add_packed_git: convert strcpy into xsnprintf
[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"
697cc8ef 11#include "lockfile.h"
1f688557 12#include "delta.h"
a733cb60 13#include "pack.h"
8e440259
PE
14#include "blob.h"
15#include "commit.h"
4dd1fbc7 16#include "run-command.h"
8e440259
PE
17#include "tag.h"
18#include "tree.h"
c879daa2 19#include "tree-walk.h"
f35a6d3b 20#include "refs.h"
70f5d5d3 21#include "pack-revindex.h"
628522ec 22#include "sha1-lookup.h"
568508e7 23#include "bulk-checkin.h"
090ea126 24#include "streaming.h"
543c5caa 25#include "dir.h"
0fcfd160 26
144bde78
LT
27#ifndef O_NOATIME
28#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
29#define O_NOATIME 01000000
30#else
31#define O_NOATIME 0
32#endif
33#endif
34
28bd70d8
JN
35#define SZ_FMT PRIuMAX
36static inline uintmax_t sz_fmt(size_t s) { return s; }
e05db0fd 37
96f1e58f 38const unsigned char null_sha1[20];
88cd621d 39
c597ba80
NTND
40/*
41 * This is meant to hold a *small* number of objects that you would
42 * want read_sha1_file() to be able to return, but yet you do not want
43 * to write them into the object store (e.g. a browse-only
44 * application).
45 */
46static struct cached_object {
47 unsigned char sha1[20];
48 enum object_type type;
49 void *buf;
50 unsigned long size;
51} *cached_objects;
52static int cached_object_nr, cached_object_alloc;
53
54static struct cached_object empty_tree = {
dab0d410 55 EMPTY_TREE_SHA1_BIN_LITERAL,
c597ba80
NTND
56 OBJ_TREE,
57 "",
58 0
59};
60
1b1005d1
MH
61/*
62 * A pointer to the last packed_git in which an object was found.
63 * When an object is sought, we look in this packfile first, because
64 * objects that are looked up at similar times are often in the same
65 * packfile as one another.
66 */
c01f51cc
NTND
67static struct packed_git *last_found_pack;
68
c597ba80
NTND
69static struct cached_object *find_cached_object(const unsigned char *sha1)
70{
71 int i;
72 struct cached_object *co = cached_objects;
73
74 for (i = 0; i < cached_object_nr; i++, co++) {
75 if (!hashcmp(co->sha1, sha1))
76 return co;
77 }
78 if (!hashcmp(sha1, empty_tree.sha1))
79 return &empty_tree;
80 return NULL;
81}
82
90a6464b
JH
83int mkdir_in_gitdir(const char *path)
84{
85 if (mkdir(path, 0777)) {
86 int saved_errno = errno;
87 struct stat st;
88 struct strbuf sb = STRBUF_INIT;
89
90 if (errno != EEXIST)
91 return -1;
92 /*
93 * Are we looking at a path in a symlinked worktree
94 * whose original repository does not yet have it?
95 * e.g. .git/rr-cache pointing at its original
96 * repository in which the user hasn't performed any
97 * conflict resolution yet?
98 */
99 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
100 strbuf_readlink(&sb, path, st.st_size) ||
101 !is_absolute_path(sb.buf) ||
102 mkdir(sb.buf, 0777)) {
103 strbuf_release(&sb);
104 errno = saved_errno;
105 return -1;
106 }
107 strbuf_release(&sb);
108 }
109 return adjust_shared_perm(path);
110}
111
0be0521b 112enum scld_error safe_create_leading_directories(char *path)
b2cb9425 113{
26c8ae2a 114 char *next_component = path + offset_1st_component(path);
0be0521b 115 enum scld_error ret = SCLD_OK;
67d42212 116
0be0521b 117 while (ret == SCLD_OK && next_component) {
f0502332 118 struct stat st;
0f527403 119 char *slash = next_component, slash_character;
f0502332 120
0f527403
MH
121 while (*slash && !is_dir_sep(*slash))
122 slash++;
123
124 if (!*slash)
b2cb9425 125 break;
bf10cf70 126
26c8ae2a 127 next_component = slash + 1;
0f527403 128 while (is_dir_sep(*next_component))
bf10cf70 129 next_component++;
26c8ae2a 130 if (!*next_component)
5f0bdf50 131 break;
831651fd 132
0f527403 133 slash_character = *slash;
831651fd 134 *slash = '\0';
67d42212
JR
135 if (!stat(path, &st)) {
136 /* path exists */
9e6f885d 137 if (!S_ISDIR(st.st_mode))
0be0521b 138 ret = SCLD_EXISTS;
53a39721 139 } else if (mkdir(path, 0777)) {
928734d9 140 if (errno == EEXIST &&
9e6f885d 141 !stat(path, &st) && S_ISDIR(st.st_mode))
928734d9 142 ; /* somebody created it since we checked */
18d37e86
MH
143 else if (errno == ENOENT)
144 /*
145 * Either mkdir() failed because
146 * somebody just pruned the containing
147 * directory, or stat() failed because
148 * the file that was in our way was
149 * just removed. Either way, inform
150 * the caller that it might be worth
151 * trying again:
152 */
153 ret = SCLD_VANISHED;
9e6f885d 154 else
0be0521b 155 ret = SCLD_FAILED;
53a39721 156 } else if (adjust_shared_perm(path)) {
0be0521b 157 ret = SCLD_PERMS;
457f06d6 158 }
0f527403 159 *slash = slash_character;
b2cb9425 160 }
9e6f885d 161 return ret;
b2cb9425 162}
723c31fe 163
0be0521b 164enum scld_error safe_create_leading_directories_const(const char *path)
8e21d63b
JK
165{
166 /* path points to cache entries, so xstrdup before messing with it */
167 char *buf = xstrdup(path);
0be0521b 168 enum scld_error result = safe_create_leading_directories(buf);
8e21d63b
JK
169 free(buf);
170 return result;
171}
172
ace1534d
JH
173static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
174{
175 int i;
176 for (i = 0; i < 20; i++) {
177 static char hex[] = "0123456789abcdef";
178 unsigned int val = sha1[i];
179 char *pos = pathbuf + i*2 + (i > 0);
180 *pos++ = hex[val >> 4];
181 *pos = hex[val & 0xf];
182 }
183}
184
30d6c6ea 185const char *sha1_file_name(const unsigned char *sha1)
0fcfd160 186{
560fb6a1
JK
187 static char buf[PATH_MAX];
188 const char *objdir;
189 int len;
0fcfd160 190
560fb6a1
JK
191 objdir = get_object_directory();
192 len = strlen(objdir);
193
194 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
195 if (len + 43 > PATH_MAX)
196 die("insanely long object directory %s", objdir);
197 memcpy(buf, objdir, len);
198 buf[len] = '/';
199 buf[len+3] = '/';
200 buf[len+42] = '\0';
201 fill_sha1_path(buf + len + 1, sha1);
202 return buf;
0fcfd160
LT
203}
204
d40d535b
MH
205/*
206 * Return the name of the pack or index file with the specified sha1
207 * in its filename. *base and *name are scratch space that must be
208 * provided by the caller. which should be "pack" or "idx".
209 */
633f43e1 210static char *sha1_get_pack_name(const unsigned char *sha1,
6eec46bd 211 char **name, char **base, const char *which)
bf592c50
DB
212{
213 static const char hex[] = "0123456789abcdef";
633f43e1 214 char *buf;
bf592c50
DB
215 int i;
216
633f43e1 217 if (!*base) {
bf592c50
DB
218 const char *sha1_file_directory = get_object_directory();
219 int len = strlen(sha1_file_directory);
633f43e1 220 *base = xmalloc(len + 60);
6eec46bd
JH
221 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
222 sha1_file_directory, which);
633f43e1 223 *name = *base + len + 11;
bf592c50
DB
224 }
225
633f43e1 226 buf = *name;
bf592c50
DB
227
228 for (i = 0; i < 20; i++) {
229 unsigned int val = *sha1++;
230 *buf++ = hex[val >> 4];
231 *buf++ = hex[val & 0xf];
232 }
a6080a0a 233
633f43e1 234 return *base;
bf592c50
DB
235}
236
633f43e1 237char *sha1_pack_name(const unsigned char *sha1)
bf592c50 238{
633f43e1 239 static char *name, *base;
bf592c50 240
6eec46bd 241 return sha1_get_pack_name(sha1, &name, &base, "pack");
633f43e1 242}
bf592c50 243
633f43e1
HO
244char *sha1_pack_index_name(const unsigned char *sha1)
245{
246 static char *name, *base;
a6080a0a 247
6eec46bd 248 return sha1_get_pack_name(sha1, &name, &base, "idx");
bf592c50
DB
249}
250
d5a63b99
JH
251struct alternate_object_database *alt_odb_list;
252static struct alternate_object_database **alt_odb_tail;
ace1534d 253
ddd5d056
JH
254/*
255 * Prepare alternate object database registry.
d5a63b99
JH
256 *
257 * The variable alt_odb_list points at the list of struct
258 * alternate_object_database. The elements on this list come from
259 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
260 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
261 * whose contents is similar to that environment variable but can be
262 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
263 * contains "/the/directory/corresponding/to/.git/objects/...", while
264 * its name points just after the slash at the end of ".git/objects/"
265 * in the example above, and has enough space to hold 40-byte hex
266 * SHA1, an extra slash for the first level indirection, and the
267 * terminating NUL.
ddd5d056 268 */
539e7506
EK
269static int link_alt_odb_entry(const char *entry, const char *relative_base,
270 int depth, const char *normalized_objdir)
ace1534d 271{
c2f493a4
MW
272 struct alternate_object_database *ent;
273 struct alternate_object_database *alt;
5bdf0a84
HW
274 int pfxlen, entlen;
275 struct strbuf pathbuf = STRBUF_INIT;
d5a63b99 276
85dadc38 277 if (!is_absolute_path(entry) && relative_base) {
5bdf0a84
HW
278 strbuf_addstr(&pathbuf, real_path(relative_base));
279 strbuf_addch(&pathbuf, '/');
c2f493a4 280 }
6eac50d8 281 strbuf_addstr(&pathbuf, entry);
c2f493a4 282
5bdf0a84
HW
283 normalize_path_copy(pathbuf.buf, pathbuf.buf);
284
285 pfxlen = strlen(pathbuf.buf);
286
287 /*
288 * The trailing slash after the directory name is given by
289 * this function at the end. Remove duplicates.
290 */
291 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
292 pfxlen -= 1;
293
294 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
295 ent = xmalloc(sizeof(*ent) + entlen);
296 memcpy(ent->base, pathbuf.buf, pfxlen);
297 strbuf_release(&pathbuf);
c2f493a4
MW
298
299 ent->name = ent->base + pfxlen + 1;
300 ent->base[pfxlen + 3] = '/';
301 ent->base[pfxlen] = ent->base[entlen-1] = 0;
302
303 /* Detect cases where alternate disappeared */
90b4a71c 304 if (!is_directory(ent->base)) {
c2f493a4
MW
305 error("object directory %s does not exist; "
306 "check .git/objects/info/alternates.",
307 ent->base);
308 free(ent);
309 return -1;
310 }
311
312 /* Prevent the common mistake of listing the same
313 * thing twice, or object directory itself.
314 */
315 for (alt = alt_odb_list; alt; alt = alt->next) {
80b47854
RS
316 if (pfxlen == alt->name - alt->base - 1 &&
317 !memcmp(ent->base, alt->base, pfxlen)) {
c2f493a4
MW
318 free(ent);
319 return -1;
320 }
321 }
539e7506 322 if (!strcmp_icase(ent->base, normalized_objdir)) {
c2f493a4
MW
323 free(ent);
324 return -1;
325 }
326
327 /* add the alternate entry */
328 *alt_odb_tail = ent;
329 alt_odb_tail = &(ent->next);
330 ent->next = NULL;
331
332 /* recursively add alternates */
333 read_info_alternates(ent->base, depth + 1);
334
335 ent->base[pfxlen] = '/';
336
337 return 0;
338}
339
c5950164 340static void link_alt_odb_entries(const char *alt, int len, int sep,
c2f493a4
MW
341 const char *relative_base, int depth)
342{
6eac50d8
MH
343 struct string_list entries = STRING_LIST_INIT_NODUP;
344 char *alt_copy;
345 int i;
539e7506 346 struct strbuf objdirbuf = STRBUF_INIT;
c2f493a4
MW
347
348 if (depth > 5) {
349 error("%s: ignoring alternate object stores, nesting too deep.",
350 relative_base);
351 return;
352 }
353
f655651e 354 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
539e7506
EK
355 normalize_path_copy(objdirbuf.buf, objdirbuf.buf);
356
c5950164 357 alt_copy = xmemdupz(alt, len);
6eac50d8
MH
358 string_list_split_in_place(&entries, alt_copy, sep, -1);
359 for (i = 0; i < entries.nr; i++) {
360 const char *entry = entries.items[i].string;
361 if (entry[0] == '\0' || entry[0] == '#')
9577e7e3 362 continue;
6eac50d8
MH
363 if (!is_absolute_path(entry) && depth) {
364 error("%s: ignoring relative alternate object store %s",
365 relative_base, entry);
366 } else {
539e7506 367 link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
9577e7e3 368 }
9577e7e3 369 }
6eac50d8
MH
370 string_list_clear(&entries, 0);
371 free(alt_copy);
539e7506 372 strbuf_release(&objdirbuf);
d5a63b99
JH
373}
374
5e73633d 375void read_info_alternates(const char * relative_base, int depth)
d5a63b99 376{
9577e7e3 377 char *map;
dc49cd76 378 size_t mapsz;
d5a63b99 379 struct stat st;
5015f01c 380 char *path;
c2f493a4 381 int fd;
d5a63b99 382
5015f01c 383 path = xstrfmt("%s/info/alternates", relative_base);
f4e51683 384 fd = git_open_noatime(path);
5015f01c 385 free(path);
d5a63b99
JH
386 if (fd < 0)
387 return;
388 if (fstat(fd, &st) || (st.st_size == 0)) {
389 close(fd);
9a217f2a 390 return;
ace1534d 391 }
dc49cd76
SP
392 mapsz = xsize_t(st.st_size);
393 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 394 close(fd);
d5a63b99 395
c5950164 396 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
c2f493a4 397
dc49cd76 398 munmap(map, mapsz);
ace1534d
JH
399}
400
bef70b22
DB
401void add_to_alternates_file(const char *reference)
402{
403 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
77b9b1d1
JK
404 char *alts = git_pathdup("objects/info/alternates");
405 FILE *in, *out;
406
407 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
408 out = fdopen_lock_file(lock, "w");
409 if (!out)
410 die_errno("unable to fdopen alternates lockfile");
411
412 in = fopen(alts, "r");
413 if (in) {
414 struct strbuf line = STRBUF_INIT;
415 int found = 0;
416
417 while (strbuf_getline(&line, in, '\n') != EOF) {
418 if (!strcmp(reference, line.buf)) {
419 found = 1;
420 break;
421 }
422 fprintf_or_die(out, "%s\n", line.buf);
423 }
424
425 strbuf_release(&line);
426 fclose(in);
427
428 if (found) {
429 rollback_lock_file(lock);
430 lock = NULL;
431 }
432 }
433 else if (errno != ENOENT)
434 die_errno("unable to read alternates file");
435
436 if (lock) {
437 fprintf_or_die(out, "%s\n", reference);
438 if (commit_lock_file(lock))
439 die_errno("unable to move new alternates file into place");
440 if (alt_odb_tail)
441 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
442 }
443 free(alts);
bef70b22
DB
444}
445
fe1b2268 446int foreach_alt_odb(alt_odb_fn fn, void *cb)
d79796bc
JH
447{
448 struct alternate_object_database *ent;
fe1b2268 449 int r = 0;
d79796bc
JH
450
451 prepare_alt_odb();
fe1b2268
JK
452 for (ent = alt_odb_list; ent; ent = ent->next) {
453 r = fn(ent, cb);
454 if (r)
455 break;
456 }
457 return r;
d79796bc
JH
458}
459
c2f493a4
MW
460void prepare_alt_odb(void)
461{
554fe20d 462 const char *alt;
c2f493a4 463
7dc24aa5
SP
464 if (alt_odb_tail)
465 return;
466
c2f493a4
MW
467 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
468 if (!alt) alt = "";
469
c2f493a4 470 alt_odb_tail = &alt_odb_list;
c5950164 471 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
472
473 read_info_alternates(get_object_directory(), 0);
474}
475
3096b2ec 476/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
33d4221c 477static int freshen_file(const char *fn)
ace1534d 478{
33d4221c
JK
479 struct utimbuf t;
480 t.actime = t.modtime = time(NULL);
481 return !utime(fn, &t);
0f4dc14a 482}
ace1534d 483
3096b2ec
JK
484/*
485 * All of the check_and_freshen functions return 1 if the file exists and was
486 * freshened (if freshening was requested), 0 otherwise. If they return
487 * 0, you should not assume that it is safe to skip a write of the object (it
488 * either does not exist on disk, or has a stale mtime and may be subject to
489 * pruning).
