Merge branch 'ef/maint-win-verify-path'
[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;
251 /* 43 = 40-byte + 2 '/' + terminating NUL */
252 int pfxlen = len;
253 int entlen = pfxlen + 43;
ccfd3e99 254 int base_len = -1;
d5a63b99 255
85dadc38 256 if (!is_absolute_path(entry) && relative_base) {
c2f493a4
MW
257 /* Relative alt-odb */
258 if (base_len < 0)
259 base_len = strlen(relative_base) + 1;
260 entlen += base_len;
261 pfxlen += base_len;
262 }
263 ent = xmalloc(sizeof(*ent) + entlen);
264
85dadc38 265 if (!is_absolute_path(entry) && relative_base) {
c2f493a4
MW
266 memcpy(ent->base, relative_base, base_len - 1);
267 ent->base[base_len - 1] = '/';
268 memcpy(ent->base + base_len, entry, len);
269 }
270 else
271 memcpy(ent->base, entry, pfxlen);
272
273 ent->name = ent->base + pfxlen + 1;
274 ent->base[pfxlen + 3] = '/';
275 ent->base[pfxlen] = ent->base[entlen-1] = 0;
276
277 /* Detect cases where alternate disappeared */
90b4a71c 278 if (!is_directory(ent->base)) {
c2f493a4
MW
279 error("object directory %s does not exist; "
280 "check .git/objects/info/alternates.",
281 ent->base);
282 free(ent);
283 return -1;
284 }
285
286 /* Prevent the common mistake of listing the same
287 * thing twice, or object directory itself.
288 */
289 for (alt = alt_odb_list; alt; alt = alt->next) {
290 if (!memcmp(ent->base, alt->base, pfxlen)) {
291 free(ent);
292 return -1;
293 }
294 }
295 if (!memcmp(ent->base, objdir, pfxlen)) {
296 free(ent);
297 return -1;
298 }
299
300 /* add the alternate entry */
301 *alt_odb_tail = ent;
302 alt_odb_tail = &(ent->next);
303 ent->next = NULL;
304
305 /* recursively add alternates */
306 read_info_alternates(ent->base, depth + 1);
307
308 ent->base[pfxlen] = '/';
309
310 return 0;
311}
312
313static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
314 const char *relative_base, int depth)
315{
316 const char *cp, *last;
317
318 if (depth > 5) {
319 error("%s: ignoring alternate object stores, nesting too deep.",
320 relative_base);
321 return;
322 }
323
d5a63b99 324 last = alt;
9577e7e3
JH
325 while (last < ep) {
326 cp = last;
327 if (cp < ep && *cp == '#') {
328 while (cp < ep && *cp != sep)
329 cp++;
330 last = cp + 1;
331 continue;
332 }
c2f493a4
MW
333 while (cp < ep && *cp != sep)
334 cp++;
d5a63b99 335 if (last != cp) {
85dadc38 336 if (!is_absolute_path(last) && depth) {
c2f493a4
MW
337 error("%s: ignoring relative alternate object store %s",
338 relative_base, last);
339 } else {
340 link_alt_odb_entry(last, cp - last,
341 relative_base, depth);
1494e038 342 }
d5a63b99 343 }
9577e7e3 344 while (cp < ep && *cp == sep)
d5a63b99
JH
345 cp++;
346 last = cp;
9577e7e3 347 }
d5a63b99
JH
348}
349
c2f493a4 350static void read_info_alternates(const char * relative_base, int depth)
d5a63b99 351{
9577e7e3 352 char *map;
dc49cd76 353 size_t mapsz;
d5a63b99 354 struct stat st;
9cb18f56
JM
355 const char alt_file_name[] = "info/alternates";
356 /* Given that relative_base is no longer than PATH_MAX,
357 ensure that "path" has enough space to append "/", the
358 file name, "info/alternates", and a trailing NUL. */
359 char path[PATH_MAX + 1 + sizeof alt_file_name];
c2f493a4 360 int fd;
d5a63b99 361
9cb18f56 362 sprintf(path, "%s/%s", relative_base, alt_file_name);
f4e51683 363 fd = git_open_noatime(path);
d5a63b99
JH
364 if (fd < 0)
365 return;
366 if (fstat(fd, &st) || (st.st_size == 0)) {
367 close(fd);
9a217f2a 368 return;
ace1534d 369 }
dc49cd76
SP
370 mapsz = xsize_t(st.st_size);
371 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 372 close(fd);
d5a63b99 373
dc49cd76 374 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
c2f493a4 375
dc49cd76 376 munmap(map, mapsz);
ace1534d
JH
377}
378
bef70b22
DB
379void add_to_alternates_file(const char *reference)
380{
381 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
acd3b9ec 382 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
bef70b22
DB
383 char *alt = mkpath("%s/objects\n", reference);
384 write_or_die(fd, alt, strlen(alt));
385 if (commit_lock_file(lock))
386 die("could not close alternates file");
387 if (alt_odb_tail)
388 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
389}
390
d79796bc
JH
391void foreach_alt_odb(alt_odb_fn fn, void *cb)
392{
393 struct alternate_object_database *ent;
394
395 prepare_alt_odb();
396 for (ent = alt_odb_list; ent; ent = ent->next)
397 if (fn(ent, cb))
398 return;
399}
400
c2f493a4
MW
401void prepare_alt_odb(void)
402{
554fe20d 403 const char *alt;
c2f493a4 404
7dc24aa5
SP
405 if (alt_odb_tail)
406 return;
407
c2f493a4
MW
408 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
409 if (!alt) alt = "";
410
c2f493a4 411 alt_odb_tail = &alt_odb_list;
80ba074f 412 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
413
414 read_info_alternates(get_object_directory(), 0);
415}
416
0f4dc14a 417static int has_loose_object_local(const unsigned char *sha1)
ace1534d 418{
ace1534d 419 char *name = sha1_file_name(sha1);
0f4dc14a
BC
420 return !access(name, F_OK);
421}
ace1534d 422
0f4dc14a
BC
423int has_loose_object_nonlocal(const unsigned char *sha1)
424{
425 struct alternate_object_database *alt;
9a217f2a 426 prepare_alt_odb();
d5a63b99 427 for (alt = alt_odb_list; alt; alt = alt->next) {
0f4dc14a 428 fill_sha1_path(alt->name, sha1);
c529d75a
LT
429 if (!access(alt->base, F_OK))
430 return 1;
ace1534d 431 }
c529d75a 432 return 0;
ace1534d
JH
433}
434
0f4dc14a
BC
435static int has_loose_object(const unsigned char *sha1)
436{
437 return has_loose_object_local(sha1) ||
438 has_loose_object_nonlocal(sha1);
439}
440
60bb8b14 441static unsigned int pack_used_ctr;
a53128b6
SP
442static unsigned int pack_mmap_calls;
443static unsigned int peak_pack_open_windows;
444static unsigned int pack_open_windows;
c7934306
SP
445static unsigned int pack_open_fds;
446static unsigned int pack_max_fds;
a53128b6 447static size_t peak_pack_mapped;
60bb8b14 448static size_t pack_mapped;
9a217f2a 449struct packed_git *packed_git;
1f688557 450
b79d18c9 451void pack_report(void)
a53128b6
SP
452{
453 fprintf(stderr,
e05db0fd
PR
454 "pack_report: getpagesize() = %10" SZ_FMT "\n"
455 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
456 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
457 sz_fmt(getpagesize()),
458 sz_fmt(packed_git_window_size),
459 sz_fmt(packed_git_limit));
a53128b6
SP
460 fprintf(stderr,
461 "pack_report: pack_used_ctr = %10u\n"
462 "pack_report: pack_mmap_calls = %10u\n"
463 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
464 "pack_report: pack_mapped = "
465 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
466 pack_used_ctr,
467 pack_mmap_calls,
468 pack_open_windows, peak_pack_open_windows,
9e42d6a1 469 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
470}
471
42873078 472static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
473{
474 void *idx_map;
42873078 475 struct pack_idx_header *hdr;
dc49cd76 476 size_t idx_size;
74e34e1f 477 uint32_t version, nr, i, *index;
f4e51683 478 int fd = git_open_noatime(path);
1f688557 479 struct stat st;
42873078 480
1f688557
JH
481 if (fd < 0)
482 return -1;
483 if (fstat(fd, &st)) {
484 close(fd);
485 return -1;
486 }
dc49cd76 487 idx_size = xsize_t(st.st_size);
2d88451b
SP
488 if (idx_size < 4 * 256 + 20 + 20) {
489 close(fd);
490 return error("index file %s is too small", path);
491 }
c4712e45 492 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 493 close(fd);
1f688557 494
42873078
NP
495 hdr = idx_map;
496 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
497 version = ntohl(hdr->idx_version);
498 if (version < 2 || version > 2) {
499 munmap(idx_map, idx_size);
6e1c2344 500 return error("index file %s is version %"PRIu32
74e34e1f
NP
501 " and is not supported by this binary"
502 " (try upgrading GIT to a newer version)",
503 path, version);
504 }
505 } else
506 version = 1;
df1b059d 507
1f688557 508 nr = 0;
42873078 509 index = idx_map;
74e34e1f
NP
510 if (version > 1)
511 index += 2; /* skip index header */
1f688557 512 for (i = 0; i < 256; i++) {
326bf396 513 uint32_t n = ntohl(index[i]);
2d88451b
SP
514 if (n < nr) {
515 munmap(idx_map, idx_size);
df1b059d 516 return error("non-monotonic index %s", path);
2d88451b 517 }
1f688557
JH
518 nr = n;
519 }
520
74e34e1f
NP
521 if (version == 1) {
522 /*
523 * Total size:
524 * - 256 index entries 4 bytes each
525 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
526 * - 20-byte SHA1 of the packfile
527 * - 20-byte SHA1 file checksum
528 */
529 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
530 munmap(idx_map, idx_size);
eef427a0 531 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
532 }
533 } else if (version == 2) {
534 /*
535 * Minimum size:
536 * - 8 bytes of header
537 * - 256 index entries 4 bytes each
538 * - 20-byte sha1 entry * nr
539 * - 4-byte crc entry * nr
540 * - 4-byte offset entry * nr
541 * - 20-byte SHA1 of the packfile
542 * - 20-byte SHA1 file checksum
543 * And after the 4-byte offset table might be a
544 * variable sized table containing 8-byte entries
545 * for offsets larger than 2^31.
546 */
547 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
548 unsigned long max_size = min_size;
549 if (nr)
550 max_size += (nr - 1)*8;
551 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 552 munmap(idx_map, idx_size);
eef427a0 553 return error("wrong index v2 file size in %s", path);
74e34e1f 554 }
7109c889
JH
555 if (idx_size != min_size &&
556 /*
557 * make sure we can deal with large pack offsets.
558 * 31-bit signed offset won't be enough, neither
559 * 32-bit unsigned one will be.
