pack-objects: make in_pack_header_size a variable of its own
[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"
14#include "tag.h"
15#include "tree.h"
0fcfd160 16
144bde78
LT
17#ifndef O_NOATIME
18#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
19#define O_NOATIME 01000000
20#else
21#define O_NOATIME 0
22#endif
23#endif
24
e05db0fd
PR
25#ifdef NO_C99_FORMAT
26#define SZ_FMT "lu"
27#else
28#define SZ_FMT "zu"
29#endif
30
96f1e58f 31const unsigned char null_sha1[20];
88cd621d 32
144bde78
LT
33static unsigned int sha1_file_open_flag = O_NOATIME;
34
e49521b5
JH
35signed char hexval_table[256] = {
36 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
37 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
38 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
42 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
43 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
44 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
45 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
47 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
48 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
68};
0fcfd160
LT
69
70int get_sha1_hex(const char *hex, unsigned char *sha1)
71{
72 int i;
73 for (i = 0; i < 20; i++) {
74 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
75 if (val & ~0xff)
76 return -1;
77 *sha1++ = val;
78 hex += 2;
79 }
80 return 0;
81}
82
b2cb9425
JH
83int safe_create_leading_directories(char *path)
84{
85 char *pos = path;
67d42212
JR
86 struct stat st;
87
67ffdf4c
JS
88 if (*pos == '/')
89 pos++;
b2cb9425
JH
90
91 while (pos) {
92 pos = strchr(pos, '/');
93 if (!pos)
94 break;
95 *pos = 0;
67d42212
JR
96 if (!stat(path, &st)) {
97 /* path exists */
98 if (!S_ISDIR(st.st_mode)) {
b2cb9425 99 *pos = '/';
67d42212 100 return -3;
b2cb9425 101 }
457f06d6 102 }
67d42212
JR
103 else if (mkdir(path, 0777)) {
104 *pos = '/';
105 return -1;
106 }
457f06d6
JS
107 else if (adjust_shared_perm(path)) {
108 *pos = '/';
109 return -2;
110 }
b2cb9425
JH
111 *pos++ = '/';
112 }
113 return 0;
114}
723c31fe 115
0fcfd160
LT
116char * sha1_to_hex(const unsigned char *sha1)
117{
dcb3450f
LT
118 static int bufno;
119 static char hexbuffer[4][50];
0fcfd160 120 static const char hex[] = "0123456789abcdef";
dcb3450f 121 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
0fcfd160
LT
122 int i;
123
124 for (i = 0; i < 20; i++) {
125 unsigned int val = *sha1++;
126 *buf++ = hex[val >> 4];
127 *buf++ = hex[val & 0xf];
128 }
1e80e044
JS
129 *buf = '\0';
130
0fcfd160
LT
131 return buffer;
132}
133
ace1534d
JH
134static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
135{
136 int i;
137 for (i = 0; i < 20; i++) {
138 static char hex[] = "0123456789abcdef";
139 unsigned int val = sha1[i];
140 char *pos = pathbuf + i*2 + (i > 0);
141 *pos++ = hex[val >> 4];
142 *pos = hex[val & 0xf];
143 }
144}
145
0fcfd160
LT
146/*
147 * NOTE! This returns a statically allocated buffer, so you have to be
9befac47 148 * careful about using it. Do a "xstrdup()" if you need to save the
0fcfd160 149 * filename.
ace1534d
JH
150 *
151 * Also note that this returns the location for creating. Reading
152 * SHA1 file can happen from any alternate directory listed in the
d19938ab 153 * DB_ENVIRONMENT environment variable if it is not found in
ace1534d 154 * the primary object database.
0fcfd160
LT
155 */
156char *sha1_file_name(const unsigned char *sha1)
157{
0fcfd160
LT
158 static char *name, *base;
159
160 if (!base) {
d19938ab 161 const char *sha1_file_directory = get_object_directory();
0fcfd160 162 int len = strlen(sha1_file_directory);
812666c8 163 base = xmalloc(len + 60);
0fcfd160
LT
164 memcpy(base, sha1_file_directory, len);
165 memset(base+len, 0, 60);
166 base[len] = '/';
167 base[len+3] = '/';
168 name = base + len + 1;
169 }
ace1534d 170 fill_sha1_path(name, sha1);
0fcfd160
LT
171 return base;
172}
173
bf592c50
DB
174char *sha1_pack_name(const unsigned char *sha1)
175{
176 static const char hex[] = "0123456789abcdef";
177 static char *name, *base, *buf;
178 int i;
179
180 if (!base) {
181 const char *sha1_file_directory = get_object_directory();
182 int len = strlen(sha1_file_directory);
183 base = xmalloc(len + 60);
184 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
185 name = base + len + 11;
186 }
187
188 buf = name;
189
190 for (i = 0; i < 20; i++) {
191 unsigned int val = *sha1++;
192 *buf++ = hex[val >> 4];
193 *buf++ = hex[val & 0xf];
194 }
195
196 return base;
197}
198
199char *sha1_pack_index_name(const unsigned char *sha1)
200{
201 static const char hex[] = "0123456789abcdef";
202 static char *name, *base, *buf;
203 int i;
204
205 if (!base) {
206 const char *sha1_file_directory = get_object_directory();
207 int len = strlen(sha1_file_directory);
208 base = xmalloc(len + 60);
209 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
210 name = base + len + 11;
211 }
212
213 buf = name;
214
215 for (i = 0; i < 20; i++) {
216 unsigned int val = *sha1++;
217 *buf++ = hex[val >> 4];
218 *buf++ = hex[val & 0xf];
219 }
220
221 return base;
222}
223
d5a63b99
JH
224struct alternate_object_database *alt_odb_list;
225static struct alternate_object_database **alt_odb_tail;
ace1534d 226
c2f493a4
MW
227static void read_info_alternates(const char * alternates, int depth);
228
ddd5d056
JH
229/*
230 * Prepare alternate object database registry.
d5a63b99
JH
231 *
232 * The variable alt_odb_list points at the list of struct
233 * alternate_object_database. The elements on this list come from
234 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
235 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
236 * whose contents is similar to that environment variable but can be
237 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
238 * contains "/the/directory/corresponding/to/.git/objects/...", while
239 * its name points just after the slash at the end of ".git/objects/"
240 * in the example above, and has enough space to hold 40-byte hex
241 * SHA1, an extra slash for the first level indirection, and the
242 * terminating NUL.
ddd5d056 243 */
c2f493a4 244static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
ace1534d 245{
c2f493a4 246 struct stat st;
1494e038 247 const char *objdir = get_object_directory();
c2f493a4
MW
248 struct alternate_object_database *ent;
249 struct alternate_object_database *alt;
250 /* 43 = 40-byte + 2 '/' + terminating NUL */
251 int pfxlen = len;
252 int entlen = pfxlen + 43;
ccfd3e99 253 int base_len = -1;
d5a63b99 254
c2f493a4
MW
255 if (*entry != '/' && relative_base) {
256 /* Relative alt-odb */
257 if (base_len < 0)
258 base_len = strlen(relative_base) + 1;
259 entlen += base_len;
260 pfxlen += base_len;
261 }
262 ent = xmalloc(sizeof(*ent) + entlen);
263
264 if (*entry != '/' && relative_base) {
265 memcpy(ent->base, relative_base, base_len - 1);
266 ent->base[base_len - 1] = '/';
267 memcpy(ent->base + base_len, entry, len);
268 }
269 else
270 memcpy(ent->base, entry, pfxlen);
271
272 ent->name = ent->base + pfxlen + 1;
273 ent->base[pfxlen + 3] = '/';
274 ent->base[pfxlen] = ent->base[entlen-1] = 0;
275
276 /* Detect cases where alternate disappeared */
277 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
278 error("object directory %s does not exist; "
279 "check .git/objects/info/alternates.",
280 ent->base);
281 free(ent);
282 return -1;
283 }
284
285 /* Prevent the common mistake of listing the same
286 * thing twice, or object directory itself.
