Correctly document the name of the global excludes file configuration
[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
192a6be2 36const signed char hexval_table[256] = {
e49521b5
JH
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;
1164f1e4
LT
512 unsigned long max_size = min_size;
513 if (nr)
514 max_size += (nr - 1)*8;
515 if (idx_size < min_size || idx_size > max_size) {
74e34e1f
NP
516 munmap(idx_map, idx_size);
517 return error("wrong index file size in %s", path);
518 }
519 if (idx_size != min_size) {
520 /* make sure we can deal with large pack offsets */
521 off_t x = 0x7fffffffUL, y = 0xffffffffUL;
522 if (x > (x + 1) || y > (y + 1)) {
523 munmap(idx_map, idx_size);
524 return error("pack too large for current definition of off_t in %s", path);
525 }
526 }
2d88451b 527 }
1f688557 528
74e34e1f 529 p->index_version = version;
42873078
NP
530 p->index_data = idx_map;
531 p->index_size = idx_size;
57059091 532 p->num_objects = nr;
1f688557
JH
533 return 0;
534}
535
11daf39b
SP
536static void scan_windows(struct packed_git *p,
537 struct packed_git **lru_p,
538 struct pack_window **lru_w,
539 struct pack_window **lru_l)
1f688557 540{
11daf39b
SP
541 struct pack_window *w, *w_l;
542
543 for (w_l = NULL, w = p->windows; w; w = w->next) {
544 if (!w->inuse_cnt) {
545 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
546 *lru_p = p;
547 *lru_w = w;
548 *lru_l = w_l;
54044bf8 549 }
54044bf8 550 }
11daf39b 551 w_l = w;
f9253394 552 }
11daf39b
SP
553}
554
d1efefa4 555static int unuse_one_window(struct packed_git *current, int keep_fd)
11daf39b
SP
556{
557 struct packed_git *p, *lru_p = NULL;
558 struct pack_window *lru_w = NULL, *lru_l = NULL;
559
560 if (current)
561 scan_windows(current, &lru_p, &lru_w, &lru_l);
562 for (p = packed_git; p; p = p->next)
563 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
564 if (lru_p) {
565 munmap(lru_w->base, lru_w->len);
566 pack_mapped -= lru_w->len;
567 if (lru_l)
568 lru_l->next = lru_w->next;
569 else {
570 lru_p->windows = lru_w->next;
d1efefa4 571 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
54044bf8
SP
572 close(lru_p->pack_fd);
573 lru_p->pack_fd = -1;
574 }
575 }
576 free(lru_w);
a53128b6 577 pack_open_windows--;
54044bf8
SP
578 return 1;
579 }
580 return 0;
f9253394
JH
581}
582
d1efefa4 583void release_pack_memory(size_t need, int fd)
97bfeb34
SP
584{
585 size_t cur = pack_mapped;
d1efefa4 586 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
97bfeb34
SP
587 ; /* nothing */
588}
589
03e79c88 590void unuse_pack(struct pack_window **w_cursor)
f9253394 591{
03e79c88
SP
592 struct pack_window *w = *w_cursor;
593 if (w) {
594 w->inuse_cnt--;
595 *w_cursor = NULL;
596 }
1f688557
JH
597}
598
3cf8b462
SP
599/*
600 * Do not call this directly as this leaks p->pack_fd on error return;
601 * call open_packed_git() instead.
602 */
603static int open_packed_git_1(struct packed_git *p)
1f688557 604{
9bc879c1
SP
605 struct stat st;
606 struct pack_header hdr;
607 unsigned char sha1[20];
608 unsigned char *idx_sha1;
2c039da8 609 long fd_flag;
9bc879c1
SP
610
611 p->pack_fd = open(p->pack_name, O_RDONLY);
612 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 613 return -1;
9bc879c1
SP
614
615 /* If we created the struct before we had the pack we lack size. */
bf592c50 616 if (!p->pack_size) {
bf592c50 617 if (!S_ISREG(st.st_mode))
072db278 618 return error("packfile %s not a regular file", p->pack_name);
bf592c50 619 p->pack_size = st.st_size;
9bc879c1 620 } else if (p->pack_size != st.st_size)
072db278 621 return error("packfile %s size changed", p->pack_name);
9bc879c1 622
2c039da8
JH
623 /* We leave these file descriptors open with sliding mmap;
624 * there is no point keeping them open across exec(), though.
