safe_create_leading_directories(): always restore slash at end of loop
[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
b2cb9425
JH
108int safe_create_leading_directories(char *path)
109{
26c8ae2a 110 char *next_component = path + offset_1st_component(path);
9e6f885d 111 int ret = 0;
67d42212 112
9e6f885d 113 while (!ret && next_component) {
f0502332 114 struct stat st;
26c8ae2a 115 char *slash = strchr(next_component, '/');
f0502332 116
831651fd 117 if (!slash)
b2cb9425 118 break;
bf10cf70 119
26c8ae2a 120 next_component = slash + 1;
bf10cf70
MH
121 while (*next_component == '/')
122 next_component++;
26c8ae2a 123 if (!*next_component)
5f0bdf50 124 break;
831651fd
MH
125
126 *slash = '\0';
67d42212
JR
127 if (!stat(path, &st)) {
128 /* path exists */
9e6f885d
MH
129 if (!S_ISDIR(st.st_mode))
130 ret = -3;
53a39721 131 } else if (mkdir(path, 0777)) {
928734d9 132 if (errno == EEXIST &&
9e6f885d 133 !stat(path, &st) && S_ISDIR(st.st_mode))
928734d9 134 ; /* somebody created it since we checked */
9e6f885d
MH
135 else
136 ret = -1;
53a39721 137 } else if (adjust_shared_perm(path)) {
9e6f885d 138 ret = -2;
457f06d6 139 }
831651fd 140 *slash = '/';
b2cb9425 141 }
9e6f885d 142 return ret;
b2cb9425 143}
723c31fe 144
8e21d63b
JK
145int safe_create_leading_directories_const(const char *path)
146{
147 /* path points to cache entries, so xstrdup before messing with it */
148 char *buf = xstrdup(path);
149 int result = safe_create_leading_directories(buf);
150 free(buf);
151 return result;
152}
153
ace1534d
JH
154static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
155{
156 int i;
157 for (i = 0; i < 20; i++) {
158 static char hex[] = "0123456789abcdef";
159 unsigned int val = sha1[i];
160 char *pos = pathbuf + i*2 + (i > 0);
161 *pos++ = hex[val >> 4];
162 *pos = hex[val & 0xf];
163 }
164}
165
0fcfd160
LT
166/*
167 * NOTE! This returns a statically allocated buffer, so you have to be
790296fd 168 * careful about using it. Do an "xstrdup()" if you need to save the
0fcfd160 169 * filename.
ace1534d
JH
170 *
171 * Also note that this returns the location for creating. Reading
172 * SHA1 file can happen from any alternate directory listed in the
d19938ab 173 * DB_ENVIRONMENT environment variable if it is not found in
ace1534d 174 * the primary object database.
0fcfd160
LT
175 */
176char *sha1_file_name(const unsigned char *sha1)
177{
560fb6a1
JK
178 static char buf[PATH_MAX];
179 const char *objdir;
180 int len;
0fcfd160 181
560fb6a1
JK
182 objdir = get_object_directory();
183 len = strlen(objdir);
184
185 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
186 if (len + 43 > PATH_MAX)
187 die("insanely long object directory %s", objdir);
188 memcpy(buf, objdir, len);
189 buf[len] = '/';
190 buf[len+3] = '/';
191 buf[len+42] = '\0';
192 fill_sha1_path(buf + len + 1, sha1);
193 return buf;
0fcfd160
LT
194}
195
633f43e1 196static char *sha1_get_pack_name(const unsigned char *sha1,
6eec46bd 197 char **name, char **base, const char *which)
bf592c50
DB
198{
199 static const char hex[] = "0123456789abcdef";
633f43e1 200 char *buf;
bf592c50
DB
201 int i;
202
633f43e1 203 if (!*base) {
bf592c50
DB
204 const char *sha1_file_directory = get_object_directory();
205 int len = strlen(sha1_file_directory);
633f43e1 206 *base = xmalloc(len + 60);
6eec46bd
JH
207 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
208 sha1_file_directory, which);
633f43e1 209 *name = *base + len + 11;
bf592c50
DB
210 }
211
633f43e1 212 buf = *name;
bf592c50
DB
213
214 for (i = 0; i < 20; i++) {
215 unsigned int val = *sha1++;
216 *buf++ = hex[val >> 4];
217 *buf++ = hex[val & 0xf];
218 }
a6080a0a 219
633f43e1 220 return *base;
bf592c50
DB
221}
222
633f43e1 223char *sha1_pack_name(const unsigned char *sha1)
bf592c50 224{
633f43e1 225 static char *name, *base;
bf592c50 226
6eec46bd 227 return sha1_get_pack_name(sha1, &name, &base, "pack");
633f43e1 228}
bf592c50 229
633f43e1
HO
230char *sha1_pack_index_name(const unsigned char *sha1)
231{
232 static char *name, *base;
a6080a0a 233
6eec46bd 234 return sha1_get_pack_name(sha1, &name, &base, "idx");
bf592c50
DB
235}
236
d5a63b99
JH
237struct alternate_object_database *alt_odb_list;
238static struct alternate_object_database **alt_odb_tail;
ace1534d 239
f4e51683 240static int git_open_noatime(const char *name);
c2f493a4 241
ddd5d056
JH
242/*
243 * Prepare alternate object database registry.
d5a63b99
JH
244 *
245 * The variable alt_odb_list points at the list of struct
246 * alternate_object_database. The elements on this list come from
247 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
248 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
249 * whose contents is similar to that environment variable but can be
250 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
251 * contains "/the/directory/corresponding/to/.git/objects/...", while
252 * its name points just after the slash at the end of ".git/objects/"
253 * in the example above, and has enough space to hold 40-byte hex
254 * SHA1, an extra slash for the first level indirection, and the
255 * terminating NUL.
ddd5d056 256 */
6eac50d8 257static int link_alt_odb_entry(const char *entry, const char *relative_base, int depth)
ace1534d 258{
1494e038 259 const char *objdir = get_object_directory();
c2f493a4
MW
260 struct alternate_object_database *ent;
261 struct alternate_object_database *alt;
5bdf0a84
HW
262 int pfxlen, entlen;
263 struct strbuf pathbuf = STRBUF_INIT;
d5a63b99 264
85dadc38 265 if (!is_absolute_path(entry) && relative_base) {
5bdf0a84
HW
266 strbuf_addstr(&pathbuf, real_path(relative_base));
267 strbuf_addch(&pathbuf, '/');
c2f493a4 268 }
6eac50d8 269 strbuf_addstr(&pathbuf, entry);
c2f493a4 270
5bdf0a84
HW
271 normalize_path_copy(pathbuf.buf, pathbuf.buf);
272
273 pfxlen = strlen(pathbuf.buf);
274
275 /*
276 * The trailing slash after the directory name is given by
277 * this function at the end. Remove duplicates.
278 */
279 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
280 pfxlen -= 1;
281
282 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
283 ent = xmalloc(sizeof(*ent) + entlen);
284 memcpy(ent->base, pathbuf.buf, pfxlen);
285 strbuf_release(&pathbuf);
c2f493a4
MW
286
287 ent->name = ent->base + pfxlen + 1;
288 ent->base[pfxlen + 3] = '/';
289 ent->base[pfxlen] = ent->base[entlen-1] = 0;
290
291 /* Detect cases where alternate disappeared */
90b4a71c 292 if (!is_directory(ent->base)) {
c2f493a4
MW
293 error("object directory %s does not exist; "
294 "check .git/objects/info/alternates.",
295 ent->base);
296 free(ent);
297 return -1;
298 }
299
300 /* Prevent the common mistake of listing the same
301 * thing twice, or object directory itself.
302 */
303 for (alt = alt_odb_list; alt; alt = alt->next) {
304 if (!memcmp(ent->base, alt->base, pfxlen)) {
305 free(ent);
306 return -1;
307 }
308 }
cb2912c3 309 if (!strcmp(ent->base, objdir)) {
c2f493a4
MW
310 free(ent);
311 return -1;
312 }
313
314 /* add the alternate entry */
315 *alt_odb_tail = ent;
316 alt_odb_tail = &(ent->next);
317 ent->next = NULL;
318
319 /* recursively add alternates */
320 read_info_alternates(ent->base, depth + 1);
321
322 ent->base[pfxlen] = '/';
323
324 return 0;
325}
326
c5950164 327static void link_alt_odb_entries(const char *alt, int len, int sep,
c2f493a4
MW
328 const char *relative_base, int depth)
329{
6eac50d8
MH
330 struct string_list entries = STRING_LIST_INIT_NODUP;
331 char *alt_copy;
332 int i;
c2f493a4
MW
333
334 if (depth > 5) {
335 error("%s: ignoring alternate object stores, nesting too deep.",
336 relative_base);
337 return;
338 }
339
c5950164 340 alt_copy = xmemdupz(alt, len);
6eac50d8
MH
341 string_list_split_in_place(&entries, alt_copy, sep, -1);
342 for (i = 0; i < entries.nr; i++) {
343 const char *entry = entries.items[i].string;
344 if (entry[0] == '\0' || entry[0] == '#')
9577e7e3 345 continue;
6eac50d8
MH
346 if (!is_absolute_path(entry) && depth) {
347 error("%s: ignoring relative alternate object store %s",
348 relative_base, entry);
349 } else {
350 link_alt_odb_entry(entry, relative_base, depth);
9577e7e3 351 }
9577e7e3 352 }
6eac50d8
MH
353 string_list_clear(&entries, 0);
354 free(alt_copy);
d5a63b99
JH
355}
356
5e73633d 357void read_info_alternates(const char * relative_base, int depth)
d5a63b99 358{
9577e7e3 359 char *map;
dc49cd76 360 size_t mapsz;
d5a63b99 361 struct stat st;
9cb18f56
JM
362 const char alt_file_name[] = "info/alternates";
363 /* Given that relative_base is no longer than PATH_MAX,
364 ensure that "path" has enough space to append "/", the
365 file name, "info/alternates", and a trailing NUL. */
366 char path[PATH_MAX + 1 + sizeof alt_file_name];
c2f493a4 367 int fd;
d5a63b99 368
9cb18f56 369 sprintf(path, "%s/%s", relative_base, alt_file_name);
f4e51683 370 fd = git_open_noatime(path);
d5a63b99
JH
371 if (fd < 0)
372 return;
373 if (fstat(fd, &st) || (st.st_size == 0)) {
374 close(fd);
9a217f2a 375 return;
ace1534d 376 }
dc49cd76
SP
377 mapsz = xsize_t(st.st_size);
378 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 379 close(fd);
d5a63b99 380
c5950164 381 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
c2f493a4 382
dc49cd76 383 munmap(map, mapsz);
ace1534d
JH
384}
385
bef70b22
DB
386void add_to_alternates_file(const char *reference)
387{
388 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
acd3b9ec 389 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
e6baf4a1 390 char *alt = mkpath("%s\n", reference);
bef70b22
DB
391 write_or_die(fd, alt, strlen(alt));
392 if (commit_lock_file(lock))
393 die("could not close alternates file");
394 if (alt_odb_tail)
c5950164 395 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
bef70b22
DB
396}
397
d79796bc
JH
398void foreach_alt_odb(alt_odb_fn fn, void *cb)
399{
400 struct alternate_object_database *ent;
401
402 prepare_alt_odb();
403 for (ent = alt_odb_list; ent; ent = ent->next)
404 if (fn(ent, cb))
405 return;
406}
407
c2f493a4
MW
408void prepare_alt_odb(void)
409{
554fe20d 410 const char *alt;
c2f493a4 411
7dc24aa5
SP
412 if (alt_odb_tail)
413 return;
414
c2f493a4
MW
415 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
416 if (!alt) alt = "";
417
c2f493a4 418 alt_odb_tail = &alt_odb_list;
c5950164 419 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
420
421 read_info_alternates(get_object_directory(), 0);
422}
423
0f4dc14a 424static int has_loose_object_local(const unsigned char *sha1)
ace1534d 425{
ace1534d 426 char *name = sha1_file_name(sha1);
0f4dc14a
BC
427 return !access(name, F_OK);
428}
ace1534d 429
0f4dc14a
BC
430int has_loose_object_nonlocal(const unsigned char *sha1)
431{
432 struct alternate_object_database *alt;
9a217f2a 433 prepare_alt_odb();
d5a63b99 434 for (alt = alt_odb_list; alt; alt = alt->next) {
0f4dc14a 435 fill_sha1_path(alt->name, sha1);
c529d75a
LT
436 if (!access(alt->base, F_OK))
437 return 1;
ace1534d 438 }
c529d75a 439 return 0;
ace1534d
JH
440}
441
0f4dc14a
BC
442static int has_loose_object(const unsigned char *sha1)
443{
444 return has_loose_object_local(sha1) ||
445 has_loose_object_nonlocal(sha1);
446}
447
60bb8b14 448static unsigned int pack_used_ctr;
a53128b6
SP
449static unsigned int pack_mmap_calls;
450static unsigned int peak_pack_open_windows;
451static unsigned int pack_open_windows;
c7934306
SP
452static unsigned int pack_open_fds;
453static unsigned int pack_max_fds;
a53128b6 454static size_t peak_pack_mapped;
60bb8b14 455static size_t pack_mapped;
9a217f2a 456struct packed_git *packed_git;
1f688557 457
b79d18c9 458void pack_report(void)
a53128b6
SP
459{
460 fprintf(stderr,
e05db0fd
PR
461 "pack_report: getpagesize() = %10" SZ_FMT "\n"
462 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
463 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
464 sz_fmt(getpagesize()),
465 sz_fmt(packed_git_window_size),
466 sz_fmt(packed_git_limit));
a53128b6
SP
467 fprintf(stderr,
468 "pack_report: pack_used_ctr = %10u\n"
469 "pack_report: pack_mmap_calls = %10u\n"
470 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
471 "pack_report: pack_mapped = "
472 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
473 pack_used_ctr,
474 pack_mmap_calls,
475 pack_open_windows, peak_pack_open_windows,
9e42d6a1 476 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
477}
478
42873078 479static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
480{
481 void *idx_map;
42873078 482 struct pack_idx_header *hdr;
dc49cd76 483 size_t idx_size;
74e34e1f 484 uint32_t version, nr, i, *index;
f4e51683 485 int fd = git_open_noatime(path);
1f688557 486 struct stat st;
42873078 487
1f688557
JH
488 if (fd < 0)
489 return -1;
490 if (fstat(fd, &st)) {
491 close(fd);
492 return -1;
493 }
dc49cd76 494 idx_size = xsize_t(st.st_size);
2d88451b
SP
495 if (idx_size < 4 * 256 + 20 + 20) {
496 close(fd);
497 return error("index file %s is too small", path);
498 }
c4712e45 499 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 500 close(fd);
1f688557 501
42873078
NP
502 hdr = idx_map;
503 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
504 version = ntohl(hdr->idx_version);
505 if (version < 2 || version > 2) {
506 munmap(idx_map, idx_size);
6e1c2344 507 return error("index file %s is version %"PRIu32
74e34e1f
NP
508 " and is not supported by this binary"
509 " (try upgrading GIT to a newer version)",
510 path, version);
511 }
512 } else
513 version = 1;
df1b059d 514
1f688557 515 nr = 0;
42873078 516 index = idx_map;
74e34e1f
NP
517 if (version > 1)
518 index += 2; /* skip index header */
1f688557 519 for (i = 0; i < 256; i++) {
326bf396 520 uint32_t n = ntohl(index[i]);
2d88451b
SP
521 if (n < nr) {
522 munmap(idx_map, idx_size);
df1b059d 523 return error("non-monotonic index %s", path);
2d88451b 524 }
1f688557
JH
525 nr = n;
526 }
527
74e34e1f
NP
528 if (version == 1) {
529 /*
530 * Total size:
531 * - 256 index entries 4 bytes each
532 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
533 * - 20-byte SHA1 of the packfile
534 * - 20-byte SHA1 file checksum
535 */
536 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
537 munmap(idx_map, idx_size);
eef427a0 538 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
539 }
540 } else if (version == 2) {
541 /*
542 * Minimum size:
543 * - 8 bytes of header
544 * - 256 index entries 4 bytes each
545 * - 20-byte sha1 entry * nr
546 * - 4-byte crc entry * nr
547 * - 4-byte offset entry * nr
548 * - 20-byte SHA1 of the packfile
549 * - 20-byte SHA1 file checksum
550 * And after the 4-byte offset table might be a
551 * variable sized table containing 8-byte entries
552 * for offsets larger than 2^31.