490 */
33d4221c
JK
491static int check_and_freshen_file(const char *fn, int freshen)
492{
493 if (access(fn, F_OK))
494 return 0;
3096b2ec 495 if (freshen && !freshen_file(fn))
33d4221c
JK
496 return 0;
497 return 1;
498}
499
500static int check_and_freshen_local(const unsigned char *sha1, int freshen)
501{
502 return check_and_freshen_file(sha1_file_name(sha1), freshen);
503}
504
505static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
0f4dc14a
BC
506{
507 struct alternate_object_database *alt;
9a217f2a 508 prepare_alt_odb();
d5a63b99 509 for (alt = alt_odb_list; alt; alt = alt->next) {
0f4dc14a 510 fill_sha1_path(alt->name, sha1);
33d4221c 511 if (check_and_freshen_file(alt->base, freshen))
c529d75a 512 return 1;
ace1534d 513 }
c529d75a 514 return 0;
ace1534d
JH
515}
516
33d4221c
JK
517static int check_and_freshen(const unsigned char *sha1, int freshen)
518{
519 return check_and_freshen_local(sha1, freshen) ||
520 check_and_freshen_nonlocal(sha1, freshen);
521}
522
523int has_loose_object_nonlocal(const unsigned char *sha1)
524{
525 return check_and_freshen_nonlocal(sha1, 0);
526}
527
0f4dc14a
BC
528static int has_loose_object(const unsigned char *sha1)
529{
33d4221c 530 return check_and_freshen(sha1, 0);
0f4dc14a
BC
531}
532
60bb8b14 533static unsigned int pack_used_ctr;
a53128b6
SP
534static unsigned int pack_mmap_calls;
535static unsigned int peak_pack_open_windows;
536static unsigned int pack_open_windows;
c7934306
SP
537static unsigned int pack_open_fds;
538static unsigned int pack_max_fds;
a53128b6 539static size_t peak_pack_mapped;
60bb8b14 540static size_t pack_mapped;
9a217f2a 541struct packed_git *packed_git;
1f688557 542
b79d18c9 543void pack_report(void)
a53128b6
SP
544{
545 fprintf(stderr,
e05db0fd
PR
546 "pack_report: getpagesize() = %10" SZ_FMT "\n"
547 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
548 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
549 sz_fmt(getpagesize()),
550 sz_fmt(packed_git_window_size),
551 sz_fmt(packed_git_limit));
a53128b6
SP
552 fprintf(stderr,
553 "pack_report: pack_used_ctr = %10u\n"
554 "pack_report: pack_mmap_calls = %10u\n"
555 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
556 "pack_report: pack_mapped = "
557 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
558 pack_used_ctr,
559 pack_mmap_calls,
560 pack_open_windows, peak_pack_open_windows,
9e42d6a1 561 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
562}
563
d40d535b
MH
564/*
565 * Open and mmap the index file at path, perform a couple of
566 * consistency checks, then record its information to p. Return 0 on
567 * success.
568 */
569static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
570{
571 void *idx_map;
42873078 572 struct pack_idx_header *hdr;
dc49cd76 573 size_t idx_size;
74e34e1f 574 uint32_t version, nr, i, *index;
f4e51683 575 int fd = git_open_noatime(path);
1f688557 576 struct stat st;
42873078 577
1f688557
JH
578 if (fd < 0)
579 return -1;
580 if (fstat(fd, &st)) {
581 close(fd);
582 return -1;
583 }
dc49cd76 584 idx_size = xsize_t(st.st_size);
2d88451b
SP
585 if (idx_size < 4 * 256 + 20 + 20) {
586 close(fd);
587 return error("index file %s is too small", path);
588 }
c4712e45 589 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 590 close(fd);
1f688557 591
42873078
NP
592 hdr = idx_map;
593 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
594 version = ntohl(hdr->idx_version);
595 if (version < 2 || version > 2) {
596 munmap(idx_map, idx_size);
6e1c2344 597 return error("index file %s is version %"PRIu32
74e34e1f
NP
598 " and is not supported by this binary"
599 " (try upgrading GIT to a newer version)",
600 path, version);
601 }
602 } else
603 version = 1;
df1b059d 604
1f688557 605 nr = 0;
42873078 606 index = idx_map;
74e34e1f
NP
607 if (version > 1)
608 index += 2; /* skip index header */
1f688557 609 for (i = 0; i < 256; i++) {
326bf396 610 uint32_t n = ntohl(index[i]);
2d88451b
SP
611 if (n < nr) {
612 munmap(idx_map, idx_size);
df1b059d 613 return error("non-monotonic index %s", path);
2d88451b 614 }
1f688557
JH
615 nr = n;
616 }
617
74e34e1f
NP
618 if (version == 1) {
619 /*
620 * Total size:
621 * - 256 index entries 4 bytes each
622 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
623 * - 20-byte SHA1 of the packfile
624 * - 20-byte SHA1 file checksum
625 */
626 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
627 munmap(idx_map, idx_size);
eef427a0 628 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
629 }
630 } else if (version == 2) {
631 /*
632 * Minimum size:
633 * - 8 bytes of header
634 * - 256 index entries 4 bytes each
635 * - 20-byte sha1 entry * nr
636 * - 4-byte crc entry * nr
637 * - 4-byte offset entry * nr
638 * - 20-byte SHA1 of the packfile
639 * - 20-byte SHA1 file checksum
640 * And after the 4-byte offset table might be a
641 * variable sized table containing 8-byte entries
642 * for offsets larger than 2^31.
643 */
644 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
645 unsigned long max_size = min_size;
646 if (nr)
647 max_size += (nr - 1)*8;
648 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 649 munmap(idx_map, idx_size);
eef427a0 650 return error("wrong index v2 file size in %s", path);
74e34e1f 651 }
7109c889
JH
652 if (idx_size != min_size &&
653 /*
654 * make sure we can deal with large pack offsets.
655 * 31-bit signed offset won't be enough, neither
656 * 32-bit unsigned one will be.
657 */
658 (sizeof(off_t) <= 4)) {
659 munmap(idx_map, idx_size);
660 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 661 }
2d88451b 662 }
1f688557 663
74e34e1f 664 p->index_version = version;
42873078
NP
665 p->index_data = idx_map;
666 p->index_size = idx_size;
57059091 667 p->num_objects = nr;
1f688557
JH
668 return 0;
669}
670
bc8e478a 671int open_pack_index(struct packed_git *p)
d079837e
SP
672{
673 char *idx_name;
674 int ret;
675
676 if (p->index_data)
677 return 0;
678
679 idx_name = xstrdup(p->pack_name);
680 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
681 ret = check_packed_git_idx(idx_name, p);
682 free(idx_name);
683 return ret;
684}
685
11daf39b
SP
686static void scan_windows(struct packed_git *p,
687 struct packed_git **lru_p,
688 struct pack_window **lru_w,
689 struct pack_window **lru_l)
1f688557 690{
11daf39b
SP
691 struct pack_window *w, *w_l;
692
693 for (w_l = NULL, w = p->windows; w; w = w->next) {
694 if (!w->inuse_cnt) {
695 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
696 *lru_p = p;
697 *lru_w = w;
698 *lru_l = w_l;
54044bf8 699 }
54044bf8 700 }
11daf39b 701 w_l = w;
f9253394 702 }
11daf39b
SP
703}
704
7c3ecb32 705static int unuse_one_window(struct packed_git *current)
11daf39b
SP
706{
707 struct packed_git *p, *lru_p = NULL;
708 struct pack_window *lru_w = NULL, *lru_l = NULL;
709
710 if (current)
711 scan_windows(current, &lru_p, &lru_w, &lru_l);
712 for (p = packed_git; p; p = p->next)
713 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
714 if (lru_p) {
715 munmap(lru_w->base, lru_w->len);
716 pack_mapped -= lru_w->len;
717 if (lru_l)
718 lru_l->next = lru_w->next;
7c3ecb32 719 else
54044bf8 720 lru_p->windows = lru_w->next;
54044bf8 721 free(lru_w);
a53128b6 722 pack_open_windows--;
54044bf8
SP
723 return 1;
724 }
725 return 0;
f9253394
JH
726}
727
7c3ecb32 728void release_pack_memory(size_t need)
97bfeb34
SP
729{
730 size_t cur = pack_mapped;
7c3ecb32 731 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
97bfeb34
SP
732 ; /* nothing */
733}
734
02710228
SP
735static void mmap_limit_check(size_t length)
736{
737 static size_t limit = 0;
738 if (!limit) {
739 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
740 if (!limit)
741 limit = SIZE_MAX;
742 }
743 if (length > limit)
744 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
745 (uintmax_t)length, (uintmax_t)limit);
746}
747
1570856b
JK
748void *xmmap_gently(void *start, size_t length,
749 int prot, int flags, int fd, off_t offset)
58ecbd5e 750{
02710228
SP
751 void *ret;
752
753 mmap_limit_check(length);
754 ret = mmap(start, length, prot, flags, fd, offset);
58ecbd5e
JN
755 if (ret == MAP_FAILED) {
756 if (!length)
757 return NULL;
7c3ecb32 758 release_pack_memory(length);
58ecbd5e 759 ret = mmap(start, length, prot, flags, fd, offset);
58ecbd5e
JN
760 }
761 return ret;
762}
763
1570856b
JK
764void *xmmap(void *start, size_t length,
765 int prot, int flags, int fd, off_t offset)
766{
767 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
768 if (ret == MAP_FAILED)
9ca0aaf6 769 die_errno("mmap failed");
1570856b
JK
770 return ret;
771}
772
c9ced051
SP
773void close_pack_windows(struct packed_git *p)
774{
775 while (p->windows) {
776 struct pack_window *w = p->windows;
777
778 if (w->inuse_cnt)
779 die("pack '%s' still has open windows to it",
780 p->pack_name);
781 munmap(w->base, w->len);
782 pack_mapped -= w->len;
783 pack_open_windows--;
784 p->windows = w->next;
785 free(w);
786 }
787}
788
88d0db55
BC
789/*
790 * The LRU pack is the one with the oldest MRU window, preferring packs
791 * with no used windows, or the oldest mtime if it has no windows allocated.
792 */
793static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
794{
795 struct pack_window *w, *this_mru_w;
796 int has_windows_inuse = 0;
797
798 /*
799 * Reject this pack if it has windows and the previously selected
800 * one does not. If this pack does not have windows, reject
801 * it if the pack file is newer than the previously selected one.
802 */
803 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
804 return;
805
806 for (w = this_mru_w = p->windows; w; w = w->next) {
807 /*
808 * Reject this pack if any of its windows are in use,
809 * but the previously selected pack did not have any
810 * inuse windows. Otherwise, record that this pack
811 * has windows in use.
812 */
813 if (w->inuse_cnt) {
814 if (*accept_windows_inuse)
815 has_windows_inuse = 1;
816 else
817 return;
818 }
819
820 if (w->last_used > this_mru_w->last_used)
821 this_mru_w = w;
822
823 /*
824 * Reject this pack if it has windows that have been
825 * used more recently than the previously selected pack.
826 * If the previously selected pack had windows inuse and
827 * we have not encountered a window in this pack that is
828 * inuse, skip this check since we prefer a pack with no
829 * inuse windows to one that has inuse windows.
830 */
831 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
832 this_mru_w->last_used > (*mru_w)->last_used)
833 return;
834 }
835
836 /*
837 * Select this pack.
838 */
839 *mru_w = this_mru_w;
840 *lru_p = p;
841 *accept_windows_inuse = has_windows_inuse;
842}
843
844static int close_one_pack(void)
845{
846 struct packed_git *p, *lru_p = NULL;
847 struct pack_window *mru_w = NULL;
848 int accept_windows_inuse = 1;
849
850 for (p = packed_git; p; p = p->next) {
851 if (p->pack_fd == -1)
852 continue;
853 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
854 }
855
856 if (lru_p) {
857 close(lru_p->pack_fd);
858 pack_open_fds--;
859 lru_p->pack_fd = -1;
860 return 1;
861 }
862
863 return 0;
864}
865
03e79c88 866void unuse_pack(struct pack_window **w_cursor)
f9253394 867{
03e79c88
SP
868 struct pack_window *w = *w_cursor;
869 if (w) {
870 w->inuse_cnt--;
871 *w_cursor = NULL;
872 }
1f688557
JH
873}
874
fa5fc15d
SP
875void close_pack_index(struct packed_git *p)
876{
877 if (p->index_data) {
878 munmap((void *)p->index_data, p->index_size);
879 p->index_data = NULL;
880 }
881}
882
c74faea1
NP
883/*
884 * This is used by git-repack in case a newly created pack happens to
885 * contain the same set of objects as an existing one. In that case
886 * the resulting file might be different even if its name would be the
887 * same. It is best to close any reference to the old pack before it is
01689909 888 * replaced on disk. Of course no index pointers or windows for given pack
c74faea1
NP
889 * must subsist at this point. If ever objects from this pack are requested
890 * again, the new version of the pack will be reinitialized through
891 * reprepare_packed_git().
892 */
893void free_pack_by_name(const char *pack_name)
894{
895 struct packed_git *p, **pp = &packed_git;
896
897 while (*pp) {
898 p = *pp;
899 if (strcmp(pack_name, p->pack_name) == 0) {
fa3a0c94 900 clear_delta_base_cache();
c74faea1 901 close_pack_windows(p);
c7934306 902 if (p->pack_fd != -1) {
c74faea1 903 close(p->pack_fd);
c7934306
SP
904 pack_open_fds--;
905 }
fa5fc15d 906 close_pack_index(p);
c74faea1
NP
907 free(p->bad_object_sha1);
908 *pp = p->next;
c01f51cc
NTND
909 if (last_found_pack == p)
910 last_found_pack = NULL;
c74faea1
NP
911 free(p);
912 return;
913 }
914 pp = &p->next;
915 }
916}
917
a0788266
JS
918static unsigned int get_max_fd_limit(void)
919{
920#ifdef RLIMIT_NOFILE
491a8dec
JH
921 {
922 struct rlimit lim;
a0788266 923
491a8dec
JH
924 if (!getrlimit(RLIMIT_NOFILE, &lim))
925 return lim.rlim_cur;
926 }
927#endif
928
929#ifdef _SC_OPEN_MAX
930 {
931 long open_max = sysconf(_SC_OPEN_MAX);
932 if (0 < open_max)
933 return open_max;
934 /*
935 * Otherwise, we got -1 for one of the two
936 * reasons:
937 *
938 * (1) sysconf() did not understand _SC_OPEN_MAX
939 * and signaled an error with -1; or
940 * (2) sysconf() said there is no limit.
