Makefile: Use generic rule to build test programs
[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"
f35a6d3b 16#include "refs.h"
0fcfd160 17
144bde78
LT
18#ifndef O_NOATIME
19#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
20#define O_NOATIME 01000000
21#else
22#define O_NOATIME 0
23#endif
24#endif
25
e05db0fd
PR
26#ifdef NO_C99_FORMAT
27#define SZ_FMT "lu"
28#else
29#define SZ_FMT "zu"
30#endif
31
96f1e58f 32const unsigned char null_sha1[20];
88cd621d 33
144bde78
LT
34static unsigned int sha1_file_open_flag = O_NOATIME;
35
e49521b5
JH
36signed char hexval_table[256] = {
37 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
38 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
43 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
44 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
45 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
47 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
48 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
49 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
69};
0fcfd160
LT
70
71int get_sha1_hex(const char *hex, unsigned char *sha1)
72{
73 int i;
74 for (i = 0; i < 20; i++) {
75 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
76 if (val & ~0xff)
77 return -1;
78 *sha1++ = val;
79 hex += 2;
80 }
81 return 0;
82}
83
b2cb9425
JH
84int safe_create_leading_directories(char *path)
85{
86 char *pos = path;
67d42212
JR
87 struct stat st;
88
67ffdf4c
JS
89 if (*pos == '/')
90 pos++;
b2cb9425
JH
91
92 while (pos) {
93 pos = strchr(pos, '/');
94 if (!pos)
95 break;
96 *pos = 0;
67d42212
JR
97 if (!stat(path, &st)) {
98 /* path exists */
99 if (!S_ISDIR(st.st_mode)) {
b2cb9425 100 *pos = '/';
67d42212 101 return -3;
b2cb9425 102 }
457f06d6 103 }
67d42212
JR
104 else if (mkdir(path, 0777)) {
105 *pos = '/';
106 return -1;
107 }
457f06d6
JS
108 else if (adjust_shared_perm(path)) {
109 *pos = '/';
110 return -2;
111 }
b2cb9425
JH
112 *pos++ = '/';
113 }
114 return 0;
115}
723c31fe 116
0fcfd160
LT
117char * sha1_to_hex(const unsigned char *sha1)
118{
dcb3450f
LT
119 static int bufno;
120 static char hexbuffer[4][50];
0fcfd160 121 static const char hex[] = "0123456789abcdef";
dcb3450f 122 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
0fcfd160
LT
123 int i;
124
125 for (i = 0; i < 20; i++) {
126 unsigned int val = *sha1++;
127 *buf++ = hex[val >> 4];
128 *buf++ = hex[val & 0xf];
129 }
1e80e044
JS
130 *buf = '\0';
131
0fcfd160
LT
132 return buffer;
133}
134
ace1534d
JH
135static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
136{
137 int i;
138 for (i = 0; i < 20; i++) {
139 static char hex[] = "0123456789abcdef";
140 unsigned int val = sha1[i];
141 char *pos = pathbuf + i*2 + (i > 0);
142 *pos++ = hex[val >> 4];
143 *pos = hex[val & 0xf];
144 }
145}
146
0fcfd160
LT
147/*
148 * NOTE! This returns a statically allocated buffer, so you have to be
9befac47 149 * careful about using it. Do a "xstrdup()" if you need to save the
0fcfd160 150 * filename.
ace1534d
JH
151 *
152 * Also note that this returns the location for creating. Reading
153 * SHA1 file can happen from any alternate directory listed in the
d19938ab 154 * DB_ENVIRONMENT environment variable if it is not found in
ace1534d 155 * the primary object database.
0fcfd160
LT
156 */
157char *sha1_file_name(const unsigned char *sha1)
158{
0fcfd160
LT
159 static char *name, *base;
160
161 if (!base) {
d19938ab 162 const char *sha1_file_directory = get_object_directory();
0fcfd160 163 int len = strlen(sha1_file_directory);
812666c8 164 base = xmalloc(len + 60);
0fcfd160
LT
165 memcpy(base, sha1_file_directory, len);
166 memset(base+len, 0, 60);
167 base[len] = '/';
168 base[len+3] = '/';
169 name = base + len + 1;
170 }
ace1534d 171 fill_sha1_path(name, sha1);
0fcfd160
LT
172 return base;
173}
174
bf592c50
DB
175char *sha1_pack_name(const unsigned char *sha1)
176{
177 static const char hex[] = "0123456789abcdef";
178 static char *name, *base, *buf;
179 int i;
180
181 if (!base) {
182 const char *sha1_file_directory = get_object_directory();
183 int len = strlen(sha1_file_directory);
184 base = xmalloc(len + 60);
185 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
186 name = base + len + 11;
187 }
188
189 buf = name;
190
191 for (i = 0; i < 20; i++) {
192 unsigned int val = *sha1++;
193 *buf++ = hex[val >> 4];
194 *buf++ = hex[val & 0xf];
195 }
196
197 return base;
198}
199
200char *sha1_pack_index_name(const unsigned char *sha1)
201{
202 static const char hex[] = "0123456789abcdef";
203 static char *name, *base, *buf;
204 int i;
205
206 if (!base) {
207 const char *sha1_file_directory = get_object_directory();
208 int len = strlen(sha1_file_directory);
209 base = xmalloc(len + 60);
210 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
211 name = base + len + 11;
212 }
213
214 buf = name;
215
216 for (i = 0; i < 20; i++) {
217 unsigned int val = *sha1++;
218 *buf++ = hex[val >> 4];
219 *buf++ = hex[val & 0xf];
220 }
221
222 return base;
223}
224
d5a63b99
JH
225struct alternate_object_database *alt_odb_list;
226static struct alternate_object_database **alt_odb_tail;
ace1534d 227
c2f493a4
MW
228static void read_info_alternates(const char * alternates, int depth);
229
ddd5d056
JH
230/*
231 * Prepare alternate object database registry.
d5a63b99
JH
232 *
233 * The variable alt_odb_list points at the list of struct
234 * alternate_object_database. The elements on this list come from
235 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
236 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
237 * whose contents is similar to that environment variable but can be
238 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
239 * contains "/the/directory/corresponding/to/.git/objects/...", while
240 * its name points just after the slash at the end of ".git/objects/"
241 * in the example above, and has enough space to hold 40-byte hex
242 * SHA1, an extra slash for the first level indirection, and the
243 * terminating NUL.
ddd5d056 244 */
c2f493a4 245static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
ace1534d 246{
c2f493a4 247 struct stat st;
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
c2f493a4
MW
256 if (*entry != '/' && relative_base) {
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
265 if (*entry != '/' && relative_base) {
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 */
278 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
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) {
c2f493a4
MW
336 if ((*last != '/') && depth) {
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;
c2f493a4
MW
355 char path[PATH_MAX];
356 int fd;
d5a63b99 357
c2f493a4 358 sprintf(path, "%s/info/alternates", relative_base);
d5a63b99
JH
359 fd = open(path, O_RDONLY);
360 if (fd < 0)
361 return;
362 if (fstat(fd, &st) || (st.st_size == 0)) {
363 close(fd);
9a217f2a 364 return;
ace1534d 365 }
dc49cd76
SP
366 mapsz = xsize_t(st.st_size);
367 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 368 close(fd);
d5a63b99 369
dc49cd76 370 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
c2f493a4 371
dc49cd76 372 munmap(map, mapsz);
ace1534d
JH
373}
374
c2f493a4
MW
375void prepare_alt_odb(void)
376{
554fe20d 377 const char *alt;
c2f493a4
MW
378
379 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
380 if (!alt) alt = "";
381
382 if (alt_odb_tail)
383 return;
384 alt_odb_tail = &alt_odb_list;
385 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
386
387 read_info_alternates(get_object_directory(), 0);
388}
389
ace1534d
JH
390static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
391{
ace1534d 392 char *name = sha1_file_name(sha1);
d5a63b99 393 struct alternate_object_database *alt;
ace1534d
JH
394
395 if (!stat(name, st))
396 return name;
9a217f2a 397 prepare_alt_odb();
d5a63b99
JH
398 for (alt = alt_odb_list; alt; alt = alt->next) {
399 name = alt->name;
ace1534d 400 fill_sha1_path(name, sha1);
d5a63b99
JH
401 if (!stat(alt->base, st))
402 return alt->base;
ace1534d
JH
403 }
404 return NULL;
405}
406
60bb8b14 407static unsigned int pack_used_ctr;
a53128b6
SP
408static unsigned int pack_mmap_calls;
409static unsigned int peak_pack_open_windows;
410static unsigned int pack_open_windows;
411static size_t peak_pack_mapped;
60bb8b14 412static size_t pack_mapped;
9a217f2a 413struct packed_git *packed_git;
1f688557 414
a53128b6
SP
415void pack_report()
416{
417 fprintf(stderr,
e05db0fd
PR
418 "pack_report: getpagesize() = %10" SZ_FMT "\n"
419 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
420 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
78a28df9 421 (size_t) getpagesize(),
a53128b6
SP
422 packed_git_window_size,
423 packed_git_limit);
424 fprintf(stderr,
425 "pack_report: pack_used_ctr = %10u\n"
426 "pack_report: pack_mmap_calls = %10u\n"
427 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
428 "pack_report: pack_mapped = "
429 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
430 pack_used_ctr,
431 pack_mmap_calls,
432 pack_open_windows, peak_pack_open_windows,
433 pack_mapped, peak_pack_mapped);
434}
435
42873078 436static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
437{
438 void *idx_map;
42873078 439 struct pack_idx_header *hdr;
dc49cd76 440 size_t idx_size;
74e34e1f 441 uint32_t version, nr, i, *index;
e93ec6f9 442 int fd = open(path, O_RDONLY);
1f688557 443 struct stat st;
42873078 444
1f688557
JH
445 if (fd < 0)
446 return -1;
447 if (fstat(fd, &st)) {
448 close(fd);
449 return -1;
450 }
dc49cd76 451 idx_size = xsize_t(st.st_size);
2d88451b
SP
452 if (idx_size < 4 * 256 + 20 + 20) {
453 close(fd);
454 return error("index file %s is too small", path);
455 }
c4712e45 456 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 457 close(fd);
1f688557 458
42873078
NP
459 hdr = idx_map;
460 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
461 version = ntohl(hdr->idx_version);
462 if (version < 2 || version > 2) {
463 munmap(idx_map, idx_size);
464 return error("index file %s is version %d"
465 " and is not supported by this binary"
466 " (try upgrading GIT to a newer version)",
467 path, version);
468 }
469 } else
470 version = 1;
df1b059d 471
1f688557 472 nr = 0;
42873078 473 index = idx_map;
74e34e1f
NP
474 if (version > 1)
475 index += 2; /* skip index header */
1f688557 476 for (i = 0; i < 256; i++) {
326bf396 477 uint32_t n = ntohl(index[i]);
2d88451b
SP
478 if (n < nr) {
479 munmap(idx_map, idx_size);
df1b059d 480 return error("non-monotonic index %s", path);
2d88451b 481 }
1f688557
JH
482 nr = n;
483 }
484
74e34e1f
NP
485 if (version == 1) {
486 /*
487 * Total size:
488 * - 256 index entries 4 bytes each
489 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
490 * - 20-byte SHA1 of the packfile
491 * - 20-byte SHA1 file checksum
492 */
493 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
494 munmap(idx_map, idx_size);
495 return error("wrong index file size in %s", path);
496 }
497 } else if (version == 2) {
498 /*
499 * Minimum size:
500 * - 8 bytes of header
501 * - 256 index entries 4 bytes each
502 * - 20-byte sha1 entry * nr
503 * - 4-byte crc entry * nr
504 * - 4-byte offset entry * nr
505 * - 20-byte SHA1 of the packfile
506 * - 20-byte SHA1 file checksum
507 * And after the 4-byte offset table might be a
508 * variable sized table containing 8-byte entries
509 * for offsets larger than 2^31.
