sha1_loose_object_info: make type lookup optional
[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"
6eac50d8 10#include "string-list.h"
1f688557 11#include "delta.h"
a733cb60 12#include "pack.h"
8e440259
PE
13#include "blob.h"
14#include "commit.h"
4dd1fbc7 15#include "run-command.h"
8e440259
PE
16#include "tag.h"
17#include "tree.h"
c879daa2 18#include "tree-walk.h"
f35a6d3b 19#include "refs.h"
70f5d5d3 20#include "pack-revindex.h"
628522ec 21#include "sha1-lookup.h"
568508e7 22#include "bulk-checkin.h"
090ea126 23#include "streaming.h"
543c5caa 24#include "dir.h"
0fcfd160 25
144bde78
LT
26#ifndef O_NOATIME
27#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
28#define O_NOATIME 01000000
29#else
30#define O_NOATIME 0
31#endif
32#endif
33
28bd70d8
JN
34#define SZ_FMT PRIuMAX
35static inline uintmax_t sz_fmt(size_t s) { return s; }
e05db0fd 36
96f1e58f 37const unsigned char null_sha1[20];
88cd621d 38
c597ba80
NTND
39/*
40 * This is meant to hold a *small* number of objects that you would
41 * want read_sha1_file() to be able to return, but yet you do not want
42 * to write them into the object store (e.g. a browse-only
43 * application).
44 */
45static struct cached_object {
46 unsigned char sha1[20];
47 enum object_type type;
48 void *buf;
49 unsigned long size;
50} *cached_objects;
51static int cached_object_nr, cached_object_alloc;
52
53static struct cached_object empty_tree = {
dab0d410 54 EMPTY_TREE_SHA1_BIN_LITERAL,
c597ba80
NTND
55 OBJ_TREE,
56 "",
57 0
58};
59
c01f51cc
NTND
60static struct packed_git *last_found_pack;
61
c597ba80
NTND
62static struct cached_object *find_cached_object(const unsigned char *sha1)
63{
64 int i;
65 struct cached_object *co = cached_objects;
66
67 for (i = 0; i < cached_object_nr; i++, co++) {
68 if (!hashcmp(co->sha1, sha1))
69 return co;
70 }
71 if (!hashcmp(sha1, empty_tree.sha1))
72 return &empty_tree;
73 return NULL;
74}
75
90a6464b
JH
76int mkdir_in_gitdir(const char *path)
77{
78 if (mkdir(path, 0777)) {
79 int saved_errno = errno;
80 struct stat st;
81 struct strbuf sb = STRBUF_INIT;
82
83 if (errno != EEXIST)
84 return -1;
85 /*
86 * Are we looking at a path in a symlinked worktree
87 * whose original repository does not yet have it?
88 * e.g. .git/rr-cache pointing at its original
89 * repository in which the user hasn't performed any
90 * conflict resolution yet?
91 */
92 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
93 strbuf_readlink(&sb, path, st.st_size) ||
94 !is_absolute_path(sb.buf) ||
95 mkdir(sb.buf, 0777)) {
96 strbuf_release(&sb);
97 errno = saved_errno;
98 return -1;
99 }
100 strbuf_release(&sb);
101 }
102 return adjust_shared_perm(path);
103}
104
b2cb9425
JH
105int safe_create_leading_directories(char *path)
106{
8385abfd 107 char *pos = path + offset_1st_component(path);
67d42212
JR
108 struct stat st;
109
b2cb9425
JH
110 while (pos) {
111 pos = strchr(pos, '/');
112 if (!pos)
113 break;
5f0bdf50
JH
114 while (*++pos == '/')
115 ;
116 if (!*pos)
117 break;
118 *--pos = '\0';
67d42212
JR
119 if (!stat(path, &st)) {
120 /* path exists */
121 if (!S_ISDIR(st.st_mode)) {
b2cb9425 122 *pos = '/';
67d42212 123 return -3;
b2cb9425 124 }
457f06d6 125 }
67d42212 126 else if (mkdir(path, 0777)) {
928734d9
SW
127 if (errno == EEXIST &&
128 !stat(path, &st) && S_ISDIR(st.st_mode)) {
129 ; /* somebody created it since we checked */
130 } else {
131 *pos = '/';
132 return -1;
133 }
67d42212 134 }
457f06d6
JS
135 else if (adjust_shared_perm(path)) {
136 *pos = '/';
137 return -2;
138 }
b2cb9425
JH
139 *pos++ = '/';
140 }
141 return 0;
142}
723c31fe 143
8e21d63b
JK
144int safe_create_leading_directories_const(const char *path)
145{
146 /* path points to cache entries, so xstrdup before messing with it */
147 char *buf = xstrdup(path);
148 int result = safe_create_leading_directories(buf);
149 free(buf);
150 return result;
151}
152
ace1534d
JH
153static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
154{
155 int i;
156 for (i = 0; i < 20; i++) {
157 static char hex[] = "0123456789abcdef";
158 unsigned int val = sha1[i];
159 char *pos = pathbuf + i*2 + (i > 0);
160 *pos++ = hex[val >> 4];
161 *pos = hex[val & 0xf];
162 }
163}
164
0fcfd160
LT
165/*
166 * NOTE! This returns a statically allocated buffer, so you have to be
790296fd 167 * careful about using it. Do an "xstrdup()" if you need to save the
0fcfd160 168 * filename.
ace1534d
JH
169 *
170 * Also note that this returns the location for creating. Reading
171 * SHA1 file can happen from any alternate directory listed in the
d19938ab 172 * DB_ENVIRONMENT environment variable if it is not found in
ace1534d 173 * the primary object database.
0fcfd160
LT
174 */
175char *sha1_file_name(const unsigned char *sha1)
176{
560fb6a1
JK
177 static char buf[PATH_MAX];
178 const char *objdir;
179 int len;
0fcfd160 180
560fb6a1
JK
181 objdir = get_object_directory();
182 len = strlen(objdir);
183
184 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
185 if (len + 43 > PATH_MAX)
186 die("insanely long object directory %s", objdir);
187 memcpy(buf, objdir, len);
188 buf[len] = '/';
189 buf[len+3] = '/';
190 buf[len+42] = '\0';
191 fill_sha1_path(buf + len + 1, sha1);
192 return buf;
0fcfd160
LT
193}
194
633f43e1 195static char *sha1_get_pack_name(const unsigned char *sha1,
6eec46bd 196 char **name, char **base, const char *which)
bf592c50
DB
197{
198 static const char hex[] = "0123456789abcdef";
633f43e1 199 char *buf;
bf592c50
DB
200 int i;
201
633f43e1 202 if (!*base) {
bf592c50
DB
203 const char *sha1_file_directory = get_object_directory();
204 int len = strlen(sha1_file_directory);
633f43e1 205 *base = xmalloc(len + 60);
6eec46bd
JH
206 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
207 sha1_file_directory, which);
633f43e1 208 *name = *base + len + 11;
bf592c50
DB
209 }
210
633f43e1 211 buf = *name;
bf592c50
DB
212
213 for (i = 0; i < 20; i++) {
214 unsigned int val = *sha1++;
215 *buf++ = hex[val >> 4];
216 *buf++ = hex[val & 0xf];
217 }
a6080a0a 218
633f43e1 219 return *base;
bf592c50
DB
220}
221
633f43e1 222char *sha1_pack_name(const unsigned char *sha1)
bf592c50 223{
633f43e1 224 static char *name, *base;
bf592c50 225
6eec46bd 226 return sha1_get_pack_name(sha1, &name, &base, "pack");
633f43e1 227}
bf592c50 228
633f43e1
HO
229char *sha1_pack_index_name(const unsigned char *sha1)
230{
231 static char *name, *base;
a6080a0a 232
6eec46bd 233 return sha1_get_pack_name(sha1, &name, &base, "idx");
bf592c50
DB
234}
235
d5a63b99
JH
236struct alternate_object_database *alt_odb_list;
237static struct alternate_object_database **alt_odb_tail;
ace1534d 238
f4e51683 239static int git_open_noatime(const char *name);
c2f493a4 240
ddd5d056
JH
241/*
242 * Prepare alternate object database registry.
d5a63b99
JH
243 *
244 * The variable alt_odb_list points at the list of struct
245 * alternate_object_database. The elements on this list come from
246 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
247 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
248 * whose contents is similar to that environment variable but can be
249 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
250 * contains "/the/directory/corresponding/to/.git/objects/...", while
251 * its name points just after the slash at the end of ".git/objects/"
252 * in the example above, and has enough space to hold 40-byte hex
253 * SHA1, an extra slash for the first level indirection, and the
254 * terminating NUL.
ddd5d056 255 */
6eac50d8 256static int link_alt_odb_entry(const char *entry, const char *relative_base, int depth)
ace1534d 257{
1494e038 258 const char *objdir = get_object_directory();
c2f493a4
MW
259 struct alternate_object_database *ent;
260 struct alternate_object_database *alt;
5bdf0a84
HW
261 int pfxlen, entlen;
262 struct strbuf pathbuf = STRBUF_INIT;
d5a63b99 263
85dadc38 264 if (!is_absolute_path(entry) && relative_base) {
5bdf0a84
HW
265 strbuf_addstr(&pathbuf, real_path(relative_base));
266 strbuf_addch(&pathbuf, '/');
c2f493a4 267 }
6eac50d8 268 strbuf_addstr(&pathbuf, entry);
c2f493a4 269
5bdf0a84
HW
270 normalize_path_copy(pathbuf.buf, pathbuf.buf);
271
272 pfxlen = strlen(pathbuf.buf);
273
274 /*
275 * The trailing slash after the directory name is given by
276 * this function at the end. Remove duplicates.
277 */
278 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
279 pfxlen -= 1;
280
281 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
282 ent = xmalloc(sizeof(*ent) + entlen);
283 memcpy(ent->base, pathbuf.buf, pfxlen);
284 strbuf_release(&pathbuf);
c2f493a4
MW
285
286 ent->name = ent->base + pfxlen + 1;
287 ent->base[pfxlen + 3] = '/';
288 ent->base[pfxlen] = ent->base[entlen-1] = 0;
289
290 /* Detect cases where alternate disappeared */
90b4a71c 291 if (!is_directory(ent->base)) {
c2f493a4
MW
292 error("object directory %s does not exist; "
293 "check .git/objects/info/alternates.",
294 ent->base);
295 free(ent);
296 return -1;
297 }
298
299 /* Prevent the common mistake of listing the same
300 * thing twice, or object directory itself.
