teach dry-run convert_to_git not to require a src buffer
[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"
1f688557 10#include "delta.h"
a733cb60 11#include "pack.h"
8e440259
PE
12#include "blob.h"
13#include "commit.h"
4dd1fbc7 14#include "run-command.h"
8e440259
PE
15#include "tag.h"
16#include "tree.h"
c879daa2 17#include "tree-walk.h"
f35a6d3b 18#include "refs.h"
70f5d5d3 19#include "pack-revindex.h"
628522ec 20#include "sha1-lookup.h"
0fcfd160 21
144bde78
LT
22#ifndef O_NOATIME
23#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
24#define O_NOATIME 01000000
25#else
26#define O_NOATIME 0
27#endif
28#endif
29
28bd70d8
JN
30#define SZ_FMT PRIuMAX
31static inline uintmax_t sz_fmt(size_t s) { return s; }
e05db0fd 32
96f1e58f 33const unsigned char null_sha1[20];
88cd621d 34
c597ba80
NTND
35/*
36 * This is meant to hold a *small* number of objects that you would
37 * want read_sha1_file() to be able to return, but yet you do not want
38 * to write them into the object store (e.g. a browse-only
39 * application).
40 */
41static struct cached_object {
42 unsigned char sha1[20];
43 enum object_type type;
44 void *buf;
45 unsigned long size;
46} *cached_objects;
47static int cached_object_nr, cached_object_alloc;
48
49static struct cached_object empty_tree = {
dab0d410 50 EMPTY_TREE_SHA1_BIN_LITERAL,
c597ba80
NTND
51 OBJ_TREE,
52 "",
53 0
54};
55
56static struct cached_object *find_cached_object(const unsigned char *sha1)
57{
58 int i;
59 struct cached_object *co = cached_objects;
60
61 for (i = 0; i < cached_object_nr; i++, co++) {
62 if (!hashcmp(co->sha1, sha1))
63 return co;
64 }
65 if (!hashcmp(sha1, empty_tree.sha1))
66 return &empty_tree;
67 return NULL;
68}
69
90a6464b
JH
70int mkdir_in_gitdir(const char *path)
71{
72 if (mkdir(path, 0777)) {
73 int saved_errno = errno;
74 struct stat st;
75 struct strbuf sb = STRBUF_INIT;
76
77 if (errno != EEXIST)
78 return -1;
79 /*
80 * Are we looking at a path in a symlinked worktree
81 * whose original repository does not yet have it?
82 * e.g. .git/rr-cache pointing at its original
83 * repository in which the user hasn't performed any
84 * conflict resolution yet?
85 */
86 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
87 strbuf_readlink(&sb, path, st.st_size) ||
88 !is_absolute_path(sb.buf) ||
89 mkdir(sb.buf, 0777)) {
90 strbuf_release(&sb);
91 errno = saved_errno;
92 return -1;
93 }
94 strbuf_release(&sb);
95 }
96 return adjust_shared_perm(path);
97}
98
b2cb9425
JH
99int safe_create_leading_directories(char *path)
100{
8385abfd 101 char *pos = path + offset_1st_component(path);
67d42212
JR
102 struct stat st;
103
b2cb9425
JH
104 while (pos) {
105 pos = strchr(pos, '/');
106 if (!pos)
107 break;
5f0bdf50
JH
108 while (*++pos == '/')
109 ;
110 if (!*pos)
111 break;
112 *--pos = '\0';
67d42212
JR
113 if (!stat(path, &st)) {
114 /* path exists */
115 if (!S_ISDIR(st.st_mode)) {
b2cb9425 116 *pos = '/';
67d42212 117 return -3;
b2cb9425 118 }
457f06d6 119 }
67d42212
JR
120 else if (mkdir(path, 0777)) {
121 *pos = '/';
122 return -1;
123 }
457f06d6
JS
124 else if (adjust_shared_perm(path)) {
125 *pos = '/';
126 return -2;
127 }
b2cb9425
JH
128 *pos++ = '/';
129 }
130 return 0;
131}
723c31fe 132
8e21d63b
JK
133int safe_create_leading_directories_const(const char *path)
134{
135 /* path points to cache entries, so xstrdup before messing with it */
136 char *buf = xstrdup(path);
137 int result = safe_create_leading_directories(buf);
138 free(buf);
139 return result;
140}
141
ace1534d
JH
142static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
143{
144 int i;
145 for (i = 0; i < 20; i++) {
146 static char hex[] = "0123456789abcdef";
147 unsigned int val = sha1[i];
148 char *pos = pathbuf + i*2 + (i > 0);
149 *pos++ = hex[val >> 4];
150 *pos = hex[val & 0xf];
151 }
152}
153
0fcfd160
LT
154/*
155 * NOTE! This returns a statically allocated buffer, so you have to be
790296fd 156 * careful about using it. Do an "xstrdup()" if you need to save the
0fcfd160 157 * filename.
ace1534d
JH
158 *
159 * Also note that this returns the location for creating. Reading
160 * SHA1 file can happen from any alternate directory listed in the
d19938ab 161 * DB_ENVIRONMENT environment variable if it is not found in
ace1534d 162 * the primary object database.
0fcfd160
LT
163 */
164char *sha1_file_name(const unsigned char *sha1)
165{
560fb6a1
JK
166 static char buf[PATH_MAX];
167 const char *objdir;
168 int len;
0fcfd160 169
560fb6a1
JK
170 objdir = get_object_directory();
171 len = strlen(objdir);
172
173 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
174 if (len + 43 > PATH_MAX)
175 die("insanely long object directory %s", objdir);
176 memcpy(buf, objdir, len);
177 buf[len] = '/';
178 buf[len+3] = '/';
179 buf[len+42] = '\0';
180 fill_sha1_path(buf + len + 1, sha1);
181 return buf;
0fcfd160
LT
182}
183
633f43e1 184static char *sha1_get_pack_name(const unsigned char *sha1,
6eec46bd 185 char **name, char **base, const char *which)
bf592c50
DB
186{
187 static const char hex[] = "0123456789abcdef";
633f43e1 188 char *buf;
bf592c50
DB
189 int i;
190
633f43e1 191 if (!*base) {
bf592c50
DB
192 const char *sha1_file_directory = get_object_directory();
193 int len = strlen(sha1_file_directory);
633f43e1 194 *base = xmalloc(len + 60);
6eec46bd
JH
195 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
196 sha1_file_directory, which);
633f43e1 197 *name = *base + len + 11;
bf592c50
DB
198 }
199
633f43e1 200 buf = *name;
bf592c50
DB
201
202 for (i = 0; i < 20; i++) {
203 unsigned int val = *sha1++;
204 *buf++ = hex[val >> 4];
205 *buf++ = hex[val & 0xf];
206 }
a6080a0a 207
633f43e1 208 return *base;
bf592c50
DB
209}
210
633f43e1 211char *sha1_pack_name(const unsigned char *sha1)
bf592c50 212{
633f43e1 213 static char *name, *base;
bf592c50 214
6eec46bd 215 return sha1_get_pack_name(sha1, &name, &base, "pack");
633f43e1 216}
bf592c50 217
633f43e1
HO
218char *sha1_pack_index_name(const unsigned char *sha1)
219{
220 static char *name, *base;
a6080a0a 221
6eec46bd 222 return sha1_get_pack_name(sha1, &name, &base, "idx");
bf592c50
DB
223}
224
d5a63b99
JH
225struct alternate_object_database *alt_odb_list;
226static struct alternate_object_database **alt_odb_tail;
ace1534d 227
c2f493a4 228static void read_info_alternates(const char * alternates, int depth);
f4e51683 229static int git_open_noatime(const char *name);
c2f493a4 230
ddd5d056
JH
231/*
232 * Prepare alternate object database registry.
d5a63b99
JH
233 *
234 * The variable alt_odb_list points at the list of struct
235 * alternate_object_database. The elements on this list come from
236 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
237 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
238 * whose contents is similar to that environment variable but can be
239 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
240 * contains "/the/directory/corresponding/to/.git/objects/...", while
241 * its name points just after the slash at the end of ".git/objects/"
242 * in the example above, and has enough space to hold 40-byte hex
243 * SHA1, an extra slash for the first level indirection, and the
244 * terminating NUL.
ddd5d056 245 */
c2f493a4 246static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
ace1534d 247{
1494e038 248 const char *objdir = get_object_directory();
c2f493a4
MW
249 struct alternate_object_database *ent;
250 struct alternate_object_database *alt;
5bdf0a84
HW
251 int pfxlen, entlen;
252 struct strbuf pathbuf = STRBUF_INIT;
d5a63b99 253
85dadc38 254 if (!is_absolute_path(entry) && relative_base) {
5bdf0a84
HW
255 strbuf_addstr(&pathbuf, real_path(relative_base));
256 strbuf_addch(&pathbuf, '/');
c2f493a4 257 }
5bdf0a84 258 strbuf_add(&pathbuf, entry, len);
c2f493a4 259
5bdf0a84
HW
260 normalize_path_copy(pathbuf.buf, pathbuf.buf);
261
262 pfxlen = strlen(pathbuf.buf);
263
264 /*
265 * The trailing slash after the directory name is given by
266 * this function at the end. Remove duplicates.
267 */
268 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
269 pfxlen -= 1;
270
271 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
272 ent = xmalloc(sizeof(*ent) + entlen);
273 memcpy(ent->base, pathbuf.buf, pfxlen);
274 strbuf_release(&pathbuf);
c2f493a4
MW
275
276 ent->name = ent->base + pfxlen + 1;
277 ent->base[pfxlen + 3] = '/';
278 ent->base[pfxlen] = ent->base[entlen-1] = 0;
279
280 /* Detect cases where alternate disappeared */
90b4a71c 281 if (!is_directory(ent->base)) {
c2f493a4
MW
282 error("object directory %s does not exist; "
283 "check .git/objects/info/alternates.",
284 ent->base);
285 free(ent);
286 return -1;
287 }
288
289 /* Prevent the common mistake of listing the same
290 * thing twice, or object directory itself.