510 */
511 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
512 if (idx_size < min_size || idx_size > min_size + (nr - 1)*8) {
513 munmap(idx_map, idx_size);
514 return error("wrong index file size in %s", path);
515 }
516 if (idx_size != min_size) {
517 /* make sure we can deal with large pack offsets */
518 off_t x = 0x7fffffffUL, y = 0xffffffffUL;
519 if (x > (x + 1) || y > (y + 1)) {
520 munmap(idx_map, idx_size);
521 return error("pack too large for current definition of off_t in %s", path);
522 }
523 }
2d88451b 524 }
1f688557 525
74e34e1f 526 p->index_version = version;
42873078
NP
527 p->index_data = idx_map;
528 p->index_size = idx_size;
57059091 529 p->num_objects = nr;
1f688557
JH
530 return 0;
531}
532
11daf39b
SP
533static void scan_windows(struct packed_git *p,
534 struct packed_git **lru_p,
535 struct pack_window **lru_w,
536 struct pack_window **lru_l)
1f688557 537{
11daf39b
SP
538 struct pack_window *w, *w_l;
539
540 for (w_l = NULL, w = p->windows; w; w = w->next) {
541 if (!w->inuse_cnt) {
542 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
543 *lru_p = p;
544 *lru_w = w;
545 *lru_l = w_l;
54044bf8 546 }
54044bf8 547 }
11daf39b 548 w_l = w;
f9253394 549 }
11daf39b
SP
550}
551
d1efefa4 552static int unuse_one_window(struct packed_git *current, int keep_fd)
11daf39b
SP
553{
554 struct packed_git *p, *lru_p = NULL;
555 struct pack_window *lru_w = NULL, *lru_l = NULL;
556
557 if (current)
558 scan_windows(current, &lru_p, &lru_w, &lru_l);
559 for (p = packed_git; p; p = p->next)
560 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
561 if (lru_p) {
562 munmap(lru_w->base, lru_w->len);
563 pack_mapped -= lru_w->len;
564 if (lru_l)
565 lru_l->next = lru_w->next;
566 else {
567 lru_p->windows = lru_w->next;
d1efefa4 568 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
54044bf8
SP
569 close(lru_p->pack_fd);
570 lru_p->pack_fd = -1;
571 }
572 }
573 free(lru_w);
a53128b6 574 pack_open_windows--;
54044bf8
SP
575 return 1;
576 }
577 return 0;
f9253394
JH
578}
579
d1efefa4 580void release_pack_memory(size_t need, int fd)
97bfeb34
SP
581{
582 size_t cur = pack_mapped;
d1efefa4 583 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
97bfeb34
SP
584 ; /* nothing */
585}
586
03e79c88 587void unuse_pack(struct pack_window **w_cursor)
f9253394 588{
03e79c88
SP
589 struct pack_window *w = *w_cursor;
590 if (w) {
591 w->inuse_cnt--;
592 *w_cursor = NULL;
593 }
1f688557
JH
594}
595
3cf8b462
SP
596/*
597 * Do not call this directly as this leaks p->pack_fd on error return;
598 * call open_packed_git() instead.
599 */
600static int open_packed_git_1(struct packed_git *p)
1f688557 601{
9bc879c1
SP
602 struct stat st;
603 struct pack_header hdr;
604 unsigned char sha1[20];
605 unsigned char *idx_sha1;
2c039da8 606 long fd_flag;
9bc879c1
SP
607
608 p->pack_fd = open(p->pack_name, O_RDONLY);
609 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 610 return -1;
9bc879c1
SP
611
612 /* If we created the struct before we had the pack we lack size. */
bf592c50 613 if (!p->pack_size) {
bf592c50 614 if (!S_ISREG(st.st_mode))
072db278 615 return error("packfile %s not a regular file", p->pack_name);
bf592c50 616 p->pack_size = st.st_size;
9bc879c1 617 } else if (p->pack_size != st.st_size)
072db278 618 return error("packfile %s size changed", p->pack_name);
9bc879c1 619
2c039da8
JH
620 /* We leave these file descriptors open with sliding mmap;
621 * there is no point keeping them open across exec(), though.
622 */
623 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
624 if (fd_flag < 0)
072db278 625 return error("cannot determine file descriptor flags");
2c039da8
JH
626 fd_flag |= FD_CLOEXEC;
627 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 628 return error("cannot set FD_CLOEXEC");
2c039da8 629
9bc879c1 630 /* Verify we recognize this pack file format. */
e6e2bd62 631 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 632 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 633 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 634 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 635 if (!pack_version_ok(hdr.hdr_version))
072db278 636 return error("packfile %s is version %u and not supported"
9bc879c1
SP
637 " (try upgrading GIT to a newer version)",
638 p->pack_name, ntohl(hdr.hdr_version));
639
640 /* Verify the pack matches its index. */
57059091 641 if (p->num_objects != ntohl(hdr.hdr_entries))
072db278 642 return error("packfile %s claims to have %u objects"
57059091
NP
643 " while index indicates %u objects",
644 p->pack_name, ntohl(hdr.hdr_entries),
645 p->num_objects);
9bc879c1 646 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 647 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 648 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 649 return error("packfile %s signature is unavailable", p->pack_name);
42873078 650 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 651 if (hashcmp(sha1, idx_sha1))
072db278
SP
652 return error("packfile %s does not match index", p->pack_name);
653 return 0;
9bc879c1
SP
654}
655
3cf8b462
SP
656static int open_packed_git(struct packed_git *p)
657{
658 if (!open_packed_git_1(p))
659 return 0;
660 if (p->pack_fd != -1) {
661 close(p->pack_fd);
662 p->pack_fd = -1;
663 }
664 return -1;
665}
666
c4001d92 667static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
668{
669 /* We must promise at least 20 bytes (one hash) after the
670 * offset is available from this window, otherwise the offset
671 * is not actually in this window and a different window (which
672 * has that one hash excess) must be used. This is to support
673 * the object header and delta base parsing routines below.
674 */
675 off_t win_off = win->offset;
676 return win_off <= offset
677 && (offset + 20) <= (win_off + win->len);
678}
679
03e79c88
SP
680unsigned char* use_pack(struct packed_git *p,
681 struct pack_window **w_cursor,
c4001d92 682 off_t offset,
03e79c88 683 unsigned int *left)
9bc879c1 684{
60bb8b14 685 struct pack_window *win = *w_cursor;
03e79c88 686
072db278
SP
687 if (p->pack_fd == -1 && open_packed_git(p))
688 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
689
690 /* Since packfiles end in a hash of their content and its
691 * pointless to ask for an offset into the middle of that
692 * hash, and the in_window function above wouldn't match
693 * don't allow an offset too close to the end of the file.
694 */
695 if (offset > (p->pack_size - 20))
696 die("offset beyond end of packfile (truncated pack?)");
697
698 if (!win || !in_window(win, offset)) {
699 if (win)
700 win->inuse_cnt--;
701 for (win = p->windows; win; win = win->next) {
702 if (in_window(win, offset))
703 break;
704 }
705 if (!win) {
78a28df9 706 size_t window_align = packed_git_window_size / 2;
dc49cd76 707 off_t len;
60bb8b14 708 win = xcalloc(1, sizeof(*win));
78a28df9 709 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
710 len = p->pack_size - win->offset;
711 if (len > packed_git_window_size)
712 len = packed_git_window_size;
713 win->len = (size_t)len;
60bb8b14 714 pack_mapped += win->len;
11daf39b 715 while (packed_git_limit < pack_mapped
d1efefa4 716 && unuse_one_window(p, p->pack_fd))
60bb8b14 717 ; /* nothing */
c4712e45 718 win->base = xmmap(NULL, win->len,
60bb8b14
SP
719 PROT_READ, MAP_PRIVATE,
720 p->pack_fd, win->offset);
721 if (win->base == MAP_FAILED)
73b4e4be
SP
722 die("packfile %s cannot be mapped: %s",
723 p->pack_name,
724 strerror(errno));
a53128b6
SP
725 pack_mmap_calls++;
726 pack_open_windows++;
727 if (pack_mapped > peak_pack_mapped)
728 peak_pack_mapped = pack_mapped;
729 if (pack_open_windows > peak_pack_open_windows)
730 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
731 win->next = p->windows;
732 p->windows = win;
733 }
1f688557 734 }
03e79c88
SP
735 if (win != *w_cursor) {
736 win->last_used = pack_used_ctr++;
737 win->inuse_cnt++;
738 *w_cursor = win;
739 }
60bb8b14 740 offset -= win->offset;
03e79c88 741 if (left)
dc49cd76 742 *left = win->len - xsize_t(offset);
03e79c88 743 return win->base + offset;
1f688557
JH
744}
745
42873078 746struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557
JH
747{
748 struct stat st;
42873078 749 struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
1f688557 750
42873078
NP
751 /*
752 * Make sure a corresponding .pack file exists and that
753 * the index looks sane.
