replace_object: use struct members instead of an array
[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
b12ca963
NTND
39static const char *no_log_pack_access = "no_log_pack_access";
40static const char *log_pack_access;
41
c597ba80
NTND
42/*
43 * This is meant to hold a *small* number of objects that you would
44 * want read_sha1_file() to be able to return, but yet you do not want
45 * to write them into the object store (e.g. a browse-only
46 * application).
47 */
48static struct cached_object {
49 unsigned char sha1[20];
50 enum object_type type;
51 void *buf;
52 unsigned long size;
53} *cached_objects;
54static int cached_object_nr, cached_object_alloc;
55
56static struct cached_object empty_tree = {
dab0d410 57 EMPTY_TREE_SHA1_BIN_LITERAL,
c597ba80
NTND
58 OBJ_TREE,
59 "",
60 0
61};
62
c01f51cc
NTND
63static struct packed_git *last_found_pack;
64
c597ba80
NTND
65static struct cached_object *find_cached_object(const unsigned char *sha1)
66{
67 int i;
68 struct cached_object *co = cached_objects;
69
70 for (i = 0; i < cached_object_nr; i++, co++) {
71 if (!hashcmp(co->sha1, sha1))
72 return co;
73 }
74 if (!hashcmp(sha1, empty_tree.sha1))
75 return &empty_tree;
76 return NULL;
77}
78
90a6464b
JH
79int mkdir_in_gitdir(const char *path)
80{
81 if (mkdir(path, 0777)) {
82 int saved_errno = errno;
83 struct stat st;
84 struct strbuf sb = STRBUF_INIT;
85
86 if (errno != EEXIST)
87 return -1;
88 /*
89 * Are we looking at a path in a symlinked worktree
90 * whose original repository does not yet have it?
91 * e.g. .git/rr-cache pointing at its original
92 * repository in which the user hasn't performed any
93 * conflict resolution yet?
94 */
95 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
96 strbuf_readlink(&sb, path, st.st_size) ||
97 !is_absolute_path(sb.buf) ||
98 mkdir(sb.buf, 0777)) {
99 strbuf_release(&sb);
100 errno = saved_errno;
101 return -1;
102 }
103 strbuf_release(&sb);
104 }
105 return adjust_shared_perm(path);
106}
107
0be0521b 108enum scld_error safe_create_leading_directories(char *path)
b2cb9425 109{
26c8ae2a 110 char *next_component = path + offset_1st_component(path);
0be0521b 111 enum scld_error ret = SCLD_OK;
67d42212 112
0be0521b 113 while (ret == SCLD_OK && next_component) {
f0502332 114 struct stat st;
0f527403 115 char *slash = next_component, slash_character;
f0502332 116
0f527403
MH
117 while (*slash && !is_dir_sep(*slash))
118 slash++;
119
120 if (!*slash)
b2cb9425 121 break;
bf10cf70 122
26c8ae2a 123 next_component = slash + 1;
0f527403 124 while (is_dir_sep(*next_component))
bf10cf70 125 next_component++;
26c8ae2a 126 if (!*next_component)
5f0bdf50 127 break;
831651fd 128
0f527403 129 slash_character = *slash;
831651fd 130 *slash = '\0';
67d42212
JR
131 if (!stat(path, &st)) {
132 /* path exists */
9e6f885d 133 if (!S_ISDIR(st.st_mode))
0be0521b 134 ret = SCLD_EXISTS;
53a39721 135 } else if (mkdir(path, 0777)) {
928734d9 136 if (errno == EEXIST &&
9e6f885d 137 !stat(path, &st) && S_ISDIR(st.st_mode))
928734d9 138 ; /* somebody created it since we checked */
18d37e86
MH
139 else if (errno == ENOENT)
140 /*
141 * Either mkdir() failed because
142 * somebody just pruned the containing
143 * directory, or stat() failed because
144 * the file that was in our way was
145 * just removed. Either way, inform
146 * the caller that it might be worth
147 * trying again:
148 */
149 ret = SCLD_VANISHED;
9e6f885d 150 else
0be0521b 151 ret = SCLD_FAILED;
53a39721 152 } else if (adjust_shared_perm(path)) {
0be0521b 153 ret = SCLD_PERMS;
457f06d6 154 }
0f527403 155 *slash = slash_character;
b2cb9425 156 }
9e6f885d 157 return ret;
b2cb9425 158}
723c31fe 159
0be0521b 160enum scld_error safe_create_leading_directories_const(const char *path)
8e21d63b
JK
161{
162 /* path points to cache entries, so xstrdup before messing with it */
163 char *buf = xstrdup(path);
0be0521b 164 enum scld_error result = safe_create_leading_directories(buf);
8e21d63b
JK
165 free(buf);
166 return result;
167}
168
ace1534d
JH
169static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
170{
171 int i;
172 for (i = 0; i < 20; i++) {
173 static char hex[] = "0123456789abcdef";
174 unsigned int val = sha1[i];
175 char *pos = pathbuf + i*2 + (i > 0);
176 *pos++ = hex[val >> 4];
177 *pos = hex[val & 0xf];
178 }
179}
180
0fcfd160
LT
181/*
182 * NOTE! This returns a statically allocated buffer, so you have to be
790296fd 183 * careful about using it. Do an "xstrdup()" if you need to save the
0fcfd160 184 * filename.
ace1534d
JH
185 *
186 * Also note that this returns the location for creating. Reading
187 * SHA1 file can happen from any alternate directory listed in the
d19938ab 188 * DB_ENVIRONMENT environment variable if it is not found in
ace1534d 189 * the primary object database.
0fcfd160
LT
190 */
191char *sha1_file_name(const unsigned char *sha1)
192{
560fb6a1
JK
193 static char buf[PATH_MAX];
194 const char *objdir;
195 int len;
0fcfd160 196
560fb6a1
JK
197 objdir = get_object_directory();
198 len = strlen(objdir);
199
200 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
201 if (len + 43 > PATH_MAX)
202 die("insanely long object directory %s", objdir);
203 memcpy(buf, objdir, len);
204 buf[len] = '/';
205 buf[len+3] = '/';
206 buf[len+42] = '\0';
207 fill_sha1_path(buf + len + 1, sha1);
208 return buf;
0fcfd160
LT
209}
210
633f43e1 211static char *sha1_get_pack_name(const unsigned char *sha1,
6eec46bd 212 char **name, char **base, const char *which)
bf592c50
DB
213{
214 static const char hex[] = "0123456789abcdef";
633f43e1 215 char *buf;
bf592c50
DB
216 int i;
217
633f43e1 218 if (!*base) {
bf592c50
DB
219 const char *sha1_file_directory = get_object_directory();
220 int len = strlen(sha1_file_directory);
633f43e1 221 *base = xmalloc(len + 60);
6eec46bd
JH
222 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
223 sha1_file_directory, which);
633f43e1 224 *name = *base + len + 11;
bf592c50
DB
225 }
226
633f43e1 227 buf = *name;
bf592c50
DB
228
229 for (i = 0; i < 20; i++) {
230 unsigned int val = *sha1++;
231 *buf++ = hex[val >> 4];
232 *buf++ = hex[val & 0xf];
233 }
a6080a0a 234
633f43e1 235 return *base;
bf592c50
DB
236}
237
633f43e1 238char *sha1_pack_name(const unsigned char *sha1)
bf592c50 239{
633f43e1 240 static char *name, *base;
bf592c50 241
6eec46bd 242 return sha1_get_pack_name(sha1, &name, &base, "pack");
633f43e1 243}
bf592c50 244
633f43e1
HO
245char *sha1_pack_index_name(const unsigned char *sha1)
246{
247 static char *name, *base;
a6080a0a 248
6eec46bd 249 return sha1_get_pack_name(sha1, &name, &base, "idx");
bf592c50
DB
250}
251
d5a63b99
JH
252struct alternate_object_database *alt_odb_list;
253static struct alternate_object_database **alt_odb_tail;
ace1534d 254
f4e51683 255static int git_open_noatime(const char *name);
c2f493a4 256
ddd5d056
JH
257/*
258 * Prepare alternate object database registry.
d5a63b99
JH
259 *
260 * The variable alt_odb_list points at the list of struct
261 * alternate_object_database. The elements on this list come from
262 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
263 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
264 * whose contents is similar to that environment variable but can be
265 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
266 * contains "/the/directory/corresponding/to/.git/objects/...", while
267 * its name points just after the slash at the end of ".git/objects/"
268 * in the example above, and has enough space to hold 40-byte hex
269 * SHA1, an extra slash for the first level indirection, and the
270 * terminating NUL.
ddd5d056 271 */
6eac50d8 272static int link_alt_odb_entry(const char *entry, const char *relative_base, int depth)
ace1534d 273{
1494e038 274 const char *objdir = get_object_directory();
c2f493a4
MW
275 struct alternate_object_database *ent;
276 struct alternate_object_database *alt;
5bdf0a84
HW
277 int pfxlen, entlen;
278 struct strbuf pathbuf = STRBUF_INIT;
d5a63b99 279
85dadc38 280 if (!is_absolute_path(entry) && relative_base) {
5bdf0a84
HW
281 strbuf_addstr(&pathbuf, real_path(relative_base));
282 strbuf_addch(&pathbuf, '/');
c2f493a4 283 }
6eac50d8 284 strbuf_addstr(&pathbuf, entry);
c2f493a4 285
5bdf0a84
HW
286 normalize_path_copy(pathbuf.buf, pathbuf.buf);
287
288 pfxlen = strlen(pathbuf.buf);
289
290 /*
291 * The trailing slash after the directory name is given by
292 * this function at the end. Remove duplicates.
293 */
294 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
295 pfxlen -= 1;
296
297 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
298 ent = xmalloc(sizeof(*ent) + entlen);
299 memcpy(ent->base, pathbuf.buf, pfxlen);
300 strbuf_release(&pathbuf);
c2f493a4
MW
301
302 ent->name = ent->base + pfxlen + 1;
303 ent->base[pfxlen + 3] = '/';
304 ent->base[pfxlen] = ent->base[entlen-1] = 0;
305
306 /* Detect cases where alternate disappeared */
90b4a71c 307 if (!is_directory(ent->base)) {
c2f493a4
MW
308 error("object directory %s does not exist; "
309 "check .git/objects/info/alternates.",
310 ent->base);
311 free(ent);
312 return -1;
313 }
314
315 /* Prevent the common mistake of listing the same
316 * thing twice, or object directory itself.
