mingw: add minimum getrlimit() compatibility stub
[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
f2e872aa 38static int git_open_noatime(const char *name, struct packed_git *p);
4865d2b6 39
c597ba80
NTND
40/*
41 * This is meant to hold a *small* number of objects that you would
42 * want read_sha1_file() to be able to return, but yet you do not want
43 * to write them into the object store (e.g. a browse-only
44 * application).
45 */
46static struct cached_object {
47 unsigned char sha1[20];
48 enum object_type type;
49 void *buf;
50 unsigned long size;
51} *cached_objects;
52static int cached_object_nr, cached_object_alloc;
53
54static struct cached_object empty_tree = {
dab0d410 55 EMPTY_TREE_SHA1_BIN_LITERAL,
c597ba80
NTND
56 OBJ_TREE,
57 "",
58 0
59};
60
61static struct cached_object *find_cached_object(const unsigned char *sha1)
62{
63 int i;
64 struct cached_object *co = cached_objects;
65
66 for (i = 0; i < cached_object_nr; i++, co++) {
67 if (!hashcmp(co->sha1, sha1))
68 return co;
69 }
70 if (!hashcmp(sha1, empty_tree.sha1))
71 return &empty_tree;
72 return NULL;
73}
74
b2cb9425
JH
75int safe_create_leading_directories(char *path)
76{
8385abfd 77 char *pos = path + offset_1st_component(path);
67d42212
JR
78 struct stat st;
79
b2cb9425
JH
80 while (pos) {
81 pos = strchr(pos, '/');
82 if (!pos)
83 break;
5f0bdf50
JH
84 while (*++pos == '/')
85 ;
86 if (!*pos)
87 break;
88 *--pos = '\0';
67d42212
JR
89 if (!stat(path, &st)) {
90 /* path exists */
91 if (!S_ISDIR(st.st_mode)) {
b2cb9425 92 *pos = '/';
67d42212 93 return -3;
b2cb9425 94 }
457f06d6 95 }
67d42212
JR
96 else if (mkdir(path, 0777)) {
97 *pos = '/';
98 return -1;
99 }
457f06d6
JS
100 else if (adjust_shared_perm(path)) {
101 *pos = '/';
102 return -2;
103 }
b2cb9425
JH
104 *pos++ = '/';
105 }
106 return 0;
107}
723c31fe 108
8e21d63b
JK
109int safe_create_leading_directories_const(const char *path)
110{
111 /* path points to cache entries, so xstrdup before messing with it */
112 char *buf = xstrdup(path);
113 int result = safe_create_leading_directories(buf);
114 free(buf);
115 return result;
116}
117
ace1534d
JH
118static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
119{
120 int i;
121 for (i = 0; i < 20; i++) {
122 static char hex[] = "0123456789abcdef";
123 unsigned int val = sha1[i];
124 char *pos = pathbuf + i*2 + (i > 0);
125 *pos++ = hex[val >> 4];
126 *pos = hex[val & 0xf];
127 }
128}
129
0fcfd160
LT
130/*
131 * NOTE! This returns a statically allocated buffer, so you have to be
790296fd 132 * careful about using it. Do an "xstrdup()" if you need to save the
0fcfd160 133 * filename.
ace1534d
JH
134 *
135 * Also note that this returns the location for creating. Reading
136 * SHA1 file can happen from any alternate directory listed in the
d19938ab 137 * DB_ENVIRONMENT environment variable if it is not found in
ace1534d 138 * the primary object database.
0fcfd160
LT
139 */
140char *sha1_file_name(const unsigned char *sha1)
141{
560fb6a1
JK
142 static char buf[PATH_MAX];
143 const char *objdir;
144 int len;
0fcfd160 145
560fb6a1
JK
146 objdir = get_object_directory();
147 len = strlen(objdir);
148
149 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
150 if (len + 43 > PATH_MAX)
151 die("insanely long object directory %s", objdir);
152 memcpy(buf, objdir, len);
153 buf[len] = '/';
154 buf[len+3] = '/';
155 buf[len+42] = '\0';
156 fill_sha1_path(buf + len + 1, sha1);
157 return buf;
0fcfd160
LT
158}
159
633f43e1 160static char *sha1_get_pack_name(const unsigned char *sha1,
6eec46bd 161 char **name, char **base, const char *which)
bf592c50
DB
162{
163 static const char hex[] = "0123456789abcdef";
633f43e1 164 char *buf;
bf592c50
DB
165 int i;
166
633f43e1 167 if (!*base) {
bf592c50
DB
168 const char *sha1_file_directory = get_object_directory();
169 int len = strlen(sha1_file_directory);
633f43e1 170 *base = xmalloc(len + 60);
6eec46bd
JH
171 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
172 sha1_file_directory, which);
633f43e1 173 *name = *base + len + 11;
bf592c50
DB
174 }
175
633f43e1 176 buf = *name;
bf592c50
DB
177
178 for (i = 0; i < 20; i++) {
179 unsigned int val = *sha1++;
180 *buf++ = hex[val >> 4];
181 *buf++ = hex[val & 0xf];
182 }
a6080a0a 183
633f43e1 184 return *base;
bf592c50
DB
185}
186
633f43e1 187char *sha1_pack_name(const unsigned char *sha1)
bf592c50 188{
633f43e1 189 static char *name, *base;
bf592c50 190
6eec46bd 191 return sha1_get_pack_name(sha1, &name, &base, "pack");
633f43e1 192}
bf592c50 193
633f43e1
HO
194char *sha1_pack_index_name(const unsigned char *sha1)
195{
196 static char *name, *base;
a6080a0a 197
6eec46bd 198 return sha1_get_pack_name(sha1, &name, &base, "idx");
bf592c50
DB
199}
200
d5a63b99
JH
201struct alternate_object_database *alt_odb_list;
202static struct alternate_object_database **alt_odb_tail;
ace1534d 203
c2f493a4
MW
204static void read_info_alternates(const char * alternates, int depth);
205
ddd5d056
JH
206/*
207 * Prepare alternate object database registry.
d5a63b99
JH
208 *
209 * The variable alt_odb_list points at the list of struct
210 * alternate_object_database. The elements on this list come from
211 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
212 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
213 * whose contents is similar to that environment variable but can be
214 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
215 * contains "/the/directory/corresponding/to/.git/objects/...", while
216 * its name points just after the slash at the end of ".git/objects/"
217 * in the example above, and has enough space to hold 40-byte hex
218 * SHA1, an extra slash for the first level indirection, and the
219 * terminating NUL.
ddd5d056 220 */
c2f493a4 221static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
ace1534d 222{
1494e038 223 const char *objdir = get_object_directory();
c2f493a4
MW
224 struct alternate_object_database *ent;
225 struct alternate_object_database *alt;
226 /* 43 = 40-byte + 2 '/' + terminating NUL */
227 int pfxlen = len;
228 int entlen = pfxlen + 43;
ccfd3e99 229 int base_len = -1;
d5a63b99 230
85dadc38 231 if (!is_absolute_path(entry) && relative_base) {
c2f493a4
MW
232 /* Relative alt-odb */
233 if (base_len < 0)
234 base_len = strlen(relative_base) + 1;
235 entlen += base_len;
236 pfxlen += base_len;
237 }
238 ent = xmalloc(sizeof(*ent) + entlen);
239
85dadc38 240 if (!is_absolute_path(entry) && relative_base) {
c2f493a4
MW
241 memcpy(ent->base, relative_base, base_len - 1);
242 ent->base[base_len - 1] = '/';
243 memcpy(ent->base + base_len, entry, len);
244 }
245 else
246 memcpy(ent->base, entry, pfxlen);
247
248 ent->name = ent->base + pfxlen + 1;
249 ent->base[pfxlen + 3] = '/';
250 ent->base[pfxlen] = ent->base[entlen-1] = 0;
251
252 /* Detect cases where alternate disappeared */
90b4a71c 253 if (!is_directory(ent->base)) {
c2f493a4
MW
254 error("object directory %s does not exist; "
255 "check .git/objects/info/alternates.",
256 ent->base);
257 free(ent);
258 return -1;
259 }
260
261 /* Prevent the common mistake of listing the same
262 * thing twice, or object directory itself.