754 */
755 path_len -= strlen(".idx");
756 if (path_len < 1)
1f688557 757 return NULL;
42873078
NP
758 memcpy(p->pack_name, path, path_len);
759 strcpy(p->pack_name + path_len, ".pack");
760 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode) ||
761 check_packed_git_idx(path, p)) {
762 free(p);
1f688557
JH
763 return NULL;
764 }
42873078 765
1f688557
JH
766 /* ok, it looks sane as far as we can check without
767 * actually mapping the pack file.
768 */
1f688557 769 p->pack_size = st.st_size;
1f688557 770 p->next = NULL;
c41ee586 771 p->windows = NULL;
9bc879c1 772 p->pack_fd = -1;
9d835df2 773 p->pack_local = local;
b867092f 774 p->mtime = st.st_mtime;
42873078
NP
775 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
776 hashclr(p->sha1);
1f688557
JH
777 return p;
778}
779
bf592c50 780struct packed_git *parse_pack_index(unsigned char *sha1)
c508df5e
DB
781{
782 char *path = sha1_pack_index_name(sha1);
783 return parse_pack_index_file(sha1, path);
784}
785
42873078
NP
786struct packed_git *parse_pack_index_file(const unsigned char *sha1,
787 const char *idx_path)
bf592c50 788{
42873078
NP
789 const char *path = sha1_pack_name(sha1);
790 struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
bf592c50 791
42873078
NP
792 if (check_packed_git_idx(idx_path, p)) {
793 free(p);
bf592c50 794 return NULL;
42873078 795 }
bf592c50 796
bf592c50 797 strcpy(p->pack_name, path);
bf592c50 798 p->pack_size = 0;
bf592c50 799 p->next = NULL;
c41ee586 800 p->windows = NULL;
9bc879c1 801 p->pack_fd = -1;
e702496e 802 hashcpy(p->sha1, sha1);
bf592c50
DB
803 return p;
804}
805
806void install_packed_git(struct packed_git *pack)
807{
808 pack->next = packed_git;
809 packed_git = pack;
810}
811
9d835df2 812static void prepare_packed_git_one(char *objdir, int local)
1f688557
JH
813{
814 char path[PATH_MAX];
815 int len;
816 DIR *dir;
817 struct dirent *de;
818
819 sprintf(path, "%s/pack", objdir);
820 len = strlen(path);
821 dir = opendir(path);
b5b16990 822 if (!dir) {
26125f6b 823 if (errno != ENOENT)
bd2afde8 824 error("unable to open object pack directory: %s: %s",
26125f6b 825 path, strerror(errno));
1f688557 826 return;
b5b16990 827 }
1f688557
JH
828 path[len++] = '/';
829 while ((de = readdir(dir)) != NULL) {
830 int namelen = strlen(de->d_name);
831 struct packed_git *p;
832
5bb1cda5 833 if (!has_extension(de->d_name, ".idx"))
1f688557
JH
834 continue;
835
54a15a8d 836 /* Don't reopen a pack we already have. */
1f688557 837 strcpy(path + len, de->d_name);
86f7780c
JK
838 for (p = packed_git; p; p = p->next) {
839 if (!memcmp(path, p->pack_name, len + namelen - 4))
840 break;
841 }
842 if (p)
843 continue;
54a15a8d
SP
844 /* See if it really is a valid .idx file with corresponding
845 * .pack file that we can map.
846 */
9d835df2 847 p = add_packed_git(path, len + namelen, local);
1f688557
JH
848 if (!p)
849 continue;
625e9421 850 install_packed_git(p);
1f688557 851 }
5b35bcd5 852 closedir(dir);
1f688557
JH
853}
854
b867092f
JH
855static int sort_pack(const void *a_, const void *b_)
856{
857 struct packed_git *a = *((struct packed_git **)a_);
858 struct packed_git *b = *((struct packed_git **)b_);
859 int st;
860
861 /*
862 * Local packs tend to contain objects specific to our
863 * variant of the project than remote ones. In addition,
864 * remote ones could be on a network mounted filesystem.
865 * Favor local ones for these reasons.
866 */
867 st = a->pack_local - b->pack_local;
868 if (st)
869 return -st;
870
871 /*
872 * Younger packs tend to contain more recent objects,
873 * and more recent objects tend to get accessed more
874 * often.
875 */
876 if (a->mtime < b->mtime)
877 return 1;
878 else if (a->mtime == b->mtime)
879 return 0;
880 return -1;
881}
882
883static void rearrange_packed_git(void)
884{
885 struct packed_git **ary, *p;
886 int i, n;
887
888 for (n = 0, p = packed_git; p; p = p->next)
889 n++;
890 if (n < 2)
891 return;
892
893 /* prepare an array of packed_git for easier sorting */
894 ary = xcalloc(n, sizeof(struct packed_git *));
895 for (n = 0, p = packed_git; p; p = p->next)
896 ary[n++] = p;
897
898 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
899
900 /* link them back again */
901 for (i = 0; i < n - 1; i++)
902 ary[i]->next = ary[i + 1];
903 ary[n - 1]->next = NULL;
904 packed_git = ary[0];
905
906 free(ary);
907}
908
637cdd9d 909static int prepare_packed_git_run_once = 0;
9a217f2a 910void prepare_packed_git(void)
1f688557 911{
d5a63b99 912 struct alternate_object_database *alt;
1f688557 913
637cdd9d 914 if (prepare_packed_git_run_once)
1f688557 915 return;
9d835df2 916 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 917 prepare_alt_odb();
d5a63b99 918 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 919 alt->name[-1] = 0;
9d835df2 920 prepare_packed_git_one(alt->base, 0);
1494e038 921 alt->name[-1] = '/';
1f688557 922 }
b867092f 923 rearrange_packed_git();
637cdd9d
JK
924 prepare_packed_git_run_once = 1;
925}
926
fc04c412 927void reprepare_packed_git(void)
637cdd9d
JK
928{
929 prepare_packed_git_run_once = 0;
930 prepare_packed_git();
1f688557
JH
931}
932
5d6ccf5c 933int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
0fcfd160
LT
934{
935 unsigned char real_sha1[20];
7cfb5f36 936 hash_sha1_file(map, size, type, real_sha1);
a89fccd2 937 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
938}
939
bb6b8e4f 940void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 941{
0fcfd160
LT
942 struct stat st;
943 void *map;
144bde78 944 int fd;
ace1534d
JH
945 char *filename = find_sha1_file(sha1, &st);
946
947 if (!filename) {
ace1534d
JH
948 return NULL;
949 }
0fcfd160 950
144bde78 951 fd = open(filename, O_RDONLY | sha1_file_open_flag);
0fcfd160 952 if (fd < 0) {
144bde78
LT
953 /* See if it works without O_NOATIME */
954 switch (sha1_file_open_flag) {
955 default:
956 fd = open(filename, O_RDONLY);
957 if (fd >= 0)
958 break;
959 /* Fallthrough */
960 case 0:
144bde78
LT
961 return NULL;
962 }
963
1f688557
JH
964 /* If it failed once, it will probably fail again.
965 * Stop using O_NOATIME
966 */
144bde78 967 sha1_file_open_flag = 0;
0fcfd160 968 }
dc49cd76
SP
969 *size = xsize_t(st.st_size);
970 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
0fcfd160 971 close(fd);
0fcfd160
LT
972 return map;
973}
974
726f852b 975static int legacy_loose_object(unsigned char *map)
bb6b8e4f
JH
976{
977 unsigned int word;
978
979 /*
980 * Is it a zlib-compressed buffer? If so, the first byte
981 * must be 0x78 (15-bit window size, deflated), and the
982 * first 16-bit word is evenly divisible by 31
983 */
984 word = (map[0] << 8) + map[1];
985 if (map[0] == 0x78 && !(word % 31))
986 return 1;
987 else
988 return 0;
989}
990
72518e9c 991unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 992{
ad1ed5ee 993 unsigned shift;
93821bd9 994 unsigned char c;
93821bd9 995 unsigned long size;
ad1ed5ee
JH
996 unsigned long used = 0;
997
998 c = buf[used++];
999 *type = (c >> 4) & 7;
1000 size = c & 15;
1001 shift = 4;
1002 while (c & 0x80) {
1003 if (len <= used)
1004 return 0;
1005 if (sizeof(long) * 8 <= shift)
1006 return 0;
1007 c = buf[used++];
1008 size += (c & 0x7f) << shift;
1009 shift += 7;
1010 }
1011 *sizep = size;
1012 return used;
1013}
1014
1015static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1016{
1017 unsigned long size, used;
1018 static const char valid_loose_object_type[8] = {
1019 0, /* OBJ_EXT */
1020 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1021 0, /* "delta" and others are invalid in a loose object */
93821bd9 1022 };
ad1ed5ee 1023 enum object_type type;
93821bd9 1024
c4483576
LT
1025 /* Get the data stream */
1026 memset(stream, 0, sizeof(*stream));
1027 stream->next_in = map;
1028 stream->avail_in = mapsize;
1029 stream->next_out = buffer;
93821bd9
LT
1030 stream->avail_out = bufsiz;
1031
bb6b8e4f 1032 if (legacy_loose_object(map)) {
93821bd9
LT
1033 inflateInit(stream);
1034 return inflate(stream, 0);
1035 }
1036
726f852b
NP
1037
1038 /*
1039 * There used to be a second loose object header format which
1040 * was meant to mimic the in-pack format, allowing for direct
1041 * copy of the object data. This format turned up not to be
1042 * really worth it and we don't write it any longer. But we
1043 * can still read it.