317 */
318 for (alt = alt_odb_list; alt; alt = alt->next) {
319 if (!memcmp(ent->base, alt->base, pfxlen)) {
320 free(ent);
321 return -1;
322 }
323 }
cb2912c3 324 if (!strcmp(ent->base, objdir)) {
c2f493a4
MW
325 free(ent);
326 return -1;
327 }
328
329 /* add the alternate entry */
330 *alt_odb_tail = ent;
331 alt_odb_tail = &(ent->next);
332 ent->next = NULL;
333
334 /* recursively add alternates */
335 read_info_alternates(ent->base, depth + 1);
336
337 ent->base[pfxlen] = '/';
338
339 return 0;
340}
341
c5950164 342static void link_alt_odb_entries(const char *alt, int len, int sep,
c2f493a4
MW
343 const char *relative_base, int depth)
344{
6eac50d8
MH
345 struct string_list entries = STRING_LIST_INIT_NODUP;
346 char *alt_copy;
347 int i;
c2f493a4
MW
348
349 if (depth > 5) {
350 error("%s: ignoring alternate object stores, nesting too deep.",
351 relative_base);
352 return;
353 }
354
c5950164 355 alt_copy = xmemdupz(alt, len);
6eac50d8
MH
356 string_list_split_in_place(&entries, alt_copy, sep, -1);
357 for (i = 0; i < entries.nr; i++) {
358 const char *entry = entries.items[i].string;
359 if (entry[0] == '\0' || entry[0] == '#')
9577e7e3 360 continue;
6eac50d8
MH
361 if (!is_absolute_path(entry) && depth) {
362 error("%s: ignoring relative alternate object store %s",
363 relative_base, entry);
364 } else {
365 link_alt_odb_entry(entry, relative_base, depth);
9577e7e3 366 }
9577e7e3 367 }
6eac50d8
MH
368 string_list_clear(&entries, 0);
369 free(alt_copy);
d5a63b99
JH
370}
371
5e73633d 372void read_info_alternates(const char * relative_base, int depth)
d5a63b99 373{
9577e7e3 374 char *map;
dc49cd76 375 size_t mapsz;
d5a63b99 376 struct stat st;
9cb18f56
JM
377 const char alt_file_name[] = "info/alternates";
378 /* Given that relative_base is no longer than PATH_MAX,
379 ensure that "path" has enough space to append "/", the
380 file name, "info/alternates", and a trailing NUL. */
381 char path[PATH_MAX + 1 + sizeof alt_file_name];
c2f493a4 382 int fd;
d5a63b99 383
9cb18f56 384 sprintf(path, "%s/%s", relative_base, alt_file_name);
f4e51683 385 fd = git_open_noatime(path);
d5a63b99
JH
386 if (fd < 0)
387 return;
388 if (fstat(fd, &st) || (st.st_size == 0)) {
389 close(fd);
9a217f2a 390 return;
ace1534d 391 }
dc49cd76
SP
392 mapsz = xsize_t(st.st_size);
393 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 394 close(fd);
d5a63b99 395
c5950164 396 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
c2f493a4 397
dc49cd76 398 munmap(map, mapsz);
ace1534d
JH
399}
400
bef70b22
DB
401void add_to_alternates_file(const char *reference)
402{
403 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
acd3b9ec 404 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
e6baf4a1 405 char *alt = mkpath("%s\n", reference);
bef70b22
DB
406 write_or_die(fd, alt, strlen(alt));
407 if (commit_lock_file(lock))
408 die("could not close alternates file");
409 if (alt_odb_tail)
c5950164 410 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
bef70b22
DB
411}
412
d79796bc
JH
413void foreach_alt_odb(alt_odb_fn fn, void *cb)
414{
415 struct alternate_object_database *ent;
416
417 prepare_alt_odb();
418 for (ent = alt_odb_list; ent; ent = ent->next)
419 if (fn(ent, cb))
420 return;
421}
422
c2f493a4
MW
423void prepare_alt_odb(void)
424{
554fe20d 425 const char *alt;
c2f493a4 426
7dc24aa5
SP
427 if (alt_odb_tail)
428 return;
429
c2f493a4
MW
430 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
431 if (!alt) alt = "";
432
c2f493a4 433 alt_odb_tail = &alt_odb_list;
c5950164 434 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
435
436 read_info_alternates(get_object_directory(), 0);
437}
438
0f4dc14a 439static int has_loose_object_local(const unsigned char *sha1)
ace1534d 440{
ace1534d 441 char *name = sha1_file_name(sha1);
0f4dc14a
BC
442 return !access(name, F_OK);
443}
ace1534d 444
0f4dc14a
BC
445int has_loose_object_nonlocal(const unsigned char *sha1)
446{
447 struct alternate_object_database *alt;
9a217f2a 448 prepare_alt_odb();
d5a63b99 449 for (alt = alt_odb_list; alt; alt = alt->next) {
0f4dc14a 450 fill_sha1_path(alt->name, sha1);
c529d75a
LT
451 if (!access(alt->base, F_OK))
452 return 1;
ace1534d 453 }
c529d75a 454 return 0;
ace1534d
JH
455}
456
0f4dc14a
BC
457static int has_loose_object(const unsigned char *sha1)
458{
459 return has_loose_object_local(sha1) ||
460 has_loose_object_nonlocal(sha1);
461}
462
60bb8b14 463static unsigned int pack_used_ctr;
a53128b6
SP
464static unsigned int pack_mmap_calls;
465static unsigned int peak_pack_open_windows;
466static unsigned int pack_open_windows;
c7934306
SP
467static unsigned int pack_open_fds;
468static unsigned int pack_max_fds;
a53128b6 469static size_t peak_pack_mapped;
60bb8b14 470static size_t pack_mapped;
9a217f2a 471struct packed_git *packed_git;
1f688557 472
b79d18c9 473void pack_report(void)
a53128b6
SP
474{
475 fprintf(stderr,
e05db0fd
PR
476 "pack_report: getpagesize() = %10" SZ_FMT "\n"
477 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
478 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
479 sz_fmt(getpagesize()),
480 sz_fmt(packed_git_window_size),
481 sz_fmt(packed_git_limit));
a53128b6
SP
482 fprintf(stderr,
483 "pack_report: pack_used_ctr = %10u\n"
484 "pack_report: pack_mmap_calls = %10u\n"
485 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
486 "pack_report: pack_mapped = "
487 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
488 pack_used_ctr,
489 pack_mmap_calls,
490 pack_open_windows, peak_pack_open_windows,
9e42d6a1 491 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
492}
493
42873078 494static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
495{
496 void *idx_map;
42873078 497 struct pack_idx_header *hdr;
dc49cd76 498 size_t idx_size;
74e34e1f 499 uint32_t version, nr, i, *index;
f4e51683 500 int fd = git_open_noatime(path);
1f688557 501 struct stat st;
42873078 502
1f688557
JH
503 if (fd < 0)
504 return -1;
505 if (fstat(fd, &st)) {
506 close(fd);
507 return -1;
508 }
dc49cd76 509 idx_size = xsize_t(st.st_size);
2d88451b
SP
510 if (idx_size < 4 * 256 + 20 + 20) {
511 close(fd);
512 return error("index file %s is too small", path);
513 }
c4712e45 514 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 515 close(fd);
1f688557 516
42873078
NP
517 hdr = idx_map;
518 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
519 version = ntohl(hdr->idx_version);
520 if (version < 2 || version > 2) {
521 munmap(idx_map, idx_size);
6e1c2344 522 return error("index file %s is version %"PRIu32
74e34e1f
NP
523 " and is not supported by this binary"
524 " (try upgrading GIT to a newer version)",
525 path, version);
526 }
527 } else
528 version = 1;
df1b059d 529
1f688557 530 nr = 0;
42873078 531 index = idx_map;
74e34e1f
NP
532 if (version > 1)
533 index += 2; /* skip index header */
1f688557 534 for (i = 0; i < 256; i++) {
326bf396 535 uint32_t n = ntohl(index[i]);
2d88451b
SP
536 if (n < nr) {
537 munmap(idx_map, idx_size);
df1b059d 538 return error("non-monotonic index %s", path);
2d88451b 539 }
1f688557
JH
540 nr = n;
541 }
542
74e34e1f
NP
543 if (version == 1) {
544 /*
545 * Total size:
546 * - 256 index entries 4 bytes each
547 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
548 * - 20-byte SHA1 of the packfile
549 * - 20-byte SHA1 file checksum
550 */
551 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
552 munmap(idx_map, idx_size);
eef427a0 553 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
554 }
555 } else if (version == 2) {
556 /*
557 * Minimum size:
558 * - 8 bytes of header
559 * - 256 index entries 4 bytes each
560 * - 20-byte sha1 entry * nr
561 * - 4-byte crc entry * nr
562 * - 4-byte offset entry * nr
563 * - 20-byte SHA1 of the packfile
564 * - 20-byte SHA1 file checksum
565 * And after the 4-byte offset table might be a
566 * variable sized table containing 8-byte entries
567 * for offsets larger than 2^31.
568 */
569 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
570 unsigned long max_size = min_size;
571 if (nr)
572 max_size += (nr - 1)*8;
573 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 574 munmap(idx_map, idx_size);
eef427a0 575 return error("wrong index v2 file size in %s", path);
74e34e1f 576 }
7109c889
JH
577 if (idx_size != min_size &&
578 /*
579 * make sure we can deal with large pack offsets.
580 * 31-bit signed offset won't be enough, neither
581 * 32-bit unsigned one will be.
582 */
583 (sizeof(off_t) <= 4)) {
584 munmap(idx_map, idx_size);
585 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 586 }
2d88451b 587 }
1f688557 588
74e34e1f 589 p->index_version = version;
42873078
NP
590 p->index_data = idx_map;
591 p->index_size = idx_size;
57059091 592 p->num_objects = nr;
1f688557
JH
593 return 0;
594}
595
bc8e478a 596int open_pack_index(struct packed_git *p)
d079837e
SP
597{
598 char *idx_name;
599 int ret;
600
601 if (p->index_data)
602 return 0;
603
604 idx_name = xstrdup(p->pack_name);
605 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
606 ret = check_packed_git_idx(idx_name, p);
607 free(idx_name);
608 return ret;
609}
610
11daf39b
SP
611static void scan_windows(struct packed_git *p,
612 struct packed_git **lru_p,
613 struct pack_window **lru_w,
614 struct pack_window **lru_l)
1f688557 615{
11daf39b
SP
616 struct pack_window *w, *w_l;
617
618 for (w_l = NULL, w = p->windows; w; w = w->next) {
619 if (!w->inuse_cnt) {
620 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
621 *lru_p = p;
622 *lru_w = w;
623 *lru_l = w_l;
54044bf8 624 }
54044bf8 625 }
11daf39b 626 w_l = w;
f9253394 627 }
11daf39b
SP
628}
629
7c3ecb32 630static int unuse_one_window(struct packed_git *current)
11daf39b
SP
631{
632 struct packed_git *p, *lru_p = NULL;
633 struct pack_window *lru_w = NULL, *lru_l = NULL;
634
635 if (current)
636 scan_windows(current, &lru_p, &lru_w, &lru_l);
637 for (p = packed_git; p; p = p->next)
638 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
639 if (lru_p) {
640 munmap(lru_w->base, lru_w->len);
641 pack_mapped -= lru_w->len;
642 if (lru_l)
643 lru_l->next = lru_w->next;
7c3ecb32 644 else
54044bf8 645 lru_p->windows = lru_w->next;
54044bf8 646 free(lru_w);
a53128b6 647 pack_open_windows--;
54044bf8
SP
648 return 1;
649 }
650 return 0;
f9253394
JH
651}
652
7c3ecb32 653void release_pack_memory(size_t need)
97bfeb34
SP
654{
655 size_t cur = pack_mapped;
7c3ecb32 656 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
97bfeb34
SP
657 ; /* nothing */
658}
659
58ecbd5e
JN
660void *xmmap(void *start, size_t length,
661 int prot, int flags, int fd, off_t offset)
662{
663 void *ret = mmap(start, length, prot, flags, fd, offset);
664 if (ret == MAP_FAILED) {
665 if (!length)
666 return NULL;
7c3ecb32 667 release_pack_memory(length);
58ecbd5e
JN
668 ret = mmap(start, length, prot, flags, fd, offset);
669 if (ret == MAP_FAILED)
670 die_errno("Out of memory? mmap failed");
671 }
672 return ret;
673}
674
c9ced051
SP
675void close_pack_windows(struct packed_git *p)
676{
677 while (p->windows) {
678 struct pack_window *w = p->windows;
679
680 if (w->inuse_cnt)
681 die("pack '%s' still has open windows to it",
682 p->pack_name);
683 munmap(w->base, w->len);
684 pack_mapped -= w->len;
685 pack_open_windows--;
686 p->windows = w->next;
687 free(w);
688 }
689}
690
88d0db55
BC
691/*
692 * The LRU pack is the one with the oldest MRU window, preferring packs
693 * with no used windows, or the oldest mtime if it has no windows allocated.
694 */
695static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
696{
697 struct pack_window *w, *this_mru_w;
698 int has_windows_inuse = 0;
699
700 /*
701 * Reject this pack if it has windows and the previously selected
702 * one does not. If this pack does not have windows, reject
703 * it if the pack file is newer than the previously selected one.
704 */
705 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
706 return;
707
708 for (w = this_mru_w = p->windows; w; w = w->next) {
709 /*
710 * Reject this pack if any of its windows are in use,
711 * but the previously selected pack did not have any
712 * inuse windows. Otherwise, record that this pack
713 * has windows in use.
714 */
715 if (w->inuse_cnt) {
716 if (*accept_windows_inuse)
717 has_windows_inuse = 1;
718 else
719 return;
720 }
721
722 if (w->last_used > this_mru_w->last_used)
723 this_mru_w = w;
724
725 /*
726 * Reject this pack if it has windows that have been
727 * used more recently than the previously selected pack.
728 * If the previously selected pack had windows inuse and
729 * we have not encountered a window in this pack that is
730 * inuse, skip this check since we prefer a pack with no
731 * inuse windows to one that has inuse windows.
732 */
733 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
734 this_mru_w->last_used > (*mru_w)->last_used)
735 return;
736 }
737
738 /*
739 * Select this pack.
740 */
741 *mru_w = this_mru_w;
742 *lru_p = p;
743 *accept_windows_inuse = has_windows_inuse;
744}
745
746static int close_one_pack(void)
747{
748 struct packed_git *p, *lru_p = NULL;
749 struct pack_window *mru_w = NULL;
750 int accept_windows_inuse = 1;
751
752 for (p = packed_git; p; p = p->next) {
753 if (p->pack_fd == -1)
754 continue;
755 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
756 }
757
758 if (lru_p) {
759 close(lru_p->pack_fd);
760 pack_open_fds--;
761 lru_p->pack_fd = -1;
762 return 1;
763 }
764
765 return 0;
766}
767
03e79c88 768void unuse_pack(struct pack_window **w_cursor)
f9253394 769{
03e79c88
SP
770 struct pack_window *w = *w_cursor;
771 if (w) {
772 w->inuse_cnt--;
773 *w_cursor = NULL;
774 }
1f688557
JH
775}
776
fa5fc15d
SP
777void close_pack_index(struct packed_git *p)
778{
779 if (p->index_data) {
780 munmap((void *)p->index_data, p->index_size);
781 p->index_data = NULL;
782 }
783}
784
c74faea1
NP
785/*
786 * This is used by git-repack in case a newly created pack happens to
787 * contain the same set of objects as an existing one. In that case
788 * the resulting file might be different even if its name would be the
789 * same. It is best to close any reference to the old pack before it is
790 * replaced on disk. Of course no index pointers nor windows for given pack
791 * must subsist at this point. If ever objects from this pack are requested
792 * again, the new version of the pack will be reinitialized through
793 * reprepare_packed_git().
794 */
795void free_pack_by_name(const char *pack_name)
796{
797 struct packed_git *p, **pp = &packed_git;
798
799 while (*pp) {
800 p = *pp;
801 if (strcmp(pack_name, p->pack_name) == 0) {
fa3a0c94 802 clear_delta_base_cache();
c74faea1 803 close_pack_windows(p);
c7934306 804 if (p->pack_fd != -1) {
c74faea1 805 close(p->pack_fd);
c7934306
SP
806 pack_open_fds--;
807 }
fa5fc15d 808 close_pack_index(p);
c74faea1
NP
809 free(p->bad_object_sha1);
810 *pp = p->next;
c01f51cc
NTND
811 if (last_found_pack == p)
812 last_found_pack = NULL;
c74faea1
NP
813 free(p);
814 return;
815 }
816 pp = &p->next;
817 }
818}
819
a0788266
JS
820static unsigned int get_max_fd_limit(void)
821{
822#ifdef RLIMIT_NOFILE
491a8dec
JH
823 {
824 struct rlimit lim;
a0788266 825
491a8dec
JH
826 if (!getrlimit(RLIMIT_NOFILE, &lim))
827 return lim.rlim_cur;
828 }
829#endif
830
831#ifdef _SC_OPEN_MAX
832 {
833 long open_max = sysconf(_SC_OPEN_MAX);
834 if (0 < open_max)
835 return open_max;
836 /*
837 * Otherwise, we got -1 for one of the two
838 * reasons:
839 *
840 * (1) sysconf() did not understand _SC_OPEN_MAX
841 * and signaled an error with -1; or
842 * (2) sysconf() said there is no limit.
843 *
844 * We _could_ clear errno before calling sysconf() to
845 * tell these two cases apart and return a huge number
846 * in the latter case to let the caller cap it to a
847 * value that is not so selfish, but letting the
848 * fallback OPEN_MAX codepath take care of these cases
849 * is a lot simpler.