263 */
264 for (alt = alt_odb_list; alt; alt = alt->next) {
265 if (!memcmp(ent->base, alt->base, pfxlen)) {
266 free(ent);
267 return -1;
268 }
269 }
270 if (!memcmp(ent->base, objdir, pfxlen)) {
271 free(ent);
272 return -1;
273 }
274
275 /* add the alternate entry */
276 *alt_odb_tail = ent;
277 alt_odb_tail = &(ent->next);
278 ent->next = NULL;
279
280 /* recursively add alternates */
281 read_info_alternates(ent->base, depth + 1);
282
283 ent->base[pfxlen] = '/';
284
285 return 0;
286}
287
288static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
289 const char *relative_base, int depth)
290{
291 const char *cp, *last;
292
293 if (depth > 5) {
294 error("%s: ignoring alternate object stores, nesting too deep.",
295 relative_base);
296 return;
297 }
298
d5a63b99 299 last = alt;
9577e7e3
JH
300 while (last < ep) {
301 cp = last;
302 if (cp < ep && *cp == '#') {
303 while (cp < ep && *cp != sep)
304 cp++;
305 last = cp + 1;
306 continue;
307 }
c2f493a4
MW
308 while (cp < ep && *cp != sep)
309 cp++;
d5a63b99 310 if (last != cp) {
85dadc38 311 if (!is_absolute_path(last) && depth) {
c2f493a4
MW
312 error("%s: ignoring relative alternate object store %s",
313 relative_base, last);
314 } else {
315 link_alt_odb_entry(last, cp - last,
316 relative_base, depth);
1494e038 317 }
d5a63b99 318 }
9577e7e3 319 while (cp < ep && *cp == sep)
d5a63b99
JH
320 cp++;
321 last = cp;
9577e7e3 322 }
d5a63b99
JH
323}
324
c2f493a4 325static void read_info_alternates(const char * relative_base, int depth)
d5a63b99 326{
9577e7e3 327 char *map;
dc49cd76 328 size_t mapsz;
d5a63b99 329 struct stat st;
9cb18f56
JM
330 const char alt_file_name[] = "info/alternates";
331 /* Given that relative_base is no longer than PATH_MAX,
332 ensure that "path" has enough space to append "/", the
333 file name, "info/alternates", and a trailing NUL. */
334 char path[PATH_MAX + 1 + sizeof alt_file_name];
c2f493a4 335 int fd;
d5a63b99 336
9cb18f56 337 sprintf(path, "%s/%s", relative_base, alt_file_name);
f2e872aa 338 fd = git_open_noatime(path, NULL);
d5a63b99
JH
339 if (fd < 0)
340 return;
341 if (fstat(fd, &st) || (st.st_size == 0)) {
342 close(fd);
9a217f2a 343 return;
ace1534d 344 }
dc49cd76
SP
345 mapsz = xsize_t(st.st_size);
346 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 347 close(fd);
d5a63b99 348
dc49cd76 349 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
c2f493a4 350
dc49cd76 351 munmap(map, mapsz);
ace1534d
JH
352}
353
bef70b22
DB
354void add_to_alternates_file(const char *reference)
355{
356 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
acd3b9ec 357 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
bef70b22
DB
358 char *alt = mkpath("%s/objects\n", reference);
359 write_or_die(fd, alt, strlen(alt));
360 if (commit_lock_file(lock))
361 die("could not close alternates file");
362 if (alt_odb_tail)
363 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
364}
365
d79796bc
JH
366void foreach_alt_odb(alt_odb_fn fn, void *cb)
367{
368 struct alternate_object_database *ent;
369
370 prepare_alt_odb();
371 for (ent = alt_odb_list; ent; ent = ent->next)
372 if (fn(ent, cb))
373 return;
374}
375
c2f493a4
MW
376void prepare_alt_odb(void)
377{
554fe20d 378 const char *alt;
c2f493a4 379
7dc24aa5
SP
380 if (alt_odb_tail)
381 return;
382
c2f493a4
MW
383 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
384 if (!alt) alt = "";
385
c2f493a4 386 alt_odb_tail = &alt_odb_list;
80ba074f 387 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
388
389 read_info_alternates(get_object_directory(), 0);
390}
391
0f4dc14a 392static int has_loose_object_local(const unsigned char *sha1)
ace1534d 393{
ace1534d 394 char *name = sha1_file_name(sha1);
0f4dc14a
BC
395 return !access(name, F_OK);
396}
ace1534d 397
0f4dc14a
BC
398int has_loose_object_nonlocal(const unsigned char *sha1)
399{
400 struct alternate_object_database *alt;
9a217f2a 401 prepare_alt_odb();
d5a63b99 402 for (alt = alt_odb_list; alt; alt = alt->next) {
0f4dc14a 403 fill_sha1_path(alt->name, sha1);
c529d75a
LT
404 if (!access(alt->base, F_OK))
405 return 1;
ace1534d 406 }
c529d75a 407 return 0;
ace1534d
JH
408}
409
0f4dc14a
BC
410static int has_loose_object(const unsigned char *sha1)
411{
412 return has_loose_object_local(sha1) ||
413 has_loose_object_nonlocal(sha1);
414}
415
60bb8b14 416static unsigned int pack_used_ctr;
a53128b6
SP
417static unsigned int pack_mmap_calls;
418static unsigned int peak_pack_open_windows;
419static unsigned int pack_open_windows;
c7934306
SP
420static unsigned int pack_open_fds;
421static unsigned int pack_max_fds;
a53128b6 422static size_t peak_pack_mapped;
60bb8b14 423static size_t pack_mapped;
9a217f2a 424struct packed_git *packed_git;
1f688557 425
b79d18c9 426void pack_report(void)
a53128b6
SP
427{
428 fprintf(stderr,
e05db0fd
PR
429 "pack_report: getpagesize() = %10" SZ_FMT "\n"
430 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
431 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
432 sz_fmt(getpagesize()),
433 sz_fmt(packed_git_window_size),
434 sz_fmt(packed_git_limit));
a53128b6
SP
435 fprintf(stderr,
436 "pack_report: pack_used_ctr = %10u\n"
437 "pack_report: pack_mmap_calls = %10u\n"
438 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
439 "pack_report: pack_mapped = "
440 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
441 pack_used_ctr,
442 pack_mmap_calls,
443 pack_open_windows, peak_pack_open_windows,
9e42d6a1 444 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
445}
446
42873078 447static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
448{
449 void *idx_map;
42873078 450 struct pack_idx_header *hdr;
dc49cd76 451 size_t idx_size;
74e34e1f 452 uint32_t version, nr, i, *index;
f2e872aa 453 int fd = git_open_noatime(path, p);
1f688557 454 struct stat st;
42873078 455
1f688557
JH
456 if (fd < 0)
457 return -1;
458 if (fstat(fd, &st)) {
459 close(fd);
460 return -1;
461 }
dc49cd76 462 idx_size = xsize_t(st.st_size);
2d88451b
SP
463 if (idx_size < 4 * 256 + 20 + 20) {
464 close(fd);
465 return error("index file %s is too small", path);
466 }
c4712e45 467 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 468 close(fd);
1f688557 469
42873078
NP
470 hdr = idx_map;
471 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
472 version = ntohl(hdr->idx_version);
473 if (version < 2 || version > 2) {
474 munmap(idx_map, idx_size);
6e1c2344 475 return error("index file %s is version %"PRIu32
74e34e1f
NP
476 " and is not supported by this binary"
477 " (try upgrading GIT to a newer version)",
478 path, version);
479 }
480 } else
481 version = 1;
df1b059d 482
1f688557 483 nr = 0;
42873078 484 index = idx_map;
74e34e1f
NP
485 if (version > 1)
486 index += 2; /* skip index header */
1f688557 487 for (i = 0; i < 256; i++) {
326bf396 488 uint32_t n = ntohl(index[i]);
2d88451b
SP
489 if (n < nr) {
490 munmap(idx_map, idx_size);
df1b059d 491 return error("non-monotonic index %s", path);
2d88451b 492 }
1f688557
JH
493 nr = n;
494 }
495
74e34e1f
NP
496 if (version == 1) {
497 /*
498 * Total size:
499 * - 256 index entries 4 bytes each
500 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
501 * - 20-byte SHA1 of the packfile
502 * - 20-byte SHA1 file checksum
503 */
504 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
505 munmap(idx_map, idx_size);
eef427a0 506 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
507 }
508 } else if (version == 2) {
509 /*
510 * Minimum size:
511 * - 8 bytes of header
512 * - 256 index entries 4 bytes each
513 * - 20-byte sha1 entry * nr
514 * - 4-byte crc entry * nr
515 * - 4-byte offset entry * nr
516 * - 20-byte SHA1 of the packfile
517 * - 20-byte SHA1 file checksum
518 * And after the 4-byte offset table might be a
519 * variable sized table containing 8-byte entries
520 * for offsets larger than 2^31.
521 */
522 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
523 unsigned long max_size = min_size;
524 if (nr)
525 max_size += (nr - 1)*8;
526 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 527 munmap(idx_map, idx_size);
eef427a0 528 return error("wrong index v2 file size in %s", path);
74e34e1f 529 }
7109c889
JH
530 if (idx_size != min_size &&
531 /*
532 * make sure we can deal with large pack offsets.
533 * 31-bit signed offset won't be enough, neither
534 * 32-bit unsigned one will be.
535 */
536 (sizeof(off_t) <= 4)) {
537 munmap(idx_map, idx_size);
538 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 539 }
2d88451b 540 }
1f688557 541
74e34e1f 542 p->index_version = version;
42873078
NP
543 p->index_data = idx_map;
544 p->index_size = idx_size;
57059091 545 p->num_objects = nr;
1f688557
JH
546 return 0;
547}
548
bc8e478a 549int open_pack_index(struct packed_git *p)
d079837e
SP
550{
551 char *idx_name;
552 int ret;
553
554 if (p->index_data)
555 return 0;
556
557 idx_name = xstrdup(p->pack_name);
558 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
559 ret = check_packed_git_idx(idx_name, p);
560 free(idx_name);
561 return ret;
562}
563
11daf39b
SP
564static void scan_windows(struct packed_git *p,
565 struct packed_git **lru_p,
566 struct pack_window **lru_w,
567 struct pack_window **lru_l)
1f688557 568{
11daf39b
SP
569 struct pack_window *w, *w_l;
570
571 for (w_l = NULL, w = p->windows; w; w = w->next) {
572 if (!w->inuse_cnt) {
573 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
574 *lru_p = p;
575 *lru_w = w;
576 *lru_l = w_l;
54044bf8 577 }
54044bf8 578 }
11daf39b 579 w_l = w;
f9253394 580 }
11daf39b
SP
581}
582
d1efefa4 583static int unuse_one_window(struct packed_git *current, int keep_fd)
11daf39b
SP
584{
585 struct packed_git *p, *lru_p = NULL;
586 struct pack_window *lru_w = NULL, *lru_l = NULL;
587
588 if (current)
589 scan_windows(current, &lru_p, &lru_w, &lru_l);
590 for (p = packed_git; p; p = p->next)
591 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
592 if (lru_p) {
593 munmap(lru_w->base, lru_w->len);
594 pack_mapped -= lru_w->len;
595 if (lru_l)
596 lru_l->next = lru_w->next;
597 else {
598 lru_p->windows = lru_w->next;
d1efefa4 599 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
54044bf8 600 close(lru_p->pack_fd);
c7934306 601 pack_open_fds--;
54044bf8
SP
602 lru_p->pack_fd = -1;
603 }
604 }
605 free(lru_w);
a53128b6 606 pack_open_windows--;
54044bf8
SP
607 return 1;
608 }
609 return 0;
f9253394
JH
610}
611
d1efefa4 612void release_pack_memory(size_t need, int fd)
97bfeb34
SP
613{
614 size_t cur = pack_mapped;
d1efefa4 615 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
97bfeb34
SP
616 ; /* nothing */
617}
618
58ecbd5e
JN
619void *xmmap(void *start, size_t length,
620 int prot, int flags, int fd, off_t offset)
621{
622 void *ret = mmap(start, length, prot, flags, fd, offset);
623 if (ret == MAP_FAILED) {
624 if (!length)
625 return NULL;
626 release_pack_memory(length, fd);
627 ret = mmap(start, length, prot, flags, fd, offset);
628 if (ret == MAP_FAILED)
629 die_errno("Out of memory? mmap failed");
630 }
631 return ret;
632}
633
c9ced051
SP
634void close_pack_windows(struct packed_git *p)
635{
636 while (p->windows) {
637 struct pack_window *w = p->windows;
638
639 if (w->inuse_cnt)
640 die("pack '%s' still has open windows to it",
641 p->pack_name);
642 munmap(w->base, w->len);
643 pack_mapped -= w->len;
644 pack_open_windows--;
645 p->windows = w->next;
646 free(w);
647 }
648}
649
03e79c88 650void unuse_pack(struct pack_window **w_cursor)
f9253394 651{
03e79c88
SP
652 struct pack_window *w = *w_cursor;
653 if (w) {
654 w->inuse_cnt--;
655 *w_cursor = NULL;
656 }
1f688557
JH
657}
658
fa5fc15d
SP
659void close_pack_index(struct packed_git *p)
660{
661 if (p->index_data) {
662 munmap((void *)p->index_data, p->index_size);
663 p->index_data = NULL;
664 }
665}
666
c74faea1
NP
667/*
668 * This is used by git-repack in case a newly created pack happens to
669 * contain the same set of objects as an existing one. In that case
670 * the resulting file might be different even if its name would be the
671 * same. It is best to close any reference to the old pack before it is
672 * replaced on disk. Of course no index pointers nor windows for given pack
673 * must subsist at this point. If ever objects from this pack are requested
674 * again, the new version of the pack will be reinitialized through
675 * reprepare_packed_git().
676 */
677void free_pack_by_name(const char *pack_name)
678{
679 struct packed_git *p, **pp = &packed_git;
680
681 while (*pp) {
682 p = *pp;
683 if (strcmp(pack_name, p->pack_name) == 0) {
fa3a0c94 684 clear_delta_base_cache();
c74faea1 685 close_pack_windows(p);
c7934306 686 if (p->pack_fd != -1) {
c74faea1 687 close(p->pack_fd);
c7934306
SP
688 pack_open_fds--;
689 }
fa5fc15d 690 close_pack_index(p);
c74faea1
NP
691 free(p->bad_object_sha1);
692 *pp = p->next;
693 free(p);
694 return;
695 }
696 pp = &p->next;
697 }
698}
699
3cf8b462
SP
700/*
701 * Do not call this directly as this leaks p->pack_fd on error return;
702 * call open_packed_git() instead.
