rename_tmp_log(): on SCLD_VANISHED, retry
[git/git.git] / sha1_file.c
CommitLineData
0fcfd160
LT
1/*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
0fcfd160 9#include "cache.h"
6eac50d8 10#include "string-list.h"
1f688557 11#include "delta.h"
a733cb60 12#include "pack.h"
8e440259
PE
13#include "blob.h"
14#include "commit.h"
4dd1fbc7 15#include "run-command.h"
8e440259
PE
16#include "tag.h"
17#include "tree.h"
c879daa2 18#include "tree-walk.h"
f35a6d3b 19#include "refs.h"
70f5d5d3 20#include "pack-revindex.h"
628522ec 21#include "sha1-lookup.h"
568508e7 22#include "bulk-checkin.h"
090ea126 23#include "streaming.h"
543c5caa 24#include "dir.h"
0fcfd160 25
144bde78
LT
26#ifndef O_NOATIME
27#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
28#define O_NOATIME 01000000
29#else
30#define O_NOATIME 0
31#endif
32#endif
33
28bd70d8
JN
34#define SZ_FMT PRIuMAX
35static inline uintmax_t sz_fmt(size_t s) { return s; }
e05db0fd 36
96f1e58f 37const unsigned char null_sha1[20];
88cd621d 38
b12ca963
NTND
39static const char *no_log_pack_access = "no_log_pack_access";
40static const char *log_pack_access;
41
c597ba80
NTND
42/*
43 * This is meant to hold a *small* number of objects that you would
44 * want read_sha1_file() to be able to return, but yet you do not want
45 * to write them into the object store (e.g. a browse-only
46 * application).
47 */
48static struct cached_object {
49 unsigned char sha1[20];
50 enum object_type type;
51 void *buf;
52 unsigned long size;
53} *cached_objects;
54static int cached_object_nr, cached_object_alloc;
55
56static struct cached_object empty_tree = {
dab0d410 57 EMPTY_TREE_SHA1_BIN_LITERAL,
c597ba80
NTND
58 OBJ_TREE,
59 "",
60 0
61};
62
c01f51cc
NTND
63static struct packed_git *last_found_pack;
64
c597ba80
NTND
65static struct cached_object *find_cached_object(const unsigned char *sha1)
66{
67 int i;
68 struct cached_object *co = cached_objects;
69
70 for (i = 0; i < cached_object_nr; i++, co++) {
71 if (!hashcmp(co->sha1, sha1))
72 return co;
73 }
74 if (!hashcmp(sha1, empty_tree.sha1))
75 return &empty_tree;
76 return NULL;
77}
78
90a6464b
JH
79int mkdir_in_gitdir(const char *path)
80{
81 if (mkdir(path, 0777)) {
82 int saved_errno = errno;
83 struct stat st;
84 struct strbuf sb = STRBUF_INIT;
85
86 if (errno != EEXIST)
87 return -1;
88 /*
89 * Are we looking at a path in a symlinked worktree
90 * whose original repository does not yet have it?
91 * e.g. .git/rr-cache pointing at its original
92 * repository in which the user hasn't performed any
93 * conflict resolution yet?
94 */
95 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
96 strbuf_readlink(&sb, path, st.st_size) ||
97 !is_absolute_path(sb.buf) ||
98 mkdir(sb.buf, 0777)) {
99 strbuf_release(&sb);
100 errno = saved_errno;
101 return -1;
102 }
103 strbuf_release(&sb);
104 }
105 return adjust_shared_perm(path);
106}
107
0be0521b 108enum scld_error safe_create_leading_directories(char *path)
b2cb9425 109{
26c8ae2a 110 char *next_component = path + offset_1st_component(path);
0be0521b 111 enum scld_error ret = SCLD_OK;
67d42212 112
0be0521b 113 while (ret == SCLD_OK && next_component) {
f0502332 114 struct stat st;
26c8ae2a 115 char *slash = strchr(next_component, '/');
f0502332 116
831651fd 117 if (!slash)
b2cb9425 118 break;
bf10cf70 119
26c8ae2a 120 next_component = slash + 1;
bf10cf70
MH
121 while (*next_component == '/')
122 next_component++;
26c8ae2a 123 if (!*next_component)
5f0bdf50 124 break;
831651fd
MH
125
126 *slash = '\0';
67d42212
JR
127 if (!stat(path, &st)) {
128 /* path exists */
9e6f885d 129 if (!S_ISDIR(st.st_mode))
0be0521b 130 ret = SCLD_EXISTS;
53a39721 131 } else if (mkdir(path, 0777)) {
928734d9 132 if (errno == EEXIST &&
9e6f885d 133 !stat(path, &st) && S_ISDIR(st.st_mode))
928734d9 134 ; /* somebody created it since we checked */
18d37e86
MH
135 else if (errno == ENOENT)
136 /*
137 * Either mkdir() failed because
138 * somebody just pruned the containing
139 * directory, or stat() failed because
140 * the file that was in our way was
141 * just removed. Either way, inform
142 * the caller that it might be worth
143 * trying again:
144 */
145 ret = SCLD_VANISHED;
9e6f885d 146 else
0be0521b 147 ret = SCLD_FAILED;
53a39721 148 } else if (adjust_shared_perm(path)) {
0be0521b 149 ret = SCLD_PERMS;
457f06d6 150 }
831651fd 151 *slash = '/';
b2cb9425 152 }
9e6f885d 153 return ret;
b2cb9425 154}
723c31fe 155
0be0521b 156enum scld_error safe_create_leading_directories_const(const char *path)
8e21d63b
JK
157{
158 /* path points to cache entries, so xstrdup before messing with it */
159 char *buf = xstrdup(path);
0be0521b 160 enum scld_error result = safe_create_leading_directories(buf);
8e21d63b
JK
161 free(buf);
162 return result;
163}
164
ace1534d
JH
165static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
166{
167 int i;
168 for (i = 0; i < 20; i++) {
169 static char hex[] = "0123456789abcdef";
170 unsigned int val = sha1[i];
171 char *pos = pathbuf + i*2 + (i > 0);
172 *pos++ = hex[val >> 4];
173 *pos = hex[val & 0xf];
174 }
175}
176
0fcfd160
LT
177/*
178 * NOTE! This returns a statically allocated buffer, so you have to be
790296fd 179 * careful about using it. Do an "xstrdup()" if you need to save the
0fcfd160 180 * filename.
ace1534d
JH
181 *
182 * Also note that this returns the location for creating. Reading
183 * SHA1 file can happen from any alternate directory listed in the
d19938ab 184 * DB_ENVIRONMENT environment variable if it is not found in
ace1534d 185 * the primary object database.
0fcfd160
LT
186 */
187char *sha1_file_name(const unsigned char *sha1)
188{
560fb6a1
JK
189 static char buf[PATH_MAX];
190 const char *objdir;
191 int len;
0fcfd160 192
560fb6a1
JK
193 objdir = get_object_directory();
194 len = strlen(objdir);
195
196 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
197 if (len + 43 > PATH_MAX)
198 die("insanely long object directory %s", objdir);
199 memcpy(buf, objdir, len);
200 buf[len] = '/';
201 buf[len+3] = '/';
202 buf[len+42] = '\0';
203 fill_sha1_path(buf + len + 1, sha1);
204 return buf;
0fcfd160
LT
205}
206
633f43e1 207static char *sha1_get_pack_name(const unsigned char *sha1,
6eec46bd 208 char **name, char **base, const char *which)
bf592c50
DB
209{
210 static const char hex[] = "0123456789abcdef";
633f43e1 211 char *buf;
bf592c50
DB
212 int i;
213
633f43e1 214 if (!*base) {
bf592c50
DB
215 const char *sha1_file_directory = get_object_directory();
216 int len = strlen(sha1_file_directory);
633f43e1 217 *base = xmalloc(len + 60);
6eec46bd
JH
218 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
219 sha1_file_directory, which);
633f43e1 220 *name = *base + len + 11;
bf592c50
DB
221 }
222
633f43e1 223 buf = *name;
bf592c50
DB
224
225 for (i = 0; i < 20; i++) {
226 unsigned int val = *sha1++;
227 *buf++ = hex[val >> 4];
228 *buf++ = hex[val & 0xf];
229 }
a6080a0a 230
633f43e1 231 return *base;
bf592c50
DB
232}
233
633f43e1 234char *sha1_pack_name(const unsigned char *sha1)
bf592c50 235{
633f43e1 236 static char *name, *base;
bf592c50 237
6eec46bd 238 return sha1_get_pack_name(sha1, &name, &base, "pack");
633f43e1 239}
bf592c50 240
633f43e1
HO
241char *sha1_pack_index_name(const unsigned char *sha1)
242{
243 static char *name, *base;
a6080a0a 244
6eec46bd 245 return sha1_get_pack_name(sha1, &name, &base, "idx");
bf592c50
DB
246}
247
d5a63b99
JH
248struct alternate_object_database *alt_odb_list;
249static struct alternate_object_database **alt_odb_tail;
ace1534d 250
f4e51683 251static int git_open_noatime(const char *name);
c2f493a4 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 */
6eac50d8 268static int link_alt_odb_entry(const char *entry, const char *relative_base, int depth)
ace1534d 269{
1494e038 270 const char *objdir = get_object_directory();
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) {
315 if (!memcmp(ent->base, alt->base, pfxlen)) {
316 free(ent);
317 return -1;
318 }
319 }
cb2912c3 320 if (!strcmp(ent->base, objdir)) {
c2f493a4
MW
321 free(ent);
322 return -1;
323 }
324
325 /* add the alternate entry */
326 *alt_odb_tail = ent;
327 alt_odb_tail = &(ent->next);
328 ent->next = NULL;
329
330 /* recursively add alternates */
331 read_info_alternates(ent->base, depth + 1);
332
333 ent->base[pfxlen] = '/';
334
335 return 0;
336}
337
c5950164 338static void link_alt_odb_entries(const char *alt, int len, int sep,
c2f493a4
MW
339 const char *relative_base, int depth)
340{
6eac50d8
MH
341 struct string_list entries = STRING_LIST_INIT_NODUP;
342 char *alt_copy;
343 int i;
c2f493a4
MW
344
345 if (depth > 5) {
346 error("%s: ignoring alternate object stores, nesting too deep.",
347 relative_base);
348 return;
349 }
350
c5950164 351 alt_copy = xmemdupz(alt, len);
6eac50d8
MH
352 string_list_split_in_place(&entries, alt_copy, sep, -1);
353 for (i = 0; i < entries.nr; i++) {
354 const char *entry = entries.items[i].string;
355 if (entry[0] == '\0' || entry[0] == '#')
9577e7e3 356 continue;
6eac50d8
MH
357 if (!is_absolute_path(entry) && depth) {
358 error("%s: ignoring relative alternate object store %s",
359 relative_base, entry);
360 } else {
361 link_alt_odb_entry(entry, relative_base, depth);
9577e7e3 362 }
9577e7e3 363 }
6eac50d8
MH
364 string_list_clear(&entries, 0);
365 free(alt_copy);
d5a63b99
JH
366}
367
5e73633d 368void read_info_alternates(const char * relative_base, int depth)
d5a63b99 369{
9577e7e3 370 char *map;
dc49cd76 371 size_t mapsz;
d5a63b99 372 struct stat st;
9cb18f56
JM
373 const char alt_file_name[] = "info/alternates";
374 /* Given that relative_base is no longer than PATH_MAX,
375 ensure that "path" has enough space to append "/", the
376 file name, "info/alternates", and a trailing NUL. */
377 char path[PATH_MAX + 1 + sizeof alt_file_name];
c2f493a4 378 int fd;
d5a63b99 379
9cb18f56 380 sprintf(path, "%s/%s", relative_base, alt_file_name);
f4e51683 381 fd = git_open_noatime(path);
d5a63b99
JH
382 if (fd < 0)
383 return;
384 if (fstat(fd, &st) || (st.st_size == 0)) {
385 close(fd);
9a217f2a 386 return;
ace1534d 387 }
dc49cd76
SP
388 mapsz = xsize_t(st.st_size);
389 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 390 close(fd);
d5a63b99 391
c5950164 392 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
c2f493a4 393
dc49cd76 394 munmap(map, mapsz);
ace1534d
JH
395}
396
bef70b22
DB
397void add_to_alternates_file(const char *reference)
398{
399 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
acd3b9ec 400 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
e6baf4a1 401 char *alt = mkpath("%s\n", reference);
bef70b22
DB
402 write_or_die(fd, alt, strlen(alt));
403 if (commit_lock_file(lock))
404 die("could not close alternates file");
405 if (alt_odb_tail)
c5950164 406 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
bef70b22
DB
407}
408
d79796bc
JH
409void foreach_alt_odb(alt_odb_fn fn, void *cb)
410{
411 struct alternate_object_database *ent;
412
413 prepare_alt_odb();
414 for (ent = alt_odb_list; ent; ent = ent->next)
415 if (fn(ent, cb))
416 return;
417}
418
c2f493a4
MW
419void prepare_alt_odb(void)
420{
554fe20d 421 const char *alt;
c2f493a4 422
7dc24aa5
SP
423 if (alt_odb_tail)
424 return;
425
c2f493a4
MW
426 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
427 if (!alt) alt = "";
428
c2f493a4 429 alt_odb_tail = &alt_odb_list;
c5950164 430 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
431
432 read_info_alternates(get_object_directory(), 0);
433}
434
0f4dc14a 435static int has_loose_object_local(const unsigned char *sha1)
ace1534d 436{
ace1534d 437 char *name = sha1_file_name(sha1);
0f4dc14a
BC
438 return !access(name, F_OK);
439}
ace1534d 440
0f4dc14a
BC
441int has_loose_object_nonlocal(const unsigned char *sha1)
442{
443 struct alternate_object_database *alt;
9a217f2a 444 prepare_alt_odb();
d5a63b99 445 for (alt = alt_odb_list; alt; alt = alt->next) {
0f4dc14a 446 fill_sha1_path(alt->name, sha1);
c529d75a
LT
447 if (!access(alt->base, F_OK))
448 return 1;
ace1534d 449 }
c529d75a 450 return 0;
ace1534d
JH
451}
452
0f4dc14a
BC
453static int has_loose_object(const unsigned char *sha1)
454{
455 return has_loose_object_local(sha1) ||
456 has_loose_object_nonlocal(sha1);
457}
458
60bb8b14 459static unsigned int pack_used_ctr;
a53128b6
SP
460static unsigned int pack_mmap_calls;
461static unsigned int peak_pack_open_windows;
462static unsigned int pack_open_windows;
c7934306
SP
463static unsigned int pack_open_fds;
464static unsigned int pack_max_fds;
a53128b6 465static size_t peak_pack_mapped;
60bb8b14 466static size_t pack_mapped;
9a217f2a 467struct packed_git *packed_git;
1f688557 468
b79d18c9 469void pack_report(void)
a53128b6
SP
470{
471 fprintf(stderr,
e05db0fd
PR
472 "pack_report: getpagesize() = %10" SZ_FMT "\n"
473 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
474 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
475 sz_fmt(getpagesize()),
476 sz_fmt(packed_git_window_size),
477 sz_fmt(packed_git_limit));
a53128b6
SP
478 fprintf(stderr,
479 "pack_report: pack_used_ctr = %10u\n"
480 "pack_report: pack_mmap_calls = %10u\n"
481 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
482 "pack_report: pack_mapped = "
483 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
484 pack_used_ctr,
485 pack_mmap_calls,
486 pack_open_windows, peak_pack_open_windows,
9e42d6a1 487 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
488}
489
42873078 490static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
491{
492 void *idx_map;
42873078 493 struct pack_idx_header *hdr;
dc49cd76 494 size_t idx_size;
74e34e1f 495 uint32_t version, nr, i, *index;
f4e51683 496 int fd = git_open_noatime(path);
1f688557 497 struct stat st;
42873078 498
1f688557
JH
499 if (fd < 0)
500 return -1;
501 if (fstat(fd, &st)) {
502 close(fd);
503 return -1;
504 }
dc49cd76 505 idx_size = xsize_t(st.st_size);
2d88451b
SP
506 if (idx_size < 4 * 256 + 20 + 20) {
507 close(fd);
508 return error("index file %s is too small", path);
509 }
c4712e45 510 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 511 close(fd);
1f688557 512
42873078
NP
513 hdr = idx_map;
514 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
515 version = ntohl(hdr->idx_version);
516 if (version < 2 || version > 2) {
517 munmap(idx_map, idx_size);
6e1c2344 518 return error("index file %s is version %"PRIu32
74e34e1f
NP
519 " and is not supported by this binary"
520 " (try upgrading GIT to a newer version)",
521 path, version);
522 }
523 } else
524 version = 1;
df1b059d 525
1f688557 526 nr = 0;
42873078 527 index = idx_map;
74e34e1f
NP
528 if (version > 1)
529 index += 2; /* skip index header */
1f688557 530 for (i = 0; i < 256; i++) {
326bf396 531 uint32_t n = ntohl(index[i]);
2d88451b
SP
532 if (n < nr) {
533 munmap(idx_map, idx_size);
df1b059d 534 return error("non-monotonic index %s", path);
2d88451b 535 }
1f688557
JH
536 nr = n;
537 }
538
74e34e1f
NP
539 if (version == 1) {
540 /*
541 * Total size:
542 * - 256 index entries 4 bytes each
543 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
544 * - 20-byte SHA1 of the packfile
545 * - 20-byte SHA1 file checksum
546 */
547 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
548 munmap(idx_map, idx_size);
eef427a0 549 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
550 }
551 } else if (version == 2) {
552 /*
553 * Minimum size:
554 * - 8 bytes of header
555 * - 256 index entries 4 bytes each
556 * - 20-byte sha1 entry * nr
557 * - 4-byte crc entry * nr
558 * - 4-byte offset entry * nr
559 * - 20-byte SHA1 of the packfile
560 * - 20-byte SHA1 file checksum
561 * And after the 4-byte offset table might be a
562 * variable sized table containing 8-byte entries
563 * for offsets larger than 2^31.