850 */
851 }
852#endif
a0788266 853
491a8dec 854#ifdef OPEN_MAX
a0788266
JS
855 return OPEN_MAX;
856#else
857 return 1; /* see the caller ;-) */
858#endif
859}
860
3cf8b462
SP
861/*
862 * Do not call this directly as this leaks p->pack_fd on error return;
863 * call open_packed_git() instead.
864 */
865static int open_packed_git_1(struct packed_git *p)
1f688557 866{
9bc879c1
SP
867 struct stat st;
868 struct pack_header hdr;
869 unsigned char sha1[20];
870 unsigned char *idx_sha1;
2c039da8 871 long fd_flag;
9bc879c1 872
d079837e
SP
873 if (!p->index_data && open_pack_index(p))
874 return error("packfile %s index unavailable", p->pack_name);
875
c7934306 876 if (!pack_max_fds) {
a0788266 877 unsigned int max_fds = get_max_fd_limit();
c7934306
SP
878
879 /* Save 3 for stdin/stdout/stderr, 22 for work */
880 if (25 < max_fds)
881 pack_max_fds = max_fds - 25;
882 else
883 pack_max_fds = 1;
884 }
885
88d0db55 886 while (pack_max_fds <= pack_open_fds && close_one_pack())
c7934306
SP
887 ; /* nothing */
888
f4e51683 889 p->pack_fd = git_open_noatime(p->pack_name);
9bc879c1 890 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 891 return -1;
c7934306 892 pack_open_fds++;
9bc879c1
SP
893
894 /* If we created the struct before we had the pack we lack size. */
bf592c50 895 if (!p->pack_size) {
bf592c50 896 if (!S_ISREG(st.st_mode))
072db278 897 return error("packfile %s not a regular file", p->pack_name);
bf592c50 898 p->pack_size = st.st_size;
9bc879c1 899 } else if (p->pack_size != st.st_size)
072db278 900 return error("packfile %s size changed", p->pack_name);
9bc879c1 901
2c039da8
JH
902 /* We leave these file descriptors open with sliding mmap;
903 * there is no point keeping them open across exec(), though.
904 */
905 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
906 if (fd_flag < 0)
072db278 907 return error("cannot determine file descriptor flags");
2c039da8
JH
908 fd_flag |= FD_CLOEXEC;
909 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 910 return error("cannot set FD_CLOEXEC");
2c039da8 911
9bc879c1 912 /* Verify we recognize this pack file format. */
e6e2bd62 913 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 914 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 915 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 916 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 917 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
918 return error("packfile %s is version %"PRIu32" and not"
919 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
920 p->pack_name, ntohl(hdr.hdr_version));
921
922 /* Verify the pack matches its index. */
57059091 923 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
924 return error("packfile %s claims to have %"PRIu32" objects"
925 " while index indicates %"PRIu32" objects",
57059091
NP
926 p->pack_name, ntohl(hdr.hdr_entries),
927 p->num_objects);
9bc879c1 928 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 929 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 930 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 931 return error("packfile %s signature is unavailable", p->pack_name);
42873078 932 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 933 if (hashcmp(sha1, idx_sha1))
072db278
SP
934 return error("packfile %s does not match index", p->pack_name);
935 return 0;
9bc879c1
SP
936}
937
3cf8b462
SP
938static int open_packed_git(struct packed_git *p)
939{
940 if (!open_packed_git_1(p))
941 return 0;
942 if (p->pack_fd != -1) {
943 close(p->pack_fd);
c7934306 944 pack_open_fds--;
3cf8b462
SP
945 p->pack_fd = -1;
946 }
947 return -1;
948}
949
c4001d92 950static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
951{
952 /* We must promise at least 20 bytes (one hash) after the
953 * offset is available from this window, otherwise the offset
954 * is not actually in this window and a different window (which
955 * has that one hash excess) must be used. This is to support
956 * the object header and delta base parsing routines below.
957 */
958 off_t win_off = win->offset;
959 return win_off <= offset
960 && (offset + 20) <= (win_off + win->len);
961}
962
4b25d091 963unsigned char *use_pack(struct packed_git *p,
03e79c88 964 struct pack_window **w_cursor,
c4001d92 965 off_t offset,
ef49a7a0 966 unsigned long *left)
9bc879c1 967{
60bb8b14 968 struct pack_window *win = *w_cursor;
03e79c88 969
a9d98a14 970 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
971 * pointless to ask for an offset into the middle of that
972 * hash, and the in_window function above wouldn't match
973 * don't allow an offset too close to the end of the file.
974 */
d131b7af
SP
975 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
976 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
977 if (offset > (p->pack_size - 20))
978 die("offset beyond end of packfile (truncated pack?)");
979
980 if (!win || !in_window(win, offset)) {
981 if (win)
982 win->inuse_cnt--;
983 for (win = p->windows; win; win = win->next) {
984 if (in_window(win, offset))
985 break;
986 }
987 if (!win) {
78a28df9 988 size_t window_align = packed_git_window_size / 2;
dc49cd76 989 off_t len;
d131b7af
SP
990
991 if (p->pack_fd == -1 && open_packed_git(p))
992 die("packfile %s cannot be accessed", p->pack_name);
993
60bb8b14 994 win = xcalloc(1, sizeof(*win));
78a28df9 995 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
996 len = p->pack_size - win->offset;
997 if (len > packed_git_window_size)
998 len = packed_git_window_size;
999 win->len = (size_t)len;
60bb8b14 1000 pack_mapped += win->len;
11daf39b 1001 while (packed_git_limit < pack_mapped
7c3ecb32 1002 && unuse_one_window(p))
60bb8b14 1003 ; /* nothing */
c4712e45 1004 win->base = xmmap(NULL, win->len,
60bb8b14
SP
1005 PROT_READ, MAP_PRIVATE,
1006 p->pack_fd, win->offset);
1007 if (win->base == MAP_FAILED)
73b4e4be
SP
1008 die("packfile %s cannot be mapped: %s",
1009 p->pack_name,
1010 strerror(errno));
d131b7af
SP
1011 if (!win->offset && win->len == p->pack_size
1012 && !p->do_not_close) {
1013 close(p->pack_fd);
1014 pack_open_fds--;
1015 p->pack_fd = -1;
1016 }
a53128b6
SP
1017 pack_mmap_calls++;
1018 pack_open_windows++;
1019 if (pack_mapped > peak_pack_mapped)
1020 peak_pack_mapped = pack_mapped;
1021 if (pack_open_windows > peak_pack_open_windows)
1022 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
1023 win->next = p->windows;
1024 p->windows = win;
1025 }
1f688557 1026 }
03e79c88
SP
1027 if (win != *w_cursor) {
1028 win->last_used = pack_used_ctr++;
1029 win->inuse_cnt++;
1030 *w_cursor = win;
1031 }
60bb8b14 1032 offset -= win->offset;
03e79c88 1033 if (left)
dc49cd76 1034 *left = win->len - xsize_t(offset);
03e79c88 1035 return win->base + offset;
1f688557
JH
1036}
1037
27d69a46
NP
1038static struct packed_git *alloc_packed_git(int extra)
1039{
1040 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1041 memset(p, 0, sizeof(*p));
1042 p->pack_fd = -1;
1043 return p;
1044}
1045
e0500293
JN
1046static void try_to_free_pack_memory(size_t size)
1047{
7c3ecb32 1048 release_pack_memory(size);
e0500293
JN
1049}
1050
42873078 1051struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557 1052{
e0500293 1053 static int have_set_try_to_free_routine;
1f688557 1054 struct stat st;
27d69a46 1055 struct packed_git *p = alloc_packed_git(path_len + 2);
1f688557 1056
e0500293
JN
1057 if (!have_set_try_to_free_routine) {
1058 have_set_try_to_free_routine = 1;
1059 set_try_to_free_routine(try_to_free_pack_memory);
1060 }
1061
42873078
NP
1062 /*
1063 * Make sure a corresponding .pack file exists and that
1064 * the index looks sane.
1065 */
1066 path_len -= strlen(".idx");
27d69a46
NP
1067 if (path_len < 1) {
1068 free(p);
1f688557 1069 return NULL;
27d69a46 1070 }
42873078 1071 memcpy(p->pack_name, path, path_len);
8d25931d
BC
1072
1073 strcpy(p->pack_name + path_len, ".keep");
1074 if (!access(p->pack_name, F_OK))
1075 p->pack_keep = 1;
1076
42873078 1077 strcpy(p->pack_name + path_len, ".pack");
d079837e 1078 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 1079 free(p);
1f688557
JH
1080 return NULL;
1081 }
42873078 1082
1f688557
JH
1083 /* ok, it looks sane as far as we can check without
1084 * actually mapping the pack file.
1085 */
1f688557 1086 p->pack_size = st.st_size;
9d835df2 1087 p->pack_local = local;
b867092f 1088 p->mtime = st.st_mtime;
42873078
NP
1089 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1090 hashclr(p->sha1);
1f688557
JH
1091 return p;
1092}
1093
7b64469a 1094struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 1095{
42873078 1096 const char *path = sha1_pack_name(sha1);
27d69a46 1097 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
bf592c50 1098
27d69a46
NP
1099 strcpy(p->pack_name, path);
1100 hashcpy(p->sha1, sha1);
42873078
NP
1101 if (check_packed_git_idx(idx_path, p)) {
1102 free(p);
bf592c50 1103 return NULL;
42873078 1104 }
bf592c50 1105
bf592c50
DB
1106 return p;
1107}
1108
1109void install_packed_git(struct packed_git *pack)
1110{
c7934306
SP
1111 if (pack->pack_fd != -1)
1112 pack_open_fds++;
1113
bf592c50
DB
1114 pack->next = packed_git;
1115 packed_git = pack;
1116}
1117
543c5caa
NTND
1118void (*report_garbage)(const char *desc, const char *path);
1119
1120static void report_helper(const struct string_list *list,
1121 int seen_bits, int first, int last)
1122{
1123 const char *msg;
1124 switch (seen_bits) {
1125 case 0:
1126 msg = "no corresponding .idx nor .pack";
1127 break;
1128 case 1:
1129 msg = "no corresponding .idx";
1130 break;
1131 case 2:
1132 msg = "no corresponding .pack";
1133 break;
1134 default:
1135 return;
1136 }
1137 for (; first < last; first++)
1138 report_garbage(msg, list->items[first].string);
1139}
1140
1141static void report_pack_garbage(struct string_list *list)
1142{
1143 int i, baselen = -1, first = 0, seen_bits = 0;
1144
1145 if (!report_garbage)
1146 return;
1147
1148 sort_string_list(list);
1149
1150 for (i = 0; i < list->nr; i++) {
1151 const char *path = list->items[i].string;
1152 if (baselen != -1 &&
1153 strncmp(path, list->items[first].string, baselen)) {
1154 report_helper(list, seen_bits, first, i);
1155 baselen = -1;
1156 seen_bits = 0;
1157 }
1158 if (baselen == -1) {
1159 const char *dot = strrchr(path, '.');
1160 if (!dot) {
1161 report_garbage("garbage found", path);
1162 continue;
1163 }
1164 baselen = dot - path + 1;
1165 first = i;
1166 }
1167 if (!strcmp(path + baselen, "pack"))
1168 seen_bits |= 1;
1169 else if (!strcmp(path + baselen, "idx"))
1170 seen_bits |= 2;
1171 }
1172 report_helper(list, seen_bits, first, list->nr);
1173}
1174
9d835df2 1175static void prepare_packed_git_one(char *objdir, int local)
1f688557 1176{
9cb18f56
JM
1177 /* Ensure that this buffer is large enough so that we can
1178 append "/pack/" without clobbering the stack even if
1179 strlen(objdir) were PATH_MAX. */
1180 char path[PATH_MAX + 1 + 4 + 1 + 1];
1f688557
JH
1181 int len;
1182 DIR *dir;
1183 struct dirent *de;
543c5caa 1184 struct string_list garbage = STRING_LIST_INIT_DUP;
1f688557
JH
1185
1186 sprintf(path, "%s/pack", objdir);
1187 len = strlen(path);
1188 dir = opendir(path);
b5b16990 1189 if (!dir) {
26125f6b 1190 if (errno != ENOENT)
bd2afde8 1191 error("unable to open object pack directory: %s: %s",
26125f6b 1192 path, strerror(errno));
1f688557 1193 return;
b5b16990 1194 }
1f688557
JH
1195 path[len++] = '/';
1196 while ((de = readdir(dir)) != NULL) {
1197 int namelen = strlen(de->d_name);
1198 struct packed_git *p;
1199
543c5caa
NTND
1200 if (len + namelen + 1 > sizeof(path)) {
1201 if (report_garbage) {
1202 struct strbuf sb = STRBUF_INIT;
1203 strbuf_addf(&sb, "%.*s/%s", len - 1, path, de->d_name);
1204 report_garbage("path too long", sb.buf);
1205 strbuf_release(&sb);
1206 }
1207 continue;
1208 }
1209
1210 if (is_dot_or_dotdot(de->d_name))
9cb18f56
JM
1211 continue;
1212
1f688557 1213 strcpy(path + len, de->d_name);
d90906a9
NTND
1214
1215 if (has_extension(de->d_name, ".idx")) {
1216 /* Don't reopen a pack we already have. */
1217 for (p = packed_git; p; p = p->next) {
1218 if (!memcmp(path, p->pack_name, len + namelen - 4))
1219 break;
1220 }
1221 if (p == NULL &&
1222 /*
1223 * See if it really is a valid .idx file with
1224 * corresponding .pack file that we can map.