941 *
942 * We _could_ clear errno before calling sysconf() to
943 * tell these two cases apart and return a huge number
944 * in the latter case to let the caller cap it to a
945 * value that is not so selfish, but letting the
946 * fallback OPEN_MAX codepath take care of these cases
947 * is a lot simpler.
948 */
949 }
950#endif
a0788266 951
491a8dec 952#ifdef OPEN_MAX
a0788266
JS
953 return OPEN_MAX;
954#else
955 return 1; /* see the caller ;-) */
956#endif
957}
958
3cf8b462
SP
959/*
960 * Do not call this directly as this leaks p->pack_fd on error return;
961 * call open_packed_git() instead.
962 */
963static int open_packed_git_1(struct packed_git *p)
1f688557 964{
9bc879c1
SP
965 struct stat st;
966 struct pack_header hdr;
967 unsigned char sha1[20];
968 unsigned char *idx_sha1;
2c039da8 969 long fd_flag;
9bc879c1 970
d079837e
SP
971 if (!p->index_data && open_pack_index(p))
972 return error("packfile %s index unavailable", p->pack_name);
973
c7934306 974 if (!pack_max_fds) {
a0788266 975 unsigned int max_fds = get_max_fd_limit();
c7934306
SP
976
977 /* Save 3 for stdin/stdout/stderr, 22 for work */
978 if (25 < max_fds)
979 pack_max_fds = max_fds - 25;
980 else
981 pack_max_fds = 1;
982 }
983
88d0db55 984 while (pack_max_fds <= pack_open_fds && close_one_pack())
c7934306
SP
985 ; /* nothing */
986
f4e51683 987 p->pack_fd = git_open_noatime(p->pack_name);
9bc879c1 988 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 989 return -1;
c7934306 990 pack_open_fds++;
9bc879c1
SP
991
992 /* If we created the struct before we had the pack we lack size. */
bf592c50 993 if (!p->pack_size) {
bf592c50 994 if (!S_ISREG(st.st_mode))
072db278 995 return error("packfile %s not a regular file", p->pack_name);
bf592c50 996 p->pack_size = st.st_size;
9bc879c1 997 } else if (p->pack_size != st.st_size)
072db278 998 return error("packfile %s size changed", p->pack_name);
9bc879c1 999
2c039da8
JH
1000 /* We leave these file descriptors open with sliding mmap;
1001 * there is no point keeping them open across exec(), though.
1002 */
1003 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
1004 if (fd_flag < 0)
072db278 1005 return error("cannot determine file descriptor flags");
2c039da8
JH
1006 fd_flag |= FD_CLOEXEC;
1007 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 1008 return error("cannot set FD_CLOEXEC");
2c039da8 1009
9bc879c1 1010 /* Verify we recognize this pack file format. */
e6e2bd62 1011 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 1012 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 1013 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 1014 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 1015 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
1016 return error("packfile %s is version %"PRIu32" and not"
1017 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
1018 p->pack_name, ntohl(hdr.hdr_version));
1019
1020 /* Verify the pack matches its index. */
57059091 1021 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
1022 return error("packfile %s claims to have %"PRIu32" objects"
1023 " while index indicates %"PRIu32" objects",
57059091
NP
1024 p->pack_name, ntohl(hdr.hdr_entries),
1025 p->num_objects);
9bc879c1 1026 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 1027 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 1028 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 1029 return error("packfile %s signature is unavailable", p->pack_name);
42873078 1030 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 1031 if (hashcmp(sha1, idx_sha1))
072db278
SP
1032 return error("packfile %s does not match index", p->pack_name);
1033 return 0;
9bc879c1
SP
1034}
1035
3cf8b462
SP
1036static int open_packed_git(struct packed_git *p)
1037{
1038 if (!open_packed_git_1(p))
1039 return 0;
1040 if (p->pack_fd != -1) {
1041 close(p->pack_fd);
c7934306 1042 pack_open_fds--;
3cf8b462
SP
1043 p->pack_fd = -1;
1044 }
1045 return -1;
1046}
1047
c4001d92 1048static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
1049{
1050 /* We must promise at least 20 bytes (one hash) after the
1051 * offset is available from this window, otherwise the offset
1052 * is not actually in this window and a different window (which
1053 * has that one hash excess) must be used. This is to support
1054 * the object header and delta base parsing routines below.
1055 */
1056 off_t win_off = win->offset;
1057 return win_off <= offset
1058 && (offset + 20) <= (win_off + win->len);
1059}
1060
4b25d091 1061unsigned char *use_pack(struct packed_git *p,
03e79c88 1062 struct pack_window **w_cursor,
c4001d92 1063 off_t offset,
ef49a7a0 1064 unsigned long *left)
9bc879c1 1065{
60bb8b14 1066 struct pack_window *win = *w_cursor;
03e79c88 1067
a9d98a14 1068 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
1069 * pointless to ask for an offset into the middle of that
1070 * hash, and the in_window function above wouldn't match
1071 * don't allow an offset too close to the end of the file.
1072 */
d131b7af
SP
1073 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1074 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
1075 if (offset > (p->pack_size - 20))
1076 die("offset beyond end of packfile (truncated pack?)");
1077
1078 if (!win || !in_window(win, offset)) {
1079 if (win)
1080 win->inuse_cnt--;
1081 for (win = p->windows; win; win = win->next) {
1082 if (in_window(win, offset))
1083 break;
1084 }
1085 if (!win) {
78a28df9 1086 size_t window_align = packed_git_window_size / 2;
dc49cd76 1087 off_t len;
d131b7af
SP
1088
1089 if (p->pack_fd == -1 && open_packed_git(p))
1090 die("packfile %s cannot be accessed", p->pack_name);
1091
60bb8b14 1092 win = xcalloc(1, sizeof(*win));
78a28df9 1093 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
1094 len = p->pack_size - win->offset;
1095 if (len > packed_git_window_size)
1096 len = packed_git_window_size;
1097 win->len = (size_t)len;
60bb8b14 1098 pack_mapped += win->len;
11daf39b 1099 while (packed_git_limit < pack_mapped
7c3ecb32 1100 && unuse_one_window(p))
60bb8b14 1101 ; /* nothing */
c4712e45 1102 win->base = xmmap(NULL, win->len,
60bb8b14
SP
1103 PROT_READ, MAP_PRIVATE,
1104 p->pack_fd, win->offset);
1105 if (win->base == MAP_FAILED)
73b4e4be
SP
1106 die("packfile %s cannot be mapped: %s",
1107 p->pack_name,
1108 strerror(errno));
d131b7af
SP
1109 if (!win->offset && win->len == p->pack_size
1110 && !p->do_not_close) {
1111 close(p->pack_fd);
1112 pack_open_fds--;
1113 p->pack_fd = -1;
1114 }
a53128b6
SP
1115 pack_mmap_calls++;
1116 pack_open_windows++;
1117 if (pack_mapped > peak_pack_mapped)
1118 peak_pack_mapped = pack_mapped;
1119 if (pack_open_windows > peak_pack_open_windows)
1120 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
1121 win->next = p->windows;
1122 p->windows = win;
1123 }
1f688557 1124 }
03e79c88
SP
1125 if (win != *w_cursor) {
1126 win->last_used = pack_used_ctr++;
1127 win->inuse_cnt++;
1128 *w_cursor = win;
1129 }
60bb8b14 1130 offset -= win->offset;
03e79c88 1131 if (left)
dc49cd76 1132 *left = win->len - xsize_t(offset);
03e79c88 1133 return win->base + offset;
1f688557
JH
1134}
1135
27d69a46
NP
1136static struct packed_git *alloc_packed_git(int extra)
1137{
1138 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1139 memset(p, 0, sizeof(*p));
1140 p->pack_fd = -1;
1141 return p;
1142}
1143
e0500293
JN
1144static void try_to_free_pack_memory(size_t size)
1145{
7c3ecb32 1146 release_pack_memory(size);
e0500293
JN
1147}
1148
48bcc1c3 1149struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
1f688557 1150{
e0500293 1151 static int have_set_try_to_free_routine;
1f688557 1152 struct stat st;
48bcc1c3
JK
1153 size_t alloc;
1154 struct packed_git *p;
1f688557 1155
e0500293
JN
1156 if (!have_set_try_to_free_routine) {
1157 have_set_try_to_free_routine = 1;
1158 set_try_to_free_routine(try_to_free_pack_memory);
1159 }
1160
42873078
NP
1161 /*
1162 * Make sure a corresponding .pack file exists and that
1163 * the index looks sane.
1164 */
48bcc1c3 1165 if (!strip_suffix_mem(path, &path_len, ".idx"))
1f688557 1166 return NULL;
48bcc1c3
JK
1167
1168 /*
1169 * ".pack" is long enough to hold any suffix we're adding (and
1170 * the use xsnprintf double-checks that)
1171 */
1172 alloc = path_len + strlen(".pack") + 1;
1173 p = alloc_packed_git(alloc);
42873078 1174 memcpy(p->pack_name, path, path_len);
8d25931d 1175
48bcc1c3 1176 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
8d25931d
BC
1177 if (!access(p->pack_name, F_OK))
1178 p->pack_keep = 1;
1179
48bcc1c3 1180 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
d079837e 1181 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 1182 free(p);
1f688557
JH
1183 return NULL;
1184 }
42873078 1185
1f688557
JH
1186 /* ok, it looks sane as far as we can check without
1187 * actually mapping the pack file.
1188 */
1f688557 1189 p->pack_size = st.st_size;
9d835df2 1190 p->pack_local = local;
b867092f 1191 p->mtime = st.st_mtime;
42873078
NP
1192 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1193 hashclr(p->sha1);
1f688557
JH
1194 return p;
1195}
1196
7b64469a 1197struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 1198{
42873078 1199 const char *path = sha1_pack_name(sha1);
27d69a46 1200 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
bf592c50 1201
27d69a46
NP
1202 strcpy(p->pack_name, path);
1203 hashcpy(p->sha1, sha1);
42873078
NP
1204 if (check_packed_git_idx(idx_path, p)) {
1205 free(p);
bf592c50 1206 return NULL;
42873078 1207 }
bf592c50 1208
bf592c50
DB
1209 return p;
1210}
1211
1212void install_packed_git(struct packed_git *pack)
1213{
c7934306
SP
1214 if (pack->pack_fd != -1)
1215 pack_open_fds++;
1216
bf592c50
DB
1217 pack->next = packed_git;
1218 packed_git = pack;
1219}
1220
543c5caa
NTND
1221void (*report_garbage)(const char *desc, const char *path);
1222
1223static void report_helper(const struct string_list *list,
1224 int seen_bits, int first, int last)
1225{
1226 const char *msg;
1227 switch (seen_bits) {
1228 case 0:
235e8d59 1229 msg = "no corresponding .idx or .pack";
543c5caa
NTND
1230 break;
1231 case 1:
1232 msg = "no corresponding .idx";
1233 break;
1234 case 2:
1235 msg = "no corresponding .pack";
1236 break;
1237 default:
1238 return;
1239 }
1240 for (; first < last; first++)
1241 report_garbage(msg, list->items[first].string);
1242}
1243
1244static void report_pack_garbage(struct string_list *list)
1245{
1246 int i, baselen = -1, first = 0, seen_bits = 0;
1247
1248 if (!report_garbage)
1249 return;
1250
3383e199 1251 string_list_sort(list);
543c5caa
NTND
1252
1253 for (i = 0; i < list->nr; i++) {
1254 const char *path = list->items[i].string;
1255 if (baselen != -1 &&
1256 strncmp(path, list->items[first].string, baselen)) {
1257 report_helper(list, seen_bits, first, i);
1258 baselen = -1;
1259 seen_bits = 0;
1260 }
1261 if (baselen == -1) {
1262 const char *dot = strrchr(path, '.');
1263 if (!dot) {
1264 report_garbage("garbage found", path);
1265 continue;
1266 }
1267 baselen = dot - path + 1;
1268 first = i;
1269 }
1270 if (!strcmp(path + baselen, "pack"))
1271 seen_bits |= 1;
1272 else if (!strcmp(path + baselen, "idx"))
1273 seen_bits |= 2;
1274 }
1275 report_helper(list, seen_bits, first, list->nr);
1276}
1277
9d835df2 1278static void prepare_packed_git_one(char *objdir, int local)
1f688557 1279{
880fb8de
RS
1280 struct strbuf path = STRBUF_INIT;
1281 size_t dirnamelen;
1f688557
JH
1282 DIR *dir;
1283 struct dirent *de;
543c5caa 1284 struct string_list garbage = STRING_LIST_INIT_DUP;
1f688557 1285
880fb8de
RS
1286 strbuf_addstr(&path, objdir);
1287 strbuf_addstr(&path, "/pack");
1288 dir = opendir(path.buf);
b5b16990 1289 if (!dir) {
26125f6b 1290 if (errno != ENOENT)
bd2afde8 1291 error("unable to open object pack directory: %s: %s",
880fb8de
RS
1292 path.buf, strerror(errno));
1293 strbuf_release(&path);
1f688557 1294 return;
b5b16990 1295 }
880fb8de
RS
1296 strbuf_addch(&path, '/');
1297 dirnamelen = path.len;
1f688557 1298 while ((de = readdir(dir)) != NULL) {
1f688557 1299 struct packed_git *p;
47bf4b0f 1300 size_t base_len;
543c5caa
NTND
1301
1302 if (is_dot_or_dotdot(de->d_name))
9cb18f56
JM
1303 continue;
1304
880fb8de
RS
1305 strbuf_setlen(&path, dirnamelen);
1306 strbuf_addstr(&path, de->d_name);
d90906a9 1307
47bf4b0f
JK
1308 base_len = path.len;
1309 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
d90906a9
NTND
1310 /* Don't reopen a pack we already have. */
1311 for (p = packed_git; p; p = p->next) {
47bf4b0f
JK
1312 size_t len;
1313 if (strip_suffix(p->pack_name, ".pack", &len) &&
1314 len == base_len &&
1315 !memcmp(p->pack_name, path.buf, len))
d90906a9
NTND
1316 break;
1317 }
1318 if (p == NULL &&
1319 /*
1320 * See if it really is a valid .idx file with
1321 * corresponding .pack file that we can map.
1322 */
880fb8de 1323 (p = add_packed_git(path.buf, path.len, local)) != NULL)
d90906a9 1324 install_packed_git(p);
86f7780c 1325 }
543c5caa
NTND
1326
1327 if (!report_garbage)
1328 continue;
1329
2975c770
JK
1330 if (ends_with(de->d_name, ".idx") ||
1331 ends_with(de->d_name, ".pack") ||
1332 ends_with(de->d_name, ".bitmap") ||
1333 ends_with(de->d_name, ".keep"))
880fb8de 1334 string_list_append(&garbage, path.buf);
543c5caa 1335 else
880fb8de 1336 report_garbage("garbage found", path.buf);
1f688557 1337 }
5b35bcd5 1338 closedir(dir);
543c5caa
NTND
1339 report_pack_garbage(&garbage);
1340 string_list_clear(&garbage, 0);
880fb8de 1341 strbuf_release(&path);
1f688557
JH
1342}
1343
b867092f
JH
1344static int sort_pack(const void *a_, const void *b_)
1345{
1346 struct packed_git *a = *((struct packed_git **)a_);
1347 struct packed_git *b = *((struct packed_git **)b_);
1348 int st;
1349
1350 /*
1351 * Local packs tend to contain objects specific to our
1352 * variant of the project than remote ones. In addition,
1353 * remote ones could be on a network mounted filesystem.