301 */
302 for (alt = alt_odb_list; alt; alt = alt->next) {
303 if (!memcmp(ent->base, alt->base, pfxlen)) {
304 free(ent);
305 return -1;
306 }
307 }
cb2912c3 308 if (!strcmp(ent->base, objdir)) {
c2f493a4
MW
309 free(ent);
310 return -1;
311 }
312
313 /* add the alternate entry */
314 *alt_odb_tail = ent;
315 alt_odb_tail = &(ent->next);
316 ent->next = NULL;
317
318 /* recursively add alternates */
319 read_info_alternates(ent->base, depth + 1);
320
321 ent->base[pfxlen] = '/';
322
323 return 0;
324}
325
c5950164 326static void link_alt_odb_entries(const char *alt, int len, int sep,
c2f493a4
MW
327 const char *relative_base, int depth)
328{
6eac50d8
MH
329 struct string_list entries = STRING_LIST_INIT_NODUP;
330 char *alt_copy;
331 int i;
c2f493a4
MW
332
333 if (depth > 5) {
334 error("%s: ignoring alternate object stores, nesting too deep.",
335 relative_base);
336 return;
337 }
338
c5950164 339 alt_copy = xmemdupz(alt, len);
6eac50d8
MH
340 string_list_split_in_place(&entries, alt_copy, sep, -1);
341 for (i = 0; i < entries.nr; i++) {
342 const char *entry = entries.items[i].string;
343 if (entry[0] == '\0' || entry[0] == '#')
9577e7e3 344 continue;
6eac50d8
MH
345 if (!is_absolute_path(entry) && depth) {
346 error("%s: ignoring relative alternate object store %s",
347 relative_base, entry);
348 } else {
349 link_alt_odb_entry(entry, relative_base, depth);
9577e7e3 350 }
9577e7e3 351 }
6eac50d8
MH
352 string_list_clear(&entries, 0);
353 free(alt_copy);
d5a63b99
JH
354}
355
5e73633d 356void read_info_alternates(const char * relative_base, int depth)
d5a63b99 357{
9577e7e3 358 char *map;
dc49cd76 359 size_t mapsz;
d5a63b99 360 struct stat st;
9cb18f56
JM
361 const char alt_file_name[] = "info/alternates";
362 /* Given that relative_base is no longer than PATH_MAX,
363 ensure that "path" has enough space to append "/", the
364 file name, "info/alternates", and a trailing NUL. */
365 char path[PATH_MAX + 1 + sizeof alt_file_name];
c2f493a4 366 int fd;
d5a63b99 367
9cb18f56 368 sprintf(path, "%s/%s", relative_base, alt_file_name);
f4e51683 369 fd = git_open_noatime(path);
d5a63b99
JH
370 if (fd < 0)
371 return;
372 if (fstat(fd, &st) || (st.st_size == 0)) {
373 close(fd);
9a217f2a 374 return;
ace1534d 375 }
dc49cd76
SP
376 mapsz = xsize_t(st.st_size);
377 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 378 close(fd);
d5a63b99 379
c5950164 380 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
c2f493a4 381
dc49cd76 382 munmap(map, mapsz);
ace1534d
JH
383}
384
bef70b22
DB
385void add_to_alternates_file(const char *reference)
386{
387 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
acd3b9ec 388 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
e6baf4a1 389 char *alt = mkpath("%s\n", reference);
bef70b22
DB
390 write_or_die(fd, alt, strlen(alt));
391 if (commit_lock_file(lock))
392 die("could not close alternates file");
393 if (alt_odb_tail)
c5950164 394 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
bef70b22
DB
395}
396
d79796bc
JH
397void foreach_alt_odb(alt_odb_fn fn, void *cb)
398{
399 struct alternate_object_database *ent;
400
401 prepare_alt_odb();
402 for (ent = alt_odb_list; ent; ent = ent->next)
403 if (fn(ent, cb))
404 return;
405}
406
c2f493a4
MW
407void prepare_alt_odb(void)
408{
554fe20d 409 const char *alt;
c2f493a4 410
7dc24aa5
SP
411 if (alt_odb_tail)
412 return;
413
c2f493a4
MW
414 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
415 if (!alt) alt = "";
416
c2f493a4 417 alt_odb_tail = &alt_odb_list;
c5950164 418 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
419
420 read_info_alternates(get_object_directory(), 0);
421}
422
0f4dc14a 423static int has_loose_object_local(const unsigned char *sha1)
ace1534d 424{
ace1534d 425 char *name = sha1_file_name(sha1);
0f4dc14a
BC
426 return !access(name, F_OK);
427}
ace1534d 428
0f4dc14a
BC
429int has_loose_object_nonlocal(const unsigned char *sha1)
430{
431 struct alternate_object_database *alt;
9a217f2a 432 prepare_alt_odb();
d5a63b99 433 for (alt = alt_odb_list; alt; alt = alt->next) {
0f4dc14a 434 fill_sha1_path(alt->name, sha1);
c529d75a
LT
435 if (!access(alt->base, F_OK))
436 return 1;
ace1534d 437 }
c529d75a 438 return 0;
ace1534d
JH
439}
440
0f4dc14a
BC
441static int has_loose_object(const unsigned char *sha1)
442{
443 return has_loose_object_local(sha1) ||
444 has_loose_object_nonlocal(sha1);
445}
446
60bb8b14 447static unsigned int pack_used_ctr;
a53128b6
SP
448static unsigned int pack_mmap_calls;
449static unsigned int peak_pack_open_windows;
450static unsigned int pack_open_windows;
c7934306
SP
451static unsigned int pack_open_fds;
452static unsigned int pack_max_fds;
a53128b6 453static size_t peak_pack_mapped;
60bb8b14 454static size_t pack_mapped;
9a217f2a 455struct packed_git *packed_git;
1f688557 456
b79d18c9 457void pack_report(void)
a53128b6
SP
458{
459 fprintf(stderr,
e05db0fd
PR
460 "pack_report: getpagesize() = %10" SZ_FMT "\n"
461 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
462 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
463 sz_fmt(getpagesize()),
464 sz_fmt(packed_git_window_size),
465 sz_fmt(packed_git_limit));
a53128b6
SP
466 fprintf(stderr,
467 "pack_report: pack_used_ctr = %10u\n"
468 "pack_report: pack_mmap_calls = %10u\n"
469 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
470 "pack_report: pack_mapped = "
471 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
472 pack_used_ctr,
473 pack_mmap_calls,
474 pack_open_windows, peak_pack_open_windows,
9e42d6a1 475 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
476}
477
42873078 478static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
479{
480 void *idx_map;
42873078 481 struct pack_idx_header *hdr;
dc49cd76 482 size_t idx_size;
74e34e1f 483 uint32_t version, nr, i, *index;
f4e51683 484 int fd = git_open_noatime(path);
1f688557 485 struct stat st;
42873078 486
1f688557
JH
487 if (fd < 0)
488 return -1;
489 if (fstat(fd, &st)) {
490 close(fd);
491 return -1;
492 }
dc49cd76 493 idx_size = xsize_t(st.st_size);
2d88451b
SP
494 if (idx_size < 4 * 256 + 20 + 20) {
495 close(fd);
496 return error("index file %s is too small", path);
497 }
c4712e45 498 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 499 close(fd);
1f688557 500
42873078
NP
501 hdr = idx_map;
502 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
503 version = ntohl(hdr->idx_version);
504 if (version < 2 || version > 2) {
505 munmap(idx_map, idx_size);
6e1c2344 506 return error("index file %s is version %"PRIu32
74e34e1f
NP
507 " and is not supported by this binary"
508 " (try upgrading GIT to a newer version)",
509 path, version);
510 }
511 } else
512 version = 1;
df1b059d 513
1f688557 514 nr = 0;
42873078 515 index = idx_map;
74e34e1f
NP
516 if (version > 1)
517 index += 2; /* skip index header */
1f688557 518 for (i = 0; i < 256; i++) {
326bf396 519 uint32_t n = ntohl(index[i]);
2d88451b
SP
520 if (n < nr) {
521 munmap(idx_map, idx_size);
df1b059d 522 return error("non-monotonic index %s", path);
2d88451b 523 }
1f688557
JH
524 nr = n;
525 }
526
74e34e1f
NP
527 if (version == 1) {
528 /*
529 * Total size:
530 * - 256 index entries 4 bytes each
531 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
532 * - 20-byte SHA1 of the packfile
533 * - 20-byte SHA1 file checksum
534 */
535 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
536 munmap(idx_map, idx_size);
eef427a0 537 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
538 }
539 } else if (version == 2) {
540 /*
541 * Minimum size:
542 * - 8 bytes of header
543 * - 256 index entries 4 bytes each
544 * - 20-byte sha1 entry * nr
545 * - 4-byte crc entry * nr
546 * - 4-byte offset entry * nr
547 * - 20-byte SHA1 of the packfile
548 * - 20-byte SHA1 file checksum
549 * And after the 4-byte offset table might be a
550 * variable sized table containing 8-byte entries
551 * for offsets larger than 2^31.
552 */
553 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
554 unsigned long max_size = min_size;
555 if (nr)
556 max_size += (nr - 1)*8;
557 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 558 munmap(idx_map, idx_size);
eef427a0 559 return error("wrong index v2 file size in %s", path);
74e34e1f 560 }
7109c889
JH
561 if (idx_size != min_size &&
562 /*
563 * make sure we can deal with large pack offsets.
564 * 31-bit signed offset won't be enough, neither
565 * 32-bit unsigned one will be.
566 */
567 (sizeof(off_t) <= 4)) {
568 munmap(idx_map, idx_size);
569 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 570 }
2d88451b 571 }
1f688557 572
74e34e1f 573 p->index_version = version;
42873078
NP
574 p->index_data = idx_map;
575 p->index_size = idx_size;
57059091 576 p->num_objects = nr;
1f688557
JH
577 return 0;
578}
579
bc8e478a 580int open_pack_index(struct packed_git *p)
d079837e
SP
581{
582 char *idx_name;
583 int ret;
584
585 if (p->index_data)
586 return 0;
587
588 idx_name = xstrdup(p->pack_name);
589 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
590 ret = check_packed_git_idx(idx_name, p);
591 free(idx_name);
592 return ret;
593}
594
11daf39b
SP
595static void scan_windows(struct packed_git *p,
596 struct packed_git **lru_p,
597 struct pack_window **lru_w,
598 struct pack_window **lru_l)
1f688557 599{
11daf39b
SP
600 struct pack_window *w, *w_l;
601
602 for (w_l = NULL, w = p->windows; w; w = w->next) {
603 if (!w->inuse_cnt) {
604 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
605 *lru_p = p;
606 *lru_w = w;
607 *lru_l = w_l;
54044bf8 608 }
54044bf8 609 }
11daf39b 610 w_l = w;
f9253394 611 }
11daf39b
SP
612}
613
d1efefa4 614static int unuse_one_window(struct packed_git *current, int keep_fd)
11daf39b
SP
615{
616 struct packed_git *p, *lru_p = NULL;
617 struct pack_window *lru_w = NULL, *lru_l = NULL;
618
619 if (current)
620 scan_windows(current, &lru_p, &lru_w, &lru_l);
621 for (p = packed_git; p; p = p->next)
622 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
623 if (lru_p) {
624 munmap(lru_w->base, lru_w->len);
625 pack_mapped -= lru_w->len;
626 if (lru_l)
627 lru_l->next = lru_w->next;
628 else {
629 lru_p->windows = lru_w->next;
d131b7af
SP
630 if (!lru_p->windows && lru_p->pack_fd != -1
631 && lru_p->pack_fd != keep_fd) {
54044bf8 632 close(lru_p->pack_fd);
c7934306 633 pack_open_fds--;
54044bf8
SP
634 lru_p->pack_fd = -1;
635 }
636 }
637 free(lru_w);
a53128b6 638 pack_open_windows--;
54044bf8
SP
639 return 1;
640 }
641 return 0;
f9253394
JH
642}
643
d1efefa4 644void release_pack_memory(size_t need, int fd)
97bfeb34
SP
645{
646 size_t cur = pack_mapped;
d1efefa4 647 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
97bfeb34
SP
648 ; /* nothing */
649}
650
58ecbd5e
JN
651void *xmmap(void *start, size_t length,
652 int prot, int flags, int fd, off_t offset)
653{
654 void *ret = mmap(start, length, prot, flags, fd, offset);
655 if (ret == MAP_FAILED) {
656 if (!length)
657 return NULL;
658 release_pack_memory(length, fd);
659 ret = mmap(start, length, prot, flags, fd, offset);
660 if (ret == MAP_FAILED)
661 die_errno("Out of memory? mmap failed");
662 }
663 return ret;
664}
665
c9ced051
SP
666void close_pack_windows(struct packed_git *p)
667{
668 while (p->windows) {
669 struct pack_window *w = p->windows;
670
671 if (w->inuse_cnt)
672 die("pack '%s' still has open windows to it",
673 p->pack_name);
674 munmap(w->base, w->len);
675 pack_mapped -= w->len;
676 pack_open_windows--;
677 p->windows = w->next;
678 free(w);
679 }
680}
681
03e79c88 682void unuse_pack(struct pack_window **w_cursor)
f9253394 683{
03e79c88
SP
684 struct pack_window *w = *w_cursor;
685 if (w) {
686 w->inuse_cnt--;
687 *w_cursor = NULL;
688 }
1f688557
JH
689}
690
fa5fc15d
SP
691void close_pack_index(struct packed_git *p)
692{
693 if (p->index_data) {
694 munmap((void *)p->index_data, p->index_size);
695 p->index_data = NULL;
696 }
697}
698
c74faea1
NP
699/*
700 * This is used by git-repack in case a newly created pack happens to
701 * contain the same set of objects as an existing one. In that case
702 * the resulting file might be different even if its name would be the
703 * same. It is best to close any reference to the old pack before it is
704 * replaced on disk. Of course no index pointers nor windows for given pack
705 * must subsist at this point. If ever objects from this pack are requested
706 * again, the new version of the pack will be reinitialized through
707 * reprepare_packed_git().
708 */
709void free_pack_by_name(const char *pack_name)
710{
711 struct packed_git *p, **pp = &packed_git;
712
713 while (*pp) {
714 p = *pp;
715 if (strcmp(pack_name, p->pack_name) == 0) {
fa3a0c94 716 clear_delta_base_cache();
c74faea1 717 close_pack_windows(p);
c7934306 718 if (p->pack_fd != -1) {
c74faea1 719 close(p->pack_fd);
c7934306
SP
720 pack_open_fds--;
721 }
fa5fc15d 722 close_pack_index(p);
c74faea1
NP
723 free(p->bad_object_sha1);
724 *pp = p->next;
c01f51cc
NTND
725 if (last_found_pack == p)
726 last_found_pack = NULL;
c74faea1
NP
727 free(p);
728 return;
729 }
730 pp = &p->next;
731 }
732}
733
a0788266
JS
734static unsigned int get_max_fd_limit(void)
735{
736#ifdef RLIMIT_NOFILE
737 struct rlimit lim;
738
739 if (getrlimit(RLIMIT_NOFILE, &lim))
740 die_errno("cannot get RLIMIT_NOFILE");
741
742 return lim.rlim_cur;
743#elif defined(_SC_OPEN_MAX)
744 return sysconf(_SC_OPEN_MAX);
745#elif defined(OPEN_MAX)
746 return OPEN_MAX;
747#else
748 return 1; /* see the caller ;-) */
749#endif
750}
751
3cf8b462
SP
752/*
753 * Do not call this directly as this leaks p->pack_fd on error return;
754 * call open_packed_git() instead.
755 */
756static int open_packed_git_1(struct packed_git *p)
1f688557 757{
9bc879c1
SP
758 struct stat st;
759 struct pack_header hdr;
760 unsigned char sha1[20];
761 unsigned char *idx_sha1;
2c039da8 762 long fd_flag;
9bc879c1 763
d079837e
SP
764 if (!p->index_data && open_pack_index(p))
765 return error("packfile %s index unavailable", p->pack_name);
766
c7934306 767 if (!pack_max_fds) {
a0788266 768 unsigned int max_fds = get_max_fd_limit();
c7934306
SP
769
770 /* Save 3 for stdin/stdout/stderr, 22 for work */
771 if (25 < max_fds)
772 pack_max_fds = max_fds - 25;
773 else
774 pack_max_fds = 1;
775 }
776
777 while (pack_max_fds <= pack_open_fds && unuse_one_window(NULL, -1))
778 ; /* nothing */
779
f4e51683 780 p->pack_fd = git_open_noatime(p->pack_name);
9bc879c1 781 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 782 return -1;
c7934306 783 pack_open_fds++;
9bc879c1
SP
784
785 /* If we created the struct before we had the pack we lack size. */
bf592c50 786 if (!p->pack_size) {
bf592c50 787 if (!S_ISREG(st.st_mode))
072db278 788 return error("packfile %s not a regular file", p->pack_name);
bf592c50 789 p->pack_size = st.st_size;
9bc879c1 790 } else if (p->pack_size != st.st_size)
072db278 791 return error("packfile %s size changed", p->pack_name);
9bc879c1 792
2c039da8
JH
793 /* We leave these file descriptors open with sliding mmap;
794 * there is no point keeping them open across exec(), though.