1225 */
1226 (p = add_packed_git(path, len + namelen, local)) != NULL)
1227 install_packed_git(p);
86f7780c 1228 }
543c5caa
NTND
1229
1230 if (!report_garbage)
1231 continue;
1232
1233 if (has_extension(de->d_name, ".idx") ||
1234 has_extension(de->d_name, ".pack") ||
1235 has_extension(de->d_name, ".keep"))
1236 string_list_append(&garbage, path);
1237 else
1238 report_garbage("garbage found", path);
1f688557 1239 }
5b35bcd5 1240 closedir(dir);
543c5caa
NTND
1241 report_pack_garbage(&garbage);
1242 string_list_clear(&garbage, 0);
1f688557
JH
1243}
1244
b867092f
JH
1245static int sort_pack(const void *a_, const void *b_)
1246{
1247 struct packed_git *a = *((struct packed_git **)a_);
1248 struct packed_git *b = *((struct packed_git **)b_);
1249 int st;
1250
1251 /*
1252 * Local packs tend to contain objects specific to our
1253 * variant of the project than remote ones. In addition,
1254 * remote ones could be on a network mounted filesystem.
1255 * Favor local ones for these reasons.
1256 */
1257 st = a->pack_local - b->pack_local;
1258 if (st)
1259 return -st;
1260
1261 /*
1262 * Younger packs tend to contain more recent objects,
1263 * and more recent objects tend to get accessed more
1264 * often.
1265 */
1266 if (a->mtime < b->mtime)
1267 return 1;
1268 else if (a->mtime == b->mtime)
1269 return 0;
1270 return -1;
1271}
1272
1273static void rearrange_packed_git(void)
1274{
1275 struct packed_git **ary, *p;
1276 int i, n;
1277
1278 for (n = 0, p = packed_git; p; p = p->next)
1279 n++;
1280 if (n < 2)
1281 return;
1282
1283 /* prepare an array of packed_git for easier sorting */
1284 ary = xcalloc(n, sizeof(struct packed_git *));
1285 for (n = 0, p = packed_git; p; p = p->next)
1286 ary[n++] = p;
1287
1288 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1289
1290 /* link them back again */
1291 for (i = 0; i < n - 1; i++)
1292 ary[i]->next = ary[i + 1];
1293 ary[n - 1]->next = NULL;
1294 packed_git = ary[0];
1295
1296 free(ary);
1297}
1298
637cdd9d 1299static int prepare_packed_git_run_once = 0;
9a217f2a 1300void prepare_packed_git(void)
1f688557 1301{
d5a63b99 1302 struct alternate_object_database *alt;
1f688557 1303
637cdd9d 1304 if (prepare_packed_git_run_once)
1f688557 1305 return;
9d835df2 1306 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 1307 prepare_alt_odb();
d5a63b99 1308 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 1309 alt->name[-1] = 0;
9d835df2 1310 prepare_packed_git_one(alt->base, 0);
1494e038 1311 alt->name[-1] = '/';
1f688557 1312 }
b867092f 1313 rearrange_packed_git();
637cdd9d
JK
1314 prepare_packed_git_run_once = 1;
1315}
1316
fc04c412 1317void reprepare_packed_git(void)
637cdd9d 1318{
4b480c67 1319 discard_revindex();
637cdd9d
JK
1320 prepare_packed_git_run_once = 0;
1321 prepare_packed_git();
1f688557
JH
1322}
1323
8eca0b47
NP
1324static void mark_bad_packed_object(struct packed_git *p,
1325 const unsigned char *sha1)
1326{
1327 unsigned i;
1328 for (i = 0; i < p->num_bad_objects; i++)
1329 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1330 return;
1331 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1332 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1333 p->num_bad_objects++;
1334}
1335
b6c4cecc 1336static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
ac939109
NP
1337{
1338 struct packed_git *p;
1339 unsigned i;
1340
1341 for (p = packed_git; p; p = p->next)
1342 for (i = 0; i < p->num_bad_objects; i++)
1343 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
b6c4cecc
JH
1344 return p;
1345 return NULL;
ac939109
NP
1346}
1347
090ea126
NTND
1348/*
1349 * With an in-core object data in "map", rehash it to make sure the
1350 * object name actually matches "sha1" to detect object corruption.
1351 * With "map" == NULL, try reading the object named with "sha1" using
1352 * the streaming interface and rehash it to do the same.
1353 */
1354int check_sha1_signature(const unsigned char *sha1, void *map,
1355 unsigned long size, const char *type)
0fcfd160
LT
1356{
1357 unsigned char real_sha1[20];
090ea126
NTND
1358 enum object_type obj_type;
1359 struct git_istream *st;
1360 git_SHA_CTX c;
1361 char hdr[32];
1362 int hdrlen;
1363
1364 if (map) {
1365 hash_sha1_file(map, size, type, real_sha1);
1366 return hashcmp(sha1, real_sha1) ? -1 : 0;
1367 }
1368
1369 st = open_istream(sha1, &obj_type, &size, NULL);
1370 if (!st)
1371 return -1;
1372
1373 /* Generate the header */
1374 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1375
1376 /* Sha1.. */
1377 git_SHA1_Init(&c);
1378 git_SHA1_Update(&c, hdr, hdrlen);
1379 for (;;) {
1380 char buf[1024 * 16];
1381 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1382
f54fac53
JK
1383 if (readlen < 0) {
1384 close_istream(st);
1385 return -1;
1386 }
090ea126
NTND
1387 if (!readlen)
1388 break;
1389 git_SHA1_Update(&c, buf, readlen);
1390 }
1391 git_SHA1_Final(real_sha1, &c);
1392 close_istream(st);
a89fccd2 1393 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1394}
1395
f4e51683 1396static int git_open_noatime(const char *name)
44d1c19e
LT
1397{
1398 static int sha1_file_open_flag = O_NOATIME;
44d1c19e 1399
f2e872aa
SP
1400 for (;;) {
1401 int fd = open(name, O_RDONLY | sha1_file_open_flag);
44d1c19e 1402 if (fd >= 0)
f2e872aa
SP
1403 return fd;
1404
f2e872aa
SP
1405 /* Might the failure be due to O_NOATIME? */
1406 if (errno != ENOENT && sha1_file_open_flag) {
44d1c19e 1407 sha1_file_open_flag = 0;
f2e872aa
SP
1408 continue;
1409 }
1410
1411 return -1;
44d1c19e 1412 }
44d1c19e
LT
1413}
1414
052fe5ea
JK
1415static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1416{
1417 char *name = sha1_file_name(sha1);
1418 struct alternate_object_database *alt;
1419
1420 if (!lstat(name, st))
1421 return 0;
1422
1423 prepare_alt_odb();
1424 errno = ENOENT;
1425 for (alt = alt_odb_list; alt; alt = alt->next) {
1426 name = alt->name;
1427 fill_sha1_path(name, sha1);
1428 if (!lstat(alt->base, st))
1429 return 0;
1430 }
1431
1432 return -1;
1433}
1434
44d1c19e
LT
1435static int open_sha1_file(const unsigned char *sha1)
1436{
1437 int fd;
1438 char *name = sha1_file_name(sha1);
1439 struct alternate_object_database *alt;
1440
f4e51683 1441 fd = git_open_noatime(name);
44d1c19e
LT
1442 if (fd >= 0)
1443 return fd;
1444
1445 prepare_alt_odb();
1446 errno = ENOENT;
1447 for (alt = alt_odb_list; alt; alt = alt->next) {
1448 name = alt->name;
1449 fill_sha1_path(name, sha1);
f4e51683 1450 fd = git_open_noatime(alt->base);
44d1c19e
LT
1451 if (fd >= 0)
1452 return fd;
1453 }
1454 return -1;
1455}
1456
f0270efd 1457void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1458{
0fcfd160 1459 void *map;
144bde78 1460 int fd;
ace1534d 1461
44d1c19e
LT
1462 fd = open_sha1_file(sha1);
1463 map = NULL;
1464 if (fd >= 0) {
1465 struct stat st;
0fcfd160 1466
44d1c19e
LT
1467 if (!fstat(fd, &st)) {
1468 *size = xsize_t(st.st_size);
33e42de0
MM
1469 if (!*size) {
1470 /* mmap() is forbidden on empty files */
1471 error("object file %s is empty", sha1_file_name(sha1));
1472 return NULL;
1473 }
44d1c19e 1474 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1475 }
44d1c19e 1476 close(fd);
0fcfd160 1477 }
0fcfd160
LT
1478 return map;
1479}
1480
09ded04b
NP
1481unsigned long unpack_object_header_buffer(const unsigned char *buf,
1482 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1483{
ad1ed5ee 1484 unsigned shift;
48fb7deb 1485 unsigned long size, c;
ad1ed5ee
JH
1486 unsigned long used = 0;
1487
1488 c = buf[used++];
1489 *type = (c >> 4) & 7;
1490 size = c & 15;
1491 shift = 4;
1492 while (c & 0x80) {
f630cfda 1493 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1494 error("bad object header");
ea4f9685
JH
1495 size = used = 0;
1496 break;
09ded04b 1497 }
ad1ed5ee
JH
1498 c = buf[used++];
1499 size += (c & 0x7f) << shift;
1500 shift += 7;
1501 }
1502 *sizep = size;
1503 return used;
1504}
1505
eb4f4076 1506int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
ad1ed5ee 1507{
c4483576
LT
1508 /* Get the data stream */
1509 memset(stream, 0, sizeof(*stream));
1510 stream->next_in = map;
1511 stream->avail_in = mapsize;
1512 stream->next_out = buffer;
93821bd9
LT
1513 stream->avail_out = bufsiz;
1514
39c68542 1515 git_inflate_init(stream);
cc5c54e7 1516 return git_inflate(stream, 0);
c4483576
LT
1517}
1518
ef49a7a0 1519static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1520{
1521 int bytes = strlen(buffer) + 1;
3aee68aa 1522 unsigned char *buf = xmallocz(size);
93821bd9 1523 unsigned long n;
7efbff75 1524 int status = Z_OK;
5180cacc 1525
93821bd9
LT
1526 n = stream->total_out - bytes;
1527 if (n > size)
1528 n = size;
1529 memcpy(buf, (char *) buffer + bytes, n);
1530 bytes = n;
456cdf6e
LT
1531 if (bytes <= size) {
1532 /*
1533 * The above condition must be (bytes <= size), not
1534 * (bytes < size). In other words, even though we
ccf5ace0 1535 * expect no more output and set avail_out to zero,
456cdf6e
LT
1536 * the input zlib stream may have bytes that express
1537 * "this concludes the stream", and we *do* want to
1538 * eat that input.
1539 *
1540 * Otherwise we would not be able to test that we
1541 * consumed all the input to reach the expected size;
1542 * we also want to check that zlib tells us that all
1543 * went well with status == Z_STREAM_END at the end.
1544 */
5180cacc
LT
1545 stream->next_out = buf + bytes;
1546 stream->avail_out = size - bytes;
7efbff75 1547 while (status == Z_OK)
39c68542 1548 status = git_inflate(stream, Z_FINISH);
5180cacc 1549 }
456cdf6e 1550 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1551 git_inflate_end(stream);
7efbff75
JH
1552 return buf;
1553 }
1554
1555 if (status < 0)
1556 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1557 else if (stream->avail_in)
1558 error("garbage at end of loose object '%s'",
1559 sha1_to_hex(sha1));
1560 free(buf);
1561 return NULL;
5180cacc
LT
1562}
1563
1564/*
1565 * We used to just use "sscanf()", but that's actually way
1566 * too permissive for what we want to check. So do an anal
1567 * object header parse by hand.
1568 */
f0270efd 1569int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1570{
21666f1a 1571 char type[10];
5180cacc
LT
1572 int i;
1573 unsigned long size;
1574
1575 /*
a6080a0a 1576 * The type can be at most ten bytes (including the
5180cacc 1577 * terminating '\0' that we add), and is followed by
21666f1a 1578 * a space.
5180cacc 1579 */
21666f1a 1580 i = 0;
5180cacc
LT
1581 for (;;) {
1582 char c = *hdr++;
1583 if (c == ' ')
1584 break;
21666f1a
NP
1585 type[i++] = c;
1586 if (i >= sizeof(type))
5180cacc 1587 return -1;
5180cacc 1588 }
21666f1a 1589 type[i] = 0;
5180cacc
LT
1590
1591 /*
1592 * The length must follow immediately, and be in canonical
1593 * decimal format (ie "010" is not valid).
1594 */
1595 size = *hdr++ - '0';
1596 if (size > 9)
1597 return -1;
1598 if (size) {
1599 for (;;) {
1600 unsigned long c = *hdr - '0';
1601 if (c > 9)
1602 break;
1603 hdr++;
1604 size = size * 10 + c;
1605 }
1606 }
1607 *sizep = size;
1608
1609 /*
1610 * The length must be followed by a zero byte
1611 */
21666f1a 1612 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1613}
1614
7efbff75 1615static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1616{
5180cacc 1617 int ret;
ef49a7a0 1618 git_zstream stream;
5180cacc 1619 char hdr[8192];
0fcfd160 1620
5180cacc 1621 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1622 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1623 return NULL;
1624
7efbff75 1625 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1626}
1627
54dab52a
NP
1628unsigned long get_size_from_delta(struct packed_git *p,
1629 struct pack_window **w_curs,
1630 off_t curpos)
1631{
1632 const unsigned char *data;
1633 unsigned char delta_head[20], *in;
ef49a7a0 1634 git_zstream stream;
54dab52a
NP
1635 int st;
1636
1637 memset(&stream, 0, sizeof(stream));
1638 stream.next_out = delta_head;
1639 stream.avail_out = sizeof(delta_head);
1640
39c68542 1641 git_inflate_init(&stream);
54dab52a
NP
1642 do {
1643 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1644 stream.next_in = in;
39c68542 1645 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1646 curpos += stream.next_in - in;
1647 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1648 stream.total_out < sizeof(delta_head));
39c68542 1649 git_inflate_end(&stream);
3d77d877
NP
1650 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1651 error("delta data unpack-initial failed");
1652 return 0;
1653 }
54dab52a
NP
1654
1655 /* Examine the initial part of the delta to figure out
1656 * the result size.