560 */
561 (sizeof(off_t) <= 4)) {
562 munmap(idx_map, idx_size);
563 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 564 }
2d88451b 565 }
1f688557 566
74e34e1f 567 p->index_version = version;
42873078
NP
568 p->index_data = idx_map;
569 p->index_size = idx_size;
57059091 570 p->num_objects = nr;
1f688557
JH
571 return 0;
572}
573
bc8e478a 574int open_pack_index(struct packed_git *p)
d079837e
SP
575{
576 char *idx_name;
577 int ret;
578
579 if (p->index_data)
580 return 0;
581
582 idx_name = xstrdup(p->pack_name);
583 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
584 ret = check_packed_git_idx(idx_name, p);
585 free(idx_name);
586 return ret;
587}
588
11daf39b
SP
589static void scan_windows(struct packed_git *p,
590 struct packed_git **lru_p,
591 struct pack_window **lru_w,
592 struct pack_window **lru_l)
1f688557 593{
11daf39b
SP
594 struct pack_window *w, *w_l;
595
596 for (w_l = NULL, w = p->windows; w; w = w->next) {
597 if (!w->inuse_cnt) {
598 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
599 *lru_p = p;
600 *lru_w = w;
601 *lru_l = w_l;
54044bf8 602 }
54044bf8 603 }
11daf39b 604 w_l = w;
f9253394 605 }
11daf39b
SP
606}
607
d1efefa4 608static int unuse_one_window(struct packed_git *current, int keep_fd)
11daf39b
SP
609{
610 struct packed_git *p, *lru_p = NULL;
611 struct pack_window *lru_w = NULL, *lru_l = NULL;
612
613 if (current)
614 scan_windows(current, &lru_p, &lru_w, &lru_l);
615 for (p = packed_git; p; p = p->next)
616 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
617 if (lru_p) {
618 munmap(lru_w->base, lru_w->len);
619 pack_mapped -= lru_w->len;
620 if (lru_l)
621 lru_l->next = lru_w->next;
622 else {
623 lru_p->windows = lru_w->next;
d131b7af
SP
624 if (!lru_p->windows && lru_p->pack_fd != -1
625 && lru_p->pack_fd != keep_fd) {
54044bf8 626 close(lru_p->pack_fd);
c7934306 627 pack_open_fds--;
54044bf8
SP
628 lru_p->pack_fd = -1;
629 }
630 }
631 free(lru_w);
a53128b6 632 pack_open_windows--;
54044bf8
SP
633 return 1;
634 }
635 return 0;
f9253394
JH
636}
637
d1efefa4 638void release_pack_memory(size_t need, int fd)
97bfeb34
SP
639{
640 size_t cur = pack_mapped;
d1efefa4 641 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
97bfeb34
SP
642 ; /* nothing */
643}
644
58ecbd5e
JN
645void *xmmap(void *start, size_t length,
646 int prot, int flags, int fd, off_t offset)
647{
648 void *ret = mmap(start, length, prot, flags, fd, offset);
649 if (ret == MAP_FAILED) {
650 if (!length)
651 return NULL;
652 release_pack_memory(length, fd);
653 ret = mmap(start, length, prot, flags, fd, offset);
654 if (ret == MAP_FAILED)
655 die_errno("Out of memory? mmap failed");
656 }
657 return ret;
658}
659
c9ced051
SP
660void close_pack_windows(struct packed_git *p)
661{
662 while (p->windows) {
663 struct pack_window *w = p->windows;
664
665 if (w->inuse_cnt)
666 die("pack '%s' still has open windows to it",
667 p->pack_name);
668 munmap(w->base, w->len);
669 pack_mapped -= w->len;
670 pack_open_windows--;
671 p->windows = w->next;
672 free(w);
673 }
674}
675
03e79c88 676void unuse_pack(struct pack_window **w_cursor)
f9253394 677{
03e79c88
SP
678 struct pack_window *w = *w_cursor;
679 if (w) {
680 w->inuse_cnt--;
681 *w_cursor = NULL;
682 }
1f688557
JH
683}
684
fa5fc15d
SP
685void close_pack_index(struct packed_git *p)
686{
687 if (p->index_data) {
688 munmap((void *)p->index_data, p->index_size);
689 p->index_data = NULL;
690 }
691}
692
c74faea1
NP
693/*
694 * This is used by git-repack in case a newly created pack happens to
695 * contain the same set of objects as an existing one. In that case
696 * the resulting file might be different even if its name would be the
697 * same. It is best to close any reference to the old pack before it is
698 * replaced on disk. Of course no index pointers nor windows for given pack
699 * must subsist at this point. If ever objects from this pack are requested
700 * again, the new version of the pack will be reinitialized through
701 * reprepare_packed_git().
702 */
703void free_pack_by_name(const char *pack_name)
704{
705 struct packed_git *p, **pp = &packed_git;
706
707 while (*pp) {
708 p = *pp;
709 if (strcmp(pack_name, p->pack_name) == 0) {
fa3a0c94 710 clear_delta_base_cache();
c74faea1 711 close_pack_windows(p);
c7934306 712 if (p->pack_fd != -1) {
c74faea1 713 close(p->pack_fd);
c7934306
SP
714 pack_open_fds--;
715 }
fa5fc15d 716 close_pack_index(p);
c74faea1
NP
717 free(p->bad_object_sha1);
718 *pp = p->next;
719 free(p);
720 return;
721 }
722 pp = &p->next;
723 }
724}
725
3cf8b462
SP
726/*
727 * Do not call this directly as this leaks p->pack_fd on error return;
728 * call open_packed_git() instead.
729 */
730static int open_packed_git_1(struct packed_git *p)
1f688557 731{
9bc879c1
SP
732 struct stat st;
733 struct pack_header hdr;
734 unsigned char sha1[20];
735 unsigned char *idx_sha1;
2c039da8 736 long fd_flag;
9bc879c1 737
d079837e
SP
738 if (!p->index_data && open_pack_index(p))
739 return error("packfile %s index unavailable", p->pack_name);
740
c7934306
SP
741 if (!pack_max_fds) {
742 struct rlimit lim;
743 unsigned int max_fds;
744
745 if (getrlimit(RLIMIT_NOFILE, &lim))
746 die_errno("cannot get RLIMIT_NOFILE");
747
748 max_fds = lim.rlim_cur;
749
750 /* Save 3 for stdin/stdout/stderr, 22 for work */
751 if (25 < max_fds)
752 pack_max_fds = max_fds - 25;
753 else
754 pack_max_fds = 1;
755 }
756
757 while (pack_max_fds <= pack_open_fds && unuse_one_window(NULL, -1))
758 ; /* nothing */
759
f4e51683 760 p->pack_fd = git_open_noatime(p->pack_name);
9bc879c1 761 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 762 return -1;
c7934306 763 pack_open_fds++;
9bc879c1
SP
764
765 /* If we created the struct before we had the pack we lack size. */
bf592c50 766 if (!p->pack_size) {
bf592c50 767 if (!S_ISREG(st.st_mode))
072db278 768 return error("packfile %s not a regular file", p->pack_name);
bf592c50 769 p->pack_size = st.st_size;
9bc879c1 770 } else if (p->pack_size != st.st_size)
072db278 771 return error("packfile %s size changed", p->pack_name);
9bc879c1 772
2c039da8
JH
773 /* We leave these file descriptors open with sliding mmap;
774 * there is no point keeping them open across exec(), though.
775 */
776 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
777 if (fd_flag < 0)
072db278 778 return error("cannot determine file descriptor flags");
2c039da8
JH
779 fd_flag |= FD_CLOEXEC;
780 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 781 return error("cannot set FD_CLOEXEC");
2c039da8 782
9bc879c1 783 /* Verify we recognize this pack file format. */
e6e2bd62 784 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 785 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 786 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 787 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 788 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
789 return error("packfile %s is version %"PRIu32" and not"
790 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
791 p->pack_name, ntohl(hdr.hdr_version));
792
793 /* Verify the pack matches its index. */
57059091 794 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
795 return error("packfile %s claims to have %"PRIu32" objects"
796 " while index indicates %"PRIu32" objects",
57059091
NP
797 p->pack_name, ntohl(hdr.hdr_entries),
798 p->num_objects);
9bc879c1 799 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 800 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 801 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 802 return error("packfile %s signature is unavailable", p->pack_name);
42873078 803 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 804 if (hashcmp(sha1, idx_sha1))
072db278
SP
805 return error("packfile %s does not match index", p->pack_name);
806 return 0;
9bc879c1
SP
807}
808
3cf8b462
SP
809static int open_packed_git(struct packed_git *p)
810{
811 if (!open_packed_git_1(p))
812 return 0;
813 if (p->pack_fd != -1) {
814 close(p->pack_fd);
c7934306 815 pack_open_fds--;
3cf8b462
SP
816 p->pack_fd = -1;
817 }
818 return -1;
819}
820
c4001d92 821static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
822{
823 /* We must promise at least 20 bytes (one hash) after the
824 * offset is available from this window, otherwise the offset
825 * is not actually in this window and a different window (which
826 * has that one hash excess) must be used. This is to support
827 * the object header and delta base parsing routines below.
828 */
829 off_t win_off = win->offset;
830 return win_off <= offset
831 && (offset + 20) <= (win_off + win->len);
832}
833
4b25d091 834unsigned char *use_pack(struct packed_git *p,
03e79c88 835 struct pack_window **w_cursor,
c4001d92 836 off_t offset,
03e79c88 837 unsigned int *left)
9bc879c1 838{
60bb8b14 839 struct pack_window *win = *w_cursor;
03e79c88 840
a9d98a14 841 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
842 * pointless to ask for an offset into the middle of that
843 * hash, and the in_window function above wouldn't match
844 * don't allow an offset too close to the end of the file.
845 */
d131b7af
SP
846 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
847 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
848 if (offset > (p->pack_size - 20))
849 die("offset beyond end of packfile (truncated pack?)");
850
851 if (!win || !in_window(win, offset)) {
852 if (win)
853 win->inuse_cnt--;
854 for (win = p->windows; win; win = win->next) {
855 if (in_window(win, offset))
856 break;
857 }
858 if (!win) {
78a28df9 859 size_t window_align = packed_git_window_size / 2;
dc49cd76 860 off_t len;
d131b7af
SP
861
862 if (p->pack_fd == -1 && open_packed_git(p))
863 die("packfile %s cannot be accessed", p->pack_name);
864
60bb8b14 865 win = xcalloc(1, sizeof(*win));
78a28df9 866 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
867 len = p->pack_size - win->offset;
868 if (len > packed_git_window_size)
869 len = packed_git_window_size;
870 win->len = (size_t)len;
60bb8b14 871 pack_mapped += win->len;
11daf39b 872 while (packed_git_limit < pack_mapped
d1efefa4 873 && unuse_one_window(p, p->pack_fd))
60bb8b14 874 ; /* nothing */
c4712e45 875 win->base = xmmap(NULL, win->len,
60bb8b14
SP
876 PROT_READ, MAP_PRIVATE,
877 p->pack_fd, win->offset);
878 if (win->base == MAP_FAILED)
73b4e4be
SP
879 die("packfile %s cannot be mapped: %s",
880 p->pack_name,
881 strerror(errno));
d131b7af
SP
882 if (!win->offset && win->len == p->pack_size
883 && !p->do_not_close) {
884 close(p->pack_fd);
885 pack_open_fds--;
886 p->pack_fd = -1;
887 }
a53128b6
SP
888 pack_mmap_calls++;
889 pack_open_windows++;
890 if (pack_mapped > peak_pack_mapped)
891 peak_pack_mapped = pack_mapped;
892 if (pack_open_windows > peak_pack_open_windows)
893 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
894 win->next = p->windows;
895 p->windows = win;
896 }
1f688557 897 }
03e79c88
SP
898 if (win != *w_cursor) {
899 win->last_used = pack_used_ctr++;
900 win->inuse_cnt++;
901 *w_cursor = win;
902 }
60bb8b14 903 offset -= win->offset;
03e79c88 904 if (left)
dc49cd76 905 *left = win->len - xsize_t(offset);
03e79c88 906 return win->base + offset;
1f688557
JH
907}
908
27d69a46
NP
909static struct packed_git *alloc_packed_git(int extra)
910{
911 struct packed_git *p = xmalloc(sizeof(*p) + extra);
912 memset(p, 0, sizeof(*p));
913 p->pack_fd = -1;
914 return p;
915}
916
e0500293
JN
917static void try_to_free_pack_memory(size_t size)
918{
919 release_pack_memory(size, -1);
920}
921
42873078 922struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557 923{
e0500293 924 static int have_set_try_to_free_routine;
1f688557 925 struct stat st;
27d69a46 926 struct packed_git *p = alloc_packed_git(path_len + 2);
1f688557 927
e0500293
JN
928 if (!have_set_try_to_free_routine) {
929 have_set_try_to_free_routine = 1;
930 set_try_to_free_routine(try_to_free_pack_memory);
931 }
932
42873078
NP
933 /*
934 * Make sure a corresponding .pack file exists and that
935 * the index looks sane.