287 */
288 for (alt = alt_odb_list; alt; alt = alt->next) {
289 if (!memcmp(ent->base, alt->base, pfxlen)) {
290 free(ent);
291 return -1;
292 }
293 }
294 if (!memcmp(ent->base, objdir, pfxlen)) {
295 free(ent);
296 return -1;
297 }
298
299 /* add the alternate entry */
300 *alt_odb_tail = ent;
301 alt_odb_tail = &(ent->next);
302 ent->next = NULL;
303
304 /* recursively add alternates */
305 read_info_alternates(ent->base, depth + 1);
306
307 ent->base[pfxlen] = '/';
308
309 return 0;
310}
311
312static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
313 const char *relative_base, int depth)
314{
315 const char *cp, *last;
316
317 if (depth > 5) {
318 error("%s: ignoring alternate object stores, nesting too deep.",
319 relative_base);
320 return;
321 }
322
d5a63b99 323 last = alt;
9577e7e3
JH
324 while (last < ep) {
325 cp = last;
326 if (cp < ep && *cp == '#') {
327 while (cp < ep && *cp != sep)
328 cp++;
329 last = cp + 1;
330 continue;
331 }
c2f493a4
MW
332 while (cp < ep && *cp != sep)
333 cp++;
d5a63b99 334 if (last != cp) {
c2f493a4
MW
335 if ((*last != '/') && depth) {
336 error("%s: ignoring relative alternate object store %s",
337 relative_base, last);
338 } else {
339 link_alt_odb_entry(last, cp - last,
340 relative_base, depth);
1494e038 341 }
d5a63b99 342 }
9577e7e3 343 while (cp < ep && *cp == sep)
d5a63b99
JH
344 cp++;
345 last = cp;
9577e7e3 346 }
d5a63b99
JH
347}
348
c2f493a4 349static void read_info_alternates(const char * relative_base, int depth)
d5a63b99 350{
9577e7e3 351 char *map;
dc49cd76 352 size_t mapsz;
d5a63b99 353 struct stat st;
c2f493a4
MW
354 char path[PATH_MAX];
355 int fd;
d5a63b99 356
c2f493a4 357 sprintf(path, "%s/info/alternates", relative_base);
d5a63b99
JH
358 fd = open(path, O_RDONLY);
359 if (fd < 0)
360 return;
361 if (fstat(fd, &st) || (st.st_size == 0)) {
362 close(fd);
9a217f2a 363 return;
ace1534d 364 }
dc49cd76
SP
365 mapsz = xsize_t(st.st_size);
366 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 367 close(fd);
d5a63b99 368
dc49cd76 369 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
c2f493a4 370
dc49cd76 371 munmap(map, mapsz);
ace1534d
JH
372}
373
c2f493a4
MW
374void prepare_alt_odb(void)
375{
554fe20d 376 const char *alt;
c2f493a4
MW
377
378 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
379 if (!alt) alt = "";
380
381 if (alt_odb_tail)
382 return;
383 alt_odb_tail = &alt_odb_list;
384 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
385
386 read_info_alternates(get_object_directory(), 0);
387}
388
ace1534d
JH
389static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
390{
ace1534d 391 char *name = sha1_file_name(sha1);
d5a63b99 392 struct alternate_object_database *alt;
ace1534d
JH
393
394 if (!stat(name, st))
395 return name;
9a217f2a 396 prepare_alt_odb();
d5a63b99
JH
397 for (alt = alt_odb_list; alt; alt = alt->next) {
398 name = alt->name;
ace1534d 399 fill_sha1_path(name, sha1);
d5a63b99
JH
400 if (!stat(alt->base, st))
401 return alt->base;
ace1534d
JH
402 }
403 return NULL;
404}
405
60bb8b14 406static unsigned int pack_used_ctr;
a53128b6
SP
407static unsigned int pack_mmap_calls;
408static unsigned int peak_pack_open_windows;
409static unsigned int pack_open_windows;
410static size_t peak_pack_mapped;
60bb8b14 411static size_t pack_mapped;
9a217f2a 412struct packed_git *packed_git;
1f688557 413
a53128b6
SP
414void pack_report()
415{
416 fprintf(stderr,
e05db0fd
PR
417 "pack_report: getpagesize() = %10" SZ_FMT "\n"
418 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
419 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
78a28df9 420 (size_t) getpagesize(),
a53128b6
SP
421 packed_git_window_size,
422 packed_git_limit);
423 fprintf(stderr,
424 "pack_report: pack_used_ctr = %10u\n"
425 "pack_report: pack_mmap_calls = %10u\n"
426 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
427 "pack_report: pack_mapped = "
428 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
429 pack_used_ctr,
430 pack_mmap_calls,
431 pack_open_windows, peak_pack_open_windows,
432 pack_mapped, peak_pack_mapped);
433}
434
42873078 435static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
436{
437 void *idx_map;
42873078 438 struct pack_idx_header *hdr;
dc49cd76 439 size_t idx_size;
74e34e1f 440 uint32_t version, nr, i, *index;
e93ec6f9 441 int fd = open(path, O_RDONLY);
1f688557 442 struct stat st;
42873078 443
1f688557
JH
444 if (fd < 0)
445 return -1;
446 if (fstat(fd, &st)) {
447 close(fd);
448 return -1;
449 }
dc49cd76 450 idx_size = xsize_t(st.st_size);
2d88451b
SP
451 if (idx_size < 4 * 256 + 20 + 20) {
452 close(fd);
453 return error("index file %s is too small", path);
454 }
c4712e45 455 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 456 close(fd);
1f688557 457
42873078
NP
458 hdr = idx_map;
459 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
460 version = ntohl(hdr->idx_version);
461 if (version < 2 || version > 2) {
462 munmap(idx_map, idx_size);
463 return error("index file %s is version %d"
464 " and is not supported by this binary"
465 " (try upgrading GIT to a newer version)",
466 path, version);
467 }
468 } else
469 version = 1;
df1b059d 470
1f688557 471 nr = 0;
42873078 472 index = idx_map;
74e34e1f
NP
473 if (version > 1)
474 index += 2; /* skip index header */
1f688557 475 for (i = 0; i < 256; i++) {
326bf396 476 uint32_t n = ntohl(index[i]);
2d88451b
SP
477 if (n < nr) {
478 munmap(idx_map, idx_size);
df1b059d 479 return error("non-monotonic index %s", path);
2d88451b 480 }
1f688557
JH
481 nr = n;
482 }
483
74e34e1f
NP
484 if (version == 1) {
485 /*
486 * Total size:
487 * - 256 index entries 4 bytes each
488 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
489 * - 20-byte SHA1 of the packfile
490 * - 20-byte SHA1 file checksum
491 */
492 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
493 munmap(idx_map, idx_size);
494 return error("wrong index file size in %s", path);
495 }
496 } else if (version == 2) {
497 /*
498 * Minimum size:
499 * - 8 bytes of header
500 * - 256 index entries 4 bytes each
501 * - 20-byte sha1 entry * nr
502 * - 4-byte crc entry * nr
503 * - 4-byte offset entry * nr
504 * - 20-byte SHA1 of the packfile
505 * - 20-byte SHA1 file checksum
506 * And after the 4-byte offset table might be a
507 * variable sized table containing 8-byte entries
508 * for offsets larger than 2^31.
509 */
510 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
511 if (idx_size < min_size || idx_size > min_size + (nr - 1)*8) {
512 munmap(idx_map, idx_size);
513 return error("wrong index file size in %s", path);
514 }
515 if (idx_size != min_size) {
516 /* make sure we can deal with large pack offsets */
517 off_t x = 0x7fffffffUL, y = 0xffffffffUL;
518 if (x > (x + 1) || y > (y + 1)) {
519 munmap(idx_map, idx_size);
520 return error("pack too large for current definition of off_t in %s", path);
521 }
522 }
2d88451b 523 }
1f688557 524
74e34e1f 525 p->index_version = version;
42873078
NP
526 p->index_data = idx_map;
527 p->index_size = idx_size;
57059091 528 p->num_objects = nr;
1f688557
JH
529 return 0;
530}
531
11daf39b
SP
532static void scan_windows(struct packed_git *p,
533 struct packed_git **lru_p,
534 struct pack_window **lru_w,
535 struct pack_window **lru_l)
1f688557 536{
11daf39b
SP
537 struct pack_window *w, *w_l;
538
539 for (w_l = NULL, w = p->windows; w; w = w->next) {
540 if (!w->inuse_cnt) {
541 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
542 *lru_p = p;
543 *lru_w = w;
544 *lru_l = w_l;
54044bf8 545 }
54044bf8 546 }
11daf39b 547 w_l = w;
f9253394 548 }
11daf39b
SP
549}
550
551static int unuse_one_window(struct packed_git *current)
552{
553 struct packed_git *p, *lru_p = NULL;
554 struct pack_window *lru_w = NULL, *lru_l = NULL;
555
556 if (current)
557 scan_windows(current, &lru_p, &lru_w, &lru_l);
558 for (p = packed_git; p; p = p->next)
559 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
560 if (lru_p) {
561 munmap(lru_w->base, lru_w->len);
562 pack_mapped -= lru_w->len;
563 if (lru_l)
564 lru_l->next = lru_w->next;
565 else {
566 lru_p->windows = lru_w->next;
11daf39b 567 if (!lru_p->windows && lru_p != current) {
54044bf8
SP
568 close(lru_p->pack_fd);
569 lru_p->pack_fd = -1;
570 }
571 }
572 free(lru_w);
a53128b6 573 pack_open_windows--;
54044bf8
SP
574 return 1;
575 }
576 return 0;
f9253394
JH
577}
578
97bfeb34
SP
579void release_pack_memory(size_t need)
580{
581 size_t cur = pack_mapped;
582 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
583 ; /* nothing */
584}
585
03e79c88 586void unuse_pack(struct pack_window **w_cursor)
f9253394 587{
03e79c88
SP
588 struct pack_window *w = *w_cursor;
589 if (w) {
590 w->inuse_cnt--;
591 *w_cursor = NULL;
592 }
1f688557
JH
593}
594
3cf8b462
SP
595/*
596 * Do not call this directly as this leaks p->pack_fd on error return;
597 * call open_packed_git() instead.
598 */
599static int open_packed_git_1(struct packed_git *p)
1f688557 600{
9bc879c1
SP
601 struct stat st;
602 struct pack_header hdr;
603 unsigned char sha1[20];
604 unsigned char *idx_sha1;
2c039da8 605 long fd_flag;
9bc879c1
SP
606
607 p->pack_fd = open(p->pack_name, O_RDONLY);
608 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 609 return -1;
9bc879c1
SP
610
611 /* If we created the struct before we had the pack we lack size. */
bf592c50 612 if (!p->pack_size) {
bf592c50 613 if (!S_ISREG(st.st_mode))
072db278 614 return error("packfile %s not a regular file", p->pack_name);
bf592c50 615 p->pack_size = st.st_size;
9bc879c1 616 } else if (p->pack_size != st.st_size)
072db278 617 return error("packfile %s size changed", p->pack_name);
9bc879c1 618
2c039da8
JH
619 /* We leave these file descriptors open with sliding mmap;
620 * there is no point keeping them open across exec(), though.
621 */
622 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
623 if (fd_flag < 0)
072db278 624 return error("cannot determine file descriptor flags");
2c039da8
JH
625 fd_flag |= FD_CLOEXEC;
626 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 627 return error("cannot set FD_CLOEXEC");
2c039da8 628
9bc879c1 629 /* Verify we recognize this pack file format. */
e6e2bd62 630 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 631 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 632 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 633 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 634 if (!pack_version_ok(hdr.hdr_version))
072db278 635 return error("packfile %s is version %u and not supported"
9bc879c1
SP
636 " (try upgrading GIT to a newer version)",
637 p->pack_name, ntohl(hdr.hdr_version));
638
639 /* Verify the pack matches its index. */
57059091 640 if (p->num_objects != ntohl(hdr.hdr_entries))
072db278 641 return error("packfile %s claims to have %u objects"
57059091
NP
642 " while index indicates %u objects",
643 p->pack_name, ntohl(hdr.hdr_entries),
644 p->num_objects);
9bc879c1 645 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 646 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 647 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 648 return error("packfile %s signature is unavailable", p->pack_name);
42873078 649 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 650 if (hashcmp(sha1, idx_sha1))
072db278
SP
651 return error("packfile %s does not match index", p->pack_name);
652 return 0;
9bc879c1
SP
653}
654
3cf8b462
SP
655static int open_packed_git(struct packed_git *p)
656{
657 if (!open_packed_git_1(p))
658 return 0;
659 if (p->pack_fd != -1) {
660 close(p->pack_fd);
661 p->pack_fd = -1;
662 }
663 return -1;
664}
665
c4001d92 666static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
667{
668 /* We must promise at least 20 bytes (one hash) after the
669 * offset is available from this window, otherwise the offset
670 * is not actually in this window and a different window (which
671 * has that one hash excess) must be used. This is to support
672 * the object header and delta base parsing routines below.