1657 */
1658 data = delta_head;
1659
1660 /* ignore base size */
1661 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1662
1663 /* Read the result size */
1664 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1665}
1666
c4001d92 1667static off_t get_delta_base(struct packed_git *p,
03e79c88 1668 struct pack_window **w_curs,
c4001d92 1669 off_t *curpos,
21666f1a 1670 enum object_type type,
c4001d92 1671 off_t delta_obj_offset)
eb32d236 1672{
2b87c45b 1673 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1674 off_t base_offset;
eb32d236 1675
8d8a4ea5
SP
1676 /* use_pack() assured us we have [base_info, base_info + 20)
1677 * as a range that we can look at without walking off the
1678 * end of the mapped window. Its actually the hash size
1679 * that is assured. An OFS_DELTA longer than the hash size
1680 * is stupid, as then a REF_DELTA would be smaller to store.
1681 */
21666f1a 1682 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1683 unsigned used = 0;
1684 unsigned char c = base_info[used++];
1685 base_offset = c & 127;
1686 while (c & 128) {
1687 base_offset += 1;
8723f216 1688 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1689 return 0; /* overflow */
eb32d236
NP
1690 c = base_info[used++];
1691 base_offset = (base_offset << 7) + (c & 127);
1692 }
1693 base_offset = delta_obj_offset - base_offset;
d8f32556 1694 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1695 return 0; /* out of bound */
2b87c45b 1696 *curpos += used;
21666f1a 1697 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1698 /* The base entry _must_ be in the same pack */
1699 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1700 *curpos += 20;
eb32d236
NP
1701 } else
1702 die("I am totally screwed");
2b87c45b 1703 return base_offset;
eb32d236
NP
1704}
1705
5d642e75
JK
1706/*
1707 * Like get_delta_base above, but we return the sha1 instead of the pack
1708 * offset. This means it is cheaper for REF deltas (we do not have to do
1709 * the final object lookup), but more expensive for OFS deltas (we
1710 * have to load the revidx to convert the offset back into a sha1).
1711 */
1712static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1713 struct pack_window **w_curs,
1714 off_t curpos,
1715 enum object_type type,
1716 off_t delta_obj_offset)
1717{
1718 if (type == OBJ_REF_DELTA) {
1719 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1720 return base;
1721 } else if (type == OBJ_OFS_DELTA) {
1722 struct revindex_entry *revidx;
1723 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1724 type, delta_obj_offset);
1725
1726 if (!base_offset)
1727 return NULL;
1728
1729 revidx = find_pack_revindex(p, base_offset);
1730 if (!revidx)
1731 return NULL;
1732
1733 return nth_packed_object_sha1(p, revidx->nr);
1734 } else
1735 return NULL;
1736}
1737
f8c8abc5
JH
1738int unpack_object_header(struct packed_git *p,
1739 struct pack_window **w_curs,
1740 off_t *curpos,
1741 unsigned long *sizep)
a733cb60 1742{
03e79c88 1743 unsigned char *base;
ef49a7a0 1744 unsigned long left;
ad1ed5ee 1745 unsigned long used;
2b87c45b 1746 enum object_type type;
a733cb60 1747
8d8a4ea5 1748 /* use_pack() assures us we have [base, base + 20) available
0353a0c4 1749 * as a range that we can look at. (Its actually the hash
3dff5379 1750 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1751 * the maximum deflated object size is 2^137, which is just
1752 * insane, so we know won't exceed what we have been given.
1753 */
2b87c45b 1754 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1755 used = unpack_object_header_buffer(base, left, &type, sizep);
1756 if (!used) {
1757 type = OBJ_BAD;
1758 } else
1759 *curpos += used;
ad1ed5ee 1760
2b87c45b 1761 return type;
a733cb60
LT
1762}
1763
790d96c0
TR
1764static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1765{
1766 int type;
1767 struct revindex_entry *revidx;
1768 const unsigned char *sha1;
1769 revidx = find_pack_revindex(p, obj_offset);
1770 if (!revidx)
1771 return OBJ_BAD;
1772 sha1 = nth_packed_object_sha1(p, revidx->nr);
1773 mark_bad_packed_object(p, sha1);
1774 type = sha1_object_info(sha1, NULL);
1775 if (type <= OBJ_NONE)
1776 return OBJ_BAD;
1777 return type;
1778}
1779
790d96c0
TR
1780#define POI_STACK_PREALLOC 64
1781
90191d37
JK
1782static enum object_type packed_to_object_type(struct packed_git *p,
1783 off_t obj_offset,
1784 enum object_type type,
1785 struct pack_window **w_curs,
1786 off_t curpos)
1f688557 1787{
790d96c0
TR
1788 off_t small_poi_stack[POI_STACK_PREALLOC];
1789 off_t *poi_stack = small_poi_stack;
1790 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
5db47c2b 1791
790d96c0
TR
1792 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1793 off_t base_offset;
90191d37 1794 unsigned long size;
790d96c0
TR
1795 /* Push the object we're going to leave behind */
1796 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1797 poi_stack_alloc = alloc_nr(poi_stack_nr);
1798 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1799 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1800 } else {
1801 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1802 }
1803 poi_stack[poi_stack_nr++] = obj_offset;
1804 /* If parsing the base offset fails, just unwind */
90191d37 1805 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
790d96c0
TR
1806 if (!base_offset)
1807 goto unwind;
1808 curpos = obj_offset = base_offset;
90191d37 1809 type = unpack_object_header(p, w_curs, &curpos, &size);
790d96c0
TR
1810 if (type <= OBJ_NONE) {
1811 /* If getting the base itself fails, we first
1812 * retry the base, otherwise unwind */
1813 type = retry_bad_packed_offset(p, base_offset);
1814 if (type > OBJ_NONE)
1815 goto out;
1816 goto unwind;
1817 }
1818 }
1819
21666f1a 1820 switch (type) {
790d96c0 1821 case OBJ_BAD:
a733cb60 1822 case OBJ_COMMIT:
a733cb60 1823 case OBJ_TREE:
a733cb60 1824 case OBJ_BLOB:
a733cb60 1825 case OBJ_TAG:
a69d0943 1826 break;
1f688557 1827 default:
3d77d877
NP
1828 error("unknown object type %i at offset %"PRIuMAX" in %s",
1829 type, (uintmax_t)obj_offset, p->pack_name);
1830 type = OBJ_BAD;
1f688557 1831 }
790d96c0
TR
1832
1833out:
1834 if (poi_stack != small_poi_stack)
1835 free(poi_stack);
21666f1a 1836 return type;
790d96c0
TR
1837
1838unwind:
1839 while (poi_stack_nr) {
1840 obj_offset = poi_stack[--poi_stack_nr];
1841 type = retry_bad_packed_offset(p, obj_offset);
1842 if (type > OBJ_NONE)
1843 goto out;
1844 }
1845 type = OBJ_BAD;
1846 goto out;
1f688557
JH
1847}
1848
90191d37 1849static int packed_object_info(struct packed_git *p, off_t obj_offset,
23c339c0 1850 struct object_info *oi)
90191d37
JK
1851{
1852 struct pack_window *w_curs = NULL;
1853 unsigned long size;
1854 off_t curpos = obj_offset;
1855 enum object_type type;
1856
412916ee
JK
1857 /*
1858 * We always get the representation type, but only convert it to
1859 * a "real" type later if the caller is interested.
1860 */
90191d37
JK
1861 type = unpack_object_header(p, &w_curs, &curpos, &size);
1862
23c339c0 1863 if (oi->sizep) {
90191d37
JK
1864 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1865 off_t tmp_pos = curpos;
1866 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1867 type, obj_offset);
1868 if (!base_offset) {
1869 type = OBJ_BAD;
1870 goto out;
1871 }
23c339c0
JK
1872 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1873 if (*oi->sizep == 0) {
90191d37
JK
1874 type = OBJ_BAD;
1875 goto out;
1876 }
1877 } else {
23c339c0 1878 *oi->sizep = size;
90191d37
JK
1879 }
1880 }
1881
23c339c0 1882 if (oi->disk_sizep) {
90191d37 1883 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
23c339c0 1884 *oi->disk_sizep = revidx[1].offset - obj_offset;
90191d37
JK
1885 }
1886
23c339c0
JK
1887 if (oi->typep) {
1888 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
1889 if (*oi->typep < 0) {
412916ee
JK
1890 type = OBJ_BAD;
1891 goto out;
1892 }
1893 }
90191d37 1894
5d642e75
JK
1895 if (oi->delta_base_sha1) {
1896 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1897 const unsigned char *base;
1898
1899 base = get_delta_base_sha1(p, &w_curs, curpos,
1900 type, obj_offset);
1901 if (!base) {
1902 type = OBJ_BAD;
1903 goto out;
1904 }
1905
1906 hashcpy(oi->delta_base_sha1, base);
1907 } else
1908 hashclr(oi->delta_base_sha1);
1909 }
1910
90191d37
JK
1911out:
1912 unuse_pack(&w_curs);
1913 return type;
1914}
1915
eb950c19 1916static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1917 struct pack_window **w_curs,
c4001d92 1918 off_t curpos,
eb950c19 1919 unsigned long size)
de530aaa
SP
1920{
1921 int st;
ef49a7a0 1922 git_zstream stream;
079afb18 1923 unsigned char *buffer, *in;
de530aaa 1924
4ab07e4d 1925 buffer = xmallocz(size);
de530aaa 1926 memset(&stream, 0, sizeof(stream));
de530aaa 1927 stream.next_out = buffer;
39eea7bd 1928 stream.avail_out = size + 1;
de530aaa 1929
39c68542 1930 git_inflate_init(&stream);
079afb18 1931 do {
2b87c45b 1932 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 1933 stream.next_in = in;
39c68542 1934 st = git_inflate(&stream, Z_FINISH);
39eea7bd
JH
1935 if (!stream.avail_out)
1936 break; /* the payload is larger than it should be */
2b87c45b 1937 curpos += stream.next_in - in;
079afb18 1938 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 1939 git_inflate_end(&stream);
de530aaa
SP
1940 if ((st != Z_STREAM_END) || stream.total_out != size) {
1941 free(buffer);
1942 return NULL;
1943 }
1944
1945 return buffer;
1946}
1947
e5e01619
LT
1948#define MAX_DELTA_CACHE (256)
1949
18bdec11 1950static size_t delta_base_cached;
5e08ecbf
NP
1951
1952static struct delta_base_cache_lru_list {
1953 struct delta_base_cache_lru_list *prev;
1954 struct delta_base_cache_lru_list *next;
1955} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1956
e5e01619 1957static struct delta_base_cache_entry {
5e08ecbf
NP
1958 struct delta_base_cache_lru_list lru;
1959 void *data;
e5e01619
LT
1960 struct packed_git *p;
1961 off_t base_offset;
1962 unsigned long size;
e5e01619
LT
1963 enum object_type type;
1964} delta_base_cache[MAX_DELTA_CACHE];
1965
1966static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1967{
1968 unsigned long hash;
1969
1970 hash = (unsigned long)p + (unsigned long)base_offset;
1971 hash += (hash >> 8) + (hash >> 16);
3358004a 1972 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1973}
1974
84dd81c1
TR
1975static struct delta_base_cache_entry *
1976get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
5266d369
JH
1977{
1978 unsigned long hash = pack_entry_hash(p, base_offset);
84dd81c1
TR
1979 return delta_base_cache + hash;
1980}
1981
1982static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
1983 struct packed_git *p, off_t base_offset)
1984{
5266d369
JH
1985 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1986}
1987
84dd81c1
TR
1988static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1989{
1990 struct delta_base_cache_entry *ent;
1991 ent = get_delta_base_cache_entry(p, base_offset);
1992 return eq_delta_base_cache_entry(ent, p, base_offset);
1993}
1994
1995static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1996{
1997 ent->data = NULL;
1998 ent->lru.next->prev = ent->lru.prev;
1999 ent->lru.prev->next = ent->lru.next;
2000 delta_base_cached -= ent->size;
2001}
2002
62f255ad 2003static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 2004 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 2005{
84dd81c1 2006 struct delta_base_cache_entry *ent;
e5e01619 2007 void *ret;
e5e01619 2008
84dd81c1
TR
2009 ent = get_delta_base_cache_entry(p, base_offset);
2010
2011 if (!eq_delta_base_cache_entry(ent, p, base_offset))
749bc58c 2012 return unpack_entry(p, base_offset, type, base_size);
e5e01619 2013
84dd81c1
TR
2014 ret = ent->data;
2015
2016 if (!keep_cache)
2017 clear_delta_base_cache_entry(ent);
2018 else
182af834 2019 ret = xmemdupz(ent->data, ent->size);
e5e01619
LT
2020 *type = ent->type;
2021 *base_size = ent->size;
2022 return ret;
62f255ad
LT
2023}
2024
18bdec11
SP
2025static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2026{
2027 if (ent->data) {
2028 free(ent->data);
2029 ent->data = NULL;
5e08ecbf
NP
2030 ent->lru.next->prev = ent->lru.prev;
2031 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
2032 delta_base_cached -= ent->size;
2033 }
2034}
2035
3d20c636
SP
2036void clear_delta_base_cache(void)
2037{
2038 unsigned long p;
2039 for (p = 0; p < MAX_DELTA_CACHE; p++)
2040 release_delta_base_cache(&delta_base_cache[p]);
2041}
2042
62f255ad
LT
2043static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2044 void *base, unsigned long base_size, enum object_type type)
2045{
5e08ecbf 2046 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 2047 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 2048 struct delta_base_cache_lru_list *lru;
e5e01619 2049
18bdec11
SP
2050 release_delta_base_cache(ent);
2051 delta_base_cached += base_size;
5e08ecbf
NP
2052
2053 for (lru = delta_base_cache_lru.next;
2054 delta_base_cached > delta_base_cache_limit
2055 && lru != &delta_base_cache_lru;
2056 lru = lru->next) {
2057 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
2058 if (f->type == OBJ_BLOB)
2059 release_delta_base_cache(f);
2060 }
5e08ecbf
NP
2061 for (lru = delta_base_cache_lru.next;
2062 delta_base_cached > delta_base_cache_limit
2063 && lru != &delta_base_cache_lru;
2064 lru = lru->next) {
