test-lib.sh: support -x option for shell-tracing
[git/git.git] / sha1_file.c
CommitLineData
0fcfd160
LT
1/*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
0fcfd160 9#include "cache.h"
6eac50d8 10#include "string-list.h"
1f688557 11#include "delta.h"
a733cb60 12#include "pack.h"
8e440259
PE
13#include "blob.h"
14#include "commit.h"
4dd1fbc7 15#include "run-command.h"
8e440259
PE
16#include "tag.h"
17#include "tree.h"
c879daa2 18#include "tree-walk.h"
f35a6d3b 19#include "refs.h"
70f5d5d3 20#include "pack-revindex.h"
628522ec 21#include "sha1-lookup.h"
568508e7 22#include "bulk-checkin.h"
090ea126 23#include "streaming.h"
543c5caa 24#include "dir.h"
0fcfd160 25
144bde78
LT
26#ifndef O_NOATIME
27#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
28#define O_NOATIME 01000000
29#else
30#define O_NOATIME 0
31#endif
32#endif
33
28bd70d8
JN
34#define SZ_FMT PRIuMAX
35static inline uintmax_t sz_fmt(size_t s) { return s; }
e05db0fd 36
96f1e58f 37const unsigned char null_sha1[20];
88cd621d 38
c597ba80
NTND
39/*
40 * This is meant to hold a *small* number of objects that you would
41 * want read_sha1_file() to be able to return, but yet you do not want
42 * to write them into the object store (e.g. a browse-only
43 * application).
44 */
45static struct cached_object {
46 unsigned char sha1[20];
47 enum object_type type;
48 void *buf;
49 unsigned long size;
50} *cached_objects;
51static int cached_object_nr, cached_object_alloc;
52
53static struct cached_object empty_tree = {
dab0d410 54 EMPTY_TREE_SHA1_BIN_LITERAL,
c597ba80
NTND
55 OBJ_TREE,
56 "",
57 0
58};
59
1b1005d1
MH
60/*
61 * A pointer to the last packed_git in which an object was found.
62 * When an object is sought, we look in this packfile first, because
63 * objects that are looked up at similar times are often in the same
64 * packfile as one another.
65 */
c01f51cc
NTND
66static struct packed_git *last_found_pack;
67
c597ba80
NTND
68static struct cached_object *find_cached_object(const unsigned char *sha1)
69{
70 int i;
71 struct cached_object *co = cached_objects;
72
73 for (i = 0; i < cached_object_nr; i++, co++) {
74 if (!hashcmp(co->sha1, sha1))
75 return co;
76 }
77 if (!hashcmp(sha1, empty_tree.sha1))
78 return &empty_tree;
79 return NULL;
80}
81
90a6464b
JH
82int mkdir_in_gitdir(const char *path)
83{
84 if (mkdir(path, 0777)) {
85 int saved_errno = errno;
86 struct stat st;
87 struct strbuf sb = STRBUF_INIT;
88
89 if (errno != EEXIST)
90 return -1;
91 /*
92 * Are we looking at a path in a symlinked worktree
93 * whose original repository does not yet have it?
94 * e.g. .git/rr-cache pointing at its original
95 * repository in which the user hasn't performed any
96 * conflict resolution yet?
97 */
98 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
99 strbuf_readlink(&sb, path, st.st_size) ||
100 !is_absolute_path(sb.buf) ||
101 mkdir(sb.buf, 0777)) {
102 strbuf_release(&sb);
103 errno = saved_errno;
104 return -1;
105 }
106 strbuf_release(&sb);
107 }
108 return adjust_shared_perm(path);
109}
110
0be0521b 111enum scld_error safe_create_leading_directories(char *path)
b2cb9425 112{
26c8ae2a 113 char *next_component = path + offset_1st_component(path);
0be0521b 114 enum scld_error ret = SCLD_OK;
67d42212 115
0be0521b 116 while (ret == SCLD_OK && next_component) {
f0502332 117 struct stat st;
0f527403 118 char *slash = next_component, slash_character;
f0502332 119
0f527403
MH
120 while (*slash && !is_dir_sep(*slash))
121 slash++;
122
123 if (!*slash)
b2cb9425 124 break;
bf10cf70 125
26c8ae2a 126 next_component = slash + 1;
0f527403 127 while (is_dir_sep(*next_component))
bf10cf70 128 next_component++;
26c8ae2a 129 if (!*next_component)
5f0bdf50 130 break;
831651fd 131
0f527403 132 slash_character = *slash;
831651fd 133 *slash = '\0';
67d42212
JR
134 if (!stat(path, &st)) {
135 /* path exists */
9e6f885d 136 if (!S_ISDIR(st.st_mode))
0be0521b 137 ret = SCLD_EXISTS;
53a39721 138 } else if (mkdir(path, 0777)) {
928734d9 139 if (errno == EEXIST &&
9e6f885d 140 !stat(path, &st) && S_ISDIR(st.st_mode))
928734d9 141 ; /* somebody created it since we checked */
18d37e86
MH
142 else if (errno == ENOENT)
143 /*
144 * Either mkdir() failed because
145 * somebody just pruned the containing
146 * directory, or stat() failed because
147 * the file that was in our way was
148 * just removed. Either way, inform
149 * the caller that it might be worth
150 * trying again:
151 */
152 ret = SCLD_VANISHED;
9e6f885d 153 else
0be0521b 154 ret = SCLD_FAILED;
53a39721 155 } else if (adjust_shared_perm(path)) {
0be0521b 156 ret = SCLD_PERMS;
457f06d6 157 }
0f527403 158 *slash = slash_character;
b2cb9425 159 }
9e6f885d 160 return ret;
b2cb9425 161}
723c31fe 162
0be0521b 163enum scld_error safe_create_leading_directories_const(const char *path)
8e21d63b
JK
164{
165 /* path points to cache entries, so xstrdup before messing with it */
166 char *buf = xstrdup(path);
0be0521b 167 enum scld_error result = safe_create_leading_directories(buf);
8e21d63b
JK
168 free(buf);
169 return result;
170}
171
ace1534d
JH
172static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
173{
174 int i;
175 for (i = 0; i < 20; i++) {
176 static char hex[] = "0123456789abcdef";
177 unsigned int val = sha1[i];
178 char *pos = pathbuf + i*2 + (i > 0);
179 *pos++ = hex[val >> 4];
180 *pos = hex[val & 0xf];
181 }
182}
183
30d6c6ea 184const char *sha1_file_name(const unsigned char *sha1)
0fcfd160 185{
560fb6a1
JK
186 static char buf[PATH_MAX];
187 const char *objdir;
188 int len;
0fcfd160 189
560fb6a1
JK
190 objdir = get_object_directory();
191 len = strlen(objdir);
192
193 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
194 if (len + 43 > PATH_MAX)
195 die("insanely long object directory %s", objdir);
196 memcpy(buf, objdir, len);
197 buf[len] = '/';
198 buf[len+3] = '/';
199 buf[len+42] = '\0';
200 fill_sha1_path(buf + len + 1, sha1);
201 return buf;
0fcfd160
LT
202}
203
d40d535b
MH
204/*
205 * Return the name of the pack or index file with the specified sha1
206 * in its filename. *base and *name are scratch space that must be
207 * provided by the caller. which should be "pack" or "idx".
208 */
633f43e1 209static char *sha1_get_pack_name(const unsigned char *sha1,
6eec46bd 210 char **name, char **base, const char *which)
bf592c50
DB
211{
212 static const char hex[] = "0123456789abcdef";
633f43e1 213 char *buf;
bf592c50
DB
214 int i;
215
633f43e1 216 if (!*base) {
bf592c50
DB
217 const char *sha1_file_directory = get_object_directory();
218 int len = strlen(sha1_file_directory);
633f43e1 219 *base = xmalloc(len + 60);
6eec46bd
JH
220 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
221 sha1_file_directory, which);
633f43e1 222 *name = *base + len + 11;
bf592c50
DB
223 }
224
633f43e1 225 buf = *name;
bf592c50
DB
226
227 for (i = 0; i < 20; i++) {
228 unsigned int val = *sha1++;
229 *buf++ = hex[val >> 4];
230 *buf++ = hex[val & 0xf];
231 }
a6080a0a 232
633f43e1 233 return *base;
bf592c50
DB
234}
235
633f43e1 236char *sha1_pack_name(const unsigned char *sha1)
bf592c50 237{
633f43e1 238 static char *name, *base;
bf592c50 239
6eec46bd 240 return sha1_get_pack_name(sha1, &name, &base, "pack");
633f43e1 241}
bf592c50 242
633f43e1
HO
243char *sha1_pack_index_name(const unsigned char *sha1)
244{
245 static char *name, *base;
a6080a0a 246
6eec46bd 247 return sha1_get_pack_name(sha1, &name, &base, "idx");
bf592c50
DB
248}
249
d5a63b99
JH
250struct alternate_object_database *alt_odb_list;
251static struct alternate_object_database **alt_odb_tail;
ace1534d 252
ddd5d056
JH
253/*
254 * Prepare alternate object database registry.
d5a63b99
JH
255 *
256 * The variable alt_odb_list points at the list of struct
257 * alternate_object_database. The elements on this list come from
258 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
259 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
260 * whose contents is similar to that environment variable but can be
261 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
262 * contains "/the/directory/corresponding/to/.git/objects/...", while
263 * its name points just after the slash at the end of ".git/objects/"
264 * in the example above, and has enough space to hold 40-byte hex
265 * SHA1, an extra slash for the first level indirection, and the
266 * terminating NUL.
ddd5d056 267 */
539e7506
EK
268static int link_alt_odb_entry(const char *entry, const char *relative_base,
269 int depth, const char *normalized_objdir)
ace1534d 270{
c2f493a4
MW
271 struct alternate_object_database *ent;
272 struct alternate_object_database *alt;
5bdf0a84
HW
273 int pfxlen, entlen;
274 struct strbuf pathbuf = STRBUF_INIT;
d5a63b99 275
85dadc38 276 if (!is_absolute_path(entry) && relative_base) {
5bdf0a84
HW
277 strbuf_addstr(&pathbuf, real_path(relative_base));
278 strbuf_addch(&pathbuf, '/');
c2f493a4 279 }
6eac50d8 280 strbuf_addstr(&pathbuf, entry);
c2f493a4 281
5bdf0a84
HW
282 normalize_path_copy(pathbuf.buf, pathbuf.buf);
283
284 pfxlen = strlen(pathbuf.buf);
285
286 /*
287 * The trailing slash after the directory name is given by
288 * this function at the end. Remove duplicates.
289 */
290 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
291 pfxlen -= 1;
292
293 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
294 ent = xmalloc(sizeof(*ent) + entlen);
295 memcpy(ent->base, pathbuf.buf, pfxlen);
296 strbuf_release(&pathbuf);
c2f493a4
MW
297
298 ent->name = ent->base + pfxlen + 1;
299 ent->base[pfxlen + 3] = '/';
300 ent->base[pfxlen] = ent->base[entlen-1] = 0;
301
302 /* Detect cases where alternate disappeared */
90b4a71c 303 if (!is_directory(ent->base)) {
c2f493a4
MW
304 error("object directory %s does not exist; "
305 "check .git/objects/info/alternates.",
306 ent->base);
307 free(ent);
308 return -1;
309 }
310
311 /* Prevent the common mistake of listing the same
312 * thing twice, or object directory itself.