1354 * Favor local ones for these reasons.
1355 */
1356 st = a->pack_local - b->pack_local;
1357 if (st)
1358 return -st;
1359
1360 /*
1361 * Younger packs tend to contain more recent objects,
1362 * and more recent objects tend to get accessed more
1363 * often.
1364 */
1365 if (a->mtime < b->mtime)
1366 return 1;
1367 else if (a->mtime == b->mtime)
1368 return 0;
1369 return -1;
1370}
1371
1372static void rearrange_packed_git(void)
1373{
1374 struct packed_git **ary, *p;
1375 int i, n;
1376
1377 for (n = 0, p = packed_git; p; p = p->next)
1378 n++;
1379 if (n < 2)
1380 return;
1381
1382 /* prepare an array of packed_git for easier sorting */
1383 ary = xcalloc(n, sizeof(struct packed_git *));
1384 for (n = 0, p = packed_git; p; p = p->next)
1385 ary[n++] = p;
1386
1387 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1388
1389 /* link them back again */
1390 for (i = 0; i < n - 1; i++)
1391 ary[i]->next = ary[i + 1];
1392 ary[n - 1]->next = NULL;
1393 packed_git = ary[0];
1394
1395 free(ary);
1396}
1397
637cdd9d 1398static int prepare_packed_git_run_once = 0;
9a217f2a 1399void prepare_packed_git(void)
1f688557 1400{
d5a63b99 1401 struct alternate_object_database *alt;
1f688557 1402
637cdd9d 1403 if (prepare_packed_git_run_once)
1f688557 1404 return;
9d835df2 1405 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 1406 prepare_alt_odb();
d5a63b99 1407 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 1408 alt->name[-1] = 0;
9d835df2 1409 prepare_packed_git_one(alt->base, 0);
1494e038 1410 alt->name[-1] = '/';
1f688557 1411 }
b867092f 1412 rearrange_packed_git();
637cdd9d
JK
1413 prepare_packed_git_run_once = 1;
1414}
1415
fc04c412 1416void reprepare_packed_git(void)
637cdd9d
JK
1417{
1418 prepare_packed_git_run_once = 0;
1419 prepare_packed_git();
1f688557
JH
1420}
1421
8eca0b47
NP
1422static void mark_bad_packed_object(struct packed_git *p,
1423 const unsigned char *sha1)
1424{
1425 unsigned i;
1426 for (i = 0; i < p->num_bad_objects; i++)
1427 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1428 return;
1429 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1430 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1431 p->num_bad_objects++;
1432}
1433
b6c4cecc 1434static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
ac939109
NP
1435{
1436 struct packed_git *p;
1437 unsigned i;
1438
1439 for (p = packed_git; p; p = p->next)
1440 for (i = 0; i < p->num_bad_objects; i++)
1441 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
b6c4cecc
JH
1442 return p;
1443 return NULL;
ac939109
NP
1444}
1445
090ea126
NTND
1446/*
1447 * With an in-core object data in "map", rehash it to make sure the
1448 * object name actually matches "sha1" to detect object corruption.
1449 * With "map" == NULL, try reading the object named with "sha1" using
1450 * the streaming interface and rehash it to do the same.
1451 */
1452int check_sha1_signature(const unsigned char *sha1, void *map,
1453 unsigned long size, const char *type)
0fcfd160
LT
1454{
1455 unsigned char real_sha1[20];
090ea126
NTND
1456 enum object_type obj_type;
1457 struct git_istream *st;
1458 git_SHA_CTX c;
1459 char hdr[32];
1460 int hdrlen;
1461
1462 if (map) {
1463 hash_sha1_file(map, size, type, real_sha1);
1464 return hashcmp(sha1, real_sha1) ? -1 : 0;
1465 }
1466
1467 st = open_istream(sha1, &obj_type, &size, NULL);
1468 if (!st)
1469 return -1;
1470
1471 /* Generate the header */
ef1286d3 1472 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
090ea126
NTND
1473
1474 /* Sha1.. */
1475 git_SHA1_Init(&c);
1476 git_SHA1_Update(&c, hdr, hdrlen);
1477 for (;;) {
1478 char buf[1024 * 16];
1479 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1480
f54fac53
JK
1481 if (readlen < 0) {
1482 close_istream(st);
1483 return -1;
1484 }
090ea126
NTND
1485 if (!readlen)
1486 break;
1487 git_SHA1_Update(&c, buf, readlen);
1488 }
1489 git_SHA1_Final(real_sha1, &c);
1490 close_istream(st);
a89fccd2 1491 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1492}
1493
ec73f580 1494int git_open_noatime(const char *name)
44d1c19e
LT
1495{
1496 static int sha1_file_open_flag = O_NOATIME;
44d1c19e 1497
f2e872aa 1498 for (;;) {
dff6f280
CB
1499 int fd;
1500
1501 errno = 0;
1502 fd = open(name, O_RDONLY | sha1_file_open_flag);
44d1c19e 1503 if (fd >= 0)
f2e872aa
SP
1504 return fd;
1505
f2e872aa
SP
1506 /* Might the failure be due to O_NOATIME? */
1507 if (errno != ENOENT && sha1_file_open_flag) {
44d1c19e 1508 sha1_file_open_flag = 0;
f2e872aa
SP
1509 continue;
1510 }
1511
1512 return -1;
44d1c19e 1513 }
44d1c19e
LT
1514}
1515
052fe5ea
JK
1516static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1517{
052fe5ea
JK
1518 struct alternate_object_database *alt;
1519
30d6c6ea 1520 if (!lstat(sha1_file_name(sha1), st))
052fe5ea
JK
1521 return 0;
1522
1523 prepare_alt_odb();
1524 errno = ENOENT;
1525 for (alt = alt_odb_list; alt; alt = alt->next) {
30d6c6ea 1526 fill_sha1_path(alt->name, sha1);
052fe5ea
JK
1527 if (!lstat(alt->base, st))
1528 return 0;
1529 }
1530
1531 return -1;
1532}
1533
44d1c19e
LT
1534static int open_sha1_file(const unsigned char *sha1)
1535{
1536 int fd;
44d1c19e 1537 struct alternate_object_database *alt;
d6c8a05b 1538 int most_interesting_errno;
44d1c19e 1539
30d6c6ea 1540 fd = git_open_noatime(sha1_file_name(sha1));
44d1c19e
LT
1541 if (fd >= 0)
1542 return fd;
d6c8a05b 1543 most_interesting_errno = errno;
44d1c19e
LT
1544
1545 prepare_alt_odb();
44d1c19e 1546 for (alt = alt_odb_list; alt; alt = alt->next) {
30d6c6ea 1547 fill_sha1_path(alt->name, sha1);
f4e51683 1548 fd = git_open_noatime(alt->base);
44d1c19e
LT
1549 if (fd >= 0)
1550 return fd;
d6c8a05b
JK
1551 if (most_interesting_errno == ENOENT)
1552 most_interesting_errno = errno;
44d1c19e 1553 }
d6c8a05b 1554 errno = most_interesting_errno;
44d1c19e
LT
1555 return -1;
1556}
1557
f0270efd 1558void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1559{
0fcfd160 1560 void *map;
144bde78 1561 int fd;
ace1534d 1562
44d1c19e
LT
1563 fd = open_sha1_file(sha1);
1564 map = NULL;
1565 if (fd >= 0) {
1566 struct stat st;
0fcfd160 1567
44d1c19e
LT
1568 if (!fstat(fd, &st)) {
1569 *size = xsize_t(st.st_size);
33e42de0
MM
1570 if (!*size) {
1571 /* mmap() is forbidden on empty files */
1572 error("object file %s is empty", sha1_file_name(sha1));
1573 return NULL;
1574 }
44d1c19e 1575 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1576 }
44d1c19e 1577 close(fd);
0fcfd160 1578 }
0fcfd160
LT
1579 return map;
1580}
1581
09ded04b
NP
1582unsigned long unpack_object_header_buffer(const unsigned char *buf,
1583 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1584{
ad1ed5ee 1585 unsigned shift;
48fb7deb 1586 unsigned long size, c;
ad1ed5ee
JH
1587 unsigned long used = 0;
1588
1589 c = buf[used++];
1590 *type = (c >> 4) & 7;
1591 size = c & 15;
1592 shift = 4;
1593 while (c & 0x80) {
f630cfda 1594 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1595 error("bad object header");
ea4f9685
JH
1596 size = used = 0;
1597 break;
09ded04b 1598 }
ad1ed5ee
JH
1599 c = buf[used++];
1600 size += (c & 0x7f) << shift;
1601 shift += 7;
1602 }
1603 *sizep = size;
1604 return used;
1605}
1606
eb4f4076 1607int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
ad1ed5ee 1608{
c4483576
LT
1609 /* Get the data stream */
1610 memset(stream, 0, sizeof(*stream));
1611 stream->next_in = map;
1612 stream->avail_in = mapsize;
1613 stream->next_out = buffer;
93821bd9
LT
1614 stream->avail_out = bufsiz;
1615
39c68542 1616 git_inflate_init(stream);
cc5c54e7 1617 return git_inflate(stream, 0);
c4483576
LT
1618}
1619
46f03448
KN
1620static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1621 unsigned long mapsize, void *buffer,
1622 unsigned long bufsiz, struct strbuf *header)
1623{
1624 int status;
1625
1626 status = unpack_sha1_header(stream, map, mapsize, buffer, bufsiz);
1627
1628 /*
1629 * Check if entire header is unpacked in the first iteration.
1630 */
1631 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1632 return 0;
1633
1634 /*
1635 * buffer[0..bufsiz] was not large enough. Copy the partial
1636 * result out to header, and then append the result of further
1637 * reading the stream.
1638 */
1639 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1640 stream->next_out = buffer;
1641 stream->avail_out = bufsiz;
1642
1643 do {
1644 status = git_inflate(stream, 0);
1645 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1646 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1647 return 0;
1648 stream->next_out = buffer;
1649 stream->avail_out = bufsiz;
1650 } while (status != Z_STREAM_END);
1651 return -1;
1652}
1653
ef49a7a0 1654static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1655{
1656 int bytes = strlen(buffer) + 1;
3aee68aa 1657 unsigned char *buf = xmallocz(size);
93821bd9 1658 unsigned long n;
7efbff75 1659 int status = Z_OK;
5180cacc 1660
93821bd9
LT
1661 n = stream->total_out - bytes;
1662 if (n > size)
1663 n = size;
1664 memcpy(buf, (char *) buffer + bytes, n);
1665 bytes = n;
456cdf6e
LT
1666 if (bytes <= size) {
1667 /*
1668 * The above condition must be (bytes <= size), not
1669 * (bytes < size). In other words, even though we
ccf5ace0 1670 * expect no more output and set avail_out to zero,
456cdf6e
LT
1671 * the input zlib stream may have bytes that express
1672 * "this concludes the stream", and we *do* want to
1673 * eat that input.
1674 *
1675 * Otherwise we would not be able to test that we
1676 * consumed all the input to reach the expected size;
1677 * we also want to check that zlib tells us that all
1678 * went well with status == Z_STREAM_END at the end.
1679 */
5180cacc
LT
1680 stream->next_out = buf + bytes;
1681 stream->avail_out = size - bytes;
7efbff75 1682 while (status == Z_OK)
39c68542 1683 status = git_inflate(stream, Z_FINISH);
5180cacc 1684 }
456cdf6e 1685 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1686 git_inflate_end(stream);
7efbff75
JH
1687 return buf;
1688 }
1689
1690 if (status < 0)
1691 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1692 else if (stream->avail_in)
1693 error("garbage at end of loose object '%s'",
1694 sha1_to_hex(sha1));
1695 free(buf);
1696 return NULL;
5180cacc
LT
1697}
1698
1699/*
1700 * We used to just use "sscanf()", but that's actually way
1701 * too permissive for what we want to check. So do an anal
1702 * object header parse by hand.
1703 */
46f03448
KN
1704static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1705 unsigned int flags)
5180cacc 1706{
46f03448 1707 const char *type_buf = hdr;
5180cacc 1708 unsigned long size;
46f03448 1709 int type, type_len = 0;
5180cacc
LT
1710
1711 /*
46f03448 1712 * The type can be of any size but is followed by
21666f1a 1713 * a space.
5180cacc 1714 */
5180cacc
LT
1715 for (;;) {
1716 char c = *hdr++;
1717 if (c == ' ')
1718 break;
46f03448 1719 type_len++;
5180cacc 1720 }
46f03448
KN
1721
1722 type = type_from_string_gently(type_buf, type_len, 1);
1723 if (oi->typename)
1724 strbuf_add(oi->typename, type_buf, type_len);
1725 /*
1726 * Set type to 0 if its an unknown object and
1727 * we're obtaining the type using '--allow-unkown-type'
1728 * option.
1729 */
1730 if ((flags & LOOKUP_UNKNOWN_OBJECT) && (type < 0))
1731 type = 0;
1732 else if (type < 0)
1733 die("invalid object type");
1734 if (oi->typep)
1735 *oi->typep = type;
5180cacc
LT
1736
1737 /*
1738 * The length must follow immediately, and be in canonical
1739 * decimal format (ie "010" is not valid).
1740 */
1741 size = *hdr++ - '0';
1742 if (size > 9)
1743 return -1;
1744 if (size) {
1745 for (;;) {
1746 unsigned long c = *hdr - '0';
1747 if (c > 9)
1748 break;
1749 hdr++;
1750 size = size * 10 + c;
1751 }
1752 }
46f03448
KN
1753
1754 if (oi->sizep)
1755 *oi->sizep = size;
5180cacc
LT
1756
1757 /*
1758 * The length must be followed by a zero byte
1759 */
46f03448
KN
1760 return *hdr ? -1 : type;
1761}
1762
1763int parse_sha1_header(const char *hdr, unsigned long *sizep)
1764{
1765 struct object_info oi;
1766
1767 oi.sizep = sizep;
1768 oi.typename = NULL;
1769 oi.typep = NULL;
1770 return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
5180cacc
LT
1771}
1772
7efbff75 1773static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1774{
5180cacc 1775 int ret;
ef49a7a0 1776 git_zstream stream;
5180cacc 1777 char hdr[8192];
0fcfd160 1778
5180cacc 1779 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1780 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1781 return NULL;
1782
7efbff75 1783 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1784}
1785
54dab52a
NP
1786unsigned long get_size_from_delta(struct packed_git *p,
1787 struct pack_window **w_curs,
1788 off_t curpos)
1789{
1790 const unsigned char *data;
1791 unsigned char delta_head[20], *in;
ef49a7a0 1792 git_zstream stream;
54dab52a
NP
1793 int st;
1794
1795 memset(&stream, 0, sizeof(stream));
1796 stream.next_out = delta_head;
1797 stream.avail_out = sizeof(delta_head);
1798
39c68542 1799 git_inflate_init(&stream);
54dab52a
NP
1800 do {
1801 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1802 stream.next_in = in;
39c68542 1803 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1804 curpos += stream.next_in - in;
1805 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1806 stream.total_out < sizeof(delta_head));
39c68542 1807 git_inflate_end(&stream);
3d77d877
NP
1808 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1809 error("delta data unpack-initial failed");
1810 return 0;
1811 }
54dab52a
NP
1812
1813 /* Examine the initial part of the delta to figure out
1814 * the result size.