703 */
704static int open_packed_git_1(struct packed_git *p)
1f688557 705{
9bc879c1
SP
706 struct stat st;
707 struct pack_header hdr;
708 unsigned char sha1[20];
709 unsigned char *idx_sha1;
2c039da8 710 long fd_flag;
9bc879c1 711
d079837e
SP
712 if (!p->index_data && open_pack_index(p))
713 return error("packfile %s index unavailable", p->pack_name);
714
c7934306
SP
715 if (!pack_max_fds) {
716 struct rlimit lim;
717 unsigned int max_fds;
718
719 if (getrlimit(RLIMIT_NOFILE, &lim))
720 die_errno("cannot get RLIMIT_NOFILE");
721
722 max_fds = lim.rlim_cur;
723
724 /* Save 3 for stdin/stdout/stderr, 22 for work */
725 if (25 < max_fds)
726 pack_max_fds = max_fds - 25;
727 else
728 pack_max_fds = 1;
729 }
730
731 while (pack_max_fds <= pack_open_fds && unuse_one_window(NULL, -1))
732 ; /* nothing */
733
f2e872aa 734 p->pack_fd = git_open_noatime(p->pack_name, p);
9bc879c1 735 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 736 return -1;
c7934306 737 pack_open_fds++;
9bc879c1
SP
738
739 /* If we created the struct before we had the pack we lack size. */
bf592c50 740 if (!p->pack_size) {
bf592c50 741 if (!S_ISREG(st.st_mode))
072db278 742 return error("packfile %s not a regular file", p->pack_name);
bf592c50 743 p->pack_size = st.st_size;
9bc879c1 744 } else if (p->pack_size != st.st_size)
072db278 745 return error("packfile %s size changed", p->pack_name);
9bc879c1 746
2c039da8
JH
747 /* We leave these file descriptors open with sliding mmap;
748 * there is no point keeping them open across exec(), though.
749 */
750 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
751 if (fd_flag < 0)
072db278 752 return error("cannot determine file descriptor flags");
2c039da8
JH
753 fd_flag |= FD_CLOEXEC;
754 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 755 return error("cannot set FD_CLOEXEC");
2c039da8 756
9bc879c1 757 /* Verify we recognize this pack file format. */
e6e2bd62 758 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 759 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 760 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 761 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 762 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
763 return error("packfile %s is version %"PRIu32" and not"
764 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
765 p->pack_name, ntohl(hdr.hdr_version));
766
767 /* Verify the pack matches its index. */
57059091 768 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
769 return error("packfile %s claims to have %"PRIu32" objects"
770 " while index indicates %"PRIu32" objects",
57059091
NP
771 p->pack_name, ntohl(hdr.hdr_entries),
772 p->num_objects);
9bc879c1 773 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 774 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 775 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 776 return error("packfile %s signature is unavailable", p->pack_name);
42873078 777 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 778 if (hashcmp(sha1, idx_sha1))
072db278
SP
779 return error("packfile %s does not match index", p->pack_name);
780 return 0;
9bc879c1
SP
781}
782
3cf8b462
SP
783static int open_packed_git(struct packed_git *p)
784{
785 if (!open_packed_git_1(p))
786 return 0;
787 if (p->pack_fd != -1) {
788 close(p->pack_fd);
c7934306 789 pack_open_fds--;
3cf8b462
SP
790 p->pack_fd = -1;
791 }
792 return -1;
793}
794
c4001d92 795static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
796{
797 /* We must promise at least 20 bytes (one hash) after the
798 * offset is available from this window, otherwise the offset
799 * is not actually in this window and a different window (which
800 * has that one hash excess) must be used. This is to support
801 * the object header and delta base parsing routines below.
802 */
803 off_t win_off = win->offset;
804 return win_off <= offset
805 && (offset + 20) <= (win_off + win->len);
806}
807
4b25d091 808unsigned char *use_pack(struct packed_git *p,
03e79c88 809 struct pack_window **w_cursor,
c4001d92 810 off_t offset,
03e79c88 811 unsigned int *left)
9bc879c1 812{
60bb8b14 813 struct pack_window *win = *w_cursor;
03e79c88 814
072db278
SP
815 if (p->pack_fd == -1 && open_packed_git(p))
816 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14 817
a9d98a14 818 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
819 * pointless to ask for an offset into the middle of that
820 * hash, and the in_window function above wouldn't match
821 * don't allow an offset too close to the end of the file.
822 */
823 if (offset > (p->pack_size - 20))
824 die("offset beyond end of packfile (truncated pack?)");
825
826 if (!win || !in_window(win, offset)) {
827 if (win)
828 win->inuse_cnt--;
829 for (win = p->windows; win; win = win->next) {
830 if (in_window(win, offset))
831 break;
832 }
833 if (!win) {
78a28df9 834 size_t window_align = packed_git_window_size / 2;
dc49cd76 835 off_t len;
60bb8b14 836 win = xcalloc(1, sizeof(*win));
78a28df9 837 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
838 len = p->pack_size - win->offset;
839 if (len > packed_git_window_size)
840 len = packed_git_window_size;
841 win->len = (size_t)len;
60bb8b14 842 pack_mapped += win->len;
11daf39b 843 while (packed_git_limit < pack_mapped
d1efefa4 844 && unuse_one_window(p, p->pack_fd))
60bb8b14 845 ; /* nothing */
c4712e45 846 win->base = xmmap(NULL, win->len,
60bb8b14
SP
847 PROT_READ, MAP_PRIVATE,
848 p->pack_fd, win->offset);
849 if (win->base == MAP_FAILED)
73b4e4be
SP
850 die("packfile %s cannot be mapped: %s",
851 p->pack_name,
852 strerror(errno));
a53128b6
SP
853 pack_mmap_calls++;
854 pack_open_windows++;
855 if (pack_mapped > peak_pack_mapped)
856 peak_pack_mapped = pack_mapped;
857 if (pack_open_windows > peak_pack_open_windows)
858 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
859 win->next = p->windows;
860 p->windows = win;
861 }
1f688557 862 }
03e79c88
SP
863 if (win != *w_cursor) {
864 win->last_used = pack_used_ctr++;
865 win->inuse_cnt++;
866 *w_cursor = win;
867 }
60bb8b14 868 offset -= win->offset;
03e79c88 869 if (left)
dc49cd76 870 *left = win->len - xsize_t(offset);
03e79c88 871 return win->base + offset;
1f688557
JH
872}
873
27d69a46
NP
874static struct packed_git *alloc_packed_git(int extra)
875{
876 struct packed_git *p = xmalloc(sizeof(*p) + extra);
877 memset(p, 0, sizeof(*p));
878 p->pack_fd = -1;
879 return p;
880}
881
e0500293
JN
882static void try_to_free_pack_memory(size_t size)
883{
884 release_pack_memory(size, -1);
885}
886
42873078 887struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557 888{
e0500293 889 static int have_set_try_to_free_routine;
1f688557 890 struct stat st;
27d69a46 891 struct packed_git *p = alloc_packed_git(path_len + 2);
1f688557 892
e0500293
JN
893 if (!have_set_try_to_free_routine) {
894 have_set_try_to_free_routine = 1;
895 set_try_to_free_routine(try_to_free_pack_memory);
896 }
897
42873078
NP
898 /*
899 * Make sure a corresponding .pack file exists and that
900 * the index looks sane.
901 */
902 path_len -= strlen(".idx");
27d69a46
NP
903 if (path_len < 1) {
904 free(p);
1f688557 905 return NULL;
27d69a46 906 }
42873078 907 memcpy(p->pack_name, path, path_len);
8d25931d
BC
908
909 strcpy(p->pack_name + path_len, ".keep");
910 if (!access(p->pack_name, F_OK))
911 p->pack_keep = 1;
912
42873078 913 strcpy(p->pack_name + path_len, ".pack");
d079837e 914 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 915 free(p);
1f688557
JH
916 return NULL;
917 }
42873078 918
1f688557
JH
919 /* ok, it looks sane as far as we can check without
920 * actually mapping the pack file.
921 */
1f688557 922 p->pack_size = st.st_size;
9d835df2 923 p->pack_local = local;
b867092f 924 p->mtime = st.st_mtime;
42873078
NP
925 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
926 hashclr(p->sha1);
1f688557
JH
927 return p;
928}
929
7b64469a 930struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 931{
42873078 932 const char *path = sha1_pack_name(sha1);
27d69a46 933 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
bf592c50 934
27d69a46
NP
935 strcpy(p->pack_name, path);
936 hashcpy(p->sha1, sha1);
42873078
NP
937 if (check_packed_git_idx(idx_path, p)) {
938 free(p);
bf592c50 939 return NULL;
42873078 940 }
bf592c50 941
bf592c50
DB
942 return p;
943}
944
945void install_packed_git(struct packed_git *pack)
946{
c7934306
SP
947 if (pack->pack_fd != -1)
948 pack_open_fds++;
949
bf592c50
DB
950 pack->next = packed_git;
951 packed_git = pack;
952}
953
9d835df2 954static void prepare_packed_git_one(char *objdir, int local)
1f688557 955{
9cb18f56
JM
956 /* Ensure that this buffer is large enough so that we can
957 append "/pack/" without clobbering the stack even if
958 strlen(objdir) were PATH_MAX. */
959 char path[PATH_MAX + 1 + 4 + 1 + 1];
1f688557
JH
960 int len;
961 DIR *dir;
962 struct dirent *de;
963
964 sprintf(path, "%s/pack", objdir);
965 len = strlen(path);
966 dir = opendir(path);
b5b16990 967 if (!dir) {
26125f6b 968 if (errno != ENOENT)
bd2afde8 969 error("unable to open object pack directory: %s: %s",
26125f6b 970 path, strerror(errno));
1f688557 971 return;
b5b16990 972 }
1f688557
JH
973 path[len++] = '/';
974 while ((de = readdir(dir)) != NULL) {
975 int namelen = strlen(de->d_name);
976 struct packed_git *p;
977
5bb1cda5 978 if (!has_extension(de->d_name, ".idx"))
1f688557
JH
979 continue;
980
9cb18f56
JM
981 if (len + namelen + 1 > sizeof(path))
982 continue;
983
54a15a8d 984 /* Don't reopen a pack we already have. */
1f688557 985 strcpy(path + len, de->d_name);
86f7780c
JK
986 for (p = packed_git; p; p = p->next) {
987 if (!memcmp(path, p->pack_name, len + namelen - 4))
988 break;
989 }
990 if (p)
991 continue;
54a15a8d
SP
992 /* See if it really is a valid .idx file with corresponding
993 * .pack file that we can map.
994 */
9d835df2 995 p = add_packed_git(path, len + namelen, local);
1f688557
JH
996 if (!p)
997 continue;
625e9421 998 install_packed_git(p);
1f688557 999 }
5b35bcd5 1000 closedir(dir);
1f688557
JH
1001}
1002
b867092f
JH
1003static int sort_pack(const void *a_, const void *b_)
1004{
1005 struct packed_git *a = *((struct packed_git **)a_);
1006 struct packed_git *b = *((struct packed_git **)b_);
1007 int st;
1008
1009 /*
1010 * Local packs tend to contain objects specific to our
1011 * variant of the project than remote ones. In addition,
1012 * remote ones could be on a network mounted filesystem.
1013 * Favor local ones for these reasons.
1014 */
1015 st = a->pack_local - b->pack_local;
1016 if (st)
1017 return -st;
1018
1019 /*
1020 * Younger packs tend to contain more recent objects,
1021 * and more recent objects tend to get accessed more
1022 * often.