291 */
292 for (alt = alt_odb_list; alt; alt = alt->next) {
293 if (!memcmp(ent->base, alt->base, pfxlen)) {
294 free(ent);
295 return -1;
296 }
297 }
298 if (!memcmp(ent->base, objdir, pfxlen)) {
299 free(ent);
300 return -1;
301 }
302
303 /* add the alternate entry */
304 *alt_odb_tail = ent;
305 alt_odb_tail = &(ent->next);
306 ent->next = NULL;
307
308 /* recursively add alternates */
309 read_info_alternates(ent->base, depth + 1);
310
311 ent->base[pfxlen] = '/';
312
313 return 0;
314}
315
316static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
317 const char *relative_base, int depth)
318{
319 const char *cp, *last;
320
321 if (depth > 5) {
322 error("%s: ignoring alternate object stores, nesting too deep.",
323 relative_base);
324 return;
325 }
326
d5a63b99 327 last = alt;
9577e7e3
JH
328 while (last < ep) {
329 cp = last;
330 if (cp < ep && *cp == '#') {
331 while (cp < ep && *cp != sep)
332 cp++;
333 last = cp + 1;
334 continue;
335 }
c2f493a4
MW
336 while (cp < ep && *cp != sep)
337 cp++;
d5a63b99 338 if (last != cp) {
85dadc38 339 if (!is_absolute_path(last) && depth) {
c2f493a4
MW
340 error("%s: ignoring relative alternate object store %s",
341 relative_base, last);
342 } else {
343 link_alt_odb_entry(last, cp - last,
344 relative_base, depth);
1494e038 345 }
d5a63b99 346 }
9577e7e3 347 while (cp < ep && *cp == sep)
d5a63b99
JH
348 cp++;
349 last = cp;
9577e7e3 350 }
d5a63b99
JH
351}
352
c2f493a4 353static void read_info_alternates(const char * relative_base, int depth)
d5a63b99 354{
9577e7e3 355 char *map;
dc49cd76 356 size_t mapsz;
d5a63b99 357 struct stat st;
9cb18f56
JM
358 const char alt_file_name[] = "info/alternates";
359 /* Given that relative_base is no longer than PATH_MAX,
360 ensure that "path" has enough space to append "/", the
361 file name, "info/alternates", and a trailing NUL. */
362 char path[PATH_MAX + 1 + sizeof alt_file_name];
c2f493a4 363 int fd;
d5a63b99 364
9cb18f56 365 sprintf(path, "%s/%s", relative_base, alt_file_name);
f4e51683 366 fd = git_open_noatime(path);
d5a63b99
JH
367 if (fd < 0)
368 return;
369 if (fstat(fd, &st) || (st.st_size == 0)) {
370 close(fd);
9a217f2a 371 return;
ace1534d 372 }
dc49cd76
SP
373 mapsz = xsize_t(st.st_size);
374 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 375 close(fd);
d5a63b99 376
dc49cd76 377 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
c2f493a4 378
dc49cd76 379 munmap(map, mapsz);
ace1534d
JH
380}
381
bef70b22
DB
382void add_to_alternates_file(const char *reference)
383{
384 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
acd3b9ec 385 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
e6baf4a1 386 char *alt = mkpath("%s\n", reference);
bef70b22
DB
387 write_or_die(fd, alt, strlen(alt));
388 if (commit_lock_file(lock))
389 die("could not close alternates file");
390 if (alt_odb_tail)
391 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
392}
393
d79796bc
JH
394void foreach_alt_odb(alt_odb_fn fn, void *cb)
395{
396 struct alternate_object_database *ent;
397
398 prepare_alt_odb();
399 for (ent = alt_odb_list; ent; ent = ent->next)
400 if (fn(ent, cb))
401 return;
402}
403
c2f493a4
MW
404void prepare_alt_odb(void)
405{
554fe20d 406 const char *alt;
c2f493a4 407
7dc24aa5
SP
408 if (alt_odb_tail)
409 return;
410
c2f493a4
MW
411 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
412 if (!alt) alt = "";
413
c2f493a4 414 alt_odb_tail = &alt_odb_list;
80ba074f 415 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
416
417 read_info_alternates(get_object_directory(), 0);
418}
419
0f4dc14a 420static int has_loose_object_local(const unsigned char *sha1)
ace1534d 421{
ace1534d 422 char *name = sha1_file_name(sha1);
0f4dc14a
BC
423 return !access(name, F_OK);
424}
ace1534d 425
0f4dc14a
BC
426int has_loose_object_nonlocal(const unsigned char *sha1)
427{
428 struct alternate_object_database *alt;
9a217f2a 429 prepare_alt_odb();
d5a63b99 430 for (alt = alt_odb_list; alt; alt = alt->next) {
0f4dc14a 431 fill_sha1_path(alt->name, sha1);
c529d75a
LT
432 if (!access(alt->base, F_OK))
433 return 1;
ace1534d 434 }
c529d75a 435 return 0;
ace1534d
JH
436}
437
0f4dc14a
BC
438static int has_loose_object(const unsigned char *sha1)
439{
440 return has_loose_object_local(sha1) ||
441 has_loose_object_nonlocal(sha1);
442}
443
60bb8b14 444static unsigned int pack_used_ctr;
a53128b6
SP
445static unsigned int pack_mmap_calls;
446static unsigned int peak_pack_open_windows;
447static unsigned int pack_open_windows;
c7934306
SP
448static unsigned int pack_open_fds;
449static unsigned int pack_max_fds;
a53128b6 450static size_t peak_pack_mapped;
60bb8b14 451static size_t pack_mapped;
9a217f2a 452struct packed_git *packed_git;
1f688557 453
b79d18c9 454void pack_report(void)
a53128b6
SP
455{
456 fprintf(stderr,
e05db0fd
PR
457 "pack_report: getpagesize() = %10" SZ_FMT "\n"
458 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
459 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
460 sz_fmt(getpagesize()),
461 sz_fmt(packed_git_window_size),
462 sz_fmt(packed_git_limit));
a53128b6
SP
463 fprintf(stderr,
464 "pack_report: pack_used_ctr = %10u\n"
465 "pack_report: pack_mmap_calls = %10u\n"
466 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
467 "pack_report: pack_mapped = "
468 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
469 pack_used_ctr,
470 pack_mmap_calls,
471 pack_open_windows, peak_pack_open_windows,
9e42d6a1 472 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
473}
474
42873078 475static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
476{
477 void *idx_map;
42873078 478 struct pack_idx_header *hdr;
dc49cd76 479 size_t idx_size;
74e34e1f 480 uint32_t version, nr, i, *index;
f4e51683 481 int fd = git_open_noatime(path);
1f688557 482 struct stat st;
42873078 483
1f688557
JH
484 if (fd < 0)
485 return -1;
486 if (fstat(fd, &st)) {
487 close(fd);
488 return -1;
489 }
dc49cd76 490 idx_size = xsize_t(st.st_size);
2d88451b
SP
491 if (idx_size < 4 * 256 + 20 + 20) {
492 close(fd);
493 return error("index file %s is too small", path);
494 }
c4712e45 495 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 496 close(fd);
1f688557 497
42873078
NP
498 hdr = idx_map;
499 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
500 version = ntohl(hdr->idx_version);
501 if (version < 2 || version > 2) {
502 munmap(idx_map, idx_size);
6e1c2344 503 return error("index file %s is version %"PRIu32
74e34e1f
NP
504 " and is not supported by this binary"
505 " (try upgrading GIT to a newer version)",
506 path, version);
507 }
508 } else
509 version = 1;
df1b059d 510
1f688557 511 nr = 0;
42873078 512 index = idx_map;
74e34e1f
NP
513 if (version > 1)
514 index += 2; /* skip index header */
1f688557 515 for (i = 0; i < 256; i++) {
326bf396 516 uint32_t n = ntohl(index[i]);
2d88451b
SP
517 if (n < nr) {
518 munmap(idx_map, idx_size);
df1b059d 519 return error("non-monotonic index %s", path);
2d88451b 520 }
1f688557
JH
521 nr = n;
522 }
523
74e34e1f
NP
524 if (version == 1) {
525 /*
526 * Total size:
527 * - 256 index entries 4 bytes each
528 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
529 * - 20-byte SHA1 of the packfile
530 * - 20-byte SHA1 file checksum
531 */
532 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
533 munmap(idx_map, idx_size);
eef427a0 534 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
535 }
536 } else if (version == 2) {
537 /*
538 * Minimum size:
539 * - 8 bytes of header
540 * - 256 index entries 4 bytes each
541 * - 20-byte sha1 entry * nr
542 * - 4-byte crc entry * nr
543 * - 4-byte offset entry * nr
544 * - 20-byte SHA1 of the packfile
545 * - 20-byte SHA1 file checksum
546 * And after the 4-byte offset table might be a
547 * variable sized table containing 8-byte entries
548 * for offsets larger than 2^31.
549 */
550 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
551 unsigned long max_size = min_size;
552 if (nr)
553 max_size += (nr - 1)*8;
554 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 555 munmap(idx_map, idx_size);
eef427a0 556 return error("wrong index v2 file size in %s", path);
74e34e1f 557 }
7109c889
JH
558 if (idx_size != min_size &&
559 /*
560 * make sure we can deal with large pack offsets.
561 * 31-bit signed offset won't be enough, neither
562 * 32-bit unsigned one will be.
563 */
564 (sizeof(off_t) <= 4)) {
565 munmap(idx_map, idx_size);
566 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 567 }
2d88451b 568 }
1f688557 569
74e34e1f 570 p->index_version = version;
42873078
NP
571 p->index_data = idx_map;
572 p->index_size = idx_size;
57059091 573 p->num_objects = nr;
1f688557
JH
574 return 0;
575}
576
bc8e478a 577int open_pack_index(struct packed_git *p)
d079837e
SP
578{
579 char *idx_name;
580 int ret;
581
582 if (p->index_data)
583 return 0;
584
585 idx_name = xstrdup(p->pack_name);
586 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
587 ret = check_packed_git_idx(idx_name, p);
588 free(idx_name);
589 return ret;
590}
591
11daf39b
SP
592static void scan_windows(struct packed_git *p,
593 struct packed_git **lru_p,
594 struct pack_window **lru_w,
595 struct pack_window **lru_l)
1f688557 596{
11daf39b
SP
597 struct pack_window *w, *w_l;
598
599 for (w_l = NULL, w = p->windows; w; w = w->next) {
600 if (!w->inuse_cnt) {
601 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
602 *lru_p = p;
603 *lru_w = w;
604 *lru_l = w_l;
54044bf8 605 }
54044bf8 606 }
11daf39b 607 w_l = w;
f9253394 608 }
11daf39b
SP
609}
610
d1efefa4 611static int unuse_one_window(struct packed_git *current, int keep_fd)
11daf39b
SP
612{
613 struct packed_git *p, *lru_p = NULL;
614 struct pack_window *lru_w = NULL, *lru_l = NULL;
615
616 if (current)
617 scan_windows(current, &lru_p, &lru_w, &lru_l);
618 for (p = packed_git; p; p = p->next)
619 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
620 if (lru_p) {
621 munmap(lru_w->base, lru_w->len);
622 pack_mapped -= lru_w->len;
623 if (lru_l)
624 lru_l->next = lru_w->next;
625 else {
626 lru_p->windows = lru_w->next;
d131b7af
SP
627 if (!lru_p->windows && lru_p->pack_fd != -1
628 && lru_p->pack_fd != keep_fd) {
54044bf8 629 close(lru_p->pack_fd);
c7934306 630 pack_open_fds--;
54044bf8
SP
631 lru_p->pack_fd = -1;
632 }
633 }
634 free(lru_w);
a53128b6 635 pack_open_windows--;
54044bf8
SP
636 return 1;
637 }
638 return 0;
f9253394
JH
639}
640
d1efefa4 641void release_pack_memory(size_t need, int fd)
97bfeb34
SP
642{
643 size_t cur = pack_mapped;
d1efefa4 644 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
97bfeb34
SP
645 ; /* nothing */
646}
647
58ecbd5e
JN
648void *xmmap(void *start, size_t length,
649 int prot, int flags, int fd, off_t offset)
650{
651 void *ret = mmap(start, length, prot, flags, fd, offset);
652 if (ret == MAP_FAILED) {
653 if (!length)
654 return NULL;
655 release_pack_memory(length, fd);
656 ret = mmap(start, length, prot, flags, fd, offset);
657 if (ret == MAP_FAILED)
658 die_errno("Out of memory? mmap failed");
659 }
660 return ret;
661}
662
c9ced051
SP
663void close_pack_windows(struct packed_git *p)
664{
665 while (p->windows) {
666 struct pack_window *w = p->windows;
667
668 if (w->inuse_cnt)
669 die("pack '%s' still has open windows to it",
670 p->pack_name);
671 munmap(w->base, w->len);
672 pack_mapped -= w->len;
673 pack_open_windows--;
674 p->windows = w->next;
675 free(w);
676 }
677}
678
03e79c88 679void unuse_pack(struct pack_window **w_cursor)
f9253394 680{
03e79c88
SP
681 struct pack_window *w = *w_cursor;
682 if (w) {
683 w->inuse_cnt--;
684 *w_cursor = NULL;
685 }
1f688557
JH
686}
687
fa5fc15d
SP
688void close_pack_index(struct packed_git *p)
689{
690 if (p->index_data) {
691 munmap((void *)p->index_data, p->index_size);
692 p->index_data = NULL;
693 }
694}
695
c74faea1
NP
696/*
697 * This is used by git-repack in case a newly created pack happens to
698 * contain the same set of objects as an existing one. In that case
699 * the resulting file might be different even if its name would be the
700 * same. It is best to close any reference to the old pack before it is
701 * replaced on disk. Of course no index pointers nor windows for given pack
702 * must subsist at this point. If ever objects from this pack are requested
703 * again, the new version of the pack will be reinitialized through
704 * reprepare_packed_git().
705 */
706void free_pack_by_name(const char *pack_name)
707{
708 struct packed_git *p, **pp = &packed_git;
709
710 while (*pp) {
711 p = *pp;
712 if (strcmp(pack_name, p->pack_name) == 0) {
fa3a0c94 713 clear_delta_base_cache();
c74faea1 714 close_pack_windows(p);
c7934306 715 if (p->pack_fd != -1) {
c74faea1 716 close(p->pack_fd);
c7934306
SP
717 pack_open_fds--;
718 }
fa5fc15d 719 close_pack_index(p);
c74faea1
NP
720 free(p->bad_object_sha1);
721 *pp = p->next;
722 free(p);
723 return;
724 }
725 pp = &p->next;
726 }
727}
728
3cf8b462
SP
729/*
730 * Do not call this directly as this leaks p->pack_fd on error return;
731 * call open_packed_git() instead.
