memory_limit: use git_env_ulong() to parse GIT_ALLOC_LIMIT
[git/git.git] / sha1_file.c
CommitLineData
0fcfd160
LT
1/*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
0fcfd160 9#include "cache.h"
6eac50d8 10#include "string-list.h"
1f688557 11#include "delta.h"
a733cb60 12#include "pack.h"
8e440259
PE
13#include "blob.h"
14#include "commit.h"
4dd1fbc7 15#include "run-command.h"
8e440259
PE
16#include "tag.h"
17#include "tree.h"
c879daa2 18#include "tree-walk.h"
f35a6d3b 19#include "refs.h"
70f5d5d3 20#include "pack-revindex.h"
628522ec 21#include "sha1-lookup.h"
568508e7 22#include "bulk-checkin.h"
090ea126 23#include "streaming.h"
543c5caa 24#include "dir.h"
0fcfd160 25
144bde78
LT
26#ifndef O_NOATIME
27#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
28#define O_NOATIME 01000000
29#else
30#define O_NOATIME 0
31#endif
32#endif
33
28bd70d8
JN
34#define SZ_FMT PRIuMAX
35static inline uintmax_t sz_fmt(size_t s) { return s; }
e05db0fd 36
96f1e58f 37const unsigned char null_sha1[20];
88cd621d 38
c597ba80
NTND
39/*
40 * This is meant to hold a *small* number of objects that you would
41 * want read_sha1_file() to be able to return, but yet you do not want
42 * to write them into the object store (e.g. a browse-only
43 * application).
44 */
45static struct cached_object {
46 unsigned char sha1[20];
47 enum object_type type;
48 void *buf;
49 unsigned long size;
50} *cached_objects;
51static int cached_object_nr, cached_object_alloc;
52
53static struct cached_object empty_tree = {
dab0d410 54 EMPTY_TREE_SHA1_BIN_LITERAL,
c597ba80
NTND
55 OBJ_TREE,
56 "",
57 0
58};
59
1b1005d1
MH
60/*
61 * A pointer to the last packed_git in which an object was found.
62 * When an object is sought, we look in this packfile first, because
63 * objects that are looked up at similar times are often in the same
64 * packfile as one another.
65 */
c01f51cc
NTND
66static struct packed_git *last_found_pack;
67
c597ba80
NTND
68static struct cached_object *find_cached_object(const unsigned char *sha1)
69{
70 int i;
71 struct cached_object *co = cached_objects;
72
73 for (i = 0; i < cached_object_nr; i++, co++) {
74 if (!hashcmp(co->sha1, sha1))
75 return co;
76 }
77 if (!hashcmp(sha1, empty_tree.sha1))
78 return &empty_tree;
79 return NULL;
80}
81
90a6464b
JH
82int mkdir_in_gitdir(const char *path)
83{
84 if (mkdir(path, 0777)) {
85 int saved_errno = errno;
86 struct stat st;
87 struct strbuf sb = STRBUF_INIT;
88
89 if (errno != EEXIST)
90 return -1;
91 /*
92 * Are we looking at a path in a symlinked worktree
93 * whose original repository does not yet have it?
94 * e.g. .git/rr-cache pointing at its original
95 * repository in which the user hasn't performed any
96 * conflict resolution yet?
97 */
98 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
99 strbuf_readlink(&sb, path, st.st_size) ||
100 !is_absolute_path(sb.buf) ||
101 mkdir(sb.buf, 0777)) {
102 strbuf_release(&sb);
103 errno = saved_errno;
104 return -1;
105 }
106 strbuf_release(&sb);
107 }
108 return adjust_shared_perm(path);
109}
110
0be0521b 111enum scld_error safe_create_leading_directories(char *path)
b2cb9425 112{
26c8ae2a 113 char *next_component = path + offset_1st_component(path);
0be0521b 114 enum scld_error ret = SCLD_OK;
67d42212 115
0be0521b 116 while (ret == SCLD_OK && next_component) {
f0502332 117 struct stat st;
0f527403 118 char *slash = next_component, slash_character;
f0502332 119
0f527403
MH
120 while (*slash && !is_dir_sep(*slash))
121 slash++;
122
123 if (!*slash)
b2cb9425 124 break;
bf10cf70 125
26c8ae2a 126 next_component = slash + 1;
0f527403 127 while (is_dir_sep(*next_component))
bf10cf70 128 next_component++;
26c8ae2a 129 if (!*next_component)
5f0bdf50 130 break;
831651fd 131
0f527403 132 slash_character = *slash;
831651fd 133 *slash = '\0';
67d42212
JR
134 if (!stat(path, &st)) {
135 /* path exists */
9e6f885d 136 if (!S_ISDIR(st.st_mode))
0be0521b 137 ret = SCLD_EXISTS;
53a39721 138 } else if (mkdir(path, 0777)) {
928734d9 139 if (errno == EEXIST &&
9e6f885d 140 !stat(path, &st) && S_ISDIR(st.st_mode))
928734d9 141 ; /* somebody created it since we checked */
18d37e86
MH
142 else if (errno == ENOENT)
143 /*
144 * Either mkdir() failed because
145 * somebody just pruned the containing
146 * directory, or stat() failed because
147 * the file that was in our way was
148 * just removed. Either way, inform
149 * the caller that it might be worth
150 * trying again:
151 */
152 ret = SCLD_VANISHED;
9e6f885d 153 else
0be0521b 154 ret = SCLD_FAILED;
53a39721 155 } else if (adjust_shared_perm(path)) {
0be0521b 156 ret = SCLD_PERMS;
457f06d6 157 }
0f527403 158 *slash = slash_character;
b2cb9425 159 }
9e6f885d 160 return ret;
b2cb9425 161}
723c31fe 162
0be0521b 163enum scld_error safe_create_leading_directories_const(const char *path)
8e21d63b
JK
164{
165 /* path points to cache entries, so xstrdup before messing with it */
166 char *buf = xstrdup(path);
0be0521b 167 enum scld_error result = safe_create_leading_directories(buf);
8e21d63b
JK
168 free(buf);
169 return result;
170}
171
ace1534d
JH
172static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
173{
174 int i;
175 for (i = 0; i < 20; i++) {
176 static char hex[] = "0123456789abcdef";
177 unsigned int val = sha1[i];
178 char *pos = pathbuf + i*2 + (i > 0);
179 *pos++ = hex[val >> 4];
180 *pos = hex[val & 0xf];
181 }
182}
183
30d6c6ea 184const char *sha1_file_name(const unsigned char *sha1)
0fcfd160 185{
560fb6a1
JK
186 static char buf[PATH_MAX];
187 const char *objdir;
188 int len;
0fcfd160 189
560fb6a1
JK
190 objdir = get_object_directory();
191 len = strlen(objdir);
192
193 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
194 if (len + 43 > PATH_MAX)
195 die("insanely long object directory %s", objdir);
196 memcpy(buf, objdir, len);
197 buf[len] = '/';
198 buf[len+3] = '/';
199 buf[len+42] = '\0';
200 fill_sha1_path(buf + len + 1, sha1);
201 return buf;
0fcfd160
LT
202}
203
d40d535b
MH
204/*
205 * Return the name of the pack or index file with the specified sha1
206 * in its filename. *base and *name are scratch space that must be
207 * provided by the caller. which should be "pack" or "idx".
208 */
633f43e1 209static char *sha1_get_pack_name(const unsigned char *sha1,
6eec46bd 210 char **name, char **base, const char *which)
bf592c50
DB
211{
212 static const char hex[] = "0123456789abcdef";
633f43e1 213 char *buf;
bf592c50
DB
214 int i;
215
633f43e1 216 if (!*base) {
bf592c50
DB
217 const char *sha1_file_directory = get_object_directory();
218 int len = strlen(sha1_file_directory);
633f43e1 219 *base = xmalloc(len + 60);
6eec46bd
JH
220 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
221 sha1_file_directory, which);
633f43e1 222 *name = *base + len + 11;
bf592c50
DB
223 }
224
633f43e1 225 buf = *name;
bf592c50
DB
226
227 for (i = 0; i < 20; i++) {
228 unsigned int val = *sha1++;
229 *buf++ = hex[val >> 4];
230 *buf++ = hex[val & 0xf];
231 }
a6080a0a 232
633f43e1 233 return *base;
bf592c50
DB
234}
235
633f43e1 236char *sha1_pack_name(const unsigned char *sha1)
bf592c50 237{
633f43e1 238 static char *name, *base;
bf592c50 239
6eec46bd 240 return sha1_get_pack_name(sha1, &name, &base, "pack");
633f43e1 241}
bf592c50 242
633f43e1
HO
243char *sha1_pack_index_name(const unsigned char *sha1)
244{
245 static char *name, *base;
a6080a0a 246
6eec46bd 247 return sha1_get_pack_name(sha1, &name, &base, "idx");
bf592c50
DB
248}
249
d5a63b99
JH
250struct alternate_object_database *alt_odb_list;
251static struct alternate_object_database **alt_odb_tail;
ace1534d 252
ddd5d056
JH
253/*
254 * Prepare alternate object database registry.
d5a63b99
JH
255 *
256 * The variable alt_odb_list points at the list of struct
257 * alternate_object_database. The elements on this list come from
258 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
259 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
260 * whose contents is similar to that environment variable but can be
261 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
262 * contains "/the/directory/corresponding/to/.git/objects/...", while
263 * its name points just after the slash at the end of ".git/objects/"
264 * in the example above, and has enough space to hold 40-byte hex
265 * SHA1, an extra slash for the first level indirection, and the
266 * terminating NUL.
ddd5d056 267 */
539e7506
EK
268static int link_alt_odb_entry(const char *entry, const char *relative_base,
269 int depth, const char *normalized_objdir)
ace1534d 270{
c2f493a4
MW
271 struct alternate_object_database *ent;
272 struct alternate_object_database *alt;
5bdf0a84
HW
273 int pfxlen, entlen;
274 struct strbuf pathbuf = STRBUF_INIT;
d5a63b99 275
85dadc38 276 if (!is_absolute_path(entry) && relative_base) {
5bdf0a84
HW
277 strbuf_addstr(&pathbuf, real_path(relative_base));
278 strbuf_addch(&pathbuf, '/');
c2f493a4 279 }
6eac50d8 280 strbuf_addstr(&pathbuf, entry);
c2f493a4 281
5bdf0a84
HW
282 normalize_path_copy(pathbuf.buf, pathbuf.buf);
283
284 pfxlen = strlen(pathbuf.buf);
285
286 /*
287 * The trailing slash after the directory name is given by
288 * this function at the end. Remove duplicates.
289 */
290 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
291 pfxlen -= 1;
292
293 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
294 ent = xmalloc(sizeof(*ent) + entlen);
295 memcpy(ent->base, pathbuf.buf, pfxlen);
296 strbuf_release(&pathbuf);
c2f493a4
MW
297
298 ent->name = ent->base + pfxlen + 1;
299 ent->base[pfxlen + 3] = '/';
300 ent->base[pfxlen] = ent->base[entlen-1] = 0;
301
302 /* Detect cases where alternate disappeared */
90b4a71c 303 if (!is_directory(ent->base)) {
c2f493a4
MW
304 error("object directory %s does not exist; "
305 "check .git/objects/info/alternates.",
306 ent->base);
307 free(ent);
308 return -1;
309 }
310
311 /* Prevent the common mistake of listing the same
312 * thing twice, or object directory itself.