313 */
314 for (alt = alt_odb_list; alt; alt = alt->next) {
80b47854
RS
315 if (pfxlen == alt->name - alt->base - 1 &&
316 !memcmp(ent->base, alt->base, pfxlen)) {
c2f493a4
MW
317 free(ent);
318 return -1;
319 }
320 }
539e7506 321 if (!strcmp_icase(ent->base, normalized_objdir)) {
c2f493a4
MW
322 free(ent);
323 return -1;
324 }
325
326 /* add the alternate entry */
327 *alt_odb_tail = ent;
328 alt_odb_tail = &(ent->next);
329 ent->next = NULL;
330
331 /* recursively add alternates */
332 read_info_alternates(ent->base, depth + 1);
333
334 ent->base[pfxlen] = '/';
335
336 return 0;
337}
338
c5950164 339static void link_alt_odb_entries(const char *alt, int len, int sep,
c2f493a4
MW
340 const char *relative_base, int depth)
341{
6eac50d8
MH
342 struct string_list entries = STRING_LIST_INIT_NODUP;
343 char *alt_copy;
344 int i;
539e7506 345 struct strbuf objdirbuf = STRBUF_INIT;
c2f493a4
MW
346
347 if (depth > 5) {
348 error("%s: ignoring alternate object stores, nesting too deep.",
349 relative_base);
350 return;
351 }
352
f655651e 353 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
539e7506
EK
354 normalize_path_copy(objdirbuf.buf, objdirbuf.buf);
355
c5950164 356 alt_copy = xmemdupz(alt, len);
6eac50d8
MH
357 string_list_split_in_place(&entries, alt_copy, sep, -1);
358 for (i = 0; i < entries.nr; i++) {
359 const char *entry = entries.items[i].string;
360 if (entry[0] == '\0' || entry[0] == '#')
9577e7e3 361 continue;
6eac50d8
MH
362 if (!is_absolute_path(entry) && depth) {
363 error("%s: ignoring relative alternate object store %s",
364 relative_base, entry);
365 } else {
539e7506 366 link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
9577e7e3 367 }
9577e7e3 368 }
6eac50d8
MH
369 string_list_clear(&entries, 0);
370 free(alt_copy);
539e7506 371 strbuf_release(&objdirbuf);
d5a63b99
JH
372}
373
5e73633d 374void read_info_alternates(const char * relative_base, int depth)
d5a63b99 375{
9577e7e3 376 char *map;
dc49cd76 377 size_t mapsz;
d5a63b99 378 struct stat st;
9cb18f56
JM
379 const char alt_file_name[] = "info/alternates";
380 /* Given that relative_base is no longer than PATH_MAX,
381 ensure that "path" has enough space to append "/", the
382 file name, "info/alternates", and a trailing NUL. */
383 char path[PATH_MAX + 1 + sizeof alt_file_name];
c2f493a4 384 int fd;
d5a63b99 385
9cb18f56 386 sprintf(path, "%s/%s", relative_base, alt_file_name);
f4e51683 387 fd = git_open_noatime(path);
d5a63b99
JH
388 if (fd < 0)
389 return;
390 if (fstat(fd, &st) || (st.st_size == 0)) {
391 close(fd);
9a217f2a 392 return;
ace1534d 393 }
dc49cd76
SP
394 mapsz = xsize_t(st.st_size);
395 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 396 close(fd);
d5a63b99 397
c5950164 398 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
c2f493a4 399
dc49cd76 400 munmap(map, mapsz);
ace1534d
JH
401}
402
bef70b22
DB
403void add_to_alternates_file(const char *reference)
404{
405 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
acd3b9ec 406 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
e6baf4a1 407 char *alt = mkpath("%s\n", reference);
bef70b22
DB
408 write_or_die(fd, alt, strlen(alt));
409 if (commit_lock_file(lock))
410 die("could not close alternates file");
411 if (alt_odb_tail)
c5950164 412 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
bef70b22
DB
413}
414
d79796bc
JH
415void foreach_alt_odb(alt_odb_fn fn, void *cb)
416{
417 struct alternate_object_database *ent;
418
419 prepare_alt_odb();
420 for (ent = alt_odb_list; ent; ent = ent->next)
421 if (fn(ent, cb))
422 return;
423}
424
c2f493a4
MW
425void prepare_alt_odb(void)
426{
554fe20d 427 const char *alt;
c2f493a4 428
7dc24aa5
SP
429 if (alt_odb_tail)
430 return;
431
c2f493a4
MW
432 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
433 if (!alt) alt = "";
434
c2f493a4 435 alt_odb_tail = &alt_odb_list;
c5950164 436 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
437
438 read_info_alternates(get_object_directory(), 0);
439}
440
0f4dc14a 441static int has_loose_object_local(const unsigned char *sha1)
ace1534d 442{
30d6c6ea 443 return !access(sha1_file_name(sha1), F_OK);
0f4dc14a 444}
ace1534d 445
0f4dc14a
BC
446int has_loose_object_nonlocal(const unsigned char *sha1)
447{
448 struct alternate_object_database *alt;
9a217f2a 449 prepare_alt_odb();
d5a63b99 450 for (alt = alt_odb_list; alt; alt = alt->next) {
0f4dc14a 451 fill_sha1_path(alt->name, sha1);
c529d75a
LT
452 if (!access(alt->base, F_OK))
453 return 1;
ace1534d 454 }
c529d75a 455 return 0;
ace1534d
JH
456}
457
0f4dc14a
BC
458static int has_loose_object(const unsigned char *sha1)
459{
460 return has_loose_object_local(sha1) ||
461 has_loose_object_nonlocal(sha1);
462}
463
60bb8b14 464static unsigned int pack_used_ctr;
a53128b6
SP
465static unsigned int pack_mmap_calls;
466static unsigned int peak_pack_open_windows;
467static unsigned int pack_open_windows;
c7934306
SP
468static unsigned int pack_open_fds;
469static unsigned int pack_max_fds;
a53128b6 470static size_t peak_pack_mapped;
60bb8b14 471static size_t pack_mapped;
9a217f2a 472struct packed_git *packed_git;
1f688557 473
b79d18c9 474void pack_report(void)
a53128b6
SP
475{
476 fprintf(stderr,
e05db0fd
PR
477 "pack_report: getpagesize() = %10" SZ_FMT "\n"
478 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
479 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
480 sz_fmt(getpagesize()),
481 sz_fmt(packed_git_window_size),
482 sz_fmt(packed_git_limit));
a53128b6
SP
483 fprintf(stderr,
484 "pack_report: pack_used_ctr = %10u\n"
485 "pack_report: pack_mmap_calls = %10u\n"
486 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
487 "pack_report: pack_mapped = "
488 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
489 pack_used_ctr,
490 pack_mmap_calls,
491 pack_open_windows, peak_pack_open_windows,
9e42d6a1 492 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
493}
494
d40d535b
MH
495/*
496 * Open and mmap the index file at path, perform a couple of
497 * consistency checks, then record its information to p. Return 0 on
498 * success.
499 */
500static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
501{
502 void *idx_map;
42873078 503 struct pack_idx_header *hdr;
dc49cd76 504 size_t idx_size;
74e34e1f 505 uint32_t version, nr, i, *index;
f4e51683 506 int fd = git_open_noatime(path);
1f688557 507 struct stat st;
42873078 508
1f688557
JH
509 if (fd < 0)
510 return -1;
511 if (fstat(fd, &st)) {
512 close(fd);
513 return -1;
514 }
dc49cd76 515 idx_size = xsize_t(st.st_size);
2d88451b
SP
516 if (idx_size < 4 * 256 + 20 + 20) {
517 close(fd);
518 return error("index file %s is too small", path);
519 }
c4712e45 520 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 521 close(fd);
1f688557 522
42873078
NP
523 hdr = idx_map;
524 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
525 version = ntohl(hdr->idx_version);
526 if (version < 2 || version > 2) {
527 munmap(idx_map, idx_size);
6e1c2344 528 return error("index file %s is version %"PRIu32
74e34e1f
NP
529 " and is not supported by this binary"
530 " (try upgrading GIT to a newer version)",
531 path, version);
532 }
533 } else
534 version = 1;
df1b059d 535
1f688557 536 nr = 0;
42873078 537 index = idx_map;
74e34e1f
NP
538 if (version > 1)
539 index += 2; /* skip index header */
1f688557 540 for (i = 0; i < 256; i++) {
326bf396 541 uint32_t n = ntohl(index[i]);
2d88451b
SP
542 if (n < nr) {
543 munmap(idx_map, idx_size);
df1b059d 544 return error("non-monotonic index %s", path);
2d88451b 545 }
1f688557
JH
546 nr = n;
547 }
548
74e34e1f
NP
549 if (version == 1) {
550 /*
551 * Total size:
552 * - 256 index entries 4 bytes each
553 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
554 * - 20-byte SHA1 of the packfile
555 * - 20-byte SHA1 file checksum
556 */
557 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
558 munmap(idx_map, idx_size);
eef427a0 559 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
560 }
561 } else if (version == 2) {
562 /*
563 * Minimum size:
564 * - 8 bytes of header
565 * - 256 index entries 4 bytes each
566 * - 20-byte sha1 entry * nr
567 * - 4-byte crc entry * nr
568 * - 4-byte offset entry * nr
569 * - 20-byte SHA1 of the packfile
570 * - 20-byte SHA1 file checksum
571 * And after the 4-byte offset table might be a
572 * variable sized table containing 8-byte entries
573 * for offsets larger than 2^31.
574 */
575 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
576 unsigned long max_size = min_size;
577 if (nr)
578 max_size += (nr - 1)*8;
579 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 580 munmap(idx_map, idx_size);
eef427a0 581 return error("wrong index v2 file size in %s", path);
74e34e1f 582 }
7109c889
JH
583 if (idx_size != min_size &&
584 /*
585 * make sure we can deal with large pack offsets.
586 * 31-bit signed offset won't be enough, neither
587 * 32-bit unsigned one will be.
588 */
589 (sizeof(off_t) <= 4)) {
590 munmap(idx_map, idx_size);
591 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 592 }
2d88451b 593 }
1f688557 594
74e34e1f 595 p->index_version = version;
42873078
NP
596 p->index_data = idx_map;
597 p->index_size = idx_size;
57059091 598 p->num_objects = nr;
1f688557
JH
599 return 0;
600}
601
bc8e478a 602int open_pack_index(struct packed_git *p)
d079837e
SP
603{
604 char *idx_name;
605 int ret;
606
607 if (p->index_data)
608 return 0;
609
610 idx_name = xstrdup(p->pack_name);
611 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
612 ret = check_packed_git_idx(idx_name, p);
613 free(idx_name);
614 return ret;
615}
616
11daf39b
SP
617static void scan_windows(struct packed_git *p,
618 struct packed_git **lru_p,
619 struct pack_window **lru_w,
620 struct pack_window **lru_l)
1f688557 621{
11daf39b
SP
622 struct pack_window *w, *w_l;
623
624 for (w_l = NULL, w = p->windows; w; w = w->next) {
625 if (!w->inuse_cnt) {
626 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
627 *lru_p = p;
628 *lru_w = w;
629 *lru_l = w_l;
54044bf8 630 }
54044bf8 631 }
11daf39b 632 w_l = w;
f9253394 633 }
11daf39b
SP
634}
635
7c3ecb32 636static int unuse_one_window(struct packed_git *current)
11daf39b
SP
637{
638 struct packed_git *p, *lru_p = NULL;
639 struct pack_window *lru_w = NULL, *lru_l = NULL;
640
641 if (current)
642 scan_windows(current, &lru_p, &lru_w, &lru_l);
643 for (p = packed_git; p; p = p->next)
644 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
645 if (lru_p) {
646 munmap(lru_w->base, lru_w->len);
647 pack_mapped -= lru_w->len;
648 if (lru_l)
649 lru_l->next = lru_w->next;
7c3ecb32 650 else
54044bf8 651 lru_p->windows = lru_w->next;
54044bf8 652 free(lru_w);
a53128b6 653 pack_open_windows--;
54044bf8
SP
654 return 1;
655 }
656 return 0;
f9253394
JH
657}
658
7c3ecb32 659void release_pack_memory(size_t need)
97bfeb34
SP
660{
661 size_t cur = pack_mapped;
7c3ecb32 662 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
97bfeb34
SP
663 ; /* nothing */
664}
665
02710228
SP
666static void mmap_limit_check(size_t length)
667{
668 static size_t limit = 0;
669 if (!limit) {
670 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
671 if (!limit)
672 limit = SIZE_MAX;
673 }
674 if (length > limit)
675 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
676 (uintmax_t)length, (uintmax_t)limit);
677}
678
58ecbd5e
JN
679void *xmmap(void *start, size_t length,
680 int prot, int flags, int fd, off_t offset)
681{
02710228
SP
682 void *ret;
683
684 mmap_limit_check(length);
685 ret = mmap(start, length, prot, flags, fd, offset);
58ecbd5e
JN
686 if (ret == MAP_FAILED) {
687 if (!length)
688 return NULL;
7c3ecb32 689 release_pack_memory(length);
58ecbd5e
JN
690 ret = mmap(start, length, prot, flags, fd, offset);
691 if (ret == MAP_FAILED)
692 die_errno("Out of memory? mmap failed");
693 }
694 return ret;
695}
696
c9ced051
SP
697void close_pack_windows(struct packed_git *p)
698{
699 while (p->windows) {
700 struct pack_window *w = p->windows;
701
702 if (w->inuse_cnt)
703 die("pack '%s' still has open windows to it",
704 p->pack_name);
705 munmap(w->base, w->len);
706 pack_mapped -= w->len;
707 pack_open_windows--;
708 p->windows = w->next;
709 free(w);
710 }
711}
712
88d0db55
BC
713/*
714 * The LRU pack is the one with the oldest MRU window, preferring packs
715 * with no used windows, or the oldest mtime if it has no windows allocated.
716 */
717static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
718{
719 struct pack_window *w, *this_mru_w;
720 int has_windows_inuse = 0;
721
722 /*
723 * Reject this pack if it has windows and the previously selected
724 * one does not. If this pack does not have windows, reject
725 * it if the pack file is newer than the previously selected one.
726 */
727 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
728 return;
729
730 for (w = this_mru_w = p->windows; w; w = w->next) {
731 /*
732 * Reject this pack if any of its windows are in use,
733 * but the previously selected pack did not have any
734 * inuse windows. Otherwise, record that this pack
735 * has windows in use.
736 */
737 if (w->inuse_cnt) {
738 if (*accept_windows_inuse)
739 has_windows_inuse = 1;
740 else
741 return;
742 }
743
744 if (w->last_used > this_mru_w->last_used)
745 this_mru_w = w;
746
747 /*
748 * Reject this pack if it has windows that have been
749 * used more recently than the previously selected pack.
750 * If the previously selected pack had windows inuse and
751 * we have not encountered a window in this pack that is
752 * inuse, skip this check since we prefer a pack with no
753 * inuse windows to one that has inuse windows.
754 */
755 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
756 this_mru_w->last_used > (*mru_w)->last_used)
757 return;
758 }
759
760 /*
761 * Select this pack.
762 */
763 *mru_w = this_mru_w;
764 *lru_p = p;
765 *accept_windows_inuse = has_windows_inuse;
766}
767
768static int close_one_pack(void)
769{
770 struct packed_git *p, *lru_p = NULL;
771 struct pack_window *mru_w = NULL;
772 int accept_windows_inuse = 1;
773
774 for (p = packed_git; p; p = p->next) {
775 if (p->pack_fd == -1)
776 continue;
777 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
778 }
779
780 if (lru_p) {
781 close(lru_p->pack_fd);
782 pack_open_fds--;
783 lru_p->pack_fd = -1;
784 return 1;
785 }
786
787 return 0;
788}
789
03e79c88 790void unuse_pack(struct pack_window **w_cursor)
f9253394 791{
03e79c88
SP
792 struct pack_window *w = *w_cursor;
793 if (w) {
794 w->inuse_cnt--;
795 *w_cursor = NULL;
796 }
1f688557
JH
797}
798
fa5fc15d
SP
799void close_pack_index(struct packed_git *p)
800{
801 if (p->index_data) {
802 munmap((void *)p->index_data, p->index_size);
803 p->index_data = NULL;
804 }
805}
806
c74faea1
NP
807/*
808 * This is used by git-repack in case a newly created pack happens to
809 * contain the same set of objects as an existing one. In that case
810 * the resulting file might be different even if its name would be the
811 * same. It is best to close any reference to the old pack before it is
01689909 812 * replaced on disk. Of course no index pointers or windows for given pack
c74faea1
NP
813 * must subsist at this point. If ever objects from this pack are requested
814 * again, the new version of the pack will be reinitialized through
815 * reprepare_packed_git().