1815 */
1816 data = delta_head;
1817
1818 /* ignore base size */
1819 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1820
1821 /* Read the result size */
1822 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1823}
1824
c4001d92 1825static off_t get_delta_base(struct packed_git *p,
03e79c88 1826 struct pack_window **w_curs,
c4001d92 1827 off_t *curpos,
21666f1a 1828 enum object_type type,
c4001d92 1829 off_t delta_obj_offset)
eb32d236 1830{
2b87c45b 1831 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1832 off_t base_offset;
eb32d236 1833
8d8a4ea5
SP
1834 /* use_pack() assured us we have [base_info, base_info + 20)
1835 * as a range that we can look at without walking off the
1836 * end of the mapped window. Its actually the hash size
1837 * that is assured. An OFS_DELTA longer than the hash size
1838 * is stupid, as then a REF_DELTA would be smaller to store.
1839 */
21666f1a 1840 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1841 unsigned used = 0;
1842 unsigned char c = base_info[used++];
1843 base_offset = c & 127;
1844 while (c & 128) {
1845 base_offset += 1;
8723f216 1846 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1847 return 0; /* overflow */
eb32d236
NP
1848 c = base_info[used++];
1849 base_offset = (base_offset << 7) + (c & 127);
1850 }
1851 base_offset = delta_obj_offset - base_offset;
d8f32556 1852 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1853 return 0; /* out of bound */
2b87c45b 1854 *curpos += used;
21666f1a 1855 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1856 /* The base entry _must_ be in the same pack */
1857 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1858 *curpos += 20;
eb32d236
NP
1859 } else
1860 die("I am totally screwed");
2b87c45b 1861 return base_offset;
eb32d236
NP
1862}
1863
5d642e75
JK
1864/*
1865 * Like get_delta_base above, but we return the sha1 instead of the pack
1866 * offset. This means it is cheaper for REF deltas (we do not have to do
1867 * the final object lookup), but more expensive for OFS deltas (we
1868 * have to load the revidx to convert the offset back into a sha1).
1869 */
1870static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1871 struct pack_window **w_curs,
1872 off_t curpos,
1873 enum object_type type,
1874 off_t delta_obj_offset)
1875{
1876 if (type == OBJ_REF_DELTA) {
1877 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1878 return base;
1879 } else if (type == OBJ_OFS_DELTA) {
1880 struct revindex_entry *revidx;
1881 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1882 type, delta_obj_offset);
1883
1884 if (!base_offset)
1885 return NULL;
1886
1887 revidx = find_pack_revindex(p, base_offset);
1888 if (!revidx)
1889 return NULL;
1890
1891 return nth_packed_object_sha1(p, revidx->nr);
1892 } else
1893 return NULL;
1894}
1895
f8c8abc5
JH
1896int unpack_object_header(struct packed_git *p,
1897 struct pack_window **w_curs,
1898 off_t *curpos,
1899 unsigned long *sizep)
a733cb60 1900{
03e79c88 1901 unsigned char *base;
ef49a7a0 1902 unsigned long left;
ad1ed5ee 1903 unsigned long used;
2b87c45b 1904 enum object_type type;
a733cb60 1905
8d8a4ea5 1906 /* use_pack() assures us we have [base, base + 20) available
0353a0c4 1907 * as a range that we can look at. (Its actually the hash
3dff5379 1908 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1909 * the maximum deflated object size is 2^137, which is just
1910 * insane, so we know won't exceed what we have been given.
1911 */
2b87c45b 1912 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1913 used = unpack_object_header_buffer(base, left, &type, sizep);
1914 if (!used) {
1915 type = OBJ_BAD;
1916 } else
1917 *curpos += used;
ad1ed5ee 1918
2b87c45b 1919 return type;
a733cb60
LT
1920}
1921
790d96c0
TR
1922static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1923{
1924 int type;
1925 struct revindex_entry *revidx;
1926 const unsigned char *sha1;
1927 revidx = find_pack_revindex(p, obj_offset);
1928 if (!revidx)
1929 return OBJ_BAD;
1930 sha1 = nth_packed_object_sha1(p, revidx->nr);
1931 mark_bad_packed_object(p, sha1);
1932 type = sha1_object_info(sha1, NULL);
1933 if (type <= OBJ_NONE)
1934 return OBJ_BAD;
1935 return type;
1936}
1937
790d96c0
TR
1938#define POI_STACK_PREALLOC 64
1939
90191d37
JK
1940static enum object_type packed_to_object_type(struct packed_git *p,
1941 off_t obj_offset,
1942 enum object_type type,
1943 struct pack_window **w_curs,
1944 off_t curpos)
1f688557 1945{
790d96c0
TR
1946 off_t small_poi_stack[POI_STACK_PREALLOC];
1947 off_t *poi_stack = small_poi_stack;
1948 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
5db47c2b 1949
790d96c0
TR
1950 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1951 off_t base_offset;
90191d37 1952 unsigned long size;
790d96c0
TR
1953 /* Push the object we're going to leave behind */
1954 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1955 poi_stack_alloc = alloc_nr(poi_stack_nr);
1956 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1957 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1958 } else {
1959 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1960 }
1961 poi_stack[poi_stack_nr++] = obj_offset;
1962 /* If parsing the base offset fails, just unwind */
90191d37 1963 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
790d96c0
TR
1964 if (!base_offset)
1965 goto unwind;
1966 curpos = obj_offset = base_offset;
90191d37 1967 type = unpack_object_header(p, w_curs, &curpos, &size);
790d96c0
TR
1968 if (type <= OBJ_NONE) {
1969 /* If getting the base itself fails, we first
1970 * retry the base, otherwise unwind */
1971 type = retry_bad_packed_offset(p, base_offset);
1972 if (type > OBJ_NONE)
1973 goto out;
1974 goto unwind;
1975 }
1976 }
1977
21666f1a 1978 switch (type) {
790d96c0 1979 case OBJ_BAD:
a733cb60 1980 case OBJ_COMMIT:
a733cb60 1981 case OBJ_TREE:
a733cb60 1982 case OBJ_BLOB:
a733cb60 1983 case OBJ_TAG:
a69d0943 1984 break;
1f688557 1985 default:
3d77d877
NP
1986 error("unknown object type %i at offset %"PRIuMAX" in %s",
1987 type, (uintmax_t)obj_offset, p->pack_name);
1988 type = OBJ_BAD;
1f688557 1989 }
790d96c0
TR
1990
1991out:
1992 if (poi_stack != small_poi_stack)
1993 free(poi_stack);
21666f1a 1994 return type;
790d96c0
TR
1995
1996unwind:
1997 while (poi_stack_nr) {
1998 obj_offset = poi_stack[--poi_stack_nr];
1999 type = retry_bad_packed_offset(p, obj_offset);
2000 if (type > OBJ_NONE)
2001 goto out;
2002 }
2003 type = OBJ_BAD;
2004 goto out;
1f688557
JH
2005}
2006
90191d37 2007static int packed_object_info(struct packed_git *p, off_t obj_offset,
23c339c0 2008 struct object_info *oi)
90191d37
JK
2009{
2010 struct pack_window *w_curs = NULL;
2011 unsigned long size;
2012 off_t curpos = obj_offset;
2013 enum object_type type;
2014
412916ee
JK
2015 /*
2016 * We always get the representation type, but only convert it to
2017 * a "real" type later if the caller is interested.
2018 */
90191d37
JK
2019 type = unpack_object_header(p, &w_curs, &curpos, &size);
2020
23c339c0 2021 if (oi->sizep) {
90191d37
JK
2022 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2023 off_t tmp_pos = curpos;
2024 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
2025 type, obj_offset);
2026 if (!base_offset) {
2027 type = OBJ_BAD;
2028 goto out;
2029 }
23c339c0
JK
2030 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2031 if (*oi->sizep == 0) {
90191d37
JK
2032 type = OBJ_BAD;
2033 goto out;
2034 }
2035 } else {
23c339c0 2036 *oi->sizep = size;
90191d37
JK
2037 }
2038 }
2039
23c339c0 2040 if (oi->disk_sizep) {
90191d37 2041 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
23c339c0 2042 *oi->disk_sizep = revidx[1].offset - obj_offset;
90191d37
JK
2043 }
2044
23c339c0
JK
2045 if (oi->typep) {
2046 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
2047 if (*oi->typep < 0) {
412916ee
JK
2048 type = OBJ_BAD;
2049 goto out;
2050 }
2051 }
90191d37 2052
5d642e75
JK
2053 if (oi->delta_base_sha1) {
2054 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2055 const unsigned char *base;
2056
2057 base = get_delta_base_sha1(p, &w_curs, curpos,
2058 type, obj_offset);
2059 if (!base) {
2060 type = OBJ_BAD;
2061 goto out;
2062 }
2063
2064 hashcpy(oi->delta_base_sha1, base);
2065 } else
2066 hashclr(oi->delta_base_sha1);
2067 }
2068
90191d37
JK
2069out:
2070 unuse_pack(&w_curs);
2071 return type;
2072}
2073
eb950c19 2074static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 2075 struct pack_window **w_curs,
c4001d92 2076 off_t curpos,
eb950c19 2077 unsigned long size)
de530aaa
SP
2078{
2079 int st;
ef49a7a0 2080 git_zstream stream;
079afb18 2081 unsigned char *buffer, *in;
de530aaa 2082
735efde8
NTND
2083 buffer = xmallocz_gently(size);
2084 if (!buffer)
2085 return NULL;
de530aaa 2086 memset(&stream, 0, sizeof(stream));
de530aaa 2087 stream.next_out = buffer;
39eea7bd 2088 stream.avail_out = size + 1;
de530aaa 2089
39c68542 2090 git_inflate_init(&stream);
079afb18 2091 do {
2b87c45b 2092 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 2093 stream.next_in = in;
39c68542 2094 st = git_inflate(&stream, Z_FINISH);
39eea7bd
JH
2095 if (!stream.avail_out)
2096 break; /* the payload is larger than it should be */
2b87c45b 2097 curpos += stream.next_in - in;
079afb18 2098 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 2099 git_inflate_end(&stream);
de530aaa
SP
2100 if ((st != Z_STREAM_END) || stream.total_out != size) {
2101 free(buffer);
2102 return NULL;
2103 }
2104
2105 return buffer;
2106}
2107
e5e01619
LT
2108#define MAX_DELTA_CACHE (256)
2109
18bdec11 2110static size_t delta_base_cached;
5e08ecbf
NP
2111
2112static struct delta_base_cache_lru_list {
2113 struct delta_base_cache_lru_list *prev;
2114 struct delta_base_cache_lru_list *next;
2115} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
2116
e5e01619 2117static struct delta_base_cache_entry {
5e08ecbf
NP
2118 struct delta_base_cache_lru_list lru;
2119 void *data;
e5e01619
LT
2120 struct packed_git *p;
2121 off_t base_offset;
2122 unsigned long size;
e5e01619
LT
2123 enum object_type type;
2124} delta_base_cache[MAX_DELTA_CACHE];
2125
2126static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
2127{
2128 unsigned long hash;
2129
2130 hash = (unsigned long)p + (unsigned long)base_offset;
2131 hash += (hash >> 8) + (hash >> 16);
3358004a 2132 return hash % MAX_DELTA_CACHE;
e5e01619
LT
2133}
2134
84dd81c1
TR
2135static struct delta_base_cache_entry *
2136get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
5266d369
JH
2137{
2138 unsigned long hash = pack_entry_hash(p, base_offset);
84dd81c1
TR
2139 return delta_base_cache + hash;
2140}
2141
2142static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
2143 struct packed_git *p, off_t base_offset)
2144{
5266d369
JH
2145 return (ent->data && ent->p == p && ent->base_offset == base_offset);
2146}
2147
84dd81c1
TR
2148static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2149{
2150 struct delta_base_cache_entry *ent;
2151 ent = get_delta_base_cache_entry(p, base_offset);
2152 return eq_delta_base_cache_entry(ent, p, base_offset);
2153}
2154
2155static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2156{
2157 ent->data = NULL;
2158 ent->lru.next->prev = ent->lru.prev;
2159 ent->lru.prev->next = ent->lru.next;
2160 delta_base_cached -= ent->size;
2161}
2162
62f255ad 2163static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 2164 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 2165{
84dd81c1 2166 struct delta_base_cache_entry *ent;
e5e01619 2167 void *ret;
e5e01619 2168
84dd81c1
TR
2169 ent = get_delta_base_cache_entry(p, base_offset);
2170
2171 if (!eq_delta_base_cache_entry(ent, p, base_offset))
749bc58c 2172 return unpack_entry(p, base_offset, type, base_size);
e5e01619 2173
84dd81c1
TR
2174 ret = ent->data;
2175
2176 if (!keep_cache)
2177 clear_delta_base_cache_entry(ent);
2178 else
182af834 2179 ret = xmemdupz(ent->data, ent->size);
e5e01619
LT
2180 *type = ent->type;
2181 *base_size = ent->size;
2182 return ret;
62f255ad
LT
2183}
2184
18bdec11
SP
2185static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2186{
2187 if (ent->data) {
2188 free(ent->data);
2189 ent->data = NULL;
5e08ecbf
NP
2190 ent->lru.next->prev = ent->lru.prev;
2191 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
2192 delta_base_cached -= ent->size;
2193 }
2194}
2195
3d20c636
SP
2196void clear_delta_base_cache(void)
2197{
2198 unsigned long p;
2199 for (p = 0; p < MAX_DELTA_CACHE; p++)
2200 release_delta_base_cache(&delta_base_cache[p]);
2201}
2202
62f255ad
LT
2203static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2204 void *base, unsigned long base_size, enum object_type type)
2205{
5e08ecbf 2206 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 2207 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 2208 struct delta_base_cache_lru_list *lru;
e5e01619 2209
18bdec11
SP
2210 release_delta_base_cache(ent);
2211 delta_base_cached += base_size;
5e08ecbf
NP
2212
2213 for (lru = delta_base_cache_lru.next;
2214 delta_base_cached > delta_base_cache_limit
2215 && lru != &delta_base_cache_lru;
2216 lru = lru->next) {
2217 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
2218 if (f->type == OBJ_BLOB)
2219 release_delta_base_cache(f);
2220 }
5e08ecbf
NP
2221 for (lru = delta_base_cache_lru.next;
2222 delta_base_cached > delta_base_cache_limit
2223 && lru != &delta_base_cache_lru;
2224 lru = lru->next) {
2225 struct delta_base_cache_entry *f = (void *)lru;
2226 release_delta_base_cache(f);
2227 }
18bdec11 2228
e5e01619
LT
2229 ent->p = p;
2230 ent->base_offset = base_offset;
2231 ent->type = type;
2232 ent->data = base;
2233 ent->size = base_size;
5e08ecbf
NP
2234 ent->lru.next = &delta_base_cache_lru;
2235 ent->lru.prev = delta_base_cache_lru.prev;
2236 delta_base_cache_lru.prev->next = &ent->lru;
2237 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
2238}
2239
c2c5b270
CC
2240static void *read_object(const unsigned char *sha1, enum object_type *type,
2241 unsigned long *size);
2242
5f44324d
JH
2243static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2244{
67dc598e
KB
2245 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2246 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2247 p->pack_name, (uintmax_t)obj_offset);
5f44324d
JH
2248}
2249
0e8189e2
NP
2250int do_check_packed_object_crc;
2251
abe601bb
TR
2252#define UNPACK_ENTRY_STACK_PREALLOC 64
2253struct unpack_entry_stack_ent {
2254 off_t obj_offset;
2255 off_t curpos;
2256 unsigned long size;
2257};
2258
c4001d92 2259void *unpack_entry(struct packed_git *p, off_t obj_offset,
abe601bb 2260 enum object_type *final_type, unsigned long *final_size)
1f688557 2261{
03e79c88 2262 struct pack_window *w_curs = NULL;
c4001d92 2263 off_t curpos = obj_offset;
abe601bb
TR
2264 void *data = NULL;
2265 unsigned long size;
2266 enum object_type type;
2267 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2268 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2269 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2270 int base_from_cache = 0;
1f688557 2271
67dc598e 2272 write_pack_access_log(p, obj_offset);
5f44324d 2273
abe601bb
TR
2274 /* PHASE 1: drill down to the innermost base object */
2275 for (;;) {
2276 off_t base_offset;
2277 int i;
2278 struct delta_base_cache_entry *ent;
2279
77965f8b
NTND
2280 ent = get_delta_base_cache_entry(p, curpos);
2281 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2282 type = ent->type;
2283 data = ent->data;
2284 size = ent->size;
2285 clear_delta_base_cache_entry(ent);
2286 base_from_cache = 1;
2287 break;
2288 }
2289
abe601bb
TR
2290 if (do_check_packed_object_crc && p->index_version > 1) {
2291 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2292 unsigned long len = revidx[1].offset - obj_offset;
2293 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2294 const unsigned char *sha1 =
2295 nth_packed_object_sha1(p, revidx->nr);
2296 error("bad packed object CRC for %s",
2297 sha1_to_hex(sha1));
2298 mark_bad_packed_object(p, sha1);
2299 unuse_pack(&w_curs);
2300 return NULL;
2301 }
2302 }
2303
abe601bb
TR
2304 type = unpack_object_header(p, &w_curs, &curpos, &size);
2305 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2306 break;
2307
2308 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2309 if (!base_offset) {
2310 error("failed to validate delta base reference "
2311 "at offset %"PRIuMAX" from %s",
2312 (uintmax_t)curpos, p->pack_name);
2313 /* bail to phase 2, in hopes of recovery */
2314 data = NULL;
2315 break;
2316 }
2317
2318 /* push object, proceed to base */
2319 if (delta_stack_nr >= delta_stack_alloc
2320 && delta_stack == small_delta_stack) {
2321 delta_stack_alloc = alloc_nr(delta_stack_nr);
2322 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2323 memcpy(delta_stack, small_delta_stack,
2324 sizeof(*delta_stack)*delta_stack_nr);
2325 } else {
2326 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2327 }
2328 i = delta_stack_nr++;
2329 delta_stack[i].obj_offset = obj_offset;
2330 delta_stack[i].curpos = curpos;
2331 delta_stack[i].size = size;
2332
2333 curpos = obj_offset = base_offset;
0e8189e2
NP
2334 }
2335
abe601bb
TR
2336 /* PHASE 2: handle the base */
2337 switch (type) {
eb32d236
NP
2338 case OBJ_OFS_DELTA:
2339 case OBJ_REF_DELTA:
abe601bb
TR
2340 if (data)
2341 die("BUG in unpack_entry: left loop at a valid delta");
4d703a1a 2342 break;
a733cb60 2343 case OBJ_COMMIT:
a733cb60 2344 case OBJ_TREE:
a733cb60 2345 case OBJ_BLOB:
a733cb60 2346 case OBJ_TAG:
abe601bb
TR
2347 if (!base_from_cache)
2348 data = unpack_compressed_entry(p, &w_curs, curpos, size);
4d703a1a 2349 break;
1f688557 2350 default:
8eca0b47
NP
2351 data = NULL;
2352 error("unknown object type %i at offset %"PRIuMAX" in %s",
abe601bb
TR
2353 type, (uintmax_t)obj_offset, p->pack_name);
2354 }
2355
2356 /* PHASE 3: apply deltas in order */
2357
2358 /* invariants:
2359 * 'data' holds the base data, or NULL if there was corruption
2360 */
2361 while (delta_stack_nr) {
2362 void *delta_data;
2363 void *base = data;
2364 unsigned long delta_size, base_size = size;
2365 int i;
2366
2367 data = NULL;
2368
2369 if (base)
2370 add_delta_base_cache(p, obj_offset, base, base_size, type);
2371
2372 if (!base) {
2373 /*
2374 * We're probably in deep shit, but let's try to fetch
2375 * the required base anyway from another pack or loose.