564 */
565 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
566 unsigned long max_size = min_size;
567 if (nr)
568 max_size += (nr - 1)*8;
569 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 570 munmap(idx_map, idx_size);
eef427a0 571 return error("wrong index v2 file size in %s", path);
74e34e1f 572 }
7109c889
JH
573 if (idx_size != min_size &&
574 /*
575 * make sure we can deal with large pack offsets.
576 * 31-bit signed offset won't be enough, neither
577 * 32-bit unsigned one will be.
578 */
579 (sizeof(off_t) <= 4)) {
580 munmap(idx_map, idx_size);
581 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 582 }
2d88451b 583 }
1f688557 584
74e34e1f 585 p->index_version = version;
42873078
NP
586 p->index_data = idx_map;
587 p->index_size = idx_size;
57059091 588 p->num_objects = nr;
1f688557
JH
589 return 0;
590}
591
bc8e478a 592int open_pack_index(struct packed_git *p)
d079837e
SP
593{
594 char *idx_name;
595 int ret;
596
597 if (p->index_data)
598 return 0;
599
600 idx_name = xstrdup(p->pack_name);
601 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
602 ret = check_packed_git_idx(idx_name, p);
603 free(idx_name);
604 return ret;
605}
606
11daf39b
SP
607static void scan_windows(struct packed_git *p,
608 struct packed_git **lru_p,
609 struct pack_window **lru_w,
610 struct pack_window **lru_l)
1f688557 611{
11daf39b
SP
612 struct pack_window *w, *w_l;
613
614 for (w_l = NULL, w = p->windows; w; w = w->next) {
615 if (!w->inuse_cnt) {
616 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
617 *lru_p = p;
618 *lru_w = w;
619 *lru_l = w_l;
54044bf8 620 }
54044bf8 621 }
11daf39b 622 w_l = w;
f9253394 623 }
11daf39b
SP
624}
625
7c3ecb32 626static int unuse_one_window(struct packed_git *current)
11daf39b
SP
627{
628 struct packed_git *p, *lru_p = NULL;
629 struct pack_window *lru_w = NULL, *lru_l = NULL;
630
631 if (current)
632 scan_windows(current, &lru_p, &lru_w, &lru_l);
633 for (p = packed_git; p; p = p->next)
634 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
635 if (lru_p) {
636 munmap(lru_w->base, lru_w->len);
637 pack_mapped -= lru_w->len;
638 if (lru_l)
639 lru_l->next = lru_w->next;
7c3ecb32 640 else
54044bf8 641 lru_p->windows = lru_w->next;
54044bf8 642 free(lru_w);
a53128b6 643 pack_open_windows--;
54044bf8
SP
644 return 1;
645 }
646 return 0;
f9253394
JH
647}
648
7c3ecb32 649void release_pack_memory(size_t need)
97bfeb34
SP
650{
651 size_t cur = pack_mapped;
7c3ecb32 652 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
97bfeb34
SP
653 ; /* nothing */
654}
655
58ecbd5e
JN
656void *xmmap(void *start, size_t length,
657 int prot, int flags, int fd, off_t offset)
658{
659 void *ret = mmap(start, length, prot, flags, fd, offset);
660 if (ret == MAP_FAILED) {
661 if (!length)
662 return NULL;
7c3ecb32 663 release_pack_memory(length);
58ecbd5e
JN
664 ret = mmap(start, length, prot, flags, fd, offset);
665 if (ret == MAP_FAILED)
666 die_errno("Out of memory? mmap failed");
667 }
668 return ret;
669}
670
c9ced051
SP
671void close_pack_windows(struct packed_git *p)
672{
673 while (p->windows) {
674 struct pack_window *w = p->windows;
675
676 if (w->inuse_cnt)
677 die("pack '%s' still has open windows to it",
678 p->pack_name);
679 munmap(w->base, w->len);
680 pack_mapped -= w->len;
681 pack_open_windows--;
682 p->windows = w->next;
683 free(w);
684 }
685}
686
88d0db55
BC
687/*
688 * The LRU pack is the one with the oldest MRU window, preferring packs
689 * with no used windows, or the oldest mtime if it has no windows allocated.
690 */
691static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
692{
693 struct pack_window *w, *this_mru_w;
694 int has_windows_inuse = 0;
695
696 /*
697 * Reject this pack if it has windows and the previously selected
698 * one does not. If this pack does not have windows, reject
699 * it if the pack file is newer than the previously selected one.
700 */
701 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
702 return;
703
704 for (w = this_mru_w = p->windows; w; w = w->next) {
705 /*
706 * Reject this pack if any of its windows are in use,
707 * but the previously selected pack did not have any
708 * inuse windows. Otherwise, record that this pack
709 * has windows in use.
710 */
711 if (w->inuse_cnt) {
712 if (*accept_windows_inuse)
713 has_windows_inuse = 1;
714 else
715 return;
716 }
717
718 if (w->last_used > this_mru_w->last_used)
719 this_mru_w = w;
720
721 /*
722 * Reject this pack if it has windows that have been
723 * used more recently than the previously selected pack.
724 * If the previously selected pack had windows inuse and
725 * we have not encountered a window in this pack that is
726 * inuse, skip this check since we prefer a pack with no
727 * inuse windows to one that has inuse windows.
728 */
729 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
730 this_mru_w->last_used > (*mru_w)->last_used)
731 return;
732 }
733
734 /*
735 * Select this pack.
736 */
737 *mru_w = this_mru_w;
738 *lru_p = p;
739 *accept_windows_inuse = has_windows_inuse;
740}
741
742static int close_one_pack(void)
743{
744 struct packed_git *p, *lru_p = NULL;
745 struct pack_window *mru_w = NULL;
746 int accept_windows_inuse = 1;
747
748 for (p = packed_git; p; p = p->next) {
749 if (p->pack_fd == -1)
750 continue;
751 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
752 }
753
754 if (lru_p) {
755 close(lru_p->pack_fd);
756 pack_open_fds--;
757 lru_p->pack_fd = -1;
758 return 1;
759 }
760
761 return 0;
762}
763
03e79c88 764void unuse_pack(struct pack_window **w_cursor)
f9253394 765{
03e79c88
SP
766 struct pack_window *w = *w_cursor;
767 if (w) {
768 w->inuse_cnt--;
769 *w_cursor = NULL;
770 }
1f688557
JH
771}
772
fa5fc15d
SP
773void close_pack_index(struct packed_git *p)
774{
775 if (p->index_data) {
776 munmap((void *)p->index_data, p->index_size);
777 p->index_data = NULL;
778 }
779}
780
c74faea1
NP
781/*
782 * This is used by git-repack in case a newly created pack happens to
783 * contain the same set of objects as an existing one. In that case
784 * the resulting file might be different even if its name would be the
785 * same. It is best to close any reference to the old pack before it is
786 * replaced on disk. Of course no index pointers nor windows for given pack
787 * must subsist at this point. If ever objects from this pack are requested
788 * again, the new version of the pack will be reinitialized through
789 * reprepare_packed_git().
790 */
791void free_pack_by_name(const char *pack_name)
792{
793 struct packed_git *p, **pp = &packed_git;
794
795 while (*pp) {
796 p = *pp;
797 if (strcmp(pack_name, p->pack_name) == 0) {
fa3a0c94 798 clear_delta_base_cache();
c74faea1 799 close_pack_windows(p);
c7934306 800 if (p->pack_fd != -1) {
c74faea1 801 close(p->pack_fd);
c7934306
SP
802 pack_open_fds--;
803 }
fa5fc15d 804 close_pack_index(p);
c74faea1
NP
805 free(p->bad_object_sha1);
806 *pp = p->next;
c01f51cc
NTND
807 if (last_found_pack == p)
808 last_found_pack = NULL;
c74faea1
NP
809 free(p);
810 return;
811 }
812 pp = &p->next;
813 }
814}
815
a0788266
JS
816static unsigned int get_max_fd_limit(void)
817{
818#ifdef RLIMIT_NOFILE
819 struct rlimit lim;
820
821 if (getrlimit(RLIMIT_NOFILE, &lim))
822 die_errno("cannot get RLIMIT_NOFILE");
823
824 return lim.rlim_cur;
825#elif defined(_SC_OPEN_MAX)
826 return sysconf(_SC_OPEN_MAX);
827#elif defined(OPEN_MAX)
828 return OPEN_MAX;
829#else
830 return 1; /* see the caller ;-) */
831#endif
832}
833
3cf8b462
SP
834/*
835 * Do not call this directly as this leaks p->pack_fd on error return;
836 * call open_packed_git() instead.
837 */
838static int open_packed_git_1(struct packed_git *p)
1f688557 839{
9bc879c1
SP
840 struct stat st;
841 struct pack_header hdr;
842 unsigned char sha1[20];
843 unsigned char *idx_sha1;
2c039da8 844 long fd_flag;
9bc879c1 845
d079837e
SP
846 if (!p->index_data && open_pack_index(p))
847 return error("packfile %s index unavailable", p->pack_name);
848
c7934306 849 if (!pack_max_fds) {
a0788266 850 unsigned int max_fds = get_max_fd_limit();
c7934306
SP
851
852 /* Save 3 for stdin/stdout/stderr, 22 for work */
853 if (25 < max_fds)
854 pack_max_fds = max_fds - 25;
855 else
856 pack_max_fds = 1;
857 }
858
88d0db55 859 while (pack_max_fds <= pack_open_fds && close_one_pack())
c7934306
SP
860 ; /* nothing */
861
f4e51683 862 p->pack_fd = git_open_noatime(p->pack_name);
9bc879c1 863 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 864 return -1;
c7934306 865 pack_open_fds++;
9bc879c1
SP
866
867 /* If we created the struct before we had the pack we lack size. */
bf592c50 868 if (!p->pack_size) {
bf592c50 869 if (!S_ISREG(st.st_mode))
072db278 870 return error("packfile %s not a regular file", p->pack_name);
bf592c50 871 p->pack_size = st.st_size;
9bc879c1 872 } else if (p->pack_size != st.st_size)
072db278 873 return error("packfile %s size changed", p->pack_name);
9bc879c1 874
2c039da8
JH
875 /* We leave these file descriptors open with sliding mmap;
876 * there is no point keeping them open across exec(), though.