313 */
314 for (alt = alt_odb_list; alt; alt = alt->next) {
80b47854
RS
315 if (pfxlen == alt->name - alt->base - 1 &&
316 !memcmp(ent->base, alt->base, pfxlen)) {
c2f493a4
MW
317 free(ent);
318 return -1;
319 }
320 }
539e7506 321 if (!strcmp_icase(ent->base, normalized_objdir)) {
c2f493a4
MW
322 free(ent);
323 return -1;
324 }
325
326 /* add the alternate entry */
327 *alt_odb_tail = ent;
328 alt_odb_tail = &(ent->next);
329 ent->next = NULL;
330
331 /* recursively add alternates */
332 read_info_alternates(ent->base, depth + 1);
333
334 ent->base[pfxlen] = '/';
335
336 return 0;
337}
338
c5950164 339static void link_alt_odb_entries(const char *alt, int len, int sep,
c2f493a4
MW
340 const char *relative_base, int depth)
341{
6eac50d8
MH
342 struct string_list entries = STRING_LIST_INIT_NODUP;
343 char *alt_copy;
344 int i;
539e7506 345 struct strbuf objdirbuf = STRBUF_INIT;
c2f493a4
MW
346
347 if (depth > 5) {
348 error("%s: ignoring alternate object stores, nesting too deep.",
349 relative_base);
350 return;
351 }
352
539e7506
EK
353 strbuf_addstr(&objdirbuf, absolute_path(get_object_directory()));
354 normalize_path_copy(objdirbuf.buf, objdirbuf.buf);
355
c5950164 356 alt_copy = xmemdupz(alt, len);
6eac50d8
MH
357 string_list_split_in_place(&entries, alt_copy, sep, -1);
358 for (i = 0; i < entries.nr; i++) {
359 const char *entry = entries.items[i].string;
360 if (entry[0] == '\0' || entry[0] == '#')
9577e7e3 361 continue;
6eac50d8
MH
362 if (!is_absolute_path(entry) && depth) {
363 error("%s: ignoring relative alternate object store %s",
364 relative_base, entry);
365 } else {
539e7506 366 link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
9577e7e3 367 }
9577e7e3 368 }
6eac50d8
MH
369 string_list_clear(&entries, 0);
370 free(alt_copy);
539e7506 371 strbuf_release(&objdirbuf);
d5a63b99
JH
372}
373
5e73633d 374void read_info_alternates(const char * relative_base, int depth)
d5a63b99 375{
9577e7e3 376 char *map;
dc49cd76 377 size_t mapsz;
d5a63b99 378 struct stat st;
9cb18f56
JM
379 const char alt_file_name[] = "info/alternates";
380 /* Given that relative_base is no longer than PATH_MAX,
381 ensure that "path" has enough space to append "/", the
382 file name, "info/alternates", and a trailing NUL. */
383 char path[PATH_MAX + 1 + sizeof alt_file_name];
c2f493a4 384 int fd;
d5a63b99 385
9cb18f56 386 sprintf(path, "%s/%s", relative_base, alt_file_name);
f4e51683 387 fd = git_open_noatime(path);
d5a63b99
JH
388 if (fd < 0)
389 return;
390 if (fstat(fd, &st) || (st.st_size == 0)) {
391 close(fd);
9a217f2a 392 return;
ace1534d 393 }
dc49cd76
SP
394 mapsz = xsize_t(st.st_size);
395 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 396 close(fd);
d5a63b99 397
c5950164 398 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
c2f493a4 399
dc49cd76 400 munmap(map, mapsz);
ace1534d
JH
401}
402
bef70b22
DB
403void add_to_alternates_file(const char *reference)
404{
405 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
acd3b9ec 406 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
e6baf4a1 407 char *alt = mkpath("%s\n", reference);
bef70b22
DB
408 write_or_die(fd, alt, strlen(alt));
409 if (commit_lock_file(lock))
410 die("could not close alternates file");
411 if (alt_odb_tail)
c5950164 412 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
bef70b22
DB
413}
414
d79796bc
JH
415void foreach_alt_odb(alt_odb_fn fn, void *cb)
416{
417 struct alternate_object_database *ent;
418
419 prepare_alt_odb();
420 for (ent = alt_odb_list; ent; ent = ent->next)
421 if (fn(ent, cb))
422 return;
423}
424
c2f493a4
MW
425void prepare_alt_odb(void)
426{
554fe20d 427 const char *alt;
c2f493a4 428
7dc24aa5
SP
429 if (alt_odb_tail)
430 return;
431
c2f493a4
MW
432 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
433 if (!alt) alt = "";
434
c2f493a4 435 alt_odb_tail = &alt_odb_list;
c5950164 436 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
437
438 read_info_alternates(get_object_directory(), 0);
439}
440
0f4dc14a 441static int has_loose_object_local(const unsigned char *sha1)
ace1534d 442{
30d6c6ea 443 return !access(sha1_file_name(sha1), F_OK);
0f4dc14a 444}
ace1534d 445
0f4dc14a
BC
446int has_loose_object_nonlocal(const unsigned char *sha1)
447{
448 struct alternate_object_database *alt;
9a217f2a 449 prepare_alt_odb();
d5a63b99 450 for (alt = alt_odb_list; alt; alt = alt->next) {
0f4dc14a 451 fill_sha1_path(alt->name, sha1);
c529d75a
LT
452 if (!access(alt->base, F_OK))
453 return 1;
ace1534d 454 }
c529d75a 455 return 0;
ace1534d
JH
456}
457
0f4dc14a
BC
458static int has_loose_object(const unsigned char *sha1)
459{
460 return has_loose_object_local(sha1) ||
461 has_loose_object_nonlocal(sha1);
462}
463
60bb8b14 464static unsigned int pack_used_ctr;
a53128b6
SP
465static unsigned int pack_mmap_calls;
466static unsigned int peak_pack_open_windows;
467static unsigned int pack_open_windows;
c7934306
SP
468static unsigned int pack_open_fds;
469static unsigned int pack_max_fds;
a53128b6 470static size_t peak_pack_mapped;
60bb8b14 471static size_t pack_mapped;
9a217f2a 472struct packed_git *packed_git;
1f688557 473
b79d18c9 474void pack_report(void)
a53128b6
SP
475{
476 fprintf(stderr,
e05db0fd
PR
477 "pack_report: getpagesize() = %10" SZ_FMT "\n"
478 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
479 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
480 sz_fmt(getpagesize()),
481 sz_fmt(packed_git_window_size),
482 sz_fmt(packed_git_limit));
a53128b6
SP
483 fprintf(stderr,
484 "pack_report: pack_used_ctr = %10u\n"
485 "pack_report: pack_mmap_calls = %10u\n"
486 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
487 "pack_report: pack_mapped = "
488 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
489 pack_used_ctr,
490 pack_mmap_calls,
491 pack_open_windows, peak_pack_open_windows,
9e42d6a1 492 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
493}
494
d40d535b
MH
495/*
496 * Open and mmap the index file at path, perform a couple of
497 * consistency checks, then record its information to p. Return 0 on
498 * success.
499 */
500static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
501{
502 void *idx_map;
42873078 503 struct pack_idx_header *hdr;
dc49cd76 504 size_t idx_size;
74e34e1f 505 uint32_t version, nr, i, *index;
f4e51683 506 int fd = git_open_noatime(path);
1f688557 507 struct stat st;
42873078 508
1f688557
JH
509 if (fd < 0)
510 return -1;
511 if (fstat(fd, &st)) {
512 close(fd);
513 return -1;
514 }
dc49cd76 515 idx_size = xsize_t(st.st_size);
2d88451b
SP
516 if (idx_size < 4 * 256 + 20 + 20) {
517 close(fd);
518 return error("index file %s is too small", path);
519 }
c4712e45 520 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 521 close(fd);
1f688557 522
42873078
NP
523 hdr = idx_map;
524 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
525 version = ntohl(hdr->idx_version);
526 if (version < 2 || version > 2) {
527 munmap(idx_map, idx_size);
6e1c2344 528 return error("index file %s is version %"PRIu32
74e34e1f
NP
529 " and is not supported by this binary"
530 " (try upgrading GIT to a newer version)",
531 path, version);
532 }
533 } else
534 version = 1;
df1b059d 535
1f688557 536 nr = 0;
42873078 537 index = idx_map;
74e34e1f
NP
538 if (version > 1)
539 index += 2; /* skip index header */
1f688557 540 for (i = 0; i < 256; i++) {
326bf396 541 uint32_t n = ntohl(index[i]);
2d88451b
SP
542 if (n < nr) {
543 munmap(idx_map, idx_size);
df1b059d 544 return error("non-monotonic index %s", path);
2d88451b 545 }
1f688557
JH
546 nr = n;
547 }
548
74e34e1f
NP
549 if (version == 1) {
550 /*
551 * Total size:
552 * - 256 index entries 4 bytes each
553 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
554 * - 20-byte SHA1 of the packfile
555 * - 20-byte SHA1 file checksum
556 */
557 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
558 munmap(idx_map, idx_size);
eef427a0 559 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
560 }
561 } else if (version == 2) {
562 /*
563 * Minimum size:
564 * - 8 bytes of header
565 * - 256 index entries 4 bytes each
566 * - 20-byte sha1 entry * nr
567 * - 4-byte crc entry * nr
568 * - 4-byte offset entry * nr
569 * - 20-byte SHA1 of the packfile
570 * - 20-byte SHA1 file checksum
571 * And after the 4-byte offset table might be a
572 * variable sized table containing 8-byte entries
573 * for offsets larger than 2^31.
574 */
575 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
576 unsigned long max_size = min_size;
577 if (nr)
578 max_size += (nr - 1)*8;
579 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 580 munmap(idx_map, idx_size);
eef427a0 581 return error("wrong index v2 file size in %s", path);
74e34e1f 582 }
7109c889
JH
583 if (idx_size != min_size &&
584 /*
585 * make sure we can deal with large pack offsets.
586 * 31-bit signed offset won't be enough, neither
587 * 32-bit unsigned one will be.
588 */
589 (sizeof(off_t) <= 4)) {
590 munmap(idx_map, idx_size);
591 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 592 }
2d88451b 593 }
1f688557 594
74e34e1f 595 p->index_version = version;
42873078
NP
596 p->index_data = idx_map;
597 p->index_size = idx_size;
57059091 598 p->num_objects = nr;
1f688557
JH
599 return 0;
600}
601
bc8e478a 602int open_pack_index(struct packed_git *p)
d079837e
SP
603{
604 char *idx_name;
605 int ret;
606
607 if (p->index_data)
608 return 0;
609
610 idx_name = xstrdup(p->pack_name);
611 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
612 ret = check_packed_git_idx(idx_name, p);
613 free(idx_name);
614 return ret;
615}
616
11daf39b
SP
617static void scan_windows(struct packed_git *p,
618 struct packed_git **lru_p,
619 struct pack_window **lru_w,
620 struct pack_window **lru_l)
1f688557 621{
11daf39b
SP
622 struct pack_window *w, *w_l;
623
624 for (w_l = NULL, w = p->windows; w; w = w->next) {
625 if (!w->inuse_cnt) {
626 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
627 *lru_p = p;
628 *lru_w = w;
629 *lru_l = w_l;
54044bf8 630 }
54044bf8 631 }
11daf39b 632 w_l = w;
f9253394 633 }
11daf39b
SP
634}
635
7c3ecb32 636static int unuse_one_window(struct packed_git *current)
11daf39b
SP
637{
638 struct packed_git *p, *lru_p = NULL;
639 struct pack_window *lru_w = NULL, *lru_l = NULL;
640
641 if (current)
642 scan_windows(current, &lru_p, &lru_w, &lru_l);
643 for (p = packed_git; p; p = p->next)
644 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
645 if (lru_p) {
646 munmap(lru_w->base, lru_w->len);
647 pack_mapped -= lru_w->len;
648 if (lru_l)
649 lru_l->next = lru_w->next;
7c3ecb32 650 else
54044bf8 651 lru_p->windows = lru_w->next;
54044bf8 652 free(lru_w);
a53128b6 653 pack_open_windows--;
54044bf8
SP
654 return 1;
655 }
656 return 0;
f9253394
JH
657}
658
7c3ecb32 659void release_pack_memory(size_t need)
97bfeb34
SP
660{
661 size_t cur = pack_mapped;
7c3ecb32 662 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
97bfeb34
SP
663 ; /* nothing */
664}
665
58ecbd5e
JN
666void *xmmap(void *start, size_t length,
667 int prot, int flags, int fd, off_t offset)
668{
669 void *ret = mmap(start, length, prot, flags, fd, offset);
670 if (ret == MAP_FAILED) {
671 if (!length)
672 return NULL;
7c3ecb32 673 release_pack_memory(length);
58ecbd5e
JN
674 ret = mmap(start, length, prot, flags, fd, offset);
675 if (ret == MAP_FAILED)
676 die_errno("Out of memory? mmap failed");
677 }
678 return ret;
679}
680
c9ced051
SP
681void close_pack_windows(struct packed_git *p)
682{
683 while (p->windows) {
684 struct pack_window *w = p->windows;
685
686 if (w->inuse_cnt)
687 die("pack '%s' still has open windows to it",
688 p->pack_name);
689 munmap(w->base, w->len);
690 pack_mapped -= w->len;
691 pack_open_windows--;
692 p->windows = w->next;
693 free(w);
694 }
695}
696
88d0db55
BC
697/*
698 * The LRU pack is the one with the oldest MRU window, preferring packs
699 * with no used windows, or the oldest mtime if it has no windows allocated.
700 */
701static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
702{
703 struct pack_window *w, *this_mru_w;
704 int has_windows_inuse = 0;
705
706 /*
707 * Reject this pack if it has windows and the previously selected
708 * one does not. If this pack does not have windows, reject
709 * it if the pack file is newer than the previously selected one.
710 */
711 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
712 return;
713
714 for (w = this_mru_w = p->windows; w; w = w->next) {
715 /*
716 * Reject this pack if any of its windows are in use,
717 * but the previously selected pack did not have any
718 * inuse windows. Otherwise, record that this pack
719 * has windows in use.
720 */
721 if (w->inuse_cnt) {
722 if (*accept_windows_inuse)
723 has_windows_inuse = 1;
724 else
725 return;
726 }
727
728 if (w->last_used > this_mru_w->last_used)
729 this_mru_w = w;
730
731 /*
732 * Reject this pack if it has windows that have been
733 * used more recently than the previously selected pack.
734 * If the previously selected pack had windows inuse and
735 * we have not encountered a window in this pack that is
736 * inuse, skip this check since we prefer a pack with no
737 * inuse windows to one that has inuse windows.
738 */
739 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
740 this_mru_w->last_used > (*mru_w)->last_used)
741 return;
742 }
743
744 /*
745 * Select this pack.
746 */
747 *mru_w = this_mru_w;
748 *lru_p = p;
749 *accept_windows_inuse = has_windows_inuse;
750}
751
752static int close_one_pack(void)
753{
754 struct packed_git *p, *lru_p = NULL;
755 struct pack_window *mru_w = NULL;
756 int accept_windows_inuse = 1;
757
758 for (p = packed_git; p; p = p->next) {
759 if (p->pack_fd == -1)
760 continue;
761 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
762 }
763
764 if (lru_p) {
765 close(lru_p->pack_fd);
766 pack_open_fds--;
767 lru_p->pack_fd = -1;
768 return 1;
769 }
770
771 return 0;
772}
773
03e79c88 774void unuse_pack(struct pack_window **w_cursor)
f9253394 775{
03e79c88
SP
776 struct pack_window *w = *w_cursor;
777 if (w) {
778 w->inuse_cnt--;
779 *w_cursor = NULL;
780 }
1f688557
JH
781}
782
fa5fc15d
SP
783void close_pack_index(struct packed_git *p)
784{
785 if (p->index_data) {
786 munmap((void *)p->index_data, p->index_size);
787 p->index_data = NULL;
788 }
789}
790
c74faea1
NP
791/*
792 * This is used by git-repack in case a newly created pack happens to
793 * contain the same set of objects as an existing one. In that case
794 * the resulting file might be different even if its name would be the
795 * same. It is best to close any reference to the old pack before it is
01689909 796 * replaced on disk. Of course no index pointers or windows for given pack
c74faea1
NP
797 * must subsist at this point. If ever objects from this pack are requested
798 * again, the new version of the pack will be reinitialized through
799 * reprepare_packed_git().