553 */
554 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
555 unsigned long max_size = min_size;
556 if (nr)
557 max_size += (nr - 1)*8;
558 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 559 munmap(idx_map, idx_size);
eef427a0 560 return error("wrong index v2 file size in %s", path);
74e34e1f 561 }
7109c889
JH
562 if (idx_size != min_size &&
563 /*
564 * make sure we can deal with large pack offsets.
565 * 31-bit signed offset won't be enough, neither
566 * 32-bit unsigned one will be.
567 */
568 (sizeof(off_t) <= 4)) {
569 munmap(idx_map, idx_size);
570 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 571 }
2d88451b 572 }
1f688557 573
74e34e1f 574 p->index_version = version;
42873078
NP
575 p->index_data = idx_map;
576 p->index_size = idx_size;
57059091 577 p->num_objects = nr;
1f688557
JH
578 return 0;
579}
580
bc8e478a 581int open_pack_index(struct packed_git *p)
d079837e
SP
582{
583 char *idx_name;
584 int ret;
585
586 if (p->index_data)
587 return 0;
588
589 idx_name = xstrdup(p->pack_name);
590 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
591 ret = check_packed_git_idx(idx_name, p);
592 free(idx_name);
593 return ret;
594}
595
11daf39b
SP
596static void scan_windows(struct packed_git *p,
597 struct packed_git **lru_p,
598 struct pack_window **lru_w,
599 struct pack_window **lru_l)
1f688557 600{
11daf39b
SP
601 struct pack_window *w, *w_l;
602
603 for (w_l = NULL, w = p->windows; w; w = w->next) {
604 if (!w->inuse_cnt) {
605 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
606 *lru_p = p;
607 *lru_w = w;
608 *lru_l = w_l;
54044bf8 609 }
54044bf8 610 }
11daf39b 611 w_l = w;
f9253394 612 }
11daf39b
SP
613}
614
7c3ecb32 615static int unuse_one_window(struct packed_git *current)
11daf39b
SP
616{
617 struct packed_git *p, *lru_p = NULL;
618 struct pack_window *lru_w = NULL, *lru_l = NULL;
619
620 if (current)
621 scan_windows(current, &lru_p, &lru_w, &lru_l);
622 for (p = packed_git; p; p = p->next)
623 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
624 if (lru_p) {
625 munmap(lru_w->base, lru_w->len);
626 pack_mapped -= lru_w->len;
627 if (lru_l)
628 lru_l->next = lru_w->next;
7c3ecb32 629 else
54044bf8 630 lru_p->windows = lru_w->next;
54044bf8 631 free(lru_w);
a53128b6 632 pack_open_windows--;
54044bf8
SP
633 return 1;
634 }
635 return 0;
f9253394
JH
636}
637
7c3ecb32 638void release_pack_memory(size_t need)
97bfeb34
SP
639{
640 size_t cur = pack_mapped;
7c3ecb32 641 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
97bfeb34
SP
642 ; /* nothing */
643}
644
58ecbd5e
JN
645void *xmmap(void *start, size_t length,
646 int prot, int flags, int fd, off_t offset)
647{
648 void *ret = mmap(start, length, prot, flags, fd, offset);
649 if (ret == MAP_FAILED) {
650 if (!length)
651 return NULL;
7c3ecb32 652 release_pack_memory(length);
58ecbd5e
JN
653 ret = mmap(start, length, prot, flags, fd, offset);
654 if (ret == MAP_FAILED)
655 die_errno("Out of memory? mmap failed");
656 }
657 return ret;
658}
659
c9ced051
SP
660void close_pack_windows(struct packed_git *p)
661{
662 while (p->windows) {
663 struct pack_window *w = p->windows;
664
665 if (w->inuse_cnt)
666 die("pack '%s' still has open windows to it",
667 p->pack_name);
668 munmap(w->base, w->len);
669 pack_mapped -= w->len;
670 pack_open_windows--;
671 p->windows = w->next;
672 free(w);
673 }
674}
675
88d0db55
BC
676/*
677 * The LRU pack is the one with the oldest MRU window, preferring packs
678 * with no used windows, or the oldest mtime if it has no windows allocated.
679 */
680static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
681{
682 struct pack_window *w, *this_mru_w;
683 int has_windows_inuse = 0;
684
685 /*
686 * Reject this pack if it has windows and the previously selected
687 * one does not. If this pack does not have windows, reject
688 * it if the pack file is newer than the previously selected one.
689 */
690 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
691 return;
692
693 for (w = this_mru_w = p->windows; w; w = w->next) {
694 /*
695 * Reject this pack if any of its windows are in use,
696 * but the previously selected pack did not have any
697 * inuse windows. Otherwise, record that this pack
698 * has windows in use.
699 */
700 if (w->inuse_cnt) {
701 if (*accept_windows_inuse)
702 has_windows_inuse = 1;
703 else
704 return;
705 }
706
707 if (w->last_used > this_mru_w->last_used)
708 this_mru_w = w;
709
710 /*
711 * Reject this pack if it has windows that have been
712 * used more recently than the previously selected pack.
713 * If the previously selected pack had windows inuse and
714 * we have not encountered a window in this pack that is
715 * inuse, skip this check since we prefer a pack with no
716 * inuse windows to one that has inuse windows.
717 */
718 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
719 this_mru_w->last_used > (*mru_w)->last_used)
720 return;
721 }
722
723 /*
724 * Select this pack.
725 */
726 *mru_w = this_mru_w;
727 *lru_p = p;
728 *accept_windows_inuse = has_windows_inuse;
729}
730
731static int close_one_pack(void)
732{
733 struct packed_git *p, *lru_p = NULL;
734 struct pack_window *mru_w = NULL;
735 int accept_windows_inuse = 1;
736
737 for (p = packed_git; p; p = p->next) {
738 if (p->pack_fd == -1)
739 continue;
740 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
741 }
742
743 if (lru_p) {
744 close(lru_p->pack_fd);
745 pack_open_fds--;
746 lru_p->pack_fd = -1;
747 return 1;
748 }
749
750 return 0;
751}
752
03e79c88 753void unuse_pack(struct pack_window **w_cursor)
f9253394 754{
03e79c88
SP
755 struct pack_window *w = *w_cursor;
756 if (w) {
757 w->inuse_cnt--;
758 *w_cursor = NULL;
759 }
1f688557
JH
760}
761
fa5fc15d
SP
762void close_pack_index(struct packed_git *p)
763{
764 if (p->index_data) {
765 munmap((void *)p->index_data, p->index_size);
766 p->index_data = NULL;
767 }
768}
769
c74faea1
NP
770/*
771 * This is used by git-repack in case a newly created pack happens to
772 * contain the same set of objects as an existing one. In that case
773 * the resulting file might be different even if its name would be the
774 * same. It is best to close any reference to the old pack before it is
775 * replaced on disk. Of course no index pointers nor windows for given pack
776 * must subsist at this point. If ever objects from this pack are requested
777 * again, the new version of the pack will be reinitialized through
778 * reprepare_packed_git().
779 */
780void free_pack_by_name(const char *pack_name)
781{
782 struct packed_git *p, **pp = &packed_git;
783
784 while (*pp) {
785 p = *pp;
786 if (strcmp(pack_name, p->pack_name) == 0) {
fa3a0c94 787 clear_delta_base_cache();
c74faea1 788 close_pack_windows(p);
c7934306 789 if (p->pack_fd != -1) {
c74faea1 790 close(p->pack_fd);
c7934306
SP
791 pack_open_fds--;
792 }
fa5fc15d 793 close_pack_index(p);
c74faea1
NP
794 free(p->bad_object_sha1);
795 *pp = p->next;
c01f51cc
NTND
796 if (last_found_pack == p)
797 last_found_pack = NULL;
c74faea1
NP
798 free(p);
799 return;
800 }
801 pp = &p->next;
802 }
803}
804
a0788266
JS
805static unsigned int get_max_fd_limit(void)
806{
807#ifdef RLIMIT_NOFILE
808 struct rlimit lim;
809
810 if (getrlimit(RLIMIT_NOFILE, &lim))
811 die_errno("cannot get RLIMIT_NOFILE");
812
813 return lim.rlim_cur;
814#elif defined(_SC_OPEN_MAX)
815 return sysconf(_SC_OPEN_MAX);
816#elif defined(OPEN_MAX)
817 return OPEN_MAX;
818#else
819 return 1; /* see the caller ;-) */
820#endif
821}
822
3cf8b462
SP
823/*
824 * Do not call this directly as this leaks p->pack_fd on error return;
825 * call open_packed_git() instead.
826 */
827static int open_packed_git_1(struct packed_git *p)
1f688557 828{
9bc879c1
SP
829 struct stat st;
830 struct pack_header hdr;
831 unsigned char sha1[20];
832 unsigned char *idx_sha1;
2c039da8 833 long fd_flag;
9bc879c1 834
d079837e
SP
835 if (!p->index_data && open_pack_index(p))
836 return error("packfile %s index unavailable", p->pack_name);
837
c7934306 838 if (!pack_max_fds) {
a0788266 839 unsigned int max_fds = get_max_fd_limit();
c7934306
SP
840
841 /* Save 3 for stdin/stdout/stderr, 22 for work */
842 if (25 < max_fds)
843 pack_max_fds = max_fds - 25;
844 else
845 pack_max_fds = 1;
846 }
847
88d0db55 848 while (pack_max_fds <= pack_open_fds && close_one_pack())
c7934306
SP
849 ; /* nothing */
850
f4e51683 851 p->pack_fd = git_open_noatime(p->pack_name);
9bc879c1 852 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 853 return -1;
c7934306 854 pack_open_fds++;
9bc879c1
SP
855
856 /* If we created the struct before we had the pack we lack size. */
bf592c50 857 if (!p->pack_size) {
bf592c50 858 if (!S_ISREG(st.st_mode))
072db278 859 return error("packfile %s not a regular file", p->pack_name);
bf592c50 860 p->pack_size = st.st_size;
9bc879c1 861 } else if (p->pack_size != st.st_size)
072db278 862 return error("packfile %s size changed", p->pack_name);
9bc879c1 863
2c039da8
JH
864 /* We leave these file descriptors open with sliding mmap;
865 * there is no point keeping them open across exec(), though.