673 */
674 off_t win_off = win->offset;
675 return win_off <= offset
676 && (offset + 20) <= (win_off + win->len);
677}
678
03e79c88
SP
679unsigned char* use_pack(struct packed_git *p,
680 struct pack_window **w_cursor,
c4001d92 681 off_t offset,
03e79c88 682 unsigned int *left)
9bc879c1 683{
60bb8b14 684 struct pack_window *win = *w_cursor;
03e79c88 685
072db278
SP
686 if (p->pack_fd == -1 && open_packed_git(p))
687 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
688
689 /* Since packfiles end in a hash of their content and its
690 * pointless to ask for an offset into the middle of that
691 * hash, and the in_window function above wouldn't match
692 * don't allow an offset too close to the end of the file.
693 */
694 if (offset > (p->pack_size - 20))
695 die("offset beyond end of packfile (truncated pack?)");
696
697 if (!win || !in_window(win, offset)) {
698 if (win)
699 win->inuse_cnt--;
700 for (win = p->windows; win; win = win->next) {
701 if (in_window(win, offset))
702 break;
703 }
704 if (!win) {
78a28df9 705 size_t window_align = packed_git_window_size / 2;
dc49cd76 706 off_t len;
60bb8b14 707 win = xcalloc(1, sizeof(*win));
78a28df9 708 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
709 len = p->pack_size - win->offset;
710 if (len > packed_git_window_size)
711 len = packed_git_window_size;
712 win->len = (size_t)len;
60bb8b14 713 pack_mapped += win->len;
11daf39b
SP
714 while (packed_git_limit < pack_mapped
715 && unuse_one_window(p))
60bb8b14 716 ; /* nothing */
c4712e45 717 win->base = xmmap(NULL, win->len,
60bb8b14
SP
718 PROT_READ, MAP_PRIVATE,
719 p->pack_fd, win->offset);
720 if (win->base == MAP_FAILED)
73b4e4be
SP
721 die("packfile %s cannot be mapped: %s",
722 p->pack_name,
723 strerror(errno));
a53128b6
SP
724 pack_mmap_calls++;
725 pack_open_windows++;
726 if (pack_mapped > peak_pack_mapped)
727 peak_pack_mapped = pack_mapped;
728 if (pack_open_windows > peak_pack_open_windows)
729 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
730 win->next = p->windows;
731 p->windows = win;
732 }
1f688557 733 }
03e79c88
SP
734 if (win != *w_cursor) {
735 win->last_used = pack_used_ctr++;
736 win->inuse_cnt++;
737 *w_cursor = win;
738 }
60bb8b14 739 offset -= win->offset;
03e79c88 740 if (left)
dc49cd76 741 *left = win->len - xsize_t(offset);
03e79c88 742 return win->base + offset;
1f688557
JH
743}
744
42873078 745struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557
JH
746{
747 struct stat st;
42873078 748 struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
1f688557 749
42873078
NP
750 /*
751 * Make sure a corresponding .pack file exists and that
752 * the index looks sane.
753 */
754 path_len -= strlen(".idx");
755 if (path_len < 1)
1f688557 756 return NULL;
42873078
NP
757 memcpy(p->pack_name, path, path_len);
758 strcpy(p->pack_name + path_len, ".pack");
759 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode) ||
760 check_packed_git_idx(path, p)) {
761 free(p);
1f688557
JH
762 return NULL;
763 }
42873078 764
1f688557
JH
765 /* ok, it looks sane as far as we can check without
766 * actually mapping the pack file.
767 */
1f688557 768 p->pack_size = st.st_size;
1f688557 769 p->next = NULL;
c41ee586 770 p->windows = NULL;
9bc879c1 771 p->pack_fd = -1;
9d835df2 772 p->pack_local = local;
b867092f 773 p->mtime = st.st_mtime;
42873078
NP
774 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
775 hashclr(p->sha1);
1f688557
JH
776 return p;
777}
778
bf592c50 779struct packed_git *parse_pack_index(unsigned char *sha1)
c508df5e
DB
780{
781 char *path = sha1_pack_index_name(sha1);
782 return parse_pack_index_file(sha1, path);
783}
784
42873078
NP
785struct packed_git *parse_pack_index_file(const unsigned char *sha1,
786 const char *idx_path)
bf592c50 787{
42873078
NP
788 const char *path = sha1_pack_name(sha1);
789 struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
bf592c50 790
42873078
NP
791 if (check_packed_git_idx(idx_path, p)) {
792 free(p);
bf592c50 793 return NULL;
42873078 794 }
bf592c50 795
bf592c50 796 strcpy(p->pack_name, path);
bf592c50 797 p->pack_size = 0;
bf592c50 798 p->next = NULL;
c41ee586 799 p->windows = NULL;
9bc879c1 800 p->pack_fd = -1;
e702496e 801 hashcpy(p->sha1, sha1);
bf592c50
DB
802 return p;
803}
804
805void install_packed_git(struct packed_git *pack)
806{
807 pack->next = packed_git;
808 packed_git = pack;
809}
810
9d835df2 811static void prepare_packed_git_one(char *objdir, int local)
1f688557
JH
812{
813 char path[PATH_MAX];
814 int len;
815 DIR *dir;
816 struct dirent *de;
817
818 sprintf(path, "%s/pack", objdir);
819 len = strlen(path);
820 dir = opendir(path);
b5b16990 821 if (!dir) {
26125f6b 822 if (errno != ENOENT)
bd2afde8 823 error("unable to open object pack directory: %s: %s",
26125f6b 824 path, strerror(errno));
1f688557 825 return;
b5b16990 826 }
1f688557
JH
827 path[len++] = '/';
828 while ((de = readdir(dir)) != NULL) {
829 int namelen = strlen(de->d_name);
830 struct packed_git *p;
831
5bb1cda5 832 if (!has_extension(de->d_name, ".idx"))
1f688557
JH
833 continue;
834
54a15a8d 835 /* Don't reopen a pack we already have. */
1f688557 836 strcpy(path + len, de->d_name);
86f7780c
JK
837 for (p = packed_git; p; p = p->next) {
838 if (!memcmp(path, p->pack_name, len + namelen - 4))
839 break;
840 }
841 if (p)
842 continue;
54a15a8d
SP
843 /* See if it really is a valid .idx file with corresponding
844 * .pack file that we can map.
845 */
9d835df2 846 p = add_packed_git(path, len + namelen, local);
1f688557
JH
847 if (!p)
848 continue;
625e9421 849 install_packed_git(p);
1f688557 850 }
5b35bcd5 851 closedir(dir);
1f688557
JH
852}
853
b867092f
JH
854static int sort_pack(const void *a_, const void *b_)
855{
856 struct packed_git *a = *((struct packed_git **)a_);
857 struct packed_git *b = *((struct packed_git **)b_);
858 int st;
859
860 /*
861 * Local packs tend to contain objects specific to our
862 * variant of the project than remote ones. In addition,
863 * remote ones could be on a network mounted filesystem.
864 * Favor local ones for these reasons.
865 */
866 st = a->pack_local - b->pack_local;
867 if (st)
868 return -st;
869
870 /*
871 * Younger packs tend to contain more recent objects,
872 * and more recent objects tend to get accessed more
873 * often.
874 */
875 if (a->mtime < b->mtime)
876 return 1;
877 else if (a->mtime == b->mtime)
878 return 0;
879 return -1;
880}
881
882static void rearrange_packed_git(void)
883{
884 struct packed_git **ary, *p;
885 int i, n;
886
887 for (n = 0, p = packed_git; p; p = p->next)
888 n++;
889 if (n < 2)
890 return;
891
892 /* prepare an array of packed_git for easier sorting */
893 ary = xcalloc(n, sizeof(struct packed_git *));
894 for (n = 0, p = packed_git; p; p = p->next)
895 ary[n++] = p;
896
897 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
898
899 /* link them back again */
900 for (i = 0; i < n - 1; i++)
901 ary[i]->next = ary[i + 1];
902 ary[n - 1]->next = NULL;
903 packed_git = ary[0];
904
905 free(ary);
906}
907
637cdd9d 908static int prepare_packed_git_run_once = 0;
9a217f2a 909void prepare_packed_git(void)
1f688557 910{
d5a63b99 911 struct alternate_object_database *alt;
1f688557 912
637cdd9d 913 if (prepare_packed_git_run_once)
1f688557 914 return;
9d835df2 915 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 916 prepare_alt_odb();
d5a63b99 917 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 918 alt->name[-1] = 0;
9d835df2 919 prepare_packed_git_one(alt->base, 0);
1494e038 920 alt->name[-1] = '/';
1f688557 921 }
b867092f 922 rearrange_packed_git();
637cdd9d
JK
923 prepare_packed_git_run_once = 1;
924}
925
fc04c412 926void reprepare_packed_git(void)
637cdd9d
JK
927{
928 prepare_packed_git_run_once = 0;
929 prepare_packed_git();
1f688557
JH
930}
931
5d6ccf5c 932int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
0fcfd160
LT
933{
934 unsigned char real_sha1[20];
7cfb5f36 935 hash_sha1_file(map, size, type, real_sha1);
a89fccd2 936 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
937}
938
bb6b8e4f 939void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 940{
0fcfd160
LT
941 struct stat st;
942 void *map;
144bde78 943 int fd;
ace1534d
JH
944 char *filename = find_sha1_file(sha1, &st);
945
946 if (!filename) {
ace1534d
JH
947 return NULL;
948 }
0fcfd160 949
144bde78 950 fd = open(filename, O_RDONLY | sha1_file_open_flag);
0fcfd160 951 if (fd < 0) {
144bde78
LT
952 /* See if it works without O_NOATIME */
953 switch (sha1_file_open_flag) {
954 default:
955 fd = open(filename, O_RDONLY);
956 if (fd >= 0)
957 break;
958 /* Fallthrough */
959 case 0:
144bde78
LT
960 return NULL;
961 }
962
1f688557
JH
963 /* If it failed once, it will probably fail again.