816 */
817void free_pack_by_name(const char *pack_name)
818{
819 struct packed_git *p, **pp = &packed_git;
820
821 while (*pp) {
822 p = *pp;
823 if (strcmp(pack_name, p->pack_name) == 0) {
fa3a0c94 824 clear_delta_base_cache();
c74faea1 825 close_pack_windows(p);
c7934306 826 if (p->pack_fd != -1) {
c74faea1 827 close(p->pack_fd);
c7934306
SP
828 pack_open_fds--;
829 }
fa5fc15d 830 close_pack_index(p);
c74faea1
NP
831 free(p->bad_object_sha1);
832 *pp = p->next;
c01f51cc
NTND
833 if (last_found_pack == p)
834 last_found_pack = NULL;
c74faea1
NP
835 free(p);
836 return;
837 }
838 pp = &p->next;
839 }
840}
841
a0788266
JS
842static unsigned int get_max_fd_limit(void)
843{
844#ifdef RLIMIT_NOFILE
491a8dec
JH
845 {
846 struct rlimit lim;
a0788266 847
491a8dec
JH
848 if (!getrlimit(RLIMIT_NOFILE, &lim))
849 return lim.rlim_cur;
850 }
851#endif
852
853#ifdef _SC_OPEN_MAX
854 {
855 long open_max = sysconf(_SC_OPEN_MAX);
856 if (0 < open_max)
857 return open_max;
858 /*
859 * Otherwise, we got -1 for one of the two
860 * reasons:
861 *
862 * (1) sysconf() did not understand _SC_OPEN_MAX
863 * and signaled an error with -1; or
864 * (2) sysconf() said there is no limit.
865 *
866 * We _could_ clear errno before calling sysconf() to
867 * tell these two cases apart and return a huge number
868 * in the latter case to let the caller cap it to a
869 * value that is not so selfish, but letting the
870 * fallback OPEN_MAX codepath take care of these cases
871 * is a lot simpler.
872 */
873 }
874#endif
a0788266 875
491a8dec 876#ifdef OPEN_MAX
a0788266
JS
877 return OPEN_MAX;
878#else
879 return 1; /* see the caller ;-) */
880#endif
881}
882
3cf8b462
SP
883/*
884 * Do not call this directly as this leaks p->pack_fd on error return;
885 * call open_packed_git() instead.
886 */
887static int open_packed_git_1(struct packed_git *p)
1f688557 888{
9bc879c1
SP
889 struct stat st;
890 struct pack_header hdr;
891 unsigned char sha1[20];
892 unsigned char *idx_sha1;
2c039da8 893 long fd_flag;
9bc879c1 894
d079837e
SP
895 if (!p->index_data && open_pack_index(p))
896 return error("packfile %s index unavailable", p->pack_name);
897
c7934306 898 if (!pack_max_fds) {
a0788266 899 unsigned int max_fds = get_max_fd_limit();
c7934306
SP
900
901 /* Save 3 for stdin/stdout/stderr, 22 for work */
902 if (25 < max_fds)
903 pack_max_fds = max_fds - 25;
904 else
905 pack_max_fds = 1;
906 }
907
88d0db55 908 while (pack_max_fds <= pack_open_fds && close_one_pack())
c7934306
SP
909 ; /* nothing */
910
f4e51683 911 p->pack_fd = git_open_noatime(p->pack_name);
9bc879c1 912 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 913 return -1;
c7934306 914 pack_open_fds++;
9bc879c1
SP
915
916 /* If we created the struct before we had the pack we lack size. */
bf592c50 917 if (!p->pack_size) {
bf592c50 918 if (!S_ISREG(st.st_mode))
072db278 919 return error("packfile %s not a regular file", p->pack_name);
bf592c50 920 p->pack_size = st.st_size;
9bc879c1 921 } else if (p->pack_size != st.st_size)
072db278 922 return error("packfile %s size changed", p->pack_name);
9bc879c1 923
2c039da8
JH
924 /* We leave these file descriptors open with sliding mmap;
925 * there is no point keeping them open across exec(), though.
926 */
927 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
928 if (fd_flag < 0)
072db278 929 return error("cannot determine file descriptor flags");
2c039da8
JH
930 fd_flag |= FD_CLOEXEC;
931 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 932 return error("cannot set FD_CLOEXEC");
2c039da8 933
9bc879c1 934 /* Verify we recognize this pack file format. */
e6e2bd62 935 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 936 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 937 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 938 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 939 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
940 return error("packfile %s is version %"PRIu32" and not"
941 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
942 p->pack_name, ntohl(hdr.hdr_version));
943
944 /* Verify the pack matches its index. */
57059091 945 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
946 return error("packfile %s claims to have %"PRIu32" objects"
947 " while index indicates %"PRIu32" objects",
57059091
NP
948 p->pack_name, ntohl(hdr.hdr_entries),
949 p->num_objects);
9bc879c1 950 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 951 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 952 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 953 return error("packfile %s signature is unavailable", p->pack_name);
42873078 954 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 955 if (hashcmp(sha1, idx_sha1))
072db278
SP
956 return error("packfile %s does not match index", p->pack_name);
957 return 0;
9bc879c1
SP
958}
959
3cf8b462
SP
960static int open_packed_git(struct packed_git *p)
961{
962 if (!open_packed_git_1(p))
963 return 0;
964 if (p->pack_fd != -1) {
965 close(p->pack_fd);
c7934306 966 pack_open_fds--;
3cf8b462
SP
967 p->pack_fd = -1;
968 }
969 return -1;
970}
971
c4001d92 972static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
973{
974 /* We must promise at least 20 bytes (one hash) after the
975 * offset is available from this window, otherwise the offset
976 * is not actually in this window and a different window (which
977 * has that one hash excess) must be used. This is to support
978 * the object header and delta base parsing routines below.
979 */
980 off_t win_off = win->offset;
981 return win_off <= offset
982 && (offset + 20) <= (win_off + win->len);
983}
984
4b25d091 985unsigned char *use_pack(struct packed_git *p,
03e79c88 986 struct pack_window **w_cursor,
c4001d92 987 off_t offset,
ef49a7a0 988 unsigned long *left)
9bc879c1 989{
60bb8b14 990 struct pack_window *win = *w_cursor;
03e79c88 991
a9d98a14 992 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
993 * pointless to ask for an offset into the middle of that
994 * hash, and the in_window function above wouldn't match
995 * don't allow an offset too close to the end of the file.
996 */
d131b7af
SP
997 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
998 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
999 if (offset > (p->pack_size - 20))
1000 die("offset beyond end of packfile (truncated pack?)");
1001
1002 if (!win || !in_window(win, offset)) {
1003 if (win)
1004 win->inuse_cnt--;
1005 for (win = p->windows; win; win = win->next) {
1006 if (in_window(win, offset))
1007 break;
1008 }
1009 if (!win) {
78a28df9 1010 size_t window_align = packed_git_window_size / 2;
dc49cd76 1011 off_t len;
d131b7af
SP
1012
1013 if (p->pack_fd == -1 && open_packed_git(p))
1014 die("packfile %s cannot be accessed", p->pack_name);
1015
60bb8b14 1016 win = xcalloc(1, sizeof(*win));
78a28df9 1017 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
1018 len = p->pack_size - win->offset;
1019 if (len > packed_git_window_size)
1020 len = packed_git_window_size;
1021 win->len = (size_t)len;
60bb8b14 1022 pack_mapped += win->len;
11daf39b 1023 while (packed_git_limit < pack_mapped
7c3ecb32 1024 && unuse_one_window(p))
60bb8b14 1025 ; /* nothing */
c4712e45 1026 win->base = xmmap(NULL, win->len,
60bb8b14
SP
1027 PROT_READ, MAP_PRIVATE,
1028 p->pack_fd, win->offset);
1029 if (win->base == MAP_FAILED)
73b4e4be
SP
1030 die("packfile %s cannot be mapped: %s",
1031 p->pack_name,
1032 strerror(errno));
d131b7af
SP
1033 if (!win->offset && win->len == p->pack_size
1034 && !p->do_not_close) {
1035 close(p->pack_fd);
1036 pack_open_fds--;
1037 p->pack_fd = -1;
1038 }
a53128b6
SP
1039 pack_mmap_calls++;
1040 pack_open_windows++;
1041 if (pack_mapped > peak_pack_mapped)
1042 peak_pack_mapped = pack_mapped;
1043 if (pack_open_windows > peak_pack_open_windows)
1044 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
1045 win->next = p->windows;
1046 p->windows = win;
1047 }
1f688557 1048 }
03e79c88
SP
1049 if (win != *w_cursor) {
1050 win->last_used = pack_used_ctr++;
1051 win->inuse_cnt++;
1052 *w_cursor = win;
1053 }
60bb8b14 1054 offset -= win->offset;
03e79c88 1055 if (left)
dc49cd76 1056 *left = win->len - xsize_t(offset);
03e79c88 1057 return win->base + offset;
1f688557
JH
1058}
1059
27d69a46
NP
1060static struct packed_git *alloc_packed_git(int extra)
1061{
1062 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1063 memset(p, 0, sizeof(*p));
1064 p->pack_fd = -1;
1065 return p;
1066}
1067
e0500293
JN
1068static void try_to_free_pack_memory(size_t size)
1069{
7c3ecb32 1070 release_pack_memory(size);
e0500293
JN
1071}
1072
42873078 1073struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557 1074{
e0500293 1075 static int have_set_try_to_free_routine;
1f688557 1076 struct stat st;
27d69a46 1077 struct packed_git *p = alloc_packed_git(path_len + 2);
1f688557 1078
e0500293
JN
1079 if (!have_set_try_to_free_routine) {
1080 have_set_try_to_free_routine = 1;
1081 set_try_to_free_routine(try_to_free_pack_memory);
1082 }
1083
42873078
NP
1084 /*
1085 * Make sure a corresponding .pack file exists and that
1086 * the index looks sane.
1087 */
1088 path_len -= strlen(".idx");
27d69a46
NP
1089 if (path_len < 1) {
1090 free(p);
1f688557 1091 return NULL;
27d69a46 1092 }
42873078 1093 memcpy(p->pack_name, path, path_len);
8d25931d
BC
1094
1095 strcpy(p->pack_name + path_len, ".keep");
1096 if (!access(p->pack_name, F_OK))
1097 p->pack_keep = 1;
1098
42873078 1099 strcpy(p->pack_name + path_len, ".pack");
d079837e 1100 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 1101 free(p);
1f688557
JH
1102 return NULL;
1103 }
42873078 1104
1f688557
JH
1105 /* ok, it looks sane as far as we can check without
1106 * actually mapping the pack file.
1107 */
1f688557 1108 p->pack_size = st.st_size;
9d835df2 1109 p->pack_local = local;
b867092f 1110 p->mtime = st.st_mtime;
42873078
NP
1111 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1112 hashclr(p->sha1);
1f688557
JH
1113 return p;
1114}
1115
7b64469a 1116struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 1117{
42873078 1118 const char *path = sha1_pack_name(sha1);
27d69a46 1119 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
bf592c50 1120
27d69a46
NP
1121 strcpy(p->pack_name, path);
1122 hashcpy(p->sha1, sha1);
42873078
NP
1123 if (check_packed_git_idx(idx_path, p)) {
1124 free(p);
bf592c50 1125 return NULL;
42873078 1126 }
bf592c50 1127
bf592c50
DB
1128 return p;
1129}
1130
1131void install_packed_git(struct packed_git *pack)
1132{
c7934306
SP
1133 if (pack->pack_fd != -1)
1134 pack_open_fds++;
1135
bf592c50
DB
1136 pack->next = packed_git;
1137 packed_git = pack;
1138}
1139
543c5caa
NTND
1140void (*report_garbage)(const char *desc, const char *path);
1141
1142static void report_helper(const struct string_list *list,
1143 int seen_bits, int first, int last)
1144{
1145 const char *msg;
1146 switch (seen_bits) {
1147 case 0:
235e8d59 1148 msg = "no corresponding .idx or .pack";
543c5caa
NTND
1149 break;
1150 case 1:
1151 msg = "no corresponding .idx";
1152 break;
1153 case 2:
1154 msg = "no corresponding .pack";
1155 break;
1156 default:
1157 return;
1158 }
1159 for (; first < last; first++)
1160 report_garbage(msg, list->items[first].string);
1161}
1162
1163static void report_pack_garbage(struct string_list *list)
1164{
1165 int i, baselen = -1, first = 0, seen_bits = 0;
1166
1167 if (!report_garbage)
1168 return;
1169
1170 sort_string_list(list);
1171
1172 for (i = 0; i < list->nr; i++) {
1173 const char *path = list->items[i].string;
1174 if (baselen != -1 &&
1175 strncmp(path, list->items[first].string, baselen)) {
1176 report_helper(list, seen_bits, first, i);
1177 baselen = -1;
1178 seen_bits = 0;
1179 }
1180 if (baselen == -1) {
1181 const char *dot = strrchr(path, '.');
1182 if (!dot) {
1183 report_garbage("garbage found", path);
1184 continue;
1185 }
1186 baselen = dot - path + 1;
1187 first = i;
1188 }
1189 if (!strcmp(path + baselen, "pack"))
1190 seen_bits |= 1;
1191 else if (!strcmp(path + baselen, "idx"))
1192 seen_bits |= 2;
1193 }
1194 report_helper(list, seen_bits, first, list->nr);
1195}
1196
9d835df2 1197static void prepare_packed_git_one(char *objdir, int local)
1f688557 1198{
880fb8de
RS
1199 struct strbuf path = STRBUF_INIT;
1200 size_t dirnamelen;
1f688557
JH
1201 DIR *dir;
1202 struct dirent *de;
543c5caa 1203 struct string_list garbage = STRING_LIST_INIT_DUP;
1f688557 1204
880fb8de
RS
1205 strbuf_addstr(&path, objdir);
1206 strbuf_addstr(&path, "/pack");
1207 dir = opendir(path.buf);
b5b16990 1208 if (!dir) {
26125f6b 1209 if (errno != ENOENT)
bd2afde8 1210 error("unable to open object pack directory: %s: %s",
880fb8de
RS
1211 path.buf, strerror(errno));
1212 strbuf_release(&path);
1f688557 1213 return;
b5b16990 1214 }
880fb8de
RS
1215 strbuf_addch(&path, '/');
1216 dirnamelen = path.len;
1f688557 1217 while ((de = readdir(dir)) != NULL) {
1f688557 1218 struct packed_git *p;
47bf4b0f 1219 size_t base_len;
543c5caa
NTND
1220
1221 if (is_dot_or_dotdot(de->d_name))
9cb18f56
JM
1222 continue;
1223
880fb8de
RS
1224 strbuf_setlen(&path, dirnamelen);
1225 strbuf_addstr(&path, de->d_name);
d90906a9 1226
47bf4b0f
JK
1227 base_len = path.len;
1228 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
d90906a9
NTND
1229 /* Don't reopen a pack we already have. */
1230 for (p = packed_git; p; p = p->next) {
47bf4b0f
JK
1231 size_t len;
1232 if (strip_suffix(p->pack_name, ".pack", &len) &&
1233 len == base_len &&
1234 !memcmp(p->pack_name, path.buf, len))
d90906a9
NTND
1235 break;
1236 }
1237 if (p == NULL &&
1238 /*
1239 * See if it really is a valid .idx file with
1240 * corresponding .pack file that we can map.