866 */
867 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
868 if (fd_flag < 0)
072db278 869 return error("cannot determine file descriptor flags");
2c039da8
JH
870 fd_flag |= FD_CLOEXEC;
871 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 872 return error("cannot set FD_CLOEXEC");
2c039da8 873
9bc879c1 874 /* Verify we recognize this pack file format. */
e6e2bd62 875 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 876 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 877 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 878 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 879 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
880 return error("packfile %s is version %"PRIu32" and not"
881 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
882 p->pack_name, ntohl(hdr.hdr_version));
883
884 /* Verify the pack matches its index. */
57059091 885 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
886 return error("packfile %s claims to have %"PRIu32" objects"
887 " while index indicates %"PRIu32" objects",
57059091
NP
888 p->pack_name, ntohl(hdr.hdr_entries),
889 p->num_objects);
9bc879c1 890 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 891 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 892 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 893 return error("packfile %s signature is unavailable", p->pack_name);
42873078 894 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 895 if (hashcmp(sha1, idx_sha1))
072db278
SP
896 return error("packfile %s does not match index", p->pack_name);
897 return 0;
9bc879c1
SP
898}
899
3cf8b462
SP
900static int open_packed_git(struct packed_git *p)
901{
902 if (!open_packed_git_1(p))
903 return 0;
904 if (p->pack_fd != -1) {
905 close(p->pack_fd);
c7934306 906 pack_open_fds--;
3cf8b462
SP
907 p->pack_fd = -1;
908 }
909 return -1;
910}
911
c4001d92 912static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
913{
914 /* We must promise at least 20 bytes (one hash) after the
915 * offset is available from this window, otherwise the offset
916 * is not actually in this window and a different window (which
917 * has that one hash excess) must be used. This is to support
918 * the object header and delta base parsing routines below.
919 */
920 off_t win_off = win->offset;
921 return win_off <= offset
922 && (offset + 20) <= (win_off + win->len);
923}
924
4b25d091 925unsigned char *use_pack(struct packed_git *p,
03e79c88 926 struct pack_window **w_cursor,
c4001d92 927 off_t offset,
ef49a7a0 928 unsigned long *left)
9bc879c1 929{
60bb8b14 930 struct pack_window *win = *w_cursor;
03e79c88 931
a9d98a14 932 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
933 * pointless to ask for an offset into the middle of that
934 * hash, and the in_window function above wouldn't match
935 * don't allow an offset too close to the end of the file.
936 */
d131b7af
SP
937 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
938 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
939 if (offset > (p->pack_size - 20))
940 die("offset beyond end of packfile (truncated pack?)");
941
942 if (!win || !in_window(win, offset)) {
943 if (win)
944 win->inuse_cnt--;
945 for (win = p->windows; win; win = win->next) {
946 if (in_window(win, offset))
947 break;
948 }
949 if (!win) {
78a28df9 950 size_t window_align = packed_git_window_size / 2;
dc49cd76 951 off_t len;
d131b7af
SP
952
953 if (p->pack_fd == -1 && open_packed_git(p))
954 die("packfile %s cannot be accessed", p->pack_name);
955
60bb8b14 956 win = xcalloc(1, sizeof(*win));
78a28df9 957 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
958 len = p->pack_size - win->offset;
959 if (len > packed_git_window_size)
960 len = packed_git_window_size;
961 win->len = (size_t)len;
60bb8b14 962 pack_mapped += win->len;
11daf39b 963 while (packed_git_limit < pack_mapped
7c3ecb32 964 && unuse_one_window(p))
60bb8b14 965 ; /* nothing */
c4712e45 966 win->base = xmmap(NULL, win->len,
60bb8b14
SP
967 PROT_READ, MAP_PRIVATE,
968 p->pack_fd, win->offset);
969 if (win->base == MAP_FAILED)
73b4e4be
SP
970 die("packfile %s cannot be mapped: %s",
971 p->pack_name,
972 strerror(errno));
d131b7af
SP
973 if (!win->offset && win->len == p->pack_size
974 && !p->do_not_close) {
975 close(p->pack_fd);
976 pack_open_fds--;
977 p->pack_fd = -1;
978 }
a53128b6
SP
979 pack_mmap_calls++;
980 pack_open_windows++;
981 if (pack_mapped > peak_pack_mapped)
982 peak_pack_mapped = pack_mapped;
983 if (pack_open_windows > peak_pack_open_windows)
984 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
985 win->next = p->windows;
986 p->windows = win;
987 }
1f688557 988 }
03e79c88
SP
989 if (win != *w_cursor) {
990 win->last_used = pack_used_ctr++;
991 win->inuse_cnt++;
992 *w_cursor = win;
993 }
60bb8b14 994 offset -= win->offset;
03e79c88 995 if (left)
dc49cd76 996 *left = win->len - xsize_t(offset);
03e79c88 997 return win->base + offset;
1f688557
JH
998}
999
27d69a46
NP
1000static struct packed_git *alloc_packed_git(int extra)
1001{
1002 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1003 memset(p, 0, sizeof(*p));
1004 p->pack_fd = -1;
1005 return p;
1006}
1007
e0500293
JN
1008static void try_to_free_pack_memory(size_t size)
1009{
7c3ecb32 1010 release_pack_memory(size);
e0500293
JN
1011}
1012
42873078 1013struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557 1014{
e0500293 1015 static int have_set_try_to_free_routine;
1f688557 1016 struct stat st;
27d69a46 1017 struct packed_git *p = alloc_packed_git(path_len + 2);
1f688557 1018
e0500293
JN
1019 if (!have_set_try_to_free_routine) {
1020 have_set_try_to_free_routine = 1;
1021 set_try_to_free_routine(try_to_free_pack_memory);
1022 }
1023
42873078
NP
1024 /*
1025 * Make sure a corresponding .pack file exists and that
1026 * the index looks sane.
1027 */
1028 path_len -= strlen(".idx");
27d69a46
NP
1029 if (path_len < 1) {
1030 free(p);
1f688557 1031 return NULL;
27d69a46 1032 }
42873078 1033 memcpy(p->pack_name, path, path_len);
8d25931d
BC
1034
1035 strcpy(p->pack_name + path_len, ".keep");
1036 if (!access(p->pack_name, F_OK))
1037 p->pack_keep = 1;
1038
42873078 1039 strcpy(p->pack_name + path_len, ".pack");
d079837e 1040 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 1041 free(p);
1f688557
JH
1042 return NULL;
1043 }
42873078 1044
1f688557
JH
1045 /* ok, it looks sane as far as we can check without
1046 * actually mapping the pack file.
1047 */
1f688557 1048 p->pack_size = st.st_size;
9d835df2 1049 p->pack_local = local;
b867092f 1050 p->mtime = st.st_mtime;
42873078
NP
1051 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1052 hashclr(p->sha1);
1f688557
JH
1053 return p;
1054}
1055
7b64469a 1056struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 1057{
42873078 1058 const char *path = sha1_pack_name(sha1);
27d69a46 1059 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
bf592c50 1060
27d69a46
NP
1061 strcpy(p->pack_name, path);
1062 hashcpy(p->sha1, sha1);
42873078
NP
1063 if (check_packed_git_idx(idx_path, p)) {
1064 free(p);
bf592c50 1065 return NULL;
42873078 1066 }
bf592c50 1067
bf592c50
DB
1068 return p;
1069}
1070
1071void install_packed_git(struct packed_git *pack)
1072{
c7934306
SP
1073 if (pack->pack_fd != -1)
1074 pack_open_fds++;
1075
bf592c50
DB
1076 pack->next = packed_git;
1077 packed_git = pack;
1078}
1079
543c5caa
NTND
1080void (*report_garbage)(const char *desc, const char *path);
1081
1082static void report_helper(const struct string_list *list,
1083 int seen_bits, int first, int last)
1084{
1085 const char *msg;
1086 switch (seen_bits) {
1087 case 0:
1088 msg = "no corresponding .idx nor .pack";
1089 break;
1090 case 1:
1091 msg = "no corresponding .idx";
1092 break;
1093 case 2:
1094 msg = "no corresponding .pack";
1095 break;
1096 default:
1097 return;
1098 }
1099 for (; first < last; first++)
1100 report_garbage(msg, list->items[first].string);
1101}
1102
1103static void report_pack_garbage(struct string_list *list)
1104{
1105 int i, baselen = -1, first = 0, seen_bits = 0;
1106
1107 if (!report_garbage)
1108 return;
1109
1110 sort_string_list(list);
1111
1112 for (i = 0; i < list->nr; i++) {
1113 const char *path = list->items[i].string;
1114 if (baselen != -1 &&
1115 strncmp(path, list->items[first].string, baselen)) {
1116 report_helper(list, seen_bits, first, i);
1117 baselen = -1;
1118 seen_bits = 0;
1119 }
1120 if (baselen == -1) {
1121 const char *dot = strrchr(path, '.');
1122 if (!dot) {
1123 report_garbage("garbage found", path);
1124 continue;
1125 }
1126 baselen = dot - path + 1;
1127 first = i;
1128 }
1129 if (!strcmp(path + baselen, "pack"))
1130 seen_bits |= 1;
1131 else if (!strcmp(path + baselen, "idx"))
1132 seen_bits |= 2;
1133 }
1134 report_helper(list, seen_bits, first, list->nr);
1135}
1136
9d835df2 1137static void prepare_packed_git_one(char *objdir, int local)
1f688557 1138{
9cb18f56
JM
1139 /* Ensure that this buffer is large enough so that we can
1140 append "/pack/" without clobbering the stack even if
1141 strlen(objdir) were PATH_MAX. */
1142 char path[PATH_MAX + 1 + 4 + 1 + 1];
1f688557
JH
1143 int len;
1144 DIR *dir;
1145 struct dirent *de;
543c5caa 1146 struct string_list garbage = STRING_LIST_INIT_DUP;
1f688557
JH
1147
1148 sprintf(path, "%s/pack", objdir);
1149 len = strlen(path);
1150 dir = opendir(path);
b5b16990 1151 if (!dir) {
26125f6b 1152 if (errno != ENOENT)
bd2afde8 1153 error("unable to open object pack directory: %s: %s",
26125f6b 1154 path, strerror(errno));
1f688557 1155 return;
b5b16990 1156 }
1f688557
JH
1157 path[len++] = '/';
1158 while ((de = readdir(dir)) != NULL) {
1159 int namelen = strlen(de->d_name);
1160 struct packed_git *p;
1161
543c5caa
NTND
1162 if (len + namelen + 1 > sizeof(path)) {
1163 if (report_garbage) {
1164 struct strbuf sb = STRBUF_INIT;
1165 strbuf_addf(&sb, "%.*s/%s", len - 1, path, de->d_name);
1166 report_garbage("path too long", sb.buf);
1167 strbuf_release(&sb);
1168 }
1169 continue;
1170 }
1171
1172 if (is_dot_or_dotdot(de->d_name))
9cb18f56
JM
1173 continue;
1174
1f688557 1175 strcpy(path + len, de->d_name);
d90906a9
NTND
1176
1177 if (has_extension(de->d_name, ".idx")) {
1178 /* Don't reopen a pack we already have. */
1179 for (p = packed_git; p; p = p->next) {
1180 if (!memcmp(path, p->pack_name, len + namelen - 4))
1181 break;
1182 }
1183 if (p == NULL &&
1184 /*
1185 * See if it really is a valid .idx file with
1186 * corresponding .pack file that we can map.
1187 */
1188 (p = add_packed_git(path, len + namelen, local)) != NULL)
1189 install_packed_git(p);
86f7780c 1190 }
543c5caa
NTND
1191
1192 if (!report_garbage)
1193 continue;
1194
1195 if (has_extension(de->d_name, ".idx") ||
1196 has_extension(de->d_name, ".pack") ||
1197 has_extension(de->d_name, ".keep"))
1198 string_list_append(&garbage, path);
1199 else
1200 report_garbage("garbage found", path);
1f688557 1201 }
5b35bcd5 1202 closedir(dir);
543c5caa
NTND
1203 report_pack_garbage(&garbage);
1204 string_list_clear(&garbage, 0);
1f688557
JH
1205}
1206
b867092f
JH
1207static int sort_pack(const void *a_, const void *b_)
1208{
1209 struct packed_git *a = *((struct packed_git **)a_);
1210 struct packed_git *b = *((struct packed_git **)b_);
1211 int st;
1212
1213 /*
1214 * Local packs tend to contain objects specific to our
1215 * variant of the project than remote ones. In addition,
1216 * remote ones could be on a network mounted filesystem.
1217 * Favor local ones for these reasons.
1218 */
1219 st = a->pack_local - b->pack_local;
1220 if (st)
1221 return -st;
1222
1223 /*
1224 * Younger packs tend to contain more recent objects,
1225 * and more recent objects tend to get accessed more
1226 * often.
1227 */
1228 if (a->mtime < b->mtime)
1229 return 1;
1230 else if (a->mtime == b->mtime)
1231 return 0;
1232 return -1;
1233}
1234
1235static void rearrange_packed_git(void)
1236{
1237 struct packed_git **ary, *p;
1238 int i, n;
1239
1240 for (n = 0, p = packed_git; p; p = p->next)
1241 n++;
1242 if (n < 2)
1243 return;
1244
1245 /* prepare an array of packed_git for easier sorting */
1246 ary = xcalloc(n, sizeof(struct packed_git *));
1247 for (n = 0, p = packed_git; p; p = p->next)
1248 ary[n++] = p;
1249
1250 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1251
1252 /* link them back again */
1253 for (i = 0; i < n - 1; i++)
1254 ary[i]->next = ary[i + 1];
1255 ary[n - 1]->next = NULL;
1256 packed_git = ary[0];
1257
1258 free(ary);
1259}
1260
637cdd9d 1261static int prepare_packed_git_run_once = 0;
9a217f2a 1262void prepare_packed_git(void)
1f688557 1263{
d5a63b99 1264 struct alternate_object_database *alt;
1f688557 1265
637cdd9d 1266 if (prepare_packed_git_run_once)
1f688557 1267 return;
9d835df2 1268 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 1269 prepare_alt_odb();
d5a63b99 1270 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 1271 alt->name[-1] = 0;
9d835df2 1272 prepare_packed_git_one(alt->base, 0);
1494e038 1273 alt->name[-1] = '/';
1f688557 1274 }
b867092f 1275 rearrange_packed_git();
637cdd9d
JK
1276 prepare_packed_git_run_once = 1;
1277}
1278
fc04c412 1279void reprepare_packed_git(void)
637cdd9d 1280{
4b480c67 1281 discard_revindex();
637cdd9d
JK
1282 prepare_packed_git_run_once = 0;
1283 prepare_packed_git();
1f688557
JH
1284}
1285
8eca0b47
NP
1286static void mark_bad_packed_object(struct packed_git *p,
1287 const unsigned char *sha1)
1288{
1289 unsigned i;
1290 for (i = 0; i < p->num_bad_objects; i++)
1291 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1292 return;
1293 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1294 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1295 p->num_bad_objects++;
1296}
1297
b6c4cecc 1298static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
ac939109
NP
1299{
1300 struct packed_git *p;
1301 unsigned i;
1302
1303 for (p = packed_git; p; p = p->next)
1304 for (i = 0; i < p->num_bad_objects; i++)
1305 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
b6c4cecc
JH
1306 return p;
1307 return NULL;
ac939109
NP
1308}
1309
090ea126
NTND
1310/*
1311 * With an in-core object data in "map", rehash it to make sure the
1312 * object name actually matches "sha1" to detect object corruption.