795 */
796 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
797 if (fd_flag < 0)
072db278 798 return error("cannot determine file descriptor flags");
2c039da8
JH
799 fd_flag |= FD_CLOEXEC;
800 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 801 return error("cannot set FD_CLOEXEC");
2c039da8 802
9bc879c1 803 /* Verify we recognize this pack file format. */
e6e2bd62 804 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 805 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 806 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 807 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 808 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
809 return error("packfile %s is version %"PRIu32" and not"
810 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
811 p->pack_name, ntohl(hdr.hdr_version));
812
813 /* Verify the pack matches its index. */
57059091 814 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
815 return error("packfile %s claims to have %"PRIu32" objects"
816 " while index indicates %"PRIu32" objects",
57059091
NP
817 p->pack_name, ntohl(hdr.hdr_entries),
818 p->num_objects);
9bc879c1 819 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 820 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 821 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 822 return error("packfile %s signature is unavailable", p->pack_name);
42873078 823 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 824 if (hashcmp(sha1, idx_sha1))
072db278
SP
825 return error("packfile %s does not match index", p->pack_name);
826 return 0;
9bc879c1
SP
827}
828
3cf8b462
SP
829static int open_packed_git(struct packed_git *p)
830{
831 if (!open_packed_git_1(p))
832 return 0;
833 if (p->pack_fd != -1) {
834 close(p->pack_fd);
c7934306 835 pack_open_fds--;
3cf8b462
SP
836 p->pack_fd = -1;
837 }
838 return -1;
839}
840
c4001d92 841static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
842{
843 /* We must promise at least 20 bytes (one hash) after the
844 * offset is available from this window, otherwise the offset
845 * is not actually in this window and a different window (which
846 * has that one hash excess) must be used. This is to support
847 * the object header and delta base parsing routines below.
848 */
849 off_t win_off = win->offset;
850 return win_off <= offset
851 && (offset + 20) <= (win_off + win->len);
852}
853
4b25d091 854unsigned char *use_pack(struct packed_git *p,
03e79c88 855 struct pack_window **w_cursor,
c4001d92 856 off_t offset,
ef49a7a0 857 unsigned long *left)
9bc879c1 858{
60bb8b14 859 struct pack_window *win = *w_cursor;
03e79c88 860
a9d98a14 861 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
862 * pointless to ask for an offset into the middle of that
863 * hash, and the in_window function above wouldn't match
864 * don't allow an offset too close to the end of the file.
865 */
d131b7af
SP
866 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
867 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
868 if (offset > (p->pack_size - 20))
869 die("offset beyond end of packfile (truncated pack?)");
870
871 if (!win || !in_window(win, offset)) {
872 if (win)
873 win->inuse_cnt--;
874 for (win = p->windows; win; win = win->next) {
875 if (in_window(win, offset))
876 break;
877 }
878 if (!win) {
78a28df9 879 size_t window_align = packed_git_window_size / 2;
dc49cd76 880 off_t len;
d131b7af
SP
881
882 if (p->pack_fd == -1 && open_packed_git(p))
883 die("packfile %s cannot be accessed", p->pack_name);
884
60bb8b14 885 win = xcalloc(1, sizeof(*win));
78a28df9 886 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
887 len = p->pack_size - win->offset;
888 if (len > packed_git_window_size)
889 len = packed_git_window_size;
890 win->len = (size_t)len;
60bb8b14 891 pack_mapped += win->len;
11daf39b 892 while (packed_git_limit < pack_mapped
d1efefa4 893 && unuse_one_window(p, p->pack_fd))
60bb8b14 894 ; /* nothing */
c4712e45 895 win->base = xmmap(NULL, win->len,
60bb8b14
SP
896 PROT_READ, MAP_PRIVATE,
897 p->pack_fd, win->offset);
898 if (win->base == MAP_FAILED)
73b4e4be
SP
899 die("packfile %s cannot be mapped: %s",
900 p->pack_name,
901 strerror(errno));
d131b7af
SP
902 if (!win->offset && win->len == p->pack_size
903 && !p->do_not_close) {
904 close(p->pack_fd);
905 pack_open_fds--;
906 p->pack_fd = -1;
907 }
a53128b6
SP
908 pack_mmap_calls++;
909 pack_open_windows++;
910 if (pack_mapped > peak_pack_mapped)
911 peak_pack_mapped = pack_mapped;
912 if (pack_open_windows > peak_pack_open_windows)
913 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
914 win->next = p->windows;
915 p->windows = win;
916 }
1f688557 917 }
03e79c88
SP
918 if (win != *w_cursor) {
919 win->last_used = pack_used_ctr++;
920 win->inuse_cnt++;
921 *w_cursor = win;
922 }
60bb8b14 923 offset -= win->offset;
03e79c88 924 if (left)
dc49cd76 925 *left = win->len - xsize_t(offset);
03e79c88 926 return win->base + offset;
1f688557
JH
927}
928
27d69a46
NP
929static struct packed_git *alloc_packed_git(int extra)
930{
931 struct packed_git *p = xmalloc(sizeof(*p) + extra);
932 memset(p, 0, sizeof(*p));
933 p->pack_fd = -1;
934 return p;
935}
936
e0500293
JN
937static void try_to_free_pack_memory(size_t size)
938{
939 release_pack_memory(size, -1);
940}
941
42873078 942struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557 943{
e0500293 944 static int have_set_try_to_free_routine;
1f688557 945 struct stat st;
27d69a46 946 struct packed_git *p = alloc_packed_git(path_len + 2);
1f688557 947
e0500293
JN
948 if (!have_set_try_to_free_routine) {
949 have_set_try_to_free_routine = 1;
950 set_try_to_free_routine(try_to_free_pack_memory);
951 }
952
42873078
NP
953 /*
954 * Make sure a corresponding .pack file exists and that
955 * the index looks sane.
956 */
957 path_len -= strlen(".idx");
27d69a46
NP
958 if (path_len < 1) {
959 free(p);
1f688557 960 return NULL;
27d69a46 961 }
42873078 962 memcpy(p->pack_name, path, path_len);
8d25931d
BC
963
964 strcpy(p->pack_name + path_len, ".keep");
965 if (!access(p->pack_name, F_OK))
966 p->pack_keep = 1;
967
42873078 968 strcpy(p->pack_name + path_len, ".pack");
d079837e 969 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 970 free(p);
1f688557
JH
971 return NULL;
972 }
42873078 973
1f688557
JH
974 /* ok, it looks sane as far as we can check without
975 * actually mapping the pack file.
976 */
1f688557 977 p->pack_size = st.st_size;
9d835df2 978 p->pack_local = local;
b867092f 979 p->mtime = st.st_mtime;
42873078
NP
980 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
981 hashclr(p->sha1);
1f688557
JH
982 return p;
983}
984
7b64469a 985struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 986{
42873078 987 const char *path = sha1_pack_name(sha1);
27d69a46 988 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
bf592c50 989
27d69a46
NP
990 strcpy(p->pack_name, path);
991 hashcpy(p->sha1, sha1);
42873078
NP
992 if (check_packed_git_idx(idx_path, p)) {
993 free(p);
bf592c50 994 return NULL;
42873078 995 }
bf592c50 996
bf592c50
DB
997 return p;
998}
999
1000void install_packed_git(struct packed_git *pack)
1001{
c7934306
SP
1002 if (pack->pack_fd != -1)
1003 pack_open_fds++;
1004
bf592c50
DB
1005 pack->next = packed_git;
1006 packed_git = pack;
1007}
1008
543c5caa
NTND
1009void (*report_garbage)(const char *desc, const char *path);
1010
1011static void report_helper(const struct string_list *list,
1012 int seen_bits, int first, int last)
1013{
1014 const char *msg;
1015 switch (seen_bits) {
1016 case 0:
1017 msg = "no corresponding .idx nor .pack";
1018 break;
1019 case 1:
1020 msg = "no corresponding .idx";
1021 break;
1022 case 2:
1023 msg = "no corresponding .pack";
1024 break;
1025 default:
1026 return;
1027 }
1028 for (; first < last; first++)
1029 report_garbage(msg, list->items[first].string);
1030}
1031
1032static void report_pack_garbage(struct string_list *list)
1033{
1034 int i, baselen = -1, first = 0, seen_bits = 0;
1035
1036 if (!report_garbage)
1037 return;
1038
1039 sort_string_list(list);
1040
1041 for (i = 0; i < list->nr; i++) {
1042 const char *path = list->items[i].string;
1043 if (baselen != -1 &&
1044 strncmp(path, list->items[first].string, baselen)) {
1045 report_helper(list, seen_bits, first, i);
1046 baselen = -1;
1047 seen_bits = 0;
1048 }
1049 if (baselen == -1) {
1050 const char *dot = strrchr(path, '.');
1051 if (!dot) {
1052 report_garbage("garbage found", path);
1053 continue;
1054 }
1055 baselen = dot - path + 1;
1056 first = i;
1057 }
1058 if (!strcmp(path + baselen, "pack"))
1059 seen_bits |= 1;
1060 else if (!strcmp(path + baselen, "idx"))
1061 seen_bits |= 2;
1062 }
1063 report_helper(list, seen_bits, first, list->nr);
1064}
1065
9d835df2 1066static void prepare_packed_git_one(char *objdir, int local)
1f688557 1067{
9cb18f56
JM
1068 /* Ensure that this buffer is large enough so that we can
1069 append "/pack/" without clobbering the stack even if
1070 strlen(objdir) were PATH_MAX. */
1071 char path[PATH_MAX + 1 + 4 + 1 + 1];
1f688557
JH
1072 int len;
1073 DIR *dir;
1074 struct dirent *de;
543c5caa 1075 struct string_list garbage = STRING_LIST_INIT_DUP;
1f688557
JH
1076
1077 sprintf(path, "%s/pack", objdir);
1078 len = strlen(path);
1079 dir = opendir(path);
b5b16990 1080 if (!dir) {
26125f6b 1081 if (errno != ENOENT)
bd2afde8 1082 error("unable to open object pack directory: %s: %s",
26125f6b 1083 path, strerror(errno));
1f688557 1084 return;
b5b16990 1085 }
1f688557
JH
1086 path[len++] = '/';
1087 while ((de = readdir(dir)) != NULL) {
1088 int namelen = strlen(de->d_name);
1089 struct packed_git *p;
1090
543c5caa
NTND
1091 if (len + namelen + 1 > sizeof(path)) {
1092 if (report_garbage) {
1093 struct strbuf sb = STRBUF_INIT;
1094 strbuf_addf(&sb, "%.*s/%s", len - 1, path, de->d_name);
1095 report_garbage("path too long", sb.buf);
1096 strbuf_release(&sb);
1097 }
1098 continue;
1099 }
1100
1101 if (is_dot_or_dotdot(de->d_name))
9cb18f56
JM
1102 continue;
1103
1f688557 1104 strcpy(path + len, de->d_name);
d90906a9
NTND
1105
1106 if (has_extension(de->d_name, ".idx")) {
1107 /* Don't reopen a pack we already have. */
1108 for (p = packed_git; p; p = p->next) {
1109 if (!memcmp(path, p->pack_name, len + namelen - 4))
1110 break;
1111 }
1112 if (p == NULL &&
1113 /*
1114 * See if it really is a valid .idx file with
1115 * corresponding .pack file that we can map.
1116 */
1117 (p = add_packed_git(path, len + namelen, local)) != NULL)
1118 install_packed_git(p);
86f7780c 1119 }
543c5caa
NTND
1120
1121 if (!report_garbage)
1122 continue;
1123
1124 if (has_extension(de->d_name, ".idx") ||
1125 has_extension(de->d_name, ".pack") ||
1126 has_extension(de->d_name, ".keep"))
1127 string_list_append(&garbage, path);
1128 else
1129 report_garbage("garbage found", path);
1f688557 1130 }
5b35bcd5 1131 closedir(dir);
543c5caa
NTND
1132 report_pack_garbage(&garbage);
1133 string_list_clear(&garbage, 0);
1f688557
JH
1134}
1135
b867092f
JH
1136static int sort_pack(const void *a_, const void *b_)
1137{
1138 struct packed_git *a = *((struct packed_git **)a_);
1139 struct packed_git *b = *((struct packed_git **)b_);
1140 int st;
1141
1142 /*
1143 * Local packs tend to contain objects specific to our
1144 * variant of the project than remote ones. In addition,
1145 * remote ones could be on a network mounted filesystem.
1146 * Favor local ones for these reasons.
1147 */
1148 st = a->pack_local - b->pack_local;
1149 if (st)
1150 return -st;
1151
1152 /*
1153 * Younger packs tend to contain more recent objects,
1154 * and more recent objects tend to get accessed more
1155 * often.