1023 */
1024 if (a->mtime < b->mtime)
1025 return 1;
1026 else if (a->mtime == b->mtime)
1027 return 0;
1028 return -1;
1029}
1030
1031static void rearrange_packed_git(void)
1032{
1033 struct packed_git **ary, *p;
1034 int i, n;
1035
1036 for (n = 0, p = packed_git; p; p = p->next)
1037 n++;
1038 if (n < 2)
1039 return;
1040
1041 /* prepare an array of packed_git for easier sorting */
1042 ary = xcalloc(n, sizeof(struct packed_git *));
1043 for (n = 0, p = packed_git; p; p = p->next)
1044 ary[n++] = p;
1045
1046 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1047
1048 /* link them back again */
1049 for (i = 0; i < n - 1; i++)
1050 ary[i]->next = ary[i + 1];
1051 ary[n - 1]->next = NULL;
1052 packed_git = ary[0];
1053
1054 free(ary);
1055}
1056
637cdd9d 1057static int prepare_packed_git_run_once = 0;
9a217f2a 1058void prepare_packed_git(void)
1f688557 1059{
d5a63b99 1060 struct alternate_object_database *alt;
1f688557 1061
637cdd9d 1062 if (prepare_packed_git_run_once)
1f688557 1063 return;
9d835df2 1064 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 1065 prepare_alt_odb();
d5a63b99 1066 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 1067 alt->name[-1] = 0;
9d835df2 1068 prepare_packed_git_one(alt->base, 0);
1494e038 1069 alt->name[-1] = '/';
1f688557 1070 }
b867092f 1071 rearrange_packed_git();
637cdd9d
JK
1072 prepare_packed_git_run_once = 1;
1073}
1074
fc04c412 1075void reprepare_packed_git(void)
637cdd9d 1076{
4b480c67 1077 discard_revindex();
637cdd9d
JK
1078 prepare_packed_git_run_once = 0;
1079 prepare_packed_git();
1f688557
JH
1080}
1081
8eca0b47
NP
1082static void mark_bad_packed_object(struct packed_git *p,
1083 const unsigned char *sha1)
1084{
1085 unsigned i;
1086 for (i = 0; i < p->num_bad_objects; i++)
1087 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1088 return;
1089 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1090 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1091 p->num_bad_objects++;
1092}
1093
b6c4cecc 1094static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
ac939109
NP
1095{
1096 struct packed_git *p;
1097 unsigned i;
1098
1099 for (p = packed_git; p; p = p->next)
1100 for (i = 0; i < p->num_bad_objects; i++)
1101 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
b6c4cecc
JH
1102 return p;
1103 return NULL;
ac939109
NP
1104}
1105
5d6ccf5c 1106int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
0fcfd160
LT
1107{
1108 unsigned char real_sha1[20];
7cfb5f36 1109 hash_sha1_file(map, size, type, real_sha1);
a89fccd2 1110 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1111}
1112
f2e872aa 1113static int git_open_noatime(const char *name, struct packed_git *p)
44d1c19e
LT
1114{
1115 static int sha1_file_open_flag = O_NOATIME;
44d1c19e 1116
f2e872aa
SP
1117 for (;;) {
1118 int fd = open(name, O_RDONLY | sha1_file_open_flag);
44d1c19e 1119 if (fd >= 0)
f2e872aa
SP
1120 return fd;
1121
f2e872aa
SP
1122 /* Might the failure be due to O_NOATIME? */
1123 if (errno != ENOENT && sha1_file_open_flag) {
44d1c19e 1124 sha1_file_open_flag = 0;
f2e872aa
SP
1125 continue;
1126 }
1127
1128 return -1;
44d1c19e 1129 }
44d1c19e
LT
1130}
1131
1132static int open_sha1_file(const unsigned char *sha1)
1133{
1134 int fd;
1135 char *name = sha1_file_name(sha1);
1136 struct alternate_object_database *alt;
1137
f2e872aa 1138 fd = git_open_noatime(name, NULL);
44d1c19e
LT
1139 if (fd >= 0)
1140 return fd;
1141
1142 prepare_alt_odb();
1143 errno = ENOENT;
1144 for (alt = alt_odb_list; alt; alt = alt->next) {
1145 name = alt->name;
1146 fill_sha1_path(name, sha1);
f2e872aa 1147 fd = git_open_noatime(alt->base, NULL);
44d1c19e
LT
1148 if (fd >= 0)
1149 return fd;
1150 }
1151 return -1;
1152}
1153
4175e9e3 1154static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1155{
0fcfd160 1156 void *map;
144bde78 1157 int fd;
ace1534d 1158
44d1c19e
LT
1159 fd = open_sha1_file(sha1);
1160 map = NULL;
1161 if (fd >= 0) {
1162 struct stat st;
0fcfd160 1163
44d1c19e
LT
1164 if (!fstat(fd, &st)) {
1165 *size = xsize_t(st.st_size);
1166 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1167 }
44d1c19e 1168 close(fd);
0fcfd160 1169 }
0fcfd160
LT
1170 return map;
1171}
1172
726f852b 1173static int legacy_loose_object(unsigned char *map)
bb6b8e4f
JH
1174{
1175 unsigned int word;
1176
1177 /*
1178 * Is it a zlib-compressed buffer? If so, the first byte
1179 * must be 0x78 (15-bit window size, deflated), and the
1180 * first 16-bit word is evenly divisible by 31
1181 */
1182 word = (map[0] << 8) + map[1];
1183 if (map[0] == 0x78 && !(word % 31))
1184 return 1;
1185 else
1186 return 0;
1187}
1188
09ded04b
NP
1189unsigned long unpack_object_header_buffer(const unsigned char *buf,
1190 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1191{
ad1ed5ee 1192 unsigned shift;
48fb7deb 1193 unsigned long size, c;
ad1ed5ee
JH
1194 unsigned long used = 0;
1195
1196 c = buf[used++];
1197 *type = (c >> 4) & 7;
1198 size = c & 15;
1199 shift = 4;
1200 while (c & 0x80) {
f630cfda 1201 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1202 error("bad object header");
ad1ed5ee 1203 return 0;
09ded04b 1204 }
ad1ed5ee
JH
1205 c = buf[used++];
1206 size += (c & 0x7f) << shift;
1207 shift += 7;
1208 }
1209 *sizep = size;
1210 return used;
1211}
1212
1213static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1214{
1215 unsigned long size, used;
1216 static const char valid_loose_object_type[8] = {
1217 0, /* OBJ_EXT */
1218 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1219 0, /* "delta" and others are invalid in a loose object */
93821bd9 1220 };
ad1ed5ee 1221 enum object_type type;
93821bd9 1222
c4483576
LT
1223 /* Get the data stream */
1224 memset(stream, 0, sizeof(*stream));
1225 stream->next_in = map;
1226 stream->avail_in = mapsize;
1227 stream->next_out = buffer;
93821bd9
LT
1228 stream->avail_out = bufsiz;
1229
bb6b8e4f 1230 if (legacy_loose_object(map)) {
39c68542
LT
1231 git_inflate_init(stream);
1232 return git_inflate(stream, 0);
93821bd9
LT
1233 }
1234
726f852b
NP
1235
1236 /*
1237 * There used to be a second loose object header format which
1238 * was meant to mimic the in-pack format, allowing for direct
1239 * copy of the object data. This format turned up not to be
1240 * really worth it and we don't write it any longer. But we
1241 * can still read it.
1242 */
09ded04b 1243 used = unpack_object_header_buffer(map, mapsize, &type, &size);
ad1ed5ee 1244 if (!used || !valid_loose_object_type[type])
93821bd9 1245 return -1;
ad1ed5ee
JH
1246 map += used;
1247 mapsize -= used;
93821bd9
LT
1248
1249 /* Set up the stream for the rest.. */
1250 stream->next_in = map;
1251 stream->avail_in = mapsize;
39c68542 1252 git_inflate_init(stream);
93821bd9
LT
1253
1254 /* And generate the fake traditional header */
ad1ed5ee 1255 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
df843662 1256 typename(type), size);
93821bd9 1257 return 0;
c4483576
LT
1258}
1259
7efbff75 1260static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1261{
1262 int bytes = strlen(buffer) + 1;
3aee68aa 1263 unsigned char *buf = xmallocz(size);
93821bd9 1264 unsigned long n;
7efbff75 1265 int status = Z_OK;
5180cacc 1266
93821bd9
LT
1267 n = stream->total_out - bytes;
1268 if (n > size)
1269 n = size;
1270 memcpy(buf, (char *) buffer + bytes, n);
1271 bytes = n;
456cdf6e
LT
1272 if (bytes <= size) {
1273 /*
1274 * The above condition must be (bytes <= size), not
1275 * (bytes < size). In other words, even though we
1276 * expect no more output and set avail_out to zer0,
1277 * the input zlib stream may have bytes that express
1278 * "this concludes the stream", and we *do* want to
1279 * eat that input.
1280 *
1281 * Otherwise we would not be able to test that we
1282 * consumed all the input to reach the expected size;
1283 * we also want to check that zlib tells us that all
1284 * went well with status == Z_STREAM_END at the end.
1285 */
5180cacc
LT
1286 stream->next_out = buf + bytes;
1287 stream->avail_out = size - bytes;
7efbff75 1288 while (status == Z_OK)
39c68542 1289 status = git_inflate(stream, Z_FINISH);
5180cacc 1290 }
456cdf6e 1291 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1292 git_inflate_end(stream);
7efbff75
JH
1293 return buf;
1294 }
1295
1296 if (status < 0)
1297 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1298 else if (stream->avail_in)
1299 error("garbage at end of loose object '%s'",
1300 sha1_to_hex(sha1));
1301 free(buf);
1302 return NULL;
5180cacc
LT
1303}
1304
1305/*
1306 * We used to just use "sscanf()", but that's actually way
1307 * too permissive for what we want to check. So do an anal
1308 * object header parse by hand.
1309 */
21666f1a 1310static int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1311{
21666f1a 1312 char type[10];
5180cacc
LT
1313 int i;
1314 unsigned long size;
1315
1316 /*
a6080a0a 1317 * The type can be at most ten bytes (including the
5180cacc 1318 * terminating '\0' that we add), and is followed by
21666f1a 1319 * a space.
5180cacc 1320 */
21666f1a 1321 i = 0;
5180cacc
LT
1322 for (;;) {
1323 char c = *hdr++;
1324 if (c == ' ')
1325 break;
21666f1a
NP
1326 type[i++] = c;
1327 if (i >= sizeof(type))
5180cacc 1328 return -1;
5180cacc 1329 }
21666f1a 1330 type[i] = 0;
5180cacc
LT
1331
1332 /*
1333 * The length must follow immediately, and be in canonical
1334 * decimal format (ie "010" is not valid).
1335 */
1336 size = *hdr++ - '0';
1337 if (size > 9)
1338 return -1;
1339 if (size) {
1340 for (;;) {
1341 unsigned long c = *hdr - '0';
1342 if (c > 9)
1343 break;
1344 hdr++;
1345 size = size * 10 + c;
1346 }
1347 }
1348 *sizep = size;
1349
1350 /*
1351 * The length must be followed by a zero byte
1352 */
21666f1a 1353 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1354}
1355
7efbff75 1356static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1357{
5180cacc 1358 int ret;
0fcfd160 1359 z_stream stream;
5180cacc 1360 char hdr[8192];
0fcfd160 1361
5180cacc 1362 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1363 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1364 return NULL;
1365
7efbff75 1366 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1367}
1368
54dab52a
NP
1369unsigned long get_size_from_delta(struct packed_git *p,
1370 struct pack_window **w_curs,
1371 off_t curpos)
1372{
1373 const unsigned char *data;
1374 unsigned char delta_head[20], *in;
1375 z_stream stream;
1376 int st;
1377
1378 memset(&stream, 0, sizeof(stream));
1379 stream.next_out = delta_head;
1380 stream.avail_out = sizeof(delta_head);
1381
39c68542 1382 git_inflate_init(&stream);
54dab52a
NP
1383 do {
1384 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1385 stream.next_in = in;
39c68542 1386 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1387 curpos += stream.next_in - in;
1388 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1389 stream.total_out < sizeof(delta_head));
39c68542 1390 git_inflate_end(&stream);
3d77d877
NP
1391 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1392 error("delta data unpack-initial failed");
1393 return 0;
1394 }
54dab52a
NP
1395
1396 /* Examine the initial part of the delta to figure out
1397 * the result size.