877 */
878 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
879 if (fd_flag < 0)
072db278 880 return error("cannot determine file descriptor flags");
2c039da8
JH
881 fd_flag |= FD_CLOEXEC;
882 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 883 return error("cannot set FD_CLOEXEC");
2c039da8 884
9bc879c1 885 /* Verify we recognize this pack file format. */
e6e2bd62 886 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 887 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 888 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 889 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 890 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
891 return error("packfile %s is version %"PRIu32" and not"
892 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
893 p->pack_name, ntohl(hdr.hdr_version));
894
895 /* Verify the pack matches its index. */
57059091 896 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
897 return error("packfile %s claims to have %"PRIu32" objects"
898 " while index indicates %"PRIu32" objects",
57059091
NP
899 p->pack_name, ntohl(hdr.hdr_entries),
900 p->num_objects);
9bc879c1 901 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 902 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 903 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 904 return error("packfile %s signature is unavailable", p->pack_name);
42873078 905 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 906 if (hashcmp(sha1, idx_sha1))
072db278
SP
907 return error("packfile %s does not match index", p->pack_name);
908 return 0;
9bc879c1
SP
909}
910
3cf8b462
SP
911static int open_packed_git(struct packed_git *p)
912{
913 if (!open_packed_git_1(p))
914 return 0;
915 if (p->pack_fd != -1) {
916 close(p->pack_fd);
c7934306 917 pack_open_fds--;
3cf8b462
SP
918 p->pack_fd = -1;
919 }
920 return -1;
921}
922
c4001d92 923static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
924{
925 /* We must promise at least 20 bytes (one hash) after the
926 * offset is available from this window, otherwise the offset
927 * is not actually in this window and a different window (which
928 * has that one hash excess) must be used. This is to support
929 * the object header and delta base parsing routines below.
930 */
931 off_t win_off = win->offset;
932 return win_off <= offset
933 && (offset + 20) <= (win_off + win->len);
934}
935
4b25d091 936unsigned char *use_pack(struct packed_git *p,
03e79c88 937 struct pack_window **w_cursor,
c4001d92 938 off_t offset,
ef49a7a0 939 unsigned long *left)
9bc879c1 940{
60bb8b14 941 struct pack_window *win = *w_cursor;
03e79c88 942
a9d98a14 943 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
944 * pointless to ask for an offset into the middle of that
945 * hash, and the in_window function above wouldn't match
946 * don't allow an offset too close to the end of the file.
947 */
d131b7af
SP
948 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
949 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
950 if (offset > (p->pack_size - 20))
951 die("offset beyond end of packfile (truncated pack?)");
952
953 if (!win || !in_window(win, offset)) {
954 if (win)
955 win->inuse_cnt--;
956 for (win = p->windows; win; win = win->next) {
957 if (in_window(win, offset))
958 break;
959 }
960 if (!win) {
78a28df9 961 size_t window_align = packed_git_window_size / 2;
dc49cd76 962 off_t len;
d131b7af
SP
963
964 if (p->pack_fd == -1 && open_packed_git(p))
965 die("packfile %s cannot be accessed", p->pack_name);
966
60bb8b14 967 win = xcalloc(1, sizeof(*win));
78a28df9 968 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
969 len = p->pack_size - win->offset;
970 if (len > packed_git_window_size)
971 len = packed_git_window_size;
972 win->len = (size_t)len;
60bb8b14 973 pack_mapped += win->len;
11daf39b 974 while (packed_git_limit < pack_mapped
7c3ecb32 975 && unuse_one_window(p))
60bb8b14 976 ; /* nothing */
c4712e45 977 win->base = xmmap(NULL, win->len,
60bb8b14
SP
978 PROT_READ, MAP_PRIVATE,
979 p->pack_fd, win->offset);
980 if (win->base == MAP_FAILED)
73b4e4be
SP
981 die("packfile %s cannot be mapped: %s",
982 p->pack_name,
983 strerror(errno));
d131b7af
SP
984 if (!win->offset && win->len == p->pack_size
985 && !p->do_not_close) {
986 close(p->pack_fd);
987 pack_open_fds--;
988 p->pack_fd = -1;
989 }
a53128b6
SP
990 pack_mmap_calls++;
991 pack_open_windows++;
992 if (pack_mapped > peak_pack_mapped)
993 peak_pack_mapped = pack_mapped;
994 if (pack_open_windows > peak_pack_open_windows)
995 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
996 win->next = p->windows;
997 p->windows = win;
998 }
1f688557 999 }
03e79c88
SP
1000 if (win != *w_cursor) {
1001 win->last_used = pack_used_ctr++;
1002 win->inuse_cnt++;
1003 *w_cursor = win;
1004 }
60bb8b14 1005 offset -= win->offset;
03e79c88 1006 if (left)
dc49cd76 1007 *left = win->len - xsize_t(offset);
03e79c88 1008 return win->base + offset;
1f688557
JH
1009}
1010
27d69a46
NP
1011static struct packed_git *alloc_packed_git(int extra)
1012{
1013 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1014 memset(p, 0, sizeof(*p));
1015 p->pack_fd = -1;
1016 return p;
1017}
1018
e0500293
JN
1019static void try_to_free_pack_memory(size_t size)
1020{
7c3ecb32 1021 release_pack_memory(size);
e0500293
JN
1022}
1023
42873078 1024struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557 1025{
e0500293 1026 static int have_set_try_to_free_routine;
1f688557 1027 struct stat st;
27d69a46 1028 struct packed_git *p = alloc_packed_git(path_len + 2);
1f688557 1029
e0500293
JN
1030 if (!have_set_try_to_free_routine) {
1031 have_set_try_to_free_routine = 1;
1032 set_try_to_free_routine(try_to_free_pack_memory);
1033 }
1034
42873078
NP
1035 /*
1036 * Make sure a corresponding .pack file exists and that
1037 * the index looks sane.
1038 */
1039 path_len -= strlen(".idx");
27d69a46
NP
1040 if (path_len < 1) {
1041 free(p);
1f688557 1042 return NULL;
27d69a46 1043 }
42873078 1044 memcpy(p->pack_name, path, path_len);
8d25931d
BC
1045
1046 strcpy(p->pack_name + path_len, ".keep");
1047 if (!access(p->pack_name, F_OK))
1048 p->pack_keep = 1;
1049
42873078 1050 strcpy(p->pack_name + path_len, ".pack");
d079837e 1051 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 1052 free(p);
1f688557
JH
1053 return NULL;
1054 }
42873078 1055
1f688557
JH
1056 /* ok, it looks sane as far as we can check without
1057 * actually mapping the pack file.
1058 */
1f688557 1059 p->pack_size = st.st_size;
9d835df2 1060 p->pack_local = local;
b867092f 1061 p->mtime = st.st_mtime;
42873078
NP
1062 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1063 hashclr(p->sha1);
1f688557
JH
1064 return p;
1065}
1066
7b64469a 1067struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 1068{
42873078 1069 const char *path = sha1_pack_name(sha1);
27d69a46 1070 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
bf592c50 1071
27d69a46
NP
1072 strcpy(p->pack_name, path);
1073 hashcpy(p->sha1, sha1);
42873078
NP
1074 if (check_packed_git_idx(idx_path, p)) {
1075 free(p);
bf592c50 1076 return NULL;
42873078 1077 }
bf592c50 1078
bf592c50
DB
1079 return p;
1080}
1081
1082void install_packed_git(struct packed_git *pack)
1083{
c7934306
SP
1084 if (pack->pack_fd != -1)
1085 pack_open_fds++;
1086
bf592c50
DB
1087 pack->next = packed_git;
1088 packed_git = pack;
1089}
1090
543c5caa
NTND
1091void (*report_garbage)(const char *desc, const char *path);
1092
1093static void report_helper(const struct string_list *list,
1094 int seen_bits, int first, int last)
1095{
1096 const char *msg;
1097 switch (seen_bits) {
1098 case 0:
1099 msg = "no corresponding .idx nor .pack";
1100 break;
1101 case 1:
1102 msg = "no corresponding .idx";
1103 break;
1104 case 2:
1105 msg = "no corresponding .pack";
1106 break;
1107 default:
1108 return;
1109 }
1110 for (; first < last; first++)
1111 report_garbage(msg, list->items[first].string);
1112}
1113
1114static void report_pack_garbage(struct string_list *list)
1115{
1116 int i, baselen = -1, first = 0, seen_bits = 0;
1117
1118 if (!report_garbage)
1119 return;
1120
1121 sort_string_list(list);
1122
1123 for (i = 0; i < list->nr; i++) {
1124 const char *path = list->items[i].string;
1125 if (baselen != -1 &&
1126 strncmp(path, list->items[first].string, baselen)) {
1127 report_helper(list, seen_bits, first, i);
1128 baselen = -1;
1129 seen_bits = 0;
1130 }
1131 if (baselen == -1) {
1132 const char *dot = strrchr(path, '.');
1133 if (!dot) {
1134 report_garbage("garbage found", path);
1135 continue;
1136 }
1137 baselen = dot - path + 1;
1138 first = i;
1139 }
1140 if (!strcmp(path + baselen, "pack"))
1141 seen_bits |= 1;
1142 else if (!strcmp(path + baselen, "idx"))
1143 seen_bits |= 2;
1144 }
1145 report_helper(list, seen_bits, first, list->nr);
1146}
1147
9d835df2 1148static void prepare_packed_git_one(char *objdir, int local)
1f688557 1149{
9cb18f56
JM
1150 /* Ensure that this buffer is large enough so that we can
1151 append "/pack/" without clobbering the stack even if
1152 strlen(objdir) were PATH_MAX. */
1153 char path[PATH_MAX + 1 + 4 + 1 + 1];
1f688557
JH
1154 int len;
1155 DIR *dir;
1156 struct dirent *de;
543c5caa 1157 struct string_list garbage = STRING_LIST_INIT_DUP;
1f688557
JH
1158
1159 sprintf(path, "%s/pack", objdir);
1160 len = strlen(path);
1161 dir = opendir(path);
b5b16990 1162 if (!dir) {
26125f6b 1163 if (errno != ENOENT)
bd2afde8 1164 error("unable to open object pack directory: %s: %s",
26125f6b 1165 path, strerror(errno));
1f688557 1166 return;
b5b16990 1167 }
1f688557
JH
1168 path[len++] = '/';
1169 while ((de = readdir(dir)) != NULL) {
1170 int namelen = strlen(de->d_name);
1171 struct packed_git *p;
1172
543c5caa
NTND
1173 if (len + namelen + 1 > sizeof(path)) {
1174 if (report_garbage) {
1175 struct strbuf sb = STRBUF_INIT;
1176 strbuf_addf(&sb, "%.*s/%s", len - 1, path, de->d_name);
1177 report_garbage("path too long", sb.buf);
1178 strbuf_release(&sb);
1179 }
1180 continue;
1181 }
1182
1183 if (is_dot_or_dotdot(de->d_name))
9cb18f56
JM
1184 continue;
1185
1f688557 1186 strcpy(path + len, de->d_name);
d90906a9
NTND
1187
1188 if (has_extension(de->d_name, ".idx")) {
1189 /* Don't reopen a pack we already have. */
1190 for (p = packed_git; p; p = p->next) {
1191 if (!memcmp(path, p->pack_name, len + namelen - 4))
1192 break;
1193 }
1194 if (p == NULL &&
1195 /*
1196 * See if it really is a valid .idx file with
1197 * corresponding .pack file that we can map.
1198 */
1199 (p = add_packed_git(path, len + namelen, local)) != NULL)
1200 install_packed_git(p);
86f7780c 1201 }
543c5caa
NTND
1202
1203 if (!report_garbage)
1204 continue;
1205
1206 if (has_extension(de->d_name, ".idx") ||
1207 has_extension(de->d_name, ".pack") ||
1208 has_extension(de->d_name, ".keep"))
1209 string_list_append(&garbage, path);
1210 else
1211 report_garbage("garbage found", path);
1f688557 1212 }
5b35bcd5 1213 closedir(dir);
543c5caa
NTND
1214 report_pack_garbage(&garbage);
1215 string_list_clear(&garbage, 0);
1f688557
JH
1216}
1217
b867092f
JH
1218static int sort_pack(const void *a_, const void *b_)
1219{
1220 struct packed_git *a = *((struct packed_git **)a_);
1221 struct packed_git *b = *((struct packed_git **)b_);
1222 int st;
1223
1224 /*
1225 * Local packs tend to contain objects specific to our
1226 * variant of the project than remote ones. In addition,
1227 * remote ones could be on a network mounted filesystem.
1228 * Favor local ones for these reasons.
1229 */
1230 st = a->pack_local - b->pack_local;
1231 if (st)
1232 return -st;
1233
1234 /*
1235 * Younger packs tend to contain more recent objects,
1236 * and more recent objects tend to get accessed more
1237 * often.