1313 * With "map" == NULL, try reading the object named with "sha1" using
1314 * the streaming interface and rehash it to do the same.
1315 */
1316int check_sha1_signature(const unsigned char *sha1, void *map,
1317 unsigned long size, const char *type)
0fcfd160
LT
1318{
1319 unsigned char real_sha1[20];
090ea126
NTND
1320 enum object_type obj_type;
1321 struct git_istream *st;
1322 git_SHA_CTX c;
1323 char hdr[32];
1324 int hdrlen;
1325
1326 if (map) {
1327 hash_sha1_file(map, size, type, real_sha1);
1328 return hashcmp(sha1, real_sha1) ? -1 : 0;
1329 }
1330
1331 st = open_istream(sha1, &obj_type, &size, NULL);
1332 if (!st)
1333 return -1;
1334
1335 /* Generate the header */
1336 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1337
1338 /* Sha1.. */
1339 git_SHA1_Init(&c);
1340 git_SHA1_Update(&c, hdr, hdrlen);
1341 for (;;) {
1342 char buf[1024 * 16];
1343 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1344
f54fac53
JK
1345 if (readlen < 0) {
1346 close_istream(st);
1347 return -1;
1348 }
090ea126
NTND
1349 if (!readlen)
1350 break;
1351 git_SHA1_Update(&c, buf, readlen);
1352 }
1353 git_SHA1_Final(real_sha1, &c);
1354 close_istream(st);
a89fccd2 1355 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1356}
1357
f4e51683 1358static int git_open_noatime(const char *name)
44d1c19e
LT
1359{
1360 static int sha1_file_open_flag = O_NOATIME;
44d1c19e 1361
f2e872aa
SP
1362 for (;;) {
1363 int fd = open(name, O_RDONLY | sha1_file_open_flag);
44d1c19e 1364 if (fd >= 0)
f2e872aa
SP
1365 return fd;
1366
f2e872aa
SP
1367 /* Might the failure be due to O_NOATIME? */
1368 if (errno != ENOENT && sha1_file_open_flag) {
44d1c19e 1369 sha1_file_open_flag = 0;
f2e872aa
SP
1370 continue;
1371 }
1372
1373 return -1;
44d1c19e 1374 }
44d1c19e
LT
1375}
1376
052fe5ea
JK
1377static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1378{
1379 char *name = sha1_file_name(sha1);
1380 struct alternate_object_database *alt;
1381
1382 if (!lstat(name, st))
1383 return 0;
1384
1385 prepare_alt_odb();
1386 errno = ENOENT;
1387 for (alt = alt_odb_list; alt; alt = alt->next) {
1388 name = alt->name;
1389 fill_sha1_path(name, sha1);
1390 if (!lstat(alt->base, st))
1391 return 0;
1392 }
1393
1394 return -1;
1395}
1396
44d1c19e
LT
1397static int open_sha1_file(const unsigned char *sha1)
1398{
1399 int fd;
1400 char *name = sha1_file_name(sha1);
1401 struct alternate_object_database *alt;
1402
f4e51683 1403 fd = git_open_noatime(name);
44d1c19e
LT
1404 if (fd >= 0)
1405 return fd;
1406
1407 prepare_alt_odb();
1408 errno = ENOENT;
1409 for (alt = alt_odb_list; alt; alt = alt->next) {
1410 name = alt->name;
1411 fill_sha1_path(name, sha1);
f4e51683 1412 fd = git_open_noatime(alt->base);
44d1c19e
LT
1413 if (fd >= 0)
1414 return fd;
1415 }
1416 return -1;
1417}
1418
f0270efd 1419void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1420{
0fcfd160 1421 void *map;
144bde78 1422 int fd;
ace1534d 1423
44d1c19e
LT
1424 fd = open_sha1_file(sha1);
1425 map = NULL;
1426 if (fd >= 0) {
1427 struct stat st;
0fcfd160 1428
44d1c19e
LT
1429 if (!fstat(fd, &st)) {
1430 *size = xsize_t(st.st_size);
33e42de0
MM
1431 if (!*size) {
1432 /* mmap() is forbidden on empty files */
1433 error("object file %s is empty", sha1_file_name(sha1));
1434 return NULL;
1435 }
44d1c19e 1436 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1437 }
44d1c19e 1438 close(fd);
0fcfd160 1439 }
0fcfd160
LT
1440 return map;
1441}
1442
cc5c54e7
JH
1443/*
1444 * There used to be a second loose object header format which
1445 * was meant to mimic the in-pack format, allowing for direct
1446 * copy of the object data. This format turned up not to be
1447 * really worth it and we no longer write loose objects in that
1448 * format.
1449 */
1450static int experimental_loose_object(unsigned char *map)
bb6b8e4f
JH
1451{
1452 unsigned int word;
1453
1454 /*
7f684a2a
RT
1455 * We must determine if the buffer contains the standard
1456 * zlib-deflated stream or the experimental format based
1457 * on the in-pack object format. Compare the header byte
1458 * for each format:
1459 *
1460 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1461 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1462 *
1463 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1464 * in standard loose-object format, UNLESS it is a Git-pack
1465 * format object *exactly* 8 bytes in size when inflated.
1466 *
1467 * However, RFC1950 also specifies that the 1st 16-bit word
1468 * must be divisible by 31 - this checksum tells us our buffer
1469 * is in the standard format, giving a false positive only if
1470 * the 1st word of the Git-pack format object happens to be
1471 * divisible by 31, ie:
1472 * ((byte0 * 256) + byte1) % 31 = 0
1473 * => 0ttt10000www1000 % 31 = 0
1474 *
1475 * As it happens, this case can only arise for www=3 & ttt=1
1476 * - ie, a Commit object, which would have to be 8 bytes in
1477 * size. As no Commit can be that small, we find that the
1478 * combination of these two criteria (bitmask & checksum)
1479 * can always correctly determine the buffer format.
bb6b8e4f
JH
1480 */
1481 word = (map[0] << 8) + map[1];
7f684a2a 1482 if ((map[0] & 0x8F) == 0x08 && !(word % 31))
bb6b8e4f 1483 return 0;
cc5c54e7
JH
1484 else
1485 return 1;
bb6b8e4f
JH
1486}
1487
09ded04b
NP
1488unsigned long unpack_object_header_buffer(const unsigned char *buf,
1489 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1490{
ad1ed5ee 1491 unsigned shift;
48fb7deb 1492 unsigned long size, c;
ad1ed5ee
JH
1493 unsigned long used = 0;
1494
1495 c = buf[used++];
1496 *type = (c >> 4) & 7;
1497 size = c & 15;
1498 shift = 4;
1499 while (c & 0x80) {
f630cfda 1500 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1501 error("bad object header");
ea4f9685
JH
1502 size = used = 0;
1503 break;
09ded04b 1504 }
ad1ed5ee
JH
1505 c = buf[used++];
1506 size += (c & 0x7f) << shift;
1507 shift += 7;
1508 }
1509 *sizep = size;
1510 return used;
1511}
1512
eb4f4076 1513int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
ad1ed5ee
JH
1514{
1515 unsigned long size, used;
1516 static const char valid_loose_object_type[8] = {
1517 0, /* OBJ_EXT */
1518 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1519 0, /* "delta" and others are invalid in a loose object */
93821bd9 1520 };
ad1ed5ee 1521 enum object_type type;
93821bd9 1522
c4483576
LT
1523 /* Get the data stream */
1524 memset(stream, 0, sizeof(*stream));
1525 stream->next_in = map;
1526 stream->avail_in = mapsize;
1527 stream->next_out = buffer;
93821bd9
LT
1528 stream->avail_out = bufsiz;
1529
cc5c54e7
JH
1530 if (experimental_loose_object(map)) {
1531 /*
1532 * The old experimental format we no longer produce;
1533 * we can still read it.
1534 */
1535 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1536 if (!used || !valid_loose_object_type[type])
1537 return -1;
1538 map += used;
1539 mapsize -= used;
726f852b 1540
cc5c54e7
JH
1541 /* Set up the stream for the rest.. */
1542 stream->next_in = map;
1543 stream->avail_in = mapsize;
1544 git_inflate_init(stream);
93821bd9 1545
cc5c54e7
JH
1546 /* And generate the fake traditional header */
1547 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1548 typename(type), size);
1549 return 0;
1550 }
39c68542 1551 git_inflate_init(stream);
cc5c54e7 1552 return git_inflate(stream, 0);
c4483576
LT
1553}
1554
ef49a7a0 1555static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1556{
1557 int bytes = strlen(buffer) + 1;
3aee68aa 1558 unsigned char *buf = xmallocz(size);
93821bd9 1559 unsigned long n;
7efbff75 1560 int status = Z_OK;
5180cacc 1561
93821bd9
LT
1562 n = stream->total_out - bytes;
1563 if (n > size)
1564 n = size;
1565 memcpy(buf, (char *) buffer + bytes, n);
1566 bytes = n;
456cdf6e
LT
1567 if (bytes <= size) {
1568 /*
1569 * The above condition must be (bytes <= size), not
1570 * (bytes < size). In other words, even though we
ccf5ace0 1571 * expect no more output and set avail_out to zero,
456cdf6e
LT
1572 * the input zlib stream may have bytes that express
1573 * "this concludes the stream", and we *do* want to
1574 * eat that input.
1575 *
1576 * Otherwise we would not be able to test that we
1577 * consumed all the input to reach the expected size;
1578 * we also want to check that zlib tells us that all
1579 * went well with status == Z_STREAM_END at the end.
1580 */
5180cacc
LT
1581 stream->next_out = buf + bytes;
1582 stream->avail_out = size - bytes;
7efbff75 1583 while (status == Z_OK)
39c68542 1584 status = git_inflate(stream, Z_FINISH);
5180cacc 1585 }
456cdf6e 1586 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1587 git_inflate_end(stream);
7efbff75
JH
1588 return buf;
1589 }
1590
1591 if (status < 0)
1592 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1593 else if (stream->avail_in)
1594 error("garbage at end of loose object '%s'",
1595 sha1_to_hex(sha1));
1596 free(buf);
1597 return NULL;
5180cacc
LT
1598}
1599
1600/*
1601 * We used to just use "sscanf()", but that's actually way
1602 * too permissive for what we want to check. So do an anal
1603 * object header parse by hand.
1604 */
f0270efd 1605int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1606{
21666f1a 1607 char type[10];
5180cacc
LT
1608 int i;
1609 unsigned long size;
1610
1611 /*
a6080a0a 1612 * The type can be at most ten bytes (including the
5180cacc 1613 * terminating '\0' that we add), and is followed by
21666f1a 1614 * a space.
5180cacc 1615 */
21666f1a 1616 i = 0;
5180cacc
LT
1617 for (;;) {
1618 char c = *hdr++;
1619 if (c == ' ')
1620 break;
21666f1a
NP
1621 type[i++] = c;
1622 if (i >= sizeof(type))
5180cacc 1623 return -1;
5180cacc 1624 }
21666f1a 1625 type[i] = 0;
5180cacc
LT
1626
1627 /*
1628 * The length must follow immediately, and be in canonical
1629 * decimal format (ie "010" is not valid).
1630 */
1631 size = *hdr++ - '0';
1632 if (size > 9)
1633 return -1;
1634 if (size) {
1635 for (;;) {
1636 unsigned long c = *hdr - '0';
1637 if (c > 9)
1638 break;
1639 hdr++;
1640 size = size * 10 + c;
1641 }
1642 }
1643 *sizep = size;
1644
1645 /*
1646 * The length must be followed by a zero byte
1647 */
21666f1a 1648 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1649}
1650
7efbff75 1651static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1652{
5180cacc 1653 int ret;
ef49a7a0 1654 git_zstream stream;
5180cacc 1655 char hdr[8192];
0fcfd160 1656
5180cacc 1657 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1658 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1659 return NULL;
1660
7efbff75 1661 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1662}
1663
54dab52a
NP
1664unsigned long get_size_from_delta(struct packed_git *p,
1665 struct pack_window **w_curs,
1666 off_t curpos)
1667{
1668 const unsigned char *data;
1669 unsigned char delta_head[20], *in;
ef49a7a0 1670 git_zstream stream;
54dab52a
NP
1671 int st;
1672
1673 memset(&stream, 0, sizeof(stream));
1674 stream.next_out = delta_head;
1675 stream.avail_out = sizeof(delta_head);
1676
39c68542 1677 git_inflate_init(&stream);
54dab52a
NP
1678 do {
1679 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1680 stream.next_in = in;
39c68542 1681 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1682 curpos += stream.next_in - in;
1683 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1684 stream.total_out < sizeof(delta_head));
39c68542 1685 git_inflate_end(&stream);
3d77d877
NP
1686 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1687 error("delta data unpack-initial failed");
1688 return 0;
1689 }
54dab52a
NP
1690
1691 /* Examine the initial part of the delta to figure out
1692 * the result size.