964 * Stop using O_NOATIME
965 */
144bde78 966 sha1_file_open_flag = 0;
0fcfd160 967 }
dc49cd76
SP
968 *size = xsize_t(st.st_size);
969 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
0fcfd160 970 close(fd);
0fcfd160
LT
971 return map;
972}
973
bb6b8e4f
JH
974int legacy_loose_object(unsigned char *map)
975{
976 unsigned int word;
977
978 /*
979 * Is it a zlib-compressed buffer? If so, the first byte
980 * must be 0x78 (15-bit window size, deflated), and the
981 * first 16-bit word is evenly divisible by 31
982 */
983 word = (map[0] << 8) + map[1];
984 if (map[0] == 0x78 && !(word % 31))
985 return 1;
986 else
987 return 0;
988}
989
72518e9c 990unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 991{
ad1ed5ee 992 unsigned shift;
93821bd9 993 unsigned char c;
93821bd9 994 unsigned long size;
ad1ed5ee
JH
995 unsigned long used = 0;
996
997 c = buf[used++];
998 *type = (c >> 4) & 7;
999 size = c & 15;
1000 shift = 4;
1001 while (c & 0x80) {
1002 if (len <= used)
1003 return 0;
1004 if (sizeof(long) * 8 <= shift)
1005 return 0;
1006 c = buf[used++];
1007 size += (c & 0x7f) << shift;
1008 shift += 7;
1009 }
1010 *sizep = size;
1011 return used;
1012}
1013
1014static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1015{
1016 unsigned long size, used;
1017 static const char valid_loose_object_type[8] = {
1018 0, /* OBJ_EXT */
1019 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1020 0, /* "delta" and others are invalid in a loose object */
93821bd9 1021 };
ad1ed5ee 1022 enum object_type type;
93821bd9 1023
c4483576
LT
1024 /* Get the data stream */
1025 memset(stream, 0, sizeof(*stream));
1026 stream->next_in = map;
1027 stream->avail_in = mapsize;
1028 stream->next_out = buffer;
93821bd9
LT
1029 stream->avail_out = bufsiz;
1030
bb6b8e4f 1031 if (legacy_loose_object(map)) {
93821bd9
LT
1032 inflateInit(stream);
1033 return inflate(stream, 0);
1034 }
1035
ad1ed5ee
JH
1036 used = unpack_object_header_gently(map, mapsize, &type, &size);
1037 if (!used || !valid_loose_object_type[type])
93821bd9 1038 return -1;
ad1ed5ee
JH
1039 map += used;
1040 mapsize -= used;
93821bd9
LT
1041
1042 /* Set up the stream for the rest.. */
1043 stream->next_in = map;
1044 stream->avail_in = mapsize;
c4483576 1045 inflateInit(stream);
93821bd9
LT
1046
1047 /* And generate the fake traditional header */
ad1ed5ee 1048 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
df843662 1049 typename(type), size);
93821bd9 1050 return 0;
c4483576
LT
1051}
1052
7efbff75 1053static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1054{
1055 int bytes = strlen(buffer) + 1;
d565b341 1056 unsigned char *buf = xmalloc(1+size);
93821bd9 1057 unsigned long n;
7efbff75 1058 int status = Z_OK;
5180cacc 1059
93821bd9
LT
1060 n = stream->total_out - bytes;
1061 if (n > size)
1062 n = size;
1063 memcpy(buf, (char *) buffer + bytes, n);
1064 bytes = n;
456cdf6e
LT
1065 if (bytes <= size) {
1066 /*
1067 * The above condition must be (bytes <= size), not
1068 * (bytes < size). In other words, even though we
1069 * expect no more output and set avail_out to zer0,
1070 * the input zlib stream may have bytes that express
1071 * "this concludes the stream", and we *do* want to
1072 * eat that input.
1073 *
1074 * Otherwise we would not be able to test that we
1075 * consumed all the input to reach the expected size;
1076 * we also want to check that zlib tells us that all
1077 * went well with status == Z_STREAM_END at the end.
1078 */
5180cacc
LT
1079 stream->next_out = buf + bytes;
1080 stream->avail_out = size - bytes;
7efbff75
JH
1081 while (status == Z_OK)
1082 status = inflate(stream, Z_FINISH);
5180cacc
LT
1083 }
1084 buf[size] = 0;
456cdf6e 1085 if (status == Z_STREAM_END && !stream->avail_in) {
7efbff75
JH
1086 inflateEnd(stream);
1087 return buf;
1088 }
1089
1090 if (status < 0)
1091 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1092 else if (stream->avail_in)
1093 error("garbage at end of loose object '%s'",
1094 sha1_to_hex(sha1));
1095 free(buf);
1096 return NULL;
5180cacc
LT
1097}
1098
1099/*
1100 * We used to just use "sscanf()", but that's actually way
1101 * too permissive for what we want to check. So do an anal
1102 * object header parse by hand.
1103 */
21666f1a 1104static int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1105{
21666f1a 1106 char type[10];
5180cacc
LT
1107 int i;
1108 unsigned long size;
1109
1110 /*
1111 * The type can be at most ten bytes (including the
1112 * terminating '\0' that we add), and is followed by
21666f1a 1113 * a space.
5180cacc 1114 */
21666f1a 1115 i = 0;
5180cacc
LT
1116 for (;;) {
1117 char c = *hdr++;
1118 if (c == ' ')
1119 break;
21666f1a
NP
1120 type[i++] = c;
1121 if (i >= sizeof(type))
5180cacc 1122 return -1;
5180cacc 1123 }
21666f1a 1124 type[i] = 0;
5180cacc
LT
1125
1126 /*
1127 * The length must follow immediately, and be in canonical
1128 * decimal format (ie "010" is not valid).
1129 */
1130 size = *hdr++ - '0';
1131 if (size > 9)
1132 return -1;
1133 if (size) {
1134 for (;;) {
1135 unsigned long c = *hdr - '0';
1136 if (c > 9)
1137 break;
1138 hdr++;
1139 size = size * 10 + c;
1140 }
1141 }
1142 *sizep = size;
1143
1144 /*
1145 * The length must be followed by a zero byte
1146 */
21666f1a 1147 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1148}
1149
7efbff75 1150static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1151{
5180cacc 1152 int ret;
0fcfd160 1153 z_stream stream;
5180cacc 1154 char hdr[8192];
0fcfd160 1155
5180cacc 1156 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1157 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1158 return NULL;
1159
7efbff75 1160 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1161}
1162
c4001d92 1163static off_t get_delta_base(struct packed_git *p,
03e79c88 1164 struct pack_window **w_curs,
c4001d92 1165 off_t *curpos,
21666f1a 1166 enum object_type type,
c4001d92 1167 off_t delta_obj_offset)
eb32d236 1168{
2b87c45b 1169 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1170 off_t base_offset;
eb32d236 1171
8d8a4ea5
SP
1172 /* use_pack() assured us we have [base_info, base_info + 20)
1173 * as a range that we can look at without walking off the
1174 * end of the mapped window. Its actually the hash size
1175 * that is assured. An OFS_DELTA longer than the hash size
1176 * is stupid, as then a REF_DELTA would be smaller to store.
1177 */
21666f1a 1178 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1179 unsigned used = 0;
1180 unsigned char c = base_info[used++];
1181 base_offset = c & 127;
1182 while (c & 128) {
1183 base_offset += 1;
8723f216 1184 if (!base_offset || MSB(base_offset, 7))
eb32d236
NP
1185 die("offset value overflow for delta base object");
1186 c = base_info[used++];
1187 base_offset = (base_offset << 7) + (c & 127);
1188 }
1189 base_offset = delta_obj_offset - base_offset;
1190 if (base_offset >= delta_obj_offset)
1191 die("delta base offset out of bound");
2b87c45b 1192 *curpos += used;
21666f1a 1193 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1194 /* The base entry _must_ be in the same pack */
1195 base_offset = find_pack_entry_one(base_info, p);
1196 if (!base_offset)
1197 die("failed to find delta-pack base object %s",
1198 sha1_to_hex(base_info));
2b87c45b 1199 *curpos += 20;
eb32d236
NP
1200 } else
1201 die("I am totally screwed");
2b87c45b 1202 return base_offset;
eb32d236
NP
1203}
1204
f3bf9224 1205/* forward declaration for a mutually recursive function */
c4001d92 1206static int packed_object_info(struct packed_git *p, off_t offset,
21666f1a 1207 unsigned long *sizep);
f3bf9224 1208
43057304 1209static int packed_delta_info(struct packed_git *p,
03e79c88 1210 struct pack_window **w_curs,
c4001d92 1211 off_t curpos,
21666f1a 1212 enum object_type type,
c4001d92 1213 off_t obj_offset,
43057304 1214 unsigned long *sizep)
5db47c2b 1215{
c4001d92 1216 off_t base_offset;
f3bf9224 1217
21666f1a
NP
1218 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1219 type = packed_object_info(p, base_offset, NULL);
f3bf9224 1220
c62266f3
JH
1221 /* We choose to only get the type of the base object and
1222 * ignore potentially corrupt pack file that expects the delta
1223 * based on a base with a wrong size. This saves tons of
1224 * inflate() calls.
1225 */
c62266f3
JH
1226 if (sizep) {
1227 const unsigned char *data;
079afb18 1228 unsigned char delta_head[20], *in;
c62266f3
JH
1229 z_stream stream;
1230 int st;
1231
1232 memset(&stream, 0, sizeof(stream));
c62266f3
JH
1233 stream.next_out = delta_head;
1234 stream.avail_out = sizeof(delta_head);
1235
1236 inflateInit(&stream);
079afb18 1237 do {
2b87c45b 1238 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18
SP
1239 stream.next_in = in;
1240 st = inflate(&stream, Z_FINISH);
2b87c45b 1241 curpos += stream.next_in - in;
079afb18
SP
1242 } while ((st == Z_OK || st == Z_BUF_ERROR)
1243 && stream.total_out < sizeof(delta_head));
c62266f3
JH
1244 inflateEnd(&stream);
1245 if ((st != Z_STREAM_END) &&
1246 stream.total_out != sizeof(delta_head))
1247 die("delta data unpack-initial failed");
1248
1249 /* Examine the initial part of the delta to figure out
1250 * the result size.
1251 */
1252 data = delta_head;
8960844a
NP
1253
1254 /* ignore base size */
1255 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
5db47c2b 1256
c62266f3 1257 /* Read the result size */
2b87c45b 1258 *sizep = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
c62266f3 1259 }
21666f1a
NP
1260
1261 return type;
5db47c2b
JH
1262}
1263
2b87c45b
NP
1264static int unpack_object_header(struct packed_git *p,
1265 struct pack_window **w_curs,
c4001d92 1266 off_t *curpos,
2b87c45b 1267 unsigned long *sizep)
a733cb60 1268{
03e79c88
SP
1269 unsigned char *base;
1270 unsigned int left;
ad1ed5ee 1271 unsigned long used;
2b87c45b 1272 enum object_type type;
a733cb60 1273
8d8a4ea5
SP
1274 /* use_pack() assures us we have [base, base + 20) available
1275 * as a range that we can look at at. (Its actually the hash
3dff5379 1276 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1277 * the maximum deflated object size is 2^137, which is just
1278 * insane, so we know won't exceed what we have been given.