625 */
626 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
627 if (fd_flag < 0)
072db278 628 return error("cannot determine file descriptor flags");
2c039da8
JH
629 fd_flag |= FD_CLOEXEC;
630 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 631 return error("cannot set FD_CLOEXEC");
2c039da8 632
9bc879c1 633 /* Verify we recognize this pack file format. */
e6e2bd62 634 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 635 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 636 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 637 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 638 if (!pack_version_ok(hdr.hdr_version))
072db278 639 return error("packfile %s is version %u and not supported"
9bc879c1
SP
640 " (try upgrading GIT to a newer version)",
641 p->pack_name, ntohl(hdr.hdr_version));
642
643 /* Verify the pack matches its index. */
57059091 644 if (p->num_objects != ntohl(hdr.hdr_entries))
072db278 645 return error("packfile %s claims to have %u objects"
57059091
NP
646 " while index indicates %u objects",
647 p->pack_name, ntohl(hdr.hdr_entries),
648 p->num_objects);
9bc879c1 649 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 650 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 651 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 652 return error("packfile %s signature is unavailable", p->pack_name);
42873078 653 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 654 if (hashcmp(sha1, idx_sha1))
072db278
SP
655 return error("packfile %s does not match index", p->pack_name);
656 return 0;
9bc879c1
SP
657}
658
3cf8b462
SP
659static int open_packed_git(struct packed_git *p)
660{
661 if (!open_packed_git_1(p))
662 return 0;
663 if (p->pack_fd != -1) {
664 close(p->pack_fd);
665 p->pack_fd = -1;
666 }
667 return -1;
668}
669
c4001d92 670static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
671{
672 /* We must promise at least 20 bytes (one hash) after the
673 * offset is available from this window, otherwise the offset
674 * is not actually in this window and a different window (which
675 * has that one hash excess) must be used. This is to support
676 * the object header and delta base parsing routines below.
677 */
678 off_t win_off = win->offset;
679 return win_off <= offset
680 && (offset + 20) <= (win_off + win->len);
681}
682
03e79c88
SP
683unsigned char* use_pack(struct packed_git *p,
684 struct pack_window **w_cursor,
c4001d92 685 off_t offset,
03e79c88 686 unsigned int *left)
9bc879c1 687{
60bb8b14 688 struct pack_window *win = *w_cursor;
03e79c88 689
072db278
SP
690 if (p->pack_fd == -1 && open_packed_git(p))
691 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
692
693 /* Since packfiles end in a hash of their content and its
694 * pointless to ask for an offset into the middle of that
695 * hash, and the in_window function above wouldn't match
696 * don't allow an offset too close to the end of the file.