2065 struct delta_base_cache_entry *f = (void *)lru;
2066 release_delta_base_cache(f);
2067 }
18bdec11 2068
e5e01619
LT
2069 ent->p = p;
2070 ent->base_offset = base_offset;
2071 ent->type = type;
2072 ent->data = base;
2073 ent->size = base_size;
5e08ecbf
NP
2074 ent->lru.next = &delta_base_cache_lru;
2075 ent->lru.prev = delta_base_cache_lru.prev;
2076 delta_base_cache_lru.prev->next = &ent->lru;
2077 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
2078}
2079
c2c5b270
CC
2080static void *read_object(const unsigned char *sha1, enum object_type *type,
2081 unsigned long *size);
2082
5f44324d
JH
2083static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2084{
2085 static FILE *log_file;
2086
b12ca963
NTND
2087 if (!log_pack_access)
2088 log_pack_access = getenv("GIT_TRACE_PACK_ACCESS");
2089 if (!log_pack_access)
2090 log_pack_access = no_log_pack_access;
2091 if (log_pack_access == no_log_pack_access)
2092 return;
2093
5f44324d
JH
2094 if (!log_file) {
2095 log_file = fopen(log_pack_access, "w");
2096 if (!log_file) {
2097 error("cannot open pack access log '%s' for writing: %s",
2098 log_pack_access, strerror(errno));
b12ca963 2099 log_pack_access = no_log_pack_access;
5f44324d
JH
2100 return;
2101 }
2102 }
2103 fprintf(log_file, "%s %"PRIuMAX"\n",
2104 p->pack_name, (uintmax_t)obj_offset);
2105 fflush(log_file);
2106}
2107
0e8189e2
NP
2108int do_check_packed_object_crc;
2109
abe601bb
TR
2110#define UNPACK_ENTRY_STACK_PREALLOC 64
2111struct unpack_entry_stack_ent {
2112 off_t obj_offset;
2113 off_t curpos;
2114 unsigned long size;
2115};
2116
c4001d92 2117void *unpack_entry(struct packed_git *p, off_t obj_offset,
abe601bb 2118 enum object_type *final_type, unsigned long *final_size)
1f688557 2119{
03e79c88 2120 struct pack_window *w_curs = NULL;
c4001d92 2121 off_t curpos = obj_offset;
abe601bb
TR
2122 void *data = NULL;
2123 unsigned long size;
2124 enum object_type type;
2125 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2126 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2127 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2128 int base_from_cache = 0;
1f688557 2129
b12ca963 2130 if (log_pack_access != no_log_pack_access)
5f44324d
JH
2131 write_pack_access_log(p, obj_offset);
2132
abe601bb
TR
2133 /* PHASE 1: drill down to the innermost base object */
2134 for (;;) {
2135 off_t base_offset;
2136 int i;
2137 struct delta_base_cache_entry *ent;
2138
77965f8b
NTND
2139 ent = get_delta_base_cache_entry(p, curpos);
2140 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2141 type = ent->type;
2142 data = ent->data;
2143 size = ent->size;
2144 clear_delta_base_cache_entry(ent);
2145 base_from_cache = 1;
2146 break;
2147 }
2148
abe601bb
TR
2149 if (do_check_packed_object_crc && p->index_version > 1) {
2150 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2151 unsigned long len = revidx[1].offset - obj_offset;
2152 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2153 const unsigned char *sha1 =
2154 nth_packed_object_sha1(p, revidx->nr);
2155 error("bad packed object CRC for %s",
2156 sha1_to_hex(sha1));
2157 mark_bad_packed_object(p, sha1);
2158 unuse_pack(&w_curs);
2159 return NULL;
2160 }
2161 }
2162
abe601bb
TR
2163 type = unpack_object_header(p, &w_curs, &curpos, &size);
2164 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2165 break;
2166
2167 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2168 if (!base_offset) {
2169 error("failed to validate delta base reference "
2170 "at offset %"PRIuMAX" from %s",
2171 (uintmax_t)curpos, p->pack_name);
2172 /* bail to phase 2, in hopes of recovery */
2173 data = NULL;
2174 break;
2175 }
2176
2177 /* push object, proceed to base */
2178 if (delta_stack_nr >= delta_stack_alloc
2179 && delta_stack == small_delta_stack) {
2180 delta_stack_alloc = alloc_nr(delta_stack_nr);
2181 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2182 memcpy(delta_stack, small_delta_stack,
2183 sizeof(*delta_stack)*delta_stack_nr);
2184 } else {
2185 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2186 }
2187 i = delta_stack_nr++;
2188 delta_stack[i].obj_offset = obj_offset;
2189 delta_stack[i].curpos = curpos;
2190 delta_stack[i].size = size;
2191
2192 curpos = obj_offset = base_offset;
0e8189e2
NP
2193 }
2194
abe601bb
TR
2195 /* PHASE 2: handle the base */
2196 switch (type) {
eb32d236
NP
2197 case OBJ_OFS_DELTA:
2198 case OBJ_REF_DELTA:
abe601bb
TR
2199 if (data)
2200 die("BUG in unpack_entry: left loop at a valid delta");
4d703a1a 2201 break;
a733cb60 2202 case OBJ_COMMIT:
a733cb60 2203 case OBJ_TREE:
a733cb60 2204 case OBJ_BLOB:
a733cb60 2205 case OBJ_TAG:
abe601bb
TR
2206 if (!base_from_cache)
2207 data = unpack_compressed_entry(p, &w_curs, curpos, size);
4d703a1a 2208 break;
1f688557 2209 default:
8eca0b47
NP
2210 data = NULL;
2211 error("unknown object type %i at offset %"PRIuMAX" in %s",
abe601bb
TR
2212 type, (uintmax_t)obj_offset, p->pack_name);
2213 }
2214
2215 /* PHASE 3: apply deltas in order */
2216
2217 /* invariants:
2218 * 'data' holds the base data, or NULL if there was corruption
2219 */
2220 while (delta_stack_nr) {
2221 void *delta_data;
2222 void *base = data;
2223 unsigned long delta_size, base_size = size;
2224 int i;
2225
2226 data = NULL;
2227
2228 if (base)
2229 add_delta_base_cache(p, obj_offset, base, base_size, type);
2230
2231 if (!base) {
2232 /*
2233 * We're probably in deep shit, but let's try to fetch
2234 * the required base anyway from another pack or loose.
2235 * This is costly but should happen only in the presence
2236 * of a corrupted pack, and is better than failing outright.
2237 */
2238 struct revindex_entry *revidx;
2239 const unsigned char *base_sha1;
2240 revidx = find_pack_revindex(p, obj_offset);
2241 if (revidx) {
2242 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2243 error("failed to read delta base object %s"
2244 " at offset %"PRIuMAX" from %s",
2245 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2246 p->pack_name);
2247 mark_bad_packed_object(p, base_sha1);
2248 base = read_object(base_sha1, &type, &base_size);
2249 }
2250 }
2251
2252 i = --delta_stack_nr;
2253 obj_offset = delta_stack[i].obj_offset;
2254 curpos = delta_stack[i].curpos;
2255 delta_size = delta_stack[i].size;
2256
2257 if (!base)
2258 continue;
2259
2260 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2261
2262 if (!delta_data) {
2263 error("failed to unpack compressed delta "
2264 "at offset %"PRIuMAX" from %s",
2265 (uintmax_t)curpos, p->pack_name);
abe601bb
TR
2266 data = NULL;
2267 continue;
2268 }
2269
2270 data = patch_delta(base, base_size,
2271 delta_data, delta_size,
2272 &size);
1ee886c1
JK
2273
2274 /*
2275 * We could not apply the delta; warn the user, but keep going.
2276 * Our failure will be noticed either in the next iteration of
2277 * the loop, or if this is the final delta, in the caller when
2278 * we return NULL. Those code paths will take care of making
2279 * a more explicit warning and retrying with another copy of
2280 * the object.
2281 */
abe601bb 2282 if (!data)
1ee886c1 2283 error("failed to apply delta");
abe601bb 2284
4b8f772c 2285 free(delta_data);
1f688557 2286 }
abe601bb
TR
2287
2288 *final_type = type;
2289 *final_size = size;
2290
03e79c88 2291 unuse_pack(&w_curs);
21666f1a 2292 return data;
1f688557
JH
2293}
2294
d079837e 2295const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 2296 uint32_t n)
9a217f2a 2297{
42873078 2298 const unsigned char *index = p->index_data;
d079837e
SP
2299 if (!index) {
2300 if (open_pack_index(p))
2301 return NULL;
2302 index = p->index_data;
2303 }
57059091 2304 if (n >= p->num_objects)
d72308e0 2305 return NULL;
74e34e1f
NP
2306 index += 4 * 256;
2307 if (p->index_version == 1) {
2308 return index + 24 * n + 4;
2309 } else {
2310 index += 8;
2311 return index + 20 * n;
2312 }
2313}
2314
99093238 2315off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
2316{
2317 const unsigned char *index = p->index_data;
2318 index += 4 * 256;
2319 if (p->index_version == 1) {
2320 return ntohl(*((uint32_t *)(index + 24 * n)));
2321 } else {
2322 uint32_t off;
2323 index += 8 + p->num_objects * (20 + 4);
2324 off = ntohl(*((uint32_t *)(index + 4 * n)));
2325 if (!(off & 0x80000000))
2326 return off;
2327 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2328 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2329 ntohl(*((uint32_t *)(index + 4)));
2330 }
9a217f2a
JH
2331}
2332
c4001d92 2333off_t find_pack_entry_one(const unsigned char *sha1,
43057304 2334 struct packed_git *p)
1f688557 2335{
42873078 2336 const uint32_t *level1_ofs = p->index_data;
42873078 2337 const unsigned char *index = p->index_data;
628522ec
JH
2338 unsigned hi, lo, stride;
2339 static int use_lookup = -1;
2340 static int debug_lookup = -1;
2341
2342 if (debug_lookup < 0)
2343 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 2344
d079837e
SP
2345 if (!index) {
2346 if (open_pack_index(p))
2347 return 0;
2348 level1_ofs = p->index_data;
2349 index = p->index_data;
2350 }
74e34e1f
NP
2351 if (p->index_version > 1) {
2352 level1_ofs += 2;
2353 index += 8;
2354 }
42873078 2355 index += 4 * 256;
74e34e1f
NP
2356 hi = ntohl(level1_ofs[*sha1]);
2357 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
2358 if (p->index_version > 1) {
2359 stride = 20;
2360 } else {
2361 stride = 24;
2362 index += 4;
2363 }
2364
2365 if (debug_lookup)
6e1c2344 2366 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
2367 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2368
2369 if (use_lookup < 0)
2370 use_lookup = !!getenv("GIT_USE_LOOKUP");
2371 if (use_lookup) {
2372 int pos = sha1_entry_pos(index, stride, 0,
2373 lo, hi, p->num_objects, sha1);
2374 if (pos < 0)
2375 return 0;
2376 return nth_packed_object_offset(p, pos);
2377 }
1f688557
JH
2378
2379 do {
74e34e1f 2380 unsigned mi = (lo + hi) / 2;
628522ec
JH
2381 int cmp = hashcmp(index + mi * stride, sha1);
2382
2383 if (debug_lookup)
2384 printf("lo %u hi %u rg %u mi %u\n",
2385 lo, hi, hi - lo, mi);
43057304 2386 if (!cmp)
74e34e1f 2387 return nth_packed_object_offset(p, mi);
1f688557
JH
2388 if (cmp > 0)
2389 hi = mi;
2390 else
2391 lo = mi+1;
2392 } while (lo < hi);
2393 return 0;
2394}
2395
4c080182 2396int is_pack_valid(struct packed_git *p)
d131b7af
SP
2397{
2398 /* An already open pack is known to be valid. */
2399 if (p->pack_fd != -1)
2400 return 1;
2401
2402 /* If the pack has one window completely covering the
2403 * file size, the pack is known to be valid even if
2404 * the descriptor is not currently open.
2405 */
2406 if (p->windows) {
2407 struct pack_window *w = p->windows;
2408
2409 if (!w->offset && w->len == p->pack_size)
2410 return 1;
2411 }
2412
2413 /* Force the pack to open to prove its valid. */
2414 return !open_packed_git(p);
2415}
2416
95099731
NTND
2417static int fill_pack_entry(const unsigned char *sha1,
2418 struct pack_entry *e,
2419 struct packed_git *p)
2420{
2421 off_t offset;
2422
2423 if (p->num_bad_objects) {
2424 unsigned i;
2425 for (i = 0; i < p->num_bad_objects; i++)
2426 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2427 return 0;
2428 }
2429
2430 offset = find_pack_entry_one(sha1, p);
2431 if (!offset)
2432 return 0;
2433
2434 /*
2435 * We are about to tell the caller where they can locate the
2436 * requested object. We better make sure the packfile is
2437 * still here and can be accessed before supplying that
2438 * answer, as it may have been deleted since the index was
2439 * loaded!
2440 */
2441 if (!is_pack_valid(p)) {
2442 warning("packfile %s cannot be accessed", p->pack_name);
2443 return 0;
2444 }
2445 e->offset = offset;
2446 e->p = p;
2447 hashcpy(e->sha1, sha1);
2448 return 1;
2449}
2450
4d6acb70 2451static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557
JH
2452{
2453 struct packed_git *p;
43057304 2454
1f688557 2455 prepare_packed_git();
f7c22cc6
NP
2456 if (!packed_git)
2457 return 0;
1f688557 2458
c01f51cc
NTND
2459 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2460 return 1;
8eca0b47 2461
c01f51cc
NTND
2462 for (p = packed_git; p; p = p->next) {
2463 if (p == last_found_pack || !fill_pack_entry(sha1, e, p))
2464 continue;
f7c22cc6 2465
c01f51cc
NTND
2466 last_found_pack = p;
2467 return 1;
2468 }
1f688557
JH
2469 return 0;
2470}
2471
a6080a0a 2472struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
2473 struct packed_git *packs)
2474{
2475 struct packed_git *p;
bf592c50
DB
2476
2477 for (p = packs; p; p = p->next) {
43057304 2478 if (find_pack_entry_one(sha1, p))
bf592c50
DB
2479 return p;
2480 }
2481 return NULL;
21666f1a 2482
bf592c50
DB
2483}
2484
052fe5ea 2485static int sha1_loose_object_info(const unsigned char *sha1,
23c339c0 2486 struct object_info *oi)
65c2e0c3 2487{
36e4d74a 2488 int status;
65c2e0c3
JH
2489 unsigned long mapsize, size;
2490 void *map;
ef49a7a0 2491 git_zstream stream;
d65a16f6 2492 char hdr[32];
65c2e0c3 2493
5d642e75
JK
2494 if (oi->delta_base_sha1)
2495 hashclr(oi->delta_base_sha1);
2496
052fe5ea
JK
2497 /*
2498 * If we don't care about type or size, then we don't
4ef8d1dd
JH
2499 * need to look inside the object at all. Note that we
2500 * do not optimize out the stat call, even if the
2501 * caller doesn't care about the disk-size, since our
2502 * return value implicitly indicates whether the
2503 * object even exists.