936 */
937 path_len -= strlen(".idx");
27d69a46
NP
938 if (path_len < 1) {
939 free(p);
1f688557 940 return NULL;
27d69a46 941 }
42873078 942 memcpy(p->pack_name, path, path_len);
8d25931d
BC
943
944 strcpy(p->pack_name + path_len, ".keep");
945 if (!access(p->pack_name, F_OK))
946 p->pack_keep = 1;
947
42873078 948 strcpy(p->pack_name + path_len, ".pack");
d079837e 949 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 950 free(p);
1f688557
JH
951 return NULL;
952 }
42873078 953
1f688557
JH
954 /* ok, it looks sane as far as we can check without
955 * actually mapping the pack file.
956 */
1f688557 957 p->pack_size = st.st_size;
9d835df2 958 p->pack_local = local;
b867092f 959 p->mtime = st.st_mtime;
42873078
NP
960 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
961 hashclr(p->sha1);
1f688557
JH
962 return p;
963}
964
7b64469a 965struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 966{
42873078 967 const char *path = sha1_pack_name(sha1);
27d69a46 968 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
bf592c50 969
27d69a46
NP
970 strcpy(p->pack_name, path);
971 hashcpy(p->sha1, sha1);
42873078
NP
972 if (check_packed_git_idx(idx_path, p)) {
973 free(p);
bf592c50 974 return NULL;
42873078 975 }
bf592c50 976
bf592c50
DB
977 return p;
978}
979
980void install_packed_git(struct packed_git *pack)
981{
c7934306
SP
982 if (pack->pack_fd != -1)
983 pack_open_fds++;
984
bf592c50
DB
985 pack->next = packed_git;
986 packed_git = pack;
987}
988
9d835df2 989static void prepare_packed_git_one(char *objdir, int local)
1f688557 990{
9cb18f56
JM
991 /* Ensure that this buffer is large enough so that we can
992 append "/pack/" without clobbering the stack even if
993 strlen(objdir) were PATH_MAX. */
994 char path[PATH_MAX + 1 + 4 + 1 + 1];
1f688557
JH
995 int len;
996 DIR *dir;
997 struct dirent *de;
998
999 sprintf(path, "%s/pack", objdir);
1000 len = strlen(path);
1001 dir = opendir(path);
b5b16990 1002 if (!dir) {
26125f6b 1003 if (errno != ENOENT)
bd2afde8 1004 error("unable to open object pack directory: %s: %s",
26125f6b 1005 path, strerror(errno));
1f688557 1006 return;
b5b16990 1007 }
1f688557
JH
1008 path[len++] = '/';
1009 while ((de = readdir(dir)) != NULL) {
1010 int namelen = strlen(de->d_name);
1011 struct packed_git *p;
1012
5bb1cda5 1013 if (!has_extension(de->d_name, ".idx"))
1f688557
JH
1014 continue;
1015
9cb18f56
JM
1016 if (len + namelen + 1 > sizeof(path))
1017 continue;
1018
54a15a8d 1019 /* Don't reopen a pack we already have. */
1f688557 1020 strcpy(path + len, de->d_name);
86f7780c
JK
1021 for (p = packed_git; p; p = p->next) {
1022 if (!memcmp(path, p->pack_name, len + namelen - 4))
1023 break;
1024 }
1025 if (p)
1026 continue;
54a15a8d
SP
1027 /* See if it really is a valid .idx file with corresponding
1028 * .pack file that we can map.
1029 */
9d835df2 1030 p = add_packed_git(path, len + namelen, local);
1f688557
JH
1031 if (!p)
1032 continue;
625e9421 1033 install_packed_git(p);
1f688557 1034 }
5b35bcd5 1035 closedir(dir);
1f688557
JH
1036}
1037
b867092f
JH
1038static int sort_pack(const void *a_, const void *b_)
1039{
1040 struct packed_git *a = *((struct packed_git **)a_);
1041 struct packed_git *b = *((struct packed_git **)b_);
1042 int st;
1043
1044 /*
1045 * Local packs tend to contain objects specific to our
1046 * variant of the project than remote ones. In addition,
1047 * remote ones could be on a network mounted filesystem.
1048 * Favor local ones for these reasons.
1049 */
1050 st = a->pack_local - b->pack_local;
1051 if (st)
1052 return -st;
1053
1054 /*
1055 * Younger packs tend to contain more recent objects,
1056 * and more recent objects tend to get accessed more
1057 * often.
1058 */
1059 if (a->mtime < b->mtime)
1060 return 1;
1061 else if (a->mtime == b->mtime)
1062 return 0;
1063 return -1;
1064}
1065
1066static void rearrange_packed_git(void)
1067{
1068 struct packed_git **ary, *p;
1069 int i, n;
1070
1071 for (n = 0, p = packed_git; p; p = p->next)
1072 n++;
1073 if (n < 2)
1074 return;
1075
1076 /* prepare an array of packed_git for easier sorting */
1077 ary = xcalloc(n, sizeof(struct packed_git *));
1078 for (n = 0, p = packed_git; p; p = p->next)
1079 ary[n++] = p;
1080
1081 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1082
1083 /* link them back again */
1084 for (i = 0; i < n - 1; i++)
1085 ary[i]->next = ary[i + 1];
1086 ary[n - 1]->next = NULL;
1087 packed_git = ary[0];
1088
1089 free(ary);
1090}
1091
637cdd9d 1092static int prepare_packed_git_run_once = 0;
9a217f2a 1093void prepare_packed_git(void)
1f688557 1094{
d5a63b99 1095 struct alternate_object_database *alt;
1f688557 1096
637cdd9d 1097 if (prepare_packed_git_run_once)
1f688557 1098 return;
9d835df2 1099 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 1100 prepare_alt_odb();
d5a63b99 1101 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 1102 alt->name[-1] = 0;
9d835df2 1103 prepare_packed_git_one(alt->base, 0);
1494e038 1104 alt->name[-1] = '/';
1f688557 1105 }
b867092f 1106 rearrange_packed_git();
637cdd9d
JK
1107 prepare_packed_git_run_once = 1;
1108}
1109
fc04c412 1110void reprepare_packed_git(void)
637cdd9d 1111{
4b480c67 1112 discard_revindex();
637cdd9d
JK
1113 prepare_packed_git_run_once = 0;
1114 prepare_packed_git();
1f688557
JH
1115}
1116
8eca0b47
NP
1117static void mark_bad_packed_object(struct packed_git *p,
1118 const unsigned char *sha1)
1119{
1120 unsigned i;
1121 for (i = 0; i < p->num_bad_objects; i++)
1122 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1123 return;
1124 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1125 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1126 p->num_bad_objects++;
1127}
1128
b6c4cecc 1129static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
ac939109
NP
1130{
1131 struct packed_git *p;
1132 unsigned i;
1133
1134 for (p = packed_git; p; p = p->next)
1135 for (i = 0; i < p->num_bad_objects; i++)
1136 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
b6c4cecc
JH
1137 return p;
1138 return NULL;
ac939109
NP
1139}
1140
5d6ccf5c 1141int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
0fcfd160
LT
1142{
1143 unsigned char real_sha1[20];
7cfb5f36 1144 hash_sha1_file(map, size, type, real_sha1);
a89fccd2 1145 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1146}
1147
f4e51683 1148static int git_open_noatime(const char *name)
44d1c19e
LT
1149{
1150 static int sha1_file_open_flag = O_NOATIME;
44d1c19e 1151
f2e872aa
SP
1152 for (;;) {
1153 int fd = open(name, O_RDONLY | sha1_file_open_flag);
44d1c19e 1154 if (fd >= 0)
f2e872aa
SP
1155 return fd;
1156
f2e872aa
SP
1157 /* Might the failure be due to O_NOATIME? */
1158 if (errno != ENOENT && sha1_file_open_flag) {
44d1c19e 1159 sha1_file_open_flag = 0;
f2e872aa
SP
1160 continue;
1161 }
1162
1163 return -1;
44d1c19e 1164 }
44d1c19e
LT
1165}
1166
1167static int open_sha1_file(const unsigned char *sha1)
1168{
1169 int fd;
1170 char *name = sha1_file_name(sha1);
1171 struct alternate_object_database *alt;
1172
f4e51683 1173 fd = git_open_noatime(name);
44d1c19e
LT
1174 if (fd >= 0)
1175 return fd;
1176
1177 prepare_alt_odb();
1178 errno = ENOENT;
1179 for (alt = alt_odb_list; alt; alt = alt->next) {
1180 name = alt->name;
1181 fill_sha1_path(name, sha1);
f4e51683 1182 fd = git_open_noatime(alt->base);
44d1c19e
LT
1183 if (fd >= 0)
1184 return fd;
1185 }
1186 return -1;
1187}
1188
4175e9e3 1189static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1190{
0fcfd160 1191 void *map;
144bde78 1192 int fd;
ace1534d 1193
44d1c19e
LT
1194 fd = open_sha1_file(sha1);
1195 map = NULL;
1196 if (fd >= 0) {
1197 struct stat st;
0fcfd160 1198
44d1c19e
LT
1199 if (!fstat(fd, &st)) {
1200 *size = xsize_t(st.st_size);
1201 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1202 }
44d1c19e 1203 close(fd);
0fcfd160 1204 }
0fcfd160
LT
1205 return map;
1206}
1207
726f852b 1208static int legacy_loose_object(unsigned char *map)
bb6b8e4f
JH
1209{
1210 unsigned int word;
1211
1212 /*
1213 * Is it a zlib-compressed buffer? If so, the first byte
1214 * must be 0x78 (15-bit window size, deflated), and the
1215 * first 16-bit word is evenly divisible by 31
1216 */
1217 word = (map[0] << 8) + map[1];
1218 if (map[0] == 0x78 && !(word % 31))
1219 return 1;
1220 else
1221 return 0;
1222}
1223
09ded04b
NP
1224unsigned long unpack_object_header_buffer(const unsigned char *buf,
1225 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1226{
ad1ed5ee 1227 unsigned shift;
48fb7deb 1228 unsigned long size, c;
ad1ed5ee
JH
1229 unsigned long used = 0;
1230
1231 c = buf[used++];
1232 *type = (c >> 4) & 7;
1233 size = c & 15;
1234 shift = 4;
1235 while (c & 0x80) {
f630cfda 1236 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1237 error("bad object header");
ad1ed5ee 1238 return 0;
09ded04b 1239 }
ad1ed5ee
JH
1240 c = buf[used++];
1241 size += (c & 0x7f) << shift;
1242 shift += 7;
1243 }
1244 *sizep = size;
1245 return used;
1246}
1247
1248static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1249{
1250 unsigned long size, used;
1251 static const char valid_loose_object_type[8] = {
1252 0, /* OBJ_EXT */
1253 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1254 0, /* "delta" and others are invalid in a loose object */
93821bd9 1255 };
ad1ed5ee 1256 enum object_type type;
93821bd9 1257
c4483576
LT
1258 /* Get the data stream */
1259 memset(stream, 0, sizeof(*stream));
1260 stream->next_in = map;
1261 stream->avail_in = mapsize;
1262 stream->next_out = buffer;
93821bd9
LT
1263 stream->avail_out = bufsiz;
1264
bb6b8e4f 1265 if (legacy_loose_object(map)) {
39c68542
LT
1266 git_inflate_init(stream);
1267 return git_inflate(stream, 0);
93821bd9
LT
1268 }
1269
726f852b
NP
1270
1271 /*
1272 * There used to be a second loose object header format which
1273 * was meant to mimic the in-pack format, allowing for direct
1274 * copy of the object data. This format turned up not to be
1275 * really worth it and we don't write it any longer. But we
1276 * can still read it.