1238 */
1239 if (a->mtime < b->mtime)
1240 return 1;
1241 else if (a->mtime == b->mtime)
1242 return 0;
1243 return -1;
1244}
1245
1246static void rearrange_packed_git(void)
1247{
1248 struct packed_git **ary, *p;
1249 int i, n;
1250
1251 for (n = 0, p = packed_git; p; p = p->next)
1252 n++;
1253 if (n < 2)
1254 return;
1255
1256 /* prepare an array of packed_git for easier sorting */
1257 ary = xcalloc(n, sizeof(struct packed_git *));
1258 for (n = 0, p = packed_git; p; p = p->next)
1259 ary[n++] = p;
1260
1261 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1262
1263 /* link them back again */
1264 for (i = 0; i < n - 1; i++)
1265 ary[i]->next = ary[i + 1];
1266 ary[n - 1]->next = NULL;
1267 packed_git = ary[0];
1268
1269 free(ary);
1270}
1271
637cdd9d 1272static int prepare_packed_git_run_once = 0;
9a217f2a 1273void prepare_packed_git(void)
1f688557 1274{
d5a63b99 1275 struct alternate_object_database *alt;
1f688557 1276
637cdd9d 1277 if (prepare_packed_git_run_once)
1f688557 1278 return;
9d835df2 1279 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 1280 prepare_alt_odb();
d5a63b99 1281 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 1282 alt->name[-1] = 0;
9d835df2 1283 prepare_packed_git_one(alt->base, 0);
1494e038 1284 alt->name[-1] = '/';
1f688557 1285 }
b867092f 1286 rearrange_packed_git();
637cdd9d
JK
1287 prepare_packed_git_run_once = 1;
1288}
1289
fc04c412 1290void reprepare_packed_git(void)
637cdd9d 1291{
4b480c67 1292 discard_revindex();
637cdd9d
JK
1293 prepare_packed_git_run_once = 0;
1294 prepare_packed_git();
1f688557
JH
1295}
1296
8eca0b47
NP
1297static void mark_bad_packed_object(struct packed_git *p,
1298 const unsigned char *sha1)
1299{
1300 unsigned i;
1301 for (i = 0; i < p->num_bad_objects; i++)
1302 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1303 return;
1304 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1305 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1306 p->num_bad_objects++;
1307}
1308
b6c4cecc 1309static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
ac939109
NP
1310{
1311 struct packed_git *p;
1312 unsigned i;
1313
1314 for (p = packed_git; p; p = p->next)
1315 for (i = 0; i < p->num_bad_objects; i++)
1316 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
b6c4cecc
JH
1317 return p;
1318 return NULL;
ac939109
NP
1319}
1320
090ea126
NTND
1321/*
1322 * With an in-core object data in "map", rehash it to make sure the
1323 * object name actually matches "sha1" to detect object corruption.
1324 * With "map" == NULL, try reading the object named with "sha1" using
1325 * the streaming interface and rehash it to do the same.
1326 */
1327int check_sha1_signature(const unsigned char *sha1, void *map,
1328 unsigned long size, const char *type)
0fcfd160
LT
1329{
1330 unsigned char real_sha1[20];
090ea126
NTND
1331 enum object_type obj_type;
1332 struct git_istream *st;
1333 git_SHA_CTX c;
1334 char hdr[32];
1335 int hdrlen;
1336
1337 if (map) {
1338 hash_sha1_file(map, size, type, real_sha1);
1339 return hashcmp(sha1, real_sha1) ? -1 : 0;
1340 }
1341
1342 st = open_istream(sha1, &obj_type, &size, NULL);
1343 if (!st)
1344 return -1;
1345
1346 /* Generate the header */
1347 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1348
1349 /* Sha1.. */
1350 git_SHA1_Init(&c);
1351 git_SHA1_Update(&c, hdr, hdrlen);
1352 for (;;) {
1353 char buf[1024 * 16];
1354 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1355
f54fac53
JK
1356 if (readlen < 0) {
1357 close_istream(st);
1358 return -1;
1359 }
090ea126
NTND
1360 if (!readlen)
1361 break;
1362 git_SHA1_Update(&c, buf, readlen);
1363 }
1364 git_SHA1_Final(real_sha1, &c);
1365 close_istream(st);
a89fccd2 1366 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1367}
1368
f4e51683 1369static int git_open_noatime(const char *name)
44d1c19e
LT
1370{
1371 static int sha1_file_open_flag = O_NOATIME;
44d1c19e 1372
f2e872aa
SP
1373 for (;;) {
1374 int fd = open(name, O_RDONLY | sha1_file_open_flag);
44d1c19e 1375 if (fd >= 0)
f2e872aa
SP
1376 return fd;
1377
f2e872aa
SP
1378 /* Might the failure be due to O_NOATIME? */
1379 if (errno != ENOENT && sha1_file_open_flag) {
44d1c19e 1380 sha1_file_open_flag = 0;
f2e872aa
SP
1381 continue;
1382 }
1383
1384 return -1;
44d1c19e 1385 }
44d1c19e
LT
1386}
1387
052fe5ea
JK
1388static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1389{
1390 char *name = sha1_file_name(sha1);
1391 struct alternate_object_database *alt;
1392
1393 if (!lstat(name, st))
1394 return 0;
1395
1396 prepare_alt_odb();
1397 errno = ENOENT;
1398 for (alt = alt_odb_list; alt; alt = alt->next) {
1399 name = alt->name;
1400 fill_sha1_path(name, sha1);
1401 if (!lstat(alt->base, st))
1402 return 0;
1403 }
1404
1405 return -1;
1406}
1407
44d1c19e
LT
1408static int open_sha1_file(const unsigned char *sha1)
1409{
1410 int fd;
1411 char *name = sha1_file_name(sha1);
1412 struct alternate_object_database *alt;
1413
f4e51683 1414 fd = git_open_noatime(name);
44d1c19e
LT
1415 if (fd >= 0)
1416 return fd;
1417
1418 prepare_alt_odb();
1419 errno = ENOENT;
1420 for (alt = alt_odb_list; alt; alt = alt->next) {
1421 name = alt->name;
1422 fill_sha1_path(name, sha1);
f4e51683 1423 fd = git_open_noatime(alt->base);
44d1c19e
LT
1424 if (fd >= 0)
1425 return fd;
1426 }
1427 return -1;
1428}
1429
f0270efd 1430void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1431{
0fcfd160 1432 void *map;
144bde78 1433 int fd;
ace1534d 1434
44d1c19e
LT
1435 fd = open_sha1_file(sha1);
1436 map = NULL;
1437 if (fd >= 0) {
1438 struct stat st;
0fcfd160 1439
44d1c19e
LT
1440 if (!fstat(fd, &st)) {
1441 *size = xsize_t(st.st_size);
33e42de0
MM
1442 if (!*size) {
1443 /* mmap() is forbidden on empty files */
1444 error("object file %s is empty", sha1_file_name(sha1));
1445 return NULL;
1446 }
44d1c19e 1447 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1448 }
44d1c19e 1449 close(fd);
0fcfd160 1450 }
0fcfd160
LT
1451 return map;
1452}
1453
cc5c54e7
JH
1454/*
1455 * There used to be a second loose object header format which
1456 * was meant to mimic the in-pack format, allowing for direct
1457 * copy of the object data. This format turned up not to be
1458 * really worth it and we no longer write loose objects in that
1459 * format.
1460 */
1461static int experimental_loose_object(unsigned char *map)
bb6b8e4f
JH
1462{
1463 unsigned int word;
1464
1465 /*
7f684a2a
RT
1466 * We must determine if the buffer contains the standard
1467 * zlib-deflated stream or the experimental format based
1468 * on the in-pack object format. Compare the header byte
1469 * for each format:
1470 *
1471 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1472 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1473 *
1474 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1475 * in standard loose-object format, UNLESS it is a Git-pack
1476 * format object *exactly* 8 bytes in size when inflated.
1477 *
1478 * However, RFC1950 also specifies that the 1st 16-bit word
1479 * must be divisible by 31 - this checksum tells us our buffer
1480 * is in the standard format, giving a false positive only if
1481 * the 1st word of the Git-pack format object happens to be
1482 * divisible by 31, ie:
1483 * ((byte0 * 256) + byte1) % 31 = 0
1484 * => 0ttt10000www1000 % 31 = 0
1485 *
1486 * As it happens, this case can only arise for www=3 & ttt=1
1487 * - ie, a Commit object, which would have to be 8 bytes in
1488 * size. As no Commit can be that small, we find that the
1489 * combination of these two criteria (bitmask & checksum)
1490 * can always correctly determine the buffer format.
bb6b8e4f
JH
1491 */
1492 word = (map[0] << 8) + map[1];
7f684a2a 1493 if ((map[0] & 0x8F) == 0x08 && !(word % 31))
bb6b8e4f 1494 return 0;
cc5c54e7
JH
1495 else
1496 return 1;
bb6b8e4f
JH
1497}
1498
09ded04b
NP
1499unsigned long unpack_object_header_buffer(const unsigned char *buf,
1500 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1501{
ad1ed5ee 1502 unsigned shift;
48fb7deb 1503 unsigned long size, c;
ad1ed5ee
JH
1504 unsigned long used = 0;
1505
1506 c = buf[used++];
1507 *type = (c >> 4) & 7;
1508 size = c & 15;
1509 shift = 4;
1510 while (c & 0x80) {
f630cfda 1511 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1512 error("bad object header");
ea4f9685
JH
1513 size = used = 0;
1514 break;
09ded04b 1515 }
ad1ed5ee
JH
1516 c = buf[used++];
1517 size += (c & 0x7f) << shift;
1518 shift += 7;
1519 }
1520 *sizep = size;
1521 return used;
1522}
1523
eb4f4076 1524int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
ad1ed5ee
JH
1525{
1526 unsigned long size, used;
1527 static const char valid_loose_object_type[8] = {
1528 0, /* OBJ_EXT */
1529 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1530 0, /* "delta" and others are invalid in a loose object */
93821bd9 1531 };
ad1ed5ee 1532 enum object_type type;
93821bd9 1533
c4483576
LT
1534 /* Get the data stream */
1535 memset(stream, 0, sizeof(*stream));
1536 stream->next_in = map;
1537 stream->avail_in = mapsize;
1538 stream->next_out = buffer;
93821bd9
LT
1539 stream->avail_out = bufsiz;
1540
cc5c54e7
JH
1541 if (experimental_loose_object(map)) {
1542 /*
1543 * The old experimental format we no longer produce;
1544 * we can still read it.
1545 */
1546 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1547 if (!used || !valid_loose_object_type[type])
1548 return -1;
1549 map += used;
1550 mapsize -= used;
726f852b 1551
cc5c54e7
JH
1552 /* Set up the stream for the rest.. */
1553 stream->next_in = map;
1554 stream->avail_in = mapsize;
1555 git_inflate_init(stream);
93821bd9 1556
cc5c54e7
JH
1557 /* And generate the fake traditional header */
1558 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1559 typename(type), size);
1560 return 0;
1561 }
39c68542 1562 git_inflate_init(stream);
cc5c54e7 1563 return git_inflate(stream, 0);
c4483576
LT
1564}
1565
ef49a7a0 1566static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1567{
1568 int bytes = strlen(buffer) + 1;
3aee68aa 1569 unsigned char *buf = xmallocz(size);
93821bd9 1570 unsigned long n;
7efbff75 1571 int status = Z_OK;
5180cacc 1572
93821bd9
LT
1573 n = stream->total_out - bytes;
1574 if (n > size)
1575 n = size;
1576 memcpy(buf, (char *) buffer + bytes, n);
1577 bytes = n;
456cdf6e
LT
1578 if (bytes <= size) {
1579 /*
1580 * The above condition must be (bytes <= size), not
1581 * (bytes < size). In other words, even though we
ccf5ace0 1582 * expect no more output and set avail_out to zero,
456cdf6e
LT
1583 * the input zlib stream may have bytes that express
1584 * "this concludes the stream", and we *do* want to
1585 * eat that input.
1586 *
1587 * Otherwise we would not be able to test that we
1588 * consumed all the input to reach the expected size;
1589 * we also want to check that zlib tells us that all
1590 * went well with status == Z_STREAM_END at the end.
1591 */
5180cacc
LT
1592 stream->next_out = buf + bytes;
1593 stream->avail_out = size - bytes;
7efbff75 1594 while (status == Z_OK)
39c68542 1595 status = git_inflate(stream, Z_FINISH);
5180cacc 1596 }
456cdf6e 1597 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1598 git_inflate_end(stream);
7efbff75
JH
1599 return buf;
1600 }
1601
1602 if (status < 0)
1603 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1604 else if (stream->avail_in)
1605 error("garbage at end of loose object '%s'",
1606 sha1_to_hex(sha1));
1607 free(buf);
1608 return NULL;
5180cacc
LT
1609}
1610
1611/*
1612 * We used to just use "sscanf()", but that's actually way
1613 * too permissive for what we want to check. So do an anal
1614 * object header parse by hand.
1615 */
f0270efd 1616int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1617{
21666f1a 1618 char type[10];
5180cacc
LT
1619 int i;
1620 unsigned long size;
1621
1622 /*
a6080a0a 1623 * The type can be at most ten bytes (including the
5180cacc 1624 * terminating '\0' that we add), and is followed by
21666f1a 1625 * a space.
5180cacc 1626 */
21666f1a 1627 i = 0;
5180cacc
LT
1628 for (;;) {
1629 char c = *hdr++;
1630 if (c == ' ')
1631 break;
21666f1a
NP
1632 type[i++] = c;
1633 if (i >= sizeof(type))
5180cacc 1634 return -1;
5180cacc 1635 }
21666f1a 1636 type[i] = 0;
5180cacc
LT
1637
1638 /*
1639 * The length must follow immediately, and be in canonical
1640 * decimal format (ie "010" is not valid).
1641 */
1642 size = *hdr++ - '0';
1643 if (size > 9)
1644 return -1;
1645 if (size) {
1646 for (;;) {
1647 unsigned long c = *hdr - '0';
1648 if (c > 9)
1649 break;
1650 hdr++;
1651 size = size * 10 + c;
1652 }
1653 }
1654 *sizep = size;
1655
1656 /*
1657 * The length must be followed by a zero byte
1658 */
21666f1a 1659 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1660}
1661
7efbff75 1662static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1663{
5180cacc 1664 int ret;
ef49a7a0 1665 git_zstream stream;
5180cacc 1666 char hdr[8192];
0fcfd160 1667
5180cacc 1668 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1669 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1670 return NULL;
1671
7efbff75 1672 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1673}
1674
54dab52a
NP
1675unsigned long get_size_from_delta(struct packed_git *p,
1676 struct pack_window **w_curs,
1677 off_t curpos)
1678{
1679 const unsigned char *data;
1680 unsigned char delta_head[20], *in;
ef49a7a0 1681 git_zstream stream;
54dab52a
NP
1682 int st;
1683
1684 memset(&stream, 0, sizeof(stream));
1685 stream.next_out = delta_head;
1686 stream.avail_out = sizeof(delta_head);
1687
39c68542 1688 git_inflate_init(&stream);
54dab52a
NP
1689 do {
1690 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1691 stream.next_in = in;
39c68542 1692 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1693 curpos += stream.next_in - in;
1694 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1695 stream.total_out < sizeof(delta_head));
39c68542 1696 git_inflate_end(&stream);
3d77d877
NP
1697 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1698 error("delta data unpack-initial failed");
1699 return 0;
1700 }
54dab52a
NP
1701
1702 /* Examine the initial part of the delta to figure out
1703 * the result size.