1398 */
1399 data = delta_head;
1400
1401 /* ignore base size */
1402 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1403
1404 /* Read the result size */
1405 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1406}
1407
c4001d92 1408static off_t get_delta_base(struct packed_git *p,
03e79c88 1409 struct pack_window **w_curs,
c4001d92 1410 off_t *curpos,
21666f1a 1411 enum object_type type,
c4001d92 1412 off_t delta_obj_offset)
eb32d236 1413{
2b87c45b 1414 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1415 off_t base_offset;
eb32d236 1416
8d8a4ea5
SP
1417 /* use_pack() assured us we have [base_info, base_info + 20)
1418 * as a range that we can look at without walking off the
1419 * end of the mapped window. Its actually the hash size
1420 * that is assured. An OFS_DELTA longer than the hash size
1421 * is stupid, as then a REF_DELTA would be smaller to store.
1422 */
21666f1a 1423 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1424 unsigned used = 0;
1425 unsigned char c = base_info[used++];
1426 base_offset = c & 127;
1427 while (c & 128) {
1428 base_offset += 1;
8723f216 1429 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1430 return 0; /* overflow */
eb32d236
NP
1431 c = base_info[used++];
1432 base_offset = (base_offset << 7) + (c & 127);
1433 }
1434 base_offset = delta_obj_offset - base_offset;
d8f32556 1435 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1436 return 0; /* out of bound */
2b87c45b 1437 *curpos += used;
21666f1a 1438 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1439 /* The base entry _must_ be in the same pack */
1440 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1441 *curpos += 20;
eb32d236
NP
1442 } else
1443 die("I am totally screwed");
2b87c45b 1444 return base_offset;
eb32d236
NP
1445}
1446
f3bf9224 1447/* forward declaration for a mutually recursive function */
c4001d92 1448static int packed_object_info(struct packed_git *p, off_t offset,
21666f1a 1449 unsigned long *sizep);
f3bf9224 1450
43057304 1451static int packed_delta_info(struct packed_git *p,
03e79c88 1452 struct pack_window **w_curs,
c4001d92 1453 off_t curpos,
21666f1a 1454 enum object_type type,
c4001d92 1455 off_t obj_offset,
43057304 1456 unsigned long *sizep)
5db47c2b 1457{
c4001d92 1458 off_t base_offset;
f3bf9224 1459
21666f1a 1460 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
3d77d877
NP
1461 if (!base_offset)
1462 return OBJ_BAD;
21666f1a 1463 type = packed_object_info(p, base_offset, NULL);
3d77d877 1464 if (type <= OBJ_NONE) {
08698b1e
NP
1465 struct revindex_entry *revidx;
1466 const unsigned char *base_sha1;
1467 revidx = find_pack_revindex(p, base_offset);
1468 if (!revidx)
1469 return OBJ_BAD;
1470 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
3d77d877
NP
1471 mark_bad_packed_object(p, base_sha1);
1472 type = sha1_object_info(base_sha1, NULL);
1473 if (type <= OBJ_NONE)
1474 return OBJ_BAD;
1475 }
f3bf9224 1476
c62266f3
JH
1477 /* We choose to only get the type of the base object and
1478 * ignore potentially corrupt pack file that expects the delta
1479 * based on a base with a wrong size. This saves tons of
1480 * inflate() calls.
1481 */
3d77d877 1482 if (sizep) {
54dab52a 1483 *sizep = get_size_from_delta(p, w_curs, curpos);
3d77d877
NP
1484 if (*sizep == 0)
1485 type = OBJ_BAD;
1486 }
21666f1a
NP
1487
1488 return type;
5db47c2b
JH
1489}
1490
2b87c45b
NP
1491static int unpack_object_header(struct packed_git *p,
1492 struct pack_window **w_curs,
c4001d92 1493 off_t *curpos,
2b87c45b 1494 unsigned long *sizep)
a733cb60 1495{
03e79c88
SP
1496 unsigned char *base;
1497 unsigned int left;
ad1ed5ee 1498 unsigned long used;
2b87c45b 1499 enum object_type type;
a733cb60 1500
8d8a4ea5
SP
1501 /* use_pack() assures us we have [base, base + 20) available
1502 * as a range that we can look at at. (Its actually the hash
3dff5379 1503 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1504 * the maximum deflated object size is 2^137, which is just
1505 * insane, so we know won't exceed what we have been given.
1506 */
2b87c45b 1507 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1508 used = unpack_object_header_buffer(base, left, &type, sizep);
1509 if (!used) {
1510 type = OBJ_BAD;
1511 } else
1512 *curpos += used;
ad1ed5ee 1513
2b87c45b 1514 return type;
a733cb60
LT
1515}
1516
21666f1a 1517const char *packed_object_info_detail(struct packed_git *p,
c4001d92 1518 off_t obj_offset,
21666f1a
NP
1519 unsigned long *size,
1520 unsigned long *store_size,
1521 unsigned int *delta_chain_length,
1522 unsigned char *base_sha1)
ad8c80a5 1523{
03e79c88 1524 struct pack_window *w_curs = NULL;
c4001d92
SP
1525 off_t curpos;
1526 unsigned long dummy;
43057304 1527 unsigned char *next_sha1;
21666f1a 1528 enum object_type type;
70f5d5d3 1529 struct revindex_entry *revidx;
ad8c80a5 1530
43057304 1531 *delta_chain_length = 0;
2b87c45b 1532 curpos = obj_offset;
21666f1a 1533 type = unpack_object_header(p, &w_curs, &curpos, size);
43057304 1534
70f5d5d3
NP
1535 revidx = find_pack_revindex(p, obj_offset);
1536 *store_size = revidx[1].offset - obj_offset;
1537
43057304 1538 for (;;) {
21666f1a 1539 switch (type) {
43057304 1540 default:
08a19d87 1541 die("pack %s contains unknown object type %d",
21666f1a 1542 p->pack_name, type);
43057304
NP
1543 case OBJ_COMMIT:
1544 case OBJ_TREE:
1545 case OBJ_BLOB:
1546 case OBJ_TAG:
03e79c88 1547 unuse_pack(&w_curs);
21666f1a 1548 return typename(type);
eb32d236 1549 case OBJ_OFS_DELTA:
21666f1a 1550 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
8eca0b47
NP
1551 if (!obj_offset)
1552 die("pack %s contains bad delta base reference of type %s",
1553 p->pack_name, typename(type));
eb32d236 1554 if (*delta_chain_length == 0) {
70f5d5d3
NP
1555 revidx = find_pack_revindex(p, obj_offset);
1556 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
eb32d236
NP
1557 }
1558 break;
1559 case OBJ_REF_DELTA:
2b87c45b 1560 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
43057304
NP
1561 if (*delta_chain_length == 0)
1562 hashcpy(base_sha1, next_sha1);
2b87c45b 1563 obj_offset = find_pack_entry_one(next_sha1, p);
43057304
NP
1564 break;
1565 }
43057304 1566 (*delta_chain_length)++;
2b87c45b 1567 curpos = obj_offset;
21666f1a 1568 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
ad8c80a5 1569 }
ad8c80a5
JH
1570}
1571
c4001d92 1572static int packed_object_info(struct packed_git *p, off_t obj_offset,
21666f1a 1573 unsigned long *sizep)
1f688557 1574{
03e79c88 1575 struct pack_window *w_curs = NULL;
c4001d92
SP
1576 unsigned long size;
1577 off_t curpos = obj_offset;
21666f1a 1578 enum object_type type;
5db47c2b 1579
21666f1a 1580 type = unpack_object_header(p, &w_curs, &curpos, &size);
5db47c2b 1581
21666f1a 1582 switch (type) {
eb32d236
NP
1583 case OBJ_OFS_DELTA:
1584 case OBJ_REF_DELTA:
21666f1a
NP
1585 type = packed_delta_info(p, &w_curs, curpos,
1586 type, obj_offset, sizep);
2b87c45b 1587 break;
a733cb60 1588 case OBJ_COMMIT:
a733cb60 1589 case OBJ_TREE:
a733cb60 1590 case OBJ_BLOB:
a733cb60 1591 case OBJ_TAG:
2b87c45b
NP
1592 if (sizep)
1593 *sizep = size;
a69d0943 1594 break;
1f688557 1595 default:
3d77d877
NP
1596 error("unknown object type %i at offset %"PRIuMAX" in %s",
1597 type, (uintmax_t)obj_offset, p->pack_name);
1598 type = OBJ_BAD;
1f688557 1599 }
2b87c45b 1600 unuse_pack(&w_curs);
21666f1a 1601 return type;
1f688557
JH
1602}
1603
eb950c19 1604static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1605 struct pack_window **w_curs,
c4001d92 1606 off_t curpos,
eb950c19 1607 unsigned long size)
de530aaa
SP
1608{
1609 int st;
1610 z_stream stream;
079afb18 1611 unsigned char *buffer, *in;
de530aaa 1612
4ab07e4d 1613 buffer = xmallocz(size);
de530aaa 1614 memset(&stream, 0, sizeof(stream));
de530aaa 1615 stream.next_out = buffer;
39eea7bd 1616 stream.avail_out = size + 1;
de530aaa 1617
39c68542 1618 git_inflate_init(&stream);
079afb18 1619 do {
2b87c45b 1620 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 1621 stream.next_in = in;
39c68542 1622 st = git_inflate(&stream, Z_FINISH);
39eea7bd
JH
1623 if (!stream.avail_out)
1624 break; /* the payload is larger than it should be */
2b87c45b 1625 curpos += stream.next_in - in;
079afb18 1626 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 1627 git_inflate_end(&stream);
de530aaa
SP
1628 if ((st != Z_STREAM_END) || stream.total_out != size) {
1629 free(buffer);
1630 return NULL;
1631 }
1632
1633 return buffer;
1634}
1635
e5e01619
LT
1636#define MAX_DELTA_CACHE (256)
1637
18bdec11 1638static size_t delta_base_cached;
5e08ecbf
NP
1639
1640static struct delta_base_cache_lru_list {
1641 struct delta_base_cache_lru_list *prev;
1642 struct delta_base_cache_lru_list *next;
1643} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1644
e5e01619 1645static struct delta_base_cache_entry {
5e08ecbf
NP
1646 struct delta_base_cache_lru_list lru;
1647 void *data;
e5e01619
LT
1648 struct packed_git *p;
1649 off_t base_offset;
1650 unsigned long size;
e5e01619
LT
1651 enum object_type type;
1652} delta_base_cache[MAX_DELTA_CACHE];
1653
1654static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1655{
1656 unsigned long hash;
1657
1658 hash = (unsigned long)p + (unsigned long)base_offset;
1659 hash += (hash >> 8) + (hash >> 16);
3358004a 1660 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1661}
1662
62f255ad 1663static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 1664 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 1665{
e5e01619
LT
1666 void *ret;
1667 unsigned long hash = pack_entry_hash(p, base_offset);
1668 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1669
1670 ret = ent->data;
749bc58c
MV
1671 if (!ret || ent->p != p || ent->base_offset != base_offset)
1672 return unpack_entry(p, base_offset, type, base_size);
e5e01619 1673
18bdec11 1674 if (!keep_cache) {
a0cba108 1675 ent->data = NULL;
5e08ecbf
NP
1676 ent->lru.next->prev = ent->lru.prev;