732 */
733static int open_packed_git_1(struct packed_git *p)
1f688557 734{
9bc879c1
SP
735 struct stat st;
736 struct pack_header hdr;
737 unsigned char sha1[20];
738 unsigned char *idx_sha1;
2c039da8 739 long fd_flag;
9bc879c1 740
d079837e
SP
741 if (!p->index_data && open_pack_index(p))
742 return error("packfile %s index unavailable", p->pack_name);
743
c7934306
SP
744 if (!pack_max_fds) {
745 struct rlimit lim;
746 unsigned int max_fds;
747
748 if (getrlimit(RLIMIT_NOFILE, &lim))
749 die_errno("cannot get RLIMIT_NOFILE");
750
751 max_fds = lim.rlim_cur;
752
753 /* Save 3 for stdin/stdout/stderr, 22 for work */
754 if (25 < max_fds)
755 pack_max_fds = max_fds - 25;
756 else
757 pack_max_fds = 1;
758 }
759
760 while (pack_max_fds <= pack_open_fds && unuse_one_window(NULL, -1))
761 ; /* nothing */
762
f4e51683 763 p->pack_fd = git_open_noatime(p->pack_name);
9bc879c1 764 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 765 return -1;
c7934306 766 pack_open_fds++;
9bc879c1
SP
767
768 /* If we created the struct before we had the pack we lack size. */
bf592c50 769 if (!p->pack_size) {
bf592c50 770 if (!S_ISREG(st.st_mode))
072db278 771 return error("packfile %s not a regular file", p->pack_name);
bf592c50 772 p->pack_size = st.st_size;
9bc879c1 773 } else if (p->pack_size != st.st_size)
072db278 774 return error("packfile %s size changed", p->pack_name);
9bc879c1 775
2c039da8
JH
776 /* We leave these file descriptors open with sliding mmap;
777 * there is no point keeping them open across exec(), though.
778 */
779 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
780 if (fd_flag < 0)
072db278 781 return error("cannot determine file descriptor flags");
2c039da8
JH
782 fd_flag |= FD_CLOEXEC;
783 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 784 return error("cannot set FD_CLOEXEC");
2c039da8 785
9bc879c1 786 /* Verify we recognize this pack file format. */
e6e2bd62 787 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 788 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 789 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 790 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 791 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
792 return error("packfile %s is version %"PRIu32" and not"
793 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
794 p->pack_name, ntohl(hdr.hdr_version));
795
796 /* Verify the pack matches its index. */
57059091 797 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
798 return error("packfile %s claims to have %"PRIu32" objects"
799 " while index indicates %"PRIu32" objects",
57059091
NP
800 p->pack_name, ntohl(hdr.hdr_entries),
801 p->num_objects);
9bc879c1 802 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 803 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 804 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 805 return error("packfile %s signature is unavailable", p->pack_name);
42873078 806 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 807 if (hashcmp(sha1, idx_sha1))
072db278
SP
808 return error("packfile %s does not match index", p->pack_name);
809 return 0;
9bc879c1
SP
810}
811
3cf8b462
SP
812static int open_packed_git(struct packed_git *p)
813{
814 if (!open_packed_git_1(p))
815 return 0;
816 if (p->pack_fd != -1) {
817 close(p->pack_fd);
c7934306 818 pack_open_fds--;
3cf8b462
SP
819 p->pack_fd = -1;
820 }
821 return -1;
822}
823
c4001d92 824static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
825{
826 /* We must promise at least 20 bytes (one hash) after the
827 * offset is available from this window, otherwise the offset
828 * is not actually in this window and a different window (which
829 * has that one hash excess) must be used. This is to support
830 * the object header and delta base parsing routines below.
831 */
832 off_t win_off = win->offset;
833 return win_off <= offset
834 && (offset + 20) <= (win_off + win->len);
835}
836
4b25d091 837unsigned char *use_pack(struct packed_git *p,
03e79c88 838 struct pack_window **w_cursor,
c4001d92 839 off_t offset,
ef49a7a0 840 unsigned long *left)
9bc879c1 841{
60bb8b14 842 struct pack_window *win = *w_cursor;
03e79c88 843
a9d98a14 844 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
845 * pointless to ask for an offset into the middle of that
846 * hash, and the in_window function above wouldn't match
847 * don't allow an offset too close to the end of the file.
848 */
d131b7af
SP
849 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
850 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
851 if (offset > (p->pack_size - 20))
852 die("offset beyond end of packfile (truncated pack?)");
853
854 if (!win || !in_window(win, offset)) {
855 if (win)
856 win->inuse_cnt--;
857 for (win = p->windows; win; win = win->next) {
858 if (in_window(win, offset))
859 break;
860 }
861 if (!win) {
78a28df9 862 size_t window_align = packed_git_window_size / 2;
dc49cd76 863 off_t len;
d131b7af
SP
864
865 if (p->pack_fd == -1 && open_packed_git(p))
866 die("packfile %s cannot be accessed", p->pack_name);
867
60bb8b14 868 win = xcalloc(1, sizeof(*win));
78a28df9 869 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
870 len = p->pack_size - win->offset;
871 if (len > packed_git_window_size)
872 len = packed_git_window_size;
873 win->len = (size_t)len;
60bb8b14 874 pack_mapped += win->len;
11daf39b 875 while (packed_git_limit < pack_mapped
d1efefa4 876 && unuse_one_window(p, p->pack_fd))
60bb8b14 877 ; /* nothing */
c4712e45 878 win->base = xmmap(NULL, win->len,
60bb8b14
SP
879 PROT_READ, MAP_PRIVATE,
880 p->pack_fd, win->offset);
881 if (win->base == MAP_FAILED)
73b4e4be
SP
882 die("packfile %s cannot be mapped: %s",
883 p->pack_name,
884 strerror(errno));
d131b7af
SP
885 if (!win->offset && win->len == p->pack_size
886 && !p->do_not_close) {
887 close(p->pack_fd);
888 pack_open_fds--;
889 p->pack_fd = -1;
890 }
a53128b6
SP
891 pack_mmap_calls++;
892 pack_open_windows++;
893 if (pack_mapped > peak_pack_mapped)
894 peak_pack_mapped = pack_mapped;
895 if (pack_open_windows > peak_pack_open_windows)
896 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
897 win->next = p->windows;
898 p->windows = win;
899 }
1f688557 900 }
03e79c88
SP
901 if (win != *w_cursor) {
902 win->last_used = pack_used_ctr++;
903 win->inuse_cnt++;
904 *w_cursor = win;
905 }
60bb8b14 906 offset -= win->offset;
03e79c88 907 if (left)
dc49cd76 908 *left = win->len - xsize_t(offset);
03e79c88 909 return win->base + offset;
1f688557
JH
910}
911
27d69a46
NP
912static struct packed_git *alloc_packed_git(int extra)
913{
914 struct packed_git *p = xmalloc(sizeof(*p) + extra);
915 memset(p, 0, sizeof(*p));
916 p->pack_fd = -1;
917 return p;
918}
919
e0500293
JN
920static void try_to_free_pack_memory(size_t size)
921{
922 release_pack_memory(size, -1);
923}
924
42873078 925struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557 926{
e0500293 927 static int have_set_try_to_free_routine;
1f688557 928 struct stat st;
27d69a46 929 struct packed_git *p = alloc_packed_git(path_len + 2);
1f688557 930
e0500293
JN
931 if (!have_set_try_to_free_routine) {
932 have_set_try_to_free_routine = 1;
933 set_try_to_free_routine(try_to_free_pack_memory);
934 }
935
42873078
NP
936 /*
937 * Make sure a corresponding .pack file exists and that
938 * the index looks sane.
939 */
940 path_len -= strlen(".idx");
27d69a46
NP
941 if (path_len < 1) {
942 free(p);
1f688557 943 return NULL;
27d69a46 944 }
42873078 945 memcpy(p->pack_name, path, path_len);
8d25931d
BC
946
947 strcpy(p->pack_name + path_len, ".keep");
948 if (!access(p->pack_name, F_OK))
949 p->pack_keep = 1;
950
42873078 951 strcpy(p->pack_name + path_len, ".pack");
d079837e 952 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 953 free(p);
1f688557
JH
954 return NULL;
955 }
42873078 956
1f688557
JH
957 /* ok, it looks sane as far as we can check without
958 * actually mapping the pack file.
959 */
1f688557 960 p->pack_size = st.st_size;
9d835df2 961 p->pack_local = local;
b867092f 962 p->mtime = st.st_mtime;
42873078
NP
963 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
964 hashclr(p->sha1);
1f688557
JH
965 return p;
966}
967
7b64469a 968struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 969{
42873078 970 const char *path = sha1_pack_name(sha1);
27d69a46 971 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
bf592c50 972
27d69a46
NP
973 strcpy(p->pack_name, path);
974 hashcpy(p->sha1, sha1);
42873078
NP
975 if (check_packed_git_idx(idx_path, p)) {
976 free(p);
bf592c50 977 return NULL;
42873078 978 }
bf592c50 979
bf592c50
DB
980 return p;
981}
982
983void install_packed_git(struct packed_git *pack)
984{
c7934306
SP
985 if (pack->pack_fd != -1)
986 pack_open_fds++;
987
bf592c50
DB
988 pack->next = packed_git;
989 packed_git = pack;
990}
991
9d835df2 992static void prepare_packed_git_one(char *objdir, int local)
1f688557 993{
9cb18f56
JM
994 /* Ensure that this buffer is large enough so that we can
995 append "/pack/" without clobbering the stack even if
996 strlen(objdir) were PATH_MAX. */
997 char path[PATH_MAX + 1 + 4 + 1 + 1];
1f688557
JH
998 int len;
999 DIR *dir;
1000 struct dirent *de;
1001
1002 sprintf(path, "%s/pack", objdir);
1003 len = strlen(path);
1004 dir = opendir(path);
b5b16990 1005 if (!dir) {
26125f6b 1006 if (errno != ENOENT)
bd2afde8 1007 error("unable to open object pack directory: %s: %s",
26125f6b 1008 path, strerror(errno));
1f688557 1009 return;
b5b16990 1010 }
1f688557
JH
1011 path[len++] = '/';
1012 while ((de = readdir(dir)) != NULL) {
1013 int namelen = strlen(de->d_name);
1014 struct packed_git *p;
1015
5bb1cda5 1016 if (!has_extension(de->d_name, ".idx"))
1f688557
JH
1017 continue;
1018
9cb18f56
JM
1019 if (len + namelen + 1 > sizeof(path))
1020 continue;
1021
54a15a8d 1022 /* Don't reopen a pack we already have. */
1f688557 1023 strcpy(path + len, de->d_name);
86f7780c
JK
1024 for (p = packed_git; p; p = p->next) {
1025 if (!memcmp(path, p->pack_name, len + namelen - 4))
1026 break;
1027 }
1028 if (p)
1029 continue;
54a15a8d
SP
1030 /* See if it really is a valid .idx file with corresponding
1031 * .pack file that we can map.
1032 */
9d835df2 1033 p = add_packed_git(path, len + namelen, local);
1f688557
JH
1034 if (!p)
1035 continue;
625e9421 1036 install_packed_git(p);
1f688557 1037 }
5b35bcd5 1038 closedir(dir);
1f688557
JH
1039}
1040
b867092f
JH
1041static int sort_pack(const void *a_, const void *b_)
1042{
1043 struct packed_git *a = *((struct packed_git **)a_);
1044 struct packed_git *b = *((struct packed_git **)b_);
1045 int st;
1046
1047 /*
1048 * Local packs tend to contain objects specific to our
1049 * variant of the project than remote ones. In addition,
1050 * remote ones could be on a network mounted filesystem.
1051 * Favor local ones for these reasons.
1052 */
1053 st = a->pack_local - b->pack_local;
1054 if (st)
1055 return -st;
1056
1057 /*
1058 * Younger packs tend to contain more recent objects,
1059 * and more recent objects tend to get accessed more
1060 * often.