1156 */
1157 if (a->mtime < b->mtime)
1158 return 1;
1159 else if (a->mtime == b->mtime)
1160 return 0;
1161 return -1;
1162}
1163
1164static void rearrange_packed_git(void)
1165{
1166 struct packed_git **ary, *p;
1167 int i, n;
1168
1169 for (n = 0, p = packed_git; p; p = p->next)
1170 n++;
1171 if (n < 2)
1172 return;
1173
1174 /* prepare an array of packed_git for easier sorting */
1175 ary = xcalloc(n, sizeof(struct packed_git *));
1176 for (n = 0, p = packed_git; p; p = p->next)
1177 ary[n++] = p;
1178
1179 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1180
1181 /* link them back again */
1182 for (i = 0; i < n - 1; i++)
1183 ary[i]->next = ary[i + 1];
1184 ary[n - 1]->next = NULL;
1185 packed_git = ary[0];
1186
1187 free(ary);
1188}
1189
637cdd9d 1190static int prepare_packed_git_run_once = 0;
9a217f2a 1191void prepare_packed_git(void)
1f688557 1192{
d5a63b99 1193 struct alternate_object_database *alt;
1f688557 1194
637cdd9d 1195 if (prepare_packed_git_run_once)
1f688557 1196 return;
9d835df2 1197 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 1198 prepare_alt_odb();
d5a63b99 1199 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 1200 alt->name[-1] = 0;
9d835df2 1201 prepare_packed_git_one(alt->base, 0);
1494e038 1202 alt->name[-1] = '/';
1f688557 1203 }
b867092f 1204 rearrange_packed_git();
637cdd9d
JK
1205 prepare_packed_git_run_once = 1;
1206}
1207
fc04c412 1208void reprepare_packed_git(void)
637cdd9d 1209{
4b480c67 1210 discard_revindex();
637cdd9d
JK
1211 prepare_packed_git_run_once = 0;
1212 prepare_packed_git();
1f688557
JH
1213}
1214
8eca0b47
NP
1215static void mark_bad_packed_object(struct packed_git *p,
1216 const unsigned char *sha1)
1217{
1218 unsigned i;
1219 for (i = 0; i < p->num_bad_objects; i++)
1220 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1221 return;
1222 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1223 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1224 p->num_bad_objects++;
1225}
1226
b6c4cecc 1227static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
ac939109
NP
1228{
1229 struct packed_git *p;
1230 unsigned i;
1231
1232 for (p = packed_git; p; p = p->next)
1233 for (i = 0; i < p->num_bad_objects; i++)
1234 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
b6c4cecc
JH
1235 return p;
1236 return NULL;
ac939109
NP
1237}
1238
090ea126
NTND
1239/*
1240 * With an in-core object data in "map", rehash it to make sure the
1241 * object name actually matches "sha1" to detect object corruption.
1242 * With "map" == NULL, try reading the object named with "sha1" using
1243 * the streaming interface and rehash it to do the same.
1244 */
1245int check_sha1_signature(const unsigned char *sha1, void *map,
1246 unsigned long size, const char *type)
0fcfd160
LT
1247{
1248 unsigned char real_sha1[20];
090ea126
NTND
1249 enum object_type obj_type;
1250 struct git_istream *st;
1251 git_SHA_CTX c;
1252 char hdr[32];
1253 int hdrlen;
1254
1255 if (map) {
1256 hash_sha1_file(map, size, type, real_sha1);
1257 return hashcmp(sha1, real_sha1) ? -1 : 0;
1258 }
1259
1260 st = open_istream(sha1, &obj_type, &size, NULL);
1261 if (!st)
1262 return -1;
1263
1264 /* Generate the header */
1265 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1266
1267 /* Sha1.. */
1268 git_SHA1_Init(&c);
1269 git_SHA1_Update(&c, hdr, hdrlen);
1270 for (;;) {
1271 char buf[1024 * 16];
1272 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1273
f54fac53
JK
1274 if (readlen < 0) {
1275 close_istream(st);
1276 return -1;
1277 }
090ea126
NTND
1278 if (!readlen)
1279 break;
1280 git_SHA1_Update(&c, buf, readlen);
1281 }
1282 git_SHA1_Final(real_sha1, &c);
1283 close_istream(st);
a89fccd2 1284 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1285}
1286
f4e51683 1287static int git_open_noatime(const char *name)
44d1c19e
LT
1288{
1289 static int sha1_file_open_flag = O_NOATIME;
44d1c19e 1290
f2e872aa
SP
1291 for (;;) {
1292 int fd = open(name, O_RDONLY | sha1_file_open_flag);
44d1c19e 1293 if (fd >= 0)
f2e872aa
SP
1294 return fd;
1295
f2e872aa
SP
1296 /* Might the failure be due to O_NOATIME? */
1297 if (errno != ENOENT && sha1_file_open_flag) {
44d1c19e 1298 sha1_file_open_flag = 0;
f2e872aa
SP
1299 continue;
1300 }
1301
1302 return -1;
44d1c19e 1303 }
44d1c19e
LT
1304}
1305
052fe5ea
JK
1306static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1307{
1308 char *name = sha1_file_name(sha1);
1309 struct alternate_object_database *alt;
1310
1311 if (!lstat(name, st))
1312 return 0;
1313
1314 prepare_alt_odb();
1315 errno = ENOENT;
1316 for (alt = alt_odb_list; alt; alt = alt->next) {
1317 name = alt->name;
1318 fill_sha1_path(name, sha1);
1319 if (!lstat(alt->base, st))
1320 return 0;
1321 }
1322
1323 return -1;
1324}
1325
44d1c19e
LT
1326static int open_sha1_file(const unsigned char *sha1)
1327{
1328 int fd;
1329 char *name = sha1_file_name(sha1);
1330 struct alternate_object_database *alt;
1331
f4e51683 1332 fd = git_open_noatime(name);
44d1c19e
LT
1333 if (fd >= 0)
1334 return fd;
1335
1336 prepare_alt_odb();
1337 errno = ENOENT;
1338 for (alt = alt_odb_list; alt; alt = alt->next) {
1339 name = alt->name;
1340 fill_sha1_path(name, sha1);
f4e51683 1341 fd = git_open_noatime(alt->base);
44d1c19e
LT
1342 if (fd >= 0)
1343 return fd;
1344 }
1345 return -1;
1346}
1347
f0270efd 1348void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1349{
0fcfd160 1350 void *map;
144bde78 1351 int fd;
ace1534d 1352
44d1c19e
LT
1353 fd = open_sha1_file(sha1);
1354 map = NULL;
1355 if (fd >= 0) {
1356 struct stat st;
0fcfd160 1357
44d1c19e
LT
1358 if (!fstat(fd, &st)) {
1359 *size = xsize_t(st.st_size);
33e42de0
MM
1360 if (!*size) {
1361 /* mmap() is forbidden on empty files */
1362 error("object file %s is empty", sha1_file_name(sha1));
1363 return NULL;
1364 }
44d1c19e 1365 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1366 }
44d1c19e 1367 close(fd);
0fcfd160 1368 }
0fcfd160
LT
1369 return map;
1370}
1371
cc5c54e7
JH
1372/*
1373 * There used to be a second loose object header format which
1374 * was meant to mimic the in-pack format, allowing for direct
1375 * copy of the object data. This format turned up not to be
1376 * really worth it and we no longer write loose objects in that
1377 * format.
1378 */
1379static int experimental_loose_object(unsigned char *map)
bb6b8e4f
JH
1380{
1381 unsigned int word;
1382
1383 /*
7f684a2a
RT
1384 * We must determine if the buffer contains the standard
1385 * zlib-deflated stream or the experimental format based
1386 * on the in-pack object format. Compare the header byte
1387 * for each format:
1388 *
1389 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1390 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1391 *
1392 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1393 * in standard loose-object format, UNLESS it is a Git-pack
1394 * format object *exactly* 8 bytes in size when inflated.
1395 *
1396 * However, RFC1950 also specifies that the 1st 16-bit word
1397 * must be divisible by 31 - this checksum tells us our buffer
1398 * is in the standard format, giving a false positive only if
1399 * the 1st word of the Git-pack format object happens to be
1400 * divisible by 31, ie:
1401 * ((byte0 * 256) + byte1) % 31 = 0
1402 * => 0ttt10000www1000 % 31 = 0
1403 *
1404 * As it happens, this case can only arise for www=3 & ttt=1
1405 * - ie, a Commit object, which would have to be 8 bytes in
1406 * size. As no Commit can be that small, we find that the
1407 * combination of these two criteria (bitmask & checksum)
1408 * can always correctly determine the buffer format.
bb6b8e4f
JH
1409 */
1410 word = (map[0] << 8) + map[1];
7f684a2a 1411 if ((map[0] & 0x8F) == 0x08 && !(word % 31))
bb6b8e4f 1412 return 0;
cc5c54e7
JH
1413 else
1414 return 1;
bb6b8e4f
JH
1415}
1416
09ded04b
NP
1417unsigned long unpack_object_header_buffer(const unsigned char *buf,
1418 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1419{
ad1ed5ee 1420 unsigned shift;
48fb7deb 1421 unsigned long size, c;
ad1ed5ee
JH
1422 unsigned long used = 0;
1423
1424 c = buf[used++];
1425 *type = (c >> 4) & 7;
1426 size = c & 15;
1427 shift = 4;
1428 while (c & 0x80) {
f630cfda 1429 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1430 error("bad object header");
ea4f9685
JH
1431 size = used = 0;
1432 break;
09ded04b 1433 }
ad1ed5ee
JH
1434 c = buf[used++];
1435 size += (c & 0x7f) << shift;
1436 shift += 7;
1437 }
1438 *sizep = size;
1439 return used;
1440}
1441
eb4f4076 1442int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
ad1ed5ee
JH
1443{
1444 unsigned long size, used;
1445 static const char valid_loose_object_type[8] = {
1446 0, /* OBJ_EXT */
1447 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1448 0, /* "delta" and others are invalid in a loose object */
93821bd9 1449 };
ad1ed5ee 1450 enum object_type type;
93821bd9 1451
c4483576
LT
1452 /* Get the data stream */
1453 memset(stream, 0, sizeof(*stream));
1454 stream->next_in = map;
1455 stream->avail_in = mapsize;
1456 stream->next_out = buffer;
93821bd9
LT
1457 stream->avail_out = bufsiz;
1458
cc5c54e7
JH
1459 if (experimental_loose_object(map)) {
1460 /*
1461 * The old experimental format we no longer produce;
1462 * we can still read it.
1463 */
1464 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1465 if (!used || !valid_loose_object_type[type])
1466 return -1;
1467 map += used;
1468 mapsize -= used;
726f852b 1469
cc5c54e7
JH
1470 /* Set up the stream for the rest.. */
1471 stream->next_in = map;
1472 stream->avail_in = mapsize;
1473 git_inflate_init(stream);
93821bd9 1474
cc5c54e7
JH
1475 /* And generate the fake traditional header */
1476 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1477 typename(type), size);
1478 return 0;
1479 }
39c68542 1480 git_inflate_init(stream);
cc5c54e7 1481 return git_inflate(stream, 0);
c4483576
LT
1482}
1483
ef49a7a0 1484static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1485{
1486 int bytes = strlen(buffer) + 1;
3aee68aa 1487 unsigned char *buf = xmallocz(size);
93821bd9 1488 unsigned long n;
7efbff75 1489 int status = Z_OK;
5180cacc 1490
93821bd9
LT
1491 n = stream->total_out - bytes;
1492 if (n > size)
1493 n = size;
1494 memcpy(buf, (char *) buffer + bytes, n);
1495 bytes = n;
456cdf6e
LT
1496 if (bytes <= size) {
1497 /*
1498 * The above condition must be (bytes <= size), not
1499 * (bytes < size). In other words, even though we
ccf5ace0 1500 * expect no more output and set avail_out to zero,
456cdf6e
LT
1501 * the input zlib stream may have bytes that express
1502 * "this concludes the stream", and we *do* want to
1503 * eat that input.
1504 *
1505 * Otherwise we would not be able to test that we
1506 * consumed all the input to reach the expected size;
1507 * we also want to check that zlib tells us that all
1508 * went well with status == Z_STREAM_END at the end.
1509 */
5180cacc
LT
1510 stream->next_out = buf + bytes;
1511 stream->avail_out = size - bytes;
7efbff75 1512 while (status == Z_OK)
39c68542 1513 status = git_inflate(stream, Z_FINISH);
5180cacc 1514 }
456cdf6e 1515 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1516 git_inflate_end(stream);
7efbff75
JH
1517 return buf;
1518 }
1519
1520 if (status < 0)
1521 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1522 else if (stream->avail_in)
1523 error("garbage at end of loose object '%s'",
1524 sha1_to_hex(sha1));
1525 free(buf);
1526 return NULL;
5180cacc
LT
1527}
1528
1529/*
1530 * We used to just use "sscanf()", but that's actually way
1531 * too permissive for what we want to check. So do an anal
1532 * object header parse by hand.
1533 */
f0270efd 1534int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1535{
21666f1a 1536 char type[10];
5180cacc
LT
1537 int i;
1538 unsigned long size;
1539
1540 /*
a6080a0a 1541 * The type can be at most ten bytes (including the
5180cacc 1542 * terminating '\0' that we add), and is followed by
21666f1a 1543 * a space.
5180cacc 1544 */
21666f1a 1545 i = 0;
5180cacc
LT
1546 for (;;) {
1547 char c = *hdr++;
1548 if (c == ' ')
1549 break;
21666f1a
NP
1550 type[i++] = c;
1551 if (i >= sizeof(type))
5180cacc 1552 return -1;
5180cacc 1553 }
21666f1a 1554 type[i] = 0;
5180cacc
LT
1555
1556 /*
1557 * The length must follow immediately, and be in canonical
1558 * decimal format (ie "010" is not valid).