1241 */
880fb8de 1242 (p = add_packed_git(path.buf, path.len, local)) != NULL)
d90906a9 1243 install_packed_git(p);
86f7780c 1244 }
543c5caa
NTND
1245
1246 if (!report_garbage)
1247 continue;
1248
2975c770
JK
1249 if (ends_with(de->d_name, ".idx") ||
1250 ends_with(de->d_name, ".pack") ||
1251 ends_with(de->d_name, ".bitmap") ||
1252 ends_with(de->d_name, ".keep"))
880fb8de 1253 string_list_append(&garbage, path.buf);
543c5caa 1254 else
880fb8de 1255 report_garbage("garbage found", path.buf);
1f688557 1256 }
5b35bcd5 1257 closedir(dir);
543c5caa
NTND
1258 report_pack_garbage(&garbage);
1259 string_list_clear(&garbage, 0);
880fb8de 1260 strbuf_release(&path);
1f688557
JH
1261}
1262
b867092f
JH
1263static int sort_pack(const void *a_, const void *b_)
1264{
1265 struct packed_git *a = *((struct packed_git **)a_);
1266 struct packed_git *b = *((struct packed_git **)b_);
1267 int st;
1268
1269 /*
1270 * Local packs tend to contain objects specific to our
1271 * variant of the project than remote ones. In addition,
1272 * remote ones could be on a network mounted filesystem.
1273 * Favor local ones for these reasons.
1274 */
1275 st = a->pack_local - b->pack_local;
1276 if (st)
1277 return -st;
1278
1279 /*
1280 * Younger packs tend to contain more recent objects,
1281 * and more recent objects tend to get accessed more
1282 * often.
1283 */
1284 if (a->mtime < b->mtime)
1285 return 1;
1286 else if (a->mtime == b->mtime)
1287 return 0;
1288 return -1;
1289}
1290
1291static void rearrange_packed_git(void)
1292{
1293 struct packed_git **ary, *p;
1294 int i, n;
1295
1296 for (n = 0, p = packed_git; p; p = p->next)
1297 n++;
1298 if (n < 2)
1299 return;
1300
1301 /* prepare an array of packed_git for easier sorting */
1302 ary = xcalloc(n, sizeof(struct packed_git *));
1303 for (n = 0, p = packed_git; p; p = p->next)
1304 ary[n++] = p;
1305
1306 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1307
1308 /* link them back again */
1309 for (i = 0; i < n - 1; i++)
1310 ary[i]->next = ary[i + 1];
1311 ary[n - 1]->next = NULL;
1312 packed_git = ary[0];
1313
1314 free(ary);
1315}
1316
637cdd9d 1317static int prepare_packed_git_run_once = 0;
9a217f2a 1318void prepare_packed_git(void)
1f688557 1319{
d5a63b99 1320 struct alternate_object_database *alt;
1f688557 1321
637cdd9d 1322 if (prepare_packed_git_run_once)
1f688557 1323 return;
9d835df2 1324 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 1325 prepare_alt_odb();
d5a63b99 1326 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 1327 alt->name[-1] = 0;
9d835df2 1328 prepare_packed_git_one(alt->base, 0);
1494e038 1329 alt->name[-1] = '/';
1f688557 1330 }
b867092f 1331 rearrange_packed_git();
637cdd9d
JK
1332 prepare_packed_git_run_once = 1;
1333}
1334
fc04c412 1335void reprepare_packed_git(void)
637cdd9d
JK
1336{
1337 prepare_packed_git_run_once = 0;
1338 prepare_packed_git();
1f688557
JH
1339}
1340
8eca0b47
NP
1341static void mark_bad_packed_object(struct packed_git *p,
1342 const unsigned char *sha1)
1343{
1344 unsigned i;
1345 for (i = 0; i < p->num_bad_objects; i++)
1346 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1347 return;
1348 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1349 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1350 p->num_bad_objects++;
1351}
1352
b6c4cecc 1353static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
ac939109
NP
1354{
1355 struct packed_git *p;
1356 unsigned i;
1357
1358 for (p = packed_git; p; p = p->next)
1359 for (i = 0; i < p->num_bad_objects; i++)
1360 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
b6c4cecc
JH
1361 return p;
1362 return NULL;
ac939109
NP
1363}
1364
090ea126
NTND
1365/*
1366 * With an in-core object data in "map", rehash it to make sure the
1367 * object name actually matches "sha1" to detect object corruption.
1368 * With "map" == NULL, try reading the object named with "sha1" using
1369 * the streaming interface and rehash it to do the same.
1370 */
1371int check_sha1_signature(const unsigned char *sha1, void *map,
1372 unsigned long size, const char *type)
0fcfd160
LT
1373{
1374 unsigned char real_sha1[20];
090ea126
NTND
1375 enum object_type obj_type;
1376 struct git_istream *st;
1377 git_SHA_CTX c;
1378 char hdr[32];
1379 int hdrlen;
1380
1381 if (map) {
1382 hash_sha1_file(map, size, type, real_sha1);
1383 return hashcmp(sha1, real_sha1) ? -1 : 0;
1384 }
1385
1386 st = open_istream(sha1, &obj_type, &size, NULL);
1387 if (!st)
1388 return -1;
1389
1390 /* Generate the header */
1391 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1392
1393 /* Sha1.. */
1394 git_SHA1_Init(&c);
1395 git_SHA1_Update(&c, hdr, hdrlen);
1396 for (;;) {
1397 char buf[1024 * 16];
1398 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1399
f54fac53
JK
1400 if (readlen < 0) {
1401 close_istream(st);
1402 return -1;
1403 }
090ea126
NTND
1404 if (!readlen)
1405 break;
1406 git_SHA1_Update(&c, buf, readlen);
1407 }
1408 git_SHA1_Final(real_sha1, &c);
1409 close_istream(st);
a89fccd2 1410 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1411}
1412
ec73f580 1413int git_open_noatime(const char *name)
44d1c19e
LT
1414{
1415 static int sha1_file_open_flag = O_NOATIME;
44d1c19e 1416
f2e872aa
SP
1417 for (;;) {
1418 int fd = open(name, O_RDONLY | sha1_file_open_flag);
44d1c19e 1419 if (fd >= 0)
f2e872aa
SP
1420 return fd;
1421
f2e872aa
SP
1422 /* Might the failure be due to O_NOATIME? */
1423 if (errno != ENOENT && sha1_file_open_flag) {
44d1c19e 1424 sha1_file_open_flag = 0;
f2e872aa
SP
1425 continue;
1426 }
1427
1428 return -1;
44d1c19e 1429 }
44d1c19e
LT
1430}
1431
052fe5ea
JK
1432static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1433{
052fe5ea
JK
1434 struct alternate_object_database *alt;
1435
30d6c6ea 1436 if (!lstat(sha1_file_name(sha1), st))
052fe5ea
JK
1437 return 0;
1438
1439 prepare_alt_odb();
1440 errno = ENOENT;
1441 for (alt = alt_odb_list; alt; alt = alt->next) {
30d6c6ea 1442 fill_sha1_path(alt->name, sha1);
052fe5ea
JK
1443 if (!lstat(alt->base, st))
1444 return 0;
1445 }
1446
1447 return -1;
1448}
1449
44d1c19e
LT
1450static int open_sha1_file(const unsigned char *sha1)
1451{
1452 int fd;
44d1c19e 1453 struct alternate_object_database *alt;
d6c8a05b 1454 int most_interesting_errno;
44d1c19e 1455
30d6c6ea 1456 fd = git_open_noatime(sha1_file_name(sha1));
44d1c19e
LT
1457 if (fd >= 0)
1458 return fd;
d6c8a05b 1459 most_interesting_errno = errno;
44d1c19e
LT
1460
1461 prepare_alt_odb();
44d1c19e 1462 for (alt = alt_odb_list; alt; alt = alt->next) {
30d6c6ea 1463 fill_sha1_path(alt->name, sha1);
f4e51683 1464 fd = git_open_noatime(alt->base);
44d1c19e
LT
1465 if (fd >= 0)
1466 return fd;
d6c8a05b
JK
1467 if (most_interesting_errno == ENOENT)
1468 most_interesting_errno = errno;
44d1c19e 1469 }
d6c8a05b 1470 errno = most_interesting_errno;
44d1c19e
LT
1471 return -1;
1472}
1473
f0270efd 1474void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1475{
0fcfd160 1476 void *map;
144bde78 1477 int fd;
ace1534d 1478
44d1c19e
LT
1479 fd = open_sha1_file(sha1);
1480 map = NULL;
1481 if (fd >= 0) {
1482 struct stat st;
0fcfd160 1483
44d1c19e
LT
1484 if (!fstat(fd, &st)) {
1485 *size = xsize_t(st.st_size);
33e42de0
MM
1486 if (!*size) {
1487 /* mmap() is forbidden on empty files */
1488 error("object file %s is empty", sha1_file_name(sha1));
1489 return NULL;
1490 }
44d1c19e 1491 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1492 }
44d1c19e 1493 close(fd);
0fcfd160 1494 }
0fcfd160
LT
1495 return map;
1496}
1497
09ded04b
NP
1498unsigned long unpack_object_header_buffer(const unsigned char *buf,
1499 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1500{
ad1ed5ee 1501 unsigned shift;
48fb7deb 1502 unsigned long size, c;
ad1ed5ee
JH
1503 unsigned long used = 0;
1504
1505 c = buf[used++];
1506 *type = (c >> 4) & 7;
1507 size = c & 15;
1508 shift = 4;
1509 while (c & 0x80) {
f630cfda 1510 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1511 error("bad object header");
ea4f9685
JH
1512 size = used = 0;
1513 break;
09ded04b 1514 }
ad1ed5ee
JH
1515 c = buf[used++];
1516 size += (c & 0x7f) << shift;
1517 shift += 7;
1518 }
1519 *sizep = size;
1520 return used;
1521}
1522
eb4f4076 1523int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
ad1ed5ee 1524{
c4483576
LT
1525 /* Get the data stream */
1526 memset(stream, 0, sizeof(*stream));
1527 stream->next_in = map;
1528 stream->avail_in = mapsize;
1529 stream->next_out = buffer;
93821bd9
LT
1530 stream->avail_out = bufsiz;
1531
39c68542 1532 git_inflate_init(stream);
cc5c54e7 1533 return git_inflate(stream, 0);
c4483576
LT
1534}
1535
ef49a7a0 1536static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1537{
1538 int bytes = strlen(buffer) + 1;
3aee68aa 1539 unsigned char *buf = xmallocz(size);
93821bd9 1540 unsigned long n;
7efbff75 1541 int status = Z_OK;
5180cacc 1542
93821bd9
LT
1543 n = stream->total_out - bytes;
1544 if (n > size)
1545 n = size;
1546 memcpy(buf, (char *) buffer + bytes, n);
1547 bytes = n;
456cdf6e
LT
1548 if (bytes <= size) {
1549 /*
1550 * The above condition must be (bytes <= size), not
1551 * (bytes < size). In other words, even though we
ccf5ace0 1552 * expect no more output and set avail_out to zero,
456cdf6e
LT
1553 * the input zlib stream may have bytes that express
1554 * "this concludes the stream", and we *do* want to
1555 * eat that input.
1556 *
1557 * Otherwise we would not be able to test that we
1558 * consumed all the input to reach the expected size;
1559 * we also want to check that zlib tells us that all
1560 * went well with status == Z_STREAM_END at the end.
1561 */
5180cacc
LT
1562 stream->next_out = buf + bytes;
1563 stream->avail_out = size - bytes;
7efbff75 1564 while (status == Z_OK)
39c68542 1565 status = git_inflate(stream, Z_FINISH);
5180cacc 1566 }
456cdf6e 1567 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1568 git_inflate_end(stream);
7efbff75
JH
1569 return buf;
1570 }
1571
1572 if (status < 0)
1573 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1574 else if (stream->avail_in)
1575 error("garbage at end of loose object '%s'",
1576 sha1_to_hex(sha1));
1577 free(buf);
1578 return NULL;
5180cacc
LT
1579}
1580
1581/*
1582 * We used to just use "sscanf()", but that's actually way
1583 * too permissive for what we want to check. So do an anal
1584 * object header parse by hand.
1585 */
f0270efd 1586int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1587{
21666f1a 1588 char type[10];
5180cacc
LT
1589 int i;
1590 unsigned long size;
1591
1592 /*
a6080a0a 1593 * The type can be at most ten bytes (including the
5180cacc 1594 * terminating '\0' that we add), and is followed by
21666f1a 1595 * a space.
5180cacc 1596 */
21666f1a 1597 i = 0;
5180cacc
LT
1598 for (;;) {
1599 char c = *hdr++;
1600 if (c == ' ')
1601 break;
21666f1a
NP
1602 type[i++] = c;
1603 if (i >= sizeof(type))
5180cacc 1604 return -1;
5180cacc 1605 }
21666f1a 1606 type[i] = 0;
5180cacc
LT
1607
1608 /*
1609 * The length must follow immediately, and be in canonical
1610 * decimal format (ie "010" is not valid).
1611 */
1612 size = *hdr++ - '0';
1613 if (size > 9)
1614 return -1;
1615 if (size) {
1616 for (;;) {
1617 unsigned long c = *hdr - '0';
1618 if (c > 9)
1619 break;
1620 hdr++;
1621 size = size * 10 + c;
1622 }
1623 }
1624 *sizep = size;
1625
1626 /*
1627 * The length must be followed by a zero byte
1628 */
21666f1a 1629 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1630}
1631
7efbff75 1632static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1633{
5180cacc 1634 int ret;
ef49a7a0 1635 git_zstream stream;
5180cacc 1636 char hdr[8192];
0fcfd160 1637
5180cacc 1638 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1639 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1640 return NULL;
1641
7efbff75 1642 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1643}
1644
54dab52a
NP
1645unsigned long get_size_from_delta(struct packed_git *p,
1646 struct pack_window **w_curs,
1647 off_t curpos)
1648{
1649 const unsigned char *data;
1650 unsigned char delta_head[20], *in;
ef49a7a0 1651 git_zstream stream;
54dab52a
NP
1652 int st;
1653
1654 memset(&stream, 0, sizeof(stream));
1655 stream.next_out = delta_head;
1656 stream.avail_out = sizeof(delta_head);
1657
39c68542 1658 git_inflate_init(&stream);
54dab52a
NP
1659 do {
1660 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1661 stream.next_in = in;
39c68542 1662 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1663 curpos += stream.next_in - in;
1664 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1665 stream.total_out < sizeof(delta_head));
39c68542 1666 git_inflate_end(&stream);
3d77d877
NP
1667 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1668 error("delta data unpack-initial failed");
1669 return 0;
1670 }
54dab52a
NP
1671
1672 /* Examine the initial part of the delta to figure out
1673 * the result size.