1677 ent->lru.prev->next = ent->lru.next;
18bdec11 1678 delta_base_cached -= ent->size;
182af834
PH
1679 } else {
1680 ret = xmemdupz(ent->data, ent->size);
a0cba108 1681 }
e5e01619
LT
1682 *type = ent->type;
1683 *base_size = ent->size;
1684 return ret;
62f255ad
LT
1685}
1686
18bdec11
SP
1687static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1688{
1689 if (ent->data) {
1690 free(ent->data);
1691 ent->data = NULL;
5e08ecbf
NP
1692 ent->lru.next->prev = ent->lru.prev;
1693 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
1694 delta_base_cached -= ent->size;
1695 }
1696}
1697
3d20c636
SP
1698void clear_delta_base_cache(void)
1699{
1700 unsigned long p;
1701 for (p = 0; p < MAX_DELTA_CACHE; p++)
1702 release_delta_base_cache(&delta_base_cache[p]);
1703}
1704
62f255ad
LT
1705static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1706 void *base, unsigned long base_size, enum object_type type)
1707{
5e08ecbf 1708 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 1709 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 1710 struct delta_base_cache_lru_list *lru;
e5e01619 1711
18bdec11
SP
1712 release_delta_base_cache(ent);
1713 delta_base_cached += base_size;
5e08ecbf
NP
1714
1715 for (lru = delta_base_cache_lru.next;
1716 delta_base_cached > delta_base_cache_limit
1717 && lru != &delta_base_cache_lru;
1718 lru = lru->next) {
1719 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
1720 if (f->type == OBJ_BLOB)
1721 release_delta_base_cache(f);
1722 }
5e08ecbf
NP
1723 for (lru = delta_base_cache_lru.next;
1724 delta_base_cached > delta_base_cache_limit
1725 && lru != &delta_base_cache_lru;
1726 lru = lru->next) {
1727 struct delta_base_cache_entry *f = (void *)lru;
1728 release_delta_base_cache(f);
1729 }
18bdec11 1730
e5e01619
LT
1731 ent->p = p;
1732 ent->base_offset = base_offset;
1733 ent->type = type;
1734 ent->data = base;
1735 ent->size = base_size;
5e08ecbf
NP
1736 ent->lru.next = &delta_base_cache_lru;
1737 ent->lru.prev = delta_base_cache_lru.prev;
1738 delta_base_cache_lru.prev->next = &ent->lru;
1739 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
1740}
1741
c2c5b270
CC
1742static void *read_object(const unsigned char *sha1, enum object_type *type,
1743 unsigned long *size);
1744
eb950c19 1745static void *unpack_delta_entry(struct packed_git *p,
03e79c88 1746 struct pack_window **w_curs,
c4001d92 1747 off_t curpos,
1f688557 1748 unsigned long delta_size,
c4001d92 1749 off_t obj_offset,
21666f1a 1750 enum object_type *type,
eb950c19 1751 unsigned long *sizep)
1f688557 1752{
7c3e8be3 1753 void *delta_data, *result, *base;
c4001d92
SP
1754 unsigned long base_size;
1755 off_t base_offset;
43057304 1756
21666f1a 1757 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
8eca0b47
NP
1758 if (!base_offset) {
1759 error("failed to validate delta base reference "
1760 "at offset %"PRIuMAX" from %s",
1761 (uintmax_t)curpos, p->pack_name);
1762 return NULL;
1763 }
eac12e2d 1764 unuse_pack(w_curs);
a0cba108 1765 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
8eca0b47
NP
1766 if (!base) {
1767 /*
1768 * We're probably in deep shit, but let's try to fetch
1769 * the required base anyway from another pack or loose.
1770 * This is costly but should happen only in the presence
1771 * of a corrupted pack, and is better than failing outright.
1772 */
08698b1e
NP
1773 struct revindex_entry *revidx;
1774 const unsigned char *base_sha1;
1775 revidx = find_pack_revindex(p, base_offset);
1776 if (!revidx)
1777 return NULL;
1778 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
8eca0b47
NP
1779 error("failed to read delta base object %s"
1780 " at offset %"PRIuMAX" from %s",
1781 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1782 p->pack_name);
1783 mark_bad_packed_object(p, base_sha1);
ac939109 1784 base = read_object(base_sha1, type, &base_size);
8eca0b47
NP
1785 if (!base)
1786 return NULL;
1787 }
67686d95 1788
2b87c45b 1789 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
8eca0b47
NP
1790 if (!delta_data) {
1791 error("failed to unpack compressed delta "
1792 "at offset %"PRIuMAX" from %s",
1793 (uintmax_t)curpos, p->pack_name);
1794 free(base);
1795 return NULL;
1796 }
1f688557
JH
1797 result = patch_delta(base, base_size,
1798 delta_data, delta_size,
21666f1a 1799 sizep);
1f688557
JH
1800 if (!result)
1801 die("failed to apply delta");
1802 free(delta_data);
62f255ad 1803 add_delta_base_cache(p, base_offset, base, base_size, *type);
1f688557
JH
1804 return result;
1805}
1806
0e8189e2
NP
1807int do_check_packed_object_crc;
1808
c4001d92 1809void *unpack_entry(struct packed_git *p, off_t obj_offset,
21666f1a 1810 enum object_type *type, unsigned long *sizep)
1f688557 1811{
03e79c88 1812 struct pack_window *w_curs = NULL;
c4001d92 1813 off_t curpos = obj_offset;
21666f1a 1814 void *data;
1f688557 1815
0e8189e2
NP
1816 if (do_check_packed_object_crc && p->index_version > 1) {
1817 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1818 unsigned long len = revidx[1].offset - obj_offset;
1819 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1820 const unsigned char *sha1 =
1821 nth_packed_object_sha1(p, revidx->nr);
1822 error("bad packed object CRC for %s",
1823 sha1_to_hex(sha1));
1824 mark_bad_packed_object(p, sha1);
f755bb99 1825 unuse_pack(&w_curs);
0e8189e2
NP
1826 return NULL;
1827 }
1828 }
1829
21666f1a
NP
1830 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1831 switch (*type) {
eb32d236
NP
1832 case OBJ_OFS_DELTA:
1833 case OBJ_REF_DELTA:
21666f1a
NP
1834 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1835 obj_offset, type, sizep);
4d703a1a 1836 break;
a733cb60 1837 case OBJ_COMMIT:
a733cb60 1838 case OBJ_TREE:
a733cb60 1839 case OBJ_BLOB:
a733cb60 1840 case OBJ_TAG:
21666f1a 1841 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
4d703a1a 1842 break;
1f688557 1843 default:
8eca0b47
NP
1844 data = NULL;
1845 error("unknown object type %i at offset %"PRIuMAX" in %s",
1846 *type, (uintmax_t)obj_offset, p->pack_name);
1f688557 1847 }
03e79c88 1848 unuse_pack(&w_curs);
21666f1a 1849 return data;
1f688557
JH
1850}
1851
d079837e 1852const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 1853 uint32_t n)
9a217f2a 1854{
42873078 1855 const unsigned char *index = p->index_data;
d079837e
SP
1856 if (!index) {
1857 if (open_pack_index(p))
1858 return NULL;
1859 index = p->index_data;
1860 }
57059091 1861 if (n >= p->num_objects)
d72308e0 1862 return NULL;
74e34e1f
NP
1863 index += 4 * 256;
1864 if (p->index_version == 1) {
1865 return index + 24 * n + 4;
1866 } else {
1867 index += 8;
1868 return index + 20 * n;
1869 }
1870}
1871
99093238 1872off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
1873{
1874 const unsigned char *index = p->index_data;
1875 index += 4 * 256;
1876 if (p->index_version == 1) {
1877 return ntohl(*((uint32_t *)(index + 24 * n)));
1878 } else {
1879 uint32_t off;
1880 index += 8 + p->num_objects * (20 + 4);
1881 off = ntohl(*((uint32_t *)(index + 4 * n)));
1882 if (!(off & 0x80000000))
1883 return off;
1884 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1885 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1886 ntohl(*((uint32_t *)(index + 4)));
1887 }
9a217f2a
JH
1888}
1889
c4001d92 1890off_t find_pack_entry_one(const unsigned char *sha1,
43057304 1891 struct packed_git *p)
1f688557 1892{
42873078 1893 const uint32_t *level1_ofs = p->index_data;
42873078 1894 const unsigned char *index = p->index_data;
628522ec
JH
1895 unsigned hi, lo, stride;
1896 static int use_lookup = -1;
1897 static int debug_lookup = -1;
1898
1899 if (debug_lookup < 0)
1900 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 1901
d079837e
SP
1902 if (!index) {
1903 if (open_pack_index(p))
1904 return 0;
1905 level1_ofs = p->index_data;
1906 index = p->index_data;
1907 }
74e34e1f
NP
1908 if (p->index_version > 1) {
1909 level1_ofs += 2;
1910 index += 8;
1911 }
42873078 1912 index += 4 * 256;
74e34e1f
NP
1913 hi = ntohl(level1_ofs[*sha1]);
1914 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
1915 if (p->index_version > 1) {
1916 stride = 20;
1917 } else {
1918 stride = 24;
1919 index += 4;
1920 }
1921
1922 if (debug_lookup)
6e1c2344 1923 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
1924 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1925
1926 if (use_lookup < 0)
1927 use_lookup = !!getenv("GIT_USE_LOOKUP");
1928 if (use_lookup) {
1929 int pos = sha1_entry_pos(index, stride, 0,
1930 lo, hi, p->num_objects, sha1);
1931 if (pos < 0)
1932 return 0;
1933 return nth_packed_object_offset(p, pos);
1934 }
1f688557
JH
1935
1936 do {
74e34e1f 1937 unsigned mi = (lo + hi) / 2;
628522ec
JH
1938 int cmp = hashcmp(index + mi * stride, sha1);
1939
1940 if (debug_lookup)
1941 printf("lo %u hi %u rg %u mi %u\n",
1942 lo, hi, hi - lo, mi);
43057304 1943 if (!cmp)
74e34e1f 1944 return nth_packed_object_offset(p, mi);
1f688557
JH
1945 if (cmp > 0)
1946 hi = mi;
1947 else
1948 lo = mi+1;
1949 } while (lo < hi);
1950 return 0;
1951}
1952
4d6acb70 1953static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557 1954{
f7c22cc6 1955 static struct packed_git *last_found = (void *)1;
1f688557 1956 struct packed_git *p;
c4001d92 1957 off_t offset;
43057304 1958
1f688557 1959 prepare_packed_git();
f7c22cc6
NP
1960 if (!packed_git)
1961 return 0;
1962 p = (last_found == (void *)1) ? packed_git : last_found;
1f688557 1963
f7c22cc6 1964 do {
8eca0b47
NP
1965 if (p->num_bad_objects) {
1966 unsigned i;
1967 for (i = 0; i < p->num_bad_objects; i++)
1968 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1969 goto next;
1970 }
1971
43057304
NP
1972 offset = find_pack_entry_one(sha1, p);
1973 if (offset) {
c715f783
SP
1974 /*
1975 * We are about to tell the caller where they can
1976 * locate the requested object. We better make
1977 * sure the packfile is still here and can be
1978 * accessed before supplying that answer, as
1979 * it may have been deleted since the index
1980 * was loaded!