1704 */
1705 data = delta_head;
1706
1707 /* ignore base size */
1708 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1709
1710 /* Read the result size */
1711 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1712}
1713
c4001d92 1714static off_t get_delta_base(struct packed_git *p,
03e79c88 1715 struct pack_window **w_curs,
c4001d92 1716 off_t *curpos,
21666f1a 1717 enum object_type type,
c4001d92 1718 off_t delta_obj_offset)
eb32d236 1719{
2b87c45b 1720 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1721 off_t base_offset;
eb32d236 1722
8d8a4ea5
SP
1723 /* use_pack() assured us we have [base_info, base_info + 20)
1724 * as a range that we can look at without walking off the
1725 * end of the mapped window. Its actually the hash size
1726 * that is assured. An OFS_DELTA longer than the hash size
1727 * is stupid, as then a REF_DELTA would be smaller to store.
1728 */
21666f1a 1729 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1730 unsigned used = 0;
1731 unsigned char c = base_info[used++];
1732 base_offset = c & 127;
1733 while (c & 128) {
1734 base_offset += 1;
8723f216 1735 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1736 return 0; /* overflow */
eb32d236
NP
1737 c = base_info[used++];
1738 base_offset = (base_offset << 7) + (c & 127);
1739 }
1740 base_offset = delta_obj_offset - base_offset;
d8f32556 1741 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1742 return 0; /* out of bound */
2b87c45b 1743 *curpos += used;
21666f1a 1744 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1745 /* The base entry _must_ be in the same pack */
1746 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1747 *curpos += 20;
eb32d236
NP
1748 } else
1749 die("I am totally screwed");
2b87c45b 1750 return base_offset;
eb32d236
NP
1751}
1752
f8c8abc5
JH
1753int unpack_object_header(struct packed_git *p,
1754 struct pack_window **w_curs,
1755 off_t *curpos,
1756 unsigned long *sizep)
a733cb60 1757{
03e79c88 1758 unsigned char *base;
ef49a7a0 1759 unsigned long left;
ad1ed5ee 1760 unsigned long used;
2b87c45b 1761 enum object_type type;
a733cb60 1762
8d8a4ea5 1763 /* use_pack() assures us we have [base, base + 20) available
0353a0c4 1764 * as a range that we can look at. (Its actually the hash
3dff5379 1765 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1766 * the maximum deflated object size is 2^137, which is just
1767 * insane, so we know won't exceed what we have been given.
1768 */
2b87c45b 1769 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1770 used = unpack_object_header_buffer(base, left, &type, sizep);
1771 if (!used) {
1772 type = OBJ_BAD;
1773 } else
1774 *curpos += used;
ad1ed5ee 1775
2b87c45b 1776 return type;
a733cb60
LT
1777}
1778
790d96c0
TR
1779static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1780{
1781 int type;
1782 struct revindex_entry *revidx;
1783 const unsigned char *sha1;
1784 revidx = find_pack_revindex(p, obj_offset);
1785 if (!revidx)
1786 return OBJ_BAD;
1787 sha1 = nth_packed_object_sha1(p, revidx->nr);
1788 mark_bad_packed_object(p, sha1);
1789 type = sha1_object_info(sha1, NULL);
1790 if (type <= OBJ_NONE)
1791 return OBJ_BAD;
1792 return type;
1793}
1794
790d96c0
TR
1795#define POI_STACK_PREALLOC 64
1796
90191d37
JK
1797static enum object_type packed_to_object_type(struct packed_git *p,
1798 off_t obj_offset,
1799 enum object_type type,
1800 struct pack_window **w_curs,
1801 off_t curpos)
1f688557 1802{
790d96c0
TR
1803 off_t small_poi_stack[POI_STACK_PREALLOC];
1804 off_t *poi_stack = small_poi_stack;
1805 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
5db47c2b 1806
790d96c0
TR
1807 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1808 off_t base_offset;
90191d37 1809 unsigned long size;
790d96c0
TR
1810 /* Push the object we're going to leave behind */
1811 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1812 poi_stack_alloc = alloc_nr(poi_stack_nr);
1813 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1814 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1815 } else {
1816 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1817 }
1818 poi_stack[poi_stack_nr++] = obj_offset;
1819 /* If parsing the base offset fails, just unwind */
90191d37 1820 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
790d96c0
TR
1821 if (!base_offset)
1822 goto unwind;
1823 curpos = obj_offset = base_offset;
90191d37 1824 type = unpack_object_header(p, w_curs, &curpos, &size);
790d96c0
TR
1825 if (type <= OBJ_NONE) {
1826 /* If getting the base itself fails, we first
1827 * retry the base, otherwise unwind */
1828 type = retry_bad_packed_offset(p, base_offset);
1829 if (type > OBJ_NONE)
1830 goto out;
1831 goto unwind;
1832 }
1833 }
1834
21666f1a 1835 switch (type) {
790d96c0 1836 case OBJ_BAD:
a733cb60 1837 case OBJ_COMMIT:
a733cb60 1838 case OBJ_TREE:
a733cb60 1839 case OBJ_BLOB:
a733cb60 1840 case OBJ_TAG:
a69d0943 1841 break;
1f688557 1842 default:
3d77d877
NP
1843 error("unknown object type %i at offset %"PRIuMAX" in %s",
1844 type, (uintmax_t)obj_offset, p->pack_name);
1845 type = OBJ_BAD;
1f688557 1846 }
790d96c0
TR
1847
1848out:
1849 if (poi_stack != small_poi_stack)
1850 free(poi_stack);
21666f1a 1851 return type;
790d96c0
TR
1852
1853unwind:
1854 while (poi_stack_nr) {
1855 obj_offset = poi_stack[--poi_stack_nr];
1856 type = retry_bad_packed_offset(p, obj_offset);
1857 if (type > OBJ_NONE)
1858 goto out;
1859 }
1860 type = OBJ_BAD;
1861 goto out;
1f688557
JH
1862}
1863
90191d37 1864static int packed_object_info(struct packed_git *p, off_t obj_offset,
23c339c0 1865 struct object_info *oi)
90191d37
JK
1866{
1867 struct pack_window *w_curs = NULL;
1868 unsigned long size;
1869 off_t curpos = obj_offset;
1870 enum object_type type;
1871
412916ee
JK
1872 /*
1873 * We always get the representation type, but only convert it to
1874 * a "real" type later if the caller is interested.
1875 */
90191d37
JK
1876 type = unpack_object_header(p, &w_curs, &curpos, &size);
1877
23c339c0 1878 if (oi->sizep) {
90191d37
JK
1879 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1880 off_t tmp_pos = curpos;
1881 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1882 type, obj_offset);
1883 if (!base_offset) {
1884 type = OBJ_BAD;
1885 goto out;
1886 }
23c339c0
JK
1887 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1888 if (*oi->sizep == 0) {
90191d37
JK
1889 type = OBJ_BAD;
1890 goto out;
1891 }
1892 } else {
23c339c0 1893 *oi->sizep = size;
90191d37
JK
1894 }
1895 }
1896
23c339c0 1897 if (oi->disk_sizep) {
90191d37 1898 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
23c339c0 1899 *oi->disk_sizep = revidx[1].offset - obj_offset;
90191d37
JK
1900 }
1901
23c339c0
JK
1902 if (oi->typep) {
1903 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
1904 if (*oi->typep < 0) {
412916ee
JK
1905 type = OBJ_BAD;
1906 goto out;
1907 }
1908 }
90191d37
JK
1909
1910out:
1911 unuse_pack(&w_curs);
1912 return type;
1913}
1914
eb950c19 1915static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1916 struct pack_window **w_curs,
c4001d92 1917 off_t curpos,
eb950c19 1918 unsigned long size)
de530aaa
SP
1919{
1920 int st;
ef49a7a0 1921 git_zstream stream;
079afb18 1922 unsigned char *buffer, *in;
de530aaa 1923
4ab07e4d 1924 buffer = xmallocz(size);
de530aaa 1925 memset(&stream, 0, sizeof(stream));
de530aaa 1926 stream.next_out = buffer;
39eea7bd 1927 stream.avail_out = size + 1;
de530aaa 1928
39c68542 1929 git_inflate_init(&stream);
079afb18 1930 do {
2b87c45b 1931 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 1932 stream.next_in = in;
39c68542 1933 st = git_inflate(&stream, Z_FINISH);
39eea7bd
JH
1934 if (!stream.avail_out)
1935 break; /* the payload is larger than it should be */
2b87c45b 1936 curpos += stream.next_in - in;
079afb18 1937 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 1938 git_inflate_end(&stream);
de530aaa
SP
1939 if ((st != Z_STREAM_END) || stream.total_out != size) {
1940 free(buffer);
1941 return NULL;
1942 }
1943
1944 return buffer;
1945}
1946
e5e01619
LT
1947#define MAX_DELTA_CACHE (256)
1948
18bdec11 1949static size_t delta_base_cached;
5e08ecbf
NP
1950
1951static struct delta_base_cache_lru_list {
1952 struct delta_base_cache_lru_list *prev;
1953 struct delta_base_cache_lru_list *next;
1954} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1955
e5e01619 1956static struct delta_base_cache_entry {
5e08ecbf
NP
1957 struct delta_base_cache_lru_list lru;
1958 void *data;
e5e01619
LT
1959 struct packed_git *p;
1960 off_t base_offset;
1961 unsigned long size;
e5e01619
LT
1962 enum object_type type;
1963} delta_base_cache[MAX_DELTA_CACHE];
1964
1965static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1966{
1967 unsigned long hash;
1968
1969 hash = (unsigned long)p + (unsigned long)base_offset;
1970 hash += (hash >> 8) + (hash >> 16);
3358004a 1971 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1972}
1973
84dd81c1
TR
1974static struct delta_base_cache_entry *
1975get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
5266d369
JH
1976{
1977 unsigned long hash = pack_entry_hash(p, base_offset);
84dd81c1
TR
1978 return delta_base_cache + hash;
1979}
1980
1981static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
1982 struct packed_git *p, off_t base_offset)
1983{
5266d369
JH
1984 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1985}
1986
84dd81c1
TR
1987static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1988{
1989 struct delta_base_cache_entry *ent;
1990 ent = get_delta_base_cache_entry(p, base_offset);
1991 return eq_delta_base_cache_entry(ent, p, base_offset);
1992}
1993
1994static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1995{
1996 ent->data = NULL;
1997 ent->lru.next->prev = ent->lru.prev;
1998 ent->lru.prev->next = ent->lru.next;
1999 delta_base_cached -= ent->size;
2000}
2001
62f255ad 2002static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 2003 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 2004{
84dd81c1 2005 struct delta_base_cache_entry *ent;
e5e01619 2006 void *ret;
e5e01619 2007
84dd81c1
TR
2008 ent = get_delta_base_cache_entry(p, base_offset);
2009
2010 if (!eq_delta_base_cache_entry(ent, p, base_offset))
749bc58c 2011 return unpack_entry(p, base_offset, type, base_size);
e5e01619 2012
84dd81c1
TR
2013 ret = ent->data;
2014
2015 if (!keep_cache)
2016 clear_delta_base_cache_entry(ent);
2017 else
182af834 2018 ret = xmemdupz(ent->data, ent->size);
e5e01619
LT
2019 *type = ent->type;
2020 *base_size = ent->size;
2021 return ret;
62f255ad
LT
2022}
2023
18bdec11
SP
2024static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2025{
2026 if (ent->data) {
2027 free(ent->data);
2028 ent->data = NULL;
5e08ecbf
NP
2029 ent->lru.next->prev = ent->lru.prev;
2030 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
2031 delta_base_cached -= ent->size;
2032 }
2033}
2034
3d20c636
SP
2035void clear_delta_base_cache(void)
2036{
2037 unsigned long p;
2038 for (p = 0; p < MAX_DELTA_CACHE; p++)
2039 release_delta_base_cache(&delta_base_cache[p]);
2040}
2041
62f255ad
LT
2042static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2043 void *base, unsigned long base_size, enum object_type type)
2044{
5e08ecbf 2045 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 2046 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 2047 struct delta_base_cache_lru_list *lru;
e5e01619 2048
18bdec11
SP
2049 release_delta_base_cache(ent);
2050 delta_base_cached += base_size;
5e08ecbf
NP
2051
2052 for (lru = delta_base_cache_lru.next;
2053 delta_base_cached > delta_base_cache_limit
2054 && lru != &delta_base_cache_lru;
2055 lru = lru->next) {
2056 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
2057 if (f->type == OBJ_BLOB)
2058 release_delta_base_cache(f);
2059 }
5e08ecbf
NP
2060 for (lru = delta_base_cache_lru.next;
2061 delta_base_cached > delta_base_cache_limit
2062 && lru != &delta_base_cache_lru;
2063 lru = lru->next) {
2064 struct delta_base_cache_entry *f = (void *)lru;
2065 release_delta_base_cache(f);
2066 }
18bdec11 2067
e5e01619
LT
2068 ent->p = p;
2069 ent->base_offset = base_offset;
2070 ent->type = type;
2071 ent->data = base;
2072 ent->size = base_size;
5e08ecbf
NP
2073 ent->lru.next = &delta_base_cache_lru;
2074 ent->lru.prev = delta_base_cache_lru.prev;
2075 delta_base_cache_lru.prev->next = &ent->lru;
2076 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
2077}
2078
c2c5b270
CC
2079static void *read_object(const unsigned char *sha1, enum object_type *type,
2080 unsigned long *size);
2081
5f44324d
JH
2082static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2083{