1674 */
1675 data = delta_head;
1676
1677 /* ignore base size */
1678 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1679
1680 /* Read the result size */
1681 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1682}
1683
c4001d92 1684static off_t get_delta_base(struct packed_git *p,
03e79c88 1685 struct pack_window **w_curs,
c4001d92 1686 off_t *curpos,
21666f1a 1687 enum object_type type,
c4001d92 1688 off_t delta_obj_offset)
eb32d236 1689{
2b87c45b 1690 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1691 off_t base_offset;
eb32d236 1692
8d8a4ea5
SP
1693 /* use_pack() assured us we have [base_info, base_info + 20)
1694 * as a range that we can look at without walking off the
1695 * end of the mapped window. Its actually the hash size
1696 * that is assured. An OFS_DELTA longer than the hash size
1697 * is stupid, as then a REF_DELTA would be smaller to store.
1698 */
21666f1a 1699 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1700 unsigned used = 0;
1701 unsigned char c = base_info[used++];
1702 base_offset = c & 127;
1703 while (c & 128) {
1704 base_offset += 1;
8723f216 1705 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1706 return 0; /* overflow */
eb32d236
NP
1707 c = base_info[used++];
1708 base_offset = (base_offset << 7) + (c & 127);
1709 }
1710 base_offset = delta_obj_offset - base_offset;
d8f32556 1711 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1712 return 0; /* out of bound */
2b87c45b 1713 *curpos += used;
21666f1a 1714 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1715 /* The base entry _must_ be in the same pack */
1716 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1717 *curpos += 20;
eb32d236
NP
1718 } else
1719 die("I am totally screwed");
2b87c45b 1720 return base_offset;
eb32d236
NP
1721}
1722
5d642e75
JK
1723/*
1724 * Like get_delta_base above, but we return the sha1 instead of the pack
1725 * offset. This means it is cheaper for REF deltas (we do not have to do
1726 * the final object lookup), but more expensive for OFS deltas (we
1727 * have to load the revidx to convert the offset back into a sha1).
1728 */
1729static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1730 struct pack_window **w_curs,
1731 off_t curpos,
1732 enum object_type type,
1733 off_t delta_obj_offset)
1734{
1735 if (type == OBJ_REF_DELTA) {
1736 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1737 return base;
1738 } else if (type == OBJ_OFS_DELTA) {
1739 struct revindex_entry *revidx;
1740 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1741 type, delta_obj_offset);
1742
1743 if (!base_offset)
1744 return NULL;
1745
1746 revidx = find_pack_revindex(p, base_offset);
1747 if (!revidx)
1748 return NULL;
1749
1750 return nth_packed_object_sha1(p, revidx->nr);
1751 } else
1752 return NULL;
1753}
1754
f8c8abc5
JH
1755int unpack_object_header(struct packed_git *p,
1756 struct pack_window **w_curs,
1757 off_t *curpos,
1758 unsigned long *sizep)
a733cb60 1759{
03e79c88 1760 unsigned char *base;
ef49a7a0 1761 unsigned long left;
ad1ed5ee 1762 unsigned long used;
2b87c45b 1763 enum object_type type;
a733cb60 1764
8d8a4ea5 1765 /* use_pack() assures us we have [base, base + 20) available
0353a0c4 1766 * as a range that we can look at. (Its actually the hash
3dff5379 1767 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1768 * the maximum deflated object size is 2^137, which is just
1769 * insane, so we know won't exceed what we have been given.
1770 */
2b87c45b 1771 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1772 used = unpack_object_header_buffer(base, left, &type, sizep);
1773 if (!used) {
1774 type = OBJ_BAD;
1775 } else
1776 *curpos += used;
ad1ed5ee 1777
2b87c45b 1778 return type;
a733cb60
LT
1779}
1780
790d96c0
TR
1781static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1782{
1783 int type;
1784 struct revindex_entry *revidx;
1785 const unsigned char *sha1;
1786 revidx = find_pack_revindex(p, obj_offset);
1787 if (!revidx)
1788 return OBJ_BAD;
1789 sha1 = nth_packed_object_sha1(p, revidx->nr);
1790 mark_bad_packed_object(p, sha1);
1791 type = sha1_object_info(sha1, NULL);
1792 if (type <= OBJ_NONE)
1793 return OBJ_BAD;
1794 return type;
1795}
1796
790d96c0
TR
1797#define POI_STACK_PREALLOC 64
1798
90191d37
JK
1799static enum object_type packed_to_object_type(struct packed_git *p,
1800 off_t obj_offset,
1801 enum object_type type,
1802 struct pack_window **w_curs,
1803 off_t curpos)
1f688557 1804{
790d96c0
TR
1805 off_t small_poi_stack[POI_STACK_PREALLOC];
1806 off_t *poi_stack = small_poi_stack;
1807 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
5db47c2b 1808
790d96c0
TR
1809 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1810 off_t base_offset;
90191d37 1811 unsigned long size;
790d96c0
TR
1812 /* Push the object we're going to leave behind */
1813 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1814 poi_stack_alloc = alloc_nr(poi_stack_nr);
1815 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1816 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1817 } else {
1818 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1819 }
1820 poi_stack[poi_stack_nr++] = obj_offset;
1821 /* If parsing the base offset fails, just unwind */
90191d37 1822 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
790d96c0
TR
1823 if (!base_offset)
1824 goto unwind;
1825 curpos = obj_offset = base_offset;
90191d37 1826 type = unpack_object_header(p, w_curs, &curpos, &size);
790d96c0
TR
1827 if (type <= OBJ_NONE) {
1828 /* If getting the base itself fails, we first
1829 * retry the base, otherwise unwind */
1830 type = retry_bad_packed_offset(p, base_offset);
1831 if (type > OBJ_NONE)
1832 goto out;
1833 goto unwind;
1834 }
1835 }
1836
21666f1a 1837 switch (type) {
790d96c0 1838 case OBJ_BAD:
a733cb60 1839 case OBJ_COMMIT:
a733cb60 1840 case OBJ_TREE:
a733cb60 1841 case OBJ_BLOB:
a733cb60 1842 case OBJ_TAG:
a69d0943 1843 break;
1f688557 1844 default:
3d77d877
NP
1845 error("unknown object type %i at offset %"PRIuMAX" in %s",
1846 type, (uintmax_t)obj_offset, p->pack_name);
1847 type = OBJ_BAD;
1f688557 1848 }
790d96c0
TR
1849
1850out:
1851 if (poi_stack != small_poi_stack)
1852 free(poi_stack);
21666f1a 1853 return type;
790d96c0
TR
1854
1855unwind:
1856 while (poi_stack_nr) {
1857 obj_offset = poi_stack[--poi_stack_nr];
1858 type = retry_bad_packed_offset(p, obj_offset);
1859 if (type > OBJ_NONE)
1860 goto out;
1861 }
1862 type = OBJ_BAD;
1863 goto out;
1f688557
JH
1864}
1865
90191d37 1866static int packed_object_info(struct packed_git *p, off_t obj_offset,
23c339c0 1867 struct object_info *oi)
90191d37
JK
1868{
1869 struct pack_window *w_curs = NULL;
1870 unsigned long size;
1871 off_t curpos = obj_offset;
1872 enum object_type type;
1873
412916ee
JK
1874 /*
1875 * We always get the representation type, but only convert it to
1876 * a "real" type later if the caller is interested.
1877 */
90191d37
JK
1878 type = unpack_object_header(p, &w_curs, &curpos, &size);
1879
23c339c0 1880 if (oi->sizep) {
90191d37
JK
1881 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1882 off_t tmp_pos = curpos;
1883 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1884 type, obj_offset);
1885 if (!base_offset) {
1886 type = OBJ_BAD;
1887 goto out;
1888 }
23c339c0
JK
1889 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1890 if (*oi->sizep == 0) {
90191d37
JK
1891 type = OBJ_BAD;
1892 goto out;
1893 }
1894 } else {
23c339c0 1895 *oi->sizep = size;
90191d37
JK
1896 }
1897 }
1898
23c339c0 1899 if (oi->disk_sizep) {
90191d37 1900 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
23c339c0 1901 *oi->disk_sizep = revidx[1].offset - obj_offset;
90191d37
JK
1902 }
1903
23c339c0
JK
1904 if (oi->typep) {
1905 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
1906 if (*oi->typep < 0) {
412916ee
JK
1907 type = OBJ_BAD;
1908 goto out;
1909 }
1910 }
90191d37 1911
5d642e75
JK
1912 if (oi->delta_base_sha1) {
1913 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1914 const unsigned char *base;
1915
1916 base = get_delta_base_sha1(p, &w_curs, curpos,
1917 type, obj_offset);
1918 if (!base) {
1919 type = OBJ_BAD;
1920 goto out;
1921 }
1922
1923 hashcpy(oi->delta_base_sha1, base);
1924 } else
1925 hashclr(oi->delta_base_sha1);
1926 }
1927
90191d37
JK
1928out:
1929 unuse_pack(&w_curs);
1930 return type;
1931}
1932
eb950c19 1933static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1934 struct pack_window **w_curs,
c4001d92 1935 off_t curpos,
eb950c19 1936 unsigned long size)
de530aaa
SP
1937{
1938 int st;
ef49a7a0 1939 git_zstream stream;
079afb18 1940 unsigned char *buffer, *in;
de530aaa 1941
735efde8
NTND
1942 buffer = xmallocz_gently(size);
1943 if (!buffer)
1944 return NULL;
de530aaa 1945 memset(&stream, 0, sizeof(stream));
de530aaa 1946 stream.next_out = buffer;
39eea7bd 1947 stream.avail_out = size + 1;
de530aaa 1948
39c68542 1949 git_inflate_init(&stream);
079afb18 1950 do {
2b87c45b 1951 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 1952 stream.next_in = in;
39c68542 1953 st = git_inflate(&stream, Z_FINISH);
39eea7bd
JH
1954 if (!stream.avail_out)
1955 break; /* the payload is larger than it should be */
2b87c45b 1956 curpos += stream.next_in - in;
079afb18 1957 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 1958 git_inflate_end(&stream);
de530aaa
SP
1959 if ((st != Z_STREAM_END) || stream.total_out != size) {
1960 free(buffer);
1961 return NULL;
1962 }
1963
1964 return buffer;
1965}
1966
e5e01619
LT
1967#define MAX_DELTA_CACHE (256)
1968
18bdec11 1969static size_t delta_base_cached;
5e08ecbf
NP
1970
1971static struct delta_base_cache_lru_list {
1972 struct delta_base_cache_lru_list *prev;
1973 struct delta_base_cache_lru_list *next;
1974} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1975
e5e01619 1976static struct delta_base_cache_entry {
5e08ecbf
NP
1977 struct delta_base_cache_lru_list lru;
1978 void *data;
e5e01619
LT
1979 struct packed_git *p;
1980 off_t base_offset;
1981 unsigned long size;
e5e01619
LT
1982 enum object_type type;
1983} delta_base_cache[MAX_DELTA_CACHE];
1984
1985static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1986{
1987 unsigned long hash;
1988
1989 hash = (unsigned long)p + (unsigned long)base_offset;
1990 hash += (hash >> 8) + (hash >> 16);
3358004a 1991 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1992}
1993
84dd81c1
TR
1994static struct delta_base_cache_entry *
1995get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
5266d369
JH
1996{
1997 unsigned long hash = pack_entry_hash(p, base_offset);
84dd81c1
TR
1998 return delta_base_cache + hash;
1999}
2000
2001static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
2002 struct packed_git *p, off_t base_offset)
2003{
5266d369
JH
2004 return (ent->data && ent->p == p && ent->base_offset == base_offset);
2005}
2006
84dd81c1
TR
2007static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2008{
2009 struct delta_base_cache_entry *ent;
2010 ent = get_delta_base_cache_entry(p, base_offset);
2011 return eq_delta_base_cache_entry(ent, p, base_offset);
2012}
2013
2014static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2015{
2016 ent->data = NULL;
2017 ent->lru.next->prev = ent->lru.prev;
2018 ent->lru.prev->next = ent->lru.next;
2019 delta_base_cached -= ent->size;
2020}
2021
62f255ad 2022static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 2023 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 2024{
84dd81c1 2025 struct delta_base_cache_entry *ent;
e5e01619 2026 void *ret;
e5e01619 2027
84dd81c1
TR
2028 ent = get_delta_base_cache_entry(p, base_offset);
2029
2030 if (!eq_delta_base_cache_entry(ent, p, base_offset))
749bc58c 2031 return unpack_entry(p, base_offset, type, base_size);
e5e01619 2032
84dd81c1
TR
2033 ret = ent->data;
2034
2035 if (!keep_cache)
2036 clear_delta_base_cache_entry(ent);
2037 else
182af834 2038 ret = xmemdupz(ent->data, ent->size);
e5e01619
LT
2039 *type = ent->type;
2040 *base_size = ent->size;
2041 return ret;
62f255ad
LT
2042}
2043
18bdec11
SP
2044static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2045{
2046 if (ent->data) {
2047 free(ent->data);
2048 ent->data = NULL;
5e08ecbf
NP
2049 ent->lru.next->prev = ent->lru.prev;
2050 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
2051 delta_base_cached -= ent->size;
2052 }
2053}
2054
3d20c636
SP
2055void clear_delta_base_cache(void)
2056{
2057 unsigned long p;
2058 for (p = 0; p < MAX_DELTA_CACHE; p++)
2059 release_delta_base_cache(&delta_base_cache[p]);
2060}
2061
62f255ad
LT
2062static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2063 void *base, unsigned long base_size, enum object_type type)
2064{
5e08ecbf 2065 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 2066 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 2067 struct delta_base_cache_lru_list *lru;
e5e01619 2068
18bdec11
SP
2069 release_delta_base_cache(ent);
2070 delta_base_cached += base_size;
5e08ecbf
NP
2071
2072 for (lru = delta_base_cache_lru.next;
2073 delta_base_cached > delta_base_cache_limit
2074 && lru != &delta_base_cache_lru;
2075 lru = lru->next) {
2076 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
2077 if (f->type == OBJ_BLOB)
2078 release_delta_base_cache(f);
2079 }