1061 */
1062 if (a->mtime < b->mtime)
1063 return 1;
1064 else if (a->mtime == b->mtime)
1065 return 0;
1066 return -1;
1067}
1068
1069static void rearrange_packed_git(void)
1070{
1071 struct packed_git **ary, *p;
1072 int i, n;
1073
1074 for (n = 0, p = packed_git; p; p = p->next)
1075 n++;
1076 if (n < 2)
1077 return;
1078
1079 /* prepare an array of packed_git for easier sorting */
1080 ary = xcalloc(n, sizeof(struct packed_git *));
1081 for (n = 0, p = packed_git; p; p = p->next)
1082 ary[n++] = p;
1083
1084 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1085
1086 /* link them back again */
1087 for (i = 0; i < n - 1; i++)
1088 ary[i]->next = ary[i + 1];
1089 ary[n - 1]->next = NULL;
1090 packed_git = ary[0];
1091
1092 free(ary);
1093}
1094
637cdd9d 1095static int prepare_packed_git_run_once = 0;
9a217f2a 1096void prepare_packed_git(void)
1f688557 1097{
d5a63b99 1098 struct alternate_object_database *alt;
1f688557 1099
637cdd9d 1100 if (prepare_packed_git_run_once)
1f688557 1101 return;
9d835df2 1102 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 1103 prepare_alt_odb();
d5a63b99 1104 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 1105 alt->name[-1] = 0;
9d835df2 1106 prepare_packed_git_one(alt->base, 0);
1494e038 1107 alt->name[-1] = '/';
1f688557 1108 }
b867092f 1109 rearrange_packed_git();
637cdd9d
JK
1110 prepare_packed_git_run_once = 1;
1111}
1112
fc04c412 1113void reprepare_packed_git(void)
637cdd9d 1114{
4b480c67 1115 discard_revindex();
637cdd9d
JK
1116 prepare_packed_git_run_once = 0;
1117 prepare_packed_git();
1f688557
JH
1118}
1119
8eca0b47
NP
1120static void mark_bad_packed_object(struct packed_git *p,
1121 const unsigned char *sha1)
1122{
1123 unsigned i;
1124 for (i = 0; i < p->num_bad_objects; i++)
1125 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1126 return;
1127 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1128 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1129 p->num_bad_objects++;
1130}
1131
b6c4cecc 1132static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
ac939109
NP
1133{
1134 struct packed_git *p;
1135 unsigned i;
1136
1137 for (p = packed_git; p; p = p->next)
1138 for (i = 0; i < p->num_bad_objects; i++)
1139 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
b6c4cecc
JH
1140 return p;
1141 return NULL;
ac939109
NP
1142}
1143
5d6ccf5c 1144int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
0fcfd160
LT
1145{
1146 unsigned char real_sha1[20];
7cfb5f36 1147 hash_sha1_file(map, size, type, real_sha1);
a89fccd2 1148 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1149}
1150
f4e51683 1151static int git_open_noatime(const char *name)
44d1c19e
LT
1152{
1153 static int sha1_file_open_flag = O_NOATIME;
44d1c19e 1154
f2e872aa
SP
1155 for (;;) {
1156 int fd = open(name, O_RDONLY | sha1_file_open_flag);
44d1c19e 1157 if (fd >= 0)
f2e872aa
SP
1158 return fd;
1159
f2e872aa
SP
1160 /* Might the failure be due to O_NOATIME? */
1161 if (errno != ENOENT && sha1_file_open_flag) {
44d1c19e 1162 sha1_file_open_flag = 0;
f2e872aa
SP
1163 continue;
1164 }
1165
1166 return -1;
44d1c19e 1167 }
44d1c19e
LT
1168}
1169
1170static int open_sha1_file(const unsigned char *sha1)
1171{
1172 int fd;
1173 char *name = sha1_file_name(sha1);
1174 struct alternate_object_database *alt;
1175
f4e51683 1176 fd = git_open_noatime(name);
44d1c19e
LT
1177 if (fd >= 0)
1178 return fd;
1179
1180 prepare_alt_odb();
1181 errno = ENOENT;
1182 for (alt = alt_odb_list; alt; alt = alt->next) {
1183 name = alt->name;
1184 fill_sha1_path(name, sha1);
f4e51683 1185 fd = git_open_noatime(alt->base);
44d1c19e
LT
1186 if (fd >= 0)
1187 return fd;
1188 }
1189 return -1;
1190}
1191
f0270efd 1192void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1193{
0fcfd160 1194 void *map;
144bde78 1195 int fd;
ace1534d 1196
44d1c19e
LT
1197 fd = open_sha1_file(sha1);
1198 map = NULL;
1199 if (fd >= 0) {
1200 struct stat st;
0fcfd160 1201
44d1c19e
LT
1202 if (!fstat(fd, &st)) {
1203 *size = xsize_t(st.st_size);
1204 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1205 }
44d1c19e 1206 close(fd);
0fcfd160 1207 }
0fcfd160
LT
1208 return map;
1209}
1210
cc5c54e7
JH
1211/*
1212 * There used to be a second loose object header format which
1213 * was meant to mimic the in-pack format, allowing for direct
1214 * copy of the object data. This format turned up not to be
1215 * really worth it and we no longer write loose objects in that
1216 * format.
1217 */
1218static int experimental_loose_object(unsigned char *map)
bb6b8e4f
JH
1219{
1220 unsigned int word;
1221
1222 /*
7f684a2a
RT
1223 * We must determine if the buffer contains the standard
1224 * zlib-deflated stream or the experimental format based
1225 * on the in-pack object format. Compare the header byte
1226 * for each format:
1227 *
1228 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1229 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1230 *
1231 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1232 * in standard loose-object format, UNLESS it is a Git-pack
1233 * format object *exactly* 8 bytes in size when inflated.
1234 *
1235 * However, RFC1950 also specifies that the 1st 16-bit word
1236 * must be divisible by 31 - this checksum tells us our buffer
1237 * is in the standard format, giving a false positive only if
1238 * the 1st word of the Git-pack format object happens to be
1239 * divisible by 31, ie:
1240 * ((byte0 * 256) + byte1) % 31 = 0
1241 * => 0ttt10000www1000 % 31 = 0
1242 *
1243 * As it happens, this case can only arise for www=3 & ttt=1
1244 * - ie, a Commit object, which would have to be 8 bytes in
1245 * size. As no Commit can be that small, we find that the
1246 * combination of these two criteria (bitmask & checksum)
1247 * can always correctly determine the buffer format.
bb6b8e4f
JH
1248 */
1249 word = (map[0] << 8) + map[1];
7f684a2a 1250 if ((map[0] & 0x8F) == 0x08 && !(word % 31))
bb6b8e4f 1251 return 0;
cc5c54e7
JH
1252 else
1253 return 1;
bb6b8e4f
JH
1254}
1255
09ded04b
NP
1256unsigned long unpack_object_header_buffer(const unsigned char *buf,
1257 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1258{
ad1ed5ee 1259 unsigned shift;
48fb7deb 1260 unsigned long size, c;
ad1ed5ee
JH
1261 unsigned long used = 0;
1262
1263 c = buf[used++];
1264 *type = (c >> 4) & 7;
1265 size = c & 15;
1266 shift = 4;
1267 while (c & 0x80) {
f630cfda 1268 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1269 error("bad object header");
ea4f9685
JH
1270 size = used = 0;
1271 break;
09ded04b 1272 }
ad1ed5ee
JH
1273 c = buf[used++];
1274 size += (c & 0x7f) << shift;
1275 shift += 7;
1276 }
1277 *sizep = size;
1278 return used;
1279}
1280
eb4f4076 1281int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
ad1ed5ee
JH
1282{
1283 unsigned long size, used;
1284 static const char valid_loose_object_type[8] = {
1285 0, /* OBJ_EXT */
1286 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1287 0, /* "delta" and others are invalid in a loose object */
93821bd9 1288 };
ad1ed5ee 1289 enum object_type type;
93821bd9 1290
c4483576
LT
1291 /* Get the data stream */
1292 memset(stream, 0, sizeof(*stream));
1293 stream->next_in = map;
1294 stream->avail_in = mapsize;
1295 stream->next_out = buffer;
93821bd9
LT
1296 stream->avail_out = bufsiz;
1297
cc5c54e7
JH
1298 if (experimental_loose_object(map)) {
1299 /*
1300 * The old experimental format we no longer produce;
1301 * we can still read it.
1302 */
1303 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1304 if (!used || !valid_loose_object_type[type])
1305 return -1;
1306 map += used;
1307 mapsize -= used;
726f852b 1308
cc5c54e7
JH
1309 /* Set up the stream for the rest.. */
1310 stream->next_in = map;
1311 stream->avail_in = mapsize;
1312 git_inflate_init(stream);
93821bd9 1313
cc5c54e7
JH
1314 /* And generate the fake traditional header */
1315 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1316 typename(type), size);
1317 return 0;
1318 }
39c68542 1319 git_inflate_init(stream);
cc5c54e7 1320 return git_inflate(stream, 0);
c4483576
LT
1321}
1322
ef49a7a0 1323static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1324{
1325 int bytes = strlen(buffer) + 1;
3aee68aa 1326 unsigned char *buf = xmallocz(size);
93821bd9 1327 unsigned long n;
7efbff75 1328 int status = Z_OK;
5180cacc 1329
93821bd9
LT
1330 n = stream->total_out - bytes;
1331 if (n > size)
1332 n = size;
1333 memcpy(buf, (char *) buffer + bytes, n);
1334 bytes = n;
456cdf6e
LT
1335 if (bytes <= size) {
1336 /*
1337 * The above condition must be (bytes <= size), not
1338 * (bytes < size). In other words, even though we
ccf5ace0 1339 * expect no more output and set avail_out to zero,
456cdf6e
LT
1340 * the input zlib stream may have bytes that express
1341 * "this concludes the stream", and we *do* want to
1342 * eat that input.
1343 *
1344 * Otherwise we would not be able to test that we
1345 * consumed all the input to reach the expected size;
1346 * we also want to check that zlib tells us that all
1347 * went well with status == Z_STREAM_END at the end.
1348 */
5180cacc
LT
1349 stream->next_out = buf + bytes;
1350 stream->avail_out = size - bytes;
7efbff75 1351 while (status == Z_OK)
39c68542 1352 status = git_inflate(stream, Z_FINISH);
5180cacc 1353 }
456cdf6e 1354 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1355 git_inflate_end(stream);
7efbff75
JH
1356 return buf;
1357 }
1358
1359 if (status < 0)
1360 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1361 else if (stream->avail_in)
1362 error("garbage at end of loose object '%s'",
1363 sha1_to_hex(sha1));
1364 free(buf);
1365 return NULL;
5180cacc
LT
1366}
1367
1368/*
1369 * We used to just use "sscanf()", but that's actually way
1370 * too permissive for what we want to check. So do an anal
1371 * object header parse by hand.
1372 */
f0270efd 1373int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1374{
21666f1a 1375 char type[10];
5180cacc
LT
1376 int i;
1377 unsigned long size;
1378
1379 /*
a6080a0a 1380 * The type can be at most ten bytes (including the
5180cacc 1381 * terminating '\0' that we add), and is followed by
21666f1a 1382 * a space.
5180cacc 1383 */
21666f1a 1384 i = 0;
5180cacc
LT
1385 for (;;) {
1386 char c = *hdr++;
1387 if (c == ' ')
1388 break;
21666f1a
NP
1389 type[i++] = c;
1390 if (i >= sizeof(type))
5180cacc 1391 return -1;
5180cacc 1392 }
21666f1a 1393 type[i] = 0;
5180cacc
LT
1394
1395 /*
1396 * The length must follow immediately, and be in canonical
1397 * decimal format (ie "010" is not valid).
1398 */
1399 size = *hdr++ - '0';
1400 if (size > 9)
1401 return -1;
1402 if (size) {
1403 for (;;) {
1404 unsigned long c = *hdr - '0';
1405 if (c > 9)
1406 break;
1407 hdr++;
1408 size = size * 10 + c;
1409 }
1410 }
1411 *sizep = size;
1412
1413 /*
1414 * The length must be followed by a zero byte
1415 */
21666f1a 1416 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1417}
1418
7efbff75 1419static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1420{
5180cacc 1421 int ret;
ef49a7a0 1422 git_zstream stream;
5180cacc 1423 char hdr[8192];
0fcfd160 1424
5180cacc 1425 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1426 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1427 return NULL;
1428
7efbff75 1429 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1430}
1431
54dab52a
NP
1432unsigned long get_size_from_delta(struct packed_git *p,
1433 struct pack_window **w_curs,
1434 off_t curpos)
1435{
1436 const unsigned char *data;
1437 unsigned char delta_head[20], *in;
ef49a7a0 1438 git_zstream stream;
54dab52a
NP
1439 int st;
1440
1441 memset(&stream, 0, sizeof(stream));
1442 stream.next_out = delta_head;
1443 stream.avail_out = sizeof(delta_head);
1444
39c68542 1445 git_inflate_init(&stream);
54dab52a
NP
1446 do {
1447 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1448 stream.next_in = in;
39c68542 1449 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1450 curpos += stream.next_in - in;
1451 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1452 stream.total_out < sizeof(delta_head));
39c68542 1453 git_inflate_end(&stream);
3d77d877
NP
1454 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1455 error("delta data unpack-initial failed");
1456 return 0;
1457 }
54dab52a
NP
1458
1459 /* Examine the initial part of the delta to figure out
1460 * the result size.
1461 */
1462 data = delta_head;
1463
1464 /* ignore base size */
1465 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1466
1467 /* Read the result size */
1468 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1469}
1470
c4001d92 1471static off_t get_delta_base(struct packed_git *p,
03e79c88 1472 struct pack_window **w_curs,
c4001d92 1473 off_t *curpos,
21666f1a 1474 enum object_type type,
c4001d92 1475 off_t delta_obj_offset)
eb32d236 1476{
2b87c45b 1477 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1478 off_t base_offset;
eb32d236 1479
8d8a4ea5
SP
1480 /* use_pack() assured us we have [base_info, base_info + 20)
1481 * as a range that we can look at without walking off the
1482 * end of the mapped window. Its actually the hash size
1483 * that is assured. An OFS_DELTA longer than the hash size
1484 * is stupid, as then a REF_DELTA would be smaller to store.