2084 static FILE *log_file;
2085
b12ca963
NTND
2086 if (!log_pack_access)
2087 log_pack_access = getenv("GIT_TRACE_PACK_ACCESS");
2088 if (!log_pack_access)
2089 log_pack_access = no_log_pack_access;
2090 if (log_pack_access == no_log_pack_access)
2091 return;
2092
5f44324d
JH
2093 if (!log_file) {
2094 log_file = fopen(log_pack_access, "w");
2095 if (!log_file) {
2096 error("cannot open pack access log '%s' for writing: %s",
2097 log_pack_access, strerror(errno));
b12ca963 2098 log_pack_access = no_log_pack_access;
5f44324d
JH
2099 return;
2100 }
2101 }
2102 fprintf(log_file, "%s %"PRIuMAX"\n",
2103 p->pack_name, (uintmax_t)obj_offset);
2104 fflush(log_file);
2105}
2106
0e8189e2
NP
2107int do_check_packed_object_crc;
2108
abe601bb
TR
2109#define UNPACK_ENTRY_STACK_PREALLOC 64
2110struct unpack_entry_stack_ent {
2111 off_t obj_offset;
2112 off_t curpos;
2113 unsigned long size;
2114};
2115
c4001d92 2116void *unpack_entry(struct packed_git *p, off_t obj_offset,
abe601bb 2117 enum object_type *final_type, unsigned long *final_size)
1f688557 2118{
03e79c88 2119 struct pack_window *w_curs = NULL;
c4001d92 2120 off_t curpos = obj_offset;
abe601bb
TR
2121 void *data = NULL;
2122 unsigned long size;
2123 enum object_type type;
2124 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2125 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2126 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2127 int base_from_cache = 0;
1f688557 2128
b12ca963 2129 if (log_pack_access != no_log_pack_access)
5f44324d
JH
2130 write_pack_access_log(p, obj_offset);
2131
abe601bb
TR
2132 /* PHASE 1: drill down to the innermost base object */
2133 for (;;) {
2134 off_t base_offset;
2135 int i;
2136 struct delta_base_cache_entry *ent;
2137
77965f8b
NTND
2138 ent = get_delta_base_cache_entry(p, curpos);
2139 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2140 type = ent->type;
2141 data = ent->data;
2142 size = ent->size;
2143 clear_delta_base_cache_entry(ent);
2144 base_from_cache = 1;
2145 break;
2146 }
2147
abe601bb
TR
2148 if (do_check_packed_object_crc && p->index_version > 1) {
2149 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2150 unsigned long len = revidx[1].offset - obj_offset;
2151 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2152 const unsigned char *sha1 =
2153 nth_packed_object_sha1(p, revidx->nr);
2154 error("bad packed object CRC for %s",
2155 sha1_to_hex(sha1));
2156 mark_bad_packed_object(p, sha1);
2157 unuse_pack(&w_curs);
2158 return NULL;
2159 }
2160 }
2161
abe601bb
TR
2162 type = unpack_object_header(p, &w_curs, &curpos, &size);
2163 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2164 break;
2165
2166 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2167 if (!base_offset) {
2168 error("failed to validate delta base reference "
2169 "at offset %"PRIuMAX" from %s",
2170 (uintmax_t)curpos, p->pack_name);
2171 /* bail to phase 2, in hopes of recovery */
2172 data = NULL;
2173 break;
2174 }
2175
2176 /* push object, proceed to base */
2177 if (delta_stack_nr >= delta_stack_alloc
2178 && delta_stack == small_delta_stack) {
2179 delta_stack_alloc = alloc_nr(delta_stack_nr);
2180 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2181 memcpy(delta_stack, small_delta_stack,
2182 sizeof(*delta_stack)*delta_stack_nr);
2183 } else {
2184 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2185 }
2186 i = delta_stack_nr++;
2187 delta_stack[i].obj_offset = obj_offset;
2188 delta_stack[i].curpos = curpos;
2189 delta_stack[i].size = size;
2190
2191 curpos = obj_offset = base_offset;
0e8189e2
NP
2192 }
2193
abe601bb
TR
2194 /* PHASE 2: handle the base */
2195 switch (type) {
eb32d236
NP
2196 case OBJ_OFS_DELTA:
2197 case OBJ_REF_DELTA:
abe601bb
TR
2198 if (data)
2199 die("BUG in unpack_entry: left loop at a valid delta");
4d703a1a 2200 break;
a733cb60 2201 case OBJ_COMMIT:
a733cb60 2202 case OBJ_TREE:
a733cb60 2203 case OBJ_BLOB:
a733cb60 2204 case OBJ_TAG:
abe601bb
TR
2205 if (!base_from_cache)
2206 data = unpack_compressed_entry(p, &w_curs, curpos, size);
4d703a1a 2207 break;
1f688557 2208 default:
8eca0b47
NP
2209 data = NULL;
2210 error("unknown object type %i at offset %"PRIuMAX" in %s",
abe601bb
TR
2211 type, (uintmax_t)obj_offset, p->pack_name);
2212 }
2213
2214 /* PHASE 3: apply deltas in order */
2215
2216 /* invariants:
2217 * 'data' holds the base data, or NULL if there was corruption
2218 */
2219 while (delta_stack_nr) {
2220 void *delta_data;
2221 void *base = data;
2222 unsigned long delta_size, base_size = size;
2223 int i;
2224
2225 data = NULL;
2226
2227 if (base)
2228 add_delta_base_cache(p, obj_offset, base, base_size, type);
2229
2230 if (!base) {
2231 /*
2232 * We're probably in deep shit, but let's try to fetch
2233 * the required base anyway from another pack or loose.
2234 * This is costly but should happen only in the presence
2235 * of a corrupted pack, and is better than failing outright.
2236 */
2237 struct revindex_entry *revidx;
2238 const unsigned char *base_sha1;
2239 revidx = find_pack_revindex(p, obj_offset);
2240 if (revidx) {
2241 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2242 error("failed to read delta base object %s"
2243 " at offset %"PRIuMAX" from %s",
2244 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2245 p->pack_name);
2246 mark_bad_packed_object(p, base_sha1);
2247 base = read_object(base_sha1, &type, &base_size);
2248 }
2249 }
2250
2251 i = --delta_stack_nr;
2252 obj_offset = delta_stack[i].obj_offset;
2253 curpos = delta_stack[i].curpos;
2254 delta_size = delta_stack[i].size;
2255
2256 if (!base)
2257 continue;
2258
2259 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2260
2261 if (!delta_data) {
2262 error("failed to unpack compressed delta "
2263 "at offset %"PRIuMAX" from %s",
2264 (uintmax_t)curpos, p->pack_name);
abe601bb
TR
2265 data = NULL;
2266 continue;
2267 }
2268
2269 data = patch_delta(base, base_size,
2270 delta_data, delta_size,
2271 &size);
1ee886c1
JK
2272
2273 /*
2274 * We could not apply the delta; warn the user, but keep going.
2275 * Our failure will be noticed either in the next iteration of
2276 * the loop, or if this is the final delta, in the caller when
2277 * we return NULL. Those code paths will take care of making
2278 * a more explicit warning and retrying with another copy of
2279 * the object.
2280 */
abe601bb 2281 if (!data)
1ee886c1 2282 error("failed to apply delta");
abe601bb 2283
4b8f772c 2284 free(delta_data);
1f688557 2285 }
abe601bb
TR
2286
2287 *final_type = type;
2288 *final_size = size;
2289
03e79c88 2290 unuse_pack(&w_curs);
21666f1a 2291 return data;
1f688557
JH
2292}
2293
d079837e 2294const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 2295 uint32_t n)
9a217f2a 2296{
42873078 2297 const unsigned char *index = p->index_data;
d079837e
SP
2298 if (!index) {
2299 if (open_pack_index(p))
2300 return NULL;
2301 index = p->index_data;
2302 }
57059091 2303 if (n >= p->num_objects)
d72308e0 2304 return NULL;
74e34e1f
NP
2305 index += 4 * 256;
2306 if (p->index_version == 1) {
2307 return index + 24 * n + 4;
2308 } else {
2309 index += 8;
2310 return index + 20 * n;
2311 }
2312}
2313
99093238 2314off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
2315{
2316 const unsigned char *index = p->index_data;
2317 index += 4 * 256;
2318 if (p->index_version == 1) {
2319 return ntohl(*((uint32_t *)(index + 24 * n)));
2320 } else {
2321 uint32_t off;
2322 index += 8 + p->num_objects * (20 + 4);
2323 off = ntohl(*((uint32_t *)(index + 4 * n)));
2324 if (!(off & 0x80000000))
2325 return off;
2326 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2327 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2328 ntohl(*((uint32_t *)(index + 4)));
2329 }
9a217f2a
JH
2330}
2331
c4001d92 2332off_t find_pack_entry_one(const unsigned char *sha1,
43057304 2333 struct packed_git *p)
1f688557 2334{
42873078 2335 const uint32_t *level1_ofs = p->index_data;
42873078 2336 const unsigned char *index = p->index_data;
628522ec
JH
2337 unsigned hi, lo, stride;
2338 static int use_lookup = -1;
2339 static int debug_lookup = -1;
2340
2341 if (debug_lookup < 0)
2342 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 2343
d079837e
SP
2344 if (!index) {
2345 if (open_pack_index(p))
2346 return 0;
2347 level1_ofs = p->index_data;
2348 index = p->index_data;
2349 }
74e34e1f
NP
2350 if (p->index_version > 1) {
2351 level1_ofs += 2;
2352 index += 8;
2353 }
42873078 2354 index += 4 * 256;
74e34e1f
NP
2355 hi = ntohl(level1_ofs[*sha1]);
2356 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
2357 if (p->index_version > 1) {
2358 stride = 20;
2359 } else {
2360 stride = 24;
2361 index += 4;
2362 }
2363
2364 if (debug_lookup)
6e1c2344 2365 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
2366 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2367
2368 if (use_lookup < 0)
2369 use_lookup = !!getenv("GIT_USE_LOOKUP");
2370 if (use_lookup) {
2371 int pos = sha1_entry_pos(index, stride, 0,
2372 lo, hi, p->num_objects, sha1);
2373 if (pos < 0)
2374 return 0;
2375 return nth_packed_object_offset(p, pos);
2376 }
1f688557
JH
2377
2378 do {
74e34e1f 2379 unsigned mi = (lo + hi) / 2;
628522ec
JH
2380 int cmp = hashcmp(index + mi * stride, sha1);
2381
2382 if (debug_lookup)
2383 printf("lo %u hi %u rg %u mi %u\n",
2384 lo, hi, hi - lo, mi);
43057304 2385 if (!cmp)
74e34e1f 2386 return nth_packed_object_offset(p, mi);
1f688557
JH
2387 if (cmp > 0)
2388 hi = mi;
2389 else
2390 lo = mi+1;
2391 } while (lo < hi);
2392 return 0;
2393}
2394
4c080182 2395int is_pack_valid(struct packed_git *p)
d131b7af
SP
2396{
2397 /* An already open pack is known to be valid. */
2398 if (p->pack_fd != -1)
2399 return 1;
2400
2401 /* If the pack has one window completely covering the
2402 * file size, the pack is known to be valid even if
2403 * the descriptor is not currently open.
2404 */
2405 if (p->windows) {
2406 struct pack_window *w = p->windows;
2407
2408 if (!w->offset && w->len == p->pack_size)
2409 return 1;
2410 }
2411
2412 /* Force the pack to open to prove its valid. */
2413 return !open_packed_git(p);
2414}
2415
95099731
NTND
2416static int fill_pack_entry(const unsigned char *sha1,
2417 struct pack_entry *e,
2418 struct packed_git *p)
2419{
2420 off_t offset;
2421
2422 if (p->num_bad_objects) {
2423 unsigned i;
2424 for (i = 0; i < p->num_bad_objects; i++)
2425 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2426 return 0;
2427 }
2428
2429 offset = find_pack_entry_one(sha1, p);
2430 if (!offset)
2431 return 0;
2432
2433 /*
2434 * We are about to tell the caller where they can locate the
2435 * requested object. We better make sure the packfile is
2436 * still here and can be accessed before supplying that
2437 * answer, as it may have been deleted since the index was
2438 * loaded!
2439 */
2440 if (!is_pack_valid(p)) {
2441 warning("packfile %s cannot be accessed", p->pack_name);
2442 return 0;
2443 }
2444 e->offset = offset;
2445 e->p = p;
2446 hashcpy(e->sha1, sha1);
2447 return 1;
2448}
2449
4d6acb70 2450static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557
JH
2451{
2452 struct packed_git *p;
43057304 2453
1f688557 2454 prepare_packed_git();
f7c22cc6
NP
2455 if (!packed_git)
2456 return 0;
1f688557 2457
c01f51cc
NTND
2458 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2459 return 1;
8eca0b47 2460
c01f51cc
NTND
2461 for (p = packed_git; p; p = p->next) {
2462 if (p == last_found_pack || !fill_pack_entry(sha1, e, p))
2463 continue;
f7c22cc6 2464
c01f51cc
NTND
2465 last_found_pack = p;
2466 return 1;
2467 }
1f688557
JH
2468 return 0;
2469}
2470
a6080a0a 2471struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
2472 struct packed_git *packs)
2473{
2474 struct packed_git *p;
bf592c50
DB
2475
2476 for (p = packs; p; p = p->next) {
43057304 2477 if (find_pack_entry_one(sha1, p))
bf592c50
DB
2478 return p;
2479 }
2480 return NULL;
21666f1a 2481
bf592c50
DB
2482}
2483
052fe5ea 2484static int sha1_loose_object_info(const unsigned char *sha1,
23c339c0 2485 struct object_info *oi)
65c2e0c3 2486{
36e4d74a 2487 int status;
65c2e0c3
JH
2488 unsigned long mapsize, size;
2489 void *map;
ef49a7a0 2490 git_zstream stream;
d65a16f6 2491 char hdr[32];
65c2e0c3 2492
052fe5ea
JK
2493 /*
2494 * If we don't care about type or size, then we don't
4ef8d1dd
JH
2495 * need to look inside the object at all. Note that we
2496 * do not optimize out the stat call, even if the
2497 * caller doesn't care about the disk-size, since our
2498 * return value implicitly indicates whether the
2499 * object even exists.