1279 */
2b87c45b
NP
1280 base = use_pack(p, w_curs, *curpos, &left);
1281 used = unpack_object_header_gently(base, left, &type, sizep);
ad1ed5ee
JH
1282 if (!used)
1283 die("object offset outside of pack file");
2b87c45b 1284 *curpos += used;
ad1ed5ee 1285
2b87c45b 1286 return type;
a733cb60
LT
1287}
1288
21666f1a 1289const char *packed_object_info_detail(struct packed_git *p,
c4001d92 1290 off_t obj_offset,
21666f1a
NP
1291 unsigned long *size,
1292 unsigned long *store_size,
1293 unsigned int *delta_chain_length,
1294 unsigned char *base_sha1)
ad8c80a5 1295{
03e79c88 1296 struct pack_window *w_curs = NULL;
c4001d92
SP
1297 off_t curpos;
1298 unsigned long dummy;
43057304 1299 unsigned char *next_sha1;
21666f1a 1300 enum object_type type;
ad8c80a5 1301
43057304 1302 *delta_chain_length = 0;
2b87c45b 1303 curpos = obj_offset;
21666f1a 1304 type = unpack_object_header(p, &w_curs, &curpos, size);
43057304
NP
1305
1306 for (;;) {
21666f1a 1307 switch (type) {
43057304 1308 default:
08a19d87 1309 die("pack %s contains unknown object type %d",
21666f1a 1310 p->pack_name, type);
43057304
NP
1311 case OBJ_COMMIT:
1312 case OBJ_TREE:
1313 case OBJ_BLOB:
1314 case OBJ_TAG:
43057304 1315 *store_size = 0; /* notyet */
03e79c88 1316 unuse_pack(&w_curs);
21666f1a 1317 return typename(type);
eb32d236 1318 case OBJ_OFS_DELTA:
21666f1a 1319 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
eb32d236 1320 if (*delta_chain_length == 0) {
2b87c45b 1321 /* TODO: find base_sha1 as pointed by curpos */
30fee062 1322 hashclr(base_sha1);
eb32d236
NP
1323 }
1324 break;
1325 case OBJ_REF_DELTA:
2b87c45b 1326 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
43057304
NP
1327 if (*delta_chain_length == 0)
1328 hashcpy(base_sha1, next_sha1);
2b87c45b 1329 obj_offset = find_pack_entry_one(next_sha1, p);
43057304
NP
1330 break;
1331 }
43057304 1332 (*delta_chain_length)++;
2b87c45b 1333 curpos = obj_offset;
21666f1a 1334 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
ad8c80a5 1335 }
ad8c80a5
JH
1336}
1337
c4001d92 1338static int packed_object_info(struct packed_git *p, off_t obj_offset,
21666f1a 1339 unsigned long *sizep)
1f688557 1340{
03e79c88 1341 struct pack_window *w_curs = NULL;
c4001d92
SP
1342 unsigned long size;
1343 off_t curpos = obj_offset;
21666f1a 1344 enum object_type type;
5db47c2b 1345
21666f1a 1346 type = unpack_object_header(p, &w_curs, &curpos, &size);
5db47c2b 1347
21666f1a 1348 switch (type) {
eb32d236
NP
1349 case OBJ_OFS_DELTA:
1350 case OBJ_REF_DELTA:
21666f1a
NP
1351 type = packed_delta_info(p, &w_curs, curpos,
1352 type, obj_offset, sizep);
2b87c45b 1353 break;
a733cb60 1354 case OBJ_COMMIT:
a733cb60 1355 case OBJ_TREE:
a733cb60 1356 case OBJ_BLOB:
a733cb60 1357 case OBJ_TAG:
2b87c45b
NP
1358 if (sizep)
1359 *sizep = size;
a69d0943 1360 break;
1f688557 1361 default:
08a19d87 1362 die("pack %s contains unknown object type %d",
21666f1a 1363 p->pack_name, type);
1f688557 1364 }
2b87c45b 1365 unuse_pack(&w_curs);
21666f1a 1366 return type;
1f688557
JH
1367}
1368
eb950c19 1369static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1370 struct pack_window **w_curs,
c4001d92 1371 off_t curpos,
eb950c19 1372 unsigned long size)
de530aaa
SP
1373{
1374 int st;
1375 z_stream stream;
079afb18 1376 unsigned char *buffer, *in;
de530aaa
SP
1377
1378 buffer = xmalloc(size + 1);
1379 buffer[size] = 0;
1380 memset(&stream, 0, sizeof(stream));
de530aaa
SP
1381 stream.next_out = buffer;
1382 stream.avail_out = size;
1383
1384 inflateInit(&stream);
079afb18 1385 do {
2b87c45b 1386 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18
SP
1387 stream.next_in = in;
1388 st = inflate(&stream, Z_FINISH);
2b87c45b 1389 curpos += stream.next_in - in;
079afb18 1390 } while (st == Z_OK || st == Z_BUF_ERROR);
de530aaa
SP
1391 inflateEnd(&stream);
1392 if ((st != Z_STREAM_END) || stream.total_out != size) {
1393 free(buffer);
1394 return NULL;
1395 }
1396
1397 return buffer;
1398}
1399
e5e01619
LT
1400#define MAX_DELTA_CACHE (256)
1401
18bdec11 1402static size_t delta_base_cached;
5e08ecbf
NP
1403
1404static struct delta_base_cache_lru_list {
1405 struct delta_base_cache_lru_list *prev;
1406 struct delta_base_cache_lru_list *next;
1407} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1408
e5e01619 1409static struct delta_base_cache_entry {
5e08ecbf
NP
1410 struct delta_base_cache_lru_list lru;
1411 void *data;
e5e01619
LT
1412 struct packed_git *p;
1413 off_t base_offset;
1414 unsigned long size;
e5e01619
LT
1415 enum object_type type;
1416} delta_base_cache[MAX_DELTA_CACHE];
1417
1418static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1419{
1420 unsigned long hash;
1421
1422 hash = (unsigned long)p + (unsigned long)base_offset;
1423 hash += (hash >> 8) + (hash >> 16);
3358004a 1424 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1425}
1426
62f255ad 1427static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 1428 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 1429{
e5e01619
LT
1430 void *ret;
1431 unsigned long hash = pack_entry_hash(p, base_offset);
1432 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1433
1434 ret = ent->data;
1435 if (ret && ent->p == p && ent->base_offset == base_offset)
1436 goto found_cache_entry;
62f255ad 1437 return unpack_entry(p, base_offset, type, base_size);
e5e01619
LT
1438
1439found_cache_entry:
18bdec11 1440 if (!keep_cache) {
a0cba108 1441 ent->data = NULL;
5e08ecbf
NP
1442 ent->lru.next->prev = ent->lru.prev;
1443 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
1444 delta_base_cached -= ent->size;
1445 }
a0cba108
NP
1446 else {
1447 ret = xmalloc(ent->size + 1);
1448 memcpy(ret, ent->data, ent->size);
1449 ((char *)ret)[ent->size] = 0;
1450 }
e5e01619
LT
1451 *type = ent->type;
1452 *base_size = ent->size;
1453 return ret;
62f255ad
LT
1454}
1455
18bdec11
SP
1456static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1457{
1458 if (ent->data) {
1459 free(ent->data);
1460 ent->data = NULL;
5e08ecbf
NP
1461 ent->lru.next->prev = ent->lru.prev;
1462 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
1463 delta_base_cached -= ent->size;
1464 }
1465}
1466
62f255ad
LT
1467static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1468 void *base, unsigned long base_size, enum object_type type)
1469{
5e08ecbf 1470 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 1471 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 1472 struct delta_base_cache_lru_list *lru;
e5e01619 1473
18bdec11
SP
1474 release_delta_base_cache(ent);
1475 delta_base_cached += base_size;
5e08ecbf
NP
1476
1477 for (lru = delta_base_cache_lru.next;
1478 delta_base_cached > delta_base_cache_limit
1479 && lru != &delta_base_cache_lru;
1480 lru = lru->next) {
1481 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
1482 if (f->type == OBJ_BLOB)
1483 release_delta_base_cache(f);
1484 }
5e08ecbf
NP
1485 for (lru = delta_base_cache_lru.next;
1486 delta_base_cached > delta_base_cache_limit
1487 && lru != &delta_base_cache_lru;
1488 lru = lru->next) {
1489 struct delta_base_cache_entry *f = (void *)lru;
1490 release_delta_base_cache(f);
1491 }
18bdec11 1492
e5e01619
LT
1493 ent->p = p;
1494 ent->base_offset = base_offset;
1495 ent->type = type;
1496 ent->data = base;
1497 ent->size = base_size;
5e08ecbf
NP
1498 ent->lru.next = &delta_base_cache_lru;
1499 ent->lru.prev = delta_base_cache_lru.prev;
1500 delta_base_cache_lru.prev->next = &ent->lru;
1501 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
1502}
1503
eb950c19 1504static void *unpack_delta_entry(struct packed_git *p,
03e79c88 1505 struct pack_window **w_curs,
c4001d92 1506 off_t curpos,
1f688557 1507 unsigned long delta_size,
c4001d92 1508 off_t obj_offset,
21666f1a 1509 enum object_type *type,
eb950c19 1510 unsigned long *sizep)
1f688557 1511{
7c3e8be3 1512 void *delta_data, *result, *base;
c4001d92
SP
1513 unsigned long base_size;
1514 off_t base_offset;
43057304 1515
21666f1a 1516 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
a0cba108 1517 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
67686d95 1518 if (!base)
c4001d92
SP
1519 die("failed to read delta base object"
1520 " at %"PRIuMAX" from %s",
1521 (uintmax_t)base_offset, p->pack_name);
67686d95 1522
2b87c45b 1523 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1f688557
JH
1524 result = patch_delta(base, base_size,
1525 delta_data, delta_size,
21666f1a 1526 sizep);
1f688557
JH
1527 if (!result)
1528 die("failed to apply delta");
1529 free(delta_data);
62f255ad 1530 add_delta_base_cache(p, base_offset, base, base_size, *type);
1f688557
JH
1531 return result;
1532}
1533
c4001d92 1534void *unpack_entry(struct packed_git *p, off_t obj_offset,
21666f1a 1535 enum object_type *type, unsigned long *sizep)
1f688557 1536{
03e79c88 1537 struct pack_window *w_curs = NULL;
c4001d92 1538 off_t curpos = obj_offset;
21666f1a 1539 void *data;
1f688557 1540
21666f1a
NP
1541 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1542 switch (*type) {
eb32d236
NP
1543 case OBJ_OFS_DELTA:
1544 case OBJ_REF_DELTA:
21666f1a
NP
1545 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1546 obj_offset, type, sizep);
4d703a1a 1547 break;
a733cb60 1548 case OBJ_COMMIT:
a733cb60 1549 case OBJ_TREE:
a733cb60 1550 case OBJ_BLOB:
a733cb60 1551 case OBJ_TAG:
21666f1a 1552 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
4d703a1a 1553 break;
1f688557 1554 default:
21666f1a 1555 die("unknown object type %i in %s", *type, p->pack_name);
1f688557 1556 }
03e79c88 1557 unuse_pack(&w_curs);
21666f1a 1558 return data;
1f688557
JH
1559}
1560
d72308e0
NP
1561const unsigned char *nth_packed_object_sha1(const struct packed_git *p,
1562 uint32_t n)
9a217f2a 1563{
