
JH
2155 ref_type = "object ";
2156 else {
2157 free(buffer);
2158 return NULL;
2159 }
2160 ref_length = strlen(ref_type);
f4913f91 2161
50974ec9
MK
2162 if (ref_length + 40 > isize ||
2163 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2164 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2165 free(buffer);
2166 return NULL;
2167 }
1cf58e72 2168 free(buffer);
40469ee9
JH
2169 /* Now we have the ID of the referred-to object in
2170 * actual_sha1. Check again. */
f4913f91 2171 }
f4913f91
JH
2172}
2173
ce9fbf16 2174static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2175 const char *type, unsigned char *sha1,
d65a16f6 2176 char *hdr, int *hdrlen)
d410c0f5 2177{
9126f009 2178 git_SHA_CTX c;
d410c0f5
JH
2179
2180 /* Generate the header */
d65a16f6 2181 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
2182
2183 /* Sha1.. */
9126f009
NP
2184 git_SHA1_Init(&c);
2185 git_SHA1_Update(&c, hdr, *hdrlen);
2186 git_SHA1_Update(&c, buf, len);
2187 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2188}
2189
230f1322 2190/*
5a688fe4
JH
2191 * Move the just written object into its final resting place.
2192 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2193 * "moving" is only a part of what it does, when no patch between
2194 * master to pu changes the call sites of this function.
230f1322 2195 */
839837b9 2196int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322 2197{
e32c0a9c 2198 int ret = 0;
5a688fe4 2199
348df166 2200 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
2201 goto try_rename;
2202 else if (link(tmpfile, filename))
e32c0a9c 2203 ret = errno;
7ebb6fca
LT
2204
2205 /*
2206 * Coda hack - coda doesn't like cross-directory links,
2207 * so we fall back to a rename, which will mean that it
2208 * won't be able to check collisions, but that's not a
2209 * big deal.
2210 *
2211 * The same holds for FAT formatted media.
2212 *
3be1f18e 2213 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
2214 * left to unlink.
2215 */
2216 if (ret && ret != EEXIST) {
be66a6c4 2217 try_rename:
7ebb6fca 2218 if (!rename(tmpfile, filename))
3be1f18e 2219 goto out;
9e48b389 2220 ret = errno;
230f1322 2221 }
691f1a28 2222 unlink_or_warn(tmpfile);
230f1322
LT
2223 if (ret) {
2224 if (ret != EEXIST) {
916d081b 2225 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
230f1322
LT
2226 }
2227 /* FIXME!!! Collision check here ? */
2228 }
2229
3be1f18e 2230out:
5256b006 2231 if (adjust_shared_perm(filename))
5a688fe4 2232 return error("unable to set permission to '%s'", filename);
230f1322
LT
2233 return 0;
2234}
2235
4d548150
LT
2236static int write_buffer(int fd, const void *buf, size_t len)
2237{
d34cf19b 2238 if (write_in_full(fd, buf, len) < 0)
93822c22 2239 return error("file write error (%s)", strerror(errno));
4d548150
LT
2240 return 0;
2241}
2242
ce9fbf16 2243int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2244 unsigned char *sha1)
2245{
d65a16f6 2246 char hdr[32];
abdc3fc8
RS
2247 int hdrlen;
2248 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2249 return 0;
2250}
2251
e9039dd3
LT
2252/* Finalize a file on disk, and close it. */
2253static void close_sha1_file(int fd)
2254{
aafe9fba
LT
2255 if (fsync_object_files)
2256 fsync_or_die(fd, "sha1 file");
e9039dd3 2257 if (close(fd) != 0)
d824cbba 2258 die_errno("error when closing sha1 file");
e9039dd3
LT
2259}
2260
5723fe7e
LT
2261/* Size of directory component, including the ending '/' */
2262static inline int directory_size(const char *filename)
2263{
2264 const char *s = strrchr(filename, '/');
2265 if (!s)
2266 return 0;
2267 return s - filename + 1;
2268}
2269
2270/*
2271 * This creates a temporary file in the same directory as the final
2272 * 'filename'
2273 *
2274 * We want to avoid cross-directory filename renames, because those
2275 * can have problems on various filesystems (FAT, NFS, Coda).