1559 */
1560 size = *hdr++ - '0';
1561 if (size > 9)
1562 return -1;
1563 if (size) {
1564 for (;;) {
1565 unsigned long c = *hdr - '0';
1566 if (c > 9)
1567 break;
1568 hdr++;
1569 size = size * 10 + c;
1570 }
1571 }
1572 *sizep = size;
1573
1574 /*
1575 * The length must be followed by a zero byte
1576 */
21666f1a 1577 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1578}
1579
7efbff75 1580static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1581{
5180cacc 1582 int ret;
ef49a7a0 1583 git_zstream stream;
5180cacc 1584 char hdr[8192];
0fcfd160 1585
5180cacc 1586 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1587 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1588 return NULL;
1589
7efbff75 1590 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1591}
1592
54dab52a
NP
1593unsigned long get_size_from_delta(struct packed_git *p,
1594 struct pack_window **w_curs,
1595 off_t curpos)
1596{
1597 const unsigned char *data;
1598 unsigned char delta_head[20], *in;
ef49a7a0 1599 git_zstream stream;
54dab52a
NP
1600 int st;
1601
1602 memset(&stream, 0, sizeof(stream));
1603 stream.next_out = delta_head;
1604 stream.avail_out = sizeof(delta_head);
1605
39c68542 1606 git_inflate_init(&stream);
54dab52a
NP
1607 do {
1608 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1609 stream.next_in = in;
39c68542 1610 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1611 curpos += stream.next_in - in;
1612 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1613 stream.total_out < sizeof(delta_head));
39c68542 1614 git_inflate_end(&stream);
3d77d877
NP
1615 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1616 error("delta data unpack-initial failed");
1617 return 0;
1618 }
54dab52a
NP
1619
1620 /* Examine the initial part of the delta to figure out
1621 * the result size.
1622 */
1623 data = delta_head;
1624
1625 /* ignore base size */
1626 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1627
1628 /* Read the result size */
1629 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1630}
1631
c4001d92 1632static off_t get_delta_base(struct packed_git *p,
03e79c88 1633 struct pack_window **w_curs,
c4001d92 1634 off_t *curpos,
21666f1a 1635 enum object_type type,
c4001d92 1636 off_t delta_obj_offset)
eb32d236 1637{
2b87c45b 1638 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1639 off_t base_offset;
eb32d236 1640
8d8a4ea5
SP
1641 /* use_pack() assured us we have [base_info, base_info + 20)
1642 * as a range that we can look at without walking off the
1643 * end of the mapped window. Its actually the hash size
1644 * that is assured. An OFS_DELTA longer than the hash size
1645 * is stupid, as then a REF_DELTA would be smaller to store.
1646 */
21666f1a 1647 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1648 unsigned used = 0;
1649 unsigned char c = base_info[used++];
1650 base_offset = c & 127;
1651 while (c & 128) {
1652 base_offset += 1;
8723f216 1653 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1654 return 0; /* overflow */
eb32d236
NP
1655 c = base_info[used++];
1656 base_offset = (base_offset << 7) + (c & 127);
1657 }
1658 base_offset = delta_obj_offset - base_offset;
d8f32556 1659 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1660 return 0; /* out of bound */
2b87c45b 1661 *curpos += used;
21666f1a 1662 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1663 /* The base entry _must_ be in the same pack */
1664 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1665 *curpos += 20;
eb32d236
NP
1666 } else
1667 die("I am totally screwed");
2b87c45b 1668 return base_offset;
eb32d236
NP
1669}
1670
f8c8abc5
JH
1671int unpack_object_header(struct packed_git *p,
1672 struct pack_window **w_curs,
1673 off_t *curpos,
1674 unsigned long *sizep)
a733cb60 1675{
03e79c88 1676 unsigned char *base;
ef49a7a0 1677 unsigned long left;
ad1ed5ee 1678 unsigned long used;
2b87c45b 1679 enum object_type type;
a733cb60 1680
8d8a4ea5 1681 /* use_pack() assures us we have [base, base + 20) available
0353a0c4 1682 * as a range that we can look at. (Its actually the hash
3dff5379 1683 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1684 * the maximum deflated object size is 2^137, which is just
1685 * insane, so we know won't exceed what we have been given.
1686 */
2b87c45b 1687 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1688 used = unpack_object_header_buffer(base, left, &type, sizep);
1689 if (!used) {
1690 type = OBJ_BAD;
1691 } else
1692 *curpos += used;
ad1ed5ee 1693
2b87c45b 1694 return type;
a733cb60
LT
1695}
1696
790d96c0
TR
1697static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1698{
1699 int type;
1700 struct revindex_entry *revidx;
1701 const unsigned char *sha1;
1702 revidx = find_pack_revindex(p, obj_offset);
1703 if (!revidx)
1704 return OBJ_BAD;
1705 sha1 = nth_packed_object_sha1(p, revidx->nr);
1706 mark_bad_packed_object(p, sha1);
1707 type = sha1_object_info(sha1, NULL);
1708 if (type <= OBJ_NONE)
1709 return OBJ_BAD;
1710 return type;
1711}
1712
1713
1714#define POI_STACK_PREALLOC 64
1715
c4001d92 1716static int packed_object_info(struct packed_git *p, off_t obj_offset,
161f00e7
JK
1717 unsigned long *sizep, int *rtype,
1718 unsigned long *disk_sizep)
1f688557 1719{
03e79c88 1720 struct pack_window *w_curs = NULL;
c4001d92
SP
1721 unsigned long size;
1722 off_t curpos = obj_offset;
21666f1a 1723 enum object_type type;
790d96c0
TR
1724 off_t small_poi_stack[POI_STACK_PREALLOC];
1725 off_t *poi_stack = small_poi_stack;
1726 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
5db47c2b 1727
21666f1a 1728 type = unpack_object_header(p, &w_curs, &curpos, &size);
790d96c0 1729
9a490590
JH
1730 if (rtype)
1731 *rtype = type; /* representation type */
5db47c2b 1732
790d96c0
TR
1733 if (sizep) {
1734 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1735 off_t tmp_pos = curpos;
1736 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1737 type, obj_offset);
1738 if (!base_offset) {
1739 type = OBJ_BAD;
1740 goto out;
1741 }
1742 *sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1743 if (*sizep == 0) {
1744 type = OBJ_BAD;
1745 goto out;
1746 }
1747 } else {
1748 *sizep = size;
1749 }
1750 }
1751
161f00e7
JK
1752 if (disk_sizep) {
1753 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1754 *disk_sizep = revidx[1].offset - obj_offset;
1755 }
1756
790d96c0
TR
1757 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1758 off_t base_offset;
1759 /* Push the object we're going to leave behind */
1760 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1761 poi_stack_alloc = alloc_nr(poi_stack_nr);
1762 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1763 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1764 } else {
1765 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1766 }
1767 poi_stack[poi_stack_nr++] = obj_offset;
1768 /* If parsing the base offset fails, just unwind */
1769 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1770 if (!base_offset)
1771 goto unwind;
1772 curpos = obj_offset = base_offset;
1773 type = unpack_object_header(p, &w_curs, &curpos, &size);
1774 if (type <= OBJ_NONE) {
1775 /* If getting the base itself fails, we first
1776 * retry the base, otherwise unwind */
1777 type = retry_bad_packed_offset(p, base_offset);
1778 if (type > OBJ_NONE)
1779 goto out;
1780 goto unwind;
1781 }
1782 }
1783
21666f1a 1784 switch (type) {
790d96c0 1785 case OBJ_BAD:
a733cb60 1786 case OBJ_COMMIT:
a733cb60 1787 case OBJ_TREE:
a733cb60 1788 case OBJ_BLOB:
a733cb60 1789 case OBJ_TAG:
a69d0943 1790 break;
1f688557 1791 default:
3d77d877
NP
1792 error("unknown object type %i at offset %"PRIuMAX" in %s",
1793 type, (uintmax_t)obj_offset, p->pack_name);
1794 type = OBJ_BAD;
1f688557 1795 }
790d96c0
TR
1796
1797out:
1798 if (poi_stack != small_poi_stack)
1799 free(poi_stack);
2b87c45b 1800 unuse_pack(&w_curs);
21666f1a 1801 return type;
790d96c0
TR
1802
1803unwind:
1804 while (poi_stack_nr) {
1805 obj_offset = poi_stack[--poi_stack_nr];
1806 type = retry_bad_packed_offset(p, obj_offset);
1807 if (type > OBJ_NONE)
1808 goto out;
1809 }
1810 type = OBJ_BAD;
1811 goto out;
1f688557
JH
1812}
1813
eb950c19 1814static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1815 struct pack_window **w_curs,
c4001d92 1816 off_t curpos,
eb950c19 1817 unsigned long size)
de530aaa
SP
1818{
1819 int st;
ef49a7a0 1820 git_zstream stream;
079afb18 1821 unsigned char *buffer, *in;
de530aaa 1822
4ab07e4d 1823 buffer = xmallocz(size);
de530aaa 1824 memset(&stream, 0, sizeof(stream));
de530aaa 1825 stream.next_out = buffer;
39eea7bd 1826 stream.avail_out = size + 1;
de530aaa 1827
39c68542 1828 git_inflate_init(&stream);
079afb18 1829 do {
2b87c45b 1830 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 1831 stream.next_in = in;
39c68542 1832 st = git_inflate(&stream, Z_FINISH);
39eea7bd
JH
1833 if (!stream.avail_out)
1834 break; /* the payload is larger than it should be */
2b87c45b 1835 curpos += stream.next_in - in;
079afb18 1836 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 1837 git_inflate_end(&stream);
de530aaa
SP
1838 if ((st != Z_STREAM_END) || stream.total_out != size) {
1839 free(buffer);
1840 return NULL;
1841 }
1842
1843 return buffer;
1844}
1845
e5e01619
LT
1846#define MAX_DELTA_CACHE (256)
1847
18bdec11 1848static size_t delta_base_cached;
5e08ecbf
NP
1849
1850static struct delta_base_cache_lru_list {
1851 struct delta_base_cache_lru_list *prev;
1852 struct delta_base_cache_lru_list *next;
1853} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1854
e5e01619 1855static struct delta_base_cache_entry {
5e08ecbf
NP
1856 struct delta_base_cache_lru_list lru;
1857 void *data;
e5e01619
LT
1858 struct packed_git *p;
1859 off_t base_offset;
1860 unsigned long size;
e5e01619
LT
1861 enum object_type type;
1862} delta_base_cache[MAX_DELTA_CACHE];
1863
1864static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1865{
1866 unsigned long hash;
1867
1868 hash = (unsigned long)p + (unsigned long)base_offset;
1869 hash += (hash >> 8) + (hash >> 16);
3358004a 1870 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1871}
1872
84dd81c1
TR
1873static struct delta_base_cache_entry *
1874get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
5266d369
JH
1875{
1876 unsigned long hash = pack_entry_hash(p, base_offset);
84dd81c1
TR
1877 return delta_base_cache + hash;
1878}
1879
1880static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
1881 struct packed_git *p, off_t base_offset)
1882{
5266d369
JH
1883 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1884}
1885
84dd81c1
TR
1886static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1887{
1888 struct delta_base_cache_entry *ent;
1889 ent = get_delta_base_cache_entry(p, base_offset);
1890 return eq_delta_base_cache_entry(ent, p, base_offset);
1891}
1892
1893static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1894{
1895 ent->data = NULL;
1896 ent->lru.next->prev = ent->lru.prev;
1897 ent->lru.prev->next = ent->lru.next;
1898 delta_base_cached -= ent->size;
1899}
1900
62f255ad 1901static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 1902 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 1903{
84dd81c1 1904 struct delta_base_cache_entry *ent;
e5e01619 1905 void *ret;
e5e01619 1906
84dd81c1
TR
1907 ent = get_delta_base_cache_entry(p, base_offset);
1908
1909 if (!eq_delta_base_cache_entry(ent, p, base_offset))
749bc58c 1910 return unpack_entry(p, base_offset, type, base_size);
e5e01619 1911
84dd81c1
TR
1912 ret = ent->data;
1913
1914 if (!keep_cache)
1915 clear_delta_base_cache_entry(ent);
1916 else
182af834 1917 ret = xmemdupz(ent->data, ent->size);
e5e01619
LT
1918 *type = ent->type;
1919 *base_size = ent->size;
1920 return ret;
62f255ad
LT
1921}
1922
18bdec11
SP
1923static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1924{
1925 if (ent->data) {
1926 free(ent->data);
1927 ent->data = NULL;
5e08ecbf
NP
1928 ent->lru.next->prev = ent->lru.prev;
1929 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
1930 delta_base_cached -= ent->size;
1931 }
1932}
1933
3d20c636
SP
1934void clear_delta_base_cache(void)
1935{
1936 unsigned long p;
1937 for (p = 0; p < MAX_DELTA_CACHE; p++)
1938 release_delta_base_cache(&delta_base_cache[p]);
1939}
1940
62f255ad
LT
1941static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1942 void *base, unsigned long base_size, enum object_type type)
1943{
5e08ecbf 1944 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 1945 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 1946 struct delta_base_cache_lru_list *lru;
e5e01619 1947
18bdec11
SP
1948 release_delta_base_cache(ent);
1949 delta_base_cached += base_size;
5e08ecbf
NP
1950
1951 for (lru = delta_base_cache_lru.next;