1044 */
ad1ed5ee
JH
1045 used = unpack_object_header_gently(map, mapsize, &type, &size);
1046 if (!used || !valid_loose_object_type[type])
93821bd9 1047 return -1;
ad1ed5ee
JH
1048 map += used;
1049 mapsize -= used;
93821bd9
LT
1050
1051 /* Set up the stream for the rest.. */
1052 stream->next_in = map;
1053 stream->avail_in = mapsize;
c4483576 1054 inflateInit(stream);
93821bd9
LT
1055
1056 /* And generate the fake traditional header */
ad1ed5ee 1057 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
df843662 1058 typename(type), size);
93821bd9 1059 return 0;
c4483576
LT
1060}
1061
7efbff75 1062static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1063{
1064 int bytes = strlen(buffer) + 1;
d565b341 1065 unsigned char *buf = xmalloc(1+size);
93821bd9 1066 unsigned long n;
7efbff75 1067 int status = Z_OK;
5180cacc 1068
93821bd9
LT
1069 n = stream->total_out - bytes;
1070 if (n > size)
1071 n = size;
1072 memcpy(buf, (char *) buffer + bytes, n);
1073 bytes = n;
456cdf6e
LT
1074 if (bytes <= size) {
1075 /*
1076 * The above condition must be (bytes <= size), not
1077 * (bytes < size). In other words, even though we
1078 * expect no more output and set avail_out to zer0,
1079 * the input zlib stream may have bytes that express
1080 * "this concludes the stream", and we *do* want to
1081 * eat that input.
1082 *
1083 * Otherwise we would not be able to test that we
1084 * consumed all the input to reach the expected size;
1085 * we also want to check that zlib tells us that all
1086 * went well with status == Z_STREAM_END at the end.
1087 */
5180cacc
LT
1088 stream->next_out = buf + bytes;
1089 stream->avail_out = size - bytes;
7efbff75
JH
1090 while (status == Z_OK)
1091 status = inflate(stream, Z_FINISH);
5180cacc
LT
1092 }
1093 buf[size] = 0;
456cdf6e 1094 if (status == Z_STREAM_END && !stream->avail_in) {
7efbff75
JH
1095 inflateEnd(stream);
1096 return buf;
1097 }
1098
1099 if (status < 0)
1100 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1101 else if (stream->avail_in)
1102 error("garbage at end of loose object '%s'",
1103 sha1_to_hex(sha1));
1104 free(buf);
1105 return NULL;
5180cacc
LT
1106}
1107
1108/*
1109 * We used to just use "sscanf()", but that's actually way
1110 * too permissive for what we want to check. So do an anal
1111 * object header parse by hand.
1112 */
21666f1a 1113static int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1114{
21666f1a 1115 char type[10];
5180cacc
LT
1116 int i;
1117 unsigned long size;
1118
1119 /*
1120 * The type can be at most ten bytes (including the
1121 * terminating '\0' that we add), and is followed by
21666f1a 1122 * a space.
5180cacc 1123 */
21666f1a 1124 i = 0;
5180cacc
LT
1125 for (;;) {
1126 char c = *hdr++;
1127 if (c == ' ')
1128 break;
21666f1a
NP
1129 type[i++] = c;
1130 if (i >= sizeof(type))
5180cacc 1131 return -1;
5180cacc 1132 }
21666f1a 1133 type[i] = 0;
5180cacc
LT
1134
1135 /*
1136 * The length must follow immediately, and be in canonical
1137 * decimal format (ie "010" is not valid).
1138 */
1139 size = *hdr++ - '0';
1140 if (size > 9)
1141 return -1;
1142 if (size) {
1143 for (;;) {
1144 unsigned long c = *hdr - '0';
1145 if (c > 9)
1146 break;
1147 hdr++;
1148 size = size * 10 + c;
1149 }
1150 }
1151 *sizep = size;
1152
1153 /*
1154 * The length must be followed by a zero byte
1155 */
21666f1a 1156 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1157}
1158
7efbff75 1159static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1160{
5180cacc 1161 int ret;
0fcfd160 1162 z_stream stream;
5180cacc 1163 char hdr[8192];
0fcfd160 1164
5180cacc 1165 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1166 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1167 return NULL;
1168
7efbff75 1169 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1170}
1171
54dab52a
NP
1172unsigned long get_size_from_delta(struct packed_git *p,
1173 struct pack_window **w_curs,
1174 off_t curpos)
1175{
1176 const unsigned char *data;
1177 unsigned char delta_head[20], *in;
1178 z_stream stream;
1179 int st;
1180
1181 memset(&stream, 0, sizeof(stream));
1182 stream.next_out = delta_head;
1183 stream.avail_out = sizeof(delta_head);
1184
1185 inflateInit(&stream);
1186 do {
1187 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1188 stream.next_in = in;
1189 st = inflate(&stream, Z_FINISH);
1190 curpos += stream.next_in - in;
1191 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1192 stream.total_out < sizeof(delta_head));
1193 inflateEnd(&stream);
1194 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1195 die("delta data unpack-initial failed");
1196
1197 /* Examine the initial part of the delta to figure out
1198 * the result size.
1199 */
1200 data = delta_head;
1201
1202 /* ignore base size */
1203 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1204
1205 /* Read the result size */
1206 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1207}
1208
c4001d92 1209static off_t get_delta_base(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 delta_obj_offset)
eb32d236 1214{
2b87c45b 1215 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1216 off_t base_offset;
eb32d236 1217
8d8a4ea5
SP
1218 /* use_pack() assured us we have [base_info, base_info + 20)
1219 * as a range that we can look at without walking off the
1220 * end of the mapped window. Its actually the hash size
1221 * that is assured. An OFS_DELTA longer than the hash size
1222 * is stupid, as then a REF_DELTA would be smaller to store.
1223 */
21666f1a 1224 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1225 unsigned used = 0;
1226 unsigned char c = base_info[used++];
1227 base_offset = c & 127;
1228 while (c & 128) {
1229 base_offset += 1;
8723f216 1230 if (!base_offset || MSB(base_offset, 7))
eb32d236
NP
1231 die("offset value overflow for delta base object");
1232 c = base_info[used++];
1233 base_offset = (base_offset << 7) + (c & 127);
1234 }
1235 base_offset = delta_obj_offset - base_offset;
1236 if (base_offset >= delta_obj_offset)
1237 die("delta base offset out of bound");
2b87c45b 1238 *curpos += used;
21666f1a 1239 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1240 /* The base entry _must_ be in the same pack */
1241 base_offset = find_pack_entry_one(base_info, p);
1242 if (!base_offset)
1243 die("failed to find delta-pack base object %s",
1244 sha1_to_hex(base_info));
2b87c45b 1245 *curpos += 20;
eb32d236
NP
1246 } else
1247 die("I am totally screwed");
2b87c45b 1248 return base_offset;
eb32d236
NP
1249}
1250
f3bf9224 1251/* forward declaration for a mutually recursive function */
c4001d92 1252static int packed_object_info(struct packed_git *p, off_t offset,
21666f1a 1253 unsigned long *sizep);
f3bf9224 1254
43057304 1255static int packed_delta_info(struct packed_git *p,
03e79c88 1256 struct pack_window **w_curs,
c4001d92 1257 off_t curpos,
21666f1a 1258 enum object_type type,
c4001d92 1259 off_t obj_offset,
43057304 1260 unsigned long *sizep)
5db47c2b 1261{
c4001d92 1262 off_t base_offset;
f3bf9224 1263
21666f1a
NP
1264 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1265 type = packed_object_info(p, base_offset, NULL);
f3bf9224 1266
c62266f3
JH
1267 /* We choose to only get the type of the base object and
1268 * ignore potentially corrupt pack file that expects the delta
1269 * based on a base with a wrong size. This saves tons of
1270 * inflate() calls.
1271 */
54dab52a
NP
1272 if (sizep)
1273 *sizep = get_size_from_delta(p, w_curs, curpos);
21666f1a
NP
1274
1275 return type;
5db47c2b
JH
1276}
1277
2b87c45b
NP
1278static int unpack_object_header(struct packed_git *p,
1279 struct pack_window **w_curs,
c4001d92 1280 off_t *curpos,
2b87c45b 1281 unsigned long *sizep)
a733cb60 1282{
03e79c88
SP
1283 unsigned char *base;
1284 unsigned int left;
ad1ed5ee 1285 unsigned long used;
2b87c45b 1286 enum object_type type;
a733cb60 1287
8d8a4ea5
SP
1288 /* use_pack() assures us we have [base, base + 20) available
1289 * as a range that we can look at at. (Its actually the hash
3dff5379 1290 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1291 * the maximum deflated object size is 2^137, which is just
1292 * insane, so we know won't exceed what we have been given.