2376 * This is costly but should happen only in the presence
2377 * of a corrupted pack, and is better than failing outright.
2378 */
2379 struct revindex_entry *revidx;
2380 const unsigned char *base_sha1;
2381 revidx = find_pack_revindex(p, obj_offset);
2382 if (revidx) {
2383 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2384 error("failed to read delta base object %s"
2385 " at offset %"PRIuMAX" from %s",
2386 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2387 p->pack_name);
2388 mark_bad_packed_object(p, base_sha1);
2389 base = read_object(base_sha1, &type, &base_size);
2390 }
2391 }
2392
2393 i = --delta_stack_nr;
2394 obj_offset = delta_stack[i].obj_offset;
2395 curpos = delta_stack[i].curpos;
2396 delta_size = delta_stack[i].size;
2397
2398 if (!base)
2399 continue;
2400
2401 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2402
2403 if (!delta_data) {
2404 error("failed to unpack compressed delta "
2405 "at offset %"PRIuMAX" from %s",
2406 (uintmax_t)curpos, p->pack_name);
abe601bb
TR
2407 data = NULL;
2408 continue;
2409 }
2410
2411 data = patch_delta(base, base_size,
2412 delta_data, delta_size,
2413 &size);
1ee886c1
JK
2414
2415 /*
2416 * We could not apply the delta; warn the user, but keep going.
2417 * Our failure will be noticed either in the next iteration of
2418 * the loop, or if this is the final delta, in the caller when
2419 * we return NULL. Those code paths will take care of making
2420 * a more explicit warning and retrying with another copy of
2421 * the object.
2422 */
abe601bb 2423 if (!data)
1ee886c1 2424 error("failed to apply delta");
abe601bb 2425
4b8f772c 2426 free(delta_data);
1f688557 2427 }
abe601bb
TR
2428
2429 *final_type = type;
2430 *final_size = size;
2431
03e79c88 2432 unuse_pack(&w_curs);
019d1e65
NTND
2433
2434 if (delta_stack != small_delta_stack)
2435 free(delta_stack);
2436
21666f1a 2437 return data;
1f688557
JH
2438}
2439
d079837e 2440const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 2441 uint32_t n)
9a217f2a 2442{
42873078 2443 const unsigned char *index = p->index_data;
d079837e
SP
2444 if (!index) {
2445 if (open_pack_index(p))
2446 return NULL;
2447 index = p->index_data;
2448 }
57059091 2449 if (n >= p->num_objects)
d72308e0 2450 return NULL;
74e34e1f
NP
2451 index += 4 * 256;
2452 if (p->index_version == 1) {
2453 return index + 24 * n + 4;
2454 } else {
2455 index += 8;
2456 return index + 20 * n;
2457 }
2458}
2459
99093238 2460off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
2461{
2462 const unsigned char *index = p->index_data;
2463 index += 4 * 256;
2464 if (p->index_version == 1) {
2465 return ntohl(*((uint32_t *)(index + 24 * n)));
2466 } else {
2467 uint32_t off;
2468 index += 8 + p->num_objects * (20 + 4);
2469 off = ntohl(*((uint32_t *)(index + 4 * n)));
2470 if (!(off & 0x80000000))
2471 return off;
2472 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2473 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2474 ntohl(*((uint32_t *)(index + 4)));
2475 }
9a217f2a
JH
2476}
2477
c4001d92 2478off_t find_pack_entry_one(const unsigned char *sha1,
43057304 2479 struct packed_git *p)
1f688557 2480{
42873078 2481 const uint32_t *level1_ofs = p->index_data;
42873078 2482 const unsigned char *index = p->index_data;
628522ec
JH
2483 unsigned hi, lo, stride;
2484 static int use_lookup = -1;
2485 static int debug_lookup = -1;
2486
2487 if (debug_lookup < 0)
2488 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 2489
d079837e
SP
2490 if (!index) {
2491 if (open_pack_index(p))
2492 return 0;
2493 level1_ofs = p->index_data;
2494 index = p->index_data;
2495 }
74e34e1f
NP
2496 if (p->index_version > 1) {
2497 level1_ofs += 2;
2498 index += 8;
2499 }
42873078 2500 index += 4 * 256;
74e34e1f
NP
2501 hi = ntohl(level1_ofs[*sha1]);
2502 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
2503 if (p->index_version > 1) {
2504 stride = 20;
2505 } else {
2506 stride = 24;
2507 index += 4;
2508 }
2509
2510 if (debug_lookup)
6e1c2344 2511 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
2512 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2513
2514 if (use_lookup < 0)
2515 use_lookup = !!getenv("GIT_USE_LOOKUP");
2516 if (use_lookup) {
2517 int pos = sha1_entry_pos(index, stride, 0,
2518 lo, hi, p->num_objects, sha1);
2519 if (pos < 0)
2520 return 0;
2521 return nth_packed_object_offset(p, pos);
2522 }
1f688557
JH
2523
2524 do {
74e34e1f 2525 unsigned mi = (lo + hi) / 2;
628522ec
JH
2526 int cmp = hashcmp(index + mi * stride, sha1);
2527
2528 if (debug_lookup)
2529 printf("lo %u hi %u rg %u mi %u\n",
2530 lo, hi, hi - lo, mi);
43057304 2531 if (!cmp)
74e34e1f 2532 return nth_packed_object_offset(p, mi);
1f688557
JH
2533 if (cmp > 0)
2534 hi = mi;
2535 else
2536 lo = mi+1;
2537 } while (lo < hi);
2538 return 0;
2539}
2540
4c080182 2541int is_pack_valid(struct packed_git *p)
d131b7af
SP
2542{
2543 /* An already open pack is known to be valid. */
2544 if (p->pack_fd != -1)
2545 return 1;
2546
2547 /* If the pack has one window completely covering the
2548 * file size, the pack is known to be valid even if
2549 * the descriptor is not currently open.
2550 */
2551 if (p->windows) {
2552 struct pack_window *w = p->windows;
2553
2554 if (!w->offset && w->len == p->pack_size)
2555 return 1;
2556 }
2557
2558 /* Force the pack to open to prove its valid. */
2559 return !open_packed_git(p);
2560}
2561
95099731
NTND
2562static int fill_pack_entry(const unsigned char *sha1,
2563 struct pack_entry *e,
2564 struct packed_git *p)
2565{
2566 off_t offset;
2567
2568 if (p->num_bad_objects) {
2569 unsigned i;
2570 for (i = 0; i < p->num_bad_objects; i++)
2571 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2572 return 0;
2573 }
2574
2575 offset = find_pack_entry_one(sha1, p);
2576 if (!offset)
2577 return 0;
2578
2579 /*
2580 * We are about to tell the caller where they can locate the
2581 * requested object. We better make sure the packfile is
2582 * still here and can be accessed before supplying that
2583 * answer, as it may have been deleted since the index was
2584 * loaded!
2585 */
319b678a 2586 if (!is_pack_valid(p))
95099731 2587 return 0;
95099731
NTND
2588 e->offset = offset;
2589 e->p = p;
2590 hashcpy(e->sha1, sha1);
2591 return 1;
2592}
2593
d40d535b
MH
2594/*
2595 * Iff a pack file contains the object named by sha1, return true and
2596 * store its location to e.
2597 */
4d6acb70 2598static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557
JH
2599{
2600 struct packed_git *p;
43057304 2601
1f688557 2602 prepare_packed_git();
f7c22cc6
NP
2603 if (!packed_git)
2604 return 0;
1f688557 2605
c01f51cc
NTND
2606 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2607 return 1;
8eca0b47 2608
c01f51cc 2609 for (p = packed_git; p; p = p->next) {
1b1005d1
MH
2610 if (p == last_found_pack)
2611 continue; /* we already checked this one */
f7c22cc6 2612
1b1005d1
MH
2613 if (fill_pack_entry(sha1, e, p)) {
2614 last_found_pack = p;
2615 return 1;
2616 }
c01f51cc 2617 }
1f688557
JH
2618 return 0;
2619}
2620
a6080a0a 2621struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
2622 struct packed_git *packs)
2623{
2624 struct packed_git *p;
bf592c50
DB
2625
2626 for (p = packs; p; p = p->next) {
43057304 2627 if (find_pack_entry_one(sha1, p))
bf592c50
DB
2628 return p;
2629 }
2630 return NULL;
21666f1a 2631
bf592c50
DB
2632}
2633
052fe5ea 2634static int sha1_loose_object_info(const unsigned char *sha1,
46f03448
KN
2635 struct object_info *oi,
2636 int flags)
65c2e0c3 2637{
46f03448
KN
2638 int status = 0;
2639 unsigned long mapsize;
65c2e0c3 2640 void *map;
ef49a7a0 2641 git_zstream stream;
d65a16f6 2642 char hdr[32];
46f03448 2643 struct strbuf hdrbuf = STRBUF_INIT;
65c2e0c3 2644
5d642e75
JK
2645 if (oi->delta_base_sha1)
2646 hashclr(oi->delta_base_sha1);
2647
052fe5ea
JK
2648 /*
2649 * If we don't care about type or size, then we don't
4ef8d1dd
JH
2650 * need to look inside the object at all. Note that we
2651 * do not optimize out the stat call, even if the
2652 * caller doesn't care about the disk-size, since our
2653 * return value implicitly indicates whether the
2654 * object even exists.
052fe5ea 2655 */
46f03448 2656 if (!oi->typep && !oi->typename && !oi->sizep) {
4ef8d1dd
JH
2657 struct stat st;
2658 if (stat_sha1_file(sha1, &st) < 0)
2659 return -1;
2660 if (oi->disk_sizep)
23c339c0 2661 *oi->disk_sizep = st.st_size;
052fe5ea
JK
2662 return 0;
2663 }
2664
bb6b8e4f 2665 map = map_sha1_file(sha1, &mapsize);
f0df4ed5 2666 if (!map)
dbea72a8 2667 return -1;
23c339c0
JK
2668 if (oi->disk_sizep)
2669 *oi->disk_sizep = mapsize;
46f03448
KN
2670 if ((flags & LOOKUP_UNKNOWN_OBJECT)) {
2671 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2672 status = error("unable to unpack %s header with --allow-unknown-type",
2673 sha1_to_hex(sha1));
2674 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
36e4d74a
JH
2675 status = error("unable to unpack %s header",
2676 sha1_to_hex(sha1));
46f03448
KN
2677 if (status < 0)
2678 ; /* Do nothing */
2679 else if (hdrbuf.len) {
2680 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2681 status = error("unable to parse %s header with --allow-unknown-type",
2682 sha1_to_hex(sha1));
2683 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
36e4d74a 2684 status = error("unable to parse %s header", sha1_to_hex(sha1));
39c68542 2685 git_inflate_end(&stream);
65c2e0c3 2686 munmap(map, mapsize);
46f03448 2687 if (status && oi->typep)
23c339c0 2688 *oi->typep = status;
46f03448 2689 strbuf_release(&hdrbuf);
052fe5ea 2690 return 0;
65c2e0c3
JH
2691}
2692
de7b5d62 2693int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
f0df4ed5 2694{
c4d9986f 2695 struct cached_object *co;
f0df4ed5 2696 struct pack_entry e;
5b086407 2697 int rtype;
46f03448 2698 enum object_type real_type;
1f7117ef 2699 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
f0df4ed5 2700
1f7117ef 2701 co = find_cached_object(real);
c4d9986f 2702 if (co) {
5b086407
JK
2703 if (oi->typep)
2704 *(oi->typep) = co->type;
9a490590
JH
2705 if (oi->sizep)
2706 *(oi->sizep) = co->size;
161f00e7
JK
2707 if (oi->disk_sizep)
2708 *(oi->disk_sizep) = 0;
5d642e75
JK
2709 if (oi->delta_base_sha1)
2710 hashclr(oi->delta_base_sha1);
46f03448
KN
2711 if (oi->typename)
2712 strbuf_addstr(oi->typename, typename(co->type));
9a490590 2713 oi->whence = OI_CACHED;
5b086407 2714 return 0;
c4d9986f
NTND
2715 }
2716
1f7117ef 2717 if (!find_pack_entry(real, &e)) {
ddd63e64 2718 /* Most likely it's a loose object. */
46f03448 2719 if (!sha1_loose_object_info(real, oi, flags)) {
9a490590 2720 oi->whence = OI_LOOSE;
5b086407 2721 return 0;
9a490590 2722 }
ddd63e64
SG
2723
2724 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 2725 reprepare_packed_git();
1f7117ef 2726 if (!find_pack_entry(real, &e))
052fe5ea 2727 return -1;
f0df4ed5 2728 }
3d77d877 2729
46f03448
KN
2730 /*
2731 * packed_object_info() does not follow the delta chain to
2732 * find out the real type, unless it is given oi->typep.