1693 */
1694 data = delta_head;
1695
1696 /* ignore base size */
1697 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1698
1699 /* Read the result size */
1700 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1701}
1702
c4001d92 1703static off_t get_delta_base(struct packed_git *p,
03e79c88 1704 struct pack_window **w_curs,
c4001d92 1705 off_t *curpos,
21666f1a 1706 enum object_type type,
c4001d92 1707 off_t delta_obj_offset)
eb32d236 1708{
2b87c45b 1709 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1710 off_t base_offset;
eb32d236 1711
8d8a4ea5
SP
1712 /* use_pack() assured us we have [base_info, base_info + 20)
1713 * as a range that we can look at without walking off the
1714 * end of the mapped window. Its actually the hash size
1715 * that is assured. An OFS_DELTA longer than the hash size
1716 * is stupid, as then a REF_DELTA would be smaller to store.
1717 */
21666f1a 1718 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1719 unsigned used = 0;
1720 unsigned char c = base_info[used++];
1721 base_offset = c & 127;
1722 while (c & 128) {
1723 base_offset += 1;
8723f216 1724 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1725 return 0; /* overflow */
eb32d236
NP
1726 c = base_info[used++];
1727 base_offset = (base_offset << 7) + (c & 127);
1728 }
1729 base_offset = delta_obj_offset - base_offset;
d8f32556 1730 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1731 return 0; /* out of bound */
2b87c45b 1732 *curpos += used;
21666f1a 1733 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1734 /* The base entry _must_ be in the same pack */
1735 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1736 *curpos += 20;
eb32d236
NP
1737 } else
1738 die("I am totally screwed");
2b87c45b 1739 return base_offset;
eb32d236
NP
1740}
1741
f8c8abc5
JH
1742int unpack_object_header(struct packed_git *p,
1743 struct pack_window **w_curs,
1744 off_t *curpos,
1745 unsigned long *sizep)
a733cb60 1746{
03e79c88 1747 unsigned char *base;
ef49a7a0 1748 unsigned long left;
ad1ed5ee 1749 unsigned long used;
2b87c45b 1750 enum object_type type;
a733cb60 1751
8d8a4ea5 1752 /* use_pack() assures us we have [base, base + 20) available
0353a0c4 1753 * as a range that we can look at. (Its actually the hash
3dff5379 1754 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1755 * the maximum deflated object size is 2^137, which is just
1756 * insane, so we know won't exceed what we have been given.
1757 */
2b87c45b 1758 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1759 used = unpack_object_header_buffer(base, left, &type, sizep);
1760 if (!used) {
1761 type = OBJ_BAD;
1762 } else
1763 *curpos += used;
ad1ed5ee 1764
2b87c45b 1765 return type;
a733cb60
LT
1766}
1767
790d96c0
TR
1768static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1769{
1770 int type;
1771 struct revindex_entry *revidx;
1772 const unsigned char *sha1;
1773 revidx = find_pack_revindex(p, obj_offset);
1774 if (!revidx)
1775 return OBJ_BAD;
1776 sha1 = nth_packed_object_sha1(p, revidx->nr);
1777 mark_bad_packed_object(p, sha1);
1778 type = sha1_object_info(sha1, NULL);
1779 if (type <= OBJ_NONE)
1780 return OBJ_BAD;
1781 return type;
1782}
1783
790d96c0
TR
1784#define POI_STACK_PREALLOC 64
1785
90191d37
JK
1786static enum object_type packed_to_object_type(struct packed_git *p,
1787 off_t obj_offset,
1788 enum object_type type,
1789 struct pack_window **w_curs,
1790 off_t curpos)
1f688557 1791{
790d96c0
TR
1792 off_t small_poi_stack[POI_STACK_PREALLOC];
1793 off_t *poi_stack = small_poi_stack;
1794 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
5db47c2b 1795
790d96c0
TR
1796 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1797 off_t base_offset;
90191d37 1798 unsigned long size;
790d96c0
TR
1799 /* Push the object we're going to leave behind */
1800 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1801 poi_stack_alloc = alloc_nr(poi_stack_nr);
1802 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1803 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1804 } else {
1805 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1806 }
1807 poi_stack[poi_stack_nr++] = obj_offset;
1808 /* If parsing the base offset fails, just unwind */
90191d37 1809 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
790d96c0
TR
1810 if (!base_offset)
1811 goto unwind;
1812 curpos = obj_offset = base_offset;
90191d37 1813 type = unpack_object_header(p, w_curs, &curpos, &size);
790d96c0
TR
1814 if (type <= OBJ_NONE) {
1815 /* If getting the base itself fails, we first
1816 * retry the base, otherwise unwind */
1817 type = retry_bad_packed_offset(p, base_offset);
1818 if (type > OBJ_NONE)
1819 goto out;
1820 goto unwind;
1821 }
1822 }
1823
21666f1a 1824 switch (type) {
790d96c0 1825 case OBJ_BAD:
a733cb60 1826 case OBJ_COMMIT:
a733cb60 1827 case OBJ_TREE:
a733cb60 1828 case OBJ_BLOB:
a733cb60 1829 case OBJ_TAG:
a69d0943 1830 break;
1f688557 1831 default:
3d77d877
NP
1832 error("unknown object type %i at offset %"PRIuMAX" in %s",
1833 type, (uintmax_t)obj_offset, p->pack_name);
1834 type = OBJ_BAD;
1f688557 1835 }
790d96c0
TR
1836
1837out:
1838 if (poi_stack != small_poi_stack)
1839 free(poi_stack);
21666f1a 1840 return type;
790d96c0
TR
1841
1842unwind:
1843 while (poi_stack_nr) {
1844 obj_offset = poi_stack[--poi_stack_nr];
1845 type = retry_bad_packed_offset(p, obj_offset);
1846 if (type > OBJ_NONE)
1847 goto out;
1848 }
1849 type = OBJ_BAD;
1850 goto out;
1f688557
JH
1851}
1852
90191d37 1853static int packed_object_info(struct packed_git *p, off_t obj_offset,
23c339c0 1854 struct object_info *oi)
90191d37
JK
1855{
1856 struct pack_window *w_curs = NULL;
1857 unsigned long size;
1858 off_t curpos = obj_offset;
1859 enum object_type type;
1860
412916ee
JK
1861 /*
1862 * We always get the representation type, but only convert it to
1863 * a "real" type later if the caller is interested.
1864 */
90191d37
JK
1865 type = unpack_object_header(p, &w_curs, &curpos, &size);
1866
23c339c0 1867 if (oi->sizep) {
90191d37
JK
1868 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1869 off_t tmp_pos = curpos;
1870 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1871 type, obj_offset);
1872 if (!base_offset) {
1873 type = OBJ_BAD;
1874 goto out;
1875 }
23c339c0
JK
1876 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1877 if (*oi->sizep == 0) {
90191d37
JK
1878 type = OBJ_BAD;
1879 goto out;
1880 }
1881 } else {
23c339c0 1882 *oi->sizep = size;
90191d37
JK
1883 }
1884 }
1885
23c339c0 1886 if (oi->disk_sizep) {
90191d37 1887 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
23c339c0 1888 *oi->disk_sizep = revidx[1].offset - obj_offset;
90191d37
JK
1889 }
1890
23c339c0
JK
1891 if (oi->typep) {
1892 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
1893 if (*oi->typep < 0) {
412916ee
JK
1894 type = OBJ_BAD;
1895 goto out;
1896 }
1897 }
90191d37
JK
1898
1899out:
1900 unuse_pack(&w_curs);
1901 return type;
1902}
1903
eb950c19 1904static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1905 struct pack_window **w_curs,
c4001d92 1906 off_t curpos,
eb950c19 1907 unsigned long size)
de530aaa
SP
1908{
1909 int st;
ef49a7a0 1910 git_zstream stream;
079afb18 1911 unsigned char *buffer, *in;
de530aaa 1912
4ab07e4d 1913 buffer = xmallocz(size);
de530aaa 1914 memset(&stream, 0, sizeof(stream));
de530aaa 1915 stream.next_out = buffer;
39eea7bd 1916 stream.avail_out = size + 1;
de530aaa 1917
39c68542 1918 git_inflate_init(&stream);
079afb18 1919 do {
2b87c45b 1920 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 1921 stream.next_in = in;
39c68542 1922 st = git_inflate(&stream, Z_FINISH);
39eea7bd
JH
1923 if (!stream.avail_out)
1924 break; /* the payload is larger than it should be */
2b87c45b 1925 curpos += stream.next_in - in;
079afb18 1926 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 1927 git_inflate_end(&stream);
de530aaa
SP
1928 if ((st != Z_STREAM_END) || stream.total_out != size) {
1929 free(buffer);
1930 return NULL;
1931 }
1932
1933 return buffer;
1934}
1935
e5e01619
LT
1936#define MAX_DELTA_CACHE (256)
1937
18bdec11 1938static size_t delta_base_cached;
5e08ecbf
NP
1939
1940static struct delta_base_cache_lru_list {
1941 struct delta_base_cache_lru_list *prev;
1942 struct delta_base_cache_lru_list *next;
1943} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1944
e5e01619 1945static struct delta_base_cache_entry {
5e08ecbf
NP
1946 struct delta_base_cache_lru_list lru;
1947 void *data;
e5e01619
LT
1948 struct packed_git *p;
1949 off_t base_offset;
1950 unsigned long size;
e5e01619
LT
1951 enum object_type type;
1952} delta_base_cache[MAX_DELTA_CACHE];
1953
1954static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1955{
1956 unsigned long hash;
1957
1958 hash = (unsigned long)p + (unsigned long)base_offset;
1959 hash += (hash >> 8) + (hash >> 16);
3358004a 1960 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1961}
1962
84dd81c1
TR
1963static struct delta_base_cache_entry *
1964get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
5266d369
JH
1965{
1966 unsigned long hash = pack_entry_hash(p, base_offset);
84dd81c1
TR
1967 return delta_base_cache + hash;
1968}
1969
1970static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
1971 struct packed_git *p, off_t base_offset)
1972{
5266d369
JH
1973 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1974}
1975
84dd81c1
TR
1976static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1977{
1978 struct delta_base_cache_entry *ent;
1979 ent = get_delta_base_cache_entry(p, base_offset);
1980 return eq_delta_base_cache_entry(ent, p, base_offset);
1981}
1982
1983static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1984{
1985 ent->data = NULL;
1986 ent->lru.next->prev = ent->lru.prev;
1987 ent->lru.prev->next = ent->lru.next;
1988 delta_base_cached -= ent->size;
1989}
1990
62f255ad 1991static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 1992 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 1993{
84dd81c1 1994 struct delta_base_cache_entry *ent;
e5e01619 1995 void *ret;
e5e01619 1996
84dd81c1
TR
1997 ent = get_delta_base_cache_entry(p, base_offset);
1998
1999 if (!eq_delta_base_cache_entry(ent, p, base_offset))
749bc58c 2000 return unpack_entry(p, base_offset, type, base_size);
e5e01619 2001
84dd81c1
TR
2002 ret = ent->data;
2003
2004 if (!keep_cache)
2005 clear_delta_base_cache_entry(ent);
2006 else
182af834 2007 ret = xmemdupz(ent->data, ent->size);
e5e01619
LT
2008 *type = ent->type;
2009 *base_size = ent->size;
2010 return ret;
62f255ad
LT
2011}
2012
18bdec11
SP
2013static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2014{
2015 if (ent->data) {
2016 free(ent->data);
2017 ent->data = NULL;
5e08ecbf
NP
2018 ent->lru.next->prev = ent->lru.prev;
2019 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
2020 delta_base_cached -= ent->size;
2021 }
2022}
2023
3d20c636
SP
2024void clear_delta_base_cache(void)
2025{
2026 unsigned long p;
2027 for (p = 0; p < MAX_DELTA_CACHE; p++)
2028 release_delta_base_cache(&delta_base_cache[p]);
2029}
2030
62f255ad
LT
2031static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2032 void *base, unsigned long base_size, enum object_type type)
2033{
5e08ecbf 2034 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 2035 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 2036 struct delta_base_cache_lru_list *lru;
e5e01619 2037
18bdec11
SP
2038 release_delta_base_cache(ent);
2039 delta_base_cached += base_size;
5e08ecbf
NP
2040
2041 for (lru = delta_base_cache_lru.next;
2042 delta_base_cached > delta_base_cache_limit
2043 && lru != &delta_base_cache_lru;
2044 lru = lru->next) {
2045 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
2046 if (f->type == OBJ_BLOB)
2047 release_delta_base_cache(f);
2048 }
5e08ecbf
NP
2049 for (lru = delta_base_cache_lru.next;
2050 delta_base_cached > delta_base_cache_limit
2051 && lru != &delta_base_cache_lru;
2052 lru = lru->next) {
2053 struct delta_base_cache_entry *f = (void *)lru;
2054 release_delta_base_cache(f);
2055 }
18bdec11 2056
e5e01619
LT
2057 ent->p = p;
2058 ent->base_offset = base_offset;
2059 ent->type = type;
2060 ent->data = base;
2061 ent->size = base_size;
5e08ecbf
NP
2062 ent->lru.next = &delta_base_cache_lru;
2063 ent->lru.prev = delta_base_cache_lru.prev;
2064 delta_base_cache_lru.prev->next = &ent->lru;
2065 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
2066}
2067
c2c5b270
CC
2068static void *read_object(const unsigned char *sha1, enum object_type *type,
2069 unsigned long *size);
2070
5f44324d
JH
2071static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2072{