800 */
801void free_pack_by_name(const char *pack_name)
802{
803 struct packed_git *p, **pp = &packed_git;
804
805 while (*pp) {
806 p = *pp;
807 if (strcmp(pack_name, p->pack_name) == 0) {
fa3a0c94 808 clear_delta_base_cache();
c74faea1 809 close_pack_windows(p);
c7934306 810 if (p->pack_fd != -1) {
c74faea1 811 close(p->pack_fd);
c7934306
SP
812 pack_open_fds--;
813 }
fa5fc15d 814 close_pack_index(p);
c74faea1
NP
815 free(p->bad_object_sha1);
816 *pp = p->next;
c01f51cc
NTND
817 if (last_found_pack == p)
818 last_found_pack = NULL;
c74faea1
NP
819 free(p);
820 return;
821 }
822 pp = &p->next;
823 }
824}
825
a0788266
JS
826static unsigned int get_max_fd_limit(void)
827{
828#ifdef RLIMIT_NOFILE
491a8dec
JH
829 {
830 struct rlimit lim;
a0788266 831
491a8dec
JH
832 if (!getrlimit(RLIMIT_NOFILE, &lim))
833 return lim.rlim_cur;
834 }
835#endif
836
837#ifdef _SC_OPEN_MAX
838 {
839 long open_max = sysconf(_SC_OPEN_MAX);
840 if (0 < open_max)
841 return open_max;
842 /*
843 * Otherwise, we got -1 for one of the two
844 * reasons:
845 *
846 * (1) sysconf() did not understand _SC_OPEN_MAX
847 * and signaled an error with -1; or
848 * (2) sysconf() said there is no limit.
849 *
850 * We _could_ clear errno before calling sysconf() to
851 * tell these two cases apart and return a huge number
852 * in the latter case to let the caller cap it to a
853 * value that is not so selfish, but letting the
854 * fallback OPEN_MAX codepath take care of these cases
855 * is a lot simpler.
856 */
857 }
858#endif
a0788266 859
491a8dec 860#ifdef OPEN_MAX
a0788266
JS
861 return OPEN_MAX;
862#else
863 return 1; /* see the caller ;-) */
864#endif
865}
866
3cf8b462
SP
867/*
868 * Do not call this directly as this leaks p->pack_fd on error return;
869 * call open_packed_git() instead.
870 */
871static int open_packed_git_1(struct packed_git *p)
1f688557 872{
9bc879c1
SP
873 struct stat st;
874 struct pack_header hdr;
875 unsigned char sha1[20];
876 unsigned char *idx_sha1;
2c039da8 877 long fd_flag;
9bc879c1 878
d079837e
SP
879 if (!p->index_data && open_pack_index(p))
880 return error("packfile %s index unavailable", p->pack_name);
881
c7934306 882 if (!pack_max_fds) {
a0788266 883 unsigned int max_fds = get_max_fd_limit();
c7934306
SP
884
885 /* Save 3 for stdin/stdout/stderr, 22 for work */
886 if (25 < max_fds)
887 pack_max_fds = max_fds - 25;
888 else
889 pack_max_fds = 1;
890 }
891
88d0db55 892 while (pack_max_fds <= pack_open_fds && close_one_pack())
c7934306
SP
893 ; /* nothing */
894
f4e51683 895 p->pack_fd = git_open_noatime(p->pack_name);
9bc879c1 896 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 897 return -1;
c7934306 898 pack_open_fds++;
9bc879c1
SP
899
900 /* If we created the struct before we had the pack we lack size. */
bf592c50 901 if (!p->pack_size) {
bf592c50 902 if (!S_ISREG(st.st_mode))
072db278 903 return error("packfile %s not a regular file", p->pack_name);
bf592c50 904 p->pack_size = st.st_size;
9bc879c1 905 } else if (p->pack_size != st.st_size)
072db278 906 return error("packfile %s size changed", p->pack_name);
9bc879c1 907
2c039da8
JH
908 /* We leave these file descriptors open with sliding mmap;
909 * there is no point keeping them open across exec(), though.
910 */
911 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
912 if (fd_flag < 0)
072db278 913 return error("cannot determine file descriptor flags");
2c039da8
JH
914 fd_flag |= FD_CLOEXEC;
915 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 916 return error("cannot set FD_CLOEXEC");
2c039da8 917
9bc879c1 918 /* Verify we recognize this pack file format. */
e6e2bd62 919 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 920 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 921 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 922 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 923 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
924 return error("packfile %s is version %"PRIu32" and not"
925 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
926 p->pack_name, ntohl(hdr.hdr_version));
927
928 /* Verify the pack matches its index. */
57059091 929 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
930 return error("packfile %s claims to have %"PRIu32" objects"
931 " while index indicates %"PRIu32" objects",
57059091
NP
932 p->pack_name, ntohl(hdr.hdr_entries),
933 p->num_objects);
9bc879c1 934 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 935 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 936 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 937 return error("packfile %s signature is unavailable", p->pack_name);
42873078 938 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 939 if (hashcmp(sha1, idx_sha1))
072db278
SP
940 return error("packfile %s does not match index", p->pack_name);
941 return 0;
9bc879c1
SP
942}
943
3cf8b462
SP
944static int open_packed_git(struct packed_git *p)
945{
946 if (!open_packed_git_1(p))
947 return 0;
948 if (p->pack_fd != -1) {
949 close(p->pack_fd);
c7934306 950 pack_open_fds--;
3cf8b462
SP
951 p->pack_fd = -1;
952 }
953 return -1;
954}
955
c4001d92 956static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
957{
958 /* We must promise at least 20 bytes (one hash) after the
959 * offset is available from this window, otherwise the offset
960 * is not actually in this window and a different window (which
961 * has that one hash excess) must be used. This is to support
962 * the object header and delta base parsing routines below.
963 */
964 off_t win_off = win->offset;
965 return win_off <= offset
966 && (offset + 20) <= (win_off + win->len);
967}
968
4b25d091 969unsigned char *use_pack(struct packed_git *p,
03e79c88 970 struct pack_window **w_cursor,
c4001d92 971 off_t offset,
ef49a7a0 972 unsigned long *left)
9bc879c1 973{
60bb8b14 974 struct pack_window *win = *w_cursor;
03e79c88 975
a9d98a14 976 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
977 * pointless to ask for an offset into the middle of that
978 * hash, and the in_window function above wouldn't match
979 * don't allow an offset too close to the end of the file.
980 */
d131b7af
SP
981 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
982 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
983 if (offset > (p->pack_size - 20))
984 die("offset beyond end of packfile (truncated pack?)");
985
986 if (!win || !in_window(win, offset)) {
987 if (win)
988 win->inuse_cnt--;
989 for (win = p->windows; win; win = win->next) {
990 if (in_window(win, offset))
991 break;
992 }
993 if (!win) {
78a28df9 994 size_t window_align = packed_git_window_size / 2;
dc49cd76 995 off_t len;
d131b7af
SP
996
997 if (p->pack_fd == -1 && open_packed_git(p))
998 die("packfile %s cannot be accessed", p->pack_name);
999
60bb8b14 1000 win = xcalloc(1, sizeof(*win));
78a28df9 1001 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
1002 len = p->pack_size - win->offset;
1003 if (len > packed_git_window_size)
1004 len = packed_git_window_size;
1005 win->len = (size_t)len;
60bb8b14 1006 pack_mapped += win->len;
11daf39b 1007 while (packed_git_limit < pack_mapped
7c3ecb32 1008 && unuse_one_window(p))
60bb8b14 1009 ; /* nothing */
c4712e45 1010 win->base = xmmap(NULL, win->len,
60bb8b14
SP
1011 PROT_READ, MAP_PRIVATE,
1012 p->pack_fd, win->offset);
1013 if (win->base == MAP_FAILED)
73b4e4be
SP
1014 die("packfile %s cannot be mapped: %s",
1015 p->pack_name,
1016 strerror(errno));
d131b7af
SP
1017 if (!win->offset && win->len == p->pack_size
1018 && !p->do_not_close) {
1019 close(p->pack_fd);
1020 pack_open_fds--;
1021 p->pack_fd = -1;
1022 }
a53128b6
SP
1023 pack_mmap_calls++;
1024 pack_open_windows++;
1025 if (pack_mapped > peak_pack_mapped)
1026 peak_pack_mapped = pack_mapped;
1027 if (pack_open_windows > peak_pack_open_windows)
1028 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
1029 win->next = p->windows;
1030 p->windows = win;
1031 }
1f688557 1032 }
03e79c88
SP
1033 if (win != *w_cursor) {
1034 win->last_used = pack_used_ctr++;
1035 win->inuse_cnt++;
1036 *w_cursor = win;
1037 }
60bb8b14 1038 offset -= win->offset;
03e79c88 1039 if (left)
dc49cd76 1040 *left = win->len - xsize_t(offset);
03e79c88 1041 return win->base + offset;
1f688557
JH
1042}
1043
27d69a46
NP
1044static struct packed_git *alloc_packed_git(int extra)
1045{
1046 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1047 memset(p, 0, sizeof(*p));
1048 p->pack_fd = -1;
1049 return p;
1050}
1051
e0500293
JN
1052static void try_to_free_pack_memory(size_t size)
1053{
7c3ecb32 1054 release_pack_memory(size);
e0500293
JN
1055}
1056
42873078 1057struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557 1058{
e0500293 1059 static int have_set_try_to_free_routine;
1f688557 1060 struct stat st;
27d69a46 1061 struct packed_git *p = alloc_packed_git(path_len + 2);
1f688557 1062
e0500293
JN
1063 if (!have_set_try_to_free_routine) {
1064 have_set_try_to_free_routine = 1;
1065 set_try_to_free_routine(try_to_free_pack_memory);
1066 }
1067
42873078
NP
1068 /*
1069 * Make sure a corresponding .pack file exists and that
1070 * the index looks sane.
1071 */
1072 path_len -= strlen(".idx");
27d69a46
NP
1073 if (path_len < 1) {
1074 free(p);
1f688557 1075 return NULL;
27d69a46 1076 }
42873078 1077 memcpy(p->pack_name, path, path_len);
8d25931d
BC
1078
1079 strcpy(p->pack_name + path_len, ".keep");
1080 if (!access(p->pack_name, F_OK))
1081 p->pack_keep = 1;
1082
42873078 1083 strcpy(p->pack_name + path_len, ".pack");
d079837e 1084 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 1085 free(p);
1f688557
JH
1086 return NULL;
1087 }
42873078 1088
1f688557
JH
1089 /* ok, it looks sane as far as we can check without
1090 * actually mapping the pack file.
1091 */
1f688557 1092 p->pack_size = st.st_size;
9d835df2 1093 p->pack_local = local;
b867092f 1094 p->mtime = st.st_mtime;
42873078
NP
1095 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1096 hashclr(p->sha1);
1f688557
JH
1097 return p;
1098}
1099
7b64469a 1100struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 1101{
42873078 1102 const char *path = sha1_pack_name(sha1);
27d69a46 1103 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
bf592c50 1104
27d69a46
NP
1105 strcpy(p->pack_name, path);
1106 hashcpy(p->sha1, sha1);
42873078
NP
1107 if (check_packed_git_idx(idx_path, p)) {
1108 free(p);
bf592c50 1109 return NULL;
42873078 1110 }
bf592c50 1111
bf592c50
DB
1112 return p;
1113}
1114
1115void install_packed_git(struct packed_git *pack)
1116{
c7934306
SP
1117 if (pack->pack_fd != -1)
1118 pack_open_fds++;
1119
bf592c50
DB
1120 pack->next = packed_git;
1121 packed_git = pack;
1122}
1123
543c5caa
NTND
1124void (*report_garbage)(const char *desc, const char *path);
1125
1126static void report_helper(const struct string_list *list,
1127 int seen_bits, int first, int last)
1128{
1129 const char *msg;
1130 switch (seen_bits) {
1131 case 0:
235e8d59 1132 msg = "no corresponding .idx or .pack";
543c5caa
NTND
1133 break;
1134 case 1:
1135 msg = "no corresponding .idx";
1136 break;
1137 case 2:
1138 msg = "no corresponding .pack";
1139 break;
1140 default:
1141 return;
1142 }
1143 for (; first < last; first++)
1144 report_garbage(msg, list->items[first].string);
1145}
1146
1147static void report_pack_garbage(struct string_list *list)
1148{
1149 int i, baselen = -1, first = 0, seen_bits = 0;
1150
1151 if (!report_garbage)
1152 return;
1153
1154 sort_string_list(list);
1155
1156 for (i = 0; i < list->nr; i++) {
1157 const char *path = list->items[i].string;
1158 if (baselen != -1 &&
1159 strncmp(path, list->items[first].string, baselen)) {
1160 report_helper(list, seen_bits, first, i);
1161 baselen = -1;
1162 seen_bits = 0;
1163 }
1164 if (baselen == -1) {
1165 const char *dot = strrchr(path, '.');
1166 if (!dot) {
1167 report_garbage("garbage found", path);
1168 continue;
1169 }
1170 baselen = dot - path + 1;
1171 first = i;
1172 }
1173 if (!strcmp(path + baselen, "pack"))
1174 seen_bits |= 1;
1175 else if (!strcmp(path + baselen, "idx"))
1176 seen_bits |= 2;
1177 }
1178 report_helper(list, seen_bits, first, list->nr);
1179}
1180
9d835df2 1181static void prepare_packed_git_one(char *objdir, int local)
1f688557 1182{
880fb8de
RS
1183 struct strbuf path = STRBUF_INIT;
1184 size_t dirnamelen;
1f688557
JH
1185 DIR *dir;
1186 struct dirent *de;
543c5caa 1187 struct string_list garbage = STRING_LIST_INIT_DUP;
1f688557 1188
880fb8de
RS
1189 strbuf_addstr(&path, objdir);
1190 strbuf_addstr(&path, "/pack");
1191 dir = opendir(path.buf);
b5b16990 1192 if (!dir) {
26125f6b 1193 if (errno != ENOENT)
bd2afde8 1194 error("unable to open object pack directory: %s: %s",
880fb8de
RS
1195 path.buf, strerror(errno));
1196 strbuf_release(&path);
1f688557 1197 return;
b5b16990 1198 }
880fb8de
RS
1199 strbuf_addch(&path, '/');
1200 dirnamelen = path.len;
1f688557 1201 while ((de = readdir(dir)) != NULL) {
1f688557 1202 struct packed_git *p;
47bf4b0f 1203 size_t base_len;
543c5caa
NTND
1204
1205 if (is_dot_or_dotdot(de->d_name))
9cb18f56
JM
1206 continue;
1207
880fb8de
RS
1208 strbuf_setlen(&path, dirnamelen);
1209 strbuf_addstr(&path, de->d_name);
d90906a9 1210
47bf4b0f
JK
1211 base_len = path.len;
1212 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
d90906a9
NTND
1213 /* Don't reopen a pack we already have. */
1214 for (p = packed_git; p; p = p->next) {
47bf4b0f
JK
1215 size_t len;
1216 if (strip_suffix(p->pack_name, ".pack", &len) &&
1217 len == base_len &&
1218 !memcmp(p->pack_name, path.buf, len))
d90906a9
NTND
1219 break;
1220 }
1221 if (p == NULL &&
1222 /*
1223 * See if it really is a valid .idx file with
1224 * corresponding .pack file that we can map.