2733 */
2734 if (oi->typename && !oi->typep)
2735 oi->typep = &real_type;
2736
23c339c0 2737 rtype = packed_object_info(e.p, e.offset, oi);
412916ee 2738 if (rtype < 0) {
1f7117ef 2739 mark_bad_packed_object(e.p, real);
46f03448
KN
2740 if (oi->typep == &real_type)
2741 oi->typep = NULL;
1f7117ef 2742 return sha1_object_info_extended(real, oi, 0);
5266d369
JH
2743 } else if (in_delta_base_cache(e.p, e.offset)) {
2744 oi->whence = OI_DBCACHED;
9a490590
JH
2745 } else {
2746 oi->whence = OI_PACKED;
2747 oi->u.packed.offset = e.offset;
2748 oi->u.packed.pack = e.p;
2749 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2750 rtype == OBJ_OFS_DELTA);
3d77d877 2751 }
46f03448
KN
2752 if (oi->typename)
2753 strbuf_addstr(oi->typename, typename(*oi->typep));
2754 if (oi->typep == &real_type)
2755 oi->typep = NULL;
3d77d877 2756
5b086407 2757 return 0;
f0df4ed5
JS
2758}
2759
3fc0dca9 2760/* returns enum object_type or negative */
9a490590
JH
2761int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2762{
5b086407 2763 enum object_type type;
d099b717 2764 struct object_info oi = {NULL};
9a490590 2765
5b086407 2766 oi.typep = &type;
9a490590 2767 oi.sizep = sizep;
de7b5d62 2768 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
5b086407
JK
2769 return -1;
2770 return type;
9a490590
JH
2771}
2772
21666f1a
NP
2773static void *read_packed_sha1(const unsigned char *sha1,
2774 enum object_type *type, unsigned long *size)
1f688557
JH
2775{
2776 struct pack_entry e;
8eca0b47 2777 void *data;
1f688557 2778
cd673c1f 2779 if (!find_pack_entry(sha1, &e))
1f688557 2780 return NULL;
8eca0b47
NP
2781 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2782 if (!data) {
2783 /*
2784 * We're probably in deep shit, but let's try to fetch
2785 * the required object anyway from another pack or loose.
2786 * This should happen only in the presence of a corrupted
2787 * pack, and is better than failing outright.
2788 */
2789 error("failed to read object %s at offset %"PRIuMAX" from %s",
2790 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2791 mark_bad_packed_object(e.p, sha1);
ac939109 2792 data = read_object(sha1, type, size);
8eca0b47
NP
2793 }
2794 return data;
1f688557
JH
2795}
2796
21666f1a
NP
2797int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2798 unsigned char *sha1)
d66b37bb
JH
2799{
2800 struct cached_object *co;
2801
21666f1a 2802 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2803 if (has_sha1_file(sha1) || find_cached_object(sha1))
2804 return 0;
c7353967 2805 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
d66b37bb
JH
2806 co = &cached_objects[cached_object_nr++];
2807 co->size = len;
21666f1a 2808 co->type = type;
efa13f7b
JH
2809 co->buf = xmalloc(len);
2810 memcpy(co->buf, buf, len);
d66b37bb
JH
2811 hashcpy(co->sha1, sha1);
2812 return 0;
2813}
2814
c2c5b270
CC
2815static void *read_object(const unsigned char *sha1, enum object_type *type,
2816 unsigned long *size)
0fcfd160
LT
2817{
2818 unsigned long mapsize;
2819 void *map, *buf;
d66b37bb
JH
2820 struct cached_object *co;
2821
2822 co = find_cached_object(sha1);
2823 if (co) {
21666f1a 2824 *type = co->type;
d66b37bb 2825 *size = co->size;
182af834 2826 return xmemdupz(co->buf, co->size);
d66b37bb 2827 }
0fcfd160 2828
8276c007
PE
2829 buf = read_packed_sha1(sha1, type, size);
2830 if (buf)
2831 return buf;
bb6b8e4f 2832 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2833 if (map) {
7efbff75 2834 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2835 munmap(map, mapsize);
2836 return buf;
2837 }
637cdd9d 2838 reprepare_packed_git();
8276c007 2839 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2840}
2841
b6c4cecc
JH
2842/*
2843 * This function dies on corrupt objects; the callers who want to
2844 * deal with them should arrange to call read_object() and give error
2845 * messages themselves.
2846 */
5bf29b95
JH
2847void *read_sha1_file_extended(const unsigned char *sha1,
2848 enum object_type *type,
2849 unsigned long *size,
2850 unsigned flag)
ac939109 2851{
3ba7a065 2852 void *data;
b6c4cecc 2853 const struct packed_git *p;
bf93eea0 2854 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
b6c4cecc 2855
3ba7a065
JH
2856 errno = 0;
2857 data = read_object(repl, type, size);
4bbf5a26 2858 if (data)
b6c4cecc 2859 return data;
68095570 2860
25f3af3f 2861 if (errno && errno != ENOENT)
3ba7a065
JH
2862 die_errno("failed to read object %s", sha1_to_hex(sha1));
2863
68095570 2864 /* die if we replaced an object with one that does not exist */
b6c4cecc 2865 if (repl != sha1)
68095570
CC
2866 die("replacement %s not found for %s",
2867 sha1_to_hex(repl), sha1_to_hex(sha1));
2868
b6c4cecc 2869 if (has_loose_object(repl)) {
30d6c6ea
MH
2870 const char *path = sha1_file_name(sha1);
2871
b6c4cecc
JH
2872 die("loose object %s (stored in %s) is corrupt",
2873 sha1_to_hex(repl), path);
e8b15e61 2874 }
68095570 2875
b6c4cecc
JH
2876 if ((p = has_packed_and_bad(repl)) != NULL)
2877 die("packed object %s (stored in %s) is corrupt",
2878 sha1_to_hex(repl), p->pack_name);
f5552aee 2879
b6c4cecc 2880 return NULL;
ac939109
NP
2881}
2882
40469ee9 2883void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2884 const char *required_type_name,
40469ee9
JH
2885 unsigned long *size,
2886 unsigned char *actual_sha1_return)
f4913f91 2887{
21666f1a 2888 enum object_type type, required_type;
f4913f91
JH
2889 void *buffer;
2890 unsigned long isize;
40469ee9 2891 unsigned char actual_sha1[20];
f4913f91 2892
21666f1a 2893 required_type = type_from_string(required_type_name);
e702496e 2894 hashcpy(actual_sha1, sha1);
40469ee9
JH
2895 while (1) {
2896 int ref_length = -1;
2897 const char *ref_type = NULL;
f4913f91 2898
21666f1a 2899 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2900 if (!buffer)
2901 return NULL;
21666f1a 2902 if (type == required_type) {
40469ee9
JH
2903 *size = isize;
2904 if (actual_sha1_return)
e702496e 2905 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2906 return buffer;
2907 }
2908 /* Handle references */
21666f1a 2909 else if (type == OBJ_COMMIT)
40469ee9 2910 ref_type = "tree ";
21666f1a 2911 else if (type == OBJ_TAG)
40469ee9
JH
2912 ref_type = "object ";
2913 else {
2914 free(buffer);
2915 return NULL;
2916 }
2917 ref_length = strlen(ref_type);
f4913f91 2918
50974ec9
MK
2919 if (ref_length + 40 > isize ||
2920 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2921 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2922 free(buffer);
2923 return NULL;
2924 }
1cf58e72 2925 free(buffer);
40469ee9
JH
2926 /* Now we have the ID of the referred-to object in
2927 * actual_sha1. Check again. */
f4913f91 2928 }
f4913f91
JH
2929}
2930
ce9fbf16 2931static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2932 const char *type, unsigned char *sha1,
d65a16f6 2933 char *hdr, int *hdrlen)
d410c0f5 2934{
9126f009 2935 git_SHA_CTX c;
d410c0f5
JH
2936
2937 /* Generate the header */
ef1286d3 2938 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
d410c0f5
JH
2939
2940 /* Sha1.. */
9126f009
NP
2941 git_SHA1_Init(&c);
2942 git_SHA1_Update(&c, hdr, *hdrlen);
2943 git_SHA1_Update(&c, buf, len);
2944 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2945}
2946
230f1322 2947/*
5a688fe4 2948 * Move the just written object into its final resting place.
230f1322 2949 */
cb5add58 2950int finalize_object_file(const char *tmpfile, const char *filename)
230f1322 2951{
e32c0a9c 2952 int ret = 0;
5a688fe4 2953
348df166 2954 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
2955 goto try_rename;
2956 else if (link(tmpfile, filename))
e32c0a9c 2957 ret = errno;
7ebb6fca
LT
2958
2959 /*
2960 * Coda hack - coda doesn't like cross-directory links,
2961 * so we fall back to a rename, which will mean that it
2962 * won't be able to check collisions, but that's not a
2963 * big deal.
2964 *
2965 * The same holds for FAT formatted media.
2966 *
3be1f18e 2967 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
2968 * left to unlink.
2969 */
2970 if (ret && ret != EEXIST) {
be66a6c4 2971 try_rename:
7ebb6fca 2972 if (!rename(tmpfile, filename))
3be1f18e 2973 goto out;
9e48b389 2974 ret = errno;
230f1322 2975 }
691f1a28 2976 unlink_or_warn(tmpfile);
230f1322
LT
2977 if (ret) {
2978 if (ret != EEXIST) {
82247e9b 2979 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
230f1322
LT
2980 }
2981 /* FIXME!!! Collision check here ? */
2982 }
2983
3be1f18e 2984out:
5256b006 2985 if (adjust_shared_perm(filename))
5a688fe4 2986 return error("unable to set permission to '%s'", filename);
230f1322
LT
2987 return 0;
2988}
2989
4d548150
LT
2990static int write_buffer(int fd, const void *buf, size_t len)
2991{
d34cf19b 2992 if (write_in_full(fd, buf, len) < 0)
93822c22 2993 return error("file write error (%s)", strerror(errno));
4d548150
LT
2994 return 0;
2995}
2996
ce9fbf16 2997int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2998 unsigned char *sha1)
2999{
d65a16f6 3000 char hdr[32];
ef1286d3 3001 int hdrlen = sizeof(hdr);
abdc3fc8
RS
3002 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3003 return 0;
3004}
3005
e9039dd3
LT
3006/* Finalize a file on disk, and close it. */
3007static void close_sha1_file(int fd)
3008{
aafe9fba
LT
3009 if (fsync_object_files)
3010 fsync_or_die(fd, "sha1 file");
e9039dd3 3011 if (close(fd) != 0)
d824cbba 3012 die_errno("error when closing sha1 file");
e9039dd3
LT
3013}
3014
5723fe7e
LT
3015/* Size of directory component, including the ending '/' */
3016static inline int directory_size(const char *filename)
3017{
3018 const char *s = strrchr(filename, '/');
3019 if (!s)
3020 return 0;
3021 return s - filename + 1;
3022}
3023
3024/*
3025 * This creates a temporary file in the same directory as the final
3026 * 'filename'
3027 *
3028 * We want to avoid cross-directory filename renames, because those
3029 * can have problems on various filesystems (FAT, NFS, Coda).
3030 */
3031static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
3032{
3033 int fd, dirlen = directory_size(filename);
3034
3035 if (dirlen + 20 > bufsiz) {
3036 errno = ENAMETOOLONG;
3037 return -1;
3038 }
3039 memcpy(buffer, filename, dirlen);
3040 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
5256b006 3041 fd = git_mkstemp_mode(buffer, 0444);
cbacbf4e 3042 if (fd < 0 && dirlen && errno == ENOENT) {
5723fe7e 3043 /* Make sure the directory exists */
6ff6af62 3044 memcpy(buffer, filename, dirlen);
5723fe7e 3045 buffer[dirlen-1] = 0;
b2476a60
JH
3046 if (mkdir(buffer, 0777) && errno != EEXIST)
3047 return -1;
3048 if (adjust_shared_perm(buffer))
5723fe7e
LT
3049 return -1;
3050
3051 /* Try again */
3052 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
5256b006 3053 fd = git_mkstemp_mode(buffer, 0444);
5723fe7e
LT
3054 }
3055 return fd;
3056}
3057
bbac7311 3058static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 3059 const void *buf, unsigned long len, time_t mtime)
0fcfd160 3060{
915308b1 3061 int fd, ret;
9892beba 3062 unsigned char compressed[4096];
ef49a7a0 3063 git_zstream stream;
748af44c
NP
3064 git_SHA_CTX c;
3065 unsigned char parano_sha1[20];
ab1900a3 3066 static char tmp_file[PATH_MAX];
30d6c6ea 3067 const char *filename = sha1_file_name(sha1);
a44c9a5e 3068
ab1900a3 3069 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
aac17941 3070 if (fd < 0) {
35243577 3071 if (errno == EACCES)
82247e9b 3072 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
916d081b 3073 else
5eaeda70 3074 return error("unable to create temporary file: %s", strerror(errno));
aac17941
LT
3075 }
3076
0fcfd160 3077 /* Set it up */
55bb5c91 3078 git_deflate_init(&stream, zlib_compression_level);
0fcfd160 3079 stream.next_out = compressed;
9892beba 3080 stream.avail_out = sizeof(compressed);
748af44c 3081 git_SHA1_Init(&c);
a44c9a5e
LT
3082
3083 /* First header.. */
d65a16f6 3084 stream.next_in = (unsigned char *)hdr;
a44c9a5e 3085 stream.avail_in = hdrlen;
55bb5c91
JH
3086 while (git_deflate(&stream, 0) == Z_OK)
3087 ; /* nothing */
748af44c 3088 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
3089
3090 /* Then the data itself.. */
c00e657d 3091 stream.next_in = (void *)buf;
a44c9a5e 3092 stream.avail_in = len;
9892beba 3093 do {
748af44c 3094 unsigned char *in0 = stream.next_in;
55bb5c91 3095 ret = git_deflate(&stream, Z_FINISH);
748af44c 3096 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
3097 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3098 die("unable to write sha1 file");
3099 stream.next_out = compressed;
3100 stream.avail_out = sizeof(compressed);
3101 } while (ret == Z_OK);
3102
ac54c277
LT
3103 if (ret != Z_STREAM_END)
3104 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
55bb5c91 3105 ret = git_deflate_end_gently(&stream);
ac54c277
LT
3106 if (ret != Z_OK)
3107 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
3108 git_SHA1_Final(parano_sha1, &c);
3109 if (hashcmp(sha1, parano_sha1) != 0)
3110 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 3111
e9039dd3 3112 close_sha1_file(fd);
0fcfd160 3113
bbac7311
NP
3114 if (mtime) {
3115 struct utimbuf utb;
3116 utb.actime = mtime;
3117 utb.modtime = mtime;
ab1900a3 3118 if (utime(tmp_file, &utb) < 0)
bbac7311 3119 warning("failed utime() on %s: %s",
ab1900a3 3120 tmp_file, strerror(errno));
bbac7311
NP
3121 }
3122
cb5add58 3123 return finalize_object_file(tmp_file, filename);
0fcfd160 3124}
8237b185 3125
33d4221c
JK
3126static int freshen_loose_object(const unsigned char *sha1)
3127{
3128 return check_and_freshen(sha1, 1);
3129}
3130
3131static int freshen_packed_object(const unsigned char *sha1)
3132{
3133 struct pack_entry e;
ee1c6c34
JK
3134 if (!find_pack_entry(sha1, &e))
3135 return 0;
3136 if (e.p->freshened)
3137 return 1;
3138 if (!freshen_file(e.p->pack_name))
3139 return 0;
3140 e.p->freshened = 1;
3141 return 1;
33d4221c
JK
3142}
3143
1427a7ff 3144int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
bbac7311 3145{
bbac7311 3146 char hdr[32];
ef1286d3 3147 int hdrlen = sizeof(hdr);
bbac7311
NP
3148
3149 /* Normally if we have it in the pack then we do not bother writing
3150 * it out into .git/objects/??/?{38} file.