1293 */
2b87c45b
NP
1294 base = use_pack(p, w_curs, *curpos, &left);
1295 used = unpack_object_header_gently(base, left, &type, sizep);
ad1ed5ee
JH
1296 if (!used)
1297 die("object offset outside of pack file");
2b87c45b 1298 *curpos += used;
ad1ed5ee 1299
2b87c45b 1300 return type;
a733cb60
LT
1301}
1302
21666f1a 1303const char *packed_object_info_detail(struct packed_git *p,
c4001d92 1304 off_t obj_offset,
21666f1a
NP
1305 unsigned long *size,
1306 unsigned long *store_size,
1307 unsigned int *delta_chain_length,
1308 unsigned char *base_sha1)
ad8c80a5 1309{
03e79c88 1310 struct pack_window *w_curs = NULL;
c4001d92
SP
1311 off_t curpos;
1312 unsigned long dummy;
43057304 1313 unsigned char *next_sha1;
21666f1a 1314 enum object_type type;
ad8c80a5 1315
43057304 1316 *delta_chain_length = 0;
2b87c45b 1317 curpos = obj_offset;
21666f1a 1318 type = unpack_object_header(p, &w_curs, &curpos, size);
43057304
NP
1319
1320 for (;;) {
21666f1a 1321 switch (type) {
43057304 1322 default:
08a19d87 1323 die("pack %s contains unknown object type %d",
21666f1a 1324 p->pack_name, type);
43057304
NP
1325 case OBJ_COMMIT:
1326 case OBJ_TREE:
1327 case OBJ_BLOB:
1328 case OBJ_TAG:
43057304 1329 *store_size = 0; /* notyet */
03e79c88 1330 unuse_pack(&w_curs);
21666f1a 1331 return typename(type);
eb32d236 1332 case OBJ_OFS_DELTA:
21666f1a 1333 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
eb32d236 1334 if (*delta_chain_length == 0) {
2b87c45b 1335 /* TODO: find base_sha1 as pointed by curpos */
30fee062 1336 hashclr(base_sha1);
eb32d236
NP
1337 }
1338 break;
1339 case OBJ_REF_DELTA:
2b87c45b 1340 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
43057304
NP
1341 if (*delta_chain_length == 0)
1342 hashcpy(base_sha1, next_sha1);
2b87c45b 1343 obj_offset = find_pack_entry_one(next_sha1, p);
43057304
NP
1344 break;
1345 }
43057304 1346 (*delta_chain_length)++;
2b87c45b 1347 curpos = obj_offset;
21666f1a 1348 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
ad8c80a5 1349 }
ad8c80a5
JH
1350}
1351
c4001d92 1352static int packed_object_info(struct packed_git *p, off_t obj_offset,
21666f1a 1353 unsigned long *sizep)
1f688557 1354{
03e79c88 1355 struct pack_window *w_curs = NULL;
c4001d92
SP
1356 unsigned long size;
1357 off_t curpos = obj_offset;
21666f1a 1358 enum object_type type;
5db47c2b 1359
21666f1a 1360 type = unpack_object_header(p, &w_curs, &curpos, &size);
5db47c2b 1361
21666f1a 1362 switch (type) {
eb32d236
NP
1363 case OBJ_OFS_DELTA:
1364 case OBJ_REF_DELTA:
21666f1a
NP
1365 type = packed_delta_info(p, &w_curs, curpos,
1366 type, obj_offset, sizep);
2b87c45b 1367 break;
a733cb60 1368 case OBJ_COMMIT:
a733cb60 1369 case OBJ_TREE:
a733cb60 1370 case OBJ_BLOB:
a733cb60 1371 case OBJ_TAG:
2b87c45b
NP
1372 if (sizep)
1373 *sizep = size;
a69d0943 1374 break;
1f688557 1375 default:
08a19d87 1376 die("pack %s contains unknown object type %d",
21666f1a 1377 p->pack_name, type);
1f688557 1378 }
2b87c45b 1379 unuse_pack(&w_curs);
21666f1a 1380 return type;
1f688557
JH
1381}
1382
eb950c19 1383static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1384 struct pack_window **w_curs,
c4001d92 1385 off_t curpos,
eb950c19 1386 unsigned long size)
de530aaa
SP
1387{
1388 int st;
1389 z_stream stream;
079afb18 1390 unsigned char *buffer, *in;
de530aaa
SP
1391
1392 buffer = xmalloc(size + 1);
1393 buffer[size] = 0;
1394 memset(&stream, 0, sizeof(stream));
de530aaa
SP
1395 stream.next_out = buffer;
1396 stream.avail_out = size;
1397
1398 inflateInit(&stream);
079afb18 1399 do {
2b87c45b 1400 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18
SP
1401 stream.next_in = in;
1402 st = inflate(&stream, Z_FINISH);
2b87c45b 1403 curpos += stream.next_in - in;
079afb18 1404 } while (st == Z_OK || st == Z_BUF_ERROR);
de530aaa
SP
1405 inflateEnd(&stream);
1406 if ((st != Z_STREAM_END) || stream.total_out != size) {
1407 free(buffer);
1408 return NULL;
1409 }
1410
1411 return buffer;
1412}
1413
e5e01619
LT
1414#define MAX_DELTA_CACHE (256)
1415
18bdec11 1416static size_t delta_base_cached;
5e08ecbf
NP
1417
1418static struct delta_base_cache_lru_list {
1419 struct delta_base_cache_lru_list *prev;
1420 struct delta_base_cache_lru_list *next;
1421} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1422
e5e01619 1423static struct delta_base_cache_entry {
5e08ecbf
NP
1424 struct delta_base_cache_lru_list lru;
1425 void *data;
e5e01619
LT
1426 struct packed_git *p;
1427 off_t base_offset;
1428 unsigned long size;
e5e01619
LT
1429 enum object_type type;
1430} delta_base_cache[MAX_DELTA_CACHE];
1431
1432static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1433{
1434 unsigned long hash;
1435
1436 hash = (unsigned long)p + (unsigned long)base_offset;
1437 hash += (hash >> 8) + (hash >> 16);
3358004a 1438 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1439}
1440
62f255ad 1441static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 1442 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 1443{
e5e01619
LT
1444 void *ret;
1445 unsigned long hash = pack_entry_hash(p, base_offset);
1446 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1447
1448 ret = ent->data;
1449 if (ret && ent->p == p && ent->base_offset == base_offset)
1450 goto found_cache_entry;
62f255ad 1451 return unpack_entry(p, base_offset, type, base_size);
e5e01619
LT
1452
1453found_cache_entry:
18bdec11 1454 if (!keep_cache) {
a0cba108 1455 ent->data = NULL;
5e08ecbf
NP
1456 ent->lru.next->prev = ent->lru.prev;
1457 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
1458 delta_base_cached -= ent->size;
1459 }
a0cba108
NP
1460 else {
1461 ret = xmalloc(ent->size + 1);
1462 memcpy(ret, ent->data, ent->size);
1463 ((char *)ret)[ent->size] = 0;
1464 }
e5e01619
LT
1465 *type = ent->type;
1466 *base_size = ent->size;
1467 return ret;
62f255ad
LT
1468}
1469
18bdec11
SP
1470static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1471{
1472 if (ent->data) {
1473 free(ent->data);
1474 ent->data = NULL;
5e08ecbf
NP
1475 ent->lru.next->prev = ent->lru.prev;
1476 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
1477 delta_base_cached -= ent->size;
1478 }
1479}
1480
62f255ad
LT
1481static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1482 void *base, unsigned long base_size, enum object_type type)
1483{
5e08ecbf 1484 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 1485 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 1486 struct delta_base_cache_lru_list *lru;
e5e01619 1487
18bdec11
SP
1488 release_delta_base_cache(ent);
1489 delta_base_cached += base_size;
5e08ecbf
NP
1490
1491 for (lru = delta_base_cache_lru.next;
1492 delta_base_cached > delta_base_cache_limit
1493 && lru != &delta_base_cache_lru;
1494 lru = lru->next) {
1495 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
1496 if (f->type == OBJ_BLOB)
1497 release_delta_base_cache(f);
1498 }
5e08ecbf
NP
1499 for (lru = delta_base_cache_lru.next;
1500 delta_base_cached > delta_base_cache_limit
1501 && lru != &delta_base_cache_lru;
1502 lru = lru->next) {
1503 struct delta_base_cache_entry *f = (void *)lru;
1504 release_delta_base_cache(f);
1505 }
18bdec11 1506
e5e01619
LT
1507 ent->p = p;
1508 ent->base_offset = base_offset;
1509 ent->type = type;
1510 ent->data = base;
1511 ent->size = base_size;
5e08ecbf
NP
1512 ent->lru.next = &delta_base_cache_lru;
1513 ent->lru.prev = delta_base_cache_lru.prev;
1514 delta_base_cache_lru.prev->next = &ent->lru;
1515 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
1516}
1517
eb950c19 1518static void *unpack_delta_entry(struct packed_git *p,
03e79c88 1519 struct pack_window **w_curs,
c4001d92 1520 off_t curpos,
1f688557 1521 unsigned long delta_size,
c4001d92 1522 off_t obj_offset,
21666f1a 1523 enum object_type *type,
eb950c19 1524 unsigned long *sizep)
1f688557 1525{
7c3e8be3 1526 void *delta_data, *result, *base;
c4001d92
SP
1527 unsigned long base_size;
1528 off_t base_offset;
43057304 1529
21666f1a 1530 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
a0cba108 1531 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
67686d95 1532 if (!base)
c4001d92
SP
1533 die("failed to read delta base object"
1534 " at %"PRIuMAX" from %s",
1535 (uintmax_t)base_offset, p->pack_name);
67686d95 1536
2b87c45b 1537 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1f688557
JH
1538 result = patch_delta(base, base_size,
1539 delta_data, delta_size,
21666f1a 1540 sizep);
1f688557
JH
1541 if (!result)
1542 die("failed to apply delta");
1543 free(delta_data);
62f255ad 1544 add_delta_base_cache(p, base_offset, base, base_size, *type);
1f688557
JH
1545 return result;
1546}
1547
c4001d92 1548void *unpack_entry(struct packed_git *p, off_t obj_offset,
21666f1a 1549 enum object_type *type, unsigned long *sizep)
1f688557 1550{
03e79c88 1551 struct pack_window *w_curs = NULL;
c4001d92 1552 off_t curpos = obj_offset;
21666f1a 1553 void *data;
1f688557 1554
21666f1a
NP
1555 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1556 switch (*type) {
eb32d236
NP
1557 case OBJ_OFS_DELTA:
1558 case OBJ_REF_DELTA:
21666f1a
NP
1559 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1560 obj_offset, type, sizep);
4d703a1a 1561 break;
a733cb60 1562 case OBJ_COMMIT:
a733cb60 1563 case OBJ_TREE:
a733cb60 1564 case OBJ_BLOB:
a733cb60 1565 case OBJ_TAG:
21666f1a 1566 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
4d703a1a 1567 break;
1f688557 1568 default:
21666f1a 1569 die("unknown object type %i in %s", *type, p->pack_name);
1f688557 1570 }
03e79c88 1571 unuse_pack(&w_curs);
21666f1a 1572 return data;
1f688557
JH
1573}
1574
d72308e0
NP
1575const unsigned char *nth_packed_object_sha1(const struct packed_git *p,
1576 uint32_t n)
9a217f2a 1577{