1485 */
21666f1a 1486 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1487 unsigned used = 0;
1488 unsigned char c = base_info[used++];
1489 base_offset = c & 127;
1490 while (c & 128) {
1491 base_offset += 1;
8723f216 1492 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1493 return 0; /* overflow */
eb32d236
NP
1494 c = base_info[used++];
1495 base_offset = (base_offset << 7) + (c & 127);
1496 }
1497 base_offset = delta_obj_offset - base_offset;
d8f32556 1498 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1499 return 0; /* out of bound */
2b87c45b 1500 *curpos += used;
21666f1a 1501 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1502 /* The base entry _must_ be in the same pack */
1503 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1504 *curpos += 20;
eb32d236
NP
1505 } else
1506 die("I am totally screwed");
2b87c45b 1507 return base_offset;
eb32d236
NP
1508}
1509
f3bf9224 1510/* forward declaration for a mutually recursive function */
c4001d92 1511static int packed_object_info(struct packed_git *p, off_t offset,
9a490590 1512 unsigned long *sizep, int *rtype);
f3bf9224 1513
43057304 1514static int packed_delta_info(struct packed_git *p,
03e79c88 1515 struct pack_window **w_curs,
c4001d92 1516 off_t curpos,
21666f1a 1517 enum object_type type,
c4001d92 1518 off_t obj_offset,
43057304 1519 unsigned long *sizep)
5db47c2b 1520{
c4001d92 1521 off_t base_offset;
f3bf9224 1522
21666f1a 1523 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
3d77d877
NP
1524 if (!base_offset)
1525 return OBJ_BAD;
9a490590 1526 type = packed_object_info(p, base_offset, NULL, NULL);
3d77d877 1527 if (type <= OBJ_NONE) {
08698b1e
NP
1528 struct revindex_entry *revidx;
1529 const unsigned char *base_sha1;
1530 revidx = find_pack_revindex(p, base_offset);
1531 if (!revidx)
1532 return OBJ_BAD;
1533 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
3d77d877
NP
1534 mark_bad_packed_object(p, base_sha1);
1535 type = sha1_object_info(base_sha1, NULL);
1536 if (type <= OBJ_NONE)
1537 return OBJ_BAD;
1538 }
f3bf9224 1539
c62266f3
JH
1540 /* We choose to only get the type of the base object and
1541 * ignore potentially corrupt pack file that expects the delta
1542 * based on a base with a wrong size. This saves tons of
1543 * inflate() calls.
1544 */
3d77d877 1545 if (sizep) {
54dab52a 1546 *sizep = get_size_from_delta(p, w_curs, curpos);
3d77d877
NP
1547 if (*sizep == 0)
1548 type = OBJ_BAD;
1549 }
21666f1a
NP
1550
1551 return type;
5db47c2b
JH
1552}
1553
f8c8abc5
JH
1554int unpack_object_header(struct packed_git *p,
1555 struct pack_window **w_curs,
1556 off_t *curpos,
1557 unsigned long *sizep)
a733cb60 1558{
03e79c88 1559 unsigned char *base;
ef49a7a0 1560 unsigned long left;
ad1ed5ee 1561 unsigned long used;
2b87c45b 1562 enum object_type type;
a733cb60 1563
8d8a4ea5 1564 /* use_pack() assures us we have [base, base + 20) available
0353a0c4 1565 * as a range that we can look at. (Its actually the hash
3dff5379 1566 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1567 * the maximum deflated object size is 2^137, which is just
1568 * insane, so we know won't exceed what we have been given.
1569 */
2b87c45b 1570 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1571 used = unpack_object_header_buffer(base, left, &type, sizep);
1572 if (!used) {
1573 type = OBJ_BAD;
1574 } else
1575 *curpos += used;
ad1ed5ee 1576
2b87c45b 1577 return type;
a733cb60
LT
1578}
1579
c4001d92 1580static int packed_object_info(struct packed_git *p, off_t obj_offset,
9a490590 1581 unsigned long *sizep, int *rtype)
1f688557 1582{
03e79c88 1583 struct pack_window *w_curs = NULL;
c4001d92
SP
1584 unsigned long size;
1585 off_t curpos = obj_offset;
21666f1a 1586 enum object_type type;
5db47c2b 1587
21666f1a 1588 type = unpack_object_header(p, &w_curs, &curpos, &size);
9a490590
JH
1589 if (rtype)
1590 *rtype = type; /* representation type */
5db47c2b 1591
21666f1a 1592 switch (type) {
eb32d236
NP
1593 case OBJ_OFS_DELTA:
1594 case OBJ_REF_DELTA:
21666f1a
NP
1595 type = packed_delta_info(p, &w_curs, curpos,
1596 type, obj_offset, sizep);
2b87c45b 1597 break;
a733cb60 1598 case OBJ_COMMIT:
a733cb60 1599 case OBJ_TREE:
a733cb60 1600 case OBJ_BLOB:
a733cb60 1601 case OBJ_TAG:
2b87c45b
NP
1602 if (sizep)
1603 *sizep = size;
a69d0943 1604 break;
1f688557 1605 default:
3d77d877
NP
1606 error("unknown object type %i at offset %"PRIuMAX" in %s",
1607 type, (uintmax_t)obj_offset, p->pack_name);
1608 type = OBJ_BAD;
1f688557 1609 }
2b87c45b 1610 unuse_pack(&w_curs);
21666f1a 1611 return type;
1f688557
JH
1612}
1613
eb950c19 1614static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1615 struct pack_window **w_curs,
c4001d92 1616 off_t curpos,
eb950c19 1617 unsigned long size)
de530aaa
SP
1618{
1619 int st;
ef49a7a0 1620 git_zstream stream;
079afb18 1621 unsigned char *buffer, *in;
de530aaa 1622
4ab07e4d 1623 buffer = xmallocz(size);
de530aaa 1624 memset(&stream, 0, sizeof(stream));
de530aaa 1625 stream.next_out = buffer;
39eea7bd 1626 stream.avail_out = size + 1;
de530aaa 1627
39c68542 1628 git_inflate_init(&stream);
079afb18 1629 do {
2b87c45b 1630 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 1631 stream.next_in = in;
39c68542 1632 st = git_inflate(&stream, Z_FINISH);
39eea7bd
JH
1633 if (!stream.avail_out)
1634 break; /* the payload is larger than it should be */
2b87c45b 1635 curpos += stream.next_in - in;
079afb18 1636 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 1637 git_inflate_end(&stream);
de530aaa
SP
1638 if ((st != Z_STREAM_END) || stream.total_out != size) {
1639 free(buffer);
1640 return NULL;
1641 }
1642
1643 return buffer;
1644}
1645
e5e01619
LT
1646#define MAX_DELTA_CACHE (256)
1647
18bdec11 1648static size_t delta_base_cached;
5e08ecbf
NP
1649
1650static struct delta_base_cache_lru_list {
1651 struct delta_base_cache_lru_list *prev;
1652 struct delta_base_cache_lru_list *next;
1653} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1654
e5e01619 1655static struct delta_base_cache_entry {
5e08ecbf
NP
1656 struct delta_base_cache_lru_list lru;
1657 void *data;
e5e01619
LT
1658 struct packed_git *p;
1659 off_t base_offset;
1660 unsigned long size;
e5e01619
LT
1661 enum object_type type;
1662} delta_base_cache[MAX_DELTA_CACHE];
1663
1664static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1665{
1666 unsigned long hash;
1667
1668 hash = (unsigned long)p + (unsigned long)base_offset;
1669 hash += (hash >> 8) + (hash >> 16);
3358004a 1670 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1671}
1672
5266d369
JH
1673static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1674{
1675 unsigned long hash = pack_entry_hash(p, base_offset);
1676 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1677 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1678}
1679
62f255ad 1680static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 1681 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 1682{
e5e01619
LT
1683 void *ret;
1684 unsigned long hash = pack_entry_hash(p, base_offset);
1685 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1686
1687 ret = ent->data;
749bc58c
MV
1688 if (!ret || ent->p != p || ent->base_offset != base_offset)
1689 return unpack_entry(p, base_offset, type, base_size);
e5e01619 1690
18bdec11 1691 if (!keep_cache) {
a0cba108 1692 ent->data = NULL;
5e08ecbf
NP
1693 ent->lru.next->prev = ent->lru.prev;
1694 ent->lru.prev->next = ent->lru.next;
18bdec11 1695 delta_base_cached -= ent->size;
182af834
PH
1696 } else {
1697 ret = xmemdupz(ent->data, ent->size);
a0cba108 1698 }
e5e01619
LT
1699 *type = ent->type;
1700 *base_size = ent->size;
1701 return ret;
62f255ad
LT
1702}
1703
18bdec11
SP
1704static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1705{
1706 if (ent->data) {
1707 free(ent->data);
1708 ent->data = NULL;
5e08ecbf
NP
1709 ent->lru.next->prev = ent->lru.prev;
1710 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
1711 delta_base_cached -= ent->size;
1712 }
1713}
1714
3d20c636
SP
1715void clear_delta_base_cache(void)
1716{
1717 unsigned long p;
1718 for (p = 0; p < MAX_DELTA_CACHE; p++)
1719 release_delta_base_cache(&delta_base_cache[p]);
1720}
1721
62f255ad
LT
1722static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1723 void *base, unsigned long base_size, enum object_type type)
1724{
5e08ecbf 1725 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 1726 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 1727 struct delta_base_cache_lru_list *lru;
e5e01619 1728
18bdec11
SP
1729 release_delta_base_cache(ent);
1730 delta_base_cached += base_size;
5e08ecbf
NP
1731
1732 for (lru = delta_base_cache_lru.next;
1733 delta_base_cached > delta_base_cache_limit
1734 && lru != &delta_base_cache_lru;
1735 lru = lru->next) {
1736 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
1737 if (f->type == OBJ_BLOB)
1738 release_delta_base_cache(f);
1739 }
5e08ecbf
NP
1740 for (lru = delta_base_cache_lru.next;
1741 delta_base_cached > delta_base_cache_limit
1742 && lru != &delta_base_cache_lru;
1743 lru = lru->next) {
1744 struct delta_base_cache_entry *f = (void *)lru;
1745 release_delta_base_cache(f);
1746 }
18bdec11 1747
e5e01619
LT
1748 ent->p = p;
1749 ent->base_offset = base_offset;
1750 ent->type = type;
1751 ent->data = base;
1752 ent->size = base_size;
5e08ecbf
NP
1753 ent->lru.next = &delta_base_cache_lru;
1754 ent->lru.prev = delta_base_cache_lru.prev;
1755 delta_base_cache_lru.prev->next = &ent->lru;
1756 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
1757}
1758
c2c5b270
CC
1759static void *read_object(const unsigned char *sha1, enum object_type *type,
1760 unsigned long *size);
1761
eb950c19 1762static void *unpack_delta_entry(struct packed_git *p,
03e79c88 1763 struct pack_window **w_curs,
c4001d92 1764 off_t curpos,
1f688557 1765 unsigned long delta_size,
c4001d92 1766 off_t obj_offset,
21666f1a 1767 enum object_type *type,
eb950c19 1768 unsigned long *sizep)
1f688557 1769{
7c3e8be3 1770 void *delta_data, *result, *base;
c4001d92
SP
1771 unsigned long base_size;
1772 off_t base_offset;
43057304 1773
21666f1a 1774 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
8eca0b47
NP
1775 if (!base_offset) {
1776 error("failed to validate delta base reference "
1777 "at offset %"PRIuMAX" from %s",
1778 (uintmax_t)curpos, p->pack_name);
1779 return NULL;
1780 }
eac12e2d 1781 unuse_pack(w_curs);
a0cba108 1782 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
8eca0b47
NP
1783 if (!base) {
1784 /*
1785 * We're probably in deep shit, but let's try to fetch
1786 * the required base anyway from another pack or loose.
1787 * This is costly but should happen only in the presence
1788 * of a corrupted pack, and is better than failing outright.