697 */
698 if (offset > (p->pack_size - 20))
699 die("offset beyond end of packfile (truncated pack?)");
700
701 if (!win || !in_window(win, offset)) {
702 if (win)
703 win->inuse_cnt--;
704 for (win = p->windows; win; win = win->next) {
705 if (in_window(win, offset))
706 break;
707 }
708 if (!win) {
78a28df9 709 size_t window_align = packed_git_window_size / 2;
dc49cd76 710 off_t len;
60bb8b14 711 win = xcalloc(1, sizeof(*win));
78a28df9 712 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
713 len = p->pack_size - win->offset;
714 if (len > packed_git_window_size)
715 len = packed_git_window_size;
716 win->len = (size_t)len;
60bb8b14 717 pack_mapped += win->len;
11daf39b 718 while (packed_git_limit < pack_mapped
d1efefa4 719 && unuse_one_window(p, p->pack_fd))
60bb8b14 720 ; /* nothing */
c4712e45 721 win->base = xmmap(NULL, win->len,
60bb8b14
SP
722 PROT_READ, MAP_PRIVATE,
723 p->pack_fd, win->offset);
724 if (win->base == MAP_FAILED)
73b4e4be
SP
725 die("packfile %s cannot be mapped: %s",
726 p->pack_name,
727 strerror(errno));
a53128b6
SP
728 pack_mmap_calls++;
729 pack_open_windows++;
730 if (pack_mapped > peak_pack_mapped)
731 peak_pack_mapped = pack_mapped;
732 if (pack_open_windows > peak_pack_open_windows)
733 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
734 win->next = p->windows;
735 p->windows = win;
736 }
1f688557 737 }
03e79c88
SP
738 if (win != *w_cursor) {
739 win->last_used = pack_used_ctr++;
740 win->inuse_cnt++;
741 *w_cursor = win;
742 }
60bb8b14 743 offset -= win->offset;
03e79c88 744 if (left)
dc49cd76 745 *left = win->len - xsize_t(offset);
03e79c88 746 return win->base + offset;
1f688557
JH
747}
748
42873078 749struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557
JH
750{
751 struct stat st;
42873078 752 struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
1f688557 753
42873078
NP
754 /*
755 * Make sure a corresponding .pack file exists and that
756 * the index looks sane.
757 */
758 path_len -= strlen(".idx");
759 if (path_len < 1)
1f688557 760 return NULL;
42873078
NP
761 memcpy(p->pack_name, path, path_len);
762 strcpy(p->pack_name + path_len, ".pack");
763 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode) ||
764 check_packed_git_idx(path, p)) {
765 free(p);
1f688557
JH
766 return NULL;
767 }
42873078 768
1f688557
JH
769 /* ok, it looks sane as far as we can check without
770 * actually mapping the pack file.
771 */
1f688557 772 p->pack_size = st.st_size;
1f688557 773 p->next = NULL;
c41ee586 774 p->windows = NULL;
9bc879c1 775 p->pack_fd = -1;
9d835df2 776 p->pack_local = local;
b867092f 777 p->mtime = st.st_mtime;
42873078
NP
778 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
779 hashclr(p->sha1);
1f688557
JH
780 return p;
781}
782
bf592c50 783struct packed_git *parse_pack_index(unsigned char *sha1)
c508df5e
DB
784{
785 char *path = sha1_pack_index_name(sha1);
786 return parse_pack_index_file(sha1, path);
787}
788
42873078
NP
789struct packed_git *parse_pack_index_file(const unsigned char *sha1,
790 const char *idx_path)
bf592c50 791{
42873078
NP
792 const char *path = sha1_pack_name(sha1);
793 struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
bf592c50 794
42873078
NP
795 if (check_packed_git_idx(idx_path, p)) {
796 free(p);
bf592c50 797 return NULL;
42873078 798 }
bf592c50 799
bf592c50 800 strcpy(p->pack_name, path);
bf592c50 801 p->pack_size = 0;
bf592c50 802 p->next = NULL;
c41ee586 803 p->windows = NULL;
9bc879c1 804 p->pack_fd = -1;
e702496e 805 hashcpy(p->sha1, sha1);
bf592c50
DB
806 return p;
807}
808
809void install_packed_git(struct packed_git *pack)
810{
811 pack->next = packed_git;
812 packed_git = pack;
813}
814
9d835df2 815static void prepare_packed_git_one(char *objdir, int local)
1f688557
JH
816{
817 char path[PATH_MAX];
818 int len;
819 DIR *dir;
820 struct dirent *de;
821
822 sprintf(path, "%s/pack", objdir);
823 len = strlen(path);
824 dir = opendir(path);
b5b16990 825 if (!dir) {
26125f6b 826 if (errno != ENOENT)
bd2afde8 827 error("unable to open object pack directory: %s: %s",
26125f6b 828 path, strerror(errno));
1f688557 829 return;
b5b16990 830 }
1f688557
JH
831 path[len++] = '/';
832 while ((de = readdir(dir)) != NULL) {
833 int namelen = strlen(de->d_name);
834 struct packed_git *p;
835
5bb1cda5 836 if (!has_extension(de->d_name, ".idx"))
1f688557
JH
837 continue;
838
54a15a8d 839 /* Don't reopen a pack we already have. */
1f688557 840 strcpy(path + len, de->d_name);
86f7780c
JK
841 for (p = packed_git; p; p = p->next) {
842 if (!memcmp(path, p->pack_name, len + namelen - 4))
843 break;
844 }
845 if (p)
846 continue;
54a15a8d
SP
847 /* See if it really is a valid .idx file with corresponding
848 * .pack file that we can map.