42873078 1578 const unsigned char *index = p->index_data;
57059091 1579 if (n >= p->num_objects)
d72308e0 1580 return NULL;
74e34e1f
NP
1581 index += 4 * 256;
1582 if (p->index_version == 1) {
1583 return index + 24 * n + 4;
1584 } else {
1585 index += 8;
1586 return index + 20 * n;
1587 }
1588}
1589
1590static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1591{
1592 const unsigned char *index = p->index_data;
1593 index += 4 * 256;
1594 if (p->index_version == 1) {
1595 return ntohl(*((uint32_t *)(index + 24 * n)));
1596 } else {
1597 uint32_t off;
1598 index += 8 + p->num_objects * (20 + 4);
1599 off = ntohl(*((uint32_t *)(index + 4 * n)));
1600 if (!(off & 0x80000000))
1601 return off;
1602 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1603 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1604 ntohl(*((uint32_t *)(index + 4)));
1605 }
9a217f2a
JH
1606}
1607
c4001d92 1608off_t find_pack_entry_one(const unsigned char *sha1,
43057304 1609 struct packed_git *p)
1f688557 1610{
42873078 1611 const uint32_t *level1_ofs = p->index_data;
42873078 1612 const unsigned char *index = p->index_data;
74e34e1f 1613 unsigned hi, lo;
42873078 1614
74e34e1f
NP
1615 if (p->index_version > 1) {
1616 level1_ofs += 2;
1617 index += 8;
1618 }
42873078 1619 index += 4 * 256;
74e34e1f
NP
1620 hi = ntohl(level1_ofs[*sha1]);
1621 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1f688557
JH
1622
1623 do {
74e34e1f
NP
1624 unsigned mi = (lo + hi) / 2;
1625 unsigned x = (p->index_version > 1) ? (mi * 20) : (mi * 24 + 4);
1626 int cmp = hashcmp(index + x, sha1);
43057304 1627 if (!cmp)
74e34e1f 1628 return nth_packed_object_offset(p, mi);
1f688557
JH
1629 if (cmp > 0)
1630 hi = mi;
1631 else
1632 lo = mi+1;
1633 } while (lo < hi);
1634 return 0;
1635}
1636
d4ff6d92
SP
1637static int matches_pack_name(struct packed_git *p, const char *ig)
1638{
1639 const char *last_c, *c;
1640
1641 if (!strcmp(p->pack_name, ig))
1642 return 0;
1643
1644 for (c = p->pack_name, last_c = c; *c;)
1645 if (*c == '/')
1646 last_c = ++c;
1647 else
1648 ++c;
1649 if (!strcmp(last_c, ig))
1650 return 0;
1651
1652 return 1;
1653}
1654
106d710b 1655static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1f688557
JH
1656{
1657 struct packed_git *p;
c4001d92 1658 off_t offset;
43057304 1659
1f688557
JH
1660 prepare_packed_git();
1661
1662 for (p = packed_git; p; p = p->next) {
106d710b
JH
1663 if (ignore_packed) {
1664 const char **ig;
1665 for (ig = ignore_packed; *ig; ig++)
d4ff6d92 1666 if (!matches_pack_name(p, *ig))
106d710b
JH
1667 break;
1668 if (*ig)
1669 continue;
1670 }
43057304
NP
1671 offset = find_pack_entry_one(sha1, p);
1672 if (offset) {
c715f783
SP
1673 /*
1674 * We are about to tell the caller where they can
1675 * locate the requested object. We better make
1676 * sure the packfile is still here and can be
1677 * accessed before supplying that answer, as
1678 * it may have been deleted since the index
1679 * was loaded!
1680 */
1681 if (p->pack_fd == -1 && open_packed_git(p)) {
1682 error("packfile %s cannot be accessed", p->pack_name);
1683 continue;
1684 }
43057304
NP
1685 e->offset = offset;
1686 e->p = p;
1687 hashcpy(e->sha1, sha1);
1f688557 1688 return 1;
43057304 1689 }
1f688557
JH
1690 }
1691 return 0;
1692}
1693
bf592c50
DB
1694struct packed_git *find_sha1_pack(const unsigned char *sha1,
1695 struct packed_git *packs)
1696{
1697 struct packed_git *p;
bf592c50
DB
1698
1699 for (p = packs; p; p = p->next) {
43057304 1700 if (find_pack_entry_one(sha1, p))
bf592c50
DB
1701 return p;
1702 }
1703 return NULL;
21666f1a 1704
bf592c50
DB
1705}
1706
21666f1a 1707static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
65c2e0c3 1708{
36e4d74a 1709 int status;
65c2e0c3
JH
1710 unsigned long mapsize, size;
1711 void *map;
1712 z_stream stream;
d65a16f6 1713 char hdr[32];
65c2e0c3 1714
bb6b8e4f 1715 map = map_sha1_file(sha1, &mapsize);
f0df4ed5
JS
1716 if (!map)
1717 return error("unable to find %s", sha1_to_hex(sha1));
36e4d74a
JH
1718 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1719 status = error("unable to unpack %s header",
1720 sha1_to_hex(sha1));
21666f1a 1721 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 1722 status = error("unable to parse %s header", sha1_to_hex(sha1));
21666f1a
NP
1723 else if (sizep)
1724 *sizep = size;
65c2e0c3
JH
1725 inflateEnd(&stream);
1726 munmap(map, mapsize);
1727 return status;
1728}
1729
21666f1a 1730int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
f0df4ed5 1731{
f0df4ed5
JS
1732 struct pack_entry e;
1733
1734 if (!find_pack_entry(sha1, &e, NULL)) {
1735 reprepare_packed_git();
1736 if (!find_pack_entry(sha1, &e, NULL))
21666f1a 1737 return sha1_loose_object_info(sha1, sizep);
f0df4ed5 1738 }
21666f1a 1739 return packed_object_info(e.p, e.offset, sizep);
f0df4ed5
JS
1740}
1741
21666f1a
NP
1742static void *read_packed_sha1(const unsigned char *sha1,
1743 enum object_type *type, unsigned long *size)
1f688557
JH
1744{
1745 struct pack_entry e;
1746
8276c007 1747 if (!find_pack_entry(sha1, &e, NULL))
1f688557 1748 return NULL;
8276c007 1749 else
a0cba108 1750 return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1f688557
JH
1751}
1752
d66b37bb
JH
1753/*
1754 * This is meant to hold a *small* number of objects that you would
1755 * want read_sha1_file() to be able to return, but yet you do not want
1756 * to write them into the object store (e.g. a browse-only
1757 * application).
1758 */
1759static struct cached_object {
1760 unsigned char sha1[20];
21666f1a 1761 enum object_type type;
d66b37bb
JH
1762 void *buf;
1763 unsigned long size;
1764} *cached_objects;
1765static int cached_object_nr, cached_object_alloc;
1766
1767static struct cached_object *find_cached_object(const unsigned char *sha1)
1768{
1769 int i;
1770 struct cached_object *co = cached_objects;
1771
1772 for (i = 0; i < cached_object_nr; i++, co++) {
1773 if (!hashcmp(co->sha1, sha1))
1774 return co;
1775 }
1776 return NULL;
1777}
1778
21666f1a
NP
1779int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1780 unsigned char *sha1)
d66b37bb
JH
1781{
1782 struct cached_object *co;
1783
21666f1a 1784 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
1785 if (has_sha1_file(sha1) || find_cached_object(sha1))
1786 return 0;
1787 if (cached_object_alloc <= cached_object_nr) {
1788 cached_object_alloc = alloc_nr(cached_object_alloc);
1789 cached_objects = xrealloc(cached_objects,
1790 sizeof(*cached_objects) *
1791 cached_object_alloc);
1792 }
1793 co = &cached_objects[cached_object_nr++];
1794 co->size = len;
21666f1a 1795 co->type = type;
efa13f7b
JH
1796 co->buf = xmalloc(len);
1797 memcpy(co->buf, buf, len);
d66b37bb
JH
1798 hashcpy(co->sha1, sha1);
1799 return 0;
1800}
1801
21666f1a
NP
1802void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1803 unsigned long *size)
0fcfd160
LT
1804{
1805 unsigned long mapsize;
1806 void *map, *buf;
d66b37bb
JH
1807 struct cached_object *co;
1808
1809 co = find_cached_object(sha1);
1810 if (co) {
1811 buf = xmalloc(co->size + 1);
1812 memcpy(buf, co->buf, co->size);
1813 ((char*)buf)[co->size] = 0;
21666f1a 1814 *type = co->type;
d66b37bb
JH
1815 *size = co->size;
1816 return buf;
1817 }
0fcfd160 1818
8276c007
PE
1819 buf = read_packed_sha1(sha1, type, size);
1820 if (buf)
1821 return buf;
bb6b8e4f 1822 map = map_sha1_file(sha1, &mapsize);
0fcfd160 1823 if (map) {
7efbff75 1824 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
1825 munmap(map, mapsize);
1826 return buf;
1827 }
637cdd9d 1828 reprepare_packed_git();
8276c007 1829 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
1830}
1831
40469ee9 1832void *read_object_with_reference(const unsigned char *sha1,
21666f1a 1833 const char *required_type_name,
40469ee9
JH
1834 unsigned long *size,
1835 unsigned char *actual_sha1_return)
f4913f91 1836{
21666f1a 1837 enum object_type type, required_type;
f4913f91
JH
1838 void *buffer;
1839 unsigned long isize;
40469ee9 1840 unsigned char actual_sha1[20];
f4913f91 1841
21666f1a 1842 required_type = type_from_string(required_type_name);
e702496e 1843 hashcpy(actual_sha1, sha1);
40469ee9
JH
1844 while (1) {
1845 int ref_length = -1;
1846 const char *ref_type = NULL;
f4913f91 1847
21666f1a 1848 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
1849 if (!buffer)
1850 return NULL;
21666f1a 1851 if (type == required_type) {
40469ee9
JH
1852 *size = isize;
1853 if (actual_sha1_return)
e702496e 1854 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
1855 return buffer;
1856 }
1857 /* Handle references */
21666f1a 1858 else if (type == OBJ_COMMIT)
40469ee9 1859 ref_type = "tree ";
21666f1a 1860 else if (type == OBJ_TAG)
40469ee9
JH
1861 ref_type = "object ";
1862 else {
1863 free(buffer);
1864 return NULL;
1865 }
1866 ref_length = strlen(ref_type);
f4913f91 1867
40469ee9 1868 if (memcmp(buffer, ref_type, ref_length) ||
1d7f171c 1869 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
1870 free(buffer);
1871 return NULL;
1872 }
1cf58e72 1873 free(buffer);
40469ee9
JH
1874 /* Now we have the ID of the referred-to object in
1875 * actual_sha1. Check again. */
f4913f91 1876 }
f4913f91
JH
1877}
1878
ce9fbf16 1879static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 1880 const char *type, unsigned char *sha1,
d65a16f6 1881 char *hdr, int *hdrlen)
d410c0f5
JH
1882{
1883 SHA_CTX c;
1884
1885 /* Generate the header */
d65a16f6 1886 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
1887
1888 /* Sha1.. */
1889 SHA1_Init(&c);
1890 SHA1_Update(&c, hdr, *hdrlen);
1891 SHA1_Update(&c, buf, len);
1892 SHA1_Final(sha1, &c);
d410c0f5
JH
1893}
1894
230f1322
LT
1895/*
1896 * Link the tempfile to the final place, possibly creating the
1897 * last directory level as you do so.