2276 */
2277static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2278{
2279 int fd, dirlen = directory_size(filename);
2280
2281 if (dirlen + 20 > bufsiz) {
2282 errno = ENAMETOOLONG;
2283 return -1;
2284 }
2285 memcpy(buffer, filename, dirlen);
2286 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
5256b006 2287 fd = git_mkstemp_mode(buffer, 0444);
cbacbf4e 2288 if (fd < 0 && dirlen && errno == ENOENT) {
5723fe7e 2289 /* Make sure the directory exists */
6ff6af62 2290 memcpy(buffer, filename, dirlen);
5723fe7e 2291 buffer[dirlen-1] = 0;
3bfaf018 2292 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
5723fe7e
LT
2293 return -1;
2294
2295 /* Try again */
2296 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
5256b006 2297 fd = git_mkstemp_mode(buffer, 0444);
5723fe7e
LT
2298 }
2299 return fd;
2300}
2301
bbac7311 2302static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 2303 const void *buf, unsigned long len, time_t mtime)
0fcfd160 2304{
915308b1 2305 int fd, ret;
9892beba 2306 unsigned char compressed[4096];
0fcfd160 2307 z_stream stream;
748af44c
NP
2308 git_SHA_CTX c;
2309 unsigned char parano_sha1[20];
706bc531 2310 char *filename;
aac17941 2311 static char tmpfile[PATH_MAX];
a44c9a5e 2312
972a9155 2313 filename = sha1_file_name(sha1);
5723fe7e 2314 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
fd73ccf2
JS
2315 while (fd < 0 && errno == EMFILE && unuse_one_window(packed_git, -1))
2316 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
aac17941 2317 if (fd < 0) {
35243577 2318 if (errno == EACCES)
916d081b
PB
2319 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2320 else
2321 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
aac17941
LT
2322 }
2323
0fcfd160
LT
2324 /* Set it up */
2325 memset(&stream, 0, sizeof(stream));
12f6c308 2326 deflateInit(&stream, zlib_compression_level);
0fcfd160 2327 stream.next_out = compressed;
9892beba 2328 stream.avail_out = sizeof(compressed);
748af44c 2329 git_SHA1_Init(&c);
a44c9a5e
LT
2330
2331 /* First header.. */
d65a16f6 2332 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2333 stream.avail_in = hdrlen;
726f852b
NP
2334 while (deflate(&stream, 0) == Z_OK)
2335 /* nothing */;
748af44c 2336 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
2337
2338 /* Then the data itself.. */
c00e657d 2339 stream.next_in = (void *)buf;
a44c9a5e 2340 stream.avail_in = len;
9892beba 2341 do {
748af44c 2342 unsigned char *in0 = stream.next_in;
9892beba 2343 ret = deflate(&stream, Z_FINISH);
748af44c 2344 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
2345 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2346 die("unable to write sha1 file");
2347 stream.next_out = compressed;
2348 stream.avail_out = sizeof(compressed);
2349 } while (ret == Z_OK);
2350
ac54c277
LT
2351 if (ret != Z_STREAM_END)
2352 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
ac54c277
LT
2353 ret = deflateEnd(&stream);
2354 if (ret != Z_OK)
2355 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
2356 git_SHA1_Final(parano_sha1, &c);
2357 if (hashcmp(sha1, parano_sha1) != 0)
2358 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 2359
e9039dd3 2360 close_sha1_file(fd);
0fcfd160 2361
bbac7311
NP
2362 if (mtime) {
2363 struct utimbuf utb;
2364 utb.actime = mtime;
2365 utb.modtime = mtime;
2366 if (utime(tmpfile, &utb) < 0)
2367 warning("failed utime() on %s: %s",
2368 tmpfile, strerror(errno));
2369 }
2370
230f1322 2371 return move_temp_to_file(tmpfile, filename);
0fcfd160 2372}
8237b185 2373
c00e657d 2374int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
bbac7311
NP
2375{
2376 unsigned char sha1[20];
2377 char hdr[32];
2378 int hdrlen;
2379
2380 /* Normally if we have it in the pack then we do not bother writing
2381 * it out into .git/objects/??/?{38} file.