1225 */
880fb8de 1226 (p = add_packed_git(path.buf, path.len, local)) != NULL)
d90906a9 1227 install_packed_git(p);
86f7780c 1228 }
543c5caa
NTND
1229
1230 if (!report_garbage)
1231 continue;
1232
2975c770
JK
1233 if (ends_with(de->d_name, ".idx") ||
1234 ends_with(de->d_name, ".pack") ||
1235 ends_with(de->d_name, ".bitmap") ||
1236 ends_with(de->d_name, ".keep"))
880fb8de 1237 string_list_append(&garbage, path.buf);
543c5caa 1238 else
880fb8de 1239 report_garbage("garbage found", path.buf);
1f688557 1240 }
5b35bcd5 1241 closedir(dir);
543c5caa
NTND
1242 report_pack_garbage(&garbage);
1243 string_list_clear(&garbage, 0);
880fb8de 1244 strbuf_release(&path);
1f688557
JH
1245}
1246
b867092f
JH
1247static int sort_pack(const void *a_, const void *b_)
1248{
1249 struct packed_git *a = *((struct packed_git **)a_);
1250 struct packed_git *b = *((struct packed_git **)b_);
1251 int st;
1252
1253 /*
1254 * Local packs tend to contain objects specific to our
1255 * variant of the project than remote ones. In addition,
1256 * remote ones could be on a network mounted filesystem.
1257 * Favor local ones for these reasons.
1258 */
1259 st = a->pack_local - b->pack_local;
1260 if (st)
1261 return -st;
1262
1263 /*
1264 * Younger packs tend to contain more recent objects,
1265 * and more recent objects tend to get accessed more
1266 * often.
1267 */
1268 if (a->mtime < b->mtime)
1269 return 1;
1270 else if (a->mtime == b->mtime)
1271 return 0;
1272 return -1;
1273}
1274
1275static void rearrange_packed_git(void)
1276{
1277 struct packed_git **ary, *p;
1278 int i, n;
1279
1280 for (n = 0, p = packed_git; p; p = p->next)
1281 n++;
1282 if (n < 2)
1283 return;
1284
1285 /* prepare an array of packed_git for easier sorting */
1286 ary = xcalloc(n, sizeof(struct packed_git *));
1287 for (n = 0, p = packed_git; p; p = p->next)
1288 ary[n++] = p;
1289
1290 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1291
1292 /* link them back again */
1293 for (i = 0; i < n - 1; i++)
1294 ary[i]->next = ary[i + 1];
1295 ary[n - 1]->next = NULL;
1296 packed_git = ary[0];
1297
1298 free(ary);
1299}
1300
637cdd9d 1301static int prepare_packed_git_run_once = 0;
9a217f2a 1302void prepare_packed_git(void)
1f688557 1303{
d5a63b99 1304 struct alternate_object_database *alt;
1f688557 1305
637cdd9d 1306 if (prepare_packed_git_run_once)
1f688557 1307 return;
9d835df2 1308 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 1309 prepare_alt_odb();
d5a63b99 1310 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 1311 alt->name[-1] = 0;
9d835df2 1312 prepare_packed_git_one(alt->base, 0);
1494e038 1313 alt->name[-1] = '/';
1f688557 1314 }
b867092f 1315 rearrange_packed_git();
637cdd9d
JK
1316 prepare_packed_git_run_once = 1;
1317}
1318
fc04c412 1319void reprepare_packed_git(void)
637cdd9d
JK
1320{
1321 prepare_packed_git_run_once = 0;
1322 prepare_packed_git();
1f688557
JH
1323}
1324
8eca0b47
NP
1325static void mark_bad_packed_object(struct packed_git *p,
1326 const unsigned char *sha1)
1327{
1328 unsigned i;
1329 for (i = 0; i < p->num_bad_objects; i++)
1330 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1331 return;
1332 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1333 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1334 p->num_bad_objects++;
1335}
1336
b6c4cecc 1337static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
ac939109
NP
1338{
1339 struct packed_git *p;
1340 unsigned i;
1341
1342 for (p = packed_git; p; p = p->next)
1343 for (i = 0; i < p->num_bad_objects; i++)
1344 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
b6c4cecc
JH
1345 return p;
1346 return NULL;
ac939109
NP
1347}
1348
090ea126
NTND
1349/*
1350 * With an in-core object data in "map", rehash it to make sure the
1351 * object name actually matches "sha1" to detect object corruption.
1352 * With "map" == NULL, try reading the object named with "sha1" using
1353 * the streaming interface and rehash it to do the same.
1354 */
1355int check_sha1_signature(const unsigned char *sha1, void *map,
1356 unsigned long size, const char *type)
0fcfd160
LT
1357{
1358 unsigned char real_sha1[20];
090ea126
NTND
1359 enum object_type obj_type;
1360 struct git_istream *st;
1361 git_SHA_CTX c;
1362 char hdr[32];
1363 int hdrlen;
1364
1365 if (map) {
1366 hash_sha1_file(map, size, type, real_sha1);
1367 return hashcmp(sha1, real_sha1) ? -1 : 0;
1368 }
1369
1370 st = open_istream(sha1, &obj_type, &size, NULL);
1371 if (!st)
1372 return -1;
1373
1374 /* Generate the header */
1375 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1376
1377 /* Sha1.. */
1378 git_SHA1_Init(&c);
1379 git_SHA1_Update(&c, hdr, hdrlen);
1380 for (;;) {
1381 char buf[1024 * 16];
1382 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1383
f54fac53
JK
1384 if (readlen < 0) {
1385 close_istream(st);
1386 return -1;
1387 }
090ea126
NTND
1388 if (!readlen)
1389 break;
1390 git_SHA1_Update(&c, buf, readlen);
1391 }
1392 git_SHA1_Final(real_sha1, &c);
1393 close_istream(st);
a89fccd2 1394 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1395}
1396
ec73f580 1397int git_open_noatime(const char *name)
44d1c19e
LT
1398{
1399 static int sha1_file_open_flag = O_NOATIME;
44d1c19e 1400
f2e872aa
SP
1401 for (;;) {
1402 int fd = open(name, O_RDONLY | sha1_file_open_flag);
44d1c19e 1403 if (fd >= 0)
f2e872aa
SP
1404 return fd;
1405
f2e872aa
SP
1406 /* Might the failure be due to O_NOATIME? */
1407 if (errno != ENOENT && sha1_file_open_flag) {
44d1c19e 1408 sha1_file_open_flag = 0;
f2e872aa
SP
1409 continue;
1410 }
1411
1412 return -1;
44d1c19e 1413 }
44d1c19e
LT
1414}
1415
052fe5ea
JK
1416static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1417{
052fe5ea
JK
1418 struct alternate_object_database *alt;
1419
30d6c6ea 1420 if (!lstat(sha1_file_name(sha1), st))
052fe5ea
JK
1421 return 0;
1422
1423 prepare_alt_odb();
1424 errno = ENOENT;
1425 for (alt = alt_odb_list; alt; alt = alt->next) {
30d6c6ea 1426 fill_sha1_path(alt->name, sha1);
052fe5ea
JK
1427 if (!lstat(alt->base, st))
1428 return 0;
1429 }
1430
1431 return -1;
1432}
1433
44d1c19e
LT
1434static int open_sha1_file(const unsigned char *sha1)
1435{
1436 int fd;
44d1c19e 1437 struct alternate_object_database *alt;
d6c8a05b 1438 int most_interesting_errno;
44d1c19e 1439
30d6c6ea 1440 fd = git_open_noatime(sha1_file_name(sha1));
44d1c19e
LT
1441 if (fd >= 0)
1442 return fd;
d6c8a05b 1443 most_interesting_errno = errno;
44d1c19e
LT
1444
1445 prepare_alt_odb();
44d1c19e 1446 for (alt = alt_odb_list; alt; alt = alt->next) {
30d6c6ea 1447 fill_sha1_path(alt->name, sha1);
f4e51683 1448 fd = git_open_noatime(alt->base);
44d1c19e
LT
1449 if (fd >= 0)
1450 return fd;
d6c8a05b
JK
1451 if (most_interesting_errno == ENOENT)
1452 most_interesting_errno = errno;
44d1c19e 1453 }
d6c8a05b 1454 errno = most_interesting_errno;
44d1c19e
LT
1455 return -1;
1456}
1457
f0270efd 1458void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1459{
0fcfd160 1460 void *map;
144bde78 1461 int fd;
ace1534d 1462
44d1c19e
LT
1463 fd = open_sha1_file(sha1);
1464 map = NULL;
1465 if (fd >= 0) {
1466 struct stat st;
0fcfd160 1467
44d1c19e
LT
1468 if (!fstat(fd, &st)) {
1469 *size = xsize_t(st.st_size);
33e42de0
MM
1470 if (!*size) {
1471 /* mmap() is forbidden on empty files */
1472 error("object file %s is empty", sha1_file_name(sha1));
1473 return NULL;
1474 }
44d1c19e 1475 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1476 }
44d1c19e 1477 close(fd);
0fcfd160 1478 }
0fcfd160
LT
1479 return map;
1480}
1481
09ded04b
NP
1482unsigned long unpack_object_header_buffer(const unsigned char *buf,
1483 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1484{
ad1ed5ee 1485 unsigned shift;
48fb7deb 1486 unsigned long size, c;
ad1ed5ee
JH
1487 unsigned long used = 0;
1488
1489 c = buf[used++];
1490 *type = (c >> 4) & 7;
1491 size = c & 15;
1492 shift = 4;
1493 while (c & 0x80) {
f630cfda 1494 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1495 error("bad object header");
ea4f9685
JH
1496 size = used = 0;
1497 break;
09ded04b 1498 }
ad1ed5ee
JH
1499 c = buf[used++];
1500 size += (c & 0x7f) << shift;
1501 shift += 7;
1502 }
1503 *sizep = size;
1504 return used;
1505}
1506
eb4f4076 1507int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
ad1ed5ee 1508{
c4483576
LT
1509 /* Get the data stream */
1510 memset(stream, 0, sizeof(*stream));
1511 stream->next_in = map;
1512 stream->avail_in = mapsize;
1513 stream->next_out = buffer;
93821bd9
LT
1514 stream->avail_out = bufsiz;
1515
39c68542 1516 git_inflate_init(stream);
cc5c54e7 1517 return git_inflate(stream, 0);
c4483576
LT
1518}
1519
ef49a7a0 1520static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1521{
1522 int bytes = strlen(buffer) + 1;
3aee68aa 1523 unsigned char *buf = xmallocz(size);
93821bd9 1524 unsigned long n;
7efbff75 1525 int status = Z_OK;
5180cacc 1526
93821bd9
LT
1527 n = stream->total_out - bytes;
1528 if (n > size)
1529 n = size;
1530 memcpy(buf, (char *) buffer + bytes, n);
1531 bytes = n;
456cdf6e
LT
1532 if (bytes <= size) {
1533 /*
1534 * The above condition must be (bytes <= size), not
1535 * (bytes < size). In other words, even though we
ccf5ace0 1536 * expect no more output and set avail_out to zero,
456cdf6e
LT
1537 * the input zlib stream may have bytes that express
1538 * "this concludes the stream", and we *do* want to
1539 * eat that input.
1540 *
1541 * Otherwise we would not be able to test that we
1542 * consumed all the input to reach the expected size;
1543 * we also want to check that zlib tells us that all
1544 * went well with status == Z_STREAM_END at the end.
1545 */
5180cacc
LT
1546 stream->next_out = buf + bytes;
1547 stream->avail_out = size - bytes;
7efbff75 1548 while (status == Z_OK)
39c68542 1549 status = git_inflate(stream, Z_FINISH);
5180cacc 1550 }
456cdf6e 1551 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1552 git_inflate_end(stream);
7efbff75
JH
1553 return buf;
1554 }
1555
1556 if (status < 0)
1557 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1558 else if (stream->avail_in)
1559 error("garbage at end of loose object '%s'",
1560 sha1_to_hex(sha1));
1561 free(buf);
1562 return NULL;
5180cacc
LT
1563}
1564
1565/*
1566 * We used to just use "sscanf()", but that's actually way
1567 * too permissive for what we want to check. So do an anal
1568 * object header parse by hand.
1569 */
f0270efd 1570int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1571{
21666f1a 1572 char type[10];
5180cacc
LT
1573 int i;
1574 unsigned long size;
1575
1576 /*
a6080a0a 1577 * The type can be at most ten bytes (including the
5180cacc 1578 * terminating '\0' that we add), and is followed by
21666f1a 1579 * a space.
5180cacc 1580 */
21666f1a 1581 i = 0;
5180cacc
LT
1582 for (;;) {
1583 char c = *hdr++;
1584 if (c == ' ')
1585 break;
21666f1a
NP
1586 type[i++] = c;
1587 if (i >= sizeof(type))
5180cacc 1588 return -1;
5180cacc 1589 }
21666f1a 1590 type[i] = 0;
5180cacc
LT
1591
1592 /*
1593 * The length must follow immediately, and be in canonical
1594 * decimal format (ie "010" is not valid).