1898 *
1899 * Returns the errno on failure, 0 on success.
1900 */
839837b9 1901static int link_temp_to_file(const char *tmpfile, const char *filename)
230f1322
LT
1902{
1903 int ret;
756aaf4a 1904 char *dir;
230f1322
LT
1905
1906 if (!link(tmpfile, filename))
1907 return 0;
1908
1909 /*
756aaf4a 1910 * Try to mkdir the last path component if that failed.
230f1322
LT
1911 *
1912 * Re-try the "link()" regardless of whether the mkdir
1913 * succeeds, since a race might mean that somebody
1914 * else succeeded.
1915 */
1916 ret = errno;
756aaf4a
SP
1917 dir = strrchr(filename, '/');
1918 if (dir) {
1919 *dir = 0;
866cae0d 1920 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
91c23e48 1921 *dir = '/';
756aaf4a 1922 return -2;
91c23e48 1923 }
756aaf4a
SP
1924 *dir = '/';
1925 if (!link(tmpfile, filename))
1926 return 0;
1927 ret = errno;
230f1322
LT
1928 }
1929 return ret;
1930}
1931
1932/*
1933 * Move the just written object into its final resting place
1934 */
839837b9 1935int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322
LT
1936{
1937 int ret = link_temp_to_file(tmpfile, filename);
7ebb6fca
LT
1938
1939 /*
1940 * Coda hack - coda doesn't like cross-directory links,
1941 * so we fall back to a rename, which will mean that it
1942 * won't be able to check collisions, but that's not a
1943 * big deal.
1944 *
1945 * The same holds for FAT formatted media.
1946 *
1947 * When this succeeds, we just return 0. We have nothing
1948 * left to unlink.
1949 */
1950 if (ret && ret != EEXIST) {
1951 if (!rename(tmpfile, filename))
230f1322 1952 return 0;
9e48b389 1953 ret = errno;
230f1322
LT
1954 }
1955 unlink(tmpfile);
1956 if (ret) {
1957 if (ret != EEXIST) {
916d081b 1958 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
230f1322
LT
1959 }
1960 /* FIXME!!! Collision check here ? */
1961 }
1962
1963 return 0;
1964}
1965
4d548150
LT
1966static int write_buffer(int fd, const void *buf, size_t len)
1967{
d34cf19b 1968 if (write_in_full(fd, buf, len) < 0)
93822c22 1969 return error("file write error (%s)", strerror(errno));
4d548150
LT
1970 return 0;
1971}
1972
ce9fbf16 1973int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
1974 unsigned char *sha1)
1975{
d65a16f6 1976 char hdr[32];
abdc3fc8
RS
1977 int hdrlen;
1978 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1979 return 0;
1980}
1981
bf0f910d 1982int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
0fcfd160 1983{
ac54c277 1984 int size, ret;
bf0f910d 1985 unsigned char *compressed;
0fcfd160
LT
1986 z_stream stream;
1987 unsigned char sha1[20];
706bc531 1988 char *filename;
aac17941 1989 static char tmpfile[PATH_MAX];
d65a16f6 1990 char hdr[32];
230f1322 1991 int fd, hdrlen;
a44c9a5e 1992
d410c0f5
JH
1993 /* Normally if we have it in the pack then we do not bother writing
1994 * it out into .git/objects/??/?{38} file.
1995 */
972a9155
RS
1996 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1997 filename = sha1_file_name(sha1);
706bc531 1998 if (returnsha1)
e702496e 1999 hashcpy(returnsha1, sha1);
d410c0f5
JH
2000 if (has_sha1_file(sha1))
2001 return 0;
aac17941
LT
2002 fd = open(filename, O_RDONLY);
2003 if (fd >= 0) {
706bc531 2004 /*
aac17941
LT
2005 * FIXME!!! We might do collision checking here, but we'd
2006 * need to uncompress the old file and check it. Later.
706bc531 2007 */
aac17941 2008 close(fd);
706bc531
LT
2009 return 0;
2010 }
2011
aac17941 2012 if (errno != ENOENT) {
916d081b 2013 return error("sha1 file %s: %s\n", filename, strerror(errno));
aac17941
LT
2014 }
2015
0e55181f 2016 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
ace1534d 2017
aac17941
LT
2018 fd = mkstemp(tmpfile);
2019 if (fd < 0) {
916d081b
PB
2020 if (errno == EPERM)
2021 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2022 else
2023 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
aac17941
LT
2024 }
2025
0fcfd160
LT
2026 /* Set it up */
2027 memset(&stream, 0, sizeof(stream));
12f6c308 2028 deflateInit(&stream, zlib_compression_level);
93821bd9 2029 size = 8 + deflateBound(&stream, len+hdrlen);
812666c8 2030 compressed = xmalloc(size);
0fcfd160
LT
2031
2032 /* Compress it */
0fcfd160
LT
2033 stream.next_out = compressed;
2034 stream.avail_out = size;
a44c9a5e
LT
2035
2036 /* First header.. */
d65a16f6 2037 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2038 stream.avail_in = hdrlen;
726f852b
NP
2039 while (deflate(&stream, 0) == Z_OK)
2040 /* nothing */;
a44c9a5e
LT
2041
2042 /* Then the data itself.. */
2043 stream.next_in = buf;
2044 stream.avail_in = len;
ac54c277
LT
2045 ret = deflate(&stream, Z_FINISH);
2046 if (ret != Z_STREAM_END)
2047 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2048
2049 ret = deflateEnd(&stream);
2050 if (ret != Z_OK)
2051 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2052
0fcfd160
LT
2053 size = stream.total_out;
2054
4d548150
LT
2055 if (write_buffer(fd, compressed, size) < 0)
2056 die("unable to write sha1 file");
aac17941 2057 fchmod(fd, 0444);
e82973cf
JH
2058 if (close(fd))
2059 die("unable to write sha1 file");
383f85b7 2060 free(compressed);
0fcfd160 2061
230f1322 2062 return move_temp_to_file(tmpfile, filename);
0fcfd160 2063}
8237b185 2064
4d548150
LT
2065/*
2066 * We need to unpack and recompress the object for writing
2067 * it out to a different file.