2073 static FILE *log_file;
2074
b12ca963
NTND
2075 if (!log_pack_access)
2076 log_pack_access = getenv("GIT_TRACE_PACK_ACCESS");
2077 if (!log_pack_access)
2078 log_pack_access = no_log_pack_access;
2079 if (log_pack_access == no_log_pack_access)
2080 return;
2081
5f44324d
JH
2082 if (!log_file) {
2083 log_file = fopen(log_pack_access, "w");
2084 if (!log_file) {
2085 error("cannot open pack access log '%s' for writing: %s",
2086 log_pack_access, strerror(errno));
b12ca963 2087 log_pack_access = no_log_pack_access;
5f44324d
JH
2088 return;
2089 }
2090 }
2091 fprintf(log_file, "%s %"PRIuMAX"\n",
2092 p->pack_name, (uintmax_t)obj_offset);
2093 fflush(log_file);
2094}
2095
0e8189e2
NP
2096int do_check_packed_object_crc;
2097
abe601bb
TR
2098#define UNPACK_ENTRY_STACK_PREALLOC 64
2099struct unpack_entry_stack_ent {
2100 off_t obj_offset;
2101 off_t curpos;
2102 unsigned long size;
2103};
2104
c4001d92 2105void *unpack_entry(struct packed_git *p, off_t obj_offset,
abe601bb 2106 enum object_type *final_type, unsigned long *final_size)
1f688557 2107{
03e79c88 2108 struct pack_window *w_curs = NULL;
c4001d92 2109 off_t curpos = obj_offset;
abe601bb
TR
2110 void *data = NULL;
2111 unsigned long size;
2112 enum object_type type;
2113 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2114 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2115 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2116 int base_from_cache = 0;
1f688557 2117
b12ca963 2118 if (log_pack_access != no_log_pack_access)
5f44324d
JH
2119 write_pack_access_log(p, obj_offset);
2120
abe601bb
TR
2121 /* PHASE 1: drill down to the innermost base object */
2122 for (;;) {
2123 off_t base_offset;
2124 int i;
2125 struct delta_base_cache_entry *ent;
2126
77965f8b
NTND
2127 ent = get_delta_base_cache_entry(p, curpos);
2128 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2129 type = ent->type;
2130 data = ent->data;
2131 size = ent->size;
2132 clear_delta_base_cache_entry(ent);
2133 base_from_cache = 1;
2134 break;
2135 }
2136
abe601bb
TR
2137 if (do_check_packed_object_crc && p->index_version > 1) {
2138 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2139 unsigned long len = revidx[1].offset - obj_offset;
2140 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2141 const unsigned char *sha1 =
2142 nth_packed_object_sha1(p, revidx->nr);
2143 error("bad packed object CRC for %s",
2144 sha1_to_hex(sha1));
2145 mark_bad_packed_object(p, sha1);
2146 unuse_pack(&w_curs);
2147 return NULL;
2148 }
2149 }
2150
abe601bb
TR
2151 type = unpack_object_header(p, &w_curs, &curpos, &size);
2152 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2153 break;
2154
2155 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2156 if (!base_offset) {
2157 error("failed to validate delta base reference "
2158 "at offset %"PRIuMAX" from %s",
2159 (uintmax_t)curpos, p->pack_name);
2160 /* bail to phase 2, in hopes of recovery */
2161 data = NULL;
2162 break;
2163 }
2164
2165 /* push object, proceed to base */
2166 if (delta_stack_nr >= delta_stack_alloc
2167 && delta_stack == small_delta_stack) {
2168 delta_stack_alloc = alloc_nr(delta_stack_nr);
2169 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2170 memcpy(delta_stack, small_delta_stack,
2171 sizeof(*delta_stack)*delta_stack_nr);
2172 } else {
2173 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2174 }
2175 i = delta_stack_nr++;
2176 delta_stack[i].obj_offset = obj_offset;
2177 delta_stack[i].curpos = curpos;
2178 delta_stack[i].size = size;
2179
2180 curpos = obj_offset = base_offset;
0e8189e2
NP
2181 }
2182
abe601bb
TR
2183 /* PHASE 2: handle the base */
2184 switch (type) {
eb32d236
NP
2185 case OBJ_OFS_DELTA:
2186 case OBJ_REF_DELTA:
abe601bb
TR
2187 if (data)
2188 die("BUG in unpack_entry: left loop at a valid delta");
4d703a1a 2189 break;
a733cb60 2190 case OBJ_COMMIT:
a733cb60 2191 case OBJ_TREE:
a733cb60 2192 case OBJ_BLOB:
a733cb60 2193 case OBJ_TAG:
abe601bb
TR
2194 if (!base_from_cache)
2195 data = unpack_compressed_entry(p, &w_curs, curpos, size);
4d703a1a 2196 break;
1f688557 2197 default:
8eca0b47
NP
2198 data = NULL;
2199 error("unknown object type %i at offset %"PRIuMAX" in %s",
abe601bb
TR
2200 type, (uintmax_t)obj_offset, p->pack_name);
2201 }
2202
2203 /* PHASE 3: apply deltas in order */
2204
2205 /* invariants:
2206 * 'data' holds the base data, or NULL if there was corruption
2207 */
2208 while (delta_stack_nr) {
2209 void *delta_data;
2210 void *base = data;
2211 unsigned long delta_size, base_size = size;
2212 int i;
2213
2214 data = NULL;
2215
2216 if (base)
2217 add_delta_base_cache(p, obj_offset, base, base_size, type);
2218
2219 if (!base) {
2220 /*
2221 * We're probably in deep shit, but let's try to fetch
2222 * the required base anyway from another pack or loose.
2223 * This is costly but should happen only in the presence
2224 * of a corrupted pack, and is better than failing outright.
2225 */
2226 struct revindex_entry *revidx;
2227 const unsigned char *base_sha1;
2228 revidx = find_pack_revindex(p, obj_offset);
2229 if (revidx) {
2230 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2231 error("failed to read delta base object %s"
2232 " at offset %"PRIuMAX" from %s",
2233 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2234 p->pack_name);
2235 mark_bad_packed_object(p, base_sha1);
2236 base = read_object(base_sha1, &type, &base_size);
2237 }
2238 }
2239
2240 i = --delta_stack_nr;
2241 obj_offset = delta_stack[i].obj_offset;
2242 curpos = delta_stack[i].curpos;
2243 delta_size = delta_stack[i].size;
2244
2245 if (!base)
2246 continue;
2247
2248 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2249
2250 if (!delta_data) {
2251 error("failed to unpack compressed delta "
2252 "at offset %"PRIuMAX" from %s",
2253 (uintmax_t)curpos, p->pack_name);
abe601bb
TR
2254 data = NULL;
2255 continue;
2256 }
2257
2258 data = patch_delta(base, base_size,
2259 delta_data, delta_size,
2260 &size);
1ee886c1
JK
2261
2262 /*
2263 * We could not apply the delta; warn the user, but keep going.
2264 * Our failure will be noticed either in the next iteration of
2265 * the loop, or if this is the final delta, in the caller when
2266 * we return NULL. Those code paths will take care of making
2267 * a more explicit warning and retrying with another copy of
2268 * the object.
2269 */
abe601bb 2270 if (!data)
1ee886c1 2271 error("failed to apply delta");
abe601bb 2272
4b8f772c 2273 free(delta_data);
1f688557 2274 }
abe601bb
TR
2275
2276 *final_type = type;
2277 *final_size = size;
2278
03e79c88 2279 unuse_pack(&w_curs);
21666f1a 2280 return data;
1f688557
JH
2281}
2282
d079837e 2283const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 2284 uint32_t n)
9a217f2a 2285{
42873078 2286 const unsigned char *index = p->index_data;
d079837e
SP
2287 if (!index) {
2288 if (open_pack_index(p))
2289 return NULL;
2290 index = p->index_data;
2291 }
57059091 2292 if (n >= p->num_objects)
d72308e0 2293 return NULL;
74e34e1f
NP
2294 index += 4 * 256;
2295 if (p->index_version == 1) {
2296 return index + 24 * n + 4;
2297 } else {
2298 index += 8;
2299 return index + 20 * n;
2300 }
2301}
2302
99093238 2303off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
2304{
2305 const unsigned char *index = p->index_data;
2306 index += 4 * 256;
2307 if (p->index_version == 1) {
2308 return ntohl(*((uint32_t *)(index + 24 * n)));
2309 } else {
2310 uint32_t off;
2311 index += 8 + p->num_objects * (20 + 4);
2312 off = ntohl(*((uint32_t *)(index + 4 * n)));
2313 if (!(off & 0x80000000))
2314 return off;
2315 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2316 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2317 ntohl(*((uint32_t *)(index + 4)));
2318 }
9a217f2a
JH
2319}
2320
c4001d92 2321off_t find_pack_entry_one(const unsigned char *sha1,
43057304 2322 struct packed_git *p)
1f688557 2323{
42873078 2324 const uint32_t *level1_ofs = p->index_data;
42873078 2325 const unsigned char *index = p->index_data;
628522ec
JH
2326 unsigned hi, lo, stride;
2327 static int use_lookup = -1;
2328 static int debug_lookup = -1;
2329
2330 if (debug_lookup < 0)
2331 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 2332
d079837e
SP
2333 if (!index) {
2334 if (open_pack_index(p))
2335 return 0;
2336 level1_ofs = p->index_data;
2337 index = p->index_data;
2338 }
74e34e1f
NP
2339 if (p->index_version > 1) {
2340 level1_ofs += 2;
2341 index += 8;
2342 }
42873078 2343 index += 4 * 256;
74e34e1f
NP
2344 hi = ntohl(level1_ofs[*sha1]);
2345 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
2346 if (p->index_version > 1) {
2347 stride = 20;
2348 } else {
2349 stride = 24;
2350 index += 4;
2351 }
2352
2353 if (debug_lookup)
6e1c2344 2354 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
2355 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2356
2357 if (use_lookup < 0)
2358 use_lookup = !!getenv("GIT_USE_LOOKUP");
2359 if (use_lookup) {
2360 int pos = sha1_entry_pos(index, stride, 0,
2361 lo, hi, p->num_objects, sha1);
2362 if (pos < 0)
2363 return 0;
2364 return nth_packed_object_offset(p, pos);
2365 }
1f688557
JH
2366
2367 do {
74e34e1f 2368 unsigned mi = (lo + hi) / 2;
628522ec
JH
2369 int cmp = hashcmp(index + mi * stride, sha1);
2370
2371 if (debug_lookup)
2372 printf("lo %u hi %u rg %u mi %u\n",
2373 lo, hi, hi - lo, mi);
43057304 2374 if (!cmp)
74e34e1f 2375 return nth_packed_object_offset(p, mi);
1f688557
JH
2376 if (cmp > 0)
2377 hi = mi;
2378 else
2379 lo = mi+1;
2380 } while (lo < hi);
2381 return 0;
2382}
2383
4c080182 2384int is_pack_valid(struct packed_git *p)
d131b7af
SP
2385{
2386 /* An already open pack is known to be valid. */
2387 if (p->pack_fd != -1)
2388 return 1;
2389
2390 /* If the pack has one window completely covering the
2391 * file size, the pack is known to be valid even if
2392 * the descriptor is not currently open.
2393 */
2394 if (p->windows) {
2395 struct pack_window *w = p->windows;
2396
2397 if (!w->offset && w->len == p->pack_size)
2398 return 1;
2399 }
2400
2401 /* Force the pack to open to prove its valid. */
2402 return !open_packed_git(p);
2403}
2404
95099731
NTND
2405static int fill_pack_entry(const unsigned char *sha1,
2406 struct pack_entry *e,
2407 struct packed_git *p)
2408{
2409 off_t offset;
2410
2411 if (p->num_bad_objects) {
2412 unsigned i;
2413 for (i = 0; i < p->num_bad_objects; i++)
2414 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2415 return 0;
2416 }
2417
2418 offset = find_pack_entry_one(sha1, p);
2419 if (!offset)
2420 return 0;
2421
2422 /*
2423 * We are about to tell the caller where they can locate the
2424 * requested object. We better make sure the packfile is
2425 * still here and can be accessed before supplying that
2426 * answer, as it may have been deleted since the index was
2427 * loaded!
2428 */
2429 if (!is_pack_valid(p)) {
2430 warning("packfile %s cannot be accessed", p->pack_name);
2431 return 0;
2432 }
2433 e->offset = offset;
2434 e->p = p;
2435 hashcpy(e->sha1, sha1);
2436 return 1;
2437}
2438
4d6acb70 2439static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557
JH
2440{
2441 struct packed_git *p;
43057304 2442
1f688557 2443 prepare_packed_git();
f7c22cc6
NP
2444 if (!packed_git)
2445 return 0;
1f688557 2446
c01f51cc
NTND
2447 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2448 return 1;
8eca0b47 2449
c01f51cc
NTND
2450 for (p = packed_git; p; p = p->next) {
2451 if (p == last_found_pack || !fill_pack_entry(sha1, e, p))
2452 continue;
f7c22cc6 2453
c01f51cc
NTND
2454 last_found_pack = p;
2455 return 1;
2456 }
1f688557
JH
2457 return 0;
2458}
2459
a6080a0a 2460struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
2461 struct packed_git *packs)
2462{
2463 struct packed_git *p;
bf592c50
DB
2464
2465 for (p = packs; p; p = p->next) {
43057304 2466 if (find_pack_entry_one(sha1, p))
bf592c50
DB
2467 return p;
2468 }
2469 return NULL;
21666f1a 2470
bf592c50
DB
2471}
2472
052fe5ea 2473static int sha1_loose_object_info(const unsigned char *sha1,
23c339c0 2474 struct object_info *oi)
65c2e0c3 2475{
36e4d74a 2476 int status;
65c2e0c3
JH
2477 unsigned long mapsize, size;
2478 void *map;
ef49a7a0 2479 git_zstream stream;
d65a16f6 2480 char hdr[32];
65c2e0c3 2481
052fe5ea
JK
2482 /*
2483 * If we don't care about type or size, then we don't
4ef8d1dd
JH
2484 * need to look inside the object at all. Note that we
2485 * do not optimize out the stat call, even if the
2486 * caller doesn't care about the disk-size, since our
2487 * return value implicitly indicates whether the
2488 * object even exists.