1595 */
1596 size = *hdr++ - '0';
1597 if (size > 9)
1598 return -1;
1599 if (size) {
1600 for (;;) {
1601 unsigned long c = *hdr - '0';
1602 if (c > 9)
1603 break;
1604 hdr++;
1605 size = size * 10 + c;
1606 }
1607 }
1608 *sizep = size;
1609
1610 /*
1611 * The length must be followed by a zero byte
1612 */
21666f1a 1613 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1614}
1615
7efbff75 1616static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1617{
5180cacc 1618 int ret;
ef49a7a0 1619 git_zstream stream;
5180cacc 1620 char hdr[8192];
0fcfd160 1621
5180cacc 1622 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1623 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1624 return NULL;
1625
7efbff75 1626 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1627}
1628
54dab52a
NP
1629unsigned long get_size_from_delta(struct packed_git *p,
1630 struct pack_window **w_curs,
1631 off_t curpos)
1632{
1633 const unsigned char *data;
1634 unsigned char delta_head[20], *in;
ef49a7a0 1635 git_zstream stream;
54dab52a
NP
1636 int st;
1637
1638 memset(&stream, 0, sizeof(stream));
1639 stream.next_out = delta_head;
1640 stream.avail_out = sizeof(delta_head);
1641
39c68542 1642 git_inflate_init(&stream);
54dab52a
NP
1643 do {
1644 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1645 stream.next_in = in;
39c68542 1646 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1647 curpos += stream.next_in - in;
1648 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1649 stream.total_out < sizeof(delta_head));
39c68542 1650 git_inflate_end(&stream);
3d77d877
NP
1651 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1652 error("delta data unpack-initial failed");
1653 return 0;
1654 }
54dab52a
NP
1655
1656 /* Examine the initial part of the delta to figure out
1657 * the result size.
1658 */
1659 data = delta_head;
1660
1661 /* ignore base size */
1662 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1663
1664 /* Read the result size */
1665 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1666}
1667
c4001d92 1668static off_t get_delta_base(struct packed_git *p,
03e79c88 1669 struct pack_window **w_curs,
c4001d92 1670 off_t *curpos,
21666f1a 1671 enum object_type type,
c4001d92 1672 off_t delta_obj_offset)
eb32d236 1673{
2b87c45b 1674 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1675 off_t base_offset;
eb32d236 1676
8d8a4ea5
SP
1677 /* use_pack() assured us we have [base_info, base_info + 20)
1678 * as a range that we can look at without walking off the
1679 * end of the mapped window. Its actually the hash size
1680 * that is assured. An OFS_DELTA longer than the hash size
1681 * is stupid, as then a REF_DELTA would be smaller to store.
1682 */
21666f1a 1683 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1684 unsigned used = 0;
1685 unsigned char c = base_info[used++];
1686 base_offset = c & 127;
1687 while (c & 128) {
1688 base_offset += 1;
8723f216 1689 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1690 return 0; /* overflow */
eb32d236
NP
1691 c = base_info[used++];
1692 base_offset = (base_offset << 7) + (c & 127);
1693 }
1694 base_offset = delta_obj_offset - base_offset;
d8f32556 1695 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1696 return 0; /* out of bound */
2b87c45b 1697 *curpos += used;
21666f1a 1698 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1699 /* The base entry _must_ be in the same pack */
1700 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1701 *curpos += 20;
eb32d236
NP
1702 } else
1703 die("I am totally screwed");
2b87c45b 1704 return base_offset;
eb32d236
NP
1705}
1706
5d642e75
JK
1707/*
1708 * Like get_delta_base above, but we return the sha1 instead of the pack
1709 * offset. This means it is cheaper for REF deltas (we do not have to do
1710 * the final object lookup), but more expensive for OFS deltas (we
1711 * have to load the revidx to convert the offset back into a sha1).
1712 */
1713static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1714 struct pack_window **w_curs,
1715 off_t curpos,
1716 enum object_type type,
1717 off_t delta_obj_offset)
1718{
1719 if (type == OBJ_REF_DELTA) {
1720 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1721 return base;
1722 } else if (type == OBJ_OFS_DELTA) {
1723 struct revindex_entry *revidx;
1724 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1725 type, delta_obj_offset);
1726
1727 if (!base_offset)
1728 return NULL;
1729
1730 revidx = find_pack_revindex(p, base_offset);
1731 if (!revidx)
1732 return NULL;
1733
1734 return nth_packed_object_sha1(p, revidx->nr);
1735 } else
1736 return NULL;
1737}
1738
f8c8abc5
JH
1739int unpack_object_header(struct packed_git *p,
1740 struct pack_window **w_curs,
1741 off_t *curpos,
1742 unsigned long *sizep)
a733cb60 1743{
03e79c88 1744 unsigned char *base;
ef49a7a0 1745 unsigned long left;
ad1ed5ee 1746 unsigned long used;
2b87c45b 1747 enum object_type type;
a733cb60 1748
8d8a4ea5 1749 /* use_pack() assures us we have [base, base + 20) available
0353a0c4 1750 * as a range that we can look at. (Its actually the hash
3dff5379 1751 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1752 * the maximum deflated object size is 2^137, which is just
1753 * insane, so we know won't exceed what we have been given.
1754 */
2b87c45b 1755 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1756 used = unpack_object_header_buffer(base, left, &type, sizep);
1757 if (!used) {
1758 type = OBJ_BAD;
1759 } else
1760 *curpos += used;
ad1ed5ee 1761
2b87c45b 1762 return type;
a733cb60
LT
1763}
1764
790d96c0
TR
1765static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1766{
1767 int type;
1768 struct revindex_entry *revidx;
1769 const unsigned char *sha1;
1770 revidx = find_pack_revindex(p, obj_offset);
1771 if (!revidx)
1772 return OBJ_BAD;
1773 sha1 = nth_packed_object_sha1(p, revidx->nr);
1774 mark_bad_packed_object(p, sha1);
1775 type = sha1_object_info(sha1, NULL);
1776 if (type <= OBJ_NONE)
1777 return OBJ_BAD;
1778 return type;
1779}
1780
790d96c0
TR
1781#define POI_STACK_PREALLOC 64
1782
90191d37
JK
1783static enum object_type packed_to_object_type(struct packed_git *p,
1784 off_t obj_offset,
1785 enum object_type type,
1786 struct pack_window **w_curs,
1787 off_t curpos)
1f688557 1788{
790d96c0
TR
1789 off_t small_poi_stack[POI_STACK_PREALLOC];
1790 off_t *poi_stack = small_poi_stack;
1791 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
5db47c2b 1792
790d96c0
TR
1793 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1794 off_t base_offset;
90191d37 1795 unsigned long size;
790d96c0
TR
1796 /* Push the object we're going to leave behind */
1797 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1798 poi_stack_alloc = alloc_nr(poi_stack_nr);
1799 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1800 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1801 } else {
1802 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1803 }
1804 poi_stack[poi_stack_nr++] = obj_offset;
1805 /* If parsing the base offset fails, just unwind */
90191d37 1806 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
790d96c0
TR
1807 if (!base_offset)
1808 goto unwind;
1809 curpos = obj_offset = base_offset;
90191d37 1810 type = unpack_object_header(p, w_curs, &curpos, &size);
790d96c0
TR
1811 if (type <= OBJ_NONE) {
1812 /* If getting the base itself fails, we first
1813 * retry the base, otherwise unwind */
1814 type = retry_bad_packed_offset(p, base_offset);
1815 if (type > OBJ_NONE)
1816 goto out;
1817 goto unwind;
1818 }
1819 }
1820
21666f1a 1821 switch (type) {
790d96c0 1822 case OBJ_BAD:
a733cb60 1823 case OBJ_COMMIT:
a733cb60 1824 case OBJ_TREE:
a733cb60 1825 case OBJ_BLOB:
a733cb60 1826 case OBJ_TAG:
a69d0943 1827 break;
1f688557 1828 default:
3d77d877
NP
1829 error("unknown object type %i at offset %"PRIuMAX" in %s",
1830 type, (uintmax_t)obj_offset, p->pack_name);
1831 type = OBJ_BAD;
1f688557 1832 }
790d96c0
TR
1833
1834out:
1835 if (poi_stack != small_poi_stack)
1836 free(poi_stack);
21666f1a 1837 return type;
790d96c0
TR
1838
1839unwind:
1840 while (poi_stack_nr) {
1841 obj_offset = poi_stack[--poi_stack_nr];
1842 type = retry_bad_packed_offset(p, obj_offset);
1843 if (type > OBJ_NONE)
1844 goto out;
1845 }
1846 type = OBJ_BAD;
1847 goto out;
1f688557
JH
1848}
1849
90191d37 1850static int packed_object_info(struct packed_git *p, off_t obj_offset,
23c339c0 1851 struct object_info *oi)
90191d37
JK
1852{
1853 struct pack_window *w_curs = NULL;
1854 unsigned long size;
1855 off_t curpos = obj_offset;
1856 enum object_type type;
1857
412916ee
JK
1858 /*
1859 * We always get the representation type, but only convert it to
1860 * a "real" type later if the caller is interested.
1861 */
90191d37
JK
1862 type = unpack_object_header(p, &w_curs, &curpos, &size);
1863
23c339c0 1864 if (oi->sizep) {
90191d37
JK
1865 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1866 off_t tmp_pos = curpos;
1867 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1868 type, obj_offset);
1869 if (!base_offset) {
1870 type = OBJ_BAD;
1871 goto out;
1872 }
23c339c0
JK
1873 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1874 if (*oi->sizep == 0) {
90191d37
JK
1875 type = OBJ_BAD;
1876 goto out;
1877 }
1878 } else {
23c339c0 1879 *oi->sizep = size;
90191d37
JK
1880 }
1881 }
1882
23c339c0 1883 if (oi->disk_sizep) {
90191d37 1884 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
23c339c0 1885 *oi->disk_sizep = revidx[1].offset - obj_offset;
90191d37
JK
1886 }
1887
23c339c0
JK
1888 if (oi->typep) {
1889 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
1890 if (*oi->typep < 0) {
412916ee
JK
1891 type = OBJ_BAD;
1892 goto out;
1893 }
1894 }
90191d37 1895
5d642e75
JK
1896 if (oi->delta_base_sha1) {
1897 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1898 const unsigned char *base;
1899
1900 base = get_delta_base_sha1(p, &w_curs, curpos,
1901 type, obj_offset);
1902 if (!base) {
1903 type = OBJ_BAD;
1904 goto out;
1905 }
1906
1907 hashcpy(oi->delta_base_sha1, base);
1908 } else
1909 hashclr(oi->delta_base_sha1);
1910 }
1911
90191d37
JK
1912out:
1913 unuse_pack(&w_curs);
1914 return type;
1915}
1916
eb950c19 1917static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1918 struct pack_window **w_curs,
c4001d92 1919 off_t curpos,
eb950c19 1920 unsigned long size)
de530aaa
SP
1921{
1922 int st;
ef49a7a0 1923 git_zstream stream;
079afb18 1924 unsigned char *buffer, *in;
de530aaa 1925
4ab07e4d 1926 buffer = xmallocz(size);
de530aaa 1927 memset(&stream, 0, sizeof(stream));
de530aaa 1928 stream.next_out = buffer;
39eea7bd 1929 stream.avail_out = size + 1;
de530aaa 1930
39c68542 1931 git_inflate_init(&stream);
079afb18 1932 do {
2b87c45b 1933 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 1934 stream.next_in = in;
39c68542 1935 st = git_inflate(&stream, Z_FINISH);
39eea7bd
JH
1936 if (!stream.avail_out)
1937 break; /* the payload is larger than it should be */
2b87c45b 1938 curpos += stream.next_in - in;
079afb18 1939 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 1940 git_inflate_end(&stream);
de530aaa
SP
1941 if ((st != Z_STREAM_END) || stream.total_out != size) {
1942 free(buffer);
1943 return NULL;
1944 }
1945
1946 return buffer;
1947}
1948
e5e01619
LT
1949#define MAX_DELTA_CACHE (256)
1950
18bdec11 1951static size_t delta_base_cached;
5e08ecbf
NP
1952
1953static struct delta_base_cache_lru_list {
1954 struct delta_base_cache_lru_list *prev;
1955 struct delta_base_cache_lru_list *next;
1956} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1957
e5e01619 1958static struct delta_base_cache_entry {
5e08ecbf
NP
1959 struct delta_base_cache_lru_list lru;
1960 void *data;
e5e01619
LT
1961 struct packed_git *p;
1962 off_t base_offset;
1963 unsigned long size;
e5e01619
LT
1964 enum object_type type;
1965} delta_base_cache[MAX_DELTA_CACHE];
1966
1967static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1968{
1969 unsigned long hash;
1970
1971 hash = (unsigned long)p + (unsigned long)base_offset;
1972 hash += (hash >> 8) + (hash >> 16);
3358004a 1973 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1974}
1975
84dd81c1
TR
1976static struct delta_base_cache_entry *
1977get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
5266d369
JH
1978{
1979 unsigned long hash = pack_entry_hash(p, base_offset);
84dd81c1
TR
1980 return delta_base_cache + hash;
1981}
1982
1983static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
1984 struct packed_git *p, off_t base_offset)
1985{
5266d369
JH
1986 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1987}
1988
84dd81c1
TR
1989static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1990{
1991 struct delta_base_cache_entry *ent;
1992 ent = get_delta_base_cache_entry(p, base_offset);
1993 return eq_delta_base_cache_entry(ent, p, base_offset);
1994}
1995
1996static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1997{
1998 ent->data = NULL;
1999 ent->lru.next->prev = ent->lru.prev;
2000 ent->lru.prev->next = ent->lru.next;
2001 delta_base_cached -= ent->size;
2002}
2003
62f255ad 2004static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 2005 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 2006{
84dd81c1 2007 struct delta_base_cache_entry *ent;
e5e01619 2008 void *ret;