2068 */
2069static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
a5eda52b 2070{
4d548150 2071 size_t size;
a5eda52b 2072 z_stream stream;
4d548150
LT
2073 unsigned char *unpacked;
2074 unsigned long len;
21666f1a 2075 enum object_type type;
d65a16f6 2076 char hdr[32];
4d548150
LT
2077 int hdrlen;
2078 void *buf;
bfc66daf 2079
a9486b02 2080 /* need to unpack and recompress it by itself */
21666f1a 2081 unpacked = read_packed_sha1(sha1, &type, &len);
8276c007
PE
2082 if (!unpacked)
2083 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
a5eda52b 2084
21666f1a 2085 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
a5eda52b 2086
4d548150
LT
2087 /* Set it up */
2088 memset(&stream, 0, sizeof(stream));
12f6c308 2089 deflateInit(&stream, zlib_compression_level);
4d548150
LT
2090 size = deflateBound(&stream, len + hdrlen);
2091 buf = xmalloc(size);
a5eda52b 2092
4d548150
LT
2093 /* Compress it */
2094 stream.next_out = buf;
2095 stream.avail_out = size;
a5eda52b 2096
4d548150
LT
2097 /* First header.. */
2098 stream.next_in = (void *)hdr;
2099 stream.avail_in = hdrlen;
2100 while (deflate(&stream, 0) == Z_OK)
2101 /* nothing */;
bfc66daf 2102
4d548150
LT
2103 /* Then the data itself.. */
2104 stream.next_in = unpacked;
2105 stream.avail_in = len;
2106 while (deflate(&stream, Z_FINISH) == Z_OK)
2107 /* nothing */;
2108 deflateEnd(&stream);
2109 free(unpacked);
bfc66daf 2110
4d548150
LT
2111 *objsize = stream.total_out;
2112 return buf;
2113}
2114
2115int write_sha1_to_fd(int fd, const unsigned char *sha1)
2116{
2117 int retval;
2118 unsigned long objsize;
bb6b8e4f 2119 void *buf = map_sha1_file(sha1, &objsize);
4d548150
LT
2120
2121 if (buf) {
2122 retval = write_buffer(fd, buf, objsize);
2123 munmap(buf, objsize);
2124 return retval;
2125 }
2126
2127 buf = repack_object(sha1, &objsize);
2128 retval = write_buffer(fd, buf, objsize);
2129 free(buf);
2130 return retval;
a5eda52b
DB
2131}
2132
70b9829e
DB
2133int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2134 size_t bufsize, size_t *bufposn)
8237b185 2135{
230f1322 2136 char tmpfile[PATH_MAX];
8237b185
DB
2137 int local;
2138 z_stream stream;
2139 unsigned char real_sha1[20];
bf0f910d 2140 unsigned char discard[4096];
8237b185
DB
2141 int ret;
2142 SHA_CTX c;
2143
0e55181f 2144 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
8237b185 2145
230f1322 2146 local = mkstemp(tmpfile);
916d081b
PB
2147 if (local < 0) {
2148 if (errno == EPERM)
2149 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2150 else
2151 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2152 }
8237b185
DB
2153
2154 memset(&stream, 0, sizeof(stream));
2155
2156 inflateInit(&stream);
2157
2158 SHA1_Init(&c);
2159
2160 do {
2161 ssize_t size;
70b9829e
DB
2162 if (*bufposn) {
2163 stream.avail_in = *bufposn;
96ad15ae 2164 stream.next_in = (unsigned char *) buffer;
70b9829e
DB
2165 do {
2166 stream.next_out = discard;
2167 stream.avail_out = sizeof(discard);
2168 ret = inflate(&stream, Z_SYNC_FLUSH);
2169 SHA1_Update(&c, discard, sizeof(discard) -
2170 stream.avail_out);
2171 } while (stream.avail_in && ret == Z_OK);
4d548150
LT
2172 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2173 die("unable to write sha1 file");
70b9829e
DB
2174 memmove(buffer, buffer + *bufposn - stream.avail_in,
2175 stream.avail_in);
2176 *bufposn = stream.avail_in;
2177 if (ret != Z_OK)
2178 break;
2179 }
93d26e4c 2180 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
8237b185
DB
2181 if (size <= 0) {
2182 close(local);
230f1322 2183 unlink(tmpfile);
8237b185
DB
2184 if (!size)
2185 return error("Connection closed?");
2186 perror("Reading from connection");
2187 return -1;
2188 }
70b9829e
DB
2189 *bufposn += size;
2190 } while (1);
8237b185
DB
2191 inflateEnd(&stream);
2192
b5b8d814 2193 fchmod(local, 0444);
0d315468
JM
2194 if (close(local) != 0)
2195 die("unable to write sha1 file");
8237b185
DB
2196 SHA1_Final(real_sha1, &c);
2197 if (ret != Z_STREAM_END) {
230f1322 2198 unlink(tmpfile);
8237b185
DB
2199 return error("File %s corrupted", sha1_to_hex(sha1));
2200 }
a89fccd2 2201 if (hashcmp(sha1, real_sha1)) {
230f1322 2202 unlink(tmpfile);
bd2afde8 2203 return error("File %s has bad hash", sha1_to_hex(sha1));
8237b185 2204 }
230f1322
LT
2205
2206 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
8237b185
DB
2207}
2208
bf592c50
DB
2209int has_pack_index(const unsigned char *sha1)
2210{
2211 struct stat st;
2212 if (stat(sha1_pack_index_name(sha1), &st))
2213 return 0;
2214 return 1;
2215}
2216
2217int has_pack_file(const unsigned char *sha1)
2218{
2219 struct stat st;
2220 if (stat(sha1_pack_name(sha1), &st))
2221 return 0;
2222 return 1;
2223}
2224
106d710b 2225int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
dade09c2
LT
2226{
2227 struct pack_entry e;
106d710b 2228 return find_pack_entry(sha1, &e, ignore_packed);
dade09c2
LT
2229}
2230
8237b185
DB
2231int has_sha1_file(const unsigned char *sha1)
2232{
8237b185 2233 struct stat st;
1f688557
JH
2234 struct pack_entry e;
2235
106d710b 2236 if (find_pack_entry(sha1, &e, NULL))
1f688557 2237 return 1;
ab90ea5d 2238 return find_sha1_file(sha1, &st) ? 1 : 0;
8237b185 2239}
74400e71 2240
e7332f96
BE
2241/*
2242 * reads from fd as long as possible into a supplied buffer of size bytes.
82e5a82f 2243 * If necessary the buffer's size is increased using realloc()
e7332f96
BE
2244 *
2245 * returns 0 if anything went fine and -1 otherwise
2246 *
2247 * NOTE: both buf and size may change, but even when -1 is returned
2248 * you still have to free() it yourself.
2249 */
2250int read_pipe(int fd, char** return_buf, unsigned long* return_size)
024510c8 2251{
e7332f96
BE
2252 char* buf = *return_buf;
2253 unsigned long size = *return_size;
8a912bcb 2254 ssize_t iret;
024510c8 2255 unsigned long off = 0;
e7332f96 2256
024510c8 2257 do {
e7332f96 2258 iret = xread(fd, buf + off, size - off);
024510c8
DB
2259 if (iret > 0) {
2260 off += iret;
2261 if (off == size) {
2262 size *= 2;
83572c1a 2263 buf = xrealloc(buf, size);
024510c8
DB
2264 }
2265 }
2266 } while (iret > 0);
e7332f96
BE
2267
2268 *return_buf = buf;
2269 *return_size = off;
2270
2271 if (iret < 0)
2272 return -1;
2273 return 0;
2274}
2275
2276int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2277{
2278 unsigned long size = 4096;
2d7320d0 2279 char *buf = xmalloc(size);
e7332f96 2280 int ret;
e7332f96
BE
2281
2282 if (read_pipe(fd, &buf, &size)) {
024510c8
DB
2283 free(buf);
2284 return -1;
2285 }
e7332f96 2286
024510c8 2287 if (!type)
8e440259 2288 type = blob_type;
024510c8 2289 if (write_object)
e7332f96 2290 ret = write_sha1_file(buf, size, type, sha1);
abdc3fc8
RS
2291 else
2292 ret = hash_sha1_file(buf, size, type, sha1);
024510c8
DB
2293 free(buf);
2294 return ret;
2295}
2296
edaec3fb 2297int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
53bca91a 2298 enum object_type type, const char *path)
74400e71 2299{
dc49cd76 2300 size_t size = xsize_t(st->st_size);
3a55602e 2301 void *buf = NULL;
6c510bee 2302 int ret, re_allocated = 0;
74400e71 2303
74400e71 2304 if (size)
c4712e45 2305 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
74400e71 2306 close(fd);
74400e71 2307
7672db20 2308 if (!type)
edaec3fb 2309 type = OBJ_BLOB;
6c510bee
LT
2310
2311 /*
2312 * Convert blobs to git internal format
2313 */
d0d8e14d 2314 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
6c510bee 2315 unsigned long nsize = size;
ac78e548
AR
2316 char *nbuf = convert_to_git(path, buf, &nsize);
2317 if (nbuf) {
2318 munmap(buf, size);
6c510bee
LT
2319 size = nsize;
2320 buf = nbuf;
2321 re_allocated = 1;
2322 }
2323 }
2324
7672db20 2325 if (write_object)
edaec3fb 2326 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 2327 else
edaec3fb 2328 ret = hash_sha1_file(buf, size, typename(type), sha1);
6c510bee
LT
2329 if (re_allocated) {
2330 free(buf);
2331 return ret;
2332 }
aac17941
LT
2333 if (size)
2334 munmap(buf, size);
2335 return ret;
74400e71 2336}
ec1fcc16
JH
2337
2338int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2339{
2340 int fd;
2341 char *target;
dc49cd76 2342 size_t len;
ec1fcc16
JH
2343
2344 switch (st->st_mode & S_IFMT) {
2345 case S_IFREG:
2346 fd = open(path, O_RDONLY);
2347 if (fd < 0)
2348 return error("open(\"%s\"): %s", path,
2349 strerror(errno));
53bca91a 2350 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
ec1fcc16
JH
2351 return error("%s: failed to insert into database",
2352 path);
2353 break;
2354 case S_IFLNK:
dc49cd76
SP
2355 len = xsize_t(st->st_size);
2356 target = xmalloc(len + 1);
2357 if (readlink(path, target, len + 1) != st->st_size) {
ec1fcc16
JH
2358 char *errstr = strerror(errno);
2359 free(target);
2360 return error("readlink(\"%s\"): %s", path,
2361 errstr);
2362 }
abdc3fc8 2363 if (!write_object)
dc49cd76
SP
2364 hash_sha1_file(target, len, blob_type, sha1);
2365 else if (write_sha1_file(target, len, blob_type, sha1))
ec1fcc16
JH
2366 return error("%s: failed to insert into database",
2367 path);
2368 free(target);
2369 break;
f35a6d3b
LT
2370 case S_IFDIR:
2371 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
2372 default:
2373 return error("%s: unsupported file type", path);
2374 }
2375 return 0;
2376}
a69e5429
JH
2377
2378int read_pack_header(int fd, struct pack_header *header)
2379{
2380 char *c = (char*)header;
2381 ssize_t remaining = sizeof(struct pack_header);
2382 do {
2383 ssize_t r = xread(fd, c, remaining);
2384 if (r <= 0)
2385 /* "eof before pack header was fully read" */
2386 return PH_ERROR_EOF;
2387 remaining -= r;
2388 c += r;
2389 } while (remaining > 0);
2390 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2391 /* "protocol error (pack signature mismatch detected)" */
2392 return PH_ERROR_PACK_SIGNATURE;
2393 if (!pack_version_ok(header->hdr_version))
2394 /* "protocol error (pack version unsupported)" */
2395 return PH_ERROR_PROTOCOL;
2396 return 0;
2397}