1789 */
08698b1e
NP
1790 struct revindex_entry *revidx;
1791 const unsigned char *base_sha1;
1792 revidx = find_pack_revindex(p, base_offset);
1793 if (!revidx)
1794 return NULL;
1795 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
8eca0b47
NP
1796 error("failed to read delta base object %s"
1797 " at offset %"PRIuMAX" from %s",
1798 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1799 p->pack_name);
1800 mark_bad_packed_object(p, base_sha1);
ac939109 1801 base = read_object(base_sha1, type, &base_size);
8eca0b47
NP
1802 if (!base)
1803 return NULL;
1804 }
67686d95 1805
2b87c45b 1806 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
8eca0b47
NP
1807 if (!delta_data) {
1808 error("failed to unpack compressed delta "
1809 "at offset %"PRIuMAX" from %s",
1810 (uintmax_t)curpos, p->pack_name);
1811 free(base);
1812 return NULL;
1813 }
1f688557
JH
1814 result = patch_delta(base, base_size,
1815 delta_data, delta_size,
21666f1a 1816 sizep);
1f688557
JH
1817 if (!result)
1818 die("failed to apply delta");
1819 free(delta_data);
62f255ad 1820 add_delta_base_cache(p, base_offset, base, base_size, *type);
1f688557
JH
1821 return result;
1822}
1823
5f44324d
JH
1824static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1825{
1826 static FILE *log_file;
1827
1828 if (!log_file) {
1829 log_file = fopen(log_pack_access, "w");
1830 if (!log_file) {
1831 error("cannot open pack access log '%s' for writing: %s",
1832 log_pack_access, strerror(errno));
1833 log_pack_access = NULL;
1834 return;
1835 }
1836 }
1837 fprintf(log_file, "%s %"PRIuMAX"\n",
1838 p->pack_name, (uintmax_t)obj_offset);
1839 fflush(log_file);
1840}
1841
0e8189e2
NP
1842int do_check_packed_object_crc;
1843
c4001d92 1844void *unpack_entry(struct packed_git *p, off_t obj_offset,
21666f1a 1845 enum object_type *type, unsigned long *sizep)
1f688557 1846{
03e79c88 1847 struct pack_window *w_curs = NULL;
c4001d92 1848 off_t curpos = obj_offset;
21666f1a 1849 void *data;
1f688557 1850
5f44324d
JH
1851 if (log_pack_access)
1852 write_pack_access_log(p, obj_offset);
1853
0e8189e2
NP
1854 if (do_check_packed_object_crc && p->index_version > 1) {
1855 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1856 unsigned long len = revidx[1].offset - obj_offset;
1857 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1858 const unsigned char *sha1 =
1859 nth_packed_object_sha1(p, revidx->nr);
1860 error("bad packed object CRC for %s",
1861 sha1_to_hex(sha1));
1862 mark_bad_packed_object(p, sha1);
f755bb99 1863 unuse_pack(&w_curs);
0e8189e2
NP
1864 return NULL;
1865 }
1866 }
1867
21666f1a
NP
1868 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1869 switch (*type) {
eb32d236
NP
1870 case OBJ_OFS_DELTA:
1871 case OBJ_REF_DELTA:
21666f1a
NP
1872 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1873 obj_offset, type, sizep);
4d703a1a 1874 break;
a733cb60 1875 case OBJ_COMMIT:
a733cb60 1876 case OBJ_TREE:
a733cb60 1877 case OBJ_BLOB:
a733cb60 1878 case OBJ_TAG:
21666f1a 1879 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
4d703a1a 1880 break;
1f688557 1881 default:
8eca0b47
NP
1882 data = NULL;
1883 error("unknown object type %i at offset %"PRIuMAX" in %s",
1884 *type, (uintmax_t)obj_offset, p->pack_name);
1f688557 1885 }
03e79c88 1886 unuse_pack(&w_curs);
21666f1a 1887 return data;
1f688557
JH
1888}
1889
d079837e 1890const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 1891 uint32_t n)
9a217f2a 1892{
42873078 1893 const unsigned char *index = p->index_data;
d079837e
SP
1894 if (!index) {
1895 if (open_pack_index(p))
1896 return NULL;
1897 index = p->index_data;
1898 }
57059091 1899 if (n >= p->num_objects)
d72308e0 1900 return NULL;
74e34e1f
NP
1901 index += 4 * 256;
1902 if (p->index_version == 1) {
1903 return index + 24 * n + 4;
1904 } else {
1905 index += 8;
1906 return index + 20 * n;
1907 }
1908}
1909
99093238 1910off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
1911{
1912 const unsigned char *index = p->index_data;
1913 index += 4 * 256;
1914 if (p->index_version == 1) {
1915 return ntohl(*((uint32_t *)(index + 24 * n)));
1916 } else {
1917 uint32_t off;
1918 index += 8 + p->num_objects * (20 + 4);
1919 off = ntohl(*((uint32_t *)(index + 4 * n)));
1920 if (!(off & 0x80000000))
1921 return off;
1922 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1923 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1924 ntohl(*((uint32_t *)(index + 4)));
1925 }
9a217f2a
JH
1926}
1927
c4001d92 1928off_t find_pack_entry_one(const unsigned char *sha1,
43057304 1929 struct packed_git *p)
1f688557 1930{
42873078 1931 const uint32_t *level1_ofs = p->index_data;
42873078 1932 const unsigned char *index = p->index_data;
628522ec
JH
1933 unsigned hi, lo, stride;
1934 static int use_lookup = -1;
1935 static int debug_lookup = -1;
1936
1937 if (debug_lookup < 0)
1938 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 1939
d079837e
SP
1940 if (!index) {
1941 if (open_pack_index(p))
1942 return 0;
1943 level1_ofs = p->index_data;
1944 index = p->index_data;
1945 }
74e34e1f
NP
1946 if (p->index_version > 1) {
1947 level1_ofs += 2;
1948 index += 8;
1949 }
42873078 1950 index += 4 * 256;
74e34e1f
NP
1951 hi = ntohl(level1_ofs[*sha1]);
1952 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
1953 if (p->index_version > 1) {
1954 stride = 20;
1955 } else {
1956 stride = 24;
1957 index += 4;
1958 }
1959
1960 if (debug_lookup)
6e1c2344 1961 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
1962 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1963
1964 if (use_lookup < 0)
1965 use_lookup = !!getenv("GIT_USE_LOOKUP");
1966 if (use_lookup) {
1967 int pos = sha1_entry_pos(index, stride, 0,
1968 lo, hi, p->num_objects, sha1);
1969 if (pos < 0)
1970 return 0;
1971 return nth_packed_object_offset(p, pos);
1972 }
1f688557
JH
1973
1974 do {
74e34e1f 1975 unsigned mi = (lo + hi) / 2;
628522ec
JH
1976 int cmp = hashcmp(index + mi * stride, sha1);
1977
1978 if (debug_lookup)
1979 printf("lo %u hi %u rg %u mi %u\n",
1980 lo, hi, hi - lo, mi);
43057304 1981 if (!cmp)
74e34e1f 1982 return nth_packed_object_offset(p, mi);
1f688557
JH
1983 if (cmp > 0)
1984 hi = mi;
1985 else
1986 lo = mi+1;
1987 } while (lo < hi);
1988 return 0;
1989}
1990
4c080182 1991int is_pack_valid(struct packed_git *p)
d131b7af
SP
1992{
1993 /* An already open pack is known to be valid. */
1994 if (p->pack_fd != -1)
1995 return 1;
1996
1997 /* If the pack has one window completely covering the
1998 * file size, the pack is known to be valid even if
1999 * the descriptor is not currently open.
2000 */
2001 if (p->windows) {
2002 struct pack_window *w = p->windows;
2003
2004 if (!w->offset && w->len == p->pack_size)
2005 return 1;
2006 }
2007
2008 /* Force the pack to open to prove its valid. */
2009 return !open_packed_git(p);
2010}
2011
4d6acb70 2012static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557 2013{
f7c22cc6 2014 static struct packed_git *last_found = (void *)1;
1f688557 2015 struct packed_git *p;
c4001d92 2016 off_t offset;
43057304 2017
1f688557 2018 prepare_packed_git();
f7c22cc6
NP
2019 if (!packed_git)
2020 return 0;
2021 p = (last_found == (void *)1) ? packed_git : last_found;
1f688557 2022
f7c22cc6 2023 do {
8eca0b47
NP
2024 if (p->num_bad_objects) {
2025 unsigned i;
2026 for (i = 0; i < p->num_bad_objects; i++)
2027 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2028 goto next;
2029 }
2030
43057304
NP
2031 offset = find_pack_entry_one(sha1, p);
2032 if (offset) {
c715f783
SP
2033 /*
2034 * We are about to tell the caller where they can
2035 * locate the requested object. We better make
2036 * sure the packfile is still here and can be
2037 * accessed before supplying that answer, as
2038 * it may have been deleted since the index
2039 * was loaded!
2040 */
d131b7af 2041 if (!is_pack_valid(p)) {
58a6a9cc 2042 warning("packfile %s cannot be accessed", p->pack_name);
f7c22cc6 2043 goto next;
c715f783 2044 }
43057304
NP
2045 e->offset = offset;
2046 e->p = p;
2047 hashcpy(e->sha1, sha1);
f7c22cc6 2048 last_found = p;
1f688557 2049 return 1;
43057304 2050 }
f7c22cc6
NP
2051
2052 next:
2053 if (p == last_found)
2054 p = packed_git;
2055 else
2056 p = p->next;
2057 if (p == last_found)
2058 p = p->next;
2059 } while (p);
1f688557
JH
2060 return 0;
2061}
2062
a6080a0a 2063struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
2064 struct packed_git *packs)
2065{
2066 struct packed_git *p;
bf592c50
DB
2067
2068 for (p = packs; p; p = p->next) {
43057304 2069 if (find_pack_entry_one(sha1, p))
bf592c50
DB
2070 return p;
2071 }
2072 return NULL;
21666f1a 2073
bf592c50
DB
2074}
2075
21666f1a 2076static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
65c2e0c3 2077{
36e4d74a 2078 int status;
65c2e0c3
JH
2079 unsigned long mapsize, size;
2080 void *map;
ef49a7a0 2081 git_zstream stream;
d65a16f6 2082 char hdr[32];
65c2e0c3 2083
bb6b8e4f 2084 map = map_sha1_file(sha1, &mapsize);
f0df4ed5
JS
2085 if (!map)
2086 return error("unable to find %s", sha1_to_hex(sha1));
36e4d74a
JH
2087 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2088 status = error("unable to unpack %s header",
2089 sha1_to_hex(sha1));
21666f1a 2090 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 2091 status = error("unable to parse %s header", sha1_to_hex(sha1));
21666f1a
NP
2092 else if (sizep)
2093 *sizep = size;
39c68542 2094 git_inflate_end(&stream);
65c2e0c3
JH
2095 munmap(map, mapsize);
2096 return status;
2097}
2098
9a490590
JH
2099/* returns enum object_type or negative */
2100int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
f0df4ed5 2101{
c4d9986f 2102 struct cached_object *co;
f0df4ed5 2103 struct pack_entry e;
9a490590 2104 int status, rtype;
f0df4ed5 2105
c4d9986f
NTND
2106 co = find_cached_object(sha1);
2107 if (co) {
9a490590
JH
2108 if (oi->sizep)
2109 *(oi->sizep) = co->size;
2110 oi->whence = OI_CACHED;
c4d9986f
NTND
2111 return co->type;
2112 }
2113
cd673c1f 2114 if (!find_pack_entry(sha1, &e)) {
ddd63e64 2115 /* Most likely it's a loose object. */
9a490590
JH
2116 status = sha1_loose_object_info(sha1, oi->sizep);
2117 if (status >= 0) {
2118 oi->whence = OI_LOOSE;
ddd63e64 2119 return status;
9a490590 2120 }
ddd63e64
SG
2121
2122 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 2123 reprepare_packed_git();
cd673c1f 2124 if (!find_pack_entry(sha1, &e))
ddd63e64 2125 return status;
f0df4ed5 2126 }
3d77d877 2127
9a490590 2128 status = packed_object_info(e.p, e.offset, oi->sizep, &rtype);
3d77d877
NP
2129 if (status < 0) {
2130 mark_bad_packed_object(e.p, sha1);
9a490590 2131 status = sha1_object_info_extended(sha1, oi);
5266d369
JH
2132 } else if (in_delta_base_cache(e.p, e.offset)) {
2133 oi->whence = OI_DBCACHED;
9a490590
JH
2134 } else {
2135 oi->whence = OI_PACKED;
2136 oi->u.packed.offset = e.offset;
2137 oi->u.packed.pack = e.p;
2138 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2139 rtype == OBJ_OFS_DELTA);
3d77d877
NP
2140 }
2141
2142 return status;
f0df4ed5
JS
2143}
2144
9a490590
JH
2145int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2146{
2147 struct object_info oi;
2148
2149 oi.sizep = sizep;
2150 return sha1_object_info_extended(sha1, &oi);
2151}
2152
21666f1a
NP
2153static void *read_packed_sha1(const unsigned char *sha1,
2154 enum object_type *type, unsigned long *size)
1f688557
JH
2155{
2156 struct pack_entry e;
8eca0b47 2157 void *data;
1f688557 2158
cd673c1f 2159 if (!find_pack_entry(sha1, &e))
1f688557 2160 return NULL;
8eca0b47
NP
2161 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2162 if (!data) {
2163 /*
2164 * We're probably in deep shit, but let's try to fetch
2165 * the required object anyway from another pack or loose.