5e08ecbf
NP
2080 for (lru = delta_base_cache_lru.next;
2081 delta_base_cached > delta_base_cache_limit
2082 && lru != &delta_base_cache_lru;
2083 lru = lru->next) {
2084 struct delta_base_cache_entry *f = (void *)lru;
2085 release_delta_base_cache(f);
2086 }
18bdec11 2087
e5e01619
LT
2088 ent->p = p;
2089 ent->base_offset = base_offset;
2090 ent->type = type;
2091 ent->data = base;
2092 ent->size = base_size;
5e08ecbf
NP
2093 ent->lru.next = &delta_base_cache_lru;
2094 ent->lru.prev = delta_base_cache_lru.prev;
2095 delta_base_cache_lru.prev->next = &ent->lru;
2096 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
2097}
2098
c2c5b270
CC
2099static void *read_object(const unsigned char *sha1, enum object_type *type,
2100 unsigned long *size);
2101
5f44324d
JH
2102static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2103{
67dc598e
KB
2104 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2105 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2106 p->pack_name, (uintmax_t)obj_offset);
5f44324d
JH
2107}
2108
0e8189e2
NP
2109int do_check_packed_object_crc;
2110
abe601bb
TR
2111#define UNPACK_ENTRY_STACK_PREALLOC 64
2112struct unpack_entry_stack_ent {
2113 off_t obj_offset;
2114 off_t curpos;
2115 unsigned long size;
2116};
2117
c4001d92 2118void *unpack_entry(struct packed_git *p, off_t obj_offset,
abe601bb 2119 enum object_type *final_type, unsigned long *final_size)
1f688557 2120{
03e79c88 2121 struct pack_window *w_curs = NULL;
c4001d92 2122 off_t curpos = obj_offset;
abe601bb
TR
2123 void *data = NULL;
2124 unsigned long size;
2125 enum object_type type;
2126 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2127 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2128 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2129 int base_from_cache = 0;
1f688557 2130
67dc598e 2131 write_pack_access_log(p, obj_offset);
5f44324d 2132
abe601bb
TR
2133 /* PHASE 1: drill down to the innermost base object */
2134 for (;;) {
2135 off_t base_offset;
2136 int i;
2137 struct delta_base_cache_entry *ent;
2138
77965f8b
NTND
2139 ent = get_delta_base_cache_entry(p, curpos);
2140 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2141 type = ent->type;
2142 data = ent->data;
2143 size = ent->size;
2144 clear_delta_base_cache_entry(ent);
2145 base_from_cache = 1;
2146 break;
2147 }
2148
abe601bb
TR
2149 if (do_check_packed_object_crc && p->index_version > 1) {
2150 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2151 unsigned long len = revidx[1].offset - obj_offset;
2152 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2153 const unsigned char *sha1 =
2154 nth_packed_object_sha1(p, revidx->nr);
2155 error("bad packed object CRC for %s",
2156 sha1_to_hex(sha1));
2157 mark_bad_packed_object(p, sha1);
2158 unuse_pack(&w_curs);
2159 return NULL;
2160 }
2161 }
2162
abe601bb
TR
2163 type = unpack_object_header(p, &w_curs, &curpos, &size);
2164 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2165 break;
2166
2167 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2168 if (!base_offset) {
2169 error("failed to validate delta base reference "
2170 "at offset %"PRIuMAX" from %s",
2171 (uintmax_t)curpos, p->pack_name);
2172 /* bail to phase 2, in hopes of recovery */
2173 data = NULL;
2174 break;
2175 }
2176
2177 /* push object, proceed to base */
2178 if (delta_stack_nr >= delta_stack_alloc
2179 && delta_stack == small_delta_stack) {
2180 delta_stack_alloc = alloc_nr(delta_stack_nr);
2181 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2182 memcpy(delta_stack, small_delta_stack,
2183 sizeof(*delta_stack)*delta_stack_nr);
2184 } else {
2185 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2186 }
2187 i = delta_stack_nr++;
2188 delta_stack[i].obj_offset = obj_offset;
2189 delta_stack[i].curpos = curpos;
2190 delta_stack[i].size = size;
2191
2192 curpos = obj_offset = base_offset;
0e8189e2
NP
2193 }
2194
abe601bb
TR
2195 /* PHASE 2: handle the base */
2196 switch (type) {
eb32d236
NP
2197 case OBJ_OFS_DELTA:
2198 case OBJ_REF_DELTA:
abe601bb
TR
2199 if (data)
2200 die("BUG in unpack_entry: left loop at a valid delta");
4d703a1a 2201 break;
a733cb60 2202 case OBJ_COMMIT:
a733cb60 2203 case OBJ_TREE:
a733cb60 2204 case OBJ_BLOB:
a733cb60 2205 case OBJ_TAG:
abe601bb
TR
2206 if (!base_from_cache)
2207 data = unpack_compressed_entry(p, &w_curs, curpos, size);
4d703a1a 2208 break;
1f688557 2209 default:
8eca0b47
NP
2210 data = NULL;
2211 error("unknown object type %i at offset %"PRIuMAX" in %s",
abe601bb
TR
2212 type, (uintmax_t)obj_offset, p->pack_name);
2213 }
2214
2215 /* PHASE 3: apply deltas in order */
2216
2217 /* invariants:
2218 * 'data' holds the base data, or NULL if there was corruption
2219 */
2220 while (delta_stack_nr) {
2221 void *delta_data;
2222 void *base = data;
2223 unsigned long delta_size, base_size = size;
2224 int i;
2225
2226 data = NULL;
2227
2228 if (base)
2229 add_delta_base_cache(p, obj_offset, base, base_size, type);
2230
2231 if (!base) {
2232 /*
2233 * We're probably in deep shit, but let's try to fetch
2234 * the required base anyway from another pack or loose.
2235 * This is costly but should happen only in the presence
2236 * of a corrupted pack, and is better than failing outright.
2237 */
2238 struct revindex_entry *revidx;
2239 const unsigned char *base_sha1;
2240 revidx = find_pack_revindex(p, obj_offset);
2241 if (revidx) {
2242 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2243 error("failed to read delta base object %s"
2244 " at offset %"PRIuMAX" from %s",
2245 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2246 p->pack_name);
2247 mark_bad_packed_object(p, base_sha1);
2248 base = read_object(base_sha1, &type, &base_size);
2249 }
2250 }
2251
2252 i = --delta_stack_nr;
2253 obj_offset = delta_stack[i].obj_offset;
2254 curpos = delta_stack[i].curpos;
2255 delta_size = delta_stack[i].size;
2256
2257 if (!base)
2258 continue;
2259
2260 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2261
2262 if (!delta_data) {
2263 error("failed to unpack compressed delta "
2264 "at offset %"PRIuMAX" from %s",
2265 (uintmax_t)curpos, p->pack_name);
abe601bb
TR
2266 data = NULL;
2267 continue;
2268 }
2269
2270 data = patch_delta(base, base_size,
2271 delta_data, delta_size,
2272 &size);
1ee886c1
JK
2273
2274 /*
2275 * We could not apply the delta; warn the user, but keep going.
2276 * Our failure will be noticed either in the next iteration of
2277 * the loop, or if this is the final delta, in the caller when
2278 * we return NULL. Those code paths will take care of making
2279 * a more explicit warning and retrying with another copy of
2280 * the object.
2281 */
abe601bb 2282 if (!data)
1ee886c1 2283 error("failed to apply delta");
abe601bb 2284
4b8f772c 2285 free(delta_data);
1f688557 2286 }
abe601bb
TR
2287
2288 *final_type = type;
2289 *final_size = size;
2290
03e79c88 2291 unuse_pack(&w_curs);
019d1e65
NTND
2292
2293 if (delta_stack != small_delta_stack)
2294 free(delta_stack);
2295
21666f1a 2296 return data;
1f688557
JH
2297}
2298
d079837e 2299const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 2300 uint32_t n)
9a217f2a 2301{
42873078 2302 const unsigned char *index = p->index_data;
d079837e
SP
2303 if (!index) {
2304 if (open_pack_index(p))
2305 return NULL;
2306 index = p->index_data;
2307 }
57059091 2308 if (n >= p->num_objects)
d72308e0 2309 return NULL;
74e34e1f
NP
2310 index += 4 * 256;
2311 if (p->index_version == 1) {
2312 return index + 24 * n + 4;
2313 } else {
2314 index += 8;
2315 return index + 20 * n;
2316 }
2317}
2318
99093238 2319off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
2320{
2321 const unsigned char *index = p->index_data;
2322 index += 4 * 256;
2323 if (p->index_version == 1) {
2324 return ntohl(*((uint32_t *)(index + 24 * n)));
2325 } else {
2326 uint32_t off;
2327 index += 8 + p->num_objects * (20 + 4);
2328 off = ntohl(*((uint32_t *)(index + 4 * n)));
2329 if (!(off & 0x80000000))
2330 return off;
2331 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2332 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2333 ntohl(*((uint32_t *)(index + 4)));
2334 }
9a217f2a
JH
2335}
2336
c4001d92 2337off_t find_pack_entry_one(const unsigned char *sha1,
43057304 2338 struct packed_git *p)
1f688557 2339{
42873078 2340 const uint32_t *level1_ofs = p->index_data;
42873078 2341 const unsigned char *index = p->index_data;
628522ec
JH
2342 unsigned hi, lo, stride;
2343 static int use_lookup = -1;
2344 static int debug_lookup = -1;
2345
2346 if (debug_lookup < 0)
2347 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 2348
d079837e
SP
2349 if (!index) {
2350 if (open_pack_index(p))
2351 return 0;
2352 level1_ofs = p->index_data;
2353 index = p->index_data;
2354 }
74e34e1f
NP
2355 if (p->index_version > 1) {
2356 level1_ofs += 2;
2357 index += 8;
2358 }
42873078 2359 index += 4 * 256;
74e34e1f
NP
2360 hi = ntohl(level1_ofs[*sha1]);
2361 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
2362 if (p->index_version > 1) {
2363 stride = 20;
2364 } else {
2365 stride = 24;
2366 index += 4;
2367 }
2368
2369 if (debug_lookup)
6e1c2344 2370 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
2371 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2372
2373 if (use_lookup < 0)
2374 use_lookup = !!getenv("GIT_USE_LOOKUP");
2375 if (use_lookup) {
2376 int pos = sha1_entry_pos(index, stride, 0,
2377 lo, hi, p->num_objects, sha1);
2378 if (pos < 0)
2379 return 0;
2380 return nth_packed_object_offset(p, pos);
2381 }
1f688557
JH
2382
2383 do {
74e34e1f 2384 unsigned mi = (lo + hi) / 2;
628522ec
JH
2385 int cmp = hashcmp(index + mi * stride, sha1);
2386
2387 if (debug_lookup)
2388 printf("lo %u hi %u rg %u mi %u\n",
2389 lo, hi, hi - lo, mi);
43057304 2390 if (!cmp)
74e34e1f 2391 return nth_packed_object_offset(p, mi);
1f688557
JH
2392 if (cmp > 0)
2393 hi = mi;
2394 else
2395 lo = mi+1;
2396 } while (lo < hi);
2397 return 0;
2398}
2399
4c080182 2400int is_pack_valid(struct packed_git *p)
d131b7af
SP
2401{
2402 /* An already open pack is known to be valid. */
2403 if (p->pack_fd != -1)
2404 return 1;
2405
2406 /* If the pack has one window completely covering the
2407 * file size, the pack is known to be valid even if
2408 * the descriptor is not currently open.
2409 */
2410 if (p->windows) {
2411 struct pack_window *w = p->windows;
2412
2413 if (!w->offset && w->len == p->pack_size)
2414 return 1;
2415 }
2416
2417 /* Force the pack to open to prove its valid. */
2418 return !open_packed_git(p);
2419}
2420
95099731
NTND
2421static int fill_pack_entry(const unsigned char *sha1,
2422 struct pack_entry *e,
2423 struct packed_git *p)
2424{
2425 off_t offset;
2426
2427 if (p->num_bad_objects) {
2428 unsigned i;
2429 for (i = 0; i < p->num_bad_objects; i++)
2430 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2431 return 0;
2432 }
2433
2434 offset = find_pack_entry_one(sha1, p);
2435 if (!offset)
2436 return 0;
2437
2438 /*
2439 * We are about to tell the caller where they can locate the
2440 * requested object. We better make sure the packfile is
2441 * still here and can be accessed before supplying that
2442 * answer, as it may have been deleted since the index was
2443 * loaded!
2444 */
2445 if (!is_pack_valid(p)) {
2446 warning("packfile %s cannot be accessed", p->pack_name);
2447 return 0;
2448 }
2449 e->offset = offset;
2450 e->p = p;
2451 hashcpy(e->sha1, sha1);
2452 return 1;
2453}
2454
d40d535b
MH
2455/*
2456 * Iff a pack file contains the object named by sha1, return true and
2457 * store its location to e.
2458 */
4d6acb70 2459static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557
JH
2460{
2461 struct packed_git *p;
43057304 2462
1f688557 2463 prepare_packed_git();
f7c22cc6
NP
2464 if (!packed_git)
2465 return 0;
1f688557 2466
c01f51cc
NTND
2467 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2468 return 1;
8eca0b47 2469
c01f51cc 2470 for (p = packed_git; p; p = p->next) {
1b1005d1
MH
2471 if (p == last_found_pack)
2472 continue; /* we already checked this one */
f7c22cc6 2473
1b1005d1
MH
2474 if (fill_pack_entry(sha1, e, p)) {
2475 last_found_pack = p;
2476 return 1;
2477 }
c01f51cc 2478 }
1f688557
JH
2479 return 0;
2480}
2481
a6080a0a 2482struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
2483 struct packed_git *packs)
2484{
2485 struct packed_git *p;
bf592c50
DB
2486
2487 for (p = packs; p; p = p->next) {
43057304 2488 if (find_pack_entry_one(sha1, p))
bf592c50
DB
2489 return p;
2490 }
2491 return NULL;
21666f1a 2492
bf592c50
DB
2493}
2494
052fe5ea 2495static int sha1_loose_object_info(const unsigned char *sha1,
23c339c0 2496 struct object_info *oi)
65c2e0c3 2497{
36e4d74a 2498 int status;
65c2e0c3
JH
2499 unsigned long mapsize, size;
2500 void *map;
ef49a7a0 2501 git_zstream stream;
d65a16f6 2502 char hdr[32];
65c2e0c3 2503
5d642e75
JK
2504 if (oi->delta_base_sha1)
2505 hashclr(oi->delta_base_sha1);
2506
052fe5ea
JK
2507 /*
2508 * If we don't care about type or size, then we don't
4ef8d1dd
JH
2509 * need to look inside the object at all. Note that we
2510 * do not optimize out the stat call, even if the
2511 * caller doesn't care about the disk-size, since our
2512 * return value implicitly indicates whether the
2513 * object even exists.