1952 delta_base_cached > delta_base_cache_limit
1953 && lru != &delta_base_cache_lru;
1954 lru = lru->next) {
1955 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
1956 if (f->type == OBJ_BLOB)
1957 release_delta_base_cache(f);
1958 }
5e08ecbf
NP
1959 for (lru = delta_base_cache_lru.next;
1960 delta_base_cached > delta_base_cache_limit
1961 && lru != &delta_base_cache_lru;
1962 lru = lru->next) {
1963 struct delta_base_cache_entry *f = (void *)lru;
1964 release_delta_base_cache(f);
1965 }
18bdec11 1966
e5e01619
LT
1967 ent->p = p;
1968 ent->base_offset = base_offset;
1969 ent->type = type;
1970 ent->data = base;
1971 ent->size = base_size;
5e08ecbf
NP
1972 ent->lru.next = &delta_base_cache_lru;
1973 ent->lru.prev = delta_base_cache_lru.prev;
1974 delta_base_cache_lru.prev->next = &ent->lru;
1975 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
1976}
1977
c2c5b270
CC
1978static void *read_object(const unsigned char *sha1, enum object_type *type,
1979 unsigned long *size);
1980
5f44324d
JH
1981static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1982{
1983 static FILE *log_file;
1984
1985 if (!log_file) {
1986 log_file = fopen(log_pack_access, "w");
1987 if (!log_file) {
1988 error("cannot open pack access log '%s' for writing: %s",
1989 log_pack_access, strerror(errno));
1990 log_pack_access = NULL;
1991 return;
1992 }
1993 }
1994 fprintf(log_file, "%s %"PRIuMAX"\n",
1995 p->pack_name, (uintmax_t)obj_offset);
1996 fflush(log_file);
1997}
1998
0e8189e2
NP
1999int do_check_packed_object_crc;
2000
abe601bb
TR
2001#define UNPACK_ENTRY_STACK_PREALLOC 64
2002struct unpack_entry_stack_ent {
2003 off_t obj_offset;
2004 off_t curpos;
2005 unsigned long size;
2006};
2007
c4001d92 2008void *unpack_entry(struct packed_git *p, off_t obj_offset,
abe601bb 2009 enum object_type *final_type, unsigned long *final_size)
1f688557 2010{
03e79c88 2011 struct pack_window *w_curs = NULL;
c4001d92 2012 off_t curpos = obj_offset;
abe601bb
TR
2013 void *data = NULL;
2014 unsigned long size;
2015 enum object_type type;
2016 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2017 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2018 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2019 int base_from_cache = 0;
1f688557 2020
5f44324d
JH
2021 if (log_pack_access)
2022 write_pack_access_log(p, obj_offset);
2023
abe601bb
TR
2024 /* PHASE 1: drill down to the innermost base object */
2025 for (;;) {
2026 off_t base_offset;
2027 int i;
2028 struct delta_base_cache_entry *ent;
2029
2030 if (do_check_packed_object_crc && p->index_version > 1) {
2031 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2032 unsigned long len = revidx[1].offset - obj_offset;
2033 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2034 const unsigned char *sha1 =
2035 nth_packed_object_sha1(p, revidx->nr);
2036 error("bad packed object CRC for %s",
2037 sha1_to_hex(sha1));
2038 mark_bad_packed_object(p, sha1);
2039 unuse_pack(&w_curs);
2040 return NULL;
2041 }
2042 }
2043
2044 ent = get_delta_base_cache_entry(p, curpos);
2045 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2046 type = ent->type;
2047 data = ent->data;
2048 size = ent->size;
2049 clear_delta_base_cache_entry(ent);
2050 base_from_cache = 1;
2051 break;
0e8189e2 2052 }
abe601bb
TR
2053
2054 type = unpack_object_header(p, &w_curs, &curpos, &size);
2055 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2056 break;
2057
2058 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2059 if (!base_offset) {
2060 error("failed to validate delta base reference "
2061 "at offset %"PRIuMAX" from %s",
2062 (uintmax_t)curpos, p->pack_name);
2063 /* bail to phase 2, in hopes of recovery */
2064 data = NULL;
2065 break;
2066 }
2067
2068 /* push object, proceed to base */
2069 if (delta_stack_nr >= delta_stack_alloc
2070 && delta_stack == small_delta_stack) {
2071 delta_stack_alloc = alloc_nr(delta_stack_nr);
2072 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2073 memcpy(delta_stack, small_delta_stack,
2074 sizeof(*delta_stack)*delta_stack_nr);
2075 } else {
2076 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2077 }
2078 i = delta_stack_nr++;
2079 delta_stack[i].obj_offset = obj_offset;
2080 delta_stack[i].curpos = curpos;
2081 delta_stack[i].size = size;
2082
2083 curpos = obj_offset = base_offset;
0e8189e2
NP
2084 }
2085
abe601bb
TR
2086 /* PHASE 2: handle the base */
2087 switch (type) {
eb32d236
NP
2088 case OBJ_OFS_DELTA:
2089 case OBJ_REF_DELTA:
abe601bb
TR
2090 if (data)
2091 die("BUG in unpack_entry: left loop at a valid delta");
4d703a1a 2092 break;
a733cb60 2093 case OBJ_COMMIT:
a733cb60 2094 case OBJ_TREE:
a733cb60 2095 case OBJ_BLOB:
a733cb60 2096 case OBJ_TAG:
abe601bb
TR
2097 if (!base_from_cache)
2098 data = unpack_compressed_entry(p, &w_curs, curpos, size);
4d703a1a 2099 break;
1f688557 2100 default:
8eca0b47
NP
2101 data = NULL;
2102 error("unknown object type %i at offset %"PRIuMAX" in %s",
abe601bb
TR
2103 type, (uintmax_t)obj_offset, p->pack_name);
2104 }
2105
2106 /* PHASE 3: apply deltas in order */
2107
2108 /* invariants:
2109 * 'data' holds the base data, or NULL if there was corruption
2110 */
2111 while (delta_stack_nr) {
2112 void *delta_data;
2113 void *base = data;
2114 unsigned long delta_size, base_size = size;
2115 int i;
2116
2117 data = NULL;
2118
2119 if (base)
2120 add_delta_base_cache(p, obj_offset, base, base_size, type);
2121
2122 if (!base) {
2123 /*
2124 * We're probably in deep shit, but let's try to fetch
2125 * the required base anyway from another pack or loose.
2126 * This is costly but should happen only in the presence
2127 * of a corrupted pack, and is better than failing outright.
2128 */
2129 struct revindex_entry *revidx;
2130 const unsigned char *base_sha1;
2131 revidx = find_pack_revindex(p, obj_offset);
2132 if (revidx) {
2133 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2134 error("failed to read delta base object %s"
2135 " at offset %"PRIuMAX" from %s",
2136 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2137 p->pack_name);
2138 mark_bad_packed_object(p, base_sha1);
2139 base = read_object(base_sha1, &type, &base_size);
2140 }
2141 }
2142
2143 i = --delta_stack_nr;
2144 obj_offset = delta_stack[i].obj_offset;
2145 curpos = delta_stack[i].curpos;
2146 delta_size = delta_stack[i].size;
2147
2148 if (!base)
2149 continue;
2150
2151 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2152
2153 if (!delta_data) {
2154 error("failed to unpack compressed delta "
2155 "at offset %"PRIuMAX" from %s",
2156 (uintmax_t)curpos, p->pack_name);
abe601bb
TR
2157 data = NULL;
2158 continue;
2159 }
2160
2161 data = patch_delta(base, base_size,
2162 delta_data, delta_size,
2163 &size);
2164 if (!data)
2165 die("failed to apply delta");
2166
2167 free (delta_data);
1f688557 2168 }
abe601bb
TR
2169
2170 *final_type = type;
2171 *final_size = size;
2172
03e79c88 2173 unuse_pack(&w_curs);
21666f1a 2174 return data;
1f688557
JH
2175}
2176
d079837e 2177const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 2178 uint32_t n)
9a217f2a 2179{
42873078 2180 const unsigned char *index = p->index_data;
d079837e
SP
2181 if (!index) {
2182 if (open_pack_index(p))
2183 return NULL;
2184 index = p->index_data;
2185 }
57059091 2186 if (n >= p->num_objects)
d72308e0 2187 return NULL;
74e34e1f
NP
2188 index += 4 * 256;
2189 if (p->index_version == 1) {
2190 return index + 24 * n + 4;
2191 } else {
2192 index += 8;
2193 return index + 20 * n;
2194 }
2195}
2196
99093238 2197off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
2198{
2199 const unsigned char *index = p->index_data;
2200 index += 4 * 256;
2201 if (p->index_version == 1) {
2202 return ntohl(*((uint32_t *)(index + 24 * n)));
2203 } else {
2204 uint32_t off;
2205 index += 8 + p->num_objects * (20 + 4);
2206 off = ntohl(*((uint32_t *)(index + 4 * n)));
2207 if (!(off & 0x80000000))
2208 return off;
2209 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2210 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2211 ntohl(*((uint32_t *)(index + 4)));
2212 }
9a217f2a
JH
2213}
2214
c4001d92 2215off_t find_pack_entry_one(const unsigned char *sha1,
43057304 2216 struct packed_git *p)
1f688557 2217{
42873078 2218 const uint32_t *level1_ofs = p->index_data;
42873078 2219 const unsigned char *index = p->index_data;
628522ec
JH
2220 unsigned hi, lo, stride;
2221 static int use_lookup = -1;
2222 static int debug_lookup = -1;
2223
2224 if (debug_lookup < 0)
2225 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 2226
d079837e
SP
2227 if (!index) {
2228 if (open_pack_index(p))
2229 return 0;
2230 level1_ofs = p->index_data;
2231 index = p->index_data;
2232 }
74e34e1f
NP
2233 if (p->index_version > 1) {
2234 level1_ofs += 2;
2235 index += 8;
2236 }
42873078 2237 index += 4 * 256;
74e34e1f
NP
2238 hi = ntohl(level1_ofs[*sha1]);
2239 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
2240 if (p->index_version > 1) {
2241 stride = 20;
2242 } else {
2243 stride = 24;
2244 index += 4;
2245 }
2246
2247 if (debug_lookup)
6e1c2344 2248 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
2249 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2250
2251 if (use_lookup < 0)
2252 use_lookup = !!getenv("GIT_USE_LOOKUP");
2253 if (use_lookup) {
2254 int pos = sha1_entry_pos(index, stride, 0,
2255 lo, hi, p->num_objects, sha1);
2256 if (pos < 0)
2257 return 0;
2258 return nth_packed_object_offset(p, pos);
2259 }
1f688557
JH
2260
2261 do {
74e34e1f 2262 unsigned mi = (lo + hi) / 2;
628522ec
JH
2263 int cmp = hashcmp(index + mi * stride, sha1);
2264
2265 if (debug_lookup)
2266 printf("lo %u hi %u rg %u mi %u\n",
2267 lo, hi, hi - lo, mi);
43057304 2268 if (!cmp)
74e34e1f 2269 return nth_packed_object_offset(p, mi);
1f688557
JH
2270 if (cmp > 0)
2271 hi = mi;
2272 else
2273 lo = mi+1;
2274 } while (lo < hi);
2275 return 0;
2276}
2277
4c080182 2278int is_pack_valid(struct packed_git *p)
d131b7af
SP
2279{
2280 /* An already open pack is known to be valid. */
2281 if (p->pack_fd != -1)
2282 return 1;
2283
2284 /* If the pack has one window completely covering the
2285 * file size, the pack is known to be valid even if
2286 * the descriptor is not currently open.
2287 */
2288 if (p->windows) {
2289 struct pack_window *w = p->windows;
2290
2291 if (!w->offset && w->len == p->pack_size)
2292 return 1;
2293 }
2294
2295 /* Force the pack to open to prove its valid. */
2296 return !open_packed_git(p);
2297}
2298
95099731
NTND
2299static int fill_pack_entry(const unsigned char *sha1,
2300 struct pack_entry *e,
2301 struct packed_git *p)
2302{
2303 off_t offset;
2304
2305 if (p->num_bad_objects) {
2306 unsigned i;
2307 for (i = 0; i < p->num_bad_objects; i++)
2308 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2309 return 0;
2310 }
2311
2312 offset = find_pack_entry_one(sha1, p);
2313 if (!offset)
2314 return 0;
2315
2316 /*
2317 * We are about to tell the caller where they can locate the
2318 * requested object. We better make sure the packfile is
2319 * still here and can be accessed before supplying that
2320 * answer, as it may have been deleted since the index was
2321 * loaded!
2322 */
2323 if (!is_pack_valid(p)) {
2324 warning("packfile %s cannot be accessed", p->pack_name);
2325 return 0;
2326 }
2327 e->offset = offset;
2328 e->p = p;
2329 hashcpy(e->sha1, sha1);
2330 return 1;
2331}
2332
4d6acb70 2333static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557
JH
2334{
2335 struct packed_git *p;
43057304 2336
1f688557 2337 prepare_packed_git();
f7c22cc6
NP
2338 if (!packed_git)
2339 return 0;
1f688557 2340
c01f51cc
NTND
2341 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2342 return 1;
8eca0b47 2343
c01f51cc
NTND
2344 for (p = packed_git; p; p = p->next) {
2345 if (p == last_found_pack || !fill_pack_entry(sha1, e, p))
2346 continue;
f7c22cc6 2347
c01f51cc
NTND
2348 last_found_pack = p;
2349 return 1;
2350 }
1f688557
JH
2351 return 0;
2352}
2353
a6080a0a 2354struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
2355 struct packed_git *packs)
2356{
2357 struct packed_git *p;
bf592c50
DB
2358
2359 for (p = packs; p; p = p->next) {
43057304 2360 if (find_pack_entry_one(sha1, p))
bf592c50
DB
2361 return p;
2362 }
2363 return NULL;
21666f1a 2364
bf592c50
DB
2365}
2366
052fe5ea
JK
2367static int sha1_loose_object_info(const unsigned char *sha1,
2368 enum object_type *typep,
2369 unsigned long *sizep,
161f00e7 2370 unsigned long *disk_sizep)
65c2e0c3 2371{
36e4d74a 2372 int status;
65c2e0c3
JH
2373 unsigned long mapsize, size;
2374 void *map;
ef49a7a0 2375 git_zstream stream;
d65a16f6 2376 char hdr[32];
65c2e0c3 2377
052fe5ea
JK
2378 /*
2379 * If we don't care about type or size, then we don't
2380 * need to look inside the object at all.