3151 */
3152 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
b5f52f37 3153 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
bbac7311
NP
3154 return 0;
3155 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3156}
3157
0c3db67c
ES
3158int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3159 unsigned char *sha1, unsigned flags)
3160{
3161 char *header;
3162 int hdrlen, status = 0;
3163
3164 /* type string, SP, %lu of the length plus NUL must fit this */
ef1286d3
JK
3165 hdrlen = strlen(type) + 32;
3166 header = xmalloc(hdrlen);
0c3db67c
ES
3167 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3168
3169 if (!(flags & HASH_WRITE_OBJECT))
3170 goto cleanup;
051086b9 3171 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
0c3db67c
ES
3172 goto cleanup;
3173 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3174
3175cleanup:
3176 free(header);
3177 return status;
3178}
3179
bbac7311
NP
3180int force_object_loose(const unsigned char *sha1, time_t mtime)
3181{
bbac7311
NP
3182 void *buf;
3183 unsigned long len;
3184 enum object_type type;
3185 char hdr[32];
3186 int hdrlen;
1fb23e65 3187 int ret;
bbac7311 3188
c529d75a 3189 if (has_loose_object(sha1))
bbac7311
NP
3190 return 0;
3191 buf = read_packed_sha1(sha1, &type, &len);
3192 if (!buf)
3193 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
ef1286d3 3194 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1fb23e65
BS
3195 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3196 free(buf);
3197
3198 return ret;
bbac7311
NP
3199}
3200
bf592c50
DB
3201int has_pack_index(const unsigned char *sha1)
3202{
3203 struct stat st;
3204 if (stat(sha1_pack_index_name(sha1), &st))
3205 return 0;
3206 return 1;
bf592c50
DB
3207}
3208
cd673c1f
JH
3209int has_sha1_pack(const unsigned char *sha1)
3210{
3211 struct pack_entry e;
3212 return find_pack_entry(sha1, &e);
3213}
3214
0eeb077b 3215int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
8237b185 3216{
1f688557
JH
3217 struct pack_entry e;
3218
cd673c1f 3219 if (find_pack_entry(sha1, &e))
1f688557 3220 return 1;
45e8a748
JK
3221 if (has_loose_object(sha1))
3222 return 1;
0eeb077b
JK
3223 if (flags & HAS_SHA1_QUICK)
3224 return 0;
45e8a748
JK
3225 reprepare_packed_git();
3226 return find_pack_entry(sha1, &e);
8237b185 3227}
74400e71 3228
c879daa2
NTND
3229static void check_tree(const void *buf, size_t size)
3230{
3231 struct tree_desc desc;
3232 struct name_entry entry;
3233
3234 init_tree_desc(&desc, buf, size);
3235 while (tree_entry(&desc, &entry))
3236 /* do nothing
3237 * tree_entry() will die() on malformed entries */
3238 ;
3239}
3240
3241static void check_commit(const void *buf, size_t size)
3242{
3243 struct commit c;
3244 memset(&c, 0, sizeof(c));
3245 if (parse_commit_buffer(&c, buf, size))
3246 die("corrupt commit");
3247}
3248
3249static void check_tag(const void *buf, size_t size)
3250{
3251 struct tag t;
3252 memset(&t, 0, sizeof(t));
3253 if (parse_tag_buffer(&t, buf, size))
3254 die("corrupt tag");
3255}
3256
43df4f86 3257static int index_mem(unsigned char *sha1, void *buf, size_t size,
c4ce46fc
JH
3258 enum object_type type,
3259 const char *path, unsigned flags)
e7332f96 3260{
6c510bee 3261 int ret, re_allocated = 0;
c4ce46fc 3262 int write_object = flags & HASH_WRITE_OBJECT;
74400e71 3263
7672db20 3264 if (!type)
edaec3fb 3265 type = OBJ_BLOB;
6c510bee
LT
3266
3267 /*
3268 * Convert blobs to git internal format
3269 */
43df4f86 3270 if ((type == OBJ_BLOB) && path) {
f285a2d7 3271 struct strbuf nbuf = STRBUF_INIT;
21e5ad50 3272 if (convert_to_git(path, buf, size, &nbuf,
5e12e78e 3273 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
b315c5c0 3274 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
3275 re_allocated = 1;
3276 }
3277 }
c4ce46fc 3278 if (flags & HASH_FORMAT_CHECK) {
c879daa2
NTND
3279 if (type == OBJ_TREE)
3280 check_tree(buf, size);
3281 if (type == OBJ_COMMIT)
3282 check_commit(buf, size);
3283 if (type == OBJ_TAG)
3284 check_tag(buf, size);
3285 }
6c510bee 3286
7672db20 3287 if (write_object)
edaec3fb 3288 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 3289 else
edaec3fb 3290 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 3291 if (re_allocated)
6c510bee 3292 free(buf);
43df4f86
DP
3293 return ret;
3294}
3295
9035d75a
SP
3296static int index_stream_convert_blob(unsigned char *sha1, int fd,
3297 const char *path, unsigned flags)
3298{
3299 int ret;
3300 const int write_object = flags & HASH_WRITE_OBJECT;
3301 struct strbuf sbuf = STRBUF_INIT;
3302
3303 assert(path);
3304 assert(would_convert_to_git_filter_fd(path));
3305
3306 convert_to_git_filter_fd(path, fd, &sbuf,
3307 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3308
3309 if (write_object)
3310 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3311 sha1);
3312 else
3313 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3314 sha1);
3315 strbuf_release(&sbuf);
3316 return ret;
3317}
3318
7b41e1e1
JH
3319static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3320 const char *path, unsigned flags)
3321{
3322 struct strbuf sbuf = STRBUF_INIT;
3323 int ret;
3324
3325 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3326 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3327 else
3328 ret = -1;
3329 strbuf_release(&sbuf);
3330 return ret;
3331}
3332
ea68b0ce
DP
3333#define SMALL_FILE_SIZE (32*1024)
3334
7b41e1e1
JH
3335static int index_core(unsigned char *sha1, int fd, size_t size,
3336 enum object_type type, const char *path,
3337 unsigned flags)
43df4f86
DP
3338{
3339 int ret;
43df4f86 3340
7b41e1e1 3341 if (!size) {
f6a1e1e2 3342 ret = index_mem(sha1, "", size, type, path, flags);
ea68b0ce
DP
3343 } else if (size <= SMALL_FILE_SIZE) {
3344 char *buf = xmalloc(size);
3345 if (size == read_in_full(fd, buf, size))
c4ce46fc 3346 ret = index_mem(sha1, buf, size, type, path, flags);
ea68b0ce
DP
3347 else
3348 ret = error("short read %s", strerror(errno));
3349 free(buf);
08bda208 3350 } else {
43df4f86 3351 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
c4ce46fc 3352 ret = index_mem(sha1, buf, size, type, path, flags);
aac17941 3353 munmap(buf, size);
08bda208 3354 }
7b41e1e1
JH
3355 return ret;
3356}
3357
4dd1fbc7 3358/*
568508e7
JH
3359 * This creates one packfile per large blob unless bulk-checkin
3360 * machinery is "plugged".
4dd1fbc7
JH
3361 *
3362 * This also bypasses the usual "convert-to-git" dance, and that is on
3363 * purpose. We could write a streaming version of the converting
3364 * functions and insert that before feeding the data to fast-import
4f22b101
JK
3365 * (or equivalent in-core API described above). However, that is
3366 * somewhat complicated, as we do not know the size of the filter
3367 * result, which we need to know beforehand when writing a git object.
3368 * Since the primary motivation for trying to stream from the working
3369 * tree file and to avoid mmaping it in core is to deal with large
3370 * binary blobs, they generally do not want to get any conversion, and
3371 * callers should avoid this code path when filters are requested.
4dd1fbc7
JH
3372 */
3373static int index_stream(unsigned char *sha1, int fd, size_t size,
3374 enum object_type type, const char *path,
3375 unsigned flags)
3376{
568508e7 3377 return index_bulk_checkin(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
3378}
3379
7b41e1e1
JH
3380int index_fd(unsigned char *sha1, int fd, struct stat *st,
3381 enum object_type type, const char *path, unsigned flags)
3382{
3383 int ret;
7b41e1e1 3384
9079ab7c
SP
3385 /*
3386 * Call xsize_t() only when needed to avoid potentially unnecessary
3387 * die() for large files.
3388 */
9035d75a
SP
3389 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3390 ret = index_stream_convert_blob(sha1, fd, path, flags);
3391 else if (!S_ISREG(st->st_mode))
7b41e1e1 3392 ret = index_pipe(sha1, fd, type, path, flags);
9079ab7c 3393 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
7ce7c760 3394 (path && would_convert_to_git(path)))
9079ab7c
SP
3395 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3396 flags);
4dd1fbc7 3397 else
9079ab7c
SP
3398 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3399 flags);
43df4f86 3400 close(fd);
aac17941 3401 return ret;
74400e71 3402}
ec1fcc16 3403
c4ce46fc 3404int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
ec1fcc16
JH
3405{
3406 int fd;
b760d3aa 3407 struct strbuf sb = STRBUF_INIT;
ec1fcc16
JH
3408
3409 switch (st->st_mode & S_IFMT) {
3410 case S_IFREG:
3411 fd = open(path, O_RDONLY);
3412 if (fd < 0)
3413 return error("open(\"%s\"): %s", path,
3414 strerror(errno));
c4ce46fc 3415 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
ec1fcc16
JH
3416 return error("%s: failed to insert into database",
3417 path);
3418 break;
3419 case S_IFLNK:
b760d3aa 3420 if (strbuf_readlink(&sb, path, st->st_size)) {
ec1fcc16 3421 char *errstr = strerror(errno);
ec1fcc16
JH
3422 return error("readlink(\"%s\"): %s", path,
3423 errstr);
3424 }
c4ce46fc 3425 if (!(flags & HASH_WRITE_OBJECT))
b760d3aa
LT
3426 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3427 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
ec1fcc16
JH
3428 return error("%s: failed to insert into database",
3429 path);
b760d3aa 3430 strbuf_release(&sb);
ec1fcc16 3431 break;
f35a6d3b
LT
3432 case S_IFDIR:
3433 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
3434 default:
3435 return error("%s: unsupported file type", path);
3436 }
3437 return 0;
3438}
a69e5429
JH
3439
3440int read_pack_header(int fd, struct pack_header *header)
3441{
c697ad14
HO
3442 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3443 /* "eof before pack header was fully read" */
3444 return PH_ERROR_EOF;
3445
a69e5429
JH
3446 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3447 /* "protocol error (pack signature mismatch detected)" */
3448 return PH_ERROR_PACK_SIGNATURE;
3449 if (!pack_version_ok(header->hdr_version))
3450 /* "protocol error (pack version unsupported)" */
3451 return PH_ERROR_PROTOCOL;
3452 return 0;
3453}
40d52ff7
JK
3454
3455void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3456{
3457 enum object_type type = sha1_object_info(sha1, NULL);
3458 if (type < 0)
3459 die("%s is not a valid object", sha1_to_hex(sha1));
3460 if (type != expect)
3461 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3462 typename(expect));
3463}
27e1e22d
JK
3464
3465static int for_each_file_in_obj_subdir(int subdir_nr,
3466 struct strbuf *path,
3467 each_loose_object_fn obj_cb,
3468 each_loose_cruft_fn cruft_cb,
3469 each_loose_subdir_fn subdir_cb,
3470 void *data)
3471{
3472 size_t baselen = path->len;
3473 DIR *dir = opendir(path->buf);
3474 struct dirent *de;
3475 int r = 0;
3476
3477 if (!dir) {
3478 if (errno == ENOENT)
3479 return 0;
3480 return error("unable to open %s: %s", path->buf, strerror(errno));
3481 }
3482
3483 while ((de = readdir(dir))) {
3484 if (is_dot_or_dotdot(de->d_name))
3485 continue;
3486
3487 strbuf_setlen(path, baselen);
3488 strbuf_addf(path, "/%s", de->d_name);
3489
3490 if (strlen(de->d_name) == 38) {
3491 char hex[41];
3492 unsigned char sha1[20];
3493
3494 snprintf(hex, sizeof(hex), "%02x%s",
3495 subdir_nr, de->d_name);
3496 if (!get_sha1_hex(hex, sha1)) {
3497 if (obj_cb) {
3498 r = obj_cb(sha1, path->buf, data);
3499 if (r)
3500 break;
3501 }
3502 continue;
3503 }
3504 }
3505
3506 if (cruft_cb) {
3507 r = cruft_cb(de->d_name, path->buf, data);
3508 if (r)
3509 break;
3510 }
3511 }
3512 strbuf_setlen(path, baselen);
3513
3514 if (!r && subdir_cb)
3515 r = subdir_cb(subdir_nr, path->buf, data);
3516
3517 closedir(dir);
3518 return r;
3519}
3520
e6f875e0 3521int for_each_loose_file_in_objdir_buf(struct strbuf *path,
27e1e22d
JK
3522 each_loose_object_fn obj_cb,
3523 each_loose_cruft_fn cruft_cb,
3524 each_loose_subdir_fn subdir_cb,
3525 void *data)
3526{
e6f875e0 3527 size_t baselen = path->len;
27e1e22d
JK
3528 int r = 0;
3529 int i;
3530
27e1e22d 3531 for (i = 0; i < 256; i++) {
e6f875e0
JK
3532 strbuf_addf(path, "/%02x", i);
3533 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
27e1e22d 3534 subdir_cb, data);
e6f875e0 3535 strbuf_setlen(path, baselen);
27e1e22d
JK
3536 if (r)
3537 break;
3538 }
3539
e6f875e0
JK
3540 return r;
3541}
3542
3543int for_each_loose_file_in_objdir(const char *path,
3544 each_loose_object_fn obj_cb,
3545 each_loose_cruft_fn cruft_cb,
3546 each_loose_subdir_fn subdir_cb,
3547 void *data)
3548{
3549 struct strbuf buf = STRBUF_INIT;
3550 int r;
3551
3552 strbuf_addstr(&buf, path);
3553 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3554 subdir_cb, data);
27e1e22d 3555 strbuf_release(&buf);
e6f875e0 3556