42873078 1564 const unsigned char *index = p->index_data;
57059091 1565 if (n >= p->num_objects)
d72308e0 1566 return NULL;
74e34e1f
NP
1567 index += 4 * 256;
1568 if (p->index_version == 1) {
1569 return index + 24 * n + 4;
1570 } else {
1571 index += 8;
1572 return index + 20 * n;
1573 }
1574}
1575
1576static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1577{
1578 const unsigned char *index = p->index_data;
1579 index += 4 * 256;
1580 if (p->index_version == 1) {
1581 return ntohl(*((uint32_t *)(index + 24 * n)));
1582 } else {
1583 uint32_t off;
1584 index += 8 + p->num_objects * (20 + 4);
1585 off = ntohl(*((uint32_t *)(index + 4 * n)));
1586 if (!(off & 0x80000000))
1587 return off;
1588 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1589 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1590 ntohl(*((uint32_t *)(index + 4)));
1591 }
9a217f2a
JH
1592}
1593
c4001d92 1594off_t find_pack_entry_one(const unsigned char *sha1,
43057304 1595 struct packed_git *p)
1f688557 1596{
42873078 1597 const uint32_t *level1_ofs = p->index_data;
42873078 1598 const unsigned char *index = p->index_data;
74e34e1f 1599 unsigned hi, lo;
42873078 1600
74e34e1f
NP
1601 if (p->index_version > 1) {
1602 level1_ofs += 2;
1603 index += 8;
1604 }
42873078 1605 index += 4 * 256;
74e34e1f
NP
1606 hi = ntohl(level1_ofs[*sha1]);
1607 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1f688557
JH
1608
1609 do {
74e34e1f
NP
1610 unsigned mi = (lo + hi) / 2;
1611 unsigned x = (p->index_version > 1) ? (mi * 20) : (mi * 24 + 4);
1612 int cmp = hashcmp(index + x, sha1);
43057304 1613 if (!cmp)
74e34e1f 1614 return nth_packed_object_offset(p, mi);
1f688557
JH
1615 if (cmp > 0)
1616 hi = mi;
1617 else
1618 lo = mi+1;
1619 } while (lo < hi);
1620 return 0;
1621}
1622
d4ff6d92
SP
1623static int matches_pack_name(struct packed_git *p, const char *ig)
1624{
1625 const char *last_c, *c;
1626
1627 if (!strcmp(p->pack_name, ig))
1628 return 0;
1629
1630 for (c = p->pack_name, last_c = c; *c;)
1631 if (*c == '/')
1632 last_c = ++c;
1633 else
1634 ++c;
1635 if (!strcmp(last_c, ig))
1636 return 0;
1637
1638 return 1;
1639}
1640
106d710b 1641static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1f688557
JH
1642{
1643 struct packed_git *p;
c4001d92 1644 off_t offset;
43057304 1645
1f688557
JH
1646 prepare_packed_git();
1647
1648 for (p = packed_git; p; p = p->next) {
106d710b
JH
1649 if (ignore_packed) {
1650 const char **ig;
1651 for (ig = ignore_packed; *ig; ig++)
d4ff6d92 1652 if (!matches_pack_name(p, *ig))
106d710b
JH
1653 break;
1654 if (*ig)
1655 continue;
1656 }
43057304
NP
1657 offset = find_pack_entry_one(sha1, p);
1658 if (offset) {
c715f783
SP
1659 /*
1660 * We are about to tell the caller where they can
1661 * locate the requested object. We better make
1662 * sure the packfile is still here and can be
1663 * accessed before supplying that answer, as
1664 * it may have been deleted since the index
1665 * was loaded!
1666 */
1667 if (p->pack_fd == -1 && open_packed_git(p)) {
1668 error("packfile %s cannot be accessed", p->pack_name);
1669 continue;
1670 }
43057304
NP
1671 e->offset = offset;
1672 e->p = p;
1673 hashcpy(e->sha1, sha1);
1f688557 1674 return 1;
43057304 1675 }
1f688557
JH
1676 }
1677 return 0;
1678}
1679
bf592c50
DB
1680struct packed_git *find_sha1_pack(const unsigned char *sha1,
1681 struct packed_git *packs)
1682{
1683 struct packed_git *p;
bf592c50
DB
1684
1685 for (p = packs; p; p = p->next) {
43057304 1686 if (find_pack_entry_one(sha1, p))
bf592c50
DB
1687 return p;
1688 }
1689 return NULL;
21666f1a 1690
bf592c50
DB
1691}
1692
21666f1a 1693static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
65c2e0c3 1694{
36e4d74a 1695 int status;
65c2e0c3
JH
1696 unsigned long mapsize, size;
1697 void *map;
1698 z_stream stream;
d65a16f6 1699 char hdr[32];
65c2e0c3 1700
bb6b8e4f 1701 map = map_sha1_file(sha1, &mapsize);
f0df4ed5
JS
1702 if (!map)
1703 return error("unable to find %s", sha1_to_hex(sha1));
36e4d74a
JH
1704 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1705 status = error("unable to unpack %s header",
1706 sha1_to_hex(sha1));
21666f1a 1707 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 1708 status = error("unable to parse %s header", sha1_to_hex(sha1));
21666f1a
NP
1709 else if (sizep)
1710 *sizep = size;
65c2e0c3
JH
1711 inflateEnd(&stream);
1712 munmap(map, mapsize);
1713 return status;
1714}
1715
21666f1a 1716int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
f0df4ed5 1717{
f0df4ed5
JS
1718 struct pack_entry e;
1719
1720 if (!find_pack_entry(sha1, &e, NULL)) {
1721 reprepare_packed_git();
1722 if (!find_pack_entry(sha1, &e, NULL))
21666f1a 1723 return sha1_loose_object_info(sha1, sizep);
f0df4ed5 1724 }
21666f1a 1725 return packed_object_info(e.p, e.offset, sizep);
f0df4ed5
JS
1726}
1727
21666f1a
NP
1728static void *read_packed_sha1(const unsigned char *sha1,
1729 enum object_type *type, unsigned long *size)
1f688557
JH
1730{
1731 struct pack_entry e;
1732
8276c007 1733 if (!find_pack_entry(sha1, &e, NULL))
1f688557 1734 return NULL;
8276c007 1735 else
a0cba108 1736 return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1f688557
JH
1737}
1738
d66b37bb
JH
1739/*
1740 * This is meant to hold a *small* number of objects that you would
1741 * want read_sha1_file() to be able to return, but yet you do not want
1742 * to write them into the object store (e.g. a browse-only
1743 * application).
1744 */
1745static struct cached_object {
1746 unsigned char sha1[20];
21666f1a 1747 enum object_type type;
d66b37bb
JH
1748 void *buf;
1749 unsigned long size;
1750} *cached_objects;
1751static int cached_object_nr, cached_object_alloc;
1752
1753static struct cached_object *find_cached_object(const unsigned char *sha1)
1754{
1755 int i;
1756 struct cached_object *co = cached_objects;
1757
1758 for (i = 0; i < cached_object_nr; i++, co++) {
1759 if (!hashcmp(co->sha1, sha1))
1760 return co;
1761 }
1762 return NULL;
1763}
1764
21666f1a
NP
1765int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1766 unsigned char *sha1)
d66b37bb
JH
1767{
1768 struct cached_object *co;
1769
21666f1a 1770 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
1771 if (has_sha1_file(sha1) || find_cached_object(sha1))
1772 return 0;
1773 if (cached_object_alloc <= cached_object_nr) {
1774 cached_object_alloc = alloc_nr(cached_object_alloc);
1775 cached_objects = xrealloc(cached_objects,
1776 sizeof(*cached_objects) *
1777 cached_object_alloc);
1778 }
1779 co = &cached_objects[cached_object_nr++];
1780 co->size = len;
21666f1a 1781 co->type = type;
efa13f7b
JH
1782 co->buf = xmalloc(len);
1783 memcpy(co->buf, buf, len);
d66b37bb
JH
1784 hashcpy(co->sha1, sha1);
1785 return 0;
1786}
1787
21666f1a
NP
1788void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1789 unsigned long *size)
0fcfd160
LT
1790{
1791 unsigned long mapsize;
1792 void *map, *buf;
d66b37bb
JH
1793 struct cached_object *co;
1794
1795 co = find_cached_object(sha1);
1796 if (co) {
1797 buf = xmalloc(co->size + 1);
1798 memcpy(buf, co->buf, co->size);
1799 ((char*)buf)[co->size] = 0;
21666f1a 1800 *type = co->type;
d66b37bb
JH
1801 *size = co->size;
1802 return buf;
1803 }
0fcfd160 1804
8276c007
PE
1805 buf = read_packed_sha1(sha1, type, size);
1806 if (buf)
1807 return buf;
bb6b8e4f 1808 map = map_sha1_file(sha1, &mapsize);
0fcfd160 1809 if (map) {
7efbff75 1810 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
1811 munmap(map, mapsize);
1812 return buf;
1813 }
637cdd9d 1814 reprepare_packed_git();
8276c007 1815 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
1816}
1817
40469ee9 1818void *read_object_with_reference(const unsigned char *sha1,
21666f1a 1819 const char *required_type_name,
40469ee9
JH
1820 unsigned long *size,
1821 unsigned char *actual_sha1_return)
f4913f91 1822{
21666f1a 1823 enum object_type type, required_type;
f4913f91
JH
1824 void *buffer;
1825 unsigned long isize;
40469ee9 1826 unsigned char actual_sha1[20];
f4913f91 1827
21666f1a 1828 required_type = type_from_string(required_type_name);
e702496e 1829 hashcpy(actual_sha1, sha1);
40469ee9
JH
1830 while (1) {
1831 int ref_length = -1;
1832 const char *ref_type = NULL;
f4913f91 1833
21666f1a 1834 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
1835 if (!buffer)
1836 return NULL;
21666f1a 1837 if (type == required_type) {
40469ee9
JH
1838 *size = isize;
1839 if (actual_sha1_return)
e702496e 1840 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
1841 return buffer;
1842 }
1843 /* Handle references */
21666f1a 1844 else if (type == OBJ_COMMIT)
40469ee9 1845 ref_type = "tree ";
21666f1a 1846 else if (type == OBJ_TAG)
40469ee9
JH
1847 ref_type = "object ";
1848 else {
1849 free(buffer);
1850 return NULL;
1851 }
1852 ref_length = strlen(ref_type);
f4913f91 1853
40469ee9 1854 if (memcmp(buffer, ref_type, ref_length) ||
1d7f171c 1855 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
1856 free(buffer);
1857 return NULL;
1858 }
1cf58e72 1859 free(buffer);
40469ee9
JH
1860 /* Now we have the ID of the referred-to object in
1861 * actual_sha1. Check again. */
f4913f91 1862 }
f4913f91
JH
1863}
1864
ce9fbf16 1865static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 1866 const char *type, unsigned char *sha1,
d65a16f6 1867 char *hdr, int *hdrlen)
d410c0f5
JH
1868{
1869 SHA_CTX c;
1870
1871 /* Generate the header */
d65a16f6 1872 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
1873
1874 /* Sha1.. */
1875 SHA1_Init(&c);
1876 SHA1_Update(&c, hdr, *hdrlen);
1877 SHA1_Update(&c, buf, len);
1878 SHA1_Final(sha1, &c);
d410c0f5
JH
1879}
1880
230f1322
LT
1881/*
1882 * Link the tempfile to the final place, possibly creating the
1883 * last directory level as you do so.