849 */
9d835df2 850 p = add_packed_git(path, len + namelen, local);
1f688557
JH
851 if (!p)
852 continue;
625e9421 853 install_packed_git(p);
1f688557 854 }
5b35bcd5 855 closedir(dir);
1f688557
JH
856}
857
b867092f
JH
858static int sort_pack(const void *a_, const void *b_)
859{
860 struct packed_git *a = *((struct packed_git **)a_);
861 struct packed_git *b = *((struct packed_git **)b_);
862 int st;
863
864 /*
865 * Local packs tend to contain objects specific to our
866 * variant of the project than remote ones. In addition,
867 * remote ones could be on a network mounted filesystem.
868 * Favor local ones for these reasons.
869 */
870 st = a->pack_local - b->pack_local;
871 if (st)
872 return -st;
873
874 /*
875 * Younger packs tend to contain more recent objects,
876 * and more recent objects tend to get accessed more
877 * often.
878 */
879 if (a->mtime < b->mtime)
880 return 1;
881 else if (a->mtime == b->mtime)
882 return 0;
883 return -1;
884}
885
886static void rearrange_packed_git(void)
887{
888 struct packed_git **ary, *p;
889 int i, n;
890
891 for (n = 0, p = packed_git; p; p = p->next)
892 n++;
893 if (n < 2)
894 return;
895
896 /* prepare an array of packed_git for easier sorting */
897 ary = xcalloc(n, sizeof(struct packed_git *));
898 for (n = 0, p = packed_git; p; p = p->next)
899 ary[n++] = p;
900
901 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
902
903 /* link them back again */
904 for (i = 0; i < n - 1; i++)
905 ary[i]->next = ary[i + 1];
906 ary[n - 1]->next = NULL;
907 packed_git = ary[0];
908
909 free(ary);
910}
911
637cdd9d 912static int prepare_packed_git_run_once = 0;
9a217f2a 913void prepare_packed_git(void)
1f688557 914{
d5a63b99 915 struct alternate_object_database *alt;
1f688557 916
637cdd9d 917 if (prepare_packed_git_run_once)
1f688557 918 return;
9d835df2 919 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 920 prepare_alt_odb();
d5a63b99 921 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 922 alt->name[-1] = 0;
9d835df2 923 prepare_packed_git_one(alt->base, 0);
1494e038 924 alt->name[-1] = '/';
1f688557 925 }
b867092f 926 rearrange_packed_git();
637cdd9d
JK
927 prepare_packed_git_run_once = 1;
928}
929
fc04c412 930void reprepare_packed_git(void)
637cdd9d
JK
931{
932 prepare_packed_git_run_once = 0;
933 prepare_packed_git();
1f688557
JH
934}
935
5d6ccf5c 936int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
0fcfd160
LT
937{
938 unsigned char real_sha1[20];
7cfb5f36 939 hash_sha1_file(map, size, type, real_sha1);
a89fccd2 940 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
941}
942
bb6b8e4f 943void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 944{
0fcfd160
LT
945 struct stat st;
946 void *map;
144bde78 947 int fd;
ace1534d
JH
948 char *filename = find_sha1_file(sha1, &st);
949
950 if (!filename) {
ace1534d
JH
951 return NULL;
952 }
0fcfd160 953
144bde78 954 fd = open(filename, O_RDONLY | sha1_file_open_flag);
0fcfd160 955 if (fd < 0) {
144bde78
LT
956 /* See if it works without O_NOATIME */
957 switch (sha1_file_open_flag) {
958 default:
959 fd = open(filename, O_RDONLY);
960 if (fd >= 0)
961 break;
962 /* Fallthrough */
963 case 0:
144bde78
LT
964 return NULL;
965 }
966
1f688557
JH
967 /* If it failed once, it will probably fail again.