052fe5ea 2500 */
23c339c0 2501 if (!oi->typep && !oi->sizep) {
4ef8d1dd
JH
2502 struct stat st;
2503 if (stat_sha1_file(sha1, &st) < 0)
2504 return -1;
2505 if (oi->disk_sizep)
23c339c0 2506 *oi->disk_sizep = st.st_size;
052fe5ea
JK
2507 return 0;
2508 }
2509
bb6b8e4f 2510 map = map_sha1_file(sha1, &mapsize);
f0df4ed5 2511 if (!map)
dbea72a8 2512 return -1;
23c339c0
JK
2513 if (oi->disk_sizep)
2514 *oi->disk_sizep = mapsize;
36e4d74a
JH
2515 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2516 status = error("unable to unpack %s header",
2517 sha1_to_hex(sha1));
21666f1a 2518 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 2519 status = error("unable to parse %s header", sha1_to_hex(sha1));
23c339c0
JK
2520 else if (oi->sizep)
2521 *oi->sizep = size;
39c68542 2522 git_inflate_end(&stream);
65c2e0c3 2523 munmap(map, mapsize);
23c339c0
JK
2524 if (oi->typep)
2525 *oi->typep = status;
052fe5ea 2526 return 0;
65c2e0c3
JH
2527}
2528
9a490590 2529int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
f0df4ed5 2530{
c4d9986f 2531 struct cached_object *co;
f0df4ed5 2532 struct pack_entry e;
5b086407 2533 int rtype;
f0df4ed5 2534
c4d9986f
NTND
2535 co = find_cached_object(sha1);
2536 if (co) {
5b086407
JK
2537 if (oi->typep)
2538 *(oi->typep) = co->type;
9a490590
JH
2539 if (oi->sizep)
2540 *(oi->sizep) = co->size;
161f00e7
JK
2541 if (oi->disk_sizep)
2542 *(oi->disk_sizep) = 0;
9a490590 2543 oi->whence = OI_CACHED;
5b086407 2544 return 0;
c4d9986f
NTND
2545 }
2546
cd673c1f 2547 if (!find_pack_entry(sha1, &e)) {
ddd63e64 2548 /* Most likely it's a loose object. */
23c339c0 2549 if (!sha1_loose_object_info(sha1, oi)) {
9a490590 2550 oi->whence = OI_LOOSE;
5b086407 2551 return 0;
9a490590 2552 }
ddd63e64
SG
2553
2554 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 2555 reprepare_packed_git();
cd673c1f 2556 if (!find_pack_entry(sha1, &e))
052fe5ea 2557 return -1;
f0df4ed5 2558 }
3d77d877 2559
23c339c0 2560 rtype = packed_object_info(e.p, e.offset, oi);
412916ee 2561 if (rtype < 0) {
3d77d877 2562 mark_bad_packed_object(e.p, sha1);
412916ee 2563 return sha1_object_info_extended(sha1, oi);
5266d369
JH
2564 } else if (in_delta_base_cache(e.p, e.offset)) {
2565 oi->whence = OI_DBCACHED;
9a490590
JH
2566 } else {
2567 oi->whence = OI_PACKED;
2568 oi->u.packed.offset = e.offset;
2569 oi->u.packed.pack = e.p;
2570 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2571 rtype == OBJ_OFS_DELTA);
3d77d877
NP
2572 }
2573
5b086407 2574 return 0;
f0df4ed5
JS
2575}
2576
3fc0dca9 2577/* returns enum object_type or negative */
9a490590
JH
2578int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2579{
5b086407 2580 enum object_type type;
d099b717 2581 struct object_info oi = {NULL};
9a490590 2582
5b086407 2583 oi.typep = &type;
9a490590 2584 oi.sizep = sizep;
5b086407
JK
2585 if (sha1_object_info_extended(sha1, &oi) < 0)
2586 return -1;
2587 return type;
9a490590
JH
2588}
2589
21666f1a
NP
2590static void *read_packed_sha1(const unsigned char *sha1,
2591 enum object_type *type, unsigned long *size)
1f688557
JH
2592{
2593 struct pack_entry e;
8eca0b47 2594 void *data;
1f688557 2595
cd673c1f 2596 if (!find_pack_entry(sha1, &e))
1f688557 2597 return NULL;
8eca0b47
NP
2598 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2599 if (!data) {
2600 /*
2601 * We're probably in deep shit, but let's try to fetch
2602 * the required object anyway from another pack or loose.
2603 * This should happen only in the presence of a corrupted
2604 * pack, and is better than failing outright.
2605 */
2606 error("failed to read object %s at offset %"PRIuMAX" from %s",
2607 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2608 mark_bad_packed_object(e.p, sha1);
ac939109 2609 data = read_object(sha1, type, size);
8eca0b47
NP
2610 }
2611 return data;
1f688557
JH
2612}
2613
21666f1a
NP
2614int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2615 unsigned char *sha1)
d66b37bb
JH
2616{
2617 struct cached_object *co;
2618
21666f1a 2619 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2620 if (has_sha1_file(sha1) || find_cached_object(sha1))
2621 return 0;
2622 if (cached_object_alloc <= cached_object_nr) {
2623 cached_object_alloc = alloc_nr(cached_object_alloc);
2624 cached_objects = xrealloc(cached_objects,
2625 sizeof(*cached_objects) *
2626 cached_object_alloc);
2627 }
2628 co = &cached_objects[cached_object_nr++];
2629 co->size = len;
21666f1a 2630 co->type = type;
efa13f7b
JH
2631 co->buf = xmalloc(len);
2632 memcpy(co->buf, buf, len);
d66b37bb
JH
2633 hashcpy(co->sha1, sha1);
2634 return 0;
2635}
2636
c2c5b270
CC
2637static void *read_object(const unsigned char *sha1, enum object_type *type,
2638 unsigned long *size)
0fcfd160
LT
2639{
2640 unsigned long mapsize;
2641 void *map, *buf;
d66b37bb
JH
2642 struct cached_object *co;
2643
2644 co = find_cached_object(sha1);
2645 if (co) {
21666f1a 2646 *type = co->type;
d66b37bb 2647 *size = co->size;
182af834 2648 return xmemdupz(co->buf, co->size);
d66b37bb 2649 }
0fcfd160 2650
8276c007
PE
2651 buf = read_packed_sha1(sha1, type, size);
2652 if (buf)
2653 return buf;
bb6b8e4f 2654 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2655 if (map) {
7efbff75 2656 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2657 munmap(map, mapsize);
2658 return buf;
2659 }
637cdd9d 2660 reprepare_packed_git();
8276c007 2661 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2662}
2663
b6c4cecc
JH
2664/*
2665 * This function dies on corrupt objects; the callers who want to
2666 * deal with them should arrange to call read_object() and give error
2667 * messages themselves.
2668 */
5bf29b95
JH
2669void *read_sha1_file_extended(const unsigned char *sha1,
2670 enum object_type *type,
2671 unsigned long *size,
2672 unsigned flag)
ac939109 2673{
3ba7a065 2674 void *data;
e8b15e61 2675 char *path;
b6c4cecc 2676 const struct packed_git *p;
5bf29b95
JH
2677 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2678 ? lookup_replace_object(sha1) : sha1;
b6c4cecc 2679
3ba7a065
JH
2680 errno = 0;
2681 data = read_object(repl, type, size);
4bbf5a26 2682 if (data)
b6c4cecc 2683 return data;
68095570 2684
25f3af3f 2685 if (errno && errno != ENOENT)
3ba7a065
JH
2686 die_errno("failed to read object %s", sha1_to_hex(sha1));
2687
68095570 2688 /* die if we replaced an object with one that does not exist */
b6c4cecc 2689 if (repl != sha1)
68095570
CC
2690 die("replacement %s not found for %s",
2691 sha1_to_hex(repl), sha1_to_hex(sha1));
2692
b6c4cecc
JH
2693 if (has_loose_object(repl)) {
2694 path = sha1_file_name(sha1);
2695 die("loose object %s (stored in %s) is corrupt",
2696 sha1_to_hex(repl), path);
e8b15e61 2697 }
68095570 2698
b6c4cecc
JH
2699 if ((p = has_packed_and_bad(repl)) != NULL)
2700 die("packed object %s (stored in %s) is corrupt",
2701 sha1_to_hex(repl), p->pack_name);
f5552aee 2702
b6c4cecc 2703 return NULL;
ac939109
NP
2704}
2705
40469ee9 2706void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2707 const char *required_type_name,
40469ee9
JH
2708 unsigned long *size,
2709 unsigned char *actual_sha1_return)
f4913f91 2710{
21666f1a 2711 enum object_type type, required_type;
f4913f91
JH
2712 void *buffer;
2713 unsigned long isize;
40469ee9 2714 unsigned char actual_sha1[20];
f4913f91 2715
21666f1a 2716 required_type = type_from_string(required_type_name);
e702496e 2717 hashcpy(actual_sha1, sha1);
40469ee9
JH
2718 while (1) {
2719 int ref_length = -1;
2720 const char *ref_type = NULL;
f4913f91 2721
21666f1a 2722 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2723 if (!buffer)
2724 return NULL;
21666f1a 2725 if (type == required_type) {
40469ee9
JH
2726 *size = isize;
2727 if (actual_sha1_return)
e702496e 2728 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2729 return buffer;
2730 }
2731 /* Handle references */
21666f1a 2732 else if (type == OBJ_COMMIT)
40469ee9 2733 ref_type = "tree ";
21666f1a 2734 else if (type == OBJ_TAG)
40469ee9
JH
2735 ref_type = "object ";
2736 else {
2737 free(buffer);
2738 return NULL;
2739 }
2740 ref_length = strlen(ref_type);
f4913f91 2741
50974ec9
MK
2742 if (ref_length + 40 > isize ||
2743 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2744 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2745 free(buffer);
2746 return NULL;
2747 }
1cf58e72 2748 free(buffer);
40469ee9
JH
2749 /* Now we have the ID of the referred-to object in
2750 * actual_sha1. Check again. */
f4913f91 2751 }
f4913f91
JH
2752}
2753
ce9fbf16 2754static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2755 const char *type, unsigned char *sha1,
d65a16f6 2756 char *hdr, int *hdrlen)
d410c0f5 2757{
9126f009 2758 git_SHA_CTX c;
d410c0f5
JH
2759
2760 /* Generate the header */
d65a16f6 2761 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
2762
2763 /* Sha1.. */
9126f009
NP
2764 git_SHA1_Init(&c);
2765 git_SHA1_Update(&c, hdr, *hdrlen);
2766 git_SHA1_Update(&c, buf, len);
2767 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2768}
2769
230f1322 2770/*
5a688fe4
JH
2771 * Move the just written object into its final resting place.
2772 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2773 * "moving" is only a part of what it does, when no patch between
2774 * master to pu changes the call sites of this function.
230f1322 2775 */
839837b9 2776int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322 2777{
e32c0a9c 2778 int ret = 0;
5a688fe4 2779
348df166 2780 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
2781 goto try_rename;
2782 else if (link(tmpfile, filename))
e32c0a9c 2783 ret = errno;
7ebb6fca
LT
2784
2785 /*
2786 * Coda hack - coda doesn't like cross-directory links,
2787 * so we fall back to a rename, which will mean that it
2788 * won't be able to check collisions, but that's not a
2789 * big deal.
2790 *
2791 * The same holds for FAT formatted media.
2792 *
3be1f18e 2793 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
2794 * left to unlink.
2795 */
2796 if (ret && ret != EEXIST) {
be66a6c4 2797 try_rename:
7ebb6fca 2798 if (!rename(tmpfile, filename))
3be1f18e 2799 goto out;
9e48b389 2800 ret = errno;
230f1322 2801 }
691f1a28 2802 unlink_or_warn(tmpfile);
230f1322
LT
2803 if (ret) {
2804 if (ret != EEXIST) {
82247e9b 2805 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
230f1322
LT
2806 }
2807 /* FIXME!!! Collision check here ? */
2808 }
2809
3be1f18e 2810out:
5256b006 2811 if (adjust_shared_perm(filename))
5a688fe4 2812 return error("unable to set permission to '%s'", filename);
230f1322
LT
2813 return 0;
2814}
2815
4d548150
LT
2816static int write_buffer(int fd, const void *buf, size_t len)
2817{
d34cf19b 2818 if (write_in_full(fd, buf, len) < 0)
93822c22 2819 return error("file write error (%s)", strerror(errno));
4d548150
LT
2820 return 0;
2821}
2822
ce9fbf16 2823int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2824 unsigned char *sha1)
2825{
d65a16f6 2826 char hdr[32];
abdc3fc8
RS
2827 int hdrlen;
2828 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2829 return 0;
2830}
2831
e9039dd3
LT
2832/* Finalize a file on disk, and close it. */
2833static void close_sha1_file(int fd)
2834{
aafe9fba
LT
2835 if (fsync_object_files)
2836 fsync_or_die(fd, "sha1 file");
e9039dd3 2837 if (close(fd) != 0)
d824cbba 2838 die_errno("error when closing sha1 file");
e9039dd3
LT
2839}
2840
5723fe7e
LT
2841/* Size of directory component, including the ending '/' */
2842static inline int directory_size(const char *filename)
2843{
2844 const char *s = strrchr(filename, '/');
2845 if (!s)
2846 return 0;
2847 return s - filename + 1;
2848}
2849
2850/*
2851 * This creates a temporary file in the same directory as the final
2852 * 'filename'
2853 *
2854 * We want to avoid cross-directory filename renames, because those
2855 * can have problems on various filesystems (FAT, NFS, Coda).