052fe5ea 2504 */
23c339c0 2505 if (!oi->typep && !oi->sizep) {
4ef8d1dd
JH
2506 struct stat st;
2507 if (stat_sha1_file(sha1, &st) < 0)
2508 return -1;
2509 if (oi->disk_sizep)
23c339c0 2510 *oi->disk_sizep = st.st_size;
052fe5ea
JK
2511 return 0;
2512 }
2513
bb6b8e4f 2514 map = map_sha1_file(sha1, &mapsize);
f0df4ed5 2515 if (!map)
dbea72a8 2516 return -1;
23c339c0
JK
2517 if (oi->disk_sizep)
2518 *oi->disk_sizep = mapsize;
36e4d74a
JH
2519 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2520 status = error("unable to unpack %s header",
2521 sha1_to_hex(sha1));
21666f1a 2522 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 2523 status = error("unable to parse %s header", sha1_to_hex(sha1));
23c339c0
JK
2524 else if (oi->sizep)
2525 *oi->sizep = size;
39c68542 2526 git_inflate_end(&stream);
65c2e0c3 2527 munmap(map, mapsize);
23c339c0
JK
2528 if (oi->typep)
2529 *oi->typep = status;
052fe5ea 2530 return 0;
65c2e0c3
JH
2531}
2532
de7b5d62 2533int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
f0df4ed5 2534{
c4d9986f 2535 struct cached_object *co;
f0df4ed5 2536 struct pack_entry e;
5b086407 2537 int rtype;
1f7117ef 2538 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
f0df4ed5 2539
1f7117ef 2540 co = find_cached_object(real);
c4d9986f 2541 if (co) {
5b086407
JK
2542 if (oi->typep)
2543 *(oi->typep) = co->type;
9a490590
JH
2544 if (oi->sizep)
2545 *(oi->sizep) = co->size;
161f00e7
JK
2546 if (oi->disk_sizep)
2547 *(oi->disk_sizep) = 0;
5d642e75
JK
2548 if (oi->delta_base_sha1)
2549 hashclr(oi->delta_base_sha1);
9a490590 2550 oi->whence = OI_CACHED;
5b086407 2551 return 0;
c4d9986f
NTND
2552 }
2553
1f7117ef 2554 if (!find_pack_entry(real, &e)) {
ddd63e64 2555 /* Most likely it's a loose object. */
1f7117ef 2556 if (!sha1_loose_object_info(real, oi)) {
9a490590 2557 oi->whence = OI_LOOSE;
5b086407 2558 return 0;
9a490590 2559 }
ddd63e64
SG
2560
2561 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 2562 reprepare_packed_git();
1f7117ef 2563 if (!find_pack_entry(real, &e))
052fe5ea 2564 return -1;
f0df4ed5 2565 }
3d77d877 2566
23c339c0 2567 rtype = packed_object_info(e.p, e.offset, oi);
412916ee 2568 if (rtype < 0) {
1f7117ef
CC
2569 mark_bad_packed_object(e.p, real);
2570 return sha1_object_info_extended(real, oi, 0);
5266d369
JH
2571 } else if (in_delta_base_cache(e.p, e.offset)) {
2572 oi->whence = OI_DBCACHED;
9a490590
JH
2573 } else {
2574 oi->whence = OI_PACKED;
2575 oi->u.packed.offset = e.offset;
2576 oi->u.packed.pack = e.p;
2577 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2578 rtype == OBJ_OFS_DELTA);
3d77d877
NP
2579 }
2580
5b086407 2581 return 0;
f0df4ed5
JS
2582}
2583
3fc0dca9 2584/* returns enum object_type or negative */
9a490590
JH
2585int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2586{
5b086407 2587 enum object_type type;
d099b717 2588 struct object_info oi = {NULL};
9a490590 2589
5b086407 2590 oi.typep = &type;
9a490590 2591 oi.sizep = sizep;
de7b5d62 2592 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
5b086407
JK
2593 return -1;
2594 return type;
9a490590
JH
2595}
2596
21666f1a
NP
2597static void *read_packed_sha1(const unsigned char *sha1,
2598 enum object_type *type, unsigned long *size)
1f688557
JH
2599{
2600 struct pack_entry e;
8eca0b47 2601 void *data;
1f688557 2602
cd673c1f 2603 if (!find_pack_entry(sha1, &e))
1f688557 2604 return NULL;
8eca0b47
NP
2605 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2606 if (!data) {
2607 /*
2608 * We're probably in deep shit, but let's try to fetch
2609 * the required object anyway from another pack or loose.
2610 * This should happen only in the presence of a corrupted
2611 * pack, and is better than failing outright.
2612 */
2613 error("failed to read object %s at offset %"PRIuMAX" from %s",
2614 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2615 mark_bad_packed_object(e.p, sha1);
ac939109 2616 data = read_object(sha1, type, size);
8eca0b47
NP
2617 }
2618 return data;
1f688557
JH
2619}
2620
21666f1a
NP
2621int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2622 unsigned char *sha1)
d66b37bb
JH
2623{
2624 struct cached_object *co;
2625
21666f1a 2626 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2627 if (has_sha1_file(sha1) || find_cached_object(sha1))
2628 return 0;
2629 if (cached_object_alloc <= cached_object_nr) {
2630 cached_object_alloc = alloc_nr(cached_object_alloc);
2631 cached_objects = xrealloc(cached_objects,
2632 sizeof(*cached_objects) *
2633 cached_object_alloc);
2634 }
2635 co = &cached_objects[cached_object_nr++];
2636 co->size = len;
21666f1a 2637 co->type = type;
efa13f7b
JH
2638 co->buf = xmalloc(len);
2639 memcpy(co->buf, buf, len);
d66b37bb
JH
2640 hashcpy(co->sha1, sha1);
2641 return 0;
2642}
2643
c2c5b270
CC
2644static void *read_object(const unsigned char *sha1, enum object_type *type,
2645 unsigned long *size)
0fcfd160
LT
2646{
2647 unsigned long mapsize;
2648 void *map, *buf;
d66b37bb
JH
2649 struct cached_object *co;
2650
2651 co = find_cached_object(sha1);
2652 if (co) {
21666f1a 2653 *type = co->type;
d66b37bb 2654 *size = co->size;
182af834 2655 return xmemdupz(co->buf, co->size);
d66b37bb 2656 }
0fcfd160 2657
8276c007
PE
2658 buf = read_packed_sha1(sha1, type, size);
2659 if (buf)
2660 return buf;
bb6b8e4f 2661 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2662 if (map) {
7efbff75 2663 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2664 munmap(map, mapsize);
2665 return buf;
2666 }
637cdd9d 2667 reprepare_packed_git();
8276c007 2668 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2669}
2670
b6c4cecc
JH
2671/*
2672 * This function dies on corrupt objects; the callers who want to
2673 * deal with them should arrange to call read_object() and give error
2674 * messages themselves.
2675 */
5bf29b95
JH
2676void *read_sha1_file_extended(const unsigned char *sha1,
2677 enum object_type *type,
2678 unsigned long *size,
2679 unsigned flag)
ac939109 2680{
3ba7a065 2681 void *data;
e8b15e61 2682 char *path;
b6c4cecc 2683 const struct packed_git *p;
bf93eea0 2684 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
b6c4cecc 2685
3ba7a065
JH
2686 errno = 0;
2687 data = read_object(repl, type, size);
4bbf5a26 2688 if (data)
b6c4cecc 2689 return data;
68095570 2690
25f3af3f 2691 if (errno && errno != ENOENT)
3ba7a065
JH
2692 die_errno("failed to read object %s", sha1_to_hex(sha1));
2693
68095570 2694 /* die if we replaced an object with one that does not exist */
b6c4cecc 2695 if (repl != sha1)
68095570
CC
2696 die("replacement %s not found for %s",
2697 sha1_to_hex(repl), sha1_to_hex(sha1));
2698
b6c4cecc
JH
2699 if (has_loose_object(repl)) {
2700 path = sha1_file_name(sha1);
2701 die("loose object %s (stored in %s) is corrupt",
2702 sha1_to_hex(repl), path);
e8b15e61 2703 }
68095570 2704
b6c4cecc
JH
2705 if ((p = has_packed_and_bad(repl)) != NULL)
2706 die("packed object %s (stored in %s) is corrupt",
2707 sha1_to_hex(repl), p->pack_name);
f5552aee 2708
b6c4cecc 2709 return NULL;
ac939109
NP
2710}
2711
40469ee9 2712void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2713 const char *required_type_name,
40469ee9
JH
2714 unsigned long *size,
2715 unsigned char *actual_sha1_return)
f4913f91 2716{
21666f1a 2717 enum object_type type, required_type;
f4913f91
JH
2718 void *buffer;
2719 unsigned long isize;
40469ee9 2720 unsigned char actual_sha1[20];
f4913f91 2721
21666f1a 2722 required_type = type_from_string(required_type_name);
e702496e 2723 hashcpy(actual_sha1, sha1);
40469ee9
JH
2724 while (1) {
2725 int ref_length = -1;
2726 const char *ref_type = NULL;
f4913f91 2727
21666f1a 2728 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2729 if (!buffer)
2730 return NULL;
21666f1a 2731 if (type == required_type) {
40469ee9
JH
2732 *size = isize;
2733 if (actual_sha1_return)
e702496e 2734 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2735 return buffer;
2736 }
2737 /* Handle references */
21666f1a 2738 else if (type == OBJ_COMMIT)
40469ee9 2739 ref_type = "tree ";
21666f1a 2740 else if (type == OBJ_TAG)
40469ee9
JH
2741 ref_type = "object ";
2742 else {
2743 free(buffer);
2744 return NULL;
2745 }
2746 ref_length = strlen(ref_type);
f4913f91 2747
50974ec9
MK
2748 if (ref_length + 40 > isize ||
2749 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2750 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2751 free(buffer);
2752 return NULL;
2753 }
1cf58e72 2754 free(buffer);
40469ee9
JH
2755 /* Now we have the ID of the referred-to object in
2756 * actual_sha1. Check again. */
f4913f91 2757 }
f4913f91
JH
2758}
2759
ce9fbf16 2760static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2761 const char *type, unsigned char *sha1,
d65a16f6 2762 char *hdr, int *hdrlen)
d410c0f5 2763{
9126f009 2764 git_SHA_CTX c;
d410c0f5
JH
2765
2766 /* Generate the header */
d65a16f6 2767 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
2768
2769 /* Sha1.. */
9126f009
NP
2770 git_SHA1_Init(&c);
2771 git_SHA1_Update(&c, hdr, *hdrlen);
2772 git_SHA1_Update(&c, buf, len);
2773 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2774}
2775
230f1322 2776/*
5a688fe4
JH
2777 * Move the just written object into its final resting place.
2778 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2779 * "moving" is only a part of what it does, when no patch between
2780 * master to pu changes the call sites of this function.
230f1322 2781 */
839837b9 2782int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322 2783{
e32c0a9c 2784 int ret = 0;
5a688fe4 2785
348df166 2786 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
2787 goto try_rename;
2788 else if (link(tmpfile, filename))
e32c0a9c 2789 ret = errno;
7ebb6fca
LT
2790
2791 /*
2792 * Coda hack - coda doesn't like cross-directory links,
2793 * so we fall back to a rename, which will mean that it
2794 * won't be able to check collisions, but that's not a
2795 * big deal.
2796 *
2797 * The same holds for FAT formatted media.
2798 *
3be1f18e 2799 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
2800 * left to unlink.
2801 */
2802 if (ret && ret != EEXIST) {
be66a6c4 2803 try_rename:
7ebb6fca 2804 if (!rename(tmpfile, filename))
3be1f18e 2805 goto out;
9e48b389 2806 ret = errno;
230f1322 2807 }
691f1a28 2808 unlink_or_warn(tmpfile);
230f1322
LT
2809 if (ret) {
2810 if (ret != EEXIST) {
82247e9b 2811 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
230f1322
LT
2812 }
2813 /* FIXME!!! Collision check here ? */
2814 }
2815
3be1f18e 2816out:
5256b006 2817 if (adjust_shared_perm(filename))
5a688fe4 2818 return error("unable to set permission to '%s'", filename);
230f1322
LT
2819 return 0;
2820}
2821
4d548150
LT
2822static int write_buffer(int fd, const void *buf, size_t len)
2823{
d34cf19b 2824 if (write_in_full(fd, buf, len) < 0)
93822c22 2825 return error("file write error (%s)", strerror(errno));
4d548150
LT
2826 return 0;
2827}
2828
ce9fbf16 2829int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2830 unsigned char *sha1)
2831{
d65a16f6 2832 char hdr[32];
abdc3fc8
RS
2833 int hdrlen;
2834 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2835 return 0;
2836}
2837
e9039dd3
LT
2838/* Finalize a file on disk, and close it. */
2839static void close_sha1_file(int fd)
2840{
aafe9fba
LT
2841 if (fsync_object_files)
2842 fsync_or_die(fd, "sha1 file");
e9039dd3 2843 if (close(fd) != 0)
d824cbba 2844 die_errno("error when closing sha1 file");
e9039dd3
LT
2845}
2846
5723fe7e
LT
2847/* Size of directory component, including the ending '/' */
2848static inline int directory_size(const char *filename)
2849{
2850 const char *s = strrchr(filename, '/');
2851 if (!s)
2852 return 0;
2853 return s - filename + 1;
2854}
2855
2856/*
2857 * This creates a temporary file in the same directory as the final
2858 * 'filename'
2859 *
2860 * We want to avoid cross-directory filename renames, because those
2861 * can have problems on various filesystems (FAT, NFS, Coda).