052fe5ea 2489 */
23c339c0 2490 if (!oi->typep && !oi->sizep) {
4ef8d1dd
JH
2491 struct stat st;
2492 if (stat_sha1_file(sha1, &st) < 0)
2493 return -1;
2494 if (oi->disk_sizep)
23c339c0 2495 *oi->disk_sizep = st.st_size;
052fe5ea
JK
2496 return 0;
2497 }
2498
bb6b8e4f 2499 map = map_sha1_file(sha1, &mapsize);
f0df4ed5 2500 if (!map)
dbea72a8 2501 return -1;
23c339c0
JK
2502 if (oi->disk_sizep)
2503 *oi->disk_sizep = mapsize;
36e4d74a
JH
2504 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2505 status = error("unable to unpack %s header",
2506 sha1_to_hex(sha1));
21666f1a 2507 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 2508 status = error("unable to parse %s header", sha1_to_hex(sha1));
23c339c0
JK
2509 else if (oi->sizep)
2510 *oi->sizep = size;
39c68542 2511 git_inflate_end(&stream);
65c2e0c3 2512 munmap(map, mapsize);
23c339c0
JK
2513 if (oi->typep)
2514 *oi->typep = status;
052fe5ea 2515 return 0;
65c2e0c3
JH
2516}
2517
9a490590 2518int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
f0df4ed5 2519{
c4d9986f 2520 struct cached_object *co;
f0df4ed5 2521 struct pack_entry e;
5b086407 2522 int rtype;
f0df4ed5 2523
c4d9986f
NTND
2524 co = find_cached_object(sha1);
2525 if (co) {
5b086407
JK
2526 if (oi->typep)
2527 *(oi->typep) = co->type;
9a490590
JH
2528 if (oi->sizep)
2529 *(oi->sizep) = co->size;
161f00e7
JK
2530 if (oi->disk_sizep)
2531 *(oi->disk_sizep) = 0;
9a490590 2532 oi->whence = OI_CACHED;
5b086407 2533 return 0;
c4d9986f
NTND
2534 }
2535
cd673c1f 2536 if (!find_pack_entry(sha1, &e)) {
ddd63e64 2537 /* Most likely it's a loose object. */
23c339c0 2538 if (!sha1_loose_object_info(sha1, oi)) {
9a490590 2539 oi->whence = OI_LOOSE;
5b086407 2540 return 0;
9a490590 2541 }
ddd63e64
SG
2542
2543 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 2544 reprepare_packed_git();
cd673c1f 2545 if (!find_pack_entry(sha1, &e))
052fe5ea 2546 return -1;
f0df4ed5 2547 }
3d77d877 2548
23c339c0 2549 rtype = packed_object_info(e.p, e.offset, oi);
412916ee 2550 if (rtype < 0) {
3d77d877 2551 mark_bad_packed_object(e.p, sha1);
412916ee 2552 return sha1_object_info_extended(sha1, oi);
5266d369
JH
2553 } else if (in_delta_base_cache(e.p, e.offset)) {
2554 oi->whence = OI_DBCACHED;
9a490590
JH
2555 } else {
2556 oi->whence = OI_PACKED;
2557 oi->u.packed.offset = e.offset;
2558 oi->u.packed.pack = e.p;
2559 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2560 rtype == OBJ_OFS_DELTA);
3d77d877
NP
2561 }
2562
5b086407 2563 return 0;
f0df4ed5
JS
2564}
2565
3fc0dca9 2566/* returns enum object_type or negative */
9a490590
JH
2567int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2568{
5b086407 2569 enum object_type type;
d099b717 2570 struct object_info oi = {NULL};
9a490590 2571
5b086407 2572 oi.typep = &type;
9a490590 2573 oi.sizep = sizep;
5b086407
JK
2574 if (sha1_object_info_extended(sha1, &oi) < 0)
2575 return -1;
2576 return type;
9a490590
JH
2577}
2578
21666f1a
NP
2579static void *read_packed_sha1(const unsigned char *sha1,
2580 enum object_type *type, unsigned long *size)
1f688557
JH
2581{
2582 struct pack_entry e;
8eca0b47 2583 void *data;
1f688557 2584
cd673c1f 2585 if (!find_pack_entry(sha1, &e))
1f688557 2586 return NULL;
8eca0b47
NP
2587 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2588 if (!data) {
2589 /*
2590 * We're probably in deep shit, but let's try to fetch
2591 * the required object anyway from another pack or loose.
2592 * This should happen only in the presence of a corrupted
2593 * pack, and is better than failing outright.
2594 */
2595 error("failed to read object %s at offset %"PRIuMAX" from %s",
2596 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2597 mark_bad_packed_object(e.p, sha1);
ac939109 2598 data = read_object(sha1, type, size);
8eca0b47
NP
2599 }
2600 return data;
1f688557
JH
2601}
2602
21666f1a
NP
2603int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2604 unsigned char *sha1)
d66b37bb
JH
2605{
2606 struct cached_object *co;
2607
21666f1a 2608 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2609 if (has_sha1_file(sha1) || find_cached_object(sha1))
2610 return 0;
2611 if (cached_object_alloc <= cached_object_nr) {
2612 cached_object_alloc = alloc_nr(cached_object_alloc);
2613 cached_objects = xrealloc(cached_objects,
2614 sizeof(*cached_objects) *
2615 cached_object_alloc);
2616 }
2617 co = &cached_objects[cached_object_nr++];
2618 co->size = len;
21666f1a 2619 co->type = type;
efa13f7b
JH
2620 co->buf = xmalloc(len);
2621 memcpy(co->buf, buf, len);
d66b37bb
JH
2622 hashcpy(co->sha1, sha1);
2623 return 0;
2624}
2625
c2c5b270
CC
2626static void *read_object(const unsigned char *sha1, enum object_type *type,
2627 unsigned long *size)
0fcfd160
LT
2628{
2629 unsigned long mapsize;
2630 void *map, *buf;
d66b37bb
JH
2631 struct cached_object *co;
2632
2633 co = find_cached_object(sha1);
2634 if (co) {
21666f1a 2635 *type = co->type;
d66b37bb 2636 *size = co->size;
182af834 2637 return xmemdupz(co->buf, co->size);
d66b37bb 2638 }
0fcfd160 2639
8276c007
PE
2640 buf = read_packed_sha1(sha1, type, size);
2641 if (buf)
2642 return buf;
bb6b8e4f 2643 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2644 if (map) {
7efbff75 2645 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2646 munmap(map, mapsize);
2647 return buf;
2648 }
637cdd9d 2649 reprepare_packed_git();
8276c007 2650 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2651}
2652
b6c4cecc
JH
2653/*
2654 * This function dies on corrupt objects; the callers who want to
2655 * deal with them should arrange to call read_object() and give error
2656 * messages themselves.
2657 */
5bf29b95
JH
2658void *read_sha1_file_extended(const unsigned char *sha1,
2659 enum object_type *type,
2660 unsigned long *size,
2661 unsigned flag)
ac939109 2662{
3ba7a065 2663 void *data;
e8b15e61 2664 char *path;
b6c4cecc 2665 const struct packed_git *p;
5bf29b95
JH
2666 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2667 ? lookup_replace_object(sha1) : sha1;
b6c4cecc 2668
3ba7a065
JH
2669 errno = 0;
2670 data = read_object(repl, type, size);
4bbf5a26 2671 if (data)
b6c4cecc 2672 return data;
68095570 2673
25f3af3f 2674 if (errno && errno != ENOENT)
3ba7a065
JH
2675 die_errno("failed to read object %s", sha1_to_hex(sha1));
2676
68095570 2677 /* die if we replaced an object with one that does not exist */
b6c4cecc 2678 if (repl != sha1)
68095570
CC
2679 die("replacement %s not found for %s",
2680 sha1_to_hex(repl), sha1_to_hex(sha1));
2681
b6c4cecc
JH
2682 if (has_loose_object(repl)) {
2683 path = sha1_file_name(sha1);
2684 die("loose object %s (stored in %s) is corrupt",
2685 sha1_to_hex(repl), path);
e8b15e61 2686 }
68095570 2687
b6c4cecc
JH
2688 if ((p = has_packed_and_bad(repl)) != NULL)
2689 die("packed object %s (stored in %s) is corrupt",
2690 sha1_to_hex(repl), p->pack_name);
f5552aee 2691
b6c4cecc 2692 return NULL;
ac939109
NP
2693}
2694
40469ee9 2695void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2696 const char *required_type_name,
40469ee9
JH
2697 unsigned long *size,
2698 unsigned char *actual_sha1_return)
f4913f91 2699{
21666f1a 2700 enum object_type type, required_type;
f4913f91
JH
2701 void *buffer;
2702 unsigned long isize;
40469ee9 2703 unsigned char actual_sha1[20];
f4913f91 2704
21666f1a 2705 required_type = type_from_string(required_type_name);
e702496e 2706 hashcpy(actual_sha1, sha1);
40469ee9
JH
2707 while (1) {
2708 int ref_length = -1;
2709 const char *ref_type = NULL;
f4913f91 2710
21666f1a 2711 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2712 if (!buffer)
2713 return NULL;
21666f1a 2714 if (type == required_type) {
40469ee9
JH
2715 *size = isize;
2716 if (actual_sha1_return)
e702496e 2717 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2718 return buffer;
2719 }
2720 /* Handle references */
21666f1a 2721 else if (type == OBJ_COMMIT)
40469ee9 2722 ref_type = "tree ";
21666f1a 2723 else if (type == OBJ_TAG)
40469ee9
JH
2724 ref_type = "object ";
2725 else {
2726 free(buffer);
2727 return NULL;
2728 }
2729 ref_length = strlen(ref_type);
f4913f91 2730
50974ec9
MK
2731 if (ref_length + 40 > isize ||
2732 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2733 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2734 free(buffer);
2735 return NULL;
2736 }
1cf58e72 2737 free(buffer);
40469ee9
JH
2738 /* Now we have the ID of the referred-to object in
2739 * actual_sha1. Check again. */
f4913f91 2740 }
f4913f91
JH
2741}
2742
ce9fbf16 2743static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2744 const char *type, unsigned char *sha1,
d65a16f6 2745 char *hdr, int *hdrlen)
d410c0f5 2746{
9126f009 2747 git_SHA_CTX c;
d410c0f5
JH
2748
2749 /* Generate the header */
d65a16f6 2750 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
2751
2752 /* Sha1.. */
9126f009
NP
2753 git_SHA1_Init(&c);
2754 git_SHA1_Update(&c, hdr, *hdrlen);
2755 git_SHA1_Update(&c, buf, len);
2756 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2757}
2758
230f1322 2759/*
5a688fe4
JH
2760 * Move the just written object into its final resting place.
2761 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2762 * "moving" is only a part of what it does, when no patch between
2763 * master to pu changes the call sites of this function.
230f1322 2764 */
839837b9 2765int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322 2766{
e32c0a9c 2767 int ret = 0;
5a688fe4 2768
348df166 2769 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
2770 goto try_rename;
2771 else if (link(tmpfile, filename))
e32c0a9c 2772 ret = errno;
7ebb6fca
LT
2773
2774 /*
2775 * Coda hack - coda doesn't like cross-directory links,
2776 * so we fall back to a rename, which will mean that it
2777 * won't be able to check collisions, but that's not a
2778 * big deal.
2779 *
2780 * The same holds for FAT formatted media.
2781 *
3be1f18e 2782 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
2783 * left to unlink.
2784 */
2785 if (ret && ret != EEXIST) {
be66a6c4 2786 try_rename:
7ebb6fca 2787 if (!rename(tmpfile, filename))
3be1f18e 2788 goto out;
9e48b389 2789 ret = errno;
230f1322 2790 }
691f1a28 2791 unlink_or_warn(tmpfile);
230f1322
LT
2792 if (ret) {
2793 if (ret != EEXIST) {
82247e9b 2794 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
230f1322
LT
2795 }
2796 /* FIXME!!! Collision check here ? */
2797 }
2798
3be1f18e 2799out:
5256b006 2800 if (adjust_shared_perm(filename))
5a688fe4 2801 return error("unable to set permission to '%s'", filename);
230f1322
LT
2802 return 0;
2803}
2804
4d548150
LT
2805static int write_buffer(int fd, const void *buf, size_t len)
2806{
d34cf19b 2807 if (write_in_full(fd, buf, len) < 0)
93822c22 2808 return error("file write error (%s)", strerror(errno));
4d548150
LT
2809 return 0;
2810}
2811
ce9fbf16 2812int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2813 unsigned char *sha1)
2814{
d65a16f6 2815 char hdr[32];
abdc3fc8
RS
2816 int hdrlen;
2817 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2818 return 0;
2819}
2820
e9039dd3
LT
2821/* Finalize a file on disk, and close it. */
2822static void close_sha1_file(int fd)
2823{
aafe9fba
LT
2824 if (fsync_object_files)
2825 fsync_or_die(fd, "sha1 file");
e9039dd3 2826 if (close(fd) != 0)
d824cbba 2827 die_errno("error when closing sha1 file");
e9039dd3
LT
2828}
2829
5723fe7e
LT
2830/* Size of directory component, including the ending '/' */
2831static inline int directory_size(const char *filename)
2832{
2833 const char *s = strrchr(filename, '/');
2834 if (!s)
2835 return 0;
2836 return s - filename + 1;
2837}
2838
2839/*
2840 * This creates a temporary file in the same directory as the final
2841 * 'filename'
2842 *
2843 * We want to avoid cross-directory filename renames, because those
2844 * can have problems on various filesystems (FAT, NFS, Coda).