2856 */
2857static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2858{
2859 int fd, dirlen = directory_size(filename);
2860
2861 if (dirlen + 20 > bufsiz) {
2862 errno = ENAMETOOLONG;
2863 return -1;
2864 }
2865 memcpy(buffer, filename, dirlen);
2866 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
5256b006 2867 fd = git_mkstemp_mode(buffer, 0444);
cbacbf4e 2868 if (fd < 0 && dirlen && errno == ENOENT) {
5723fe7e 2869 /* Make sure the directory exists */
6ff6af62 2870 memcpy(buffer, filename, dirlen);
5723fe7e 2871 buffer[dirlen-1] = 0;
b2476a60
JH
2872 if (mkdir(buffer, 0777) && errno != EEXIST)
2873 return -1;
2874 if (adjust_shared_perm(buffer))
5723fe7e
LT
2875 return -1;
2876
2877 /* Try again */
2878 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
5256b006 2879 fd = git_mkstemp_mode(buffer, 0444);
5723fe7e
LT
2880 }
2881 return fd;
2882}
2883
bbac7311 2884static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 2885 const void *buf, unsigned long len, time_t mtime)
0fcfd160 2886{
915308b1 2887 int fd, ret;
9892beba 2888 unsigned char compressed[4096];
ef49a7a0 2889 git_zstream stream;
748af44c
NP
2890 git_SHA_CTX c;
2891 unsigned char parano_sha1[20];
706bc531 2892 char *filename;
ab1900a3 2893 static char tmp_file[PATH_MAX];
a44c9a5e 2894
972a9155 2895 filename = sha1_file_name(sha1);
ab1900a3 2896 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
aac17941 2897 if (fd < 0) {
35243577 2898 if (errno == EACCES)
82247e9b 2899 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
916d081b 2900 else
5eaeda70 2901 return error("unable to create temporary file: %s", strerror(errno));
aac17941
LT
2902 }
2903
0fcfd160
LT
2904 /* Set it up */
2905 memset(&stream, 0, sizeof(stream));
55bb5c91 2906 git_deflate_init(&stream, zlib_compression_level);
0fcfd160 2907 stream.next_out = compressed;
9892beba 2908 stream.avail_out = sizeof(compressed);
748af44c 2909 git_SHA1_Init(&c);
a44c9a5e
LT
2910
2911 /* First header.. */
d65a16f6 2912 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2913 stream.avail_in = hdrlen;
55bb5c91
JH
2914 while (git_deflate(&stream, 0) == Z_OK)
2915 ; /* nothing */
748af44c 2916 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
2917
2918 /* Then the data itself.. */
c00e657d 2919 stream.next_in = (void *)buf;
a44c9a5e 2920 stream.avail_in = len;
9892beba 2921 do {
748af44c 2922 unsigned char *in0 = stream.next_in;
55bb5c91 2923 ret = git_deflate(&stream, Z_FINISH);
748af44c 2924 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
2925 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2926 die("unable to write sha1 file");
2927 stream.next_out = compressed;
2928 stream.avail_out = sizeof(compressed);
2929 } while (ret == Z_OK);
2930
ac54c277
LT
2931 if (ret != Z_STREAM_END)
2932 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
55bb5c91 2933 ret = git_deflate_end_gently(&stream);
ac54c277
LT
2934 if (ret != Z_OK)
2935 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
2936 git_SHA1_Final(parano_sha1, &c);
2937 if (hashcmp(sha1, parano_sha1) != 0)
2938 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 2939
e9039dd3 2940 close_sha1_file(fd);
0fcfd160 2941
bbac7311
NP
2942 if (mtime) {
2943 struct utimbuf utb;
2944 utb.actime = mtime;
2945 utb.modtime = mtime;
ab1900a3 2946 if (utime(tmp_file, &utb) < 0)
bbac7311 2947 warning("failed utime() on %s: %s",
ab1900a3 2948 tmp_file, strerror(errno));
bbac7311
NP
2949 }
2950
ab1900a3 2951 return move_temp_to_file(tmp_file, filename);
0fcfd160 2952}
8237b185 2953
c00e657d 2954int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
bbac7311
NP
2955{
2956 unsigned char sha1[20];
2957 char hdr[32];
2958 int hdrlen;
2959
2960 /* Normally if we have it in the pack then we do not bother writing
2961 * it out into .git/objects/??/?{38} file.
2962 */
2963 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2964 if (returnsha1)
2965 hashcpy(returnsha1, sha1);
2966 if (has_sha1_file(sha1))
2967 return 0;
2968 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2969}
2970
2971int force_object_loose(const unsigned char *sha1, time_t mtime)
2972{
bbac7311
NP
2973 void *buf;
2974 unsigned long len;
2975 enum object_type type;
2976 char hdr[32];
2977 int hdrlen;
1fb23e65 2978 int ret;
bbac7311 2979
c529d75a 2980 if (has_loose_object(sha1))
bbac7311
NP
2981 return 0;
2982 buf = read_packed_sha1(sha1, &type, &len);
2983 if (!buf)
2984 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2985 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1fb23e65
BS
2986 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2987 free(buf);
2988
2989 return ret;
bbac7311
NP
2990}
2991
bf592c50
DB
2992int has_pack_index(const unsigned char *sha1)
2993{
2994 struct stat st;
2995 if (stat(sha1_pack_index_name(sha1), &st))
2996 return 0;
2997 return 1;
bf592c50
DB
2998}
2999
cd673c1f
JH
3000int has_sha1_pack(const unsigned char *sha1)
3001{
3002 struct pack_entry e;
3003 return find_pack_entry(sha1, &e);
3004}
3005
8237b185
DB
3006int has_sha1_file(const unsigned char *sha1)
3007{
1f688557
JH
3008 struct pack_entry e;
3009
cd673c1f 3010 if (find_pack_entry(sha1, &e))
1f688557 3011 return 1;
45e8a748
JK
3012 if (has_loose_object(sha1))
3013 return 1;
3014 reprepare_packed_git();
3015 return find_pack_entry(sha1, &e);
8237b185 3016}
74400e71 3017
c879daa2
NTND
3018static void check_tree(const void *buf, size_t size)
3019{
3020 struct tree_desc desc;
3021 struct name_entry entry;
3022
3023 init_tree_desc(&desc, buf, size);
3024 while (tree_entry(&desc, &entry))
3025 /* do nothing
3026 * tree_entry() will die() on malformed entries */
3027 ;
3028}
3029
3030static void check_commit(const void *buf, size_t size)
3031{
3032 struct commit c;
3033 memset(&c, 0, sizeof(c));
3034 if (parse_commit_buffer(&c, buf, size))
3035 die("corrupt commit");
3036}
3037
3038static void check_tag(const void *buf, size_t size)
3039{
3040 struct tag t;
3041 memset(&t, 0, sizeof(t));
3042 if (parse_tag_buffer(&t, buf, size))
3043 die("corrupt tag");
3044}
3045
43df4f86 3046static int index_mem(unsigned char *sha1, void *buf, size_t size,
c4ce46fc
JH
3047 enum object_type type,
3048 const char *path, unsigned flags)
e7332f96 3049{
6c510bee 3050 int ret, re_allocated = 0;
c4ce46fc 3051 int write_object = flags & HASH_WRITE_OBJECT;
74400e71 3052
7672db20 3053 if (!type)
edaec3fb 3054 type = OBJ_BLOB;
6c510bee
LT
3055
3056 /*
3057 * Convert blobs to git internal format
3058 */
43df4f86 3059 if ((type == OBJ_BLOB) && path) {
f285a2d7 3060 struct strbuf nbuf = STRBUF_INIT;
21e5ad50 3061 if (convert_to_git(path, buf, size, &nbuf,
5e12e78e 3062 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
b315c5c0 3063 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
3064 re_allocated = 1;
3065 }
3066 }
c4ce46fc 3067 if (flags & HASH_FORMAT_CHECK) {
c879daa2
NTND
3068 if (type == OBJ_TREE)
3069 check_tree(buf, size);
3070 if (type == OBJ_COMMIT)
3071 check_commit(buf, size);
3072 if (type == OBJ_TAG)
3073 check_tag(buf, size);
3074 }
6c510bee 3075
7672db20 3076 if (write_object)
edaec3fb 3077 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 3078 else
edaec3fb 3079 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 3080 if (re_allocated)
6c510bee 3081 free(buf);
43df4f86
DP
3082 return ret;
3083}
3084
7b41e1e1
JH
3085static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3086 const char *path, unsigned flags)
3087{
3088 struct strbuf sbuf = STRBUF_INIT;
3089 int ret;
3090
3091 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3092 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3093 else
3094 ret = -1;
3095 strbuf_release(&sbuf);
3096 return ret;
3097}
3098
ea68b0ce
DP
3099#define SMALL_FILE_SIZE (32*1024)
3100
7b41e1e1
JH
3101static int index_core(unsigned char *sha1, int fd, size_t size,
3102 enum object_type type, const char *path,
3103 unsigned flags)
43df4f86
DP
3104{
3105 int ret;
43df4f86 3106
7b41e1e1 3107 if (!size) {
c4ce46fc 3108 ret = index_mem(sha1, NULL, size, type, path, flags);
ea68b0ce
DP
3109 } else if (size <= SMALL_FILE_SIZE) {
3110 char *buf = xmalloc(size);
3111 if (size == read_in_full(fd, buf, size))
c4ce46fc 3112 ret = index_mem(sha1, buf, size, type, path, flags);
ea68b0ce
DP
3113 else
3114 ret = error("short read %s", strerror(errno));
3115 free(buf);
08bda208 3116 } else {
43df4f86 3117 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
c4ce46fc 3118 ret = index_mem(sha1, buf, size, type, path, flags);
aac17941 3119 munmap(buf, size);
08bda208 3120 }
7b41e1e1
JH
3121 return ret;
3122}
3123
4dd1fbc7 3124/*
568508e7
JH
3125 * This creates one packfile per large blob unless bulk-checkin
3126 * machinery is "plugged".
4dd1fbc7
JH
3127 *
3128 * This also bypasses the usual "convert-to-git" dance, and that is on
3129 * purpose. We could write a streaming version of the converting
3130 * functions and insert that before feeding the data to fast-import
4f22b101
JK
3131 * (or equivalent in-core API described above). However, that is
3132 * somewhat complicated, as we do not know the size of the filter
3133 * result, which we need to know beforehand when writing a git object.
3134 * Since the primary motivation for trying to stream from the working
3135 * tree file and to avoid mmaping it in core is to deal with large
3136 * binary blobs, they generally do not want to get any conversion, and
3137 * callers should avoid this code path when filters are requested.
4dd1fbc7
JH
3138 */
3139static int index_stream(unsigned char *sha1, int fd, size_t size,
3140 enum object_type type, const char *path,
3141 unsigned flags)
3142{
568508e7 3143 return index_bulk_checkin(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
3144}
3145
7b41e1e1
JH
3146int index_fd(unsigned char *sha1, int fd, struct stat *st,
3147 enum object_type type, const char *path, unsigned flags)
3148{
3149 int ret;
3150 size_t size = xsize_t(st->st_size);
3151
3152 if (!S_ISREG(st->st_mode))
3153 ret = index_pipe(sha1, fd, type, path, flags);
4f22b101
JK
3154 else if (size <= big_file_threshold || type != OBJ_BLOB ||
3155 (path && would_convert_to_git(path, NULL, 0, 0)))
7b41e1e1 3156 ret = index_core(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
3157 else
3158 ret = index_stream(sha1, fd, size, type, path, flags);
43df4f86 3159 close(fd);
aac17941 3160 return ret;
74400e71 3161}
ec1fcc16 3162
c4ce46fc 3163int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
ec1fcc16
JH
3164{
3165 int fd;
b760d3aa 3166 struct strbuf sb = STRBUF_INIT;
ec1fcc16
JH
3167
3168 switch (st->st_mode & S_IFMT) {
3169 case S_IFREG:
3170 fd = open(path, O_RDONLY);
3171 if (fd < 0)
3172 return error("open(\"%s\"): %s", path,
3173 strerror(errno));
c4ce46fc 3174 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
ec1fcc16
JH
3175 return error("%s: failed to insert into database",
3176 path);
3177 break;
3178 case S_IFLNK:
b760d3aa 3179 if (strbuf_readlink(&sb, path, st->st_size)) {
ec1fcc16 3180 char *errstr = strerror(errno);
ec1fcc16
JH
3181 return error("readlink(\"%s\"): %s", path,
3182 errstr);
3183 }
c4ce46fc 3184 if (!(flags & HASH_WRITE_OBJECT))
b760d3aa
LT
3185 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3186 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
ec1fcc16
JH
3187 return error("%s: failed to insert into database",
3188 path);
b760d3aa 3189 strbuf_release(&sb);
ec1fcc16 3190 break;
f35a6d3b
LT
3191 case S_IFDIR:
3192 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
3193 default:
3194 return error("%s: unsupported file type", path);
3195 }
3196 return 0;
3197}
a69e5429
JH
3198
3199int read_pack_header(int fd, struct pack_header *header)
3200{
c697ad14
HO
3201 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3202 /* "eof before pack header was fully read" */
3203 return PH_ERROR_EOF;
3204
a69e5429
JH
3205 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3206 /* "protocol error (pack signature mismatch detected)" */
3207 return PH_ERROR_PACK_SIGNATURE;
3208 if (!pack_version_ok(header->hdr_version))
3209 /* "protocol error (pack version unsupported)" */
3210 return PH_ERROR_PROTOCOL;
3211 return 0;
3212}
40d52ff7
JK
3213
3214void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3215{
3216 enum object_type type = sha1_object_info(sha1, NULL);
3217 if (type < 0)
3218 die("%s is not a valid object", sha1_to_hex(sha1));
3219 if (type != expect)
3220 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3221 typename(expect));
3222}