1277 */
09ded04b 1278 used = unpack_object_header_buffer(map, mapsize, &type, &size);
ad1ed5ee 1279 if (!used || !valid_loose_object_type[type])
93821bd9 1280 return -1;
ad1ed5ee
JH
1281 map += used;
1282 mapsize -= used;
93821bd9
LT
1283
1284 /* Set up the stream for the rest.. */
1285 stream->next_in = map;
1286 stream->avail_in = mapsize;
39c68542 1287 git_inflate_init(stream);
93821bd9
LT
1288
1289 /* And generate the fake traditional header */
ad1ed5ee 1290 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
df843662 1291 typename(type), size);
93821bd9 1292 return 0;
c4483576
LT
1293}
1294
7efbff75 1295static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1296{
1297 int bytes = strlen(buffer) + 1;
3aee68aa 1298 unsigned char *buf = xmallocz(size);
93821bd9 1299 unsigned long n;
7efbff75 1300 int status = Z_OK;
5180cacc 1301
93821bd9
LT
1302 n = stream->total_out - bytes;
1303 if (n > size)
1304 n = size;
1305 memcpy(buf, (char *) buffer + bytes, n);
1306 bytes = n;
456cdf6e
LT
1307 if (bytes <= size) {
1308 /*
1309 * The above condition must be (bytes <= size), not
1310 * (bytes < size). In other words, even though we
ccf5ace0 1311 * expect no more output and set avail_out to zero,
456cdf6e
LT
1312 * the input zlib stream may have bytes that express
1313 * "this concludes the stream", and we *do* want to
1314 * eat that input.
1315 *
1316 * Otherwise we would not be able to test that we
1317 * consumed all the input to reach the expected size;
1318 * we also want to check that zlib tells us that all
1319 * went well with status == Z_STREAM_END at the end.
1320 */
5180cacc
LT
1321 stream->next_out = buf + bytes;
1322 stream->avail_out = size - bytes;
7efbff75 1323 while (status == Z_OK)
39c68542 1324 status = git_inflate(stream, Z_FINISH);
5180cacc 1325 }
456cdf6e 1326 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1327 git_inflate_end(stream);
7efbff75
JH
1328 return buf;
1329 }
1330
1331 if (status < 0)
1332 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1333 else if (stream->avail_in)
1334 error("garbage at end of loose object '%s'",
1335 sha1_to_hex(sha1));
1336 free(buf);
1337 return NULL;
5180cacc
LT
1338}
1339
1340/*
1341 * We used to just use "sscanf()", but that's actually way
1342 * too permissive for what we want to check. So do an anal
1343 * object header parse by hand.
1344 */
21666f1a 1345static int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1346{
21666f1a 1347 char type[10];
5180cacc
LT
1348 int i;
1349 unsigned long size;
1350
1351 /*
a6080a0a 1352 * The type can be at most ten bytes (including the
5180cacc 1353 * terminating '\0' that we add), and is followed by
21666f1a 1354 * a space.
5180cacc 1355 */
21666f1a 1356 i = 0;
5180cacc
LT
1357 for (;;) {
1358 char c = *hdr++;
1359 if (c == ' ')
1360 break;
21666f1a
NP
1361 type[i++] = c;
1362 if (i >= sizeof(type))
5180cacc 1363 return -1;
5180cacc 1364 }
21666f1a 1365 type[i] = 0;
5180cacc
LT
1366
1367 /*
1368 * The length must follow immediately, and be in canonical
1369 * decimal format (ie "010" is not valid).
1370 */
1371 size = *hdr++ - '0';
1372 if (size > 9)
1373 return -1;
1374 if (size) {
1375 for (;;) {
1376 unsigned long c = *hdr - '0';
1377 if (c > 9)
1378 break;
1379 hdr++;
1380 size = size * 10 + c;
1381 }
1382 }
1383 *sizep = size;
1384
1385 /*
1386 * The length must be followed by a zero byte
1387 */
21666f1a 1388 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1389}
1390
7efbff75 1391static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1392{
5180cacc 1393 int ret;
0fcfd160 1394 z_stream stream;
5180cacc 1395 char hdr[8192];
0fcfd160 1396
5180cacc 1397 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1398 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1399 return NULL;
1400
7efbff75 1401 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1402}
1403
54dab52a
NP
1404unsigned long get_size_from_delta(struct packed_git *p,
1405 struct pack_window **w_curs,
1406 off_t curpos)
1407{
1408 const unsigned char *data;
1409 unsigned char delta_head[20], *in;
1410 z_stream stream;
1411 int st;
1412
1413 memset(&stream, 0, sizeof(stream));
1414 stream.next_out = delta_head;
1415 stream.avail_out = sizeof(delta_head);
1416
39c68542 1417 git_inflate_init(&stream);
54dab52a
NP
1418 do {
1419 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1420 stream.next_in = in;
39c68542 1421 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1422 curpos += stream.next_in - in;
1423 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1424 stream.total_out < sizeof(delta_head));
39c68542 1425 git_inflate_end(&stream);
3d77d877
NP
1426 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1427 error("delta data unpack-initial failed");
1428 return 0;
1429 }
54dab52a
NP
1430
1431 /* Examine the initial part of the delta to figure out
1432 * the result size.
1433 */
1434 data = delta_head;
1435
1436 /* ignore base size */
1437 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1438
1439 /* Read the result size */
1440 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1441}
1442
c4001d92 1443static off_t get_delta_base(struct packed_git *p,
03e79c88 1444 struct pack_window **w_curs,
c4001d92 1445 off_t *curpos,
21666f1a 1446 enum object_type type,
c4001d92 1447 off_t delta_obj_offset)
eb32d236 1448{
2b87c45b 1449 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1450 off_t base_offset;
eb32d236 1451
8d8a4ea5
SP
1452 /* use_pack() assured us we have [base_info, base_info + 20)
1453 * as a range that we can look at without walking off the
1454 * end of the mapped window. Its actually the hash size
1455 * that is assured. An OFS_DELTA longer than the hash size
1456 * is stupid, as then a REF_DELTA would be smaller to store.
1457 */
21666f1a 1458 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1459 unsigned used = 0;
1460 unsigned char c = base_info[used++];
1461 base_offset = c & 127;
1462 while (c & 128) {
1463 base_offset += 1;
8723f216 1464 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1465 return 0; /* overflow */
eb32d236
NP
1466 c = base_info[used++];
1467 base_offset = (base_offset << 7) + (c & 127);
1468 }
1469 base_offset = delta_obj_offset - base_offset;
d8f32556 1470 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1471 return 0; /* out of bound */
2b87c45b 1472 *curpos += used;
21666f1a 1473 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1474 /* The base entry _must_ be in the same pack */
1475 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1476 *curpos += 20;
eb32d236
NP
1477 } else
1478 die("I am totally screwed");
2b87c45b 1479 return base_offset;
eb32d236
NP
1480}
1481
f3bf9224 1482/* forward declaration for a mutually recursive function */
c4001d92 1483static int packed_object_info(struct packed_git *p, off_t offset,
21666f1a 1484 unsigned long *sizep);
f3bf9224 1485
43057304 1486static int packed_delta_info(struct packed_git *p,
03e79c88 1487 struct pack_window **w_curs,
c4001d92 1488 off_t curpos,
21666f1a 1489 enum object_type type,
c4001d92 1490 off_t obj_offset,
43057304 1491 unsigned long *sizep)
5db47c2b 1492{
c4001d92 1493 off_t base_offset;
f3bf9224 1494
21666f1a 1495 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
3d77d877
NP
1496 if (!base_offset)
1497 return OBJ_BAD;
21666f1a 1498 type = packed_object_info(p, base_offset, NULL);
3d77d877 1499 if (type <= OBJ_NONE) {
08698b1e
NP
1500 struct revindex_entry *revidx;
1501 const unsigned char *base_sha1;
1502 revidx = find_pack_revindex(p, base_offset);
1503 if (!revidx)
1504 return OBJ_BAD;
1505 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
3d77d877
NP
1506 mark_bad_packed_object(p, base_sha1);
1507 type = sha1_object_info(base_sha1, NULL);
1508 if (type <= OBJ_NONE)
1509 return OBJ_BAD;
1510 }
f3bf9224 1511
c62266f3
JH
1512 /* We choose to only get the type of the base object and
1513 * ignore potentially corrupt pack file that expects the delta
1514 * based on a base with a wrong size. This saves tons of
1515 * inflate() calls.
1516 */
3d77d877 1517 if (sizep) {
54dab52a 1518 *sizep = get_size_from_delta(p, w_curs, curpos);
3d77d877
NP
1519 if (*sizep == 0)
1520 type = OBJ_BAD;
1521 }
21666f1a
NP
1522
1523 return type;
5db47c2b
JH
1524}
1525
2b87c45b
NP
1526static int unpack_object_header(struct packed_git *p,
1527 struct pack_window **w_curs,
c4001d92 1528 off_t *curpos,
2b87c45b 1529 unsigned long *sizep)
a733cb60 1530{
03e79c88
SP
1531 unsigned char *base;
1532 unsigned int left;
ad1ed5ee 1533 unsigned long used;
2b87c45b 1534 enum object_type type;
a733cb60 1535
8d8a4ea5 1536 /* use_pack() assures us we have [base, base + 20) available
0353a0c4 1537 * as a range that we can look at. (Its actually the hash
3dff5379 1538 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1539 * the maximum deflated object size is 2^137, which is just
1540 * insane, so we know won't exceed what we have been given.