2166 * This should happen only in the presence of a corrupted
2167 * pack, and is better than failing outright.
2168 */
2169 error("failed to read object %s at offset %"PRIuMAX" from %s",
2170 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2171 mark_bad_packed_object(e.p, sha1);
ac939109 2172 data = read_object(sha1, type, size);
8eca0b47
NP
2173 }
2174 return data;
1f688557
JH
2175}
2176
21666f1a
NP
2177int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2178 unsigned char *sha1)
d66b37bb
JH
2179{
2180 struct cached_object *co;
2181
21666f1a 2182 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2183 if (has_sha1_file(sha1) || find_cached_object(sha1))
2184 return 0;
2185 if (cached_object_alloc <= cached_object_nr) {
2186 cached_object_alloc = alloc_nr(cached_object_alloc);
2187 cached_objects = xrealloc(cached_objects,
2188 sizeof(*cached_objects) *
2189 cached_object_alloc);
2190 }
2191 co = &cached_objects[cached_object_nr++];
2192 co->size = len;
21666f1a 2193 co->type = type;
efa13f7b
JH
2194 co->buf = xmalloc(len);
2195 memcpy(co->buf, buf, len);
d66b37bb
JH
2196 hashcpy(co->sha1, sha1);
2197 return 0;
2198}
2199
c2c5b270
CC
2200static void *read_object(const unsigned char *sha1, enum object_type *type,
2201 unsigned long *size)
0fcfd160
LT
2202{
2203 unsigned long mapsize;
2204 void *map, *buf;
d66b37bb
JH
2205 struct cached_object *co;
2206
2207 co = find_cached_object(sha1);
2208 if (co) {
21666f1a 2209 *type = co->type;
d66b37bb 2210 *size = co->size;
182af834 2211 return xmemdupz(co->buf, co->size);
d66b37bb 2212 }
0fcfd160 2213
8276c007
PE
2214 buf = read_packed_sha1(sha1, type, size);
2215 if (buf)
2216 return buf;
bb6b8e4f 2217 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2218 if (map) {
7efbff75 2219 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2220 munmap(map, mapsize);
2221 return buf;
2222 }
637cdd9d 2223 reprepare_packed_git();
8276c007 2224 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2225}
2226
b6c4cecc
JH
2227/*
2228 * This function dies on corrupt objects; the callers who want to
2229 * deal with them should arrange to call read_object() and give error
2230 * messages themselves.
2231 */
5bf29b95
JH
2232void *read_sha1_file_extended(const unsigned char *sha1,
2233 enum object_type *type,
2234 unsigned long *size,
2235 unsigned flag)
ac939109 2236{
3ba7a065 2237 void *data;
e8b15e61 2238 char *path;
b6c4cecc 2239 const struct packed_git *p;
5bf29b95
JH
2240 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2241 ? lookup_replace_object(sha1) : sha1;
b6c4cecc 2242
3ba7a065
JH
2243 errno = 0;
2244 data = read_object(repl, type, size);
4bbf5a26 2245 if (data)
b6c4cecc 2246 return data;
68095570 2247
25f3af3f 2248 if (errno && errno != ENOENT)
3ba7a065
JH
2249 die_errno("failed to read object %s", sha1_to_hex(sha1));
2250
68095570 2251 /* die if we replaced an object with one that does not exist */
b6c4cecc 2252 if (repl != sha1)
68095570
CC
2253 die("replacement %s not found for %s",
2254 sha1_to_hex(repl), sha1_to_hex(sha1));
2255
b6c4cecc
JH
2256 if (has_loose_object(repl)) {
2257 path = sha1_file_name(sha1);
2258 die("loose object %s (stored in %s) is corrupt",
2259 sha1_to_hex(repl), path);
e8b15e61 2260 }
68095570 2261
b6c4cecc
JH
2262 if ((p = has_packed_and_bad(repl)) != NULL)
2263 die("packed object %s (stored in %s) is corrupt",
2264 sha1_to_hex(repl), p->pack_name);
f5552aee 2265
b6c4cecc 2266 return NULL;
ac939109
NP
2267}
2268
40469ee9 2269void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2270 const char *required_type_name,
40469ee9
JH
2271 unsigned long *size,
2272 unsigned char *actual_sha1_return)
f4913f91 2273{
21666f1a 2274 enum object_type type, required_type;
f4913f91
JH
2275 void *buffer;
2276 unsigned long isize;
40469ee9 2277 unsigned char actual_sha1[20];
f4913f91 2278
21666f1a 2279 required_type = type_from_string(required_type_name);
e702496e 2280 hashcpy(actual_sha1, sha1);
40469ee9
JH
2281 while (1) {
2282 int ref_length = -1;
2283 const char *ref_type = NULL;
f4913f91 2284
21666f1a 2285 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2286 if (!buffer)
2287 return NULL;
21666f1a 2288 if (type == required_type) {
40469ee9
JH
2289 *size = isize;
2290 if (actual_sha1_return)
e702496e 2291 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2292 return buffer;
2293 }
2294 /* Handle references */
21666f1a 2295 else if (type == OBJ_COMMIT)
40469ee9 2296 ref_type = "tree ";
21666f1a 2297 else if (type == OBJ_TAG)
40469ee9
JH
2298 ref_type = "object ";
2299 else {
2300 free(buffer);
2301 return NULL;
2302 }
2303 ref_length = strlen(ref_type);
f4913f91 2304
50974ec9
MK
2305 if (ref_length + 40 > isize ||
2306 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2307 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2308 free(buffer);
2309 return NULL;
2310 }
1cf58e72 2311 free(buffer);
40469ee9
JH
2312 /* Now we have the ID of the referred-to object in
2313 * actual_sha1. Check again. */
f4913f91 2314 }
f4913f91
JH
2315}
2316
ce9fbf16 2317static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2318 const char *type, unsigned char *sha1,
d65a16f6 2319 char *hdr, int *hdrlen)
d410c0f5 2320{
9126f009 2321 git_SHA_CTX c;
d410c0f5
JH
2322
2323 /* Generate the header */
d65a16f6 2324 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
2325
2326 /* Sha1.. */
9126f009
NP
2327 git_SHA1_Init(&c);
2328 git_SHA1_Update(&c, hdr, *hdrlen);
2329 git_SHA1_Update(&c, buf, len);
2330 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2331}
2332
230f1322 2333/*
5a688fe4
JH
2334 * Move the just written object into its final resting place.
2335 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2336 * "moving" is only a part of what it does, when no patch between
2337 * master to pu changes the call sites of this function.
230f1322 2338 */
839837b9 2339int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322 2340{
e32c0a9c 2341 int ret = 0;
5a688fe4 2342
348df166 2343 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
2344 goto try_rename;
2345 else if (link(tmpfile, filename))
e32c0a9c 2346 ret = errno;
7ebb6fca
LT
2347
2348 /*
2349 * Coda hack - coda doesn't like cross-directory links,
2350 * so we fall back to a rename, which will mean that it
2351 * won't be able to check collisions, but that's not a
2352 * big deal.
2353 *
2354 * The same holds for FAT formatted media.
2355 *
3be1f18e 2356 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
2357 * left to unlink.
2358 */
2359 if (ret && ret != EEXIST) {
be66a6c4 2360 try_rename:
7ebb6fca 2361 if (!rename(tmpfile, filename))
3be1f18e 2362 goto out;
9e48b389 2363 ret = errno;
230f1322 2364 }
691f1a28 2365 unlink_or_warn(tmpfile);
230f1322
LT
2366 if (ret) {
2367 if (ret != EEXIST) {
916d081b 2368 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
230f1322
LT
2369 }
2370 /* FIXME!!! Collision check here ? */
2371 }
2372
3be1f18e 2373out:
5256b006 2374 if (adjust_shared_perm(filename))
5a688fe4 2375 return error("unable to set permission to '%s'", filename);
230f1322
LT
2376 return 0;
2377}
2378
4d548150
LT
2379static int write_buffer(int fd, const void *buf, size_t len)
2380{
d34cf19b 2381 if (write_in_full(fd, buf, len) < 0)
93822c22 2382 return error("file write error (%s)", strerror(errno));
4d548150
LT
2383 return 0;
2384}
2385
ce9fbf16 2386int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2387 unsigned char *sha1)
2388{
d65a16f6 2389 char hdr[32];
abdc3fc8
RS
2390 int hdrlen;
2391 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2392 return 0;
2393}
2394
e9039dd3
LT
2395/* Finalize a file on disk, and close it. */
2396static void close_sha1_file(int fd)
2397{
aafe9fba
LT
2398 if (fsync_object_files)
2399 fsync_or_die(fd, "sha1 file");
e9039dd3 2400 if (close(fd) != 0)
d824cbba 2401 die_errno("error when closing sha1 file");
e9039dd3
LT
2402}
2403
5723fe7e
LT
2404/* Size of directory component, including the ending '/' */
2405static inline int directory_size(const char *filename)
2406{
2407 const char *s = strrchr(filename, '/');
2408 if (!s)
2409 return 0;
2410 return s - filename + 1;
2411}
2412
2413/*
2414 * This creates a temporary file in the same directory as the final
2415 * 'filename'
2416 *
2417 * We want to avoid cross-directory filename renames, because those
2418 * can have problems on various filesystems (FAT, NFS, Coda).
2419 */
2420static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2421{
2422 int fd, dirlen = directory_size(filename);
2423
2424 if (dirlen + 20 > bufsiz) {
2425 errno = ENAMETOOLONG;
2426 return -1;
2427 }
2428 memcpy(buffer, filename, dirlen);
2429 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
5256b006 2430 fd = git_mkstemp_mode(buffer, 0444);
cbacbf4e 2431 if (fd < 0 && dirlen && errno == ENOENT) {
5723fe7e 2432 /* Make sure the directory exists */
6ff6af62 2433 memcpy(buffer, filename, dirlen);
5723fe7e 2434 buffer[dirlen-1] = 0;
3bfaf018 2435 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
5723fe7e
LT
2436 return -1;
2437
2438 /* Try again */
2439 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
5256b006 2440 fd = git_mkstemp_mode(buffer, 0444);
5723fe7e
LT
2441 }
2442 return fd;
2443}
2444
bbac7311 2445static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 2446 const void *buf, unsigned long len, time_t mtime)
0fcfd160 2447{
915308b1 2448 int fd, ret;
9892beba 2449 unsigned char compressed[4096];
ef49a7a0 2450 git_zstream stream;
748af44c
NP
2451 git_SHA_CTX c;
2452 unsigned char parano_sha1[20];
706bc531 2453 char *filename;
aac17941 2454 static char tmpfile[PATH_MAX];
a44c9a5e 2455
972a9155 2456 filename = sha1_file_name(sha1);
5723fe7e 2457 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
aac17941 2458 if (fd < 0) {
35243577 2459 if (errno == EACCES)
916d081b
PB
2460 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2461 else
2462 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
aac17941
LT
2463 }
2464
0fcfd160
LT
2465 /* Set it up */
2466 memset(&stream, 0, sizeof(stream));
55bb5c91 2467 git_deflate_init(&stream, zlib_compression_level);
0fcfd160 2468 stream.next_out = compressed;
9892beba 2469 stream.avail_out = sizeof(compressed);
748af44c 2470 git_SHA1_Init(&c);
a44c9a5e
LT
2471
2472 /* First header.. */
d65a16f6 2473 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2474 stream.avail_in = hdrlen;
55bb5c91
JH
2475 while (git_deflate(&stream, 0) == Z_OK)
2476 ; /* nothing */
748af44c 2477 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
2478
2479 /* Then the data itself.. */
c00e657d 2480 stream.next_in = (void *)buf;
a44c9a5e 2481 stream.avail_in = len;
9892beba 2482 do {
748af44c 2483 unsigned char *in0 = stream.next_in;
55bb5c91 2484 ret = git_deflate(&stream, Z_FINISH);
748af44c 2485 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
2486 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2487 die("unable to write sha1 file");
2488 stream.next_out = compressed;
2489 stream.avail_out = sizeof(compressed);
2490 } while (ret == Z_OK);
2491
ac54c277
LT
2492 if (ret != Z_STREAM_END)
2493 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
55bb5c91 2494 ret = git_deflate_end_gently(&stream);
ac54c277
LT
2495 if (ret != Z_OK)
2496 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
2497 git_SHA1_Final(parano_sha1, &c);
2498 if (hashcmp(sha1, parano_sha1) != 0)
2499 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 2500
e9039dd3 2501 close_sha1_file(fd);
0fcfd160 2502
bbac7311
NP
2503 if (mtime) {
2504 struct utimbuf utb;
2505 utb.actime = mtime;
2506 utb.modtime = mtime;
2507 if (utime(tmpfile, &utb) < 0)
2508 warning("failed utime() on %s: %s",
2509 tmpfile, strerror(errno));
2510 }
2511
230f1322 2512 return move_temp_to_file(tmpfile, filename);
0fcfd160 2513}
8237b185 2514
c00e657d 2515int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
bbac7311
NP
2516{
2517 unsigned char sha1[20];
2518 char hdr[32];
2519 int hdrlen;
2520
2521 /* Normally if we have it in the pack then we do not bother writing
2522 * it out into .git/objects/??/?{38} file.