1981 */
1982 if (p->pack_fd == -1 && open_packed_git(p)) {
1983 error("packfile %s cannot be accessed", p->pack_name);
f7c22cc6 1984 goto next;
c715f783 1985 }
43057304
NP
1986 e->offset = offset;
1987 e->p = p;
1988 hashcpy(e->sha1, sha1);
f7c22cc6 1989 last_found = p;
1f688557 1990 return 1;
43057304 1991 }
f7c22cc6
NP
1992
1993 next:
1994 if (p == last_found)
1995 p = packed_git;
1996 else
1997 p = p->next;
1998 if (p == last_found)
1999 p = p->next;
2000 } while (p);
1f688557
JH
2001 return 0;
2002}
2003
a6080a0a 2004struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
2005 struct packed_git *packs)
2006{
2007 struct packed_git *p;
bf592c50
DB
2008
2009 for (p = packs; p; p = p->next) {
43057304 2010 if (find_pack_entry_one(sha1, p))
bf592c50
DB
2011 return p;
2012 }
2013 return NULL;
21666f1a 2014
bf592c50
DB
2015}
2016
21666f1a 2017static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
65c2e0c3 2018{
36e4d74a 2019 int status;
65c2e0c3
JH
2020 unsigned long mapsize, size;
2021 void *map;
2022 z_stream stream;
d65a16f6 2023 char hdr[32];
65c2e0c3 2024
bb6b8e4f 2025 map = map_sha1_file(sha1, &mapsize);
f0df4ed5
JS
2026 if (!map)
2027 return error("unable to find %s", sha1_to_hex(sha1));
36e4d74a
JH
2028 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2029 status = error("unable to unpack %s header",
2030 sha1_to_hex(sha1));
21666f1a 2031 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 2032 status = error("unable to parse %s header", sha1_to_hex(sha1));
21666f1a
NP
2033 else if (sizep)
2034 *sizep = size;
39c68542 2035 git_inflate_end(&stream);
65c2e0c3
JH
2036 munmap(map, mapsize);
2037 return status;
2038}
2039
21666f1a 2040int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
f0df4ed5 2041{
c4d9986f 2042 struct cached_object *co;
f0df4ed5 2043 struct pack_entry e;
ddd63e64 2044 int status;
f0df4ed5 2045
c4d9986f
NTND
2046 co = find_cached_object(sha1);
2047 if (co) {
2048 if (sizep)
2049 *sizep = co->size;
2050 return co->type;
2051 }
2052
cd673c1f 2053 if (!find_pack_entry(sha1, &e)) {
ddd63e64
SG
2054 /* Most likely it's a loose object. */
2055 status = sha1_loose_object_info(sha1, sizep);
2056 if (status >= 0)
2057 return status;
2058
2059 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 2060 reprepare_packed_git();
cd673c1f 2061 if (!find_pack_entry(sha1, &e))
ddd63e64 2062 return status;
f0df4ed5 2063 }
3d77d877
NP
2064
2065 status = packed_object_info(e.p, e.offset, sizep);
2066 if (status < 0) {
2067 mark_bad_packed_object(e.p, sha1);
2068 status = sha1_object_info(sha1, sizep);
2069 }
2070
2071 return status;
f0df4ed5
JS
2072}
2073
21666f1a
NP
2074static void *read_packed_sha1(const unsigned char *sha1,
2075 enum object_type *type, unsigned long *size)
1f688557
JH
2076{
2077 struct pack_entry e;
8eca0b47 2078 void *data;
1f688557 2079
cd673c1f 2080 if (!find_pack_entry(sha1, &e))
1f688557 2081 return NULL;
8eca0b47
NP
2082 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2083 if (!data) {
2084 /*
2085 * We're probably in deep shit, but let's try to fetch
2086 * the required object anyway from another pack or loose.
2087 * This should happen only in the presence of a corrupted
2088 * pack, and is better than failing outright.
2089 */
2090 error("failed to read object %s at offset %"PRIuMAX" from %s",
2091 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2092 mark_bad_packed_object(e.p, sha1);
ac939109 2093 data = read_object(sha1, type, size);
8eca0b47
NP
2094 }
2095 return data;
1f688557
JH
2096}
2097
21666f1a
NP
2098int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2099 unsigned char *sha1)
d66b37bb
JH
2100{
2101 struct cached_object *co;
2102
21666f1a 2103 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2104 if (has_sha1_file(sha1) || find_cached_object(sha1))
2105 return 0;
2106 if (cached_object_alloc <= cached_object_nr) {
2107 cached_object_alloc = alloc_nr(cached_object_alloc);
2108 cached_objects = xrealloc(cached_objects,
2109 sizeof(*cached_objects) *
2110 cached_object_alloc);
2111 }
2112 co = &cached_objects[cached_object_nr++];
2113 co->size = len;
21666f1a 2114 co->type = type;
efa13f7b
JH
2115 co->buf = xmalloc(len);
2116 memcpy(co->buf, buf, len);
d66b37bb
JH
2117 hashcpy(co->sha1, sha1);
2118 return 0;
2119}
2120
c2c5b270
CC
2121static void *read_object(const unsigned char *sha1, enum object_type *type,
2122 unsigned long *size)
0fcfd160
LT
2123{
2124 unsigned long mapsize;
2125 void *map, *buf;
d66b37bb
JH
2126 struct cached_object *co;
2127
2128 co = find_cached_object(sha1);
2129 if (co) {
21666f1a 2130 *type = co->type;
d66b37bb 2131 *size = co->size;
182af834 2132 return xmemdupz(co->buf, co->size);
d66b37bb 2133 }
0fcfd160 2134
8276c007
PE
2135 buf = read_packed_sha1(sha1, type, size);
2136 if (buf)
2137 return buf;
bb6b8e4f 2138 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2139 if (map) {
7efbff75 2140 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2141 munmap(map, mapsize);
2142 return buf;
2143 }
637cdd9d 2144 reprepare_packed_git();
8276c007 2145 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2146}
2147
b6c4cecc
JH
2148/*
2149 * This function dies on corrupt objects; the callers who want to
2150 * deal with them should arrange to call read_object() and give error
2151 * messages themselves.
2152 */
f5552aee
CC
2153void *read_sha1_file_repl(const unsigned char *sha1,
2154 enum object_type *type,
2155 unsigned long *size,
2156 const unsigned char **replacement)
ac939109 2157{
68095570 2158 const unsigned char *repl = lookup_replace_object(sha1);
3ba7a065 2159 void *data;
e8b15e61 2160 char *path;
b6c4cecc
JH
2161 const struct packed_git *p;
2162
3ba7a065
JH
2163 errno = 0;
2164 data = read_object(repl, type, size);
b6c4cecc
JH
2165 if (data) {
2166 if (replacement)
2167 *replacement = repl;
2168 return data;
2169 }
68095570 2170
25f3af3f 2171 if (errno && errno != ENOENT)
3ba7a065
JH
2172 die_errno("failed to read object %s", sha1_to_hex(sha1));
2173
68095570 2174 /* die if we replaced an object with one that does not exist */
b6c4cecc 2175 if (repl != sha1)
68095570
CC
2176 die("replacement %s not found for %s",
2177 sha1_to_hex(repl), sha1_to_hex(sha1));
2178
b6c4cecc
JH
2179 if (has_loose_object(repl)) {
2180 path = sha1_file_name(sha1);
2181 die("loose object %s (stored in %s) is corrupt",
2182 sha1_to_hex(repl), path);
e8b15e61 2183 }
68095570 2184
b6c4cecc
JH
2185 if ((p = has_packed_and_bad(repl)) != NULL)
2186 die("packed object %s (stored in %s) is corrupt",
2187 sha1_to_hex(repl), p->pack_name);
f5552aee 2188
b6c4cecc 2189 return NULL;
ac939109
NP
2190}
2191
40469ee9 2192void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2193 const char *required_type_name,
40469ee9
JH
2194 unsigned long *size,
2195 unsigned char *actual_sha1_return)
f4913f91 2196{
21666f1a 2197 enum object_type type, required_type;
f4913f91
JH
2198 void *buffer;
2199 unsigned long isize;
40469ee9 2200 unsigned char actual_sha1[20];
f4913f91 2201
21666f1a 2202 required_type = type_from_string(required_type_name);
e702496e 2203 hashcpy(actual_sha1, sha1);
40469ee9
JH
2204 while (1) {
2205 int ref_length = -1;
2206 const char *ref_type = NULL;
f4913f91 2207
21666f1a 2208 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2209 if (!buffer)
2210 return NULL;
21666f1a 2211 if (type == required_type) {
40469ee9
JH
2212 *size = isize;
2213 if (actual_sha1_return)
e702496e 2214 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2215 return buffer;
2216 }
2217 /* Handle references */
21666f1a 2218 else if (type == OBJ_COMMIT)
40469ee9 2219 ref_type = "tree ";
21666f1a 2220 else if (type == OBJ_TAG)
40469ee9
JH
2221 ref_type = "object ";
2222 else {
2223 free(buffer);
2224 return NULL;
2225 }
2226 ref_length = strlen(ref_type);
f4913f91 2227
50974ec9
MK
2228 if (ref_length + 40 > isize ||
2229 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2230 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2231 free(buffer);
2232 return NULL;
2233 }
1cf58e72 2234 free(buffer);
40469ee9
JH
2235 /* Now we have the ID of the referred-to object in
2236 * actual_sha1. Check again. */
f4913f91 2237 }
f4913f91
JH
2238}
2239
ce9fbf16 2240static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2241 const char *type, unsigned char *sha1,
d65a16f6 2242 char *hdr, int *hdrlen)
d410c0f5 2243{
9126f009 2244 git_SHA_CTX c;
d410c0f5
JH
2245
2246 /* Generate the header */
d65a16f6 2247 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
2248
2249 /* Sha1.. */
9126f009
NP
2250 git_SHA1_Init(&c);
2251 git_SHA1_Update(&c, hdr, *hdrlen);
2252 git_SHA1_Update(&c, buf, len);
2253 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2254}
2255
230f1322 2256/*
5a688fe4
JH
2257 * Move the just written object into its final resting place.
2258 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2259 * "moving" is only a part of what it does, when no patch between
2260 * master to pu changes the call sites of this function.
230f1322 2261 */
839837b9 2262int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322 2263{
e32c0a9c 2264 int ret = 0;
5a688fe4 2265
348df166 2266 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
2267 goto try_rename;
2268 else if (link(tmpfile, filename))
e32c0a9c 2269 ret = errno;
7ebb6fca
LT
2270
2271 /*
2272 * Coda hack - coda doesn't like cross-directory links,
2273 * so we fall back to a rename, which will mean that it
2274 * won't be able to check collisions, but that's not a
2275 * big deal.
2276 *
2277 * The same holds for FAT formatted media.
2278 *
3be1f18e 2279 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
2280 * left to unlink.