1884 *
1885 * Returns the errno on failure, 0 on success.
1886 */
839837b9 1887static int link_temp_to_file(const char *tmpfile, const char *filename)
230f1322
LT
1888{
1889 int ret;
756aaf4a 1890 char *dir;
230f1322
LT
1891
1892 if (!link(tmpfile, filename))
1893 return 0;
1894
1895 /*
756aaf4a 1896 * Try to mkdir the last path component if that failed.
230f1322
LT
1897 *
1898 * Re-try the "link()" regardless of whether the mkdir
1899 * succeeds, since a race might mean that somebody
1900 * else succeeded.
1901 */
1902 ret = errno;
756aaf4a
SP
1903 dir = strrchr(filename, '/');
1904 if (dir) {
1905 *dir = 0;
866cae0d 1906 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
91c23e48 1907 *dir = '/';
756aaf4a 1908 return -2;
91c23e48 1909 }
756aaf4a
SP
1910 *dir = '/';
1911 if (!link(tmpfile, filename))
1912 return 0;
1913 ret = errno;
230f1322
LT
1914 }
1915 return ret;
1916}
1917
1918/*
1919 * Move the just written object into its final resting place
1920 */
839837b9 1921int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322
LT
1922{
1923 int ret = link_temp_to_file(tmpfile, filename);
7ebb6fca
LT
1924
1925 /*
1926 * Coda hack - coda doesn't like cross-directory links,
1927 * so we fall back to a rename, which will mean that it
1928 * won't be able to check collisions, but that's not a
1929 * big deal.
1930 *
1931 * The same holds for FAT formatted media.
1932 *
1933 * When this succeeds, we just return 0. We have nothing
1934 * left to unlink.
1935 */
1936 if (ret && ret != EEXIST) {
1937 if (!rename(tmpfile, filename))
230f1322 1938 return 0;
9e48b389 1939 ret = errno;
230f1322
LT
1940 }
1941 unlink(tmpfile);
1942 if (ret) {
1943 if (ret != EEXIST) {
916d081b 1944 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
230f1322
LT
1945 }
1946 /* FIXME!!! Collision check here ? */
1947 }
1948
1949 return 0;
1950}
1951
4d548150
LT
1952static int write_buffer(int fd, const void *buf, size_t len)
1953{
d34cf19b 1954 if (write_in_full(fd, buf, len) < 0)
93822c22 1955 return error("file write error (%s)", strerror(errno));
4d548150
LT
1956 return 0;
1957}
1958
93821bd9
LT
1959static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1960{
1961 int hdr_len;
1962 unsigned char c;
1963
1964 c = (type << 4) | (len & 15);
1965 len >>= 4;
1966 hdr_len = 1;
1967 while (len) {
1968 *hdr++ = c | 0x80;
1969 hdr_len++;
1970 c = (len & 0x7f);
1971 len >>= 7;
1972 }
1973 *hdr = c;
1974 return hdr_len;
1975}
1976
1977static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1978{
d65a16f6 1979 int obj_type, hdrlen;
93821bd9
LT
1980
1981 if (use_legacy_headers) {
1982 while (deflate(stream, 0) == Z_OK)
1983 /* nothing */;
1984 return;
1985 }
df843662 1986 obj_type = type_from_string(type);
d65a16f6
NP
1987 hdrlen = write_binary_header(stream->next_out, obj_type, len);
1988 stream->total_out = hdrlen;
1989 stream->next_out += hdrlen;
1990 stream->avail_out -= hdrlen;
93821bd9
LT
1991}
1992
ce9fbf16 1993int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
1994 unsigned char *sha1)
1995{
d65a16f6 1996 char hdr[32];
abdc3fc8
RS
1997 int hdrlen;
1998 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1999 return 0;
2000}
2001
bf0f910d 2002int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
0fcfd160 2003{
ac54c277 2004 int size, ret;
bf0f910d 2005 unsigned char *compressed;
0fcfd160
LT
2006 z_stream stream;
2007 unsigned char sha1[20];
706bc531 2008 char *filename;
aac17941 2009 static char tmpfile[PATH_MAX];
d65a16f6 2010 char hdr[32];
230f1322 2011 int fd, hdrlen;
a44c9a5e 2012
d410c0f5
JH
2013 /* Normally if we have it in the pack then we do not bother writing
2014 * it out into .git/objects/??/?{38} file.
2015 */
972a9155
RS
2016 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2017 filename = sha1_file_name(sha1);
706bc531 2018 if (returnsha1)
e702496e 2019 hashcpy(returnsha1, sha1);
d410c0f5
JH
2020 if (has_sha1_file(sha1))
2021 return 0;
aac17941
LT
2022 fd = open(filename, O_RDONLY);
2023 if (fd >= 0) {
706bc531 2024 /*
aac17941
LT
2025 * FIXME!!! We might do collision checking here, but we'd
2026 * need to uncompress the old file and check it. Later.
706bc531 2027 */
aac17941 2028 close(fd);
706bc531
LT
2029 return 0;
2030 }
2031
aac17941 2032 if (errno != ENOENT) {
916d081b 2033 return error("sha1 file %s: %s\n", filename, strerror(errno));
aac17941
LT
2034 }
2035
0e55181f 2036 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
ace1534d 2037
aac17941
LT
2038 fd = mkstemp(tmpfile);
2039 if (fd < 0) {
916d081b
PB
2040 if (errno == EPERM)
2041 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2042 else
2043 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
aac17941
LT
2044 }
2045
0fcfd160
LT
2046 /* Set it up */
2047 memset(&stream, 0, sizeof(stream));
12f6c308 2048 deflateInit(&stream, zlib_compression_level);
93821bd9 2049 size = 8 + deflateBound(&stream, len+hdrlen);
812666c8 2050 compressed = xmalloc(size);
0fcfd160
LT
2051
2052 /* Compress it */
0fcfd160
LT
2053 stream.next_out = compressed;
2054 stream.avail_out = size;
a44c9a5e
LT
2055
2056 /* First header.. */
d65a16f6 2057 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2058 stream.avail_in = hdrlen;
93821bd9 2059 setup_object_header(&stream, type, len);
a44c9a5e
LT
2060
2061 /* Then the data itself.. */
2062 stream.next_in = buf;
2063 stream.avail_in = len;
ac54c277
LT
2064 ret = deflate(&stream, Z_FINISH);
2065 if (ret != Z_STREAM_END)
2066 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2067
2068 ret = deflateEnd(&stream);
2069 if (ret != Z_OK)
2070 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2071
0fcfd160
LT
2072 size = stream.total_out;
2073
4d548150
LT
2074 if (write_buffer(fd, compressed, size) < 0)
2075 die("unable to write sha1 file");
aac17941 2076 fchmod(fd, 0444);
e82973cf
JH
2077 if (close(fd))
2078 die("unable to write sha1 file");
383f85b7 2079 free(compressed);
0fcfd160 2080
230f1322 2081 return move_temp_to_file(tmpfile, filename);
0fcfd160 2082}
8237b185 2083
4d548150
LT
2084/*
2085 * We need to unpack and recompress the object for writing
2086 * it out to a different file.