968 * Stop using O_NOATIME
969 */
144bde78 970 sha1_file_open_flag = 0;
0fcfd160 971 }
dc49cd76
SP
972 *size = xsize_t(st.st_size);
973 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
0fcfd160 974 close(fd);
0fcfd160
LT
975 return map;
976}
977
bb6b8e4f
JH
978int legacy_loose_object(unsigned char *map)
979{
980 unsigned int word;
981
982 /*
983 * Is it a zlib-compressed buffer? If so, the first byte
984 * must be 0x78 (15-bit window size, deflated), and the
985 * first 16-bit word is evenly divisible by 31
986 */
987 word = (map[0] << 8) + map[1];
988 if (map[0] == 0x78 && !(word % 31))
989 return 1;
990 else
991 return 0;
992}
993
72518e9c 994unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 995{
ad1ed5ee 996 unsigned shift;
93821bd9 997 unsigned char c;
93821bd9 998 unsigned long size;
ad1ed5ee
JH
999 unsigned long used = 0;
1000
1001 c = buf[used++];
1002 *type = (c >> 4) & 7;
1003 size = c & 15;
1004 shift = 4;
1005 while (c & 0x80) {
1006 if (len <= used)
1007 return 0;
1008 if (sizeof(long) * 8 <= shift)
1009 return 0;
1010 c = buf[used++];
1011 size += (c & 0x7f) << shift;
1012 shift += 7;
1013 }
1014 *sizep = size;
1015 return used;
1016}
1017
1018static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1019{
1020 unsigned long size, used;
1021 static const char valid_loose_object_type[8] = {
1022 0, /* OBJ_EXT */
1023 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1024 0, /* "delta" and others are invalid in a loose object */
93821bd9 1025 };
ad1ed5ee 1026 enum object_type type;
93821bd9 1027
c4483576
LT
1028 /* Get the data stream */
1029 memset(stream, 0, sizeof(*stream));
1030 stream->next_in = map;
1031 stream->avail_in = mapsize;
1032 stream->next_out = buffer;
93821bd9
LT
1033 stream->avail_out = bufsiz;
1034
bb6b8e4f 1035 if (legacy_loose_object(map)) {
93821bd9
LT
1036 inflateInit(stream);
1037 return inflate(stream, 0);
1038 }
1039
ad1ed5ee
JH
1040 used = unpack_object_header_gently(map, mapsize, &type, &size);
1041 if (!used || !valid_loose_object_type[type])
93821bd9 1042 return -1;
ad1ed5ee
JH
1043 map += used;
1044 mapsize -= used;
93821bd9
LT
1045
1046 /* Set up the stream for the rest.. */
1047 stream->next_in = map;
1048 stream->avail_in = mapsize;
c4483576 1049 inflateInit(stream);
93821bd9
LT
1050
1051 /* And generate the fake traditional header */
ad1ed5ee 1052 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
df843662 1053 typename(type), size);
93821bd9 1054 return 0;
c4483576
LT
1055}
1056
7efbff75 1057static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1058{
1059 int bytes = strlen(buffer) + 1;
d565b341 1060 unsigned char *buf = xmalloc(1+size);
93821bd9 1061 unsigned long n;
7efbff75 1062 int status = Z_OK;
5180cacc 1063
93821bd9
LT
1064 n = stream->total_out - bytes;
1065 if (n > size)
1066 n = size;
1067 memcpy(buf, (char *) buffer + bytes, n);
1068 bytes = n;
456cdf6e
LT
1069 if (bytes <= size) {
1070 /*
1071 * The above condition must be (bytes <= size), not
1072 * (bytes < size). In other words, even though we
1073 * expect no more output and set avail_out to zer0,
1074 * the input zlib stream may have bytes that express
1075 * "this concludes the stream", and we *do* want to
1076 * eat that input.