052fe5ea 2514 */
23c339c0 2515 if (!oi->typep && !oi->sizep) {
4ef8d1dd
JH
2516 struct stat st;
2517 if (stat_sha1_file(sha1, &st) < 0)
2518 return -1;
2519 if (oi->disk_sizep)
23c339c0 2520 *oi->disk_sizep = st.st_size;
052fe5ea
JK
2521 return 0;
2522 }
2523
bb6b8e4f 2524 map = map_sha1_file(sha1, &mapsize);
f0df4ed5 2525 if (!map)
dbea72a8 2526 return -1;
23c339c0
JK
2527 if (oi->disk_sizep)
2528 *oi->disk_sizep = mapsize;
36e4d74a
JH
2529 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2530 status = error("unable to unpack %s header",
2531 sha1_to_hex(sha1));
21666f1a 2532 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 2533 status = error("unable to parse %s header", sha1_to_hex(sha1));
23c339c0
JK
2534 else if (oi->sizep)
2535 *oi->sizep = size;
39c68542 2536 git_inflate_end(&stream);
65c2e0c3 2537 munmap(map, mapsize);
23c339c0
JK
2538 if (oi->typep)
2539 *oi->typep = status;
052fe5ea 2540 return 0;
65c2e0c3
JH
2541}
2542
de7b5d62 2543int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
f0df4ed5 2544{
c4d9986f 2545 struct cached_object *co;
f0df4ed5 2546 struct pack_entry e;
5b086407 2547 int rtype;
1f7117ef 2548 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
f0df4ed5 2549
1f7117ef 2550 co = find_cached_object(real);
c4d9986f 2551 if (co) {
5b086407
JK
2552 if (oi->typep)
2553 *(oi->typep) = co->type;
9a490590
JH
2554 if (oi->sizep)
2555 *(oi->sizep) = co->size;
161f00e7
JK
2556 if (oi->disk_sizep)
2557 *(oi->disk_sizep) = 0;
5d642e75
JK
2558 if (oi->delta_base_sha1)
2559 hashclr(oi->delta_base_sha1);
9a490590 2560 oi->whence = OI_CACHED;
5b086407 2561 return 0;
c4d9986f
NTND
2562 }
2563
1f7117ef 2564 if (!find_pack_entry(real, &e)) {
ddd63e64 2565 /* Most likely it's a loose object. */
1f7117ef 2566 if (!sha1_loose_object_info(real, oi)) {
9a490590 2567 oi->whence = OI_LOOSE;
5b086407 2568 return 0;
9a490590 2569 }
ddd63e64
SG
2570
2571 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 2572 reprepare_packed_git();
1f7117ef 2573 if (!find_pack_entry(real, &e))
052fe5ea 2574 return -1;
f0df4ed5 2575 }
3d77d877 2576
23c339c0 2577 rtype = packed_object_info(e.p, e.offset, oi);
412916ee 2578 if (rtype < 0) {
1f7117ef
CC
2579 mark_bad_packed_object(e.p, real);
2580 return sha1_object_info_extended(real, oi, 0);
5266d369
JH
2581 } else if (in_delta_base_cache(e.p, e.offset)) {
2582 oi->whence = OI_DBCACHED;
9a490590
JH
2583 } else {
2584 oi->whence = OI_PACKED;
2585 oi->u.packed.offset = e.offset;
2586 oi->u.packed.pack = e.p;
2587 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2588 rtype == OBJ_OFS_DELTA);
3d77d877
NP
2589 }
2590
5b086407 2591 return 0;
f0df4ed5
JS
2592}
2593
3fc0dca9 2594/* returns enum object_type or negative */
9a490590
JH
2595int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2596{
5b086407 2597 enum object_type type;
d099b717 2598 struct object_info oi = {NULL};
9a490590 2599
5b086407 2600 oi.typep = &type;
9a490590 2601 oi.sizep = sizep;
de7b5d62 2602 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
5b086407
JK
2603 return -1;
2604 return type;
9a490590
JH
2605}
2606
21666f1a
NP
2607static void *read_packed_sha1(const unsigned char *sha1,
2608 enum object_type *type, unsigned long *size)
1f688557
JH
2609{
2610 struct pack_entry e;
8eca0b47 2611 void *data;
1f688557 2612
cd673c1f 2613 if (!find_pack_entry(sha1, &e))
1f688557 2614 return NULL;
8eca0b47
NP
2615 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2616 if (!data) {
2617 /*
2618 * We're probably in deep shit, but let's try to fetch
2619 * the required object anyway from another pack or loose.
2620 * This should happen only in the presence of a corrupted
2621 * pack, and is better than failing outright.
2622 */
2623 error("failed to read object %s at offset %"PRIuMAX" from %s",
2624 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2625 mark_bad_packed_object(e.p, sha1);
ac939109 2626 data = read_object(sha1, type, size);
8eca0b47
NP
2627 }
2628 return data;
1f688557
JH
2629}
2630
21666f1a
NP
2631int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2632 unsigned char *sha1)
d66b37bb
JH
2633{
2634 struct cached_object *co;
2635
21666f1a 2636 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2637 if (has_sha1_file(sha1) || find_cached_object(sha1))
2638 return 0;
c7353967 2639 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
d66b37bb
JH
2640 co = &cached_objects[cached_object_nr++];
2641 co->size = len;
21666f1a 2642 co->type = type;
efa13f7b
JH
2643 co->buf = xmalloc(len);
2644 memcpy(co->buf, buf, len);
d66b37bb
JH
2645 hashcpy(co->sha1, sha1);
2646 return 0;
2647}
2648
c2c5b270
CC
2649static void *read_object(const unsigned char *sha1, enum object_type *type,
2650 unsigned long *size)
0fcfd160
LT
2651{
2652 unsigned long mapsize;
2653 void *map, *buf;
d66b37bb
JH
2654 struct cached_object *co;
2655
2656 co = find_cached_object(sha1);
2657 if (co) {
21666f1a 2658 *type = co->type;
d66b37bb 2659 *size = co->size;
182af834 2660 return xmemdupz(co->buf, co->size);
d66b37bb 2661 }
0fcfd160 2662
8276c007
PE
2663 buf = read_packed_sha1(sha1, type, size);
2664 if (buf)
2665 return buf;
bb6b8e4f 2666 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2667 if (map) {
7efbff75 2668 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2669 munmap(map, mapsize);
2670 return buf;
2671 }
637cdd9d 2672 reprepare_packed_git();
8276c007 2673 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2674}
2675
b6c4cecc
JH
2676/*
2677 * This function dies on corrupt objects; the callers who want to
2678 * deal with them should arrange to call read_object() and give error
2679 * messages themselves.
2680 */
5bf29b95
JH
2681void *read_sha1_file_extended(const unsigned char *sha1,
2682 enum object_type *type,
2683 unsigned long *size,
2684 unsigned flag)
ac939109 2685{
3ba7a065 2686 void *data;
b6c4cecc 2687 const struct packed_git *p;
bf93eea0 2688 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
b6c4cecc 2689
3ba7a065
JH
2690 errno = 0;
2691 data = read_object(repl, type, size);
4bbf5a26 2692 if (data)
b6c4cecc 2693 return data;
68095570 2694
25f3af3f 2695 if (errno && errno != ENOENT)
3ba7a065
JH
2696 die_errno("failed to read object %s", sha1_to_hex(sha1));
2697
68095570 2698 /* die if we replaced an object with one that does not exist */
b6c4cecc 2699 if (repl != sha1)
68095570
CC
2700 die("replacement %s not found for %s",
2701 sha1_to_hex(repl), sha1_to_hex(sha1));
2702
b6c4cecc 2703 if (has_loose_object(repl)) {
30d6c6ea
MH
2704 const char *path = sha1_file_name(sha1);
2705
b6c4cecc
JH
2706 die("loose object %s (stored in %s) is corrupt",
2707 sha1_to_hex(repl), path);
e8b15e61 2708 }
68095570 2709
b6c4cecc
JH
2710 if ((p = has_packed_and_bad(repl)) != NULL)
2711 die("packed object %s (stored in %s) is corrupt",
2712 sha1_to_hex(repl), p->pack_name);
f5552aee 2713
b6c4cecc 2714 return NULL;
ac939109
NP
2715}
2716
40469ee9 2717void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2718 const char *required_type_name,
40469ee9
JH
2719 unsigned long *size,
2720 unsigned char *actual_sha1_return)
f4913f91 2721{
21666f1a 2722 enum object_type type, required_type;
f4913f91
JH
2723 void *buffer;
2724 unsigned long isize;
40469ee9 2725 unsigned char actual_sha1[20];
f4913f91 2726
21666f1a 2727 required_type = type_from_string(required_type_name);
e702496e 2728 hashcpy(actual_sha1, sha1);
40469ee9
JH
2729 while (1) {
2730 int ref_length = -1;
2731 const char *ref_type = NULL;
f4913f91 2732
21666f1a 2733 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2734 if (!buffer)
2735 return NULL;
21666f1a 2736 if (type == required_type) {
40469ee9
JH
2737 *size = isize;
2738 if (actual_sha1_return)
e702496e 2739 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2740 return buffer;
2741 }
2742 /* Handle references */
21666f1a 2743 else if (type == OBJ_COMMIT)
40469ee9 2744 ref_type = "tree ";
21666f1a 2745 else if (type == OBJ_TAG)
40469ee9
JH
2746 ref_type = "object ";
2747 else {
2748 free(buffer);
2749 return NULL;
2750 }
2751 ref_length = strlen(ref_type);
f4913f91 2752
50974ec9
MK
2753 if (ref_length + 40 > isize ||
2754 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2755 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2756 free(buffer);
2757 return NULL;
2758 }
1cf58e72 2759 free(buffer);
40469ee9
JH
2760 /* Now we have the ID of the referred-to object in
2761 * actual_sha1. Check again. */
f4913f91 2762 }
f4913f91
JH
2763}
2764
ce9fbf16 2765static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2766 const char *type, unsigned char *sha1,
d65a16f6 2767 char *hdr, int *hdrlen)
d410c0f5 2768{
9126f009 2769 git_SHA_CTX c;
d410c0f5
JH
2770
2771 /* Generate the header */
d65a16f6 2772 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
2773
2774 /* Sha1.. */
9126f009
NP
2775 git_SHA1_Init(&c);
2776 git_SHA1_Update(&c, hdr, *hdrlen);
2777 git_SHA1_Update(&c, buf, len);
2778 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2779}
2780
230f1322 2781/*
5a688fe4
JH
2782 * Move the just written object into its final resting place.
2783 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2784 * "moving" is only a part of what it does, when no patch between
2785 * master to pu changes the call sites of this function.
230f1322 2786 */
839837b9 2787int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322 2788{
e32c0a9c 2789 int ret = 0;
5a688fe4 2790
348df166 2791 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
2792 goto try_rename;
2793 else if (link(tmpfile, filename))
e32c0a9c 2794 ret = errno;
7ebb6fca
LT
2795
2796 /*
2797 * Coda hack - coda doesn't like cross-directory links,
2798 * so we fall back to a rename, which will mean that it
2799 * won't be able to check collisions, but that's not a
2800 * big deal.
2801 *
2802 * The same holds for FAT formatted media.
2803 *
3be1f18e 2804 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
2805 * left to unlink.
2806 */
2807 if (ret && ret != EEXIST) {
be66a6c4 2808 try_rename:
7ebb6fca 2809 if (!rename(tmpfile, filename))
3be1f18e 2810 goto out;
9e48b389 2811 ret = errno;
230f1322 2812 }
691f1a28 2813 unlink_or_warn(tmpfile);
230f1322
LT
2814 if (ret) {
2815 if (ret != EEXIST) {
82247e9b 2816 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
230f1322
LT
2817 }
2818 /* FIXME!!! Collision check here ? */
2819 }
2820
3be1f18e 2821out:
5256b006 2822 if (adjust_shared_perm(filename))
5a688fe4 2823 return error("unable to set permission to '%s'", filename);
230f1322
LT
2824 return 0;
2825}
2826
4d548150
LT
2827static int write_buffer(int fd, const void *buf, size_t len)
2828{
d34cf19b 2829 if (write_in_full(fd, buf, len) < 0)
93822c22 2830 return error("file write error (%s)", strerror(errno));
4d548150
LT
2831 return 0;
2832}
2833
ce9fbf16 2834int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2835 unsigned char *sha1)
2836{
d65a16f6 2837 char hdr[32];
abdc3fc8
RS
2838 int hdrlen;
2839 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2840 return 0;
2841}
2842
e9039dd3
LT
2843/* Finalize a file on disk, and close it. */
2844static void close_sha1_file(int fd)
2845{
aafe9fba
LT
2846 if (fsync_object_files)
2847 fsync_or_die(fd, "sha1 file");
e9039dd3 2848 if (close(fd) != 0)
d824cbba 2849 die_errno("error when closing sha1 file");
e9039dd3
LT
2850}
2851
5723fe7e
LT
2852/* Size of directory component, including the ending '/' */
2853static inline int directory_size(const char *filename)
2854{
2855 const char *s = strrchr(filename, '/');
2856 if (!s)
2857 return 0;
2858 return s - filename + 1;
2859}
2860
2861/*
2862 * This creates a temporary file in the same directory as the final
2863 * 'filename'
2864 *
2865 * We want to avoid cross-directory filename renames, because those
2866 * can have problems on various filesystems (FAT, NFS, Coda).