1541 */
2b87c45b 1542 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1543 used = unpack_object_header_buffer(base, left, &type, sizep);
1544 if (!used) {
1545 type = OBJ_BAD;
1546 } else
1547 *curpos += used;
ad1ed5ee 1548
2b87c45b 1549 return type;
a733cb60
LT
1550}
1551
21666f1a 1552const char *packed_object_info_detail(struct packed_git *p,
c4001d92 1553 off_t obj_offset,
21666f1a
NP
1554 unsigned long *size,
1555 unsigned long *store_size,
1556 unsigned int *delta_chain_length,
1557 unsigned char *base_sha1)
ad8c80a5 1558{
03e79c88 1559 struct pack_window *w_curs = NULL;
c4001d92
SP
1560 off_t curpos;
1561 unsigned long dummy;
43057304 1562 unsigned char *next_sha1;
21666f1a 1563 enum object_type type;
70f5d5d3 1564 struct revindex_entry *revidx;
ad8c80a5 1565
43057304 1566 *delta_chain_length = 0;
2b87c45b 1567 curpos = obj_offset;
21666f1a 1568 type = unpack_object_header(p, &w_curs, &curpos, size);
43057304 1569
70f5d5d3
NP
1570 revidx = find_pack_revindex(p, obj_offset);
1571 *store_size = revidx[1].offset - obj_offset;
1572
43057304 1573 for (;;) {
21666f1a 1574 switch (type) {
43057304 1575 default:
08a19d87 1576 die("pack %s contains unknown object type %d",
21666f1a 1577 p->pack_name, type);
43057304
NP
1578 case OBJ_COMMIT:
1579 case OBJ_TREE:
1580 case OBJ_BLOB:
1581 case OBJ_TAG:
03e79c88 1582 unuse_pack(&w_curs);
21666f1a 1583 return typename(type);
eb32d236 1584 case OBJ_OFS_DELTA:
21666f1a 1585 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
8eca0b47
NP
1586 if (!obj_offset)
1587 die("pack %s contains bad delta base reference of type %s",
1588 p->pack_name, typename(type));
eb32d236 1589 if (*delta_chain_length == 0) {
70f5d5d3
NP
1590 revidx = find_pack_revindex(p, obj_offset);
1591 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
eb32d236
NP
1592 }
1593 break;
1594 case OBJ_REF_DELTA:
2b87c45b 1595 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
43057304
NP
1596 if (*delta_chain_length == 0)
1597 hashcpy(base_sha1, next_sha1);
2b87c45b 1598 obj_offset = find_pack_entry_one(next_sha1, p);
43057304
NP
1599 break;
1600 }
43057304 1601 (*delta_chain_length)++;
2b87c45b 1602 curpos = obj_offset;
21666f1a 1603 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
ad8c80a5 1604 }
ad8c80a5
JH
1605}
1606
c4001d92 1607static int packed_object_info(struct packed_git *p, off_t obj_offset,
21666f1a 1608 unsigned long *sizep)
1f688557 1609{
03e79c88 1610 struct pack_window *w_curs = NULL;
c4001d92
SP
1611 unsigned long size;
1612 off_t curpos = obj_offset;
21666f1a 1613 enum object_type type;
5db47c2b 1614
21666f1a 1615 type = unpack_object_header(p, &w_curs, &curpos, &size);
5db47c2b 1616
21666f1a 1617 switch (type) {
eb32d236
NP
1618 case OBJ_OFS_DELTA:
1619 case OBJ_REF_DELTA:
21666f1a
NP
1620 type = packed_delta_info(p, &w_curs, curpos,
1621 type, obj_offset, sizep);
2b87c45b 1622 break;
a733cb60 1623 case OBJ_COMMIT:
a733cb60 1624 case OBJ_TREE:
a733cb60 1625 case OBJ_BLOB:
a733cb60 1626 case OBJ_TAG:
2b87c45b
NP
1627 if (sizep)
1628 *sizep = size;
a69d0943 1629 break;
1f688557 1630 default:
3d77d877
NP
1631 error("unknown object type %i at offset %"PRIuMAX" in %s",
1632 type, (uintmax_t)obj_offset, p->pack_name);
1633 type = OBJ_BAD;
1f688557 1634 }
2b87c45b 1635 unuse_pack(&w_curs);
21666f1a 1636 return type;
1f688557
JH
1637}
1638
eb950c19 1639static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1640 struct pack_window **w_curs,
c4001d92 1641 off_t curpos,
eb950c19 1642 unsigned long size)
de530aaa
SP
1643{
1644 int st;
1645 z_stream stream;
079afb18 1646 unsigned char *buffer, *in;
de530aaa 1647
4ab07e4d 1648 buffer = xmallocz(size);
de530aaa 1649 memset(&stream, 0, sizeof(stream));
de530aaa 1650 stream.next_out = buffer;
39eea7bd 1651 stream.avail_out = size + 1;
de530aaa 1652
39c68542 1653 git_inflate_init(&stream);
079afb18 1654 do {
2b87c45b 1655 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 1656 stream.next_in = in;
39c68542 1657 st = git_inflate(&stream, Z_FINISH);
39eea7bd
JH
1658 if (!stream.avail_out)
1659 break; /* the payload is larger than it should be */
2b87c45b 1660 curpos += stream.next_in - in;
079afb18 1661 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 1662 git_inflate_end(&stream);
de530aaa
SP
1663 if ((st != Z_STREAM_END) || stream.total_out != size) {
1664 free(buffer);
1665 return NULL;
1666 }
1667
1668 return buffer;
1669}
1670
e5e01619
LT
1671#define MAX_DELTA_CACHE (256)
1672
18bdec11 1673static size_t delta_base_cached;
5e08ecbf
NP
1674
1675static struct delta_base_cache_lru_list {
1676 struct delta_base_cache_lru_list *prev;
1677 struct delta_base_cache_lru_list *next;
1678} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1679
e5e01619 1680static struct delta_base_cache_entry {
5e08ecbf
NP
1681 struct delta_base_cache_lru_list lru;
1682 void *data;
e5e01619
LT
1683 struct packed_git *p;
1684 off_t base_offset;
1685 unsigned long size;
e5e01619
LT
1686 enum object_type type;
1687} delta_base_cache[MAX_DELTA_CACHE];
1688
1689static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1690{
1691 unsigned long hash;
1692
1693 hash = (unsigned long)p + (unsigned long)base_offset;
1694 hash += (hash >> 8) + (hash >> 16);
3358004a 1695 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1696}
1697
62f255ad 1698static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 1699 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 1700{
e5e01619
LT
1701 void *ret;
1702 unsigned long hash = pack_entry_hash(p, base_offset);
1703 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1704
1705 ret = ent->data;
749bc58c
MV
1706 if (!ret || ent->p != p || ent->base_offset != base_offset)
1707 return unpack_entry(p, base_offset, type, base_size);
e5e01619 1708
18bdec11 1709 if (!keep_cache) {
a0cba108 1710 ent->data = NULL;
5e08ecbf
NP
1711 ent->lru.next->prev = ent->lru.prev;
1712 ent->lru.prev->next = ent->lru.next;
18bdec11 1713 delta_base_cached -= ent->size;
182af834
PH
1714 } else {
1715 ret = xmemdupz(ent->data, ent->size);
a0cba108 1716 }
e5e01619
LT
1717 *type = ent->type;
1718 *base_size = ent->size;
1719 return ret;
62f255ad
LT
1720}
1721
18bdec11
SP
1722static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1723{
1724 if (ent->data) {
1725 free(ent->data);
1726 ent->data = NULL;
5e08ecbf
NP
1727 ent->lru.next->prev = ent->lru.prev;
1728 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
1729 delta_base_cached -= ent->size;
1730 }
1731}
1732
3d20c636
SP
1733void clear_delta_base_cache(void)
1734{
1735 unsigned long p;
1736 for (p = 0; p < MAX_DELTA_CACHE; p++)
1737 release_delta_base_cache(&delta_base_cache[p]);
1738}
1739
62f255ad
LT
1740static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1741 void *base, unsigned long base_size, enum object_type type)
1742{
5e08ecbf 1743 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 1744 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 1745 struct delta_base_cache_lru_list *lru;
e5e01619 1746
18bdec11
SP
1747 release_delta_base_cache(ent);
1748 delta_base_cached += base_size;
5e08ecbf
NP
1749
1750 for (lru = delta_base_cache_lru.next;
1751 delta_base_cached > delta_base_cache_limit
1752 && lru != &delta_base_cache_lru;
1753 lru = lru->next) {
1754 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
1755 if (f->type == OBJ_BLOB)
1756 release_delta_base_cache(f);
1757 }
5e08ecbf
NP
1758 for (lru = delta_base_cache_lru.next;
1759 delta_base_cached > delta_base_cache_limit
1760 && lru != &delta_base_cache_lru;
1761 lru = lru->next) {
1762 struct delta_base_cache_entry *f = (void *)lru;
1763 release_delta_base_cache(f);
1764 }
18bdec11 1765
e5e01619
LT
1766 ent->p = p;
1767 ent->base_offset = base_offset;
1768 ent->type = type;
1769 ent->data = base;
1770 ent->size = base_size;
5e08ecbf
NP
1771 ent->lru.next = &delta_base_cache_lru;
1772 ent->lru.prev = delta_base_cache_lru.prev;
1773 delta_base_cache_lru.prev->next = &ent->lru;
1774 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
1775}
1776
c2c5b270
CC
1777static void *read_object(const unsigned char *sha1, enum object_type *type,
1778 unsigned long *size);
1779
eb950c19 1780static void *unpack_delta_entry(struct packed_git *p,
03e79c88 1781 struct pack_window **w_curs,
c4001d92 1782 off_t curpos,
1f688557 1783 unsigned long delta_size,
c4001d92 1784 off_t obj_offset,
21666f1a 1785 enum object_type *type,
eb950c19 1786 unsigned long *sizep)
1f688557 1787{
7c3e8be3 1788 void *delta_data, *result, *base;
c4001d92
SP
1789 unsigned long base_size;
1790 off_t base_offset;
43057304 1791
21666f1a 1792 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
8eca0b47
NP
1793 if (!base_offset) {
1794 error("failed to validate delta base reference "
1795 "at offset %"PRIuMAX" from %s",
1796 (uintmax_t)curpos, p->pack_name);
1797 return NULL;
1798 }
eac12e2d 1799 unuse_pack(w_curs);
a0cba108 1800 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
8eca0b47
NP
1801 if (!base) {
1802 /*
1803 * We're probably in deep shit, but let's try to fetch
1804 * the required base anyway from another pack or loose.
1805 * This is costly but should happen only in the presence
1806 * of a corrupted pack, and is better than failing outright.