e5e01619 2009
84dd81c1
TR
2010 ent = get_delta_base_cache_entry(p, base_offset);
2011
2012 if (!eq_delta_base_cache_entry(ent, p, base_offset))
749bc58c 2013 return unpack_entry(p, base_offset, type, base_size);
e5e01619 2014
84dd81c1
TR
2015 ret = ent->data;
2016
2017 if (!keep_cache)
2018 clear_delta_base_cache_entry(ent);
2019 else
182af834 2020 ret = xmemdupz(ent->data, ent->size);
e5e01619
LT
2021 *type = ent->type;
2022 *base_size = ent->size;
2023 return ret;
62f255ad
LT
2024}
2025
18bdec11
SP
2026static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2027{
2028 if (ent->data) {
2029 free(ent->data);
2030 ent->data = NULL;
5e08ecbf
NP
2031 ent->lru.next->prev = ent->lru.prev;
2032 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
2033 delta_base_cached -= ent->size;
2034 }
2035}
2036
3d20c636
SP
2037void clear_delta_base_cache(void)
2038{
2039 unsigned long p;
2040 for (p = 0; p < MAX_DELTA_CACHE; p++)
2041 release_delta_base_cache(&delta_base_cache[p]);
2042}
2043
62f255ad
LT
2044static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2045 void *base, unsigned long base_size, enum object_type type)
2046{
5e08ecbf 2047 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 2048 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 2049 struct delta_base_cache_lru_list *lru;
e5e01619 2050
18bdec11
SP
2051 release_delta_base_cache(ent);
2052 delta_base_cached += base_size;
5e08ecbf
NP
2053
2054 for (lru = delta_base_cache_lru.next;
2055 delta_base_cached > delta_base_cache_limit
2056 && lru != &delta_base_cache_lru;
2057 lru = lru->next) {
2058 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
2059 if (f->type == OBJ_BLOB)
2060 release_delta_base_cache(f);
2061 }
5e08ecbf
NP
2062 for (lru = delta_base_cache_lru.next;
2063 delta_base_cached > delta_base_cache_limit
2064 && lru != &delta_base_cache_lru;
2065 lru = lru->next) {
2066 struct delta_base_cache_entry *f = (void *)lru;
2067 release_delta_base_cache(f);
2068 }
18bdec11 2069
e5e01619
LT
2070 ent->p = p;
2071 ent->base_offset = base_offset;
2072 ent->type = type;
2073 ent->data = base;
2074 ent->size = base_size;
5e08ecbf
NP
2075 ent->lru.next = &delta_base_cache_lru;
2076 ent->lru.prev = delta_base_cache_lru.prev;
2077 delta_base_cache_lru.prev->next = &ent->lru;
2078 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
2079}
2080
c2c5b270
CC
2081static void *read_object(const unsigned char *sha1, enum object_type *type,
2082 unsigned long *size);
2083
5f44324d
JH
2084static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2085{
67dc598e
KB
2086 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2087 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2088 p->pack_name, (uintmax_t)obj_offset);
5f44324d
JH
2089}
2090
0e8189e2
NP
2091int do_check_packed_object_crc;
2092
abe601bb
TR
2093#define UNPACK_ENTRY_STACK_PREALLOC 64
2094struct unpack_entry_stack_ent {
2095 off_t obj_offset;
2096 off_t curpos;
2097 unsigned long size;
2098};
2099
c4001d92 2100void *unpack_entry(struct packed_git *p, off_t obj_offset,
abe601bb 2101 enum object_type *final_type, unsigned long *final_size)
1f688557 2102{
03e79c88 2103 struct pack_window *w_curs = NULL;
c4001d92 2104 off_t curpos = obj_offset;
abe601bb
TR
2105 void *data = NULL;
2106 unsigned long size;
2107 enum object_type type;
2108 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2109 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2110 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2111 int base_from_cache = 0;
1f688557 2112
67dc598e 2113 write_pack_access_log(p, obj_offset);
5f44324d 2114
abe601bb
TR
2115 /* PHASE 1: drill down to the innermost base object */
2116 for (;;) {
2117 off_t base_offset;
2118 int i;
2119 struct delta_base_cache_entry *ent;
2120
77965f8b
NTND
2121 ent = get_delta_base_cache_entry(p, curpos);
2122 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2123 type = ent->type;
2124 data = ent->data;
2125 size = ent->size;
2126 clear_delta_base_cache_entry(ent);
2127 base_from_cache = 1;
2128 break;
2129 }
2130
abe601bb
TR
2131 if (do_check_packed_object_crc && p->index_version > 1) {
2132 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2133 unsigned long len = revidx[1].offset - obj_offset;
2134 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2135 const unsigned char *sha1 =
2136 nth_packed_object_sha1(p, revidx->nr);
2137 error("bad packed object CRC for %s",
2138 sha1_to_hex(sha1));
2139 mark_bad_packed_object(p, sha1);
2140 unuse_pack(&w_curs);
2141 return NULL;
2142 }
2143 }
2144
abe601bb
TR
2145 type = unpack_object_header(p, &w_curs, &curpos, &size);
2146 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2147 break;
2148
2149 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2150 if (!base_offset) {
2151 error("failed to validate delta base reference "
2152 "at offset %"PRIuMAX" from %s",
2153 (uintmax_t)curpos, p->pack_name);
2154 /* bail to phase 2, in hopes of recovery */
2155 data = NULL;
2156 break;
2157 }
2158
2159 /* push object, proceed to base */
2160 if (delta_stack_nr >= delta_stack_alloc
2161 && delta_stack == small_delta_stack) {
2162 delta_stack_alloc = alloc_nr(delta_stack_nr);
2163 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2164 memcpy(delta_stack, small_delta_stack,
2165 sizeof(*delta_stack)*delta_stack_nr);
2166 } else {
2167 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2168 }
2169 i = delta_stack_nr++;
2170 delta_stack[i].obj_offset = obj_offset;
2171 delta_stack[i].curpos = curpos;
2172 delta_stack[i].size = size;
2173
2174 curpos = obj_offset = base_offset;
0e8189e2
NP
2175 }
2176
abe601bb
TR
2177 /* PHASE 2: handle the base */
2178 switch (type) {
eb32d236
NP
2179 case OBJ_OFS_DELTA:
2180 case OBJ_REF_DELTA:
abe601bb
TR
2181 if (data)
2182 die("BUG in unpack_entry: left loop at a valid delta");
4d703a1a 2183 break;
a733cb60 2184 case OBJ_COMMIT:
a733cb60 2185 case OBJ_TREE:
a733cb60 2186 case OBJ_BLOB:
a733cb60 2187 case OBJ_TAG:
abe601bb
TR
2188 if (!base_from_cache)
2189 data = unpack_compressed_entry(p, &w_curs, curpos, size);
4d703a1a 2190 break;
1f688557 2191 default:
8eca0b47
NP
2192 data = NULL;
2193 error("unknown object type %i at offset %"PRIuMAX" in %s",
abe601bb
TR
2194 type, (uintmax_t)obj_offset, p->pack_name);
2195 }
2196
2197 /* PHASE 3: apply deltas in order */
2198
2199 /* invariants:
2200 * 'data' holds the base data, or NULL if there was corruption
2201 */
2202 while (delta_stack_nr) {
2203 void *delta_data;
2204 void *base = data;
2205 unsigned long delta_size, base_size = size;
2206 int i;
2207
2208 data = NULL;
2209
2210 if (base)
2211 add_delta_base_cache(p, obj_offset, base, base_size, type);
2212
2213 if (!base) {
2214 /*
2215 * We're probably in deep shit, but let's try to fetch
2216 * the required base anyway from another pack or loose.
2217 * This is costly but should happen only in the presence
2218 * of a corrupted pack, and is better than failing outright.
2219 */
2220 struct revindex_entry *revidx;
2221 const unsigned char *base_sha1;
2222 revidx = find_pack_revindex(p, obj_offset);
2223 if (revidx) {
2224 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2225 error("failed to read delta base object %s"
2226 " at offset %"PRIuMAX" from %s",
2227 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2228 p->pack_name);
2229 mark_bad_packed_object(p, base_sha1);
2230 base = read_object(base_sha1, &type, &base_size);
2231 }
2232 }
2233
2234 i = --delta_stack_nr;
2235 obj_offset = delta_stack[i].obj_offset;
2236 curpos = delta_stack[i].curpos;
2237 delta_size = delta_stack[i].size;
2238
2239 if (!base)
2240 continue;
2241
2242 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2243
2244 if (!delta_data) {
2245 error("failed to unpack compressed delta "
2246 "at offset %"PRIuMAX" from %s",
2247 (uintmax_t)curpos, p->pack_name);
abe601bb
TR
2248 data = NULL;
2249 continue;
2250 }
2251
2252 data = patch_delta(base, base_size,
2253 delta_data, delta_size,
2254 &size);
1ee886c1
JK
2255
2256 /*
2257 * We could not apply the delta; warn the user, but keep going.
2258 * Our failure will be noticed either in the next iteration of
2259 * the loop, or if this is the final delta, in the caller when
2260 * we return NULL. Those code paths will take care of making
2261 * a more explicit warning and retrying with another copy of
2262 * the object.
2263 */
abe601bb 2264 if (!data)
1ee886c1 2265 error("failed to apply delta");
abe601bb 2266
4b8f772c 2267 free(delta_data);
1f688557 2268 }
abe601bb
TR
2269
2270 *final_type = type;
2271 *final_size = size;
2272
03e79c88 2273 unuse_pack(&w_curs);
019d1e65
NTND
2274
2275 if (delta_stack != small_delta_stack)
2276 free(delta_stack);
2277
21666f1a 2278 return data;
1f688557
JH
2279}
2280
d079837e 2281const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 2282 uint32_t n)
9a217f2a 2283{
42873078 2284 const unsigned char *index = p->index_data;
d079837e
SP
2285 if (!index) {
2286 if (open_pack_index(p))
2287 return NULL;
2288 index = p->index_data;
2289 }
57059091 2290 if (n >= p->num_objects)
d72308e0 2291 return NULL;
74e34e1f
NP
2292 index += 4 * 256;
2293 if (p->index_version == 1) {
2294 return index + 24 * n + 4;
2295 } else {
2296 index += 8;
2297 return index + 20 * n;
2298 }
2299}
2300
99093238 2301off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
2302{
2303 const unsigned char *index = p->index_data;
2304 index += 4 * 256;
2305 if (p->index_version == 1) {
2306 return ntohl(*((uint32_t *)(index + 24 * n)));
2307 } else {
2308 uint32_t off;
2309 index += 8 + p->num_objects * (20 + 4);
2310 off = ntohl(*((uint32_t *)(index + 4 * n)));
2311 if (!(off & 0x80000000))
2312 return off;
2313 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2314 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2315 ntohl(*((uint32_t *)(index + 4)));
2316 }
9a217f2a
JH
2317}
2318
c4001d92 2319off_t find_pack_entry_one(const unsigned char *sha1,
43057304 2320 struct packed_git *p)
1f688557 2321{
42873078 2322 const uint32_t *level1_ofs = p->index_data;
42873078 2323 const unsigned char *index = p->index_data;
628522ec
JH
2324 unsigned hi, lo, stride;
2325 static int use_lookup = -1;
2326 static int debug_lookup = -1;
2327
2328 if (debug_lookup < 0)
2329 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 2330
d079837e
SP
2331 if (!index) {
2332 if (open_pack_index(p))
2333 return 0;
2334 level1_ofs = p->index_data;
2335 index = p->index_data;
2336 }
74e34e1f
NP
2337 if (p->index_version > 1) {
2338 level1_ofs += 2;
2339 index += 8;
2340 }
42873078 2341 index += 4 * 256;
74e34e1f
NP
2342 hi = ntohl(level1_ofs[*sha1]);
2343 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
2344 if (p->index_version > 1) {
2345 stride = 20;
2346 } else {
2347 stride = 24;
2348 index += 4;
2349 }
2350
2351 if (debug_lookup)
6e1c2344 2352 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
2353 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2354
2355 if (use_lookup < 0)
2356 use_lookup = !!getenv("GIT_USE_LOOKUP");
2357 if (use_lookup) {
2358 int pos = sha1_entry_pos(index, stride, 0,
2359 lo, hi, p->num_objects, sha1);
2360 if (pos < 0)
2361 return 0;
2362 return nth_packed_object_offset(p, pos);
2363 }
1f688557
JH
2364
2365 do {
74e34e1f 2366 unsigned mi = (lo + hi) / 2;
628522ec
JH
2367 int cmp = hashcmp(index + mi * stride, sha1);
2368
2369 if (debug_lookup)
2370 printf("lo %u hi %u rg %u mi %u\n",
2371 lo, hi, hi - lo, mi);
43057304 2372 if (!cmp)
74e34e1f 2373 return nth_packed_object_offset(p, mi);
1f688557
JH
2374 if (cmp > 0)
2375 hi = mi;
2376 else
2377 lo = mi+1;
2378 } while (lo < hi);
2379 return 0;
2380}
2381
4c080182 2382int is_pack_valid(struct packed_git *p)
d131b7af
SP
2383{
2384 /* An already open pack is known to be valid. */
2385 if (p->pack_fd != -1)
2386 return 1;
2387
2388 /* If the pack has one window completely covering the
2389 * file size, the pack is known to be valid even if
2390 * the descriptor is not currently open.
2391 */
2392 if (p->windows) {
2393 struct pack_window *w = p->windows;
2394
2395 if (!w->offset && w->len == p->pack_size)
2396 return 1;
2397 }
2398
2399 /* Force the pack to open to prove its valid. */
2400 return !open_packed_git(p);
2401}
2402
95099731
NTND
2403static int fill_pack_entry(const unsigned char *sha1,
2404 struct pack_entry *e,
2405 struct packed_git *p)
2406{
2407 off_t offset;
2408
2409 if (p->num_bad_objects) {
2410 unsigned i;
2411 for (i = 0; i < p->num_bad_objects; i++)
2412 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2413 return 0;
2414 }
2415
2416 offset = find_pack_entry_one(sha1, p);
2417 if (!offset)
2418 return 0;
2419
2420 /*
2421 * We are about to tell the caller where they can locate the
2422 * requested object. We better make sure the packfile is
2423 * still here and can be accessed before supplying that
2424 * answer, as it may have been deleted since the index was
2425 * loaded!
2426 */
2427 if (!is_pack_valid(p)) {
2428 warning("packfile %s cannot be accessed", p->pack_name);
2429 return 0;
2430 }
2431 e->offset = offset;
2432 e->p = p;
2433 hashcpy(e->sha1, sha1);
2434 return 1;
2435}
2436
d40d535b
MH
2437/*
2438 * Iff a pack file contains the object named by sha1, return true and
2439 * store its location to e.