2845 */
2846static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2847{
2848 int fd, dirlen = directory_size(filename);
2849
2850 if (dirlen + 20 > bufsiz) {
2851 errno = ENAMETOOLONG;
2852 return -1;
2853 }
2854 memcpy(buffer, filename, dirlen);
2855 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
5256b006 2856 fd = git_mkstemp_mode(buffer, 0444);
cbacbf4e 2857 if (fd < 0 && dirlen && errno == ENOENT) {
5723fe7e 2858 /* Make sure the directory exists */
6ff6af62 2859 memcpy(buffer, filename, dirlen);
5723fe7e 2860 buffer[dirlen-1] = 0;
b2476a60
JH
2861 if (mkdir(buffer, 0777) && errno != EEXIST)
2862 return -1;
2863 if (adjust_shared_perm(buffer))
5723fe7e
LT
2864 return -1;
2865
2866 /* Try again */
2867 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
5256b006 2868 fd = git_mkstemp_mode(buffer, 0444);
5723fe7e
LT
2869 }
2870 return fd;
2871}
2872
bbac7311 2873static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 2874 const void *buf, unsigned long len, time_t mtime)
0fcfd160 2875{
915308b1 2876 int fd, ret;
9892beba 2877 unsigned char compressed[4096];
ef49a7a0 2878 git_zstream stream;
748af44c
NP
2879 git_SHA_CTX c;
2880 unsigned char parano_sha1[20];
706bc531 2881 char *filename;
ab1900a3 2882 static char tmp_file[PATH_MAX];
a44c9a5e 2883
972a9155 2884 filename = sha1_file_name(sha1);
ab1900a3 2885 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
aac17941 2886 if (fd < 0) {
35243577 2887 if (errno == EACCES)
82247e9b 2888 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
916d081b 2889 else
5eaeda70 2890 return error("unable to create temporary file: %s", strerror(errno));
aac17941
LT
2891 }
2892
0fcfd160
LT
2893 /* Set it up */
2894 memset(&stream, 0, sizeof(stream));
55bb5c91 2895 git_deflate_init(&stream, zlib_compression_level);
0fcfd160 2896 stream.next_out = compressed;
9892beba 2897 stream.avail_out = sizeof(compressed);
748af44c 2898 git_SHA1_Init(&c);
a44c9a5e
LT
2899
2900 /* First header.. */
d65a16f6 2901 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2902 stream.avail_in = hdrlen;
55bb5c91
JH
2903 while (git_deflate(&stream, 0) == Z_OK)
2904 ; /* nothing */
748af44c 2905 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
2906
2907 /* Then the data itself.. */
c00e657d 2908 stream.next_in = (void *)buf;
a44c9a5e 2909 stream.avail_in = len;
9892beba 2910 do {
748af44c 2911 unsigned char *in0 = stream.next_in;
55bb5c91 2912 ret = git_deflate(&stream, Z_FINISH);
748af44c 2913 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
2914 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2915 die("unable to write sha1 file");
2916 stream.next_out = compressed;
2917 stream.avail_out = sizeof(compressed);
2918 } while (ret == Z_OK);
2919
ac54c277
LT
2920 if (ret != Z_STREAM_END)
2921 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
55bb5c91 2922 ret = git_deflate_end_gently(&stream);
ac54c277
LT
2923 if (ret != Z_OK)
2924 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
2925 git_SHA1_Final(parano_sha1, &c);
2926 if (hashcmp(sha1, parano_sha1) != 0)
2927 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 2928
e9039dd3 2929 close_sha1_file(fd);
0fcfd160 2930
bbac7311
NP
2931 if (mtime) {
2932 struct utimbuf utb;
2933 utb.actime = mtime;
2934 utb.modtime = mtime;
ab1900a3 2935 if (utime(tmp_file, &utb) < 0)
bbac7311 2936 warning("failed utime() on %s: %s",
ab1900a3 2937 tmp_file, strerror(errno));
bbac7311
NP
2938 }
2939
ab1900a3 2940 return move_temp_to_file(tmp_file, filename);
0fcfd160 2941}
8237b185 2942
c00e657d 2943int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
bbac7311
NP
2944{
2945 unsigned char sha1[20];
2946 char hdr[32];
2947 int hdrlen;
2948
2949 /* Normally if we have it in the pack then we do not bother writing
2950 * it out into .git/objects/??/?{38} file.
2951 */
2952 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2953 if (returnsha1)
2954 hashcpy(returnsha1, sha1);
2955 if (has_sha1_file(sha1))
2956 return 0;
2957 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2958}
2959
2960int force_object_loose(const unsigned char *sha1, time_t mtime)
2961{
bbac7311
NP
2962 void *buf;
2963 unsigned long len;
2964 enum object_type type;
2965 char hdr[32];
2966 int hdrlen;
1fb23e65 2967 int ret;
bbac7311 2968
c529d75a 2969 if (has_loose_object(sha1))
bbac7311
NP
2970 return 0;
2971 buf = read_packed_sha1(sha1, &type, &len);
2972 if (!buf)
2973 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2974 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1fb23e65
BS
2975 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2976 free(buf);
2977
2978 return ret;
bbac7311
NP
2979}
2980
bf592c50
DB
2981int has_pack_index(const unsigned char *sha1)
2982{
2983 struct stat st;
2984 if (stat(sha1_pack_index_name(sha1), &st))
2985 return 0;
2986 return 1;
bf592c50
DB
2987}
2988
cd673c1f
JH
2989int has_sha1_pack(const unsigned char *sha1)
2990{
2991 struct pack_entry e;
2992 return find_pack_entry(sha1, &e);
2993}
2994
8237b185
DB
2995int has_sha1_file(const unsigned char *sha1)
2996{
1f688557
JH
2997 struct pack_entry e;
2998
cd673c1f 2999 if (find_pack_entry(sha1, &e))
1f688557 3000 return 1;
45e8a748
JK
3001 if (has_loose_object(sha1))
3002 return 1;
3003 reprepare_packed_git();
3004 return find_pack_entry(sha1, &e);
8237b185 3005}
74400e71 3006
c879daa2
NTND
3007static void check_tree(const void *buf, size_t size)
3008{
3009 struct tree_desc desc;
3010 struct name_entry entry;
3011
3012 init_tree_desc(&desc, buf, size);
3013 while (tree_entry(&desc, &entry))
3014 /* do nothing
3015 * tree_entry() will die() on malformed entries */
3016 ;
3017}
3018
3019static void check_commit(const void *buf, size_t size)
3020{
3021 struct commit c;
3022 memset(&c, 0, sizeof(c));
3023 if (parse_commit_buffer(&c, buf, size))
3024 die("corrupt commit");
3025}
3026
3027static void check_tag(const void *buf, size_t size)
3028{
3029 struct tag t;
3030 memset(&t, 0, sizeof(t));
3031 if (parse_tag_buffer(&t, buf, size))
3032 die("corrupt tag");
3033}
3034
43df4f86 3035static int index_mem(unsigned char *sha1, void *buf, size_t size,
c4ce46fc
JH
3036 enum object_type type,
3037 const char *path, unsigned flags)
e7332f96 3038{
6c510bee 3039 int ret, re_allocated = 0;
c4ce46fc 3040 int write_object = flags & HASH_WRITE_OBJECT;
74400e71 3041
7672db20 3042 if (!type)
edaec3fb 3043 type = OBJ_BLOB;
6c510bee
LT
3044
3045 /*
3046 * Convert blobs to git internal format
3047 */
43df4f86 3048 if ((type == OBJ_BLOB) && path) {
f285a2d7 3049 struct strbuf nbuf = STRBUF_INIT;
21e5ad50 3050 if (convert_to_git(path, buf, size, &nbuf,
5e12e78e 3051 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
b315c5c0 3052 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
3053 re_allocated = 1;
3054 }
3055 }
c4ce46fc 3056 if (flags & HASH_FORMAT_CHECK) {
c879daa2
NTND
3057 if (type == OBJ_TREE)
3058 check_tree(buf, size);
3059 if (type == OBJ_COMMIT)
3060 check_commit(buf, size);
3061 if (type == OBJ_TAG)
3062 check_tag(buf, size);
3063 }
6c510bee 3064
7672db20 3065 if (write_object)
edaec3fb 3066 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 3067 else
edaec3fb 3068 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 3069 if (re_allocated)
6c510bee 3070 free(buf);
43df4f86
DP
3071 return ret;
3072}
3073
7b41e1e1
JH
3074static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3075 const char *path, unsigned flags)
3076{
3077 struct strbuf sbuf = STRBUF_INIT;
3078 int ret;
3079
3080 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3081 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3082 else
3083 ret = -1;
3084 strbuf_release(&sbuf);
3085 return ret;
3086}
3087
ea68b0ce
DP
3088#define SMALL_FILE_SIZE (32*1024)
3089
7b41e1e1
JH
3090static int index_core(unsigned char *sha1, int fd, size_t size,
3091 enum object_type type, const char *path,
3092 unsigned flags)
43df4f86
DP
3093{
3094 int ret;
43df4f86 3095
7b41e1e1 3096 if (!size) {
c4ce46fc 3097 ret = index_mem(sha1, NULL, size, type, path, flags);
ea68b0ce
DP
3098 } else if (size <= SMALL_FILE_SIZE) {
3099 char *buf = xmalloc(size);
3100 if (size == read_in_full(fd, buf, size))
c4ce46fc 3101 ret = index_mem(sha1, buf, size, type, path, flags);
ea68b0ce
DP
3102 else
3103 ret = error("short read %s", strerror(errno));
3104 free(buf);
08bda208 3105 } else {
43df4f86 3106 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
c4ce46fc 3107 ret = index_mem(sha1, buf, size, type, path, flags);
aac17941 3108 munmap(buf, size);
08bda208 3109 }
7b41e1e1
JH
3110 return ret;
3111}
3112
4dd1fbc7 3113/*
568508e7
JH
3114 * This creates one packfile per large blob unless bulk-checkin
3115 * machinery is "plugged".
4dd1fbc7
JH
3116 *
3117 * This also bypasses the usual "convert-to-git" dance, and that is on
3118 * purpose. We could write a streaming version of the converting
3119 * functions and insert that before feeding the data to fast-import
4f22b101
JK
3120 * (or equivalent in-core API described above). However, that is
3121 * somewhat complicated, as we do not know the size of the filter
3122 * result, which we need to know beforehand when writing a git object.
3123 * Since the primary motivation for trying to stream from the working
3124 * tree file and to avoid mmaping it in core is to deal with large
3125 * binary blobs, they generally do not want to get any conversion, and
3126 * callers should avoid this code path when filters are requested.
4dd1fbc7
JH
3127 */
3128static int index_stream(unsigned char *sha1, int fd, size_t size,
3129 enum object_type type, const char *path,
3130 unsigned flags)
3131{
568508e7 3132 return index_bulk_checkin(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
3133}
3134
7b41e1e1
JH
3135int index_fd(unsigned char *sha1, int fd, struct stat *st,
3136 enum object_type type, const char *path, unsigned flags)
3137{
3138 int ret;
3139 size_t size = xsize_t(st->st_size);
3140
3141 if (!S_ISREG(st->st_mode))
3142 ret = index_pipe(sha1, fd, type, path, flags);
4f22b101
JK
3143 else if (size <= big_file_threshold || type != OBJ_BLOB ||
3144 (path && would_convert_to_git(path, NULL, 0, 0)))
7b41e1e1 3145 ret = index_core(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
3146 else
3147 ret = index_stream(sha1, fd, size, type, path, flags);
43df4f86 3148 close(fd);
aac17941 3149 return ret;
74400e71 3150}
ec1fcc16 3151
c4ce46fc 3152int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
ec1fcc16
JH
3153{
3154 int fd;
b760d3aa 3155 struct strbuf sb = STRBUF_INIT;
ec1fcc16
JH
3156
3157 switch (st->st_mode & S_IFMT) {
3158 case S_IFREG:
3159 fd = open(path, O_RDONLY);
3160 if (fd < 0)
3161 return error("open(\"%s\"): %s", path,
3162 strerror(errno));
c4ce46fc 3163 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
ec1fcc16
JH
3164 return error("%s: failed to insert into database",
3165 path);
3166 break;
3167 case S_IFLNK:
b760d3aa 3168 if (strbuf_readlink(&sb, path, st->st_size)) {
ec1fcc16 3169 char *errstr = strerror(errno);
ec1fcc16
JH
3170 return error("readlink(\"%s\"): %s", path,
3171 errstr);
3172 }
c4ce46fc 3173 if (!(flags & HASH_WRITE_OBJECT))
b760d3aa
LT
3174 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3175 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
ec1fcc16
JH
3176 return error("%s: failed to insert into database",
3177 path);
b760d3aa 3178 strbuf_release(&sb);
ec1fcc16 3179 break;
f35a6d3b
LT
3180 case S_IFDIR:
3181 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
3182 default:
3183 return error("%s: unsupported file type", path);
3184 }
3185 return 0;
3186}
a69e5429
JH
3187
3188int read_pack_header(int fd, struct pack_header *header)
3189{
c697ad14
HO
3190 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3191 /* "eof before pack header was fully read" */
3192 return PH_ERROR_EOF;
3193
a69e5429
JH
3194 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3195 /* "protocol error (pack signature mismatch detected)" */
3196 return PH_ERROR_PACK_SIGNATURE;
3197 if (!pack_version_ok(header->hdr_version))
3198 /* "protocol error (pack version unsupported)" */
3199 return PH_ERROR_PROTOCOL;
3200 return 0;
3201}
40d52ff7
JK
3202
3203void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3204{
3205 enum object_type type = sha1_object_info(sha1, NULL);
3206 if (type < 0)
3207 die("%s is not a valid object", sha1_to_hex(sha1));
3208 if (type != expect)
3209 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3210 typename(expect));
3211}