1807 */
08698b1e
NP
1808 struct revindex_entry *revidx;
1809 const unsigned char *base_sha1;
1810 revidx = find_pack_revindex(p, base_offset);
1811 if (!revidx)
1812 return NULL;
1813 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
8eca0b47
NP
1814 error("failed to read delta base object %s"
1815 " at offset %"PRIuMAX" from %s",
1816 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1817 p->pack_name);
1818 mark_bad_packed_object(p, base_sha1);
ac939109 1819 base = read_object(base_sha1, type, &base_size);
8eca0b47
NP
1820 if (!base)
1821 return NULL;
1822 }
67686d95 1823
2b87c45b 1824 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
8eca0b47
NP
1825 if (!delta_data) {
1826 error("failed to unpack compressed delta "
1827 "at offset %"PRIuMAX" from %s",
1828 (uintmax_t)curpos, p->pack_name);
1829 free(base);
1830 return NULL;
1831 }
1f688557
JH
1832 result = patch_delta(base, base_size,
1833 delta_data, delta_size,
21666f1a 1834 sizep);
1f688557
JH
1835 if (!result)
1836 die("failed to apply delta");
1837 free(delta_data);
62f255ad 1838 add_delta_base_cache(p, base_offset, base, base_size, *type);
1f688557
JH
1839 return result;
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
0e8189e2
NP
1851 if (do_check_packed_object_crc && p->index_version > 1) {
1852 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1853 unsigned long len = revidx[1].offset - obj_offset;
1854 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1855 const unsigned char *sha1 =
1856 nth_packed_object_sha1(p, revidx->nr);
1857 error("bad packed object CRC for %s",
1858 sha1_to_hex(sha1));
1859 mark_bad_packed_object(p, sha1);
f755bb99 1860 unuse_pack(&w_curs);
0e8189e2
NP
1861 return NULL;
1862 }
1863 }
1864
21666f1a
NP
1865 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1866 switch (*type) {
eb32d236
NP
1867 case OBJ_OFS_DELTA:
1868 case OBJ_REF_DELTA:
21666f1a
NP
1869 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1870 obj_offset, type, sizep);
4d703a1a 1871 break;
a733cb60 1872 case OBJ_COMMIT:
a733cb60 1873 case OBJ_TREE:
a733cb60 1874 case OBJ_BLOB:
a733cb60 1875 case OBJ_TAG:
21666f1a 1876 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
4d703a1a 1877 break;
1f688557 1878 default:
8eca0b47
NP
1879 data = NULL;
1880 error("unknown object type %i at offset %"PRIuMAX" in %s",
1881 *type, (uintmax_t)obj_offset, p->pack_name);
1f688557 1882 }
03e79c88 1883 unuse_pack(&w_curs);
21666f1a 1884 return data;
1f688557
JH
1885}
1886
d079837e 1887const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 1888 uint32_t n)
9a217f2a 1889{
42873078 1890 const unsigned char *index = p->index_data;
d079837e
SP
1891 if (!index) {
1892 if (open_pack_index(p))
1893 return NULL;
1894 index = p->index_data;
1895 }
57059091 1896 if (n >= p->num_objects)
d72308e0 1897 return NULL;
74e34e1f
NP
1898 index += 4 * 256;
1899 if (p->index_version == 1) {
1900 return index + 24 * n + 4;
1901 } else {
1902 index += 8;
1903 return index + 20 * n;
1904 }
1905}
1906
99093238 1907off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
1908{
1909 const unsigned char *index = p->index_data;
1910 index += 4 * 256;
1911 if (p->index_version == 1) {
1912 return ntohl(*((uint32_t *)(index + 24 * n)));
1913 } else {
1914 uint32_t off;
1915 index += 8 + p->num_objects * (20 + 4);
1916 off = ntohl(*((uint32_t *)(index + 4 * n)));
1917 if (!(off & 0x80000000))
1918 return off;
1919 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1920 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1921 ntohl(*((uint32_t *)(index + 4)));
1922 }
9a217f2a
JH
1923}
1924
c4001d92 1925off_t find_pack_entry_one(const unsigned char *sha1,
43057304 1926 struct packed_git *p)
1f688557 1927{
42873078 1928 const uint32_t *level1_ofs = p->index_data;
42873078 1929 const unsigned char *index = p->index_data;
628522ec
JH
1930 unsigned hi, lo, stride;
1931 static int use_lookup = -1;
1932 static int debug_lookup = -1;
1933
1934 if (debug_lookup < 0)
1935 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 1936
d079837e
SP
1937 if (!index) {
1938 if (open_pack_index(p))
1939 return 0;
1940 level1_ofs = p->index_data;
1941 index = p->index_data;
1942 }
74e34e1f
NP
1943 if (p->index_version > 1) {
1944 level1_ofs += 2;
1945 index += 8;
1946 }
42873078 1947 index += 4 * 256;
74e34e1f
NP
1948 hi = ntohl(level1_ofs[*sha1]);
1949 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
1950 if (p->index_version > 1) {
1951 stride = 20;
1952 } else {
1953 stride = 24;
1954 index += 4;
1955 }
1956
1957 if (debug_lookup)
6e1c2344 1958 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
1959 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1960
1961 if (use_lookup < 0)
1962 use_lookup = !!getenv("GIT_USE_LOOKUP");
1963 if (use_lookup) {
1964 int pos = sha1_entry_pos(index, stride, 0,
1965 lo, hi, p->num_objects, sha1);
1966 if (pos < 0)
1967 return 0;
1968 return nth_packed_object_offset(p, pos);
1969 }
1f688557
JH
1970
1971 do {
74e34e1f 1972 unsigned mi = (lo + hi) / 2;
628522ec
JH
1973 int cmp = hashcmp(index + mi * stride, sha1);
1974
1975 if (debug_lookup)
1976 printf("lo %u hi %u rg %u mi %u\n",
1977 lo, hi, hi - lo, mi);
43057304 1978 if (!cmp)
74e34e1f 1979 return nth_packed_object_offset(p, mi);
1f688557
JH
1980 if (cmp > 0)
1981 hi = mi;
1982 else
1983 lo = mi+1;
1984 } while (lo < hi);
1985 return 0;
1986}
1987
d131b7af
SP
1988static int is_pack_valid(struct packed_git *p)
1989{
1990 /* An already open pack is known to be valid. */
1991 if (p->pack_fd != -1)
1992 return 1;
1993
1994 /* If the pack has one window completely covering the
1995 * file size, the pack is known to be valid even if
1996 * the descriptor is not currently open.
1997 */
1998 if (p->windows) {
1999 struct pack_window *w = p->windows;
2000
2001 if (!w->offset && w->len == p->pack_size)
2002 return 1;
2003 }
2004
2005 /* Force the pack to open to prove its valid. */
2006 return !open_packed_git(p);
2007}
2008
4d6acb70 2009static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557 2010{
f7c22cc6 2011 static struct packed_git *last_found = (void *)1;
1f688557 2012 struct packed_git *p;
c4001d92 2013 off_t offset;
43057304 2014
1f688557 2015 prepare_packed_git();
f7c22cc6
NP
2016 if (!packed_git)
2017 return 0;
2018 p = (last_found == (void *)1) ? packed_git : last_found;
1f688557 2019
f7c22cc6 2020 do {
8eca0b47
NP
2021 if (p->num_bad_objects) {
2022 unsigned i;
2023 for (i = 0; i < p->num_bad_objects; i++)
2024 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2025 goto next;
2026 }
2027
43057304
NP
2028 offset = find_pack_entry_one(sha1, p);
2029 if (offset) {
c715f783
SP
2030 /*
2031 * We are about to tell the caller where they can
2032 * locate the requested object. We better make
2033 * sure the packfile is still here and can be
2034 * accessed before supplying that answer, as
2035 * it may have been deleted since the index
2036 * was loaded!
2037 */
d131b7af 2038 if (!is_pack_valid(p)) {
c715f783 2039 error("packfile %s cannot be accessed", p->pack_name);
f7c22cc6 2040 goto next;
c715f783 2041 }
43057304
NP
2042 e->offset = offset;
2043 e->p = p;
2044 hashcpy(e->sha1, sha1);
f7c22cc6 2045 last_found = p;
1f688557 2046 return 1;
43057304 2047 }
f7c22cc6
NP
2048
2049 next:
2050 if (p == last_found)
2051 p = packed_git;
2052 else
2053 p = p->next;
2054 if (p == last_found)
2055 p = p->next;
2056 } while (p);
1f688557
JH
2057 return 0;
2058}
2059
a6080a0a 2060struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
2061 struct packed_git *packs)
2062{
2063 struct packed_git *p;
bf592c50
DB
2064
2065 for (p = packs; p; p = p->next) {
43057304 2066 if (find_pack_entry_one(sha1, p))
bf592c50
DB
2067 return p;
2068 }
2069 return NULL;
21666f1a 2070
bf592c50
DB
2071}
2072
21666f1a 2073static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
65c2e0c3 2074{
36e4d74a 2075 int status;
65c2e0c3
JH
2076 unsigned long mapsize, size;
2077 void *map;
2078 z_stream stream;
d65a16f6 2079 char hdr[32];
65c2e0c3 2080
bb6b8e4f 2081 map = map_sha1_file(sha1, &mapsize);
f0df4ed5
JS
2082 if (!map)
2083 return error("unable to find %s", sha1_to_hex(sha1));
36e4d74a
JH
2084 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2085 status = error("unable to unpack %s header",
2086 sha1_to_hex(sha1));
21666f1a 2087 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 2088 status = error("unable to parse %s header", sha1_to_hex(sha1));
21666f1a
NP
2089 else if (sizep)
2090 *sizep = size;
39c68542 2091 git_inflate_end(&stream);
65c2e0c3
JH
2092 munmap(map, mapsize);
2093 return status;
2094}
2095
21666f1a 2096int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
f0df4ed5 2097{
c4d9986f 2098 struct cached_object *co;
f0df4ed5 2099 struct pack_entry e;
ddd63e64 2100 int status;
f0df4ed5 2101
c4d9986f
NTND
2102 co = find_cached_object(sha1);
2103 if (co) {
2104 if (sizep)
2105 *sizep = co->size;
2106 return co->type;
2107 }
2108
cd673c1f 2109 if (!find_pack_entry(sha1, &e)) {
ddd63e64
SG
2110 /* Most likely it's a loose object. */
2111 status = sha1_loose_object_info(sha1, sizep);
2112 if (status >= 0)
2113 return status;
2114
2115 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 2116 reprepare_packed_git();
cd673c1f 2117 if (!find_pack_entry(sha1, &e))
ddd63e64 2118 return status;
f0df4ed5 2119 }
3d77d877
NP
2120
2121 status = packed_object_info(e.p, e.offset, sizep);
2122 if (status < 0) {
2123 mark_bad_packed_object(e.p, sha1);
2124 status = sha1_object_info(sha1, sizep);
2125 }
2126
2127 return status;
f0df4ed5
JS
2128}
2129
21666f1a
NP
2130static void *read_packed_sha1(const unsigned char *sha1,
2131 enum object_type *type, unsigned long *size)
1f688557
JH
2132{
2133 struct pack_entry e;
8eca0b47 2134 void *data;
1f688557 2135
cd673c1f 2136 if (!find_pack_entry(sha1, &e))
1f688557 2137 return NULL;
8eca0b47
NP
2138 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2139 if (!data) {
2140 /*
2141 * We're probably in deep shit, but let's try to fetch
2142 * the required object anyway from another pack or loose.
2143 * This should happen only in the presence of a corrupted
2144 * pack, and is better than failing outright.
2145 */
2146 error("failed to read object %s at offset %"PRIuMAX" from %s",
2147 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2148 mark_bad_packed_object(e.p, sha1);
ac939109 2149 data = read_object(sha1, type, size);
8eca0b47
NP
2150 }
2151 return data;
1f688557
JH
2152}
2153
21666f1a
NP
2154int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2155 unsigned char *sha1)
d66b37bb
JH
2156{
2157 struct cached_object *co;
2158
21666f1a 2159 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2160 if (has_sha1_file(sha1) || find_cached_object(sha1))
2161 return 0;
2162 if (cached_object_alloc <= cached_object_nr) {
2163 cached_object_alloc = alloc_nr(cached_object_alloc);
2164 cached_objects = xrealloc(cached_objects,
2165 sizeof(*cached_objects) *
2166 cached_object_alloc);
2167 }
2168 co = &cached_objects[cached_object_nr++];
2169 co->size = len;
21666f1a 2170 co->type = type;
efa13f7b
JH
2171 co->buf = xmalloc(len);
2172 memcpy(co->buf, buf, len);
d66b37bb
JH
2173 hashcpy(co->sha1, sha1);
2174 return 0;
2175}
2176
c2c5b270
CC
2177static void *read_object(const unsigned char *sha1, enum object_type *type,
2178 unsigned long *size)
0fcfd160
LT
2179{
2180 unsigned long mapsize;
2181 void *map, *buf;
d66b37bb
JH
2182 struct cached_object *co;
2183
2184 co = find_cached_object(sha1);
2185 if (co) {
21666f1a 2186 *type = co->type;
d66b37bb 2187 *size = co->size;
182af834 2188 return xmemdupz(co->buf, co->size);
d66b37bb 2189 }
0fcfd160 2190
8276c007
PE
2191 buf = read_packed_sha1(sha1, type, size);
2192 if (buf)
2193 return buf;
bb6b8e4f 2194 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2195 if (map) {
7efbff75 2196 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2197 munmap(map, mapsize);
2198 return buf;
2199 }
637cdd9d 2200 reprepare_packed_git();
8276c007 2201 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2202}
2203
b6c4cecc
JH
2204/*
2205 * This function dies on corrupt objects; the callers who want to
2206 * deal with them should arrange to call read_object() and give error
2207 * messages themselves.