2440 */
4d6acb70 2441static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557
JH
2442{
2443 struct packed_git *p;
43057304 2444
1f688557 2445 prepare_packed_git();
f7c22cc6
NP
2446 if (!packed_git)
2447 return 0;
1f688557 2448
c01f51cc
NTND
2449 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2450 return 1;
8eca0b47 2451
c01f51cc 2452 for (p = packed_git; p; p = p->next) {
1b1005d1
MH
2453 if (p == last_found_pack)
2454 continue; /* we already checked this one */
f7c22cc6 2455
1b1005d1
MH
2456 if (fill_pack_entry(sha1, e, p)) {
2457 last_found_pack = p;
2458 return 1;
2459 }
c01f51cc 2460 }
1f688557
JH
2461 return 0;
2462}
2463
a6080a0a 2464struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
2465 struct packed_git *packs)
2466{
2467 struct packed_git *p;
bf592c50
DB
2468
2469 for (p = packs; p; p = p->next) {
43057304 2470 if (find_pack_entry_one(sha1, p))
bf592c50
DB
2471 return p;
2472 }
2473 return NULL;
21666f1a 2474
bf592c50
DB
2475}
2476
052fe5ea 2477static int sha1_loose_object_info(const unsigned char *sha1,
23c339c0 2478 struct object_info *oi)
65c2e0c3 2479{
36e4d74a 2480 int status;
65c2e0c3
JH
2481 unsigned long mapsize, size;
2482 void *map;
ef49a7a0 2483 git_zstream stream;
d65a16f6 2484 char hdr[32];
65c2e0c3 2485
5d642e75
JK
2486 if (oi->delta_base_sha1)
2487 hashclr(oi->delta_base_sha1);
2488
052fe5ea
JK
2489 /*
2490 * If we don't care about type or size, then we don't
4ef8d1dd
JH
2491 * need to look inside the object at all. Note that we
2492 * do not optimize out the stat call, even if the
2493 * caller doesn't care about the disk-size, since our
2494 * return value implicitly indicates whether the
2495 * object even exists.
052fe5ea 2496 */
23c339c0 2497 if (!oi->typep && !oi->sizep) {
4ef8d1dd
JH
2498 struct stat st;
2499 if (stat_sha1_file(sha1, &st) < 0)
2500 return -1;
2501 if (oi->disk_sizep)
23c339c0 2502 *oi->disk_sizep = st.st_size;
052fe5ea
JK
2503 return 0;
2504 }
2505
bb6b8e4f 2506 map = map_sha1_file(sha1, &mapsize);
f0df4ed5 2507 if (!map)
dbea72a8 2508 return -1;
23c339c0
JK
2509 if (oi->disk_sizep)
2510 *oi->disk_sizep = mapsize;
36e4d74a
JH
2511 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2512 status = error("unable to unpack %s header",
2513 sha1_to_hex(sha1));
21666f1a 2514 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 2515 status = error("unable to parse %s header", sha1_to_hex(sha1));
23c339c0
JK
2516 else if (oi->sizep)
2517 *oi->sizep = size;
39c68542 2518 git_inflate_end(&stream);
65c2e0c3 2519 munmap(map, mapsize);
23c339c0
JK
2520 if (oi->typep)
2521 *oi->typep = status;
052fe5ea 2522 return 0;
65c2e0c3
JH
2523}
2524
de7b5d62 2525int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
f0df4ed5 2526{
c4d9986f 2527 struct cached_object *co;
f0df4ed5 2528 struct pack_entry e;
5b086407 2529 int rtype;
1f7117ef 2530 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
f0df4ed5 2531
1f7117ef 2532 co = find_cached_object(real);
c4d9986f 2533 if (co) {
5b086407
JK
2534 if (oi->typep)
2535 *(oi->typep) = co->type;
9a490590
JH
2536 if (oi->sizep)
2537 *(oi->sizep) = co->size;
161f00e7
JK
2538 if (oi->disk_sizep)
2539 *(oi->disk_sizep) = 0;
5d642e75
JK
2540 if (oi->delta_base_sha1)
2541 hashclr(oi->delta_base_sha1);
9a490590 2542 oi->whence = OI_CACHED;
5b086407 2543 return 0;
c4d9986f
NTND
2544 }
2545
1f7117ef 2546 if (!find_pack_entry(real, &e)) {
ddd63e64 2547 /* Most likely it's a loose object. */
1f7117ef 2548 if (!sha1_loose_object_info(real, oi)) {
9a490590 2549 oi->whence = OI_LOOSE;
5b086407 2550 return 0;
9a490590 2551 }
ddd63e64
SG
2552
2553 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 2554 reprepare_packed_git();
1f7117ef 2555 if (!find_pack_entry(real, &e))
052fe5ea 2556 return -1;
f0df4ed5 2557 }
3d77d877 2558
23c339c0 2559 rtype = packed_object_info(e.p, e.offset, oi);
412916ee 2560 if (rtype < 0) {
1f7117ef
CC
2561 mark_bad_packed_object(e.p, real);
2562 return sha1_object_info_extended(real, oi, 0);
5266d369
JH
2563 } else if (in_delta_base_cache(e.p, e.offset)) {
2564 oi->whence = OI_DBCACHED;
9a490590
JH
2565 } else {
2566 oi->whence = OI_PACKED;
2567 oi->u.packed.offset = e.offset;
2568 oi->u.packed.pack = e.p;
2569 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2570 rtype == OBJ_OFS_DELTA);
3d77d877
NP
2571 }
2572
5b086407 2573 return 0;
f0df4ed5
JS
2574}
2575
3fc0dca9 2576/* returns enum object_type or negative */
9a490590
JH
2577int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2578{
5b086407 2579 enum object_type type;
d099b717 2580 struct object_info oi = {NULL};
9a490590 2581
5b086407 2582 oi.typep = &type;
9a490590 2583 oi.sizep = sizep;
de7b5d62 2584 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
5b086407
JK
2585 return -1;
2586 return type;
9a490590
JH
2587}
2588
21666f1a
NP
2589static void *read_packed_sha1(const unsigned char *sha1,
2590 enum object_type *type, unsigned long *size)
1f688557
JH
2591{
2592 struct pack_entry e;
8eca0b47 2593 void *data;
1f688557 2594
cd673c1f 2595 if (!find_pack_entry(sha1, &e))
1f688557 2596 return NULL;
8eca0b47
NP
2597 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2598 if (!data) {
2599 /*
2600 * We're probably in deep shit, but let's try to fetch
2601 * the required object anyway from another pack or loose.
2602 * This should happen only in the presence of a corrupted
2603 * pack, and is better than failing outright.
2604 */
2605 error("failed to read object %s at offset %"PRIuMAX" from %s",
2606 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2607 mark_bad_packed_object(e.p, sha1);
ac939109 2608 data = read_object(sha1, type, size);
8eca0b47
NP
2609 }
2610 return data;
1f688557
JH
2611}
2612
21666f1a
NP
2613int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2614 unsigned char *sha1)
d66b37bb
JH
2615{
2616 struct cached_object *co;
2617
21666f1a 2618 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2619 if (has_sha1_file(sha1) || find_cached_object(sha1))
2620 return 0;
c7353967 2621 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
d66b37bb
JH
2622 co = &cached_objects[cached_object_nr++];
2623 co->size = len;
21666f1a 2624 co->type = type;
efa13f7b
JH
2625 co->buf = xmalloc(len);
2626 memcpy(co->buf, buf, len);
d66b37bb
JH
2627 hashcpy(co->sha1, sha1);
2628 return 0;
2629}
2630
c2c5b270
CC
2631static void *read_object(const unsigned char *sha1, enum object_type *type,
2632 unsigned long *size)
0fcfd160
LT
2633{
2634 unsigned long mapsize;
2635 void *map, *buf;
d66b37bb
JH
2636 struct cached_object *co;
2637
2638 co = find_cached_object(sha1);
2639 if (co) {
21666f1a 2640 *type = co->type;
d66b37bb 2641 *size = co->size;
182af834 2642 return xmemdupz(co->buf, co->size);
d66b37bb 2643 }
0fcfd160 2644
8276c007
PE
2645 buf = read_packed_sha1(sha1, type, size);
2646 if (buf)
2647 return buf;
bb6b8e4f 2648 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2649 if (map) {
7efbff75 2650 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2651 munmap(map, mapsize);
2652 return buf;
2653 }
637cdd9d 2654 reprepare_packed_git();
8276c007 2655 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2656}
2657
b6c4cecc
JH
2658/*
2659 * This function dies on corrupt objects; the callers who want to
2660 * deal with them should arrange to call read_object() and give error
2661 * messages themselves.
2662 */
5bf29b95
JH
2663void *read_sha1_file_extended(const unsigned char *sha1,
2664 enum object_type *type,
2665 unsigned long *size,
2666 unsigned flag)
ac939109 2667{
3ba7a065 2668 void *data;
b6c4cecc 2669 const struct packed_git *p;
bf93eea0 2670 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
b6c4cecc 2671
3ba7a065
JH
2672 errno = 0;
2673 data = read_object(repl, type, size);
4bbf5a26 2674 if (data)
b6c4cecc 2675 return data;
68095570 2676
25f3af3f 2677 if (errno && errno != ENOENT)
3ba7a065
JH
2678 die_errno("failed to read object %s", sha1_to_hex(sha1));
2679
68095570 2680 /* die if we replaced an object with one that does not exist */
b6c4cecc 2681 if (repl != sha1)
68095570
CC
2682 die("replacement %s not found for %s",
2683 sha1_to_hex(repl), sha1_to_hex(sha1));
2684
b6c4cecc 2685 if (has_loose_object(repl)) {
30d6c6ea
MH
2686 const char *path = sha1_file_name(sha1);
2687
b6c4cecc
JH
2688 die("loose object %s (stored in %s) is corrupt",
2689 sha1_to_hex(repl), path);
e8b15e61 2690 }
68095570 2691
b6c4cecc
JH
2692 if ((p = has_packed_and_bad(repl)) != NULL)
2693 die("packed object %s (stored in %s) is corrupt",
2694 sha1_to_hex(repl), p->pack_name);
f5552aee 2695
b6c4cecc 2696 return NULL;
ac939109
NP
2697}
2698
40469ee9 2699void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2700 const char *required_type_name,
40469ee9
JH
2701 unsigned long *size,
2702 unsigned char *actual_sha1_return)
f4913f91 2703{
21666f1a 2704 enum object_type type, required_type;
f4913f91
JH
2705 void *buffer;
2706 unsigned long isize;
40469ee9 2707 unsigned char actual_sha1[20];
f4913f91 2708
21666f1a 2709 required_type = type_from_string(required_type_name);
e702496e 2710 hashcpy(actual_sha1, sha1);
40469ee9
JH
2711 while (1) {
2712 int ref_length = -1;
2713 const char *ref_type = NULL;
f4913f91 2714
21666f1a 2715 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2716 if (!buffer)
2717 return NULL;
21666f1a 2718 if (type == required_type) {
40469ee9
JH
2719 *size = isize;
2720 if (actual_sha1_return)
e702496e 2721 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2722 return buffer;
2723 }
2724 /* Handle references */
21666f1a 2725 else if (type == OBJ_COMMIT)
40469ee9 2726 ref_type = "tree ";
21666f1a 2727 else if (type == OBJ_TAG)
40469ee9
JH
2728 ref_type = "object ";
2729 else {
2730 free(buffer);
2731 return NULL;
2732 }
2733 ref_length = strlen(ref_type);
f4913f91 2734
50974ec9
MK
2735 if (ref_length + 40 > isize ||
2736 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2737 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2738 free(buffer);
2739 return NULL;
2740 }
1cf58e72 2741 free(buffer);
40469ee9
JH
2742 /* Now we have the ID of the referred-to object in
2743 * actual_sha1. Check again. */
f4913f91 2744 }
f4913f91
JH
2745}
2746
ce9fbf16 2747static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2748 const char *type, unsigned char *sha1,
d65a16f6 2749 char *hdr, int *hdrlen)
d410c0f5 2750{
9126f009 2751 git_SHA_CTX c;
d410c0f5
JH
2752
2753 /* Generate the header */
d65a16f6 2754 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
2755
2756 /* Sha1.. */
9126f009
NP
2757 git_SHA1_Init(&c);
2758 git_SHA1_Update(&c, hdr, *hdrlen);
2759 git_SHA1_Update(&c, buf, len);
2760 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2761}
2762
230f1322 2763/*
5a688fe4
JH
2764 * Move the just written object into its final resting place.
2765 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2766 * "moving" is only a part of what it does, when no patch between
2767 * master to pu changes the call sites of this function.
230f1322 2768 */
839837b9 2769int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322 2770{
e32c0a9c 2771 int ret = 0;
5a688fe4 2772
348df166 2773 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
2774 goto try_rename;
2775 else if (link(tmpfile, filename))
e32c0a9c 2776 ret = errno;
7ebb6fca
LT
2777
2778 /*
2779 * Coda hack - coda doesn't like cross-directory links,
2780 * so we fall back to a rename, which will mean that it
2781 * won't be able to check collisions, but that's not a
2782 * big deal.
2783 *
2784 * The same holds for FAT formatted media.
2785 *
3be1f18e 2786 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
2787 * left to unlink.
2788 */
2789 if (ret && ret != EEXIST) {
be66a6c4 2790 try_rename:
7ebb6fca 2791 if (!rename(tmpfile, filename))
3be1f18e 2792 goto out;
9e48b389 2793 ret = errno;
230f1322 2794 }
691f1a28 2795 unlink_or_warn(tmpfile);
230f1322
LT
2796 if (ret) {
2797 if (ret != EEXIST) {
82247e9b 2798 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
230f1322
LT
2799 }
2800 /* FIXME!!! Collision check here ? */
2801 }
2802
3be1f18e 2803out:
5256b006 2804 if (adjust_shared_perm(filename))
5a688fe4 2805 return error("unable to set permission to '%s'", filename);
230f1322
LT
2806 return 0;
2807}
2808
4d548150
LT
2809static int write_buffer(int fd, const void *buf, size_t len)
2810{
d34cf19b 2811 if (write_in_full(fd, buf, len) < 0)
93822c22 2812 return error("file write error (%s)", strerror(errno));
4d548150
LT
2813 return 0;
2814}
2815
ce9fbf16 2816int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2817 unsigned char *sha1)
2818{
d65a16f6 2819 char hdr[32];
abdc3fc8
RS
2820 int hdrlen;
2821 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2822 return 0;
2823}
2824
e9039dd3
LT
2825/* Finalize a file on disk, and close it. */
2826static void close_sha1_file(int fd)
2827{
aafe9fba
LT
2828 if (fsync_object_files)
2829 fsync_or_die(fd, "sha1 file");
e9039dd3 2830 if (close(fd) != 0)
d824cbba 2831 die_errno("error when closing sha1 file");
e9039dd3
LT
2832}
2833
5723fe7e
LT
2834/* Size of directory component, including the ending '/' */
2835static inline int directory_size(const char *filename)
2836{
2837 const char *s = strrchr(filename, '/');
2838 if (!s)
2839 return 0;
2840 return s - filename + 1;
2841}
2842
2843/*
2844 * This creates a temporary file in the same directory as the final
2845 * 'filename'
2846 *
2847 * We want to avoid cross-directory filename renames, because those
2848 * can have problems on various filesystems (FAT, NFS, Coda).