2867 */
2868static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2869{
2870 int fd, dirlen = directory_size(filename);
2871
2872 if (dirlen + 20 > bufsiz) {
2873 errno = ENAMETOOLONG;
2874 return -1;
2875 }
2876 memcpy(buffer, filename, dirlen);
2877 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
5256b006 2878 fd = git_mkstemp_mode(buffer, 0444);
cbacbf4e 2879 if (fd < 0 && dirlen && errno == ENOENT) {
5723fe7e 2880 /* Make sure the directory exists */
6ff6af62 2881 memcpy(buffer, filename, dirlen);
5723fe7e 2882 buffer[dirlen-1] = 0;
b2476a60
JH
2883 if (mkdir(buffer, 0777) && errno != EEXIST)
2884 return -1;
2885 if (adjust_shared_perm(buffer))
5723fe7e
LT
2886 return -1;
2887
2888 /* Try again */
2889 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
5256b006 2890 fd = git_mkstemp_mode(buffer, 0444);
5723fe7e
LT
2891 }
2892 return fd;
2893}
2894
bbac7311 2895static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 2896 const void *buf, unsigned long len, time_t mtime)
0fcfd160 2897{
915308b1 2898 int fd, ret;
9892beba 2899 unsigned char compressed[4096];
ef49a7a0 2900 git_zstream stream;
748af44c
NP
2901 git_SHA_CTX c;
2902 unsigned char parano_sha1[20];
ab1900a3 2903 static char tmp_file[PATH_MAX];
30d6c6ea 2904 const char *filename = sha1_file_name(sha1);
a44c9a5e 2905
ab1900a3 2906 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
aac17941 2907 if (fd < 0) {
35243577 2908 if (errno == EACCES)
82247e9b 2909 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
916d081b 2910 else
5eaeda70 2911 return error("unable to create temporary file: %s", strerror(errno));
aac17941
LT
2912 }
2913
0fcfd160
LT
2914 /* Set it up */
2915 memset(&stream, 0, sizeof(stream));
55bb5c91 2916 git_deflate_init(&stream, zlib_compression_level);
0fcfd160 2917 stream.next_out = compressed;
9892beba 2918 stream.avail_out = sizeof(compressed);
748af44c 2919 git_SHA1_Init(&c);
a44c9a5e
LT
2920
2921 /* First header.. */
d65a16f6 2922 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2923 stream.avail_in = hdrlen;
55bb5c91
JH
2924 while (git_deflate(&stream, 0) == Z_OK)
2925 ; /* nothing */
748af44c 2926 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
2927
2928 /* Then the data itself.. */
c00e657d 2929 stream.next_in = (void *)buf;
a44c9a5e 2930 stream.avail_in = len;
9892beba 2931 do {
748af44c 2932 unsigned char *in0 = stream.next_in;
55bb5c91 2933 ret = git_deflate(&stream, Z_FINISH);
748af44c 2934 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
2935 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2936 die("unable to write sha1 file");
2937 stream.next_out = compressed;
2938 stream.avail_out = sizeof(compressed);
2939 } while (ret == Z_OK);
2940
ac54c277
LT
2941 if (ret != Z_STREAM_END)
2942 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
55bb5c91 2943 ret = git_deflate_end_gently(&stream);
ac54c277
LT
2944 if (ret != Z_OK)
2945 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
2946 git_SHA1_Final(parano_sha1, &c);
2947 if (hashcmp(sha1, parano_sha1) != 0)
2948 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 2949
e9039dd3 2950 close_sha1_file(fd);
0fcfd160 2951
bbac7311
NP
2952 if (mtime) {
2953 struct utimbuf utb;
2954 utb.actime = mtime;
2955 utb.modtime = mtime;
ab1900a3 2956 if (utime(tmp_file, &utb) < 0)
bbac7311 2957 warning("failed utime() on %s: %s",
ab1900a3 2958 tmp_file, strerror(errno));
bbac7311
NP
2959 }
2960
ab1900a3 2961 return move_temp_to_file(tmp_file, filename);
0fcfd160 2962}
8237b185 2963
c00e657d 2964int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
bbac7311
NP
2965{
2966 unsigned char sha1[20];
2967 char hdr[32];
2968 int hdrlen;
2969
2970 /* Normally if we have it in the pack then we do not bother writing
2971 * it out into .git/objects/??/?{38} file.
2972 */
2973 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2974 if (returnsha1)
2975 hashcpy(returnsha1, sha1);
2976 if (has_sha1_file(sha1))
2977 return 0;
2978 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2979}
2980
2981int force_object_loose(const unsigned char *sha1, time_t mtime)
2982{
bbac7311
NP
2983 void *buf;
2984 unsigned long len;
2985 enum object_type type;
2986 char hdr[32];
2987 int hdrlen;
1fb23e65 2988 int ret;
bbac7311 2989
c529d75a 2990 if (has_loose_object(sha1))
bbac7311
NP
2991 return 0;
2992 buf = read_packed_sha1(sha1, &type, &len);
2993 if (!buf)
2994 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2995 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1fb23e65
BS
2996 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2997 free(buf);
2998
2999 return ret;
bbac7311
NP
3000}
3001
bf592c50
DB
3002int has_pack_index(const unsigned char *sha1)
3003{
3004 struct stat st;
3005 if (stat(sha1_pack_index_name(sha1), &st))
3006 return 0;
3007 return 1;
bf592c50
DB
3008}
3009
cd673c1f
JH
3010int has_sha1_pack(const unsigned char *sha1)
3011{
3012 struct pack_entry e;
3013 return find_pack_entry(sha1, &e);
3014}
3015
8237b185
DB
3016int has_sha1_file(const unsigned char *sha1)
3017{
1f688557
JH
3018 struct pack_entry e;
3019
cd673c1f 3020 if (find_pack_entry(sha1, &e))
1f688557 3021 return 1;
45e8a748
JK
3022 if (has_loose_object(sha1))
3023 return 1;
3024 reprepare_packed_git();
3025 return find_pack_entry(sha1, &e);
8237b185 3026}
74400e71 3027
c879daa2
NTND
3028static void check_tree(const void *buf, size_t size)
3029{
3030 struct tree_desc desc;
3031 struct name_entry entry;
3032
3033 init_tree_desc(&desc, buf, size);
3034 while (tree_entry(&desc, &entry))
3035 /* do nothing
3036 * tree_entry() will die() on malformed entries */
3037 ;
3038}
3039
3040static void check_commit(const void *buf, size_t size)
3041{
3042 struct commit c;
3043 memset(&c, 0, sizeof(c));
3044 if (parse_commit_buffer(&c, buf, size))
3045 die("corrupt commit");
3046}
3047
3048static void check_tag(const void *buf, size_t size)
3049{
3050 struct tag t;
3051 memset(&t, 0, sizeof(t));
3052 if (parse_tag_buffer(&t, buf, size))
3053 die("corrupt tag");
3054}
3055
43df4f86 3056static int index_mem(unsigned char *sha1, void *buf, size_t size,
c4ce46fc
JH
3057 enum object_type type,
3058 const char *path, unsigned flags)
e7332f96 3059{
6c510bee 3060 int ret, re_allocated = 0;
c4ce46fc 3061 int write_object = flags & HASH_WRITE_OBJECT;
74400e71 3062
7672db20 3063 if (!type)
edaec3fb 3064 type = OBJ_BLOB;
6c510bee
LT
3065
3066 /*
3067 * Convert blobs to git internal format
3068 */
43df4f86 3069 if ((type == OBJ_BLOB) && path) {
f285a2d7 3070 struct strbuf nbuf = STRBUF_INIT;
21e5ad50 3071 if (convert_to_git(path, buf, size, &nbuf,
5e12e78e 3072 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
b315c5c0 3073 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
3074 re_allocated = 1;
3075 }
3076 }
c4ce46fc 3077 if (flags & HASH_FORMAT_CHECK) {
c879daa2
NTND
3078 if (type == OBJ_TREE)
3079 check_tree(buf, size);
3080 if (type == OBJ_COMMIT)
3081 check_commit(buf, size);
3082 if (type == OBJ_TAG)
3083 check_tag(buf, size);
3084 }
6c510bee 3085
7672db20 3086 if (write_object)
edaec3fb 3087 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 3088 else
edaec3fb 3089 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 3090 if (re_allocated)
6c510bee 3091 free(buf);
43df4f86
DP
3092 return ret;
3093}
3094
9035d75a
SP
3095static int index_stream_convert_blob(unsigned char *sha1, int fd,
3096 const char *path, unsigned flags)
3097{
3098 int ret;
3099 const int write_object = flags & HASH_WRITE_OBJECT;
3100 struct strbuf sbuf = STRBUF_INIT;
3101
3102 assert(path);
3103 assert(would_convert_to_git_filter_fd(path));
3104
3105 convert_to_git_filter_fd(path, fd, &sbuf,
3106 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3107
3108 if (write_object)
3109 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3110 sha1);
3111 else
3112 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3113 sha1);
3114 strbuf_release(&sbuf);
3115 return ret;
3116}
3117
7b41e1e1
JH
3118static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3119 const char *path, unsigned flags)
3120{
3121 struct strbuf sbuf = STRBUF_INIT;
3122 int ret;
3123
3124 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3125 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3126 else
3127 ret = -1;
3128 strbuf_release(&sbuf);
3129 return ret;
3130}
3131
ea68b0ce
DP
3132#define SMALL_FILE_SIZE (32*1024)
3133
7b41e1e1
JH
3134static int index_core(unsigned char *sha1, int fd, size_t size,
3135 enum object_type type, const char *path,
3136 unsigned flags)
43df4f86
DP
3137{
3138 int ret;
43df4f86 3139
7b41e1e1 3140 if (!size) {
c4ce46fc 3141 ret = index_mem(sha1, NULL, size, type, path, flags);
ea68b0ce
DP
3142 } else if (size <= SMALL_FILE_SIZE) {
3143 char *buf = xmalloc(size);
3144 if (size == read_in_full(fd, buf, size))
c4ce46fc 3145 ret = index_mem(sha1, buf, size, type, path, flags);
ea68b0ce
DP
3146 else
3147 ret = error("short read %s", strerror(errno));
3148 free(buf);
08bda208 3149 } else {
43df4f86 3150 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
c4ce46fc 3151 ret = index_mem(sha1, buf, size, type, path, flags);
aac17941 3152 munmap(buf, size);
08bda208 3153 }
7b41e1e1
JH
3154 return ret;
3155}
3156
4dd1fbc7 3157/*
568508e7
JH
3158 * This creates one packfile per large blob unless bulk-checkin
3159 * machinery is "plugged".
4dd1fbc7
JH
3160 *
3161 * This also bypasses the usual "convert-to-git" dance, and that is on
3162 * purpose. We could write a streaming version of the converting
3163 * functions and insert that before feeding the data to fast-import
4f22b101
JK
3164 * (or equivalent in-core API described above). However, that is
3165 * somewhat complicated, as we do not know the size of the filter
3166 * result, which we need to know beforehand when writing a git object.
3167 * Since the primary motivation for trying to stream from the working
3168 * tree file and to avoid mmaping it in core is to deal with large
3169 * binary blobs, they generally do not want to get any conversion, and
3170 * callers should avoid this code path when filters are requested.
4dd1fbc7
JH
3171 */
3172static int index_stream(unsigned char *sha1, int fd, size_t size,
3173 enum object_type type, const char *path,
3174 unsigned flags)
3175{
568508e7 3176 return index_bulk_checkin(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
3177}
3178
7b41e1e1
JH
3179int index_fd(unsigned char *sha1, int fd, struct stat *st,
3180 enum object_type type, const char *path, unsigned flags)
3181{
3182 int ret;
7b41e1e1 3183
9079ab7c
SP
3184 /*
3185 * Call xsize_t() only when needed to avoid potentially unnecessary
3186 * die() for large files.
3187 */
9035d75a
SP
3188 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3189 ret = index_stream_convert_blob(sha1, fd, path, flags);
3190 else if (!S_ISREG(st->st_mode))
7b41e1e1 3191 ret = index_pipe(sha1, fd, type, path, flags);
9079ab7c 3192 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
7ce7c760 3193 (path && would_convert_to_git(path)))
9079ab7c
SP
3194 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3195 flags);
4dd1fbc7 3196 else
9079ab7c
SP
3197 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3198 flags);
43df4f86 3199 close(fd);
aac17941 3200 return ret;
74400e71 3201}
ec1fcc16 3202
c4ce46fc 3203int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
ec1fcc16
JH
3204{
3205 int fd;
b760d3aa 3206 struct strbuf sb = STRBUF_INIT;
ec1fcc16
JH
3207
3208 switch (st->st_mode & S_IFMT) {
3209 case S_IFREG:
3210 fd = open(path, O_RDONLY);
3211 if (fd < 0)
3212 return error("open(\"%s\"): %s", path,
3213 strerror(errno));
c4ce46fc 3214 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
ec1fcc16
JH
3215 return error("%s: failed to insert into database",
3216 path);
3217 break;
3218 case S_IFLNK:
b760d3aa 3219 if (strbuf_readlink(&sb, path, st->st_size)) {
ec1fcc16 3220 char *errstr = strerror(errno);
ec1fcc16
JH
3221 return error("readlink(\"%s\"): %s", path,
3222 errstr);
3223 }
c4ce46fc 3224 if (!(flags & HASH_WRITE_OBJECT))
b760d3aa
LT
3225 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3226 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
ec1fcc16
JH
3227 return error("%s: failed to insert into database",
3228 path);
b760d3aa 3229 strbuf_release(&sb);
ec1fcc16 3230 break;
f35a6d3b
LT
3231 case S_IFDIR:
3232 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
3233 default:
3234 return error("%s: unsupported file type", path);
3235 }
3236 return 0;
3237}
a69e5429
JH
3238
3239int read_pack_header(int fd, struct pack_header *header)
3240{
c697ad14
HO
3241 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3242 /* "eof before pack header was fully read" */
3243 return PH_ERROR_EOF;
3244
a69e5429
JH
3245 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3246 /* "protocol error (pack signature mismatch detected)" */
3247 return PH_ERROR_PACK_SIGNATURE;
3248 if (!pack_version_ok(header->hdr_version))
3249 /* "protocol error (pack version unsupported)" */
3250 return PH_ERROR_PROTOCOL;
3251 return 0;
3252}
40d52ff7
JK
3253
3254void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3255{
3256 enum object_type type = sha1_object_info(sha1, NULL);
3257 if (type < 0)
3258 die("%s is not a valid object", sha1_to_hex(sha1));
3259 if (type != expect)
3260 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3261 typename(expect));
3262}