1077 *
1078 * Otherwise we would not be able to test that we
1079 * consumed all the input to reach the expected size;
1080 * we also want to check that zlib tells us that all
1081 * went well with status == Z_STREAM_END at the end.
1082 */
5180cacc
LT
1083 stream->next_out = buf + bytes;
1084 stream->avail_out = size - bytes;
7efbff75
JH
1085 while (status == Z_OK)
1086 status = inflate(stream, Z_FINISH);
5180cacc
LT
1087 }
1088 buf[size] = 0;
456cdf6e 1089 if (status == Z_STREAM_END && !stream->avail_in) {
7efbff75
JH
1090 inflateEnd(stream);
1091 return buf;
1092 }
1093
1094 if (status < 0)
1095 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1096 else if (stream->avail_in)
1097 error("garbage at end of loose object '%s'",
1098 sha1_to_hex(sha1));
1099 free(buf);
1100 return NULL;
5180cacc
LT
1101}
1102
1103/*
1104 * We used to just use "sscanf()", but that's actually way
1105 * too permissive for what we want to check. So do an anal
1106 * object header parse by hand.
1107 */
21666f1a 1108static int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1109{
21666f1a 1110 char type[10];
5180cacc
LT
1111 int i;
1112 unsigned long size;
1113
1114 /*
1115 * The type can be at most ten bytes (including the
1116 * terminating '\0' that we add), and is followed by
21666f1a 1117 * a space.
5180cacc 1118 */
21666f1a 1119 i = 0;
5180cacc
LT
1120 for (;;) {
1121 char c = *hdr++;
1122 if (c == ' ')
1123 break;
21666f1a
NP
1124 type[i++] = c;
1125 if (i >= sizeof(type))
5180cacc 1126 return -1;
5180cacc 1127 }
21666f1a 1128 type[i] = 0;
5180cacc
LT
1129
1130 /*
1131 * The length must follow immediately, and be in canonical
1132 * decimal format (ie "010" is not valid).
1133 */
1134 size = *hdr++ - '0';
1135 if (size > 9)
1136 return -1;
1137 if (size) {
1138 for (;;) {
1139 unsigned long c = *hdr - '0';
1140 if (c > 9)
1141 break;
1142 hdr++;
1143 size = size * 10 + c;
1144 }
1145 }
1146 *sizep = size;
1147
1148 /*
1149 * The length must be followed by a zero byte
1150 */
21666f1a 1151 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1152}
1153
7efbff75 1154static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1155{
5180cacc 1156 int ret;
0fcfd160 1157 z_stream stream;
5180cacc 1158 char hdr[8192];
0fcfd160 1159
5180cacc 1160 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1161 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1162 return NULL;
1163
7efbff75 1164 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1165}
1166
54dab52a
NP
1167unsigned long get_size_from_delta(struct packed_git *p,
1168 struct pack_window **w_curs,
1169 off_t curpos)
1170{
1171 const unsigned char *data;
1172 unsigned char delta_head[20], *in;
1173 z_stream stream;
1174 int st;
1175
1176 memset(&stream, 0, sizeof(stream));
1177 stream.next_out = delta_head;
1178 stream.avail_out = sizeof(delta_head);
1179
1180 inflateInit(&stream);
1181 do {
1182 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1183 stream.next_in = in;
1184 st = inflate(&stream, Z_FINISH);
1185 curpos += stream.next_in - in;
1186 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1187 stream.total_out < sizeof(delta_head));
1188 inflateEnd(&stream);
1189 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1190 die("delta data unpack-initial failed");
1191
1192 /* Examine the initial part of the delta to figure out
1193 * the result size.