2523 */
2524 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2525 if (returnsha1)
2526 hashcpy(returnsha1, sha1);
2527 if (has_sha1_file(sha1))
2528 return 0;
2529 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2530}
2531
2532int force_object_loose(const unsigned char *sha1, time_t mtime)
2533{
bbac7311
NP
2534 void *buf;
2535 unsigned long len;
2536 enum object_type type;
2537 char hdr[32];
2538 int hdrlen;
1fb23e65 2539 int ret;
bbac7311 2540
c529d75a 2541 if (has_loose_object(sha1))
bbac7311
NP
2542 return 0;
2543 buf = read_packed_sha1(sha1, &type, &len);
2544 if (!buf)
2545 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2546 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1fb23e65
BS
2547 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2548 free(buf);
2549
2550 return ret;
bbac7311
NP
2551}
2552
bf592c50
DB
2553int has_pack_index(const unsigned char *sha1)
2554{
2555 struct stat st;
2556 if (stat(sha1_pack_index_name(sha1), &st))
2557 return 0;
2558 return 1;
bf592c50
DB
2559}
2560
cd673c1f
JH
2561int has_sha1_pack(const unsigned char *sha1)
2562{
2563 struct pack_entry e;
2564 return find_pack_entry(sha1, &e);
2565}
2566
8237b185
DB
2567int has_sha1_file(const unsigned char *sha1)
2568{
1f688557
JH
2569 struct pack_entry e;
2570
cd673c1f 2571 if (find_pack_entry(sha1, &e))
1f688557 2572 return 1;
c529d75a 2573 return has_loose_object(sha1);
8237b185 2574}
74400e71 2575
c879daa2
NTND
2576static void check_tree(const void *buf, size_t size)
2577{
2578 struct tree_desc desc;
2579 struct name_entry entry;
2580
2581 init_tree_desc(&desc, buf, size);
2582 while (tree_entry(&desc, &entry))
2583 /* do nothing
2584 * tree_entry() will die() on malformed entries */
2585 ;
2586}
2587
2588static void check_commit(const void *buf, size_t size)
2589{
2590 struct commit c;
2591 memset(&c, 0, sizeof(c));
2592 if (parse_commit_buffer(&c, buf, size))
2593 die("corrupt commit");
2594}
2595
2596static void check_tag(const void *buf, size_t size)
2597{
2598 struct tag t;
2599 memset(&t, 0, sizeof(t));
2600 if (parse_tag_buffer(&t, buf, size))
2601 die("corrupt tag");
2602}
2603
43df4f86 2604static int index_mem(unsigned char *sha1, void *buf, size_t size,
c4ce46fc
JH
2605 enum object_type type,
2606 const char *path, unsigned flags)
e7332f96 2607{
6c510bee 2608 int ret, re_allocated = 0;
c4ce46fc 2609 int write_object = flags & HASH_WRITE_OBJECT;
74400e71 2610
7672db20 2611 if (!type)
edaec3fb 2612 type = OBJ_BLOB;
6c510bee
LT
2613
2614 /*
2615 * Convert blobs to git internal format
2616 */
43df4f86 2617 if ((type == OBJ_BLOB) && path) {
f285a2d7 2618 struct strbuf nbuf = STRBUF_INIT;
21e5ad50 2619 if (convert_to_git(path, buf, size, &nbuf,
5e12e78e 2620 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
b315c5c0 2621 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
2622 re_allocated = 1;
2623 }
2624 }
c4ce46fc 2625 if (flags & HASH_FORMAT_CHECK) {
c879daa2
NTND
2626 if (type == OBJ_TREE)
2627 check_tree(buf, size);
2628 if (type == OBJ_COMMIT)
2629 check_commit(buf, size);
2630 if (type == OBJ_TAG)
2631 check_tag(buf, size);
2632 }
6c510bee 2633
7672db20 2634 if (write_object)
edaec3fb 2635 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 2636 else
edaec3fb 2637 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 2638 if (re_allocated)
6c510bee 2639 free(buf);
43df4f86
DP
2640 return ret;
2641}
2642
7b41e1e1
JH
2643static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2644 const char *path, unsigned flags)
2645{
2646 struct strbuf sbuf = STRBUF_INIT;
2647 int ret;
2648
2649 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2650 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2651 else
2652 ret = -1;
2653 strbuf_release(&sbuf);
2654 return ret;
2655}
2656
ea68b0ce
DP
2657#define SMALL_FILE_SIZE (32*1024)
2658
7b41e1e1
JH
2659static int index_core(unsigned char *sha1, int fd, size_t size,
2660 enum object_type type, const char *path,
2661 unsigned flags)
43df4f86
DP
2662{
2663 int ret;
43df4f86 2664
7b41e1e1 2665 if (!size) {
c4ce46fc 2666 ret = index_mem(sha1, NULL, size, type, path, flags);
ea68b0ce
DP
2667 } else if (size <= SMALL_FILE_SIZE) {
2668 char *buf = xmalloc(size);
2669 if (size == read_in_full(fd, buf, size))
c4ce46fc 2670 ret = index_mem(sha1, buf, size, type, path, flags);
ea68b0ce
DP
2671 else
2672 ret = error("short read %s", strerror(errno));
2673 free(buf);
08bda208 2674 } else {
43df4f86 2675 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
c4ce46fc 2676 ret = index_mem(sha1, buf, size, type, path, flags);
aac17941 2677 munmap(buf, size);
08bda208 2678 }
7b41e1e1
JH
2679 return ret;
2680}
2681
4dd1fbc7
JH
2682/*
2683 * This creates one packfile per large blob, because the caller
2684 * immediately wants the result sha1, and fast-import can report the
2685 * object name via marks mechanism only by closing the created
2686 * packfile.
2687 *
2688 * This also bypasses the usual "convert-to-git" dance, and that is on
2689 * purpose. We could write a streaming version of the converting
2690 * functions and insert that before feeding the data to fast-import
2691 * (or equivalent in-core API described above), but the primary
2692 * motivation for trying to stream from the working tree file and to
2693 * avoid mmaping it in core is to deal with large binary blobs, and
2694 * by definition they do _not_ want to get any conversion.
2695 */
2696static int index_stream(unsigned char *sha1, int fd, size_t size,
2697 enum object_type type, const char *path,
2698 unsigned flags)
2699{
2700 struct child_process fast_import;
2701 char export_marks[512];
2702 const char *argv[] = { "fast-import", "--quiet", export_marks, NULL };
2703 char tmpfile[512];
2704 char fast_import_cmd[512];
2705 char buf[512];
2706 int len, tmpfd;
2707
2708 strcpy(tmpfile, git_path("hashstream_XXXXXX"));
2709 tmpfd = git_mkstemp_mode(tmpfile, 0600);
2710 if (tmpfd < 0)
2711 die_errno("cannot create tempfile: %s", tmpfile);
2712 if (close(tmpfd))
2713 die_errno("cannot close tempfile: %s", tmpfile);
2714 sprintf(export_marks, "--export-marks=%s", tmpfile);
2715
2716 memset(&fast_import, 0, sizeof(fast_import));
2717 fast_import.in = -1;
2718 fast_import.argv = argv;
2719 fast_import.git_cmd = 1;
2720 if (start_command(&fast_import))
2721 die_errno("index-stream: git fast-import failed");
2722
2723 len = sprintf(fast_import_cmd, "blob\nmark :1\ndata %lu\n",
2724 (unsigned long) size);
2725 write_or_whine(fast_import.in, fast_import_cmd, len,
2726 "index-stream: feeding fast-import");
2727 while (size) {
2728 char buf[10240];
2729 size_t sz = size < sizeof(buf) ? size : sizeof(buf);
23c7df6b 2730 ssize_t actual;
4dd1fbc7
JH
2731
2732 actual = read_in_full(fd, buf, sz);
2733 if (actual < 0)
2734 die_errno("index-stream: reading input");
2735 if (write_in_full(fast_import.in, buf, actual) != actual)
2736 die_errno("index-stream: feeding fast-import");
2737 size -= actual;
2738 }
2739 if (close(fast_import.in))
2740 die_errno("index-stream: closing fast-import");
2741 if (finish_command(&fast_import))
2742 die_errno("index-stream: finishing fast-import");
2743
2744 tmpfd = open(tmpfile, O_RDONLY);
2745 if (tmpfd < 0)
2746 die_errno("index-stream: cannot open fast-import mark");
2747 len = read(tmpfd, buf, sizeof(buf));
2748 if (len < 0)
2749 die_errno("index-stream: reading fast-import mark");
2750 if (close(tmpfd) < 0)
2751 die_errno("index-stream: closing fast-import mark");
2752 if (unlink(tmpfile))
2753 die_errno("index-stream: unlinking fast-import mark");
2754 if (len != 44 ||
2755 memcmp(":1 ", buf, 3) ||
2756 get_sha1_hex(buf + 3, sha1))
2757 die_errno("index-stream: unexpected fast-import mark: <%s>", buf);
2758 return 0;
2759}
2760
7b41e1e1
JH
2761int index_fd(unsigned char *sha1, int fd, struct stat *st,
2762 enum object_type type, const char *path, unsigned flags)
2763{
2764 int ret;
2765 size_t size = xsize_t(st->st_size);
2766
2767 if (!S_ISREG(st->st_mode))
2768 ret = index_pipe(sha1, fd, type, path, flags);
4dd1fbc7 2769 else if (size <= big_file_threshold || type != OBJ_BLOB)
7b41e1e1 2770 ret = index_core(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
2771 else
2772 ret = index_stream(sha1, fd, size, type, path, flags);
43df4f86 2773 close(fd);
aac17941 2774 return ret;
74400e71 2775}
ec1fcc16 2776
c4ce46fc 2777int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
ec1fcc16
JH
2778{
2779 int fd;
b760d3aa 2780 struct strbuf sb = STRBUF_INIT;
ec1fcc16
JH
2781
2782 switch (st->st_mode & S_IFMT) {
2783 case S_IFREG:
2784 fd = open(path, O_RDONLY);
2785 if (fd < 0)
2786 return error("open(\"%s\"): %s", path,
2787 strerror(errno));
c4ce46fc 2788 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
ec1fcc16
JH
2789 return error("%s: failed to insert into database",
2790 path);
2791 break;
2792 case S_IFLNK:
b760d3aa 2793 if (strbuf_readlink(&sb, path, st->st_size)) {
ec1fcc16 2794 char *errstr = strerror(errno);
ec1fcc16
JH
2795 return error("readlink(\"%s\"): %s", path,
2796 errstr);
2797 }
c4ce46fc 2798 if (!(flags & HASH_WRITE_OBJECT))
b760d3aa
LT
2799 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2800 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
ec1fcc16
JH
2801 return error("%s: failed to insert into database",
2802 path);
b760d3aa 2803 strbuf_release(&sb);
ec1fcc16 2804 break;
f35a6d3b
LT
2805 case S_IFDIR:
2806 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
2807 default:
2808 return error("%s: unsupported file type", path);
2809 }
2810 return 0;
2811}
a69e5429
JH
2812
2813int read_pack_header(int fd, struct pack_header *header)
2814{
c697ad14
HO
2815 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2816 /* "eof before pack header was fully read" */
2817 return PH_ERROR_EOF;
2818
a69e5429
JH
2819 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2820 /* "protocol error (pack signature mismatch detected)" */
2821 return PH_ERROR_PACK_SIGNATURE;
2822 if (!pack_version_ok(header->hdr_version))
2823 /* "protocol error (pack version unsupported)" */
2824 return PH_ERROR_PROTOCOL;
2825 return 0;
2826}
40d52ff7
JK
2827
2828void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2829{
2830 enum object_type type = sha1_object_info(sha1, NULL);
2831 if (type < 0)
2832 die("%s is not a valid object", sha1_to_hex(sha1));
2833 if (type != expect)
2834 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2835 typename(expect));
2836}