2381 */
2382 if (!typep && !sizep) {
2383 if (disk_sizep) {
2384 struct stat st;
2385 if (stat_sha1_file(sha1, &st) < 0)
2386 return -1;
2387 *disk_sizep = st.st_size;
2388 }
2389 return 0;
2390 }
2391
bb6b8e4f 2392 map = map_sha1_file(sha1, &mapsize);
f0df4ed5
JS
2393 if (!map)
2394 return error("unable to find %s", sha1_to_hex(sha1));
161f00e7
JK
2395 if (disk_sizep)
2396 *disk_sizep = mapsize;
36e4d74a
JH
2397 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2398 status = error("unable to unpack %s header",
2399 sha1_to_hex(sha1));
21666f1a 2400 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 2401 status = error("unable to parse %s header", sha1_to_hex(sha1));
21666f1a
NP
2402 else if (sizep)
2403 *sizep = size;
39c68542 2404 git_inflate_end(&stream);
65c2e0c3 2405 munmap(map, mapsize);
052fe5ea
JK
2406 if (typep)
2407 *typep = status;
2408 return 0;
65c2e0c3
JH
2409}
2410
9a490590
JH
2411/* returns enum object_type or negative */
2412int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
f0df4ed5 2413{
c4d9986f 2414 struct cached_object *co;
f0df4ed5 2415 struct pack_entry e;
f2f57e31 2416 int type, rtype;
f0df4ed5 2417
c4d9986f
NTND
2418 co = find_cached_object(sha1);
2419 if (co) {
9a490590
JH
2420 if (oi->sizep)
2421 *(oi->sizep) = co->size;
161f00e7
JK
2422 if (oi->disk_sizep)
2423 *(oi->disk_sizep) = 0;
9a490590 2424 oi->whence = OI_CACHED;
c4d9986f
NTND
2425 return co->type;
2426 }
2427
cd673c1f 2428 if (!find_pack_entry(sha1, &e)) {
ddd63e64 2429 /* Most likely it's a loose object. */
052fe5ea
JK
2430 if (!sha1_loose_object_info(sha1, &type,
2431 oi->sizep, oi->disk_sizep)) {
9a490590 2432 oi->whence = OI_LOOSE;
f2f57e31 2433 return type;
9a490590 2434 }
ddd63e64
SG
2435
2436 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 2437 reprepare_packed_git();
cd673c1f 2438 if (!find_pack_entry(sha1, &e))
052fe5ea 2439 return -1;
f0df4ed5 2440 }
3d77d877 2441
f2f57e31
JK
2442 type = packed_object_info(e.p, e.offset, oi->sizep, &rtype,
2443 oi->disk_sizep);
2444 if (type < 0) {
3d77d877 2445 mark_bad_packed_object(e.p, sha1);
f2f57e31 2446 type = sha1_object_info_extended(sha1, oi);
5266d369
JH
2447 } else if (in_delta_base_cache(e.p, e.offset)) {
2448 oi->whence = OI_DBCACHED;
9a490590
JH
2449 } else {
2450 oi->whence = OI_PACKED;
2451 oi->u.packed.offset = e.offset;
2452 oi->u.packed.pack = e.p;
2453 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2454 rtype == OBJ_OFS_DELTA);
3d77d877
NP
2455 }
2456
f2f57e31 2457 return type;
f0df4ed5
JS
2458}
2459
9a490590
JH
2460int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2461{
7c07385d 2462 struct object_info oi = {0};
9a490590
JH
2463
2464 oi.sizep = sizep;
2465 return sha1_object_info_extended(sha1, &oi);
2466}
2467
21666f1a
NP
2468static void *read_packed_sha1(const unsigned char *sha1,
2469 enum object_type *type, unsigned long *size)
1f688557
JH
2470{
2471 struct pack_entry e;
8eca0b47 2472 void *data;
1f688557 2473
cd673c1f 2474 if (!find_pack_entry(sha1, &e))
1f688557 2475 return NULL;
8eca0b47
NP
2476 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2477 if (!data) {
2478 /*
2479 * We're probably in deep shit, but let's try to fetch
2480 * the required object anyway from another pack or loose.
2481 * This should happen only in the presence of a corrupted
2482 * pack, and is better than failing outright.
2483 */
2484 error("failed to read object %s at offset %"PRIuMAX" from %s",
2485 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2486 mark_bad_packed_object(e.p, sha1);
ac939109 2487 data = read_object(sha1, type, size);
8eca0b47
NP
2488 }
2489 return data;
1f688557
JH
2490}
2491
21666f1a
NP
2492int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2493 unsigned char *sha1)
d66b37bb
JH
2494{
2495 struct cached_object *co;
2496
21666f1a 2497 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2498 if (has_sha1_file(sha1) || find_cached_object(sha1))
2499 return 0;
2500 if (cached_object_alloc <= cached_object_nr) {
2501 cached_object_alloc = alloc_nr(cached_object_alloc);
2502 cached_objects = xrealloc(cached_objects,
2503 sizeof(*cached_objects) *
2504 cached_object_alloc);
2505 }
2506 co = &cached_objects[cached_object_nr++];
2507 co->size = len;
21666f1a 2508 co->type = type;
efa13f7b
JH
2509 co->buf = xmalloc(len);
2510 memcpy(co->buf, buf, len);
d66b37bb
JH
2511 hashcpy(co->sha1, sha1);
2512 return 0;
2513}
2514
c2c5b270
CC
2515static void *read_object(const unsigned char *sha1, enum object_type *type,
2516 unsigned long *size)
0fcfd160
LT
2517{
2518 unsigned long mapsize;
2519 void *map, *buf;
d66b37bb
JH
2520 struct cached_object *co;
2521
2522 co = find_cached_object(sha1);
2523 if (co) {
21666f1a 2524 *type = co->type;
d66b37bb 2525 *size = co->size;
182af834 2526 return xmemdupz(co->buf, co->size);
d66b37bb 2527 }
0fcfd160 2528
8276c007
PE
2529 buf = read_packed_sha1(sha1, type, size);
2530 if (buf)
2531 return buf;
bb6b8e4f 2532 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2533 if (map) {
7efbff75 2534 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2535 munmap(map, mapsize);
2536 return buf;
2537 }
637cdd9d 2538 reprepare_packed_git();
8276c007 2539 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2540}
2541
b6c4cecc
JH
2542/*
2543 * This function dies on corrupt objects; the callers who want to
2544 * deal with them should arrange to call read_object() and give error
2545 * messages themselves.
2546 */
5bf29b95
JH
2547void *read_sha1_file_extended(const unsigned char *sha1,
2548 enum object_type *type,
2549 unsigned long *size,
2550 unsigned flag)
ac939109 2551{
3ba7a065 2552 void *data;
e8b15e61 2553 char *path;
b6c4cecc 2554 const struct packed_git *p;
5bf29b95
JH
2555 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2556 ? lookup_replace_object(sha1) : sha1;
b6c4cecc 2557
3ba7a065
JH
2558 errno = 0;
2559 data = read_object(repl, type, size);
4bbf5a26 2560 if (data)
b6c4cecc 2561 return data;
68095570 2562
25f3af3f 2563 if (errno && errno != ENOENT)
3ba7a065
JH
2564 die_errno("failed to read object %s", sha1_to_hex(sha1));
2565
68095570 2566 /* die if we replaced an object with one that does not exist */
b6c4cecc 2567 if (repl != sha1)
68095570
CC
2568 die("replacement %s not found for %s",
2569 sha1_to_hex(repl), sha1_to_hex(sha1));
2570
b6c4cecc
JH
2571 if (has_loose_object(repl)) {
2572 path = sha1_file_name(sha1);
2573 die("loose object %s (stored in %s) is corrupt",
2574 sha1_to_hex(repl), path);
e8b15e61 2575 }
68095570 2576
b6c4cecc
JH
2577 if ((p = has_packed_and_bad(repl)) != NULL)
2578 die("packed object %s (stored in %s) is corrupt",
2579 sha1_to_hex(repl), p->pack_name);
f5552aee 2580
b6c4cecc 2581 return NULL;
ac939109
NP
2582}
2583
40469ee9 2584void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2585 const char *required_type_name,
40469ee9
JH
2586 unsigned long *size,
2587 unsigned char *actual_sha1_return)
f4913f91 2588{
21666f1a 2589 enum object_type type, required_type;
f4913f91
JH
2590 void *buffer;
2591 unsigned long isize;
40469ee9 2592 unsigned char actual_sha1[20];
f4913f91 2593
21666f1a 2594 required_type = type_from_string(required_type_name);
e702496e 2595 hashcpy(actual_sha1, sha1);
40469ee9
JH
2596 while (1) {
2597 int ref_length = -1;
2598 const char *ref_type = NULL;
f4913f91 2599
21666f1a 2600 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2601 if (!buffer)
2602 return NULL;
21666f1a 2603 if (type == required_type) {
40469ee9
JH
2604 *size = isize;
2605 if (actual_sha1_return)
e702496e 2606 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2607 return buffer;
2608 }
2609 /* Handle references */
21666f1a 2610 else if (type == OBJ_COMMIT)
40469ee9 2611 ref_type = "tree ";
21666f1a 2612 else if (type == OBJ_TAG)
40469ee9
JH
2613 ref_type = "object ";
2614 else {
2615 free(buffer);
2616 return NULL;
2617 }
2618 ref_length = strlen(ref_type);
f4913f91 2619
50974ec9
MK
2620 if (ref_length + 40 > isize ||
2621 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2622 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2623 free(buffer);
2624 return NULL;
2625 }
1cf58e72 2626 free(buffer);
40469ee9
JH
2627 /* Now we have the ID of the referred-to object in
2628 * actual_sha1. Check again. */
f4913f91 2629 }
f4913f91
JH
2630}
2631
ce9fbf16 2632static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2633 const char *type, unsigned char *sha1,
d65a16f6 2634 char *hdr, int *hdrlen)
d410c0f5 2635{
9126f009 2636 git_SHA_CTX c;
d410c0f5
JH
2637
2638 /* Generate the header */
d65a16f6 2639 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
2640
2641 /* Sha1.. */
9126f009
NP
2642 git_SHA1_Init(&c);
2643 git_SHA1_Update(&c, hdr, *hdrlen);
2644 git_SHA1_Update(&c, buf, len);
2645 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2646}
2647
230f1322 2648/*
5a688fe4
JH
2649 * Move the just written object into its final resting place.
2650 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2651 * "moving" is only a part of what it does, when no patch between
2652 * master to pu changes the call sites of this function.
230f1322 2653 */
839837b9 2654int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322 2655{
e32c0a9c 2656 int ret = 0;
5a688fe4 2657
348df166 2658 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
2659 goto try_rename;
2660 else if (link(tmpfile, filename))
e32c0a9c 2661 ret = errno;
7ebb6fca
LT
2662
2663 /*
2664 * Coda hack - coda doesn't like cross-directory links,
2665 * so we fall back to a rename, which will mean that it
2666 * won't be able to check collisions, but that's not a
2667 * big deal.
2668 *
2669 * The same holds for FAT formatted media.
2670 *
3be1f18e 2671 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
2672 * left to unlink.
2673 */
2674 if (ret && ret != EEXIST) {
be66a6c4 2675 try_rename:
7ebb6fca 2676 if (!rename(tmpfile, filename))
3be1f18e 2677 goto out;
9e48b389 2678 ret = errno;
230f1322 2679 }
691f1a28 2680 unlink_or_warn(tmpfile);
230f1322
LT
2681 if (ret) {
2682 if (ret != EEXIST) {
82247e9b 2683 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
230f1322
LT
2684 }
2685 /* FIXME!!! Collision check here ? */
2686 }
2687
3be1f18e 2688out:
5256b006 2689 if (adjust_shared_perm(filename))
5a688fe4 2690 return error("unable to set permission to '%s'", filename);
230f1322
LT
2691 return 0;
2692}
2693
4d548150
LT
2694static int write_buffer(int fd, const void *buf, size_t len)
2695{
d34cf19b 2696 if (write_in_full(fd, buf, len) < 0)
93822c22 2697 return error("file write error (%s)", strerror(errno));
4d548150
LT
2698 return 0;
2699}
2700
ce9fbf16 2701int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2702 unsigned char *sha1)
2703{
d65a16f6 2704 char hdr[32];
abdc3fc8
RS
2705 int hdrlen;
2706 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2707 return 0;
2708}
2709
e9039dd3
LT
2710/* Finalize a file on disk, and close it. */
2711static void close_sha1_file(int fd)
2712{
aafe9fba
LT
2713 if (fsync_object_files)
2714 fsync_or_die(fd, "sha1 file");
e9039dd3 2715 if (close(fd) != 0)
d824cbba 2716 die_errno("error when closing sha1 file");
e9039dd3
LT
2717}
2718
5723fe7e
LT
2719/* Size of directory component, including the ending '/' */
2720static inline int directory_size(const char *filename)
2721{
2722 const char *s = strrchr(filename, '/');
2723 if (!s)
2724 return 0;
2725 return s - filename + 1;
2726}
2727
2728/*
2729 * This creates a temporary file in the same directory as the final
2730 * 'filename'
2731 *
2732 * We want to avoid cross-directory filename renames, because those
2733 * can have problems on various filesystems (FAT, NFS, Coda).