2849 */
2850static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2851{
2852 int fd, dirlen = directory_size(filename);
2853
2854 if (dirlen + 20 > bufsiz) {
2855 errno = ENAMETOOLONG;
2856 return -1;
2857 }
2858 memcpy(buffer, filename, dirlen);
2859 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
5256b006 2860 fd = git_mkstemp_mode(buffer, 0444);
cbacbf4e 2861 if (fd < 0 && dirlen && errno == ENOENT) {
5723fe7e 2862 /* Make sure the directory exists */
6ff6af62 2863 memcpy(buffer, filename, dirlen);
5723fe7e 2864 buffer[dirlen-1] = 0;
b2476a60
JH
2865 if (mkdir(buffer, 0777) && errno != EEXIST)
2866 return -1;
2867 if (adjust_shared_perm(buffer))
5723fe7e
LT
2868 return -1;
2869
2870 /* Try again */
2871 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
5256b006 2872 fd = git_mkstemp_mode(buffer, 0444);
5723fe7e
LT
2873 }
2874 return fd;
2875}
2876
bbac7311 2877static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 2878 const void *buf, unsigned long len, time_t mtime)
0fcfd160 2879{
915308b1 2880 int fd, ret;
9892beba 2881 unsigned char compressed[4096];
ef49a7a0 2882 git_zstream stream;
748af44c
NP
2883 git_SHA_CTX c;
2884 unsigned char parano_sha1[20];
ab1900a3 2885 static char tmp_file[PATH_MAX];
30d6c6ea 2886 const char *filename = sha1_file_name(sha1);
a44c9a5e 2887
ab1900a3 2888 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
aac17941 2889 if (fd < 0) {
35243577 2890 if (errno == EACCES)
82247e9b 2891 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
916d081b 2892 else
5eaeda70 2893 return error("unable to create temporary file: %s", strerror(errno));
aac17941
LT
2894 }
2895
0fcfd160
LT
2896 /* Set it up */
2897 memset(&stream, 0, sizeof(stream));
55bb5c91 2898 git_deflate_init(&stream, zlib_compression_level);
0fcfd160 2899 stream.next_out = compressed;
9892beba 2900 stream.avail_out = sizeof(compressed);
748af44c 2901 git_SHA1_Init(&c);
a44c9a5e
LT
2902
2903 /* First header.. */
d65a16f6 2904 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2905 stream.avail_in = hdrlen;
55bb5c91
JH
2906 while (git_deflate(&stream, 0) == Z_OK)
2907 ; /* nothing */
748af44c 2908 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
2909
2910 /* Then the data itself.. */
c00e657d 2911 stream.next_in = (void *)buf;
a44c9a5e 2912 stream.avail_in = len;
9892beba 2913 do {
748af44c 2914 unsigned char *in0 = stream.next_in;
55bb5c91 2915 ret = git_deflate(&stream, Z_FINISH);
748af44c 2916 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
2917 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2918 die("unable to write sha1 file");
2919 stream.next_out = compressed;
2920 stream.avail_out = sizeof(compressed);
2921 } while (ret == Z_OK);
2922
ac54c277
LT
2923 if (ret != Z_STREAM_END)
2924 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
55bb5c91 2925 ret = git_deflate_end_gently(&stream);
ac54c277
LT
2926 if (ret != Z_OK)
2927 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
2928 git_SHA1_Final(parano_sha1, &c);
2929 if (hashcmp(sha1, parano_sha1) != 0)
2930 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 2931
e9039dd3 2932 close_sha1_file(fd);
0fcfd160 2933
bbac7311
NP
2934 if (mtime) {
2935 struct utimbuf utb;
2936 utb.actime = mtime;
2937 utb.modtime = mtime;
ab1900a3 2938 if (utime(tmp_file, &utb) < 0)
bbac7311 2939 warning("failed utime() on %s: %s",
ab1900a3 2940 tmp_file, strerror(errno));
bbac7311
NP
2941 }
2942
ab1900a3 2943 return move_temp_to_file(tmp_file, filename);
0fcfd160 2944}
8237b185 2945
c00e657d 2946int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
bbac7311
NP
2947{
2948 unsigned char sha1[20];
2949 char hdr[32];
2950 int hdrlen;
2951
2952 /* Normally if we have it in the pack then we do not bother writing
2953 * it out into .git/objects/??/?{38} file.
2954 */
2955 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2956 if (returnsha1)
2957 hashcpy(returnsha1, sha1);
2958 if (has_sha1_file(sha1))
2959 return 0;
2960 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2961}
2962
2963int force_object_loose(const unsigned char *sha1, time_t mtime)
2964{
bbac7311
NP
2965 void *buf;
2966 unsigned long len;
2967 enum object_type type;
2968 char hdr[32];
2969 int hdrlen;
1fb23e65 2970 int ret;
bbac7311 2971
c529d75a 2972 if (has_loose_object(sha1))
bbac7311
NP
2973 return 0;
2974 buf = read_packed_sha1(sha1, &type, &len);
2975 if (!buf)
2976 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2977 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1fb23e65
BS
2978 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2979 free(buf);
2980
2981 return ret;
bbac7311
NP
2982}
2983
bf592c50
DB
2984int has_pack_index(const unsigned char *sha1)
2985{
2986 struct stat st;
2987 if (stat(sha1_pack_index_name(sha1), &st))
2988 return 0;
2989 return 1;
bf592c50
DB
2990}
2991
cd673c1f
JH
2992int has_sha1_pack(const unsigned char *sha1)
2993{
2994 struct pack_entry e;
2995 return find_pack_entry(sha1, &e);
2996}
2997
8237b185
DB
2998int has_sha1_file(const unsigned char *sha1)
2999{
1f688557
JH
3000 struct pack_entry e;
3001
cd673c1f 3002 if (find_pack_entry(sha1, &e))
1f688557 3003 return 1;
45e8a748
JK
3004 if (has_loose_object(sha1))
3005 return 1;
3006 reprepare_packed_git();
3007 return find_pack_entry(sha1, &e);
8237b185 3008}
74400e71 3009
c879daa2
NTND
3010static void check_tree(const void *buf, size_t size)
3011{
3012 struct tree_desc desc;
3013 struct name_entry entry;
3014
3015 init_tree_desc(&desc, buf, size);
3016 while (tree_entry(&desc, &entry))
3017 /* do nothing
3018 * tree_entry() will die() on malformed entries */
3019 ;
3020}
3021
3022static void check_commit(const void *buf, size_t size)
3023{
3024 struct commit c;
3025 memset(&c, 0, sizeof(c));
3026 if (parse_commit_buffer(&c, buf, size))
3027 die("corrupt commit");
3028}
3029
3030static void check_tag(const void *buf, size_t size)
3031{
3032 struct tag t;
3033 memset(&t, 0, sizeof(t));
3034 if (parse_tag_buffer(&t, buf, size))
3035 die("corrupt tag");
3036}
3037
43df4f86 3038static int index_mem(unsigned char *sha1, void *buf, size_t size,
c4ce46fc
JH
3039 enum object_type type,
3040 const char *path, unsigned flags)
e7332f96 3041{
6c510bee 3042 int ret, re_allocated = 0;
c4ce46fc 3043 int write_object = flags & HASH_WRITE_OBJECT;
74400e71 3044
7672db20 3045 if (!type)
edaec3fb 3046 type = OBJ_BLOB;
6c510bee
LT
3047
3048 /*
3049 * Convert blobs to git internal format
3050 */
43df4f86 3051 if ((type == OBJ_BLOB) && path) {
f285a2d7 3052 struct strbuf nbuf = STRBUF_INIT;
21e5ad50 3053 if (convert_to_git(path, buf, size, &nbuf,
5e12e78e 3054 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
b315c5c0 3055 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
3056 re_allocated = 1;
3057 }
3058 }
c4ce46fc 3059 if (flags & HASH_FORMAT_CHECK) {
c879daa2
NTND
3060 if (type == OBJ_TREE)
3061 check_tree(buf, size);
3062 if (type == OBJ_COMMIT)
3063 check_commit(buf, size);
3064 if (type == OBJ_TAG)
3065 check_tag(buf, size);
3066 }
6c510bee 3067
7672db20 3068 if (write_object)
edaec3fb 3069 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 3070 else
edaec3fb 3071 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 3072 if (re_allocated)
6c510bee 3073 free(buf);
43df4f86
DP
3074 return ret;
3075}
3076
7b41e1e1
JH
3077static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3078 const char *path, unsigned flags)
3079{
3080 struct strbuf sbuf = STRBUF_INIT;
3081 int ret;
3082
3083 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3084 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3085 else
3086 ret = -1;
3087 strbuf_release(&sbuf);
3088 return ret;
3089}
3090
ea68b0ce
DP
3091#define SMALL_FILE_SIZE (32*1024)
3092
7b41e1e1
JH
3093static int index_core(unsigned char *sha1, int fd, size_t size,
3094 enum object_type type, const char *path,
3095 unsigned flags)
43df4f86
DP
3096{
3097 int ret;
43df4f86 3098
7b41e1e1 3099 if (!size) {
c4ce46fc 3100 ret = index_mem(sha1, NULL, size, type, path, flags);
ea68b0ce
DP
3101 } else if (size <= SMALL_FILE_SIZE) {
3102 char *buf = xmalloc(size);
3103 if (size == read_in_full(fd, buf, size))
c4ce46fc 3104 ret = index_mem(sha1, buf, size, type, path, flags);
ea68b0ce
DP
3105 else
3106 ret = error("short read %s", strerror(errno));
3107 free(buf);
08bda208 3108 } else {
43df4f86 3109 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
c4ce46fc 3110 ret = index_mem(sha1, buf, size, type, path, flags);
aac17941 3111 munmap(buf, size);
08bda208 3112 }
7b41e1e1
JH
3113 return ret;
3114}
3115
4dd1fbc7 3116/*
568508e7
JH
3117 * This creates one packfile per large blob unless bulk-checkin
3118 * machinery is "plugged".
4dd1fbc7
JH
3119 *
3120 * This also bypasses the usual "convert-to-git" dance, and that is on
3121 * purpose. We could write a streaming version of the converting
3122 * functions and insert that before feeding the data to fast-import
4f22b101
JK
3123 * (or equivalent in-core API described above). However, that is
3124 * somewhat complicated, as we do not know the size of the filter
3125 * result, which we need to know beforehand when writing a git object.
3126 * Since the primary motivation for trying to stream from the working
3127 * tree file and to avoid mmaping it in core is to deal with large
3128 * binary blobs, they generally do not want to get any conversion, and
3129 * callers should avoid this code path when filters are requested.
4dd1fbc7
JH
3130 */
3131static int index_stream(unsigned char *sha1, int fd, size_t size,
3132 enum object_type type, const char *path,
3133 unsigned flags)
3134{
568508e7 3135 return index_bulk_checkin(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
3136}
3137
7b41e1e1
JH
3138int index_fd(unsigned char *sha1, int fd, struct stat *st,
3139 enum object_type type, const char *path, unsigned flags)
3140{
3141 int ret;
3142 size_t size = xsize_t(st->st_size);
3143
3144 if (!S_ISREG(st->st_mode))
3145 ret = index_pipe(sha1, fd, type, path, flags);
4f22b101 3146 else if (size <= big_file_threshold || type != OBJ_BLOB ||
7ce7c760 3147 (path && would_convert_to_git(path)))
7b41e1e1 3148 ret = index_core(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
3149 else
3150 ret = index_stream(sha1, fd, size, type, path, flags);
43df4f86 3151 close(fd);
aac17941 3152 return ret;
74400e71 3153}
ec1fcc16 3154
c4ce46fc 3155int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
ec1fcc16
JH
3156{
3157 int fd;
b760d3aa 3158 struct strbuf sb = STRBUF_INIT;
ec1fcc16
JH
3159
3160 switch (st->st_mode & S_IFMT) {
3161 case S_IFREG:
3162 fd = open(path, O_RDONLY);
3163 if (fd < 0)
3164 return error("open(\"%s\"): %s", path,
3165 strerror(errno));
c4ce46fc 3166 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
ec1fcc16
JH
3167 return error("%s: failed to insert into database",
3168 path);
3169 break;
3170 case S_IFLNK:
b760d3aa 3171 if (strbuf_readlink(&sb, path, st->st_size)) {
ec1fcc16 3172 char *errstr = strerror(errno);
ec1fcc16
JH
3173 return error("readlink(\"%s\"): %s", path,
3174 errstr);
3175 }
c4ce46fc 3176 if (!(flags & HASH_WRITE_OBJECT))
b760d3aa
LT
3177 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3178 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
ec1fcc16
JH
3179 return error("%s: failed to insert into database",
3180 path);
b760d3aa 3181 strbuf_release(&sb);
ec1fcc16 3182 break;
f35a6d3b
LT
3183 case S_IFDIR:
3184 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
3185 default:
3186 return error("%s: unsupported file type", path);
3187 }
3188 return 0;
3189}
a69e5429
JH
3190
3191int read_pack_header(int fd, struct pack_header *header)
3192{
c697ad14
HO
3193 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3194 /* "eof before pack header was fully read" */
3195 return PH_ERROR_EOF;
3196
a69e5429
JH
3197 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3198 /* "protocol error (pack signature mismatch detected)" */
3199 return PH_ERROR_PACK_SIGNATURE;
3200 if (!pack_version_ok(header->hdr_version))
3201 /* "protocol error (pack version unsupported)" */
3202 return PH_ERROR_PROTOCOL;
3203 return 0;
3204}
40d52ff7
JK
3205
3206void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3207{
3208 enum object_type type = sha1_object_info(sha1, NULL);
3209 if (type < 0)
3210 die("%s is not a valid object", sha1_to_hex(sha1));
3211 if (type != expect)
3212 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3213 typename(expect));
3214}