2087 */
2088static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
a5eda52b 2089{
4d548150 2090 size_t size;
a5eda52b 2091 z_stream stream;
4d548150
LT
2092 unsigned char *unpacked;
2093 unsigned long len;
21666f1a 2094 enum object_type type;
d65a16f6 2095 char hdr[32];
4d548150
LT
2096 int hdrlen;
2097 void *buf;
bfc66daf 2098
a9486b02 2099 /* need to unpack and recompress it by itself */
21666f1a 2100 unpacked = read_packed_sha1(sha1, &type, &len);
8276c007
PE
2101 if (!unpacked)
2102 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
a5eda52b 2103
21666f1a 2104 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
a5eda52b 2105
4d548150
LT
2106 /* Set it up */
2107 memset(&stream, 0, sizeof(stream));
12f6c308 2108 deflateInit(&stream, zlib_compression_level);
4d548150
LT
2109 size = deflateBound(&stream, len + hdrlen);
2110 buf = xmalloc(size);
a5eda52b 2111
4d548150
LT
2112 /* Compress it */
2113 stream.next_out = buf;
2114 stream.avail_out = size;
a5eda52b 2115
4d548150
LT
2116 /* First header.. */
2117 stream.next_in = (void *)hdr;
2118 stream.avail_in = hdrlen;
2119 while (deflate(&stream, 0) == Z_OK)
2120 /* nothing */;
bfc66daf 2121
4d548150
LT
2122 /* Then the data itself.. */
2123 stream.next_in = unpacked;
2124 stream.avail_in = len;
2125 while (deflate(&stream, Z_FINISH) == Z_OK)
2126 /* nothing */;
2127 deflateEnd(&stream);
2128 free(unpacked);
bfc66daf 2129
4d548150
LT
2130 *objsize = stream.total_out;
2131 return buf;
2132}
2133
2134int write_sha1_to_fd(int fd, const unsigned char *sha1)
2135{
2136 int retval;
2137 unsigned long objsize;
bb6b8e4f 2138 void *buf = map_sha1_file(sha1, &objsize);
4d548150
LT
2139
2140 if (buf) {
2141 retval = write_buffer(fd, buf, objsize);
2142 munmap(buf, objsize);
2143 return retval;
2144 }
2145
2146 buf = repack_object(sha1, &objsize);
2147 retval = write_buffer(fd, buf, objsize);
2148 free(buf);
2149 return retval;
a5eda52b
DB
2150}
2151
70b9829e
DB
2152int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2153 size_t bufsize, size_t *bufposn)
8237b185 2154{
230f1322 2155 char tmpfile[PATH_MAX];
8237b185
DB
2156 int local;
2157 z_stream stream;
2158 unsigned char real_sha1[20];
bf0f910d 2159 unsigned char discard[4096];
8237b185
DB
2160 int ret;
2161 SHA_CTX c;
2162
0e55181f 2163 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
8237b185 2164
230f1322 2165 local = mkstemp(tmpfile);
916d081b
PB
2166 if (local < 0) {
2167 if (errno == EPERM)
2168 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2169 else
2170 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2171 }
8237b185
DB
2172
2173 memset(&stream, 0, sizeof(stream));
2174
2175 inflateInit(&stream);
2176
2177 SHA1_Init(&c);
2178
2179 do {
2180 ssize_t size;
70b9829e
DB
2181 if (*bufposn) {
2182 stream.avail_in = *bufposn;
96ad15ae 2183 stream.next_in = (unsigned char *) buffer;
70b9829e
DB
2184 do {
2185 stream.next_out = discard;
2186 stream.avail_out = sizeof(discard);
2187 ret = inflate(&stream, Z_SYNC_FLUSH);
2188 SHA1_Update(&c, discard, sizeof(discard) -
2189 stream.avail_out);
2190 } while (stream.avail_in && ret == Z_OK);
4d548150
LT
2191 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2192 die("unable to write sha1 file");
70b9829e
DB
2193 memmove(buffer, buffer + *bufposn - stream.avail_in,
2194 stream.avail_in);
2195 *bufposn = stream.avail_in;
2196 if (ret != Z_OK)
2197 break;
2198 }
93d26e4c 2199 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
8237b185
DB
2200 if (size <= 0) {
2201 close(local);
230f1322 2202 unlink(tmpfile);
8237b185
DB
2203 if (!size)
2204 return error("Connection closed?");
2205 perror("Reading from connection");
2206 return -1;
2207 }
70b9829e
DB
2208 *bufposn += size;
2209 } while (1);
8237b185
DB
2210 inflateEnd(&stream);
2211
b5b8d814 2212 fchmod(local, 0444);
0d315468
JM
2213 if (close(local) != 0)
2214 die("unable to write sha1 file");
8237b185
DB
2215 SHA1_Final(real_sha1, &c);
2216 if (ret != Z_STREAM_END) {
230f1322 2217 unlink(tmpfile);
8237b185
DB
2218 return error("File %s corrupted", sha1_to_hex(sha1));
2219 }
a89fccd2 2220 if (hashcmp(sha1, real_sha1)) {
230f1322 2221 unlink(tmpfile);
bd2afde8 2222 return error("File %s has bad hash", sha1_to_hex(sha1));
8237b185 2223 }
230f1322
LT
2224
2225 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
8237b185
DB
2226}
2227
bf592c50
DB
2228int has_pack_index(const unsigned char *sha1)
2229{
2230 struct stat st;
2231 if (stat(sha1_pack_index_name(sha1), &st))
2232 return 0;
2233 return 1;
2234}
2235
2236int has_pack_file(const unsigned char *sha1)
2237{
2238 struct stat st;
2239 if (stat(sha1_pack_name(sha1), &st))
2240 return 0;
2241 return 1;
2242}
2243
106d710b 2244int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
dade09c2
LT
2245{
2246 struct pack_entry e;
106d710b 2247 return find_pack_entry(sha1, &e, ignore_packed);
dade09c2
LT
2248}
2249
8237b185
DB
2250int has_sha1_file(const unsigned char *sha1)
2251{
8237b185 2252 struct stat st;
1f688557
JH
2253 struct pack_entry e;
2254
106d710b 2255 if (find_pack_entry(sha1, &e, NULL))
1f688557 2256 return 1;
ab90ea5d 2257 return find_sha1_file(sha1, &st) ? 1 : 0;
8237b185 2258}
74400e71 2259
e7332f96
BE
2260/*
2261 * reads from fd as long as possible into a supplied buffer of size bytes.
82e5a82f 2262 * If necessary the buffer's size is increased using realloc()
e7332f96
BE
2263 *
2264 * returns 0 if anything went fine and -1 otherwise
2265 *
2266 * NOTE: both buf and size may change, but even when -1 is returned
2267 * you still have to free() it yourself.
2268 */
2269int read_pipe(int fd, char** return_buf, unsigned long* return_size)
024510c8 2270{
e7332f96
BE
2271 char* buf = *return_buf;
2272 unsigned long size = *return_size;
2273 int iret;
024510c8 2274 unsigned long off = 0;
e7332f96 2275
024510c8 2276 do {
e7332f96 2277 iret = xread(fd, buf + off, size - off);
024510c8
DB
2278 if (iret > 0) {
2279 off += iret;
2280 if (off == size) {
2281 size *= 2;
83572c1a 2282 buf = xrealloc(buf, size);
024510c8
DB
2283 }
2284 }
2285 } while (iret > 0);
e7332f96
BE
2286
2287 *return_buf = buf;
2288 *return_size = off;
2289
2290 if (iret < 0)
2291 return -1;
2292 return 0;
2293}
2294
2295int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2296{
2297 unsigned long size = 4096;
2d7320d0 2298 char *buf = xmalloc(size);
e7332f96 2299 int ret;
e7332f96
BE
2300
2301 if (read_pipe(fd, &buf, &size)) {
024510c8
DB
2302 free(buf);
2303 return -1;
2304 }
e7332f96 2305
024510c8 2306 if (!type)
8e440259 2307 type = blob_type;
024510c8 2308 if (write_object)
e7332f96 2309 ret = write_sha1_file(buf, size, type, sha1);
abdc3fc8
RS
2310 else
2311 ret = hash_sha1_file(buf, size, type, sha1);
024510c8
DB
2312 free(buf);
2313 return ret;
2314}
2315
edaec3fb 2316int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
53bca91a 2317 enum object_type type, const char *path)
74400e71 2318{
dc49cd76 2319 size_t size = xsize_t(st->st_size);
3a55602e 2320 void *buf = NULL;
6c510bee 2321 int ret, re_allocated = 0;
74400e71 2322
74400e71 2323 if (size)
c4712e45 2324 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
74400e71 2325 close(fd);
74400e71 2326
7672db20 2327 if (!type)
edaec3fb 2328 type = OBJ_BLOB;
6c510bee
LT
2329
2330 /*
2331 * Convert blobs to git internal format
2332 */
d0d8e14d 2333 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
6c510bee
LT
2334 unsigned long nsize = size;
2335 char *nbuf = buf;
53bca91a 2336 if (convert_to_git(path, &nbuf, &nsize)) {
6c510bee
LT
2337 if (size)
2338 munmap(buf, size);
2339 size = nsize;
2340 buf = nbuf;
2341 re_allocated = 1;
2342 }
2343 }
2344
7672db20 2345 if (write_object)
edaec3fb 2346 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 2347 else
edaec3fb 2348 ret = hash_sha1_file(buf, size, typename(type), sha1);
6c510bee
LT
2349 if (re_allocated) {
2350 free(buf);
2351 return ret;
2352 }
aac17941
LT
2353 if (size)
2354 munmap(buf, size);
2355 return ret;
74400e71 2356}
ec1fcc16
JH
2357
2358int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2359{
2360 int fd;
2361 char *target;
dc49cd76 2362 size_t len;
ec1fcc16
JH
2363
2364 switch (st->st_mode & S_IFMT) {
2365 case S_IFREG:
2366 fd = open(path, O_RDONLY);
2367 if (fd < 0)
2368 return error("open(\"%s\"): %s", path,
2369 strerror(errno));
53bca91a 2370 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
ec1fcc16
JH
2371 return error("%s: failed to insert into database",
2372 path);
2373 break;
2374 case S_IFLNK:
dc49cd76
SP
2375 len = xsize_t(st->st_size);
2376 target = xmalloc(len + 1);
2377 if (readlink(path, target, len + 1) != st->st_size) {
ec1fcc16
JH
2378 char *errstr = strerror(errno);
2379 free(target);
2380 return error("readlink(\"%s\"): %s", path,
2381 errstr);
2382 }
abdc3fc8 2383 if (!write_object)
dc49cd76
SP
2384 hash_sha1_file(target, len, blob_type, sha1);
2385 else if (write_sha1_file(target, len, blob_type, sha1))
ec1fcc16
JH
2386 return error("%s: failed to insert into database",
2387 path);
2388 free(target);
2389 break;
2390 default:
2391 return error("%s: unsupported file type", path);
2392 }
2393 return 0;
2394}
a69e5429
JH
2395
2396int read_pack_header(int fd, struct pack_header *header)
2397{
2398 char *c = (char*)header;
2399 ssize_t remaining = sizeof(struct pack_header);
2400 do {
2401 ssize_t r = xread(fd, c, remaining);
2402 if (r <= 0)
2403 /* "eof before pack header was fully read" */
2404 return PH_ERROR_EOF;
2405 remaining -= r;
2406 c += r;
2407 } while (remaining > 0);
2408 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2409 /* "protocol error (pack signature mismatch detected)" */
2410 return PH_ERROR_PACK_SIGNATURE;
2411 if (!pack_version_ok(header->hdr_version))
2412 /* "protocol error (pack version unsupported)" */
2413 return PH_ERROR_PROTOCOL;
2414 return 0;
2415}