2734 */
2735static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2736{
2737 int fd, dirlen = directory_size(filename);
2738
2739 if (dirlen + 20 > bufsiz) {
2740 errno = ENAMETOOLONG;
2741 return -1;
2742 }
2743 memcpy(buffer, filename, dirlen);
2744 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
5256b006 2745 fd = git_mkstemp_mode(buffer, 0444);
cbacbf4e 2746 if (fd < 0 && dirlen && errno == ENOENT) {
5723fe7e 2747 /* Make sure the directory exists */
6ff6af62 2748 memcpy(buffer, filename, dirlen);
5723fe7e 2749 buffer[dirlen-1] = 0;
3bfaf018 2750 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
5723fe7e
LT
2751 return -1;
2752
2753 /* Try again */
2754 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
5256b006 2755 fd = git_mkstemp_mode(buffer, 0444);
5723fe7e
LT
2756 }
2757 return fd;
2758}
2759
bbac7311 2760static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 2761 const void *buf, unsigned long len, time_t mtime)
0fcfd160 2762{
915308b1 2763 int fd, ret;
9892beba 2764 unsigned char compressed[4096];
ef49a7a0 2765 git_zstream stream;
748af44c
NP
2766 git_SHA_CTX c;
2767 unsigned char parano_sha1[20];
706bc531 2768 char *filename;
ab1900a3 2769 static char tmp_file[PATH_MAX];
a44c9a5e 2770
972a9155 2771 filename = sha1_file_name(sha1);
ab1900a3 2772 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
aac17941 2773 if (fd < 0) {
35243577 2774 if (errno == EACCES)
82247e9b 2775 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
916d081b 2776 else
5eaeda70 2777 return error("unable to create temporary file: %s", strerror(errno));
aac17941
LT
2778 }
2779
0fcfd160
LT
2780 /* Set it up */
2781 memset(&stream, 0, sizeof(stream));
55bb5c91 2782 git_deflate_init(&stream, zlib_compression_level);
0fcfd160 2783 stream.next_out = compressed;
9892beba 2784 stream.avail_out = sizeof(compressed);
748af44c 2785 git_SHA1_Init(&c);
a44c9a5e
LT
2786
2787 /* First header.. */
d65a16f6 2788 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2789 stream.avail_in = hdrlen;
55bb5c91
JH
2790 while (git_deflate(&stream, 0) == Z_OK)
2791 ; /* nothing */
748af44c 2792 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
2793
2794 /* Then the data itself.. */
c00e657d 2795 stream.next_in = (void *)buf;
a44c9a5e 2796 stream.avail_in = len;
9892beba 2797 do {
748af44c 2798 unsigned char *in0 = stream.next_in;
55bb5c91 2799 ret = git_deflate(&stream, Z_FINISH);
748af44c 2800 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
2801 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2802 die("unable to write sha1 file");
2803 stream.next_out = compressed;
2804 stream.avail_out = sizeof(compressed);
2805 } while (ret == Z_OK);
2806
ac54c277
LT
2807 if (ret != Z_STREAM_END)
2808 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
55bb5c91 2809 ret = git_deflate_end_gently(&stream);
ac54c277
LT
2810 if (ret != Z_OK)
2811 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
2812 git_SHA1_Final(parano_sha1, &c);
2813 if (hashcmp(sha1, parano_sha1) != 0)
2814 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 2815
e9039dd3 2816 close_sha1_file(fd);
0fcfd160 2817
bbac7311
NP
2818 if (mtime) {
2819 struct utimbuf utb;
2820 utb.actime = mtime;
2821 utb.modtime = mtime;
ab1900a3 2822 if (utime(tmp_file, &utb) < 0)
bbac7311 2823 warning("failed utime() on %s: %s",
ab1900a3 2824 tmp_file, strerror(errno));
bbac7311
NP
2825 }
2826
ab1900a3 2827 return move_temp_to_file(tmp_file, filename);
0fcfd160 2828}
8237b185 2829
c00e657d 2830int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
bbac7311
NP
2831{
2832 unsigned char sha1[20];
2833 char hdr[32];
2834 int hdrlen;
2835
2836 /* Normally if we have it in the pack then we do not bother writing
2837 * it out into .git/objects/??/?{38} file.
2838 */
2839 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2840 if (returnsha1)
2841 hashcpy(returnsha1, sha1);
2842 if (has_sha1_file(sha1))
2843 return 0;
2844 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2845}
2846
2847int force_object_loose(const unsigned char *sha1, time_t mtime)
2848{
bbac7311
NP
2849 void *buf;
2850 unsigned long len;
2851 enum object_type type;
2852 char hdr[32];
2853 int hdrlen;
1fb23e65 2854 int ret;
bbac7311 2855
c529d75a 2856 if (has_loose_object(sha1))
bbac7311
NP
2857 return 0;
2858 buf = read_packed_sha1(sha1, &type, &len);
2859 if (!buf)
2860 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2861 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1fb23e65
BS
2862 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2863 free(buf);
2864
2865 return ret;
bbac7311
NP
2866}
2867
bf592c50
DB
2868int has_pack_index(const unsigned char *sha1)
2869{
2870 struct stat st;
2871 if (stat(sha1_pack_index_name(sha1), &st))
2872 return 0;
2873 return 1;
bf592c50
DB
2874}
2875
cd673c1f
JH
2876int has_sha1_pack(const unsigned char *sha1)
2877{
2878 struct pack_entry e;
2879 return find_pack_entry(sha1, &e);
2880}
2881
8237b185
DB
2882int has_sha1_file(const unsigned char *sha1)
2883{
1f688557
JH
2884 struct pack_entry e;
2885
cd673c1f 2886 if (find_pack_entry(sha1, &e))
1f688557 2887 return 1;
c529d75a 2888 return has_loose_object(sha1);
8237b185 2889}
74400e71 2890
c879daa2
NTND
2891static void check_tree(const void *buf, size_t size)
2892{
2893 struct tree_desc desc;
2894 struct name_entry entry;
2895
2896 init_tree_desc(&desc, buf, size);
2897 while (tree_entry(&desc, &entry))
2898 /* do nothing
2899 * tree_entry() will die() on malformed entries */
2900 ;
2901}
2902
2903static void check_commit(const void *buf, size_t size)
2904{
2905 struct commit c;
2906 memset(&c, 0, sizeof(c));
2907 if (parse_commit_buffer(&c, buf, size))
2908 die("corrupt commit");
2909}
2910
2911static void check_tag(const void *buf, size_t size)
2912{
2913 struct tag t;
2914 memset(&t, 0, sizeof(t));
2915 if (parse_tag_buffer(&t, buf, size))
2916 die("corrupt tag");
2917}
2918
43df4f86 2919static int index_mem(unsigned char *sha1, void *buf, size_t size,
c4ce46fc
JH
2920 enum object_type type,
2921 const char *path, unsigned flags)
e7332f96 2922{
6c510bee 2923 int ret, re_allocated = 0;
c4ce46fc 2924 int write_object = flags & HASH_WRITE_OBJECT;
74400e71 2925
7672db20 2926 if (!type)
edaec3fb 2927 type = OBJ_BLOB;
6c510bee
LT
2928
2929 /*
2930 * Convert blobs to git internal format
2931 */
43df4f86 2932 if ((type == OBJ_BLOB) && path) {
f285a2d7 2933 struct strbuf nbuf = STRBUF_INIT;
21e5ad50 2934 if (convert_to_git(path, buf, size, &nbuf,
5e12e78e 2935 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
b315c5c0 2936 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
2937 re_allocated = 1;
2938 }
2939 }
c4ce46fc 2940 if (flags & HASH_FORMAT_CHECK) {
c879daa2
NTND
2941 if (type == OBJ_TREE)
2942 check_tree(buf, size);
2943 if (type == OBJ_COMMIT)
2944 check_commit(buf, size);
2945 if (type == OBJ_TAG)
2946 check_tag(buf, size);
2947 }
6c510bee 2948
7672db20 2949 if (write_object)
edaec3fb 2950 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 2951 else
edaec3fb 2952 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 2953 if (re_allocated)
6c510bee 2954 free(buf);
43df4f86
DP
2955 return ret;
2956}
2957
7b41e1e1
JH
2958static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2959 const char *path, unsigned flags)
2960{
2961 struct strbuf sbuf = STRBUF_INIT;
2962 int ret;
2963
2964 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2965 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2966 else
2967 ret = -1;
2968 strbuf_release(&sbuf);
2969 return ret;
2970}
2971
ea68b0ce
DP
2972#define SMALL_FILE_SIZE (32*1024)
2973
7b41e1e1
JH
2974static int index_core(unsigned char *sha1, int fd, size_t size,
2975 enum object_type type, const char *path,
2976 unsigned flags)
43df4f86
DP
2977{
2978 int ret;
43df4f86 2979
7b41e1e1 2980 if (!size) {
c4ce46fc 2981 ret = index_mem(sha1, NULL, size, type, path, flags);
ea68b0ce
DP
2982 } else if (size <= SMALL_FILE_SIZE) {
2983 char *buf = xmalloc(size);
2984 if (size == read_in_full(fd, buf, size))
c4ce46fc 2985 ret = index_mem(sha1, buf, size, type, path, flags);
ea68b0ce
DP
2986 else
2987 ret = error("short read %s", strerror(errno));
2988 free(buf);
08bda208 2989 } else {
43df4f86 2990 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
c4ce46fc 2991 ret = index_mem(sha1, buf, size, type, path, flags);
aac17941 2992 munmap(buf, size);
08bda208 2993 }
7b41e1e1
JH
2994 return ret;
2995}
2996
4dd1fbc7 2997/*
568508e7
JH
2998 * This creates one packfile per large blob unless bulk-checkin
2999 * machinery is "plugged".
4dd1fbc7
JH
3000 *
3001 * This also bypasses the usual "convert-to-git" dance, and that is on
3002 * purpose. We could write a streaming version of the converting
3003 * functions and insert that before feeding the data to fast-import
4f22b101
JK
3004 * (or equivalent in-core API described above). However, that is
3005 * somewhat complicated, as we do not know the size of the filter
3006 * result, which we need to know beforehand when writing a git object.
3007 * Since the primary motivation for trying to stream from the working
3008 * tree file and to avoid mmaping it in core is to deal with large
3009 * binary blobs, they generally do not want to get any conversion, and
3010 * callers should avoid this code path when filters are requested.
4dd1fbc7
JH
3011 */
3012static int index_stream(unsigned char *sha1, int fd, size_t size,
3013 enum object_type type, const char *path,
3014 unsigned flags)
3015{
568508e7 3016 return index_bulk_checkin(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
3017}
3018
7b41e1e1
JH
3019int index_fd(unsigned char *sha1, int fd, struct stat *st,
3020 enum object_type type, const char *path, unsigned flags)
3021{
3022 int ret;
3023 size_t size = xsize_t(st->st_size);
3024
3025 if (!S_ISREG(st->st_mode))
3026 ret = index_pipe(sha1, fd, type, path, flags);
4f22b101
JK
3027 else if (size <= big_file_threshold || type != OBJ_BLOB ||
3028 (path && would_convert_to_git(path, NULL, 0, 0)))
7b41e1e1 3029 ret = index_core(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
3030 else
3031 ret = index_stream(sha1, fd, size, type, path, flags);
43df4f86 3032 close(fd);
aac17941 3033 return ret;
74400e71 3034}
ec1fcc16 3035
c4ce46fc 3036int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
ec1fcc16
JH
3037{
3038 int fd;
b760d3aa 3039 struct strbuf sb = STRBUF_INIT;
ec1fcc16
JH
3040
3041 switch (st->st_mode & S_IFMT) {
3042 case S_IFREG:
3043 fd = open(path, O_RDONLY);
3044 if (fd < 0)
3045 return error("open(\"%s\"): %s", path,
3046 strerror(errno));
c4ce46fc 3047 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
ec1fcc16
JH
3048 return error("%s: failed to insert into database",
3049 path);
3050 break;
3051 case S_IFLNK:
b760d3aa 3052 if (strbuf_readlink(&sb, path, st->st_size)) {
ec1fcc16 3053 char *errstr = strerror(errno);
ec1fcc16
JH
3054 return error("readlink(\"%s\"): %s", path,
3055 errstr);
3056 }
c4ce46fc 3057 if (!(flags & HASH_WRITE_OBJECT))
b760d3aa
LT
3058 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3059 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
ec1fcc16
JH
3060 return error("%s: failed to insert into database",
3061 path);
b760d3aa 3062 strbuf_release(&sb);
ec1fcc16 3063 break;
f35a6d3b
LT
3064 case S_IFDIR:
3065 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
3066 default:
3067 return error("%s: unsupported file type", path);
3068 }
3069 return 0;
3070}
a69e5429
JH
3071
3072int read_pack_header(int fd, struct pack_header *header)
3073{
c697ad14
HO
3074 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3075 /* "eof before pack header was fully read" */
3076 return PH_ERROR_EOF;
3077
a69e5429
JH
3078 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3079 /* "protocol error (pack signature mismatch detected)" */
3080 return PH_ERROR_PACK_SIGNATURE;
3081 if (!pack_version_ok(header->hdr_version))
3082 /* "protocol error (pack version unsupported)" */
3083 return PH_ERROR_PROTOCOL;
3084 return 0;
3085}
40d52ff7
JK
3086
3087void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3088{
3089 enum object_type type = sha1_object_info(sha1, NULL);
3090 if (type < 0)
3091 die("%s is not a valid object", sha1_to_hex(sha1));
3092 if (type != expect)
3093 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3094 typename(expect));
3095}