2382 */
2383 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2384 if (returnsha1)
2385 hashcpy(returnsha1, sha1);
2386 if (has_sha1_file(sha1))
2387 return 0;
2388 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2389}
2390
2391int force_object_loose(const unsigned char *sha1, time_t mtime)
2392{
bbac7311
NP
2393 void *buf;
2394 unsigned long len;
2395 enum object_type type;
2396 char hdr[32];
2397 int hdrlen;
1fb23e65 2398 int ret;
bbac7311 2399
c529d75a 2400 if (has_loose_object(sha1))
bbac7311
NP
2401 return 0;
2402 buf = read_packed_sha1(sha1, &type, &len);
2403 if (!buf)
2404 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2405 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1fb23e65
BS
2406 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2407 free(buf);
2408
2409 return ret;
bbac7311
NP
2410}
2411
bf592c50
DB
2412int has_pack_index(const unsigned char *sha1)
2413{
2414 struct stat st;
2415 if (stat(sha1_pack_index_name(sha1), &st))
2416 return 0;
2417 return 1;
bf592c50
DB
2418}
2419
cd673c1f
JH
2420int has_sha1_pack(const unsigned char *sha1)
2421{
2422 struct pack_entry e;
2423 return find_pack_entry(sha1, &e);
2424}
2425
8237b185
DB
2426int has_sha1_file(const unsigned char *sha1)
2427{
1f688557
JH
2428 struct pack_entry e;
2429
cd673c1f 2430 if (find_pack_entry(sha1, &e))
1f688557 2431 return 1;
c529d75a 2432 return has_loose_object(sha1);
8237b185 2433}
74400e71 2434
43df4f86
DP
2435static int index_mem(unsigned char *sha1, void *buf, size_t size,
2436 int write_object, enum object_type type, const char *path)
e7332f96 2437{
6c510bee 2438 int ret, re_allocated = 0;
74400e71 2439
7672db20 2440 if (!type)
edaec3fb 2441 type = OBJ_BLOB;
6c510bee
LT
2442
2443 /*
2444 * Convert blobs to git internal format
2445 */
43df4f86 2446 if ((type == OBJ_BLOB) && path) {
f285a2d7 2447 struct strbuf nbuf = STRBUF_INIT;
21e5ad50
SP
2448 if (convert_to_git(path, buf, size, &nbuf,
2449 write_object ? safe_crlf : 0)) {
b315c5c0 2450 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
2451 re_allocated = 1;
2452 }
2453 }
2454
7672db20 2455 if (write_object)
edaec3fb 2456 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 2457 else
edaec3fb 2458 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 2459 if (re_allocated)
6c510bee 2460 free(buf);
43df4f86
DP
2461 return ret;
2462}
2463
ea68b0ce
DP
2464#define SMALL_FILE_SIZE (32*1024)
2465
43df4f86
DP
2466int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2467 enum object_type type, const char *path)
2468{
2469 int ret;
2470 size_t size = xsize_t(st->st_size);
2471
2472 if (!S_ISREG(st->st_mode)) {
f285a2d7 2473 struct strbuf sbuf = STRBUF_INIT;
43df4f86
DP
2474 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2475 ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2476 type, path);
2477 else
2478 ret = -1;
2479 strbuf_release(&sbuf);
08bda208
DP
2480 } else if (!size) {
2481 ret = index_mem(sha1, NULL, size, write_object, type, path);
ea68b0ce
DP
2482 } else if (size <= SMALL_FILE_SIZE) {
2483 char *buf = xmalloc(size);
2484 if (size == read_in_full(fd, buf, size))
2485 ret = index_mem(sha1, buf, size, write_object, type,
2486 path);
2487 else
2488 ret = error("short read %s", strerror(errno));
2489 free(buf);
08bda208 2490 } else {
43df4f86
DP
2491 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2492 ret = index_mem(sha1, buf, size, write_object, type, path);
aac17941 2493 munmap(buf, size);
08bda208 2494 }
43df4f86 2495 close(fd);
aac17941 2496 return ret;
74400e71 2497}
ec1fcc16
JH
2498
2499int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2500{
2501 int fd;
b760d3aa 2502 struct strbuf sb = STRBUF_INIT;
ec1fcc16
JH
2503
2504 switch (st->st_mode & S_IFMT) {
2505 case S_IFREG:
2506 fd = open(path, O_RDONLY);
2507 if (fd < 0)
2508 return error("open(\"%s\"): %s", path,
2509 strerror(errno));
53bca91a 2510 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
ec1fcc16
JH
2511 return error("%s: failed to insert into database",
2512 path);
2513 break;
2514 case S_IFLNK:
b760d3aa 2515 if (strbuf_readlink(&sb, path, st->st_size)) {
ec1fcc16 2516 char *errstr = strerror(errno);
ec1fcc16
JH
2517 return error("readlink(\"%s\"): %s", path,
2518 errstr);
2519 }
abdc3fc8 2520 if (!write_object)
b760d3aa
LT
2521 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2522 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
ec1fcc16
JH
2523 return error("%s: failed to insert into database",
2524 path);
b760d3aa 2525 strbuf_release(&sb);
ec1fcc16 2526 break;
f35a6d3b
LT
2527 case S_IFDIR:
2528 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
2529 default:
2530 return error("%s: unsupported file type", path);
2531 }
2532 return 0;
2533}
a69e5429
JH
2534
2535int read_pack_header(int fd, struct pack_header *header)
2536{
c697ad14
HO
2537 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2538 /* "eof before pack header was fully read" */
2539 return PH_ERROR_EOF;
2540
a69e5429
JH
2541 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2542 /* "protocol error (pack signature mismatch detected)" */
2543 return PH_ERROR_PACK_SIGNATURE;
2544 if (!pack_version_ok(header->hdr_version))
2545 /* "protocol error (pack version unsupported)" */
2546 return PH_ERROR_PROTOCOL;
2547 return 0;
2548}
40d52ff7
JK
2549
2550void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2551{
2552 enum object_type type = sha1_object_info(sha1, NULL);
2553 if (type < 0)
2554 die("%s is not a valid object", sha1_to_hex(sha1));
2555 if (type != expect)
2556 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2557 typename(expect));
2558}