2208 */
5bf29b95
JH
2209void *read_sha1_file_extended(const unsigned char *sha1,
2210 enum object_type *type,
2211 unsigned long *size,
2212 unsigned flag)
ac939109 2213{
3ba7a065 2214 void *data;
e8b15e61 2215 char *path;
b6c4cecc 2216 const struct packed_git *p;
5bf29b95
JH
2217 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2218 ? lookup_replace_object(sha1) : sha1;
b6c4cecc 2219
3ba7a065
JH
2220 errno = 0;
2221 data = read_object(repl, type, size);
4bbf5a26 2222 if (data)
b6c4cecc 2223 return data;
68095570 2224
25f3af3f 2225 if (errno && errno != ENOENT)
3ba7a065
JH
2226 die_errno("failed to read object %s", sha1_to_hex(sha1));
2227
68095570 2228 /* die if we replaced an object with one that does not exist */
b6c4cecc 2229 if (repl != sha1)
68095570
CC
2230 die("replacement %s not found for %s",
2231 sha1_to_hex(repl), sha1_to_hex(sha1));
2232
b6c4cecc
JH
2233 if (has_loose_object(repl)) {
2234 path = sha1_file_name(sha1);
2235 die("loose object %s (stored in %s) is corrupt",
2236 sha1_to_hex(repl), path);
e8b15e61 2237 }
68095570 2238
b6c4cecc
JH
2239 if ((p = has_packed_and_bad(repl)) != NULL)
2240 die("packed object %s (stored in %s) is corrupt",
2241 sha1_to_hex(repl), p->pack_name);
f5552aee 2242
b6c4cecc 2243 return NULL;
ac939109
NP
2244}
2245
40469ee9 2246void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2247 const char *required_type_name,
40469ee9
JH
2248 unsigned long *size,
2249 unsigned char *actual_sha1_return)
f4913f91 2250{
21666f1a 2251 enum object_type type, required_type;
f4913f91
JH
2252 void *buffer;
2253 unsigned long isize;
40469ee9 2254 unsigned char actual_sha1[20];
f4913f91 2255
21666f1a 2256 required_type = type_from_string(required_type_name);
e702496e 2257 hashcpy(actual_sha1, sha1);
40469ee9
JH
2258 while (1) {
2259 int ref_length = -1;
2260 const char *ref_type = NULL;
f4913f91 2261
21666f1a 2262 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2263 if (!buffer)
2264 return NULL;
21666f1a 2265 if (type == required_type) {
40469ee9
JH
2266 *size = isize;
2267 if (actual_sha1_return)
e702496e 2268 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2269 return buffer;
2270 }
2271 /* Handle references */
21666f1a 2272 else if (type == OBJ_COMMIT)
40469ee9 2273 ref_type = "tree ";
21666f1a 2274 else if (type == OBJ_TAG)
40469ee9
JH
2275 ref_type = "object ";
2276 else {
2277 free(buffer);
2278 return NULL;
2279 }
2280 ref_length = strlen(ref_type);
f4913f91 2281
50974ec9
MK
2282 if (ref_length + 40 > isize ||
2283 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2284 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2285 free(buffer);
2286 return NULL;
2287 }
1cf58e72 2288 free(buffer);
40469ee9
JH
2289 /* Now we have the ID of the referred-to object in
2290 * actual_sha1. Check again. */
f4913f91 2291 }
f4913f91
JH
2292}
2293
ce9fbf16 2294static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2295 const char *type, unsigned char *sha1,
d65a16f6 2296 char *hdr, int *hdrlen)
d410c0f5 2297{
9126f009 2298 git_SHA_CTX c;
d410c0f5
JH
2299
2300 /* Generate the header */
d65a16f6 2301 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
2302
2303 /* Sha1.. */
9126f009
NP
2304 git_SHA1_Init(&c);
2305 git_SHA1_Update(&c, hdr, *hdrlen);
2306 git_SHA1_Update(&c, buf, len);
2307 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2308}
2309
230f1322 2310/*
5a688fe4
JH
2311 * Move the just written object into its final resting place.
2312 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2313 * "moving" is only a part of what it does, when no patch between
2314 * master to pu changes the call sites of this function.
230f1322 2315 */
839837b9 2316int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322 2317{
e32c0a9c 2318 int ret = 0;
5a688fe4 2319
348df166 2320 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
2321 goto try_rename;
2322 else if (link(tmpfile, filename))
e32c0a9c 2323 ret = errno;
7ebb6fca
LT
2324
2325 /*
2326 * Coda hack - coda doesn't like cross-directory links,
2327 * so we fall back to a rename, which will mean that it
2328 * won't be able to check collisions, but that's not a
2329 * big deal.
2330 *
2331 * The same holds for FAT formatted media.
2332 *
3be1f18e 2333 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
2334 * left to unlink.
2335 */
2336 if (ret && ret != EEXIST) {
be66a6c4 2337 try_rename:
7ebb6fca 2338 if (!rename(tmpfile, filename))
3be1f18e 2339 goto out;
9e48b389 2340 ret = errno;
230f1322 2341 }
691f1a28 2342 unlink_or_warn(tmpfile);
230f1322
LT
2343 if (ret) {
2344 if (ret != EEXIST) {
916d081b 2345 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
230f1322
LT
2346 }
2347 /* FIXME!!! Collision check here ? */
2348 }
2349
3be1f18e 2350out:
5256b006 2351 if (adjust_shared_perm(filename))
5a688fe4 2352 return error("unable to set permission to '%s'", filename);
230f1322
LT
2353 return 0;
2354}
2355
4d548150
LT
2356static int write_buffer(int fd, const void *buf, size_t len)
2357{
d34cf19b 2358 if (write_in_full(fd, buf, len) < 0)
93822c22 2359 return error("file write error (%s)", strerror(errno));
4d548150
LT
2360 return 0;
2361}
2362
ce9fbf16 2363int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2364 unsigned char *sha1)
2365{
d65a16f6 2366 char hdr[32];
abdc3fc8
RS
2367 int hdrlen;
2368 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2369 return 0;
2370}
2371
e9039dd3
LT
2372/* Finalize a file on disk, and close it. */
2373static void close_sha1_file(int fd)
2374{
aafe9fba
LT
2375 if (fsync_object_files)
2376 fsync_or_die(fd, "sha1 file");
e9039dd3 2377 if (close(fd) != 0)
d824cbba 2378 die_errno("error when closing sha1 file");
e9039dd3
LT
2379}
2380
5723fe7e
LT
2381/* Size of directory component, including the ending '/' */
2382static inline int directory_size(const char *filename)
2383{
2384 const char *s = strrchr(filename, '/');
2385 if (!s)
2386 return 0;
2387 return s - filename + 1;
2388}
2389
2390/*
2391 * This creates a temporary file in the same directory as the final
2392 * 'filename'
2393 *
2394 * We want to avoid cross-directory filename renames, because those
2395 * can have problems on various filesystems (FAT, NFS, Coda).
2396 */
2397static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2398{
2399 int fd, dirlen = directory_size(filename);
2400
2401 if (dirlen + 20 > bufsiz) {
2402 errno = ENAMETOOLONG;
2403 return -1;
2404 }
2405 memcpy(buffer, filename, dirlen);
2406 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
5256b006 2407 fd = git_mkstemp_mode(buffer, 0444);
cbacbf4e 2408 if (fd < 0 && dirlen && errno == ENOENT) {
5723fe7e 2409 /* Make sure the directory exists */
6ff6af62 2410 memcpy(buffer, filename, dirlen);
5723fe7e 2411 buffer[dirlen-1] = 0;
3bfaf018 2412 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
5723fe7e
LT
2413 return -1;
2414
2415 /* Try again */
2416 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
5256b006 2417 fd = git_mkstemp_mode(buffer, 0444);
5723fe7e
LT
2418 }
2419 return fd;
2420}
2421
bbac7311 2422static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 2423 const void *buf, unsigned long len, time_t mtime)
0fcfd160 2424{
915308b1 2425 int fd, ret;
9892beba 2426 unsigned char compressed[4096];
0fcfd160 2427 z_stream stream;
748af44c
NP
2428 git_SHA_CTX c;
2429 unsigned char parano_sha1[20];
706bc531 2430 char *filename;
aac17941 2431 static char tmpfile[PATH_MAX];
a44c9a5e 2432
972a9155 2433 filename = sha1_file_name(sha1);
5723fe7e 2434 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
aac17941 2435 if (fd < 0) {
35243577 2436 if (errno == EACCES)
916d081b
PB
2437 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2438 else
2439 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
aac17941
LT
2440 }
2441
0fcfd160
LT
2442 /* Set it up */
2443 memset(&stream, 0, sizeof(stream));
12f6c308 2444 deflateInit(&stream, zlib_compression_level);
0fcfd160 2445 stream.next_out = compressed;
9892beba 2446 stream.avail_out = sizeof(compressed);
748af44c 2447 git_SHA1_Init(&c);
a44c9a5e
LT
2448
2449 /* First header.. */
d65a16f6 2450 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2451 stream.avail_in = hdrlen;
726f852b
NP
2452 while (deflate(&stream, 0) == Z_OK)
2453 /* nothing */;
748af44c 2454 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
2455
2456 /* Then the data itself.. */
c00e657d 2457 stream.next_in = (void *)buf;
a44c9a5e 2458 stream.avail_in = len;
9892beba 2459 do {
748af44c 2460 unsigned char *in0 = stream.next_in;
9892beba 2461 ret = deflate(&stream, Z_FINISH);
748af44c 2462 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
2463 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2464 die("unable to write sha1 file");
2465 stream.next_out = compressed;
2466 stream.avail_out = sizeof(compressed);
2467 } while (ret == Z_OK);
2468
ac54c277
LT
2469 if (ret != Z_STREAM_END)
2470 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
ac54c277
LT
2471 ret = deflateEnd(&stream);
2472 if (ret != Z_OK)
2473 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
2474 git_SHA1_Final(parano_sha1, &c);
2475 if (hashcmp(sha1, parano_sha1) != 0)
2476 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 2477
e9039dd3 2478 close_sha1_file(fd);
0fcfd160 2479
bbac7311
NP
2480 if (mtime) {
2481 struct utimbuf utb;
2482 utb.actime = mtime;
2483 utb.modtime = mtime;
2484 if (utime(tmpfile, &utb) < 0)
2485 warning("failed utime() on %s: %s",
2486 tmpfile, strerror(errno));
2487 }
2488
230f1322 2489 return move_temp_to_file(tmpfile, filename);
0fcfd160 2490}
8237b185 2491
c00e657d 2492int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
bbac7311
NP
2493{
2494 unsigned char sha1[20];
2495 char hdr[32];
2496 int hdrlen;
2497
2498 /* Normally if we have it in the pack then we do not bother writing
2499 * it out into .git/objects/??/?{38} file.