2281 */
2282 if (ret && ret != EEXIST) {
be66a6c4 2283 try_rename:
7ebb6fca 2284 if (!rename(tmpfile, filename))
3be1f18e 2285 goto out;
9e48b389 2286 ret = errno;
230f1322 2287 }
691f1a28 2288 unlink_or_warn(tmpfile);
230f1322
LT
2289 if (ret) {
2290 if (ret != EEXIST) {
916d081b 2291 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
230f1322
LT
2292 }
2293 /* FIXME!!! Collision check here ? */
2294 }
2295
3be1f18e 2296out:
5256b006 2297 if (adjust_shared_perm(filename))
5a688fe4 2298 return error("unable to set permission to '%s'", filename);
230f1322
LT
2299 return 0;
2300}
2301
4d548150
LT
2302static int write_buffer(int fd, const void *buf, size_t len)
2303{
d34cf19b 2304 if (write_in_full(fd, buf, len) < 0)
93822c22 2305 return error("file write error (%s)", strerror(errno));
4d548150
LT
2306 return 0;
2307}
2308
ce9fbf16 2309int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2310 unsigned char *sha1)
2311{
d65a16f6 2312 char hdr[32];
abdc3fc8
RS
2313 int hdrlen;
2314 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2315 return 0;
2316}
2317
e9039dd3
LT
2318/* Finalize a file on disk, and close it. */
2319static void close_sha1_file(int fd)
2320{
aafe9fba
LT
2321 if (fsync_object_files)
2322 fsync_or_die(fd, "sha1 file");
e9039dd3 2323 if (close(fd) != 0)
d824cbba 2324 die_errno("error when closing sha1 file");
e9039dd3
LT
2325}
2326
5723fe7e
LT
2327/* Size of directory component, including the ending '/' */
2328static inline int directory_size(const char *filename)
2329{
2330 const char *s = strrchr(filename, '/');
2331 if (!s)
2332 return 0;
2333 return s - filename + 1;
2334}
2335
2336/*
2337 * This creates a temporary file in the same directory as the final
2338 * 'filename'
2339 *
2340 * We want to avoid cross-directory filename renames, because those
2341 * can have problems on various filesystems (FAT, NFS, Coda).
2342 */
2343static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2344{
2345 int fd, dirlen = directory_size(filename);
2346
2347 if (dirlen + 20 > bufsiz) {
2348 errno = ENAMETOOLONG;
2349 return -1;
2350 }
2351 memcpy(buffer, filename, dirlen);
2352 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
5256b006 2353 fd = git_mkstemp_mode(buffer, 0444);
cbacbf4e 2354 if (fd < 0 && dirlen && errno == ENOENT) {
5723fe7e 2355 /* Make sure the directory exists */
6ff6af62 2356 memcpy(buffer, filename, dirlen);
5723fe7e 2357 buffer[dirlen-1] = 0;
3bfaf018 2358 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
5723fe7e
LT
2359 return -1;
2360
2361 /* Try again */
2362 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
5256b006 2363 fd = git_mkstemp_mode(buffer, 0444);
5723fe7e
LT
2364 }
2365 return fd;
2366}
2367
bbac7311 2368static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 2369 const void *buf, unsigned long len, time_t mtime)
0fcfd160 2370{
915308b1 2371 int fd, ret;
9892beba 2372 unsigned char compressed[4096];
0fcfd160 2373 z_stream stream;
748af44c
NP
2374 git_SHA_CTX c;
2375 unsigned char parano_sha1[20];
706bc531 2376 char *filename;
aac17941 2377 static char tmpfile[PATH_MAX];
a44c9a5e 2378
972a9155 2379 filename = sha1_file_name(sha1);
5723fe7e 2380 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
aac17941 2381 if (fd < 0) {
35243577 2382 if (errno == EACCES)
916d081b
PB
2383 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2384 else
2385 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
aac17941
LT
2386 }
2387
0fcfd160
LT
2388 /* Set it up */
2389 memset(&stream, 0, sizeof(stream));
12f6c308 2390 deflateInit(&stream, zlib_compression_level);
0fcfd160 2391 stream.next_out = compressed;
9892beba 2392 stream.avail_out = sizeof(compressed);
748af44c 2393 git_SHA1_Init(&c);
a44c9a5e
LT
2394
2395 /* First header.. */
d65a16f6 2396 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2397 stream.avail_in = hdrlen;
726f852b
NP
2398 while (deflate(&stream, 0) == Z_OK)
2399 /* nothing */;
748af44c 2400 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
2401
2402 /* Then the data itself.. */
c00e657d 2403 stream.next_in = (void *)buf;
a44c9a5e 2404 stream.avail_in = len;
9892beba 2405 do {
748af44c 2406 unsigned char *in0 = stream.next_in;
9892beba 2407 ret = deflate(&stream, Z_FINISH);
748af44c 2408 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
2409 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2410 die("unable to write sha1 file");
2411 stream.next_out = compressed;
2412 stream.avail_out = sizeof(compressed);
2413 } while (ret == Z_OK);
2414
ac54c277
LT
2415 if (ret != Z_STREAM_END)
2416 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
ac54c277
LT
2417 ret = deflateEnd(&stream);
2418 if (ret != Z_OK)
2419 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
2420 git_SHA1_Final(parano_sha1, &c);
2421 if (hashcmp(sha1, parano_sha1) != 0)
2422 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 2423
e9039dd3 2424 close_sha1_file(fd);
0fcfd160 2425
bbac7311
NP
2426 if (mtime) {
2427 struct utimbuf utb;
2428 utb.actime = mtime;
2429 utb.modtime = mtime;
2430 if (utime(tmpfile, &utb) < 0)
2431 warning("failed utime() on %s: %s",
2432 tmpfile, strerror(errno));
2433 }
2434
230f1322 2435 return move_temp_to_file(tmpfile, filename);
0fcfd160 2436}
8237b185 2437
c00e657d 2438int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
bbac7311
NP
2439{
2440 unsigned char sha1[20];
2441 char hdr[32];
2442 int hdrlen;
2443
2444 /* Normally if we have it in the pack then we do not bother writing
2445 * it out into .git/objects/??/?{38} file.
2446 */
2447 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2448 if (returnsha1)
2449 hashcpy(returnsha1, sha1);
2450 if (has_sha1_file(sha1))
2451 return 0;
2452 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2453}
2454
2455int force_object_loose(const unsigned char *sha1, time_t mtime)
2456{
bbac7311
NP
2457 void *buf;
2458 unsigned long len;
2459 enum object_type type;
2460 char hdr[32];
2461 int hdrlen;
1fb23e65 2462 int ret;
bbac7311 2463
c529d75a 2464 if (has_loose_object(sha1))
bbac7311
NP
2465 return 0;
2466 buf = read_packed_sha1(sha1, &type, &len);
2467 if (!buf)
2468 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2469 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1fb23e65
BS
2470 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2471 free(buf);
2472
2473 return ret;
bbac7311
NP
2474}
2475
bf592c50
DB
2476int has_pack_index(const unsigned char *sha1)
2477{
2478 struct stat st;
2479 if (stat(sha1_pack_index_name(sha1), &st))
2480 return 0;
2481 return 1;
bf592c50
DB
2482}
2483
cd673c1f
JH
2484int has_sha1_pack(const unsigned char *sha1)
2485{
2486 struct pack_entry e;
2487 return find_pack_entry(sha1, &e);
2488}
2489
8237b185
DB
2490int has_sha1_file(const unsigned char *sha1)
2491{
1f688557
JH
2492 struct pack_entry e;
2493
cd673c1f 2494 if (find_pack_entry(sha1, &e))
1f688557 2495 return 1;
c529d75a 2496 return has_loose_object(sha1);
8237b185 2497}
74400e71 2498
43df4f86
DP
2499static int index_mem(unsigned char *sha1, void *buf, size_t size,
2500 int write_object, enum object_type type, const char *path)
e7332f96 2501{
6c510bee 2502 int ret, re_allocated = 0;
74400e71 2503
7672db20 2504 if (!type)
edaec3fb 2505 type = OBJ_BLOB;
6c510bee
LT
2506
2507 /*
2508 * Convert blobs to git internal format
2509 */
43df4f86 2510 if ((type == OBJ_BLOB) && path) {
f285a2d7 2511 struct strbuf nbuf = STRBUF_INIT;
21e5ad50
SP
2512 if (convert_to_git(path, buf, size, &nbuf,
2513 write_object ? safe_crlf : 0)) {
b315c5c0 2514 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
2515 re_allocated = 1;
2516 }
2517 }
2518
7672db20 2519 if (write_object)
edaec3fb 2520 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 2521 else
edaec3fb 2522 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 2523 if (re_allocated)
6c510bee 2524 free(buf);
43df4f86
DP
2525 return ret;
2526}
2527
ea68b0ce
DP
2528#define SMALL_FILE_SIZE (32*1024)
2529
43df4f86
DP
2530int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2531 enum object_type type, const char *path)
2532{
2533 int ret;
2534 size_t size = xsize_t(st->st_size);
2535
2536 if (!S_ISREG(st->st_mode)) {
f285a2d7 2537 struct strbuf sbuf = STRBUF_INIT;
43df4f86
DP
2538 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2539 ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2540 type, path);
2541 else
2542 ret = -1;
2543 strbuf_release(&sbuf);
08bda208
DP
2544 } else if (!size) {
2545 ret = index_mem(sha1, NULL, size, write_object, type, path);
ea68b0ce
DP
2546 } else if (size <= SMALL_FILE_SIZE) {
2547 char *buf = xmalloc(size);
2548 if (size == read_in_full(fd, buf, size))
2549 ret = index_mem(sha1, buf, size, write_object, type,
2550 path);
2551 else
2552 ret = error("short read %s", strerror(errno));
2553 free(buf);
08bda208 2554 } else {
43df4f86
DP
2555 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2556 ret = index_mem(sha1, buf, size, write_object, type, path);
aac17941 2557 munmap(buf, size);
08bda208 2558 }
43df4f86 2559 close(fd);
aac17941 2560 return ret;
74400e71 2561}
ec1fcc16
JH
2562
2563int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2564{
2565 int fd;
b760d3aa 2566 struct strbuf sb = STRBUF_INIT;
ec1fcc16
JH
2567
2568 switch (st->st_mode & S_IFMT) {
2569 case S_IFREG:
2570 fd = open(path, O_RDONLY);
2571 if (fd < 0)
2572 return error("open(\"%s\"): %s", path,
2573 strerror(errno));
53bca91a 2574 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
ec1fcc16
JH
2575 return error("%s: failed to insert into database",
2576 path);
2577 break;
2578 case S_IFLNK:
b760d3aa 2579 if (strbuf_readlink(&sb, path, st->st_size)) {
ec1fcc16 2580 char *errstr = strerror(errno);
ec1fcc16
JH
2581 return error("readlink(\"%s\"): %s", path,
2582 errstr);
2583 }
abdc3fc8 2584 if (!write_object)
b760d3aa
LT
2585 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2586 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
ec1fcc16
JH
2587 return error("%s: failed to insert into database",
2588 path);
b760d3aa 2589 strbuf_release(&sb);
ec1fcc16 2590 break;
f35a6d3b
LT
2591 case S_IFDIR:
2592 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
2593 default:
2594 return error("%s: unsupported file type", path);
2595 }
2596 return 0;
2597}
a69e5429
JH
2598
2599int read_pack_header(int fd, struct pack_header *header)
2600{
c697ad14
HO
2601 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2602 /* "eof before pack header was fully read" */
2603 return PH_ERROR_EOF;
2604
a69e5429
JH
2605 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2606 /* "protocol error (pack signature mismatch detected)" */
2607 return PH_ERROR_PACK_SIGNATURE;
2608 if (!pack_version_ok(header->hdr_version))
2609 /* "protocol error (pack version unsupported)" */
2610 return PH_ERROR_PROTOCOL;
2611 return 0;
2612}
40d52ff7
JK
2613
2614void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2615{
2616 enum object_type type = sha1_object_info(sha1, NULL);
2617 if (type < 0)
2618 die("%s is not a valid object", sha1_to_hex(sha1));
2619 if (type != expect)
2620 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2621 typename(expect));
2622}