2862 */
2863static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2864{
2865 int fd, dirlen = directory_size(filename);
2866
2867 if (dirlen + 20 > bufsiz) {
2868 errno = ENAMETOOLONG;
2869 return -1;
2870 }
2871 memcpy(buffer, filename, dirlen);
2872 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
5256b006 2873 fd = git_mkstemp_mode(buffer, 0444);
cbacbf4e 2874 if (fd < 0 && dirlen && errno == ENOENT) {
5723fe7e 2875 /* Make sure the directory exists */
6ff6af62 2876 memcpy(buffer, filename, dirlen);
5723fe7e 2877 buffer[dirlen-1] = 0;
b2476a60
JH
2878 if (mkdir(buffer, 0777) && errno != EEXIST)
2879 return -1;
2880 if (adjust_shared_perm(buffer))
5723fe7e
LT
2881 return -1;
2882
2883 /* Try again */
2884 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
5256b006 2885 fd = git_mkstemp_mode(buffer, 0444);
5723fe7e
LT
2886 }
2887 return fd;
2888}
2889
bbac7311 2890static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 2891 const void *buf, unsigned long len, time_t mtime)
0fcfd160 2892{
915308b1 2893 int fd, ret;
9892beba 2894 unsigned char compressed[4096];
ef49a7a0 2895 git_zstream stream;
748af44c
NP
2896 git_SHA_CTX c;
2897 unsigned char parano_sha1[20];
706bc531 2898 char *filename;
ab1900a3 2899 static char tmp_file[PATH_MAX];
a44c9a5e 2900
972a9155 2901 filename = sha1_file_name(sha1);
ab1900a3 2902 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
aac17941 2903 if (fd < 0) {
35243577 2904 if (errno == EACCES)
82247e9b 2905 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
916d081b 2906 else
5eaeda70 2907 return error("unable to create temporary file: %s", strerror(errno));
aac17941
LT
2908 }
2909
0fcfd160
LT
2910 /* Set it up */
2911 memset(&stream, 0, sizeof(stream));
55bb5c91 2912 git_deflate_init(&stream, zlib_compression_level);
0fcfd160 2913 stream.next_out = compressed;
9892beba 2914 stream.avail_out = sizeof(compressed);
748af44c 2915 git_SHA1_Init(&c);
a44c9a5e
LT
2916
2917 /* First header.. */
d65a16f6 2918 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2919 stream.avail_in = hdrlen;
55bb5c91
JH
2920 while (git_deflate(&stream, 0) == Z_OK)
2921 ; /* nothing */
748af44c 2922 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
2923
2924 /* Then the data itself.. */
c00e657d 2925 stream.next_in = (void *)buf;
a44c9a5e 2926 stream.avail_in = len;
9892beba 2927 do {
748af44c 2928 unsigned char *in0 = stream.next_in;
55bb5c91 2929 ret = git_deflate(&stream, Z_FINISH);
748af44c 2930 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
2931 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2932 die("unable to write sha1 file");
2933 stream.next_out = compressed;
2934 stream.avail_out = sizeof(compressed);
2935 } while (ret == Z_OK);
2936
ac54c277
LT
2937 if (ret != Z_STREAM_END)
2938 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
55bb5c91 2939 ret = git_deflate_end_gently(&stream);
ac54c277
LT
2940 if (ret != Z_OK)
2941 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
2942 git_SHA1_Final(parano_sha1, &c);
2943 if (hashcmp(sha1, parano_sha1) != 0)
2944 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 2945
e9039dd3 2946 close_sha1_file(fd);
0fcfd160 2947
bbac7311
NP
2948 if (mtime) {
2949 struct utimbuf utb;
2950 utb.actime = mtime;
2951 utb.modtime = mtime;
ab1900a3 2952 if (utime(tmp_file, &utb) < 0)
bbac7311 2953 warning("failed utime() on %s: %s",
ab1900a3 2954 tmp_file, strerror(errno));
bbac7311
NP
2955 }
2956
ab1900a3 2957 return move_temp_to_file(tmp_file, filename);
0fcfd160 2958}
8237b185 2959
c00e657d 2960int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
bbac7311
NP
2961{
2962 unsigned char sha1[20];
2963 char hdr[32];
2964 int hdrlen;
2965
2966 /* Normally if we have it in the pack then we do not bother writing
2967 * it out into .git/objects/??/?{38} file.
2968 */
2969 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2970 if (returnsha1)
2971 hashcpy(returnsha1, sha1);
2972 if (has_sha1_file(sha1))
2973 return 0;
2974 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2975}
2976
2977int force_object_loose(const unsigned char *sha1, time_t mtime)
2978{
bbac7311
NP
2979 void *buf;
2980 unsigned long len;
2981 enum object_type type;
2982 char hdr[32];
2983 int hdrlen;
1fb23e65 2984 int ret;
bbac7311 2985
c529d75a 2986 if (has_loose_object(sha1))
bbac7311
NP
2987 return 0;
2988 buf = read_packed_sha1(sha1, &type, &len);
2989 if (!buf)
2990 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2991 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1fb23e65
BS
2992 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2993 free(buf);
2994
2995 return ret;
bbac7311
NP
2996}
2997
bf592c50
DB
2998int has_pack_index(const unsigned char *sha1)
2999{
3000 struct stat st;
3001 if (stat(sha1_pack_index_name(sha1), &st))
3002 return 0;
3003 return 1;
bf592c50
DB
3004}
3005
cd673c1f
JH
3006int has_sha1_pack(const unsigned char *sha1)
3007{
3008 struct pack_entry e;
3009 return find_pack_entry(sha1, &e);
3010}
3011
8237b185
DB
3012int has_sha1_file(const unsigned char *sha1)
3013{
1f688557
JH
3014 struct pack_entry e;
3015
cd673c1f 3016 if (find_pack_entry(sha1, &e))
1f688557 3017 return 1;
45e8a748
JK
3018 if (has_loose_object(sha1))
3019 return 1;
3020 reprepare_packed_git();
3021 return find_pack_entry(sha1, &e);
8237b185 3022}
74400e71 3023
c879daa2
NTND
3024static void check_tree(const void *buf, size_t size)
3025{
3026 struct tree_desc desc;
3027 struct name_entry entry;
3028
3029 init_tree_desc(&desc, buf, size);
3030 while (tree_entry(&desc, &entry))
3031 /* do nothing
3032 * tree_entry() will die() on malformed entries */
3033 ;
3034}
3035
3036static void check_commit(const void *buf, size_t size)
3037{
3038 struct commit c;
3039 memset(&c, 0, sizeof(c));
3040 if (parse_commit_buffer(&c, buf, size))
3041 die("corrupt commit");
3042}
3043
3044static void check_tag(const void *buf, size_t size)
3045{
3046 struct tag t;
3047 memset(&t, 0, sizeof(t));
3048 if (parse_tag_buffer(&t, buf, size))
3049 die("corrupt tag");
3050}
3051
43df4f86 3052static int index_mem(unsigned char *sha1, void *buf, size_t size,
c4ce46fc
JH
3053 enum object_type type,
3054 const char *path, unsigned flags)
e7332f96 3055{
6c510bee 3056 int ret, re_allocated = 0;
c4ce46fc 3057 int write_object = flags & HASH_WRITE_OBJECT;
74400e71 3058
7672db20 3059 if (!type)
edaec3fb 3060 type = OBJ_BLOB;
6c510bee
LT
3061
3062 /*
3063 * Convert blobs to git internal format
3064 */
43df4f86 3065 if ((type == OBJ_BLOB) && path) {
f285a2d7 3066 struct strbuf nbuf = STRBUF_INIT;
21e5ad50 3067 if (convert_to_git(path, buf, size, &nbuf,
5e12e78e 3068 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
b315c5c0 3069 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
3070 re_allocated = 1;
3071 }
3072 }
c4ce46fc 3073 if (flags & HASH_FORMAT_CHECK) {
c879daa2
NTND
3074 if (type == OBJ_TREE)
3075 check_tree(buf, size);
3076 if (type == OBJ_COMMIT)
3077 check_commit(buf, size);
3078 if (type == OBJ_TAG)
3079 check_tag(buf, size);
3080 }
6c510bee 3081
7672db20 3082 if (write_object)
edaec3fb 3083 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 3084 else
edaec3fb 3085 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 3086 if (re_allocated)
6c510bee 3087 free(buf);
43df4f86
DP
3088 return ret;
3089}
3090
7b41e1e1
JH
3091static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3092 const char *path, unsigned flags)
3093{
3094 struct strbuf sbuf = STRBUF_INIT;
3095 int ret;
3096
3097 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3098 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3099 else
3100 ret = -1;
3101 strbuf_release(&sbuf);
3102 return ret;
3103}
3104
ea68b0ce
DP
3105#define SMALL_FILE_SIZE (32*1024)
3106
7b41e1e1
JH
3107static int index_core(unsigned char *sha1, int fd, size_t size,
3108 enum object_type type, const char *path,
3109 unsigned flags)
43df4f86
DP
3110{
3111 int ret;
43df4f86 3112
7b41e1e1 3113 if (!size) {
c4ce46fc 3114 ret = index_mem(sha1, NULL, size, type, path, flags);
ea68b0ce
DP
3115 } else if (size <= SMALL_FILE_SIZE) {
3116 char *buf = xmalloc(size);
3117 if (size == read_in_full(fd, buf, size))
c4ce46fc 3118 ret = index_mem(sha1, buf, size, type, path, flags);
ea68b0ce
DP
3119 else
3120 ret = error("short read %s", strerror(errno));
3121 free(buf);
08bda208 3122 } else {
43df4f86 3123 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
c4ce46fc 3124 ret = index_mem(sha1, buf, size, type, path, flags);
aac17941 3125 munmap(buf, size);
08bda208 3126 }
7b41e1e1
JH
3127 return ret;
3128}
3129
4dd1fbc7 3130/*
568508e7
JH
3131 * This creates one packfile per large blob unless bulk-checkin
3132 * machinery is "plugged".
4dd1fbc7
JH
3133 *
3134 * This also bypasses the usual "convert-to-git" dance, and that is on
3135 * purpose. We could write a streaming version of the converting
3136 * functions and insert that before feeding the data to fast-import
4f22b101
JK
3137 * (or equivalent in-core API described above). However, that is
3138 * somewhat complicated, as we do not know the size of the filter
3139 * result, which we need to know beforehand when writing a git object.
3140 * Since the primary motivation for trying to stream from the working
3141 * tree file and to avoid mmaping it in core is to deal with large
3142 * binary blobs, they generally do not want to get any conversion, and
3143 * callers should avoid this code path when filters are requested.
4dd1fbc7
JH
3144 */
3145static int index_stream(unsigned char *sha1, int fd, size_t size,
3146 enum object_type type, const char *path,
3147 unsigned flags)
3148{
568508e7 3149 return index_bulk_checkin(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
3150}
3151
7b41e1e1
JH
3152int index_fd(unsigned char *sha1, int fd, struct stat *st,
3153 enum object_type type, const char *path, unsigned flags)
3154{
3155 int ret;
3156 size_t size = xsize_t(st->st_size);
3157
3158 if (!S_ISREG(st->st_mode))
3159 ret = index_pipe(sha1, fd, type, path, flags);
4f22b101
JK
3160 else if (size <= big_file_threshold || type != OBJ_BLOB ||
3161 (path && would_convert_to_git(path, NULL, 0, 0)))
7b41e1e1 3162 ret = index_core(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
3163 else
3164 ret = index_stream(sha1, fd, size, type, path, flags);
43df4f86 3165 close(fd);
aac17941 3166 return ret;
74400e71 3167}
ec1fcc16 3168
c4ce46fc 3169int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
ec1fcc16
JH
3170{
3171 int fd;
b760d3aa 3172 struct strbuf sb = STRBUF_INIT;
ec1fcc16
JH
3173
3174 switch (st->st_mode & S_IFMT) {
3175 case S_IFREG:
3176 fd = open(path, O_RDONLY);
3177 if (fd < 0)
3178 return error("open(\"%s\"): %s", path,
3179 strerror(errno));
c4ce46fc 3180 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
ec1fcc16
JH
3181 return error("%s: failed to insert into database",
3182 path);
3183 break;
3184 case S_IFLNK:
b760d3aa 3185 if (strbuf_readlink(&sb, path, st->st_size)) {
ec1fcc16 3186 char *errstr = strerror(errno);
ec1fcc16
JH
3187 return error("readlink(\"%s\"): %s", path,
3188 errstr);
3189 }
c4ce46fc 3190 if (!(flags & HASH_WRITE_OBJECT))
b760d3aa
LT
3191 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3192 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
ec1fcc16
JH
3193 return error("%s: failed to insert into database",
3194 path);
b760d3aa 3195 strbuf_release(&sb);
ec1fcc16 3196 break;
f35a6d3b
LT
3197 case S_IFDIR:
3198 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
3199 default:
3200 return error("%s: unsupported file type", path);
3201 }
3202 return 0;
3203}
a69e5429
JH
3204
3205int read_pack_header(int fd, struct pack_header *header)
3206{
c697ad14
HO
3207 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3208 /* "eof before pack header was fully read" */
3209 return PH_ERROR_EOF;
3210
a69e5429
JH
3211 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3212 /* "protocol error (pack signature mismatch detected)" */
3213 return PH_ERROR_PACK_SIGNATURE;
3214 if (!pack_version_ok(header->hdr_version))
3215 /* "protocol error (pack version unsupported)" */
3216 return PH_ERROR_PROTOCOL;
3217 return 0;
3218}
40d52ff7
JK
3219
3220void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3221{
3222 enum object_type type = sha1_object_info(sha1, NULL);
3223 if (type < 0)
3224 die("%s is not a valid object", sha1_to_hex(sha1));
3225 if (type != expect)
3226 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3227 typename(expect));
3228}