2500 */
2501 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2502 if (returnsha1)
2503 hashcpy(returnsha1, sha1);
2504 if (has_sha1_file(sha1))
2505 return 0;
2506 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2507}
2508
2509int force_object_loose(const unsigned char *sha1, time_t mtime)
2510{
bbac7311
NP
2511 void *buf;
2512 unsigned long len;
2513 enum object_type type;
2514 char hdr[32];
2515 int hdrlen;
1fb23e65 2516 int ret;
bbac7311 2517
c529d75a 2518 if (has_loose_object(sha1))
bbac7311
NP
2519 return 0;
2520 buf = read_packed_sha1(sha1, &type, &len);
2521 if (!buf)
2522 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2523 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1fb23e65
BS
2524 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2525 free(buf);
2526
2527 return ret;
bbac7311
NP
2528}
2529
bf592c50
DB
2530int has_pack_index(const unsigned char *sha1)
2531{
2532 struct stat st;
2533 if (stat(sha1_pack_index_name(sha1), &st))
2534 return 0;
2535 return 1;
bf592c50
DB
2536}
2537
cd673c1f
JH
2538int has_sha1_pack(const unsigned char *sha1)
2539{
2540 struct pack_entry e;
2541 return find_pack_entry(sha1, &e);
2542}
2543
8237b185
DB
2544int has_sha1_file(const unsigned char *sha1)
2545{
1f688557
JH
2546 struct pack_entry e;
2547
cd673c1f 2548 if (find_pack_entry(sha1, &e))
1f688557 2549 return 1;
c529d75a 2550 return has_loose_object(sha1);
8237b185 2551}
74400e71 2552
c879daa2
NTND
2553static void check_tree(const void *buf, size_t size)
2554{
2555 struct tree_desc desc;
2556 struct name_entry entry;
2557
2558 init_tree_desc(&desc, buf, size);
2559 while (tree_entry(&desc, &entry))
2560 /* do nothing
2561 * tree_entry() will die() on malformed entries */
2562 ;
2563}
2564
2565static void check_commit(const void *buf, size_t size)
2566{
2567 struct commit c;
2568 memset(&c, 0, sizeof(c));
2569 if (parse_commit_buffer(&c, buf, size))
2570 die("corrupt commit");
2571}
2572
2573static void check_tag(const void *buf, size_t size)
2574{
2575 struct tag t;
2576 memset(&t, 0, sizeof(t));
2577 if (parse_tag_buffer(&t, buf, size))
2578 die("corrupt tag");
2579}
2580
43df4f86 2581static int index_mem(unsigned char *sha1, void *buf, size_t size,
c4ce46fc
JH
2582 enum object_type type,
2583 const char *path, unsigned flags)
e7332f96 2584{
6c510bee 2585 int ret, re_allocated = 0;
c4ce46fc 2586 int write_object = flags & HASH_WRITE_OBJECT;
74400e71 2587
7672db20 2588 if (!type)
edaec3fb 2589 type = OBJ_BLOB;
6c510bee
LT
2590
2591 /*
2592 * Convert blobs to git internal format
2593 */
43df4f86 2594 if ((type == OBJ_BLOB) && path) {
f285a2d7 2595 struct strbuf nbuf = STRBUF_INIT;
21e5ad50
SP
2596 if (convert_to_git(path, buf, size, &nbuf,
2597 write_object ? safe_crlf : 0)) {
b315c5c0 2598 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
2599 re_allocated = 1;
2600 }
2601 }
c4ce46fc 2602 if (flags & HASH_FORMAT_CHECK) {
c879daa2
NTND
2603 if (type == OBJ_TREE)
2604 check_tree(buf, size);
2605 if (type == OBJ_COMMIT)
2606 check_commit(buf, size);
2607 if (type == OBJ_TAG)
2608 check_tag(buf, size);
2609 }
6c510bee 2610
7672db20 2611 if (write_object)
edaec3fb 2612 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 2613 else
edaec3fb 2614 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 2615 if (re_allocated)
6c510bee 2616 free(buf);
43df4f86
DP
2617 return ret;
2618}
2619
7b41e1e1
JH
2620static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2621 const char *path, unsigned flags)
2622{
2623 struct strbuf sbuf = STRBUF_INIT;
2624 int ret;
2625
2626 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2627 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2628 else
2629 ret = -1;
2630 strbuf_release(&sbuf);
2631 return ret;
2632}
2633
ea68b0ce
DP
2634#define SMALL_FILE_SIZE (32*1024)
2635
7b41e1e1
JH
2636static int index_core(unsigned char *sha1, int fd, size_t size,
2637 enum object_type type, const char *path,
2638 unsigned flags)
43df4f86
DP
2639{
2640 int ret;
43df4f86 2641
7b41e1e1 2642 if (!size) {
c4ce46fc 2643 ret = index_mem(sha1, NULL, size, type, path, flags);
ea68b0ce
DP
2644 } else if (size <= SMALL_FILE_SIZE) {
2645 char *buf = xmalloc(size);
2646 if (size == read_in_full(fd, buf, size))
c4ce46fc 2647 ret = index_mem(sha1, buf, size, type, path, flags);
ea68b0ce
DP
2648 else
2649 ret = error("short read %s", strerror(errno));
2650 free(buf);
08bda208 2651 } else {
43df4f86 2652 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
c4ce46fc 2653 ret = index_mem(sha1, buf, size, type, path, flags);
aac17941 2654 munmap(buf, size);
08bda208 2655 }
7b41e1e1
JH
2656 return ret;
2657}
2658
4dd1fbc7
JH
2659/*
2660 * This creates one packfile per large blob, because the caller
2661 * immediately wants the result sha1, and fast-import can report the
2662 * object name via marks mechanism only by closing the created
2663 * packfile.
2664 *
2665 * This also bypasses the usual "convert-to-git" dance, and that is on
2666 * purpose. We could write a streaming version of the converting
2667 * functions and insert that before feeding the data to fast-import
2668 * (or equivalent in-core API described above), but the primary
2669 * motivation for trying to stream from the working tree file and to
2670 * avoid mmaping it in core is to deal with large binary blobs, and
2671 * by definition they do _not_ want to get any conversion.
2672 */
2673static int index_stream(unsigned char *sha1, int fd, size_t size,
2674 enum object_type type, const char *path,
2675 unsigned flags)
2676{
2677 struct child_process fast_import;
2678 char export_marks[512];
2679 const char *argv[] = { "fast-import", "--quiet", export_marks, NULL };
2680 char tmpfile[512];
2681 char fast_import_cmd[512];
2682 char buf[512];
2683 int len, tmpfd;
2684
2685 strcpy(tmpfile, git_path("hashstream_XXXXXX"));
2686 tmpfd = git_mkstemp_mode(tmpfile, 0600);
2687 if (tmpfd < 0)
2688 die_errno("cannot create tempfile: %s", tmpfile);
2689 if (close(tmpfd))
2690 die_errno("cannot close tempfile: %s", tmpfile);
2691 sprintf(export_marks, "--export-marks=%s", tmpfile);
2692
2693 memset(&fast_import, 0, sizeof(fast_import));
2694 fast_import.in = -1;
2695 fast_import.argv = argv;
2696 fast_import.git_cmd = 1;
2697 if (start_command(&fast_import))
2698 die_errno("index-stream: git fast-import failed");
2699
2700 len = sprintf(fast_import_cmd, "blob\nmark :1\ndata %lu\n",
2701 (unsigned long) size);
2702 write_or_whine(fast_import.in, fast_import_cmd, len,
2703 "index-stream: feeding fast-import");
2704 while (size) {
2705 char buf[10240];
2706 size_t sz = size < sizeof(buf) ? size : sizeof(buf);
2707 size_t actual;
2708
2709 actual = read_in_full(fd, buf, sz);
2710 if (actual < 0)
2711 die_errno("index-stream: reading input");
2712 if (write_in_full(fast_import.in, buf, actual) != actual)
2713 die_errno("index-stream: feeding fast-import");
2714 size -= actual;
2715 }
2716 if (close(fast_import.in))
2717 die_errno("index-stream: closing fast-import");
2718 if (finish_command(&fast_import))
2719 die_errno("index-stream: finishing fast-import");
2720
2721 tmpfd = open(tmpfile, O_RDONLY);
2722 if (tmpfd < 0)
2723 die_errno("index-stream: cannot open fast-import mark");
2724 len = read(tmpfd, buf, sizeof(buf));
2725 if (len < 0)
2726 die_errno("index-stream: reading fast-import mark");
2727 if (close(tmpfd) < 0)
2728 die_errno("index-stream: closing fast-import mark");
2729 if (unlink(tmpfile))
2730 die_errno("index-stream: unlinking fast-import mark");
2731 if (len != 44 ||
2732 memcmp(":1 ", buf, 3) ||
2733 get_sha1_hex(buf + 3, sha1))
2734 die_errno("index-stream: unexpected fast-import mark: <%s>", buf);
2735 return 0;
2736}
2737
7b41e1e1
JH
2738int index_fd(unsigned char *sha1, int fd, struct stat *st,
2739 enum object_type type, const char *path, unsigned flags)
2740{
2741 int ret;
2742 size_t size = xsize_t(st->st_size);
2743
2744 if (!S_ISREG(st->st_mode))
2745 ret = index_pipe(sha1, fd, type, path, flags);
4dd1fbc7 2746 else if (size <= big_file_threshold || type != OBJ_BLOB)
7b41e1e1 2747 ret = index_core(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
2748 else
2749 ret = index_stream(sha1, fd, size, type, path, flags);
43df4f86 2750 close(fd);
aac17941 2751 return ret;
74400e71 2752}
ec1fcc16 2753
c4ce46fc 2754int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
ec1fcc16
JH
2755{
2756 int fd;
b760d3aa 2757 struct strbuf sb = STRBUF_INIT;
ec1fcc16
JH
2758
2759 switch (st->st_mode & S_IFMT) {
2760 case S_IFREG:
2761 fd = open(path, O_RDONLY);
2762 if (fd < 0)
2763 return error("open(\"%s\"): %s", path,
2764 strerror(errno));
c4ce46fc 2765 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
ec1fcc16
JH
2766 return error("%s: failed to insert into database",
2767 path);
2768 break;
2769 case S_IFLNK:
b760d3aa 2770 if (strbuf_readlink(&sb, path, st->st_size)) {
ec1fcc16 2771 char *errstr = strerror(errno);
ec1fcc16
JH
2772 return error("readlink(\"%s\"): %s", path,
2773 errstr);
2774 }
c4ce46fc 2775 if (!(flags & HASH_WRITE_OBJECT))
b760d3aa
LT
2776 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2777 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
ec1fcc16
JH
2778 return error("%s: failed to insert into database",
2779 path);
b760d3aa 2780 strbuf_release(&sb);
ec1fcc16 2781 break;
f35a6d3b
LT
2782 case S_IFDIR:
2783 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
2784 default:
2785 return error("%s: unsupported file type", path);
2786 }
2787 return 0;
2788}
a69e5429
JH
2789
2790int read_pack_header(int fd, struct pack_header *header)
2791{
c697ad14
HO
2792 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2793 /* "eof before pack header was fully read" */
2794 return PH_ERROR_EOF;
2795
a69e5429
JH
2796 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2797 /* "protocol error (pack signature mismatch detected)" */
2798 return PH_ERROR_PACK_SIGNATURE;
2799 if (!pack_version_ok(header->hdr_version))
2800 /* "protocol error (pack version unsupported)" */
2801 return PH_ERROR_PROTOCOL;
2802 return 0;
2803}
40d52ff7
JK
2804
2805void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2806{
2807 enum object_type type = sha1_object_info(sha1, NULL);
2808 if (type < 0)
2809 die("%s is not a valid object", sha1_to_hex(sha1));
2810 if (type != expect)
2811 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2812 typename(expect));
2813}