1194 */
1195 data = delta_head;
1196
1197 /* ignore base size */
1198 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1199
1200 /* Read the result size */
1201 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1202}
1203
c4001d92 1204static off_t get_delta_base(struct packed_git *p,
03e79c88 1205 struct pack_window **w_curs,
c4001d92 1206 off_t *curpos,
21666f1a 1207 enum object_type type,
c4001d92 1208 off_t delta_obj_offset)
eb32d236 1209{
2b87c45b 1210 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1211 off_t base_offset;
eb32d236 1212
8d8a4ea5
SP
1213 /* use_pack() assured us we have [base_info, base_info + 20)
1214 * as a range that we can look at without walking off the
1215 * end of the mapped window. Its actually the hash size
1216 * that is assured. An OFS_DELTA longer than the hash size
1217 * is stupid, as then a REF_DELTA would be smaller to store.
1218 */
21666f1a 1219 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1220 unsigned used = 0;
1221 unsigned char c = base_info[used++];
1222 base_offset = c & 127;
1223 while (c & 128) {
1224 base_offset += 1;
8723f216 1225 if (!base_offset || MSB(base_offset, 7))
eb32d236
NP
1226 die("offset value overflow for delta base object");
1227 c = base_info[used++];
1228 base_offset = (base_offset << 7) + (c & 127);
1229 }
1230 base_offset = delta_obj_offset - base_offset;
1231 if (base_offset >= delta_obj_offset)
1232 die("delta base offset out of bound");
2b87c45b 1233 *curpos += used;
21666f1a 1234 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1235 /* The base entry _must_ be in the same pack */
1236 base_offset = find_pack_entry_one(base_info, p);
1237 if (!base_offset)
1238 die("failed to find delta-pack base object %s",
1239 sha1_to_hex(base_info));
2b87c45b 1240 *curpos += 20;
eb32d236
NP
1241 } else
1242 die("I am totally screwed");
2b87c45b 1243 return base_offset;
eb32d236
NP
1244}
1245
f3bf9224 1246/* forward declaration for a mutually recursive function */
c4001d92 1247static int packed_object_info(struct packed_git *p, off_t offset,
21666f1a 1248 unsigned long *sizep);
f3bf9224 1249
43057304 1250static int packed_delta_info(struct packed_git *p,
03e79c88 1251 struct pack_window **w_curs,
c4001d92 1252 off_t curpos,
21666f1a 1253 enum object_type type,
c4001d92 1254 off_t obj_offset,
43057304 1255 unsigned long *sizep)
5db47c2b 1256{
c4001d92 1257 off_t base_offset;
f3bf9224 1258
21666f1a
NP
1259 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1260 type = packed_object_info(p, base_offset, NULL);
f3bf9224 1261
c62266f3
JH
1262 /* We choose to only get the type of the base object and
1263 * ignore potentially corrupt pack file that expects the delta
1264 * based on a base with a wrong size. This saves tons of
1265 * inflate() calls.
1266 */
54dab52a
NP
1267 if (sizep)
1268 *sizep = get_size_from_delta(p, w_curs, curpos);
21666f1a
NP
1269
1270 return type;
5db47c2b
JH
1271}
1272
2b87c45b
NP
1273static int unpack_object_header(struct packed_git *p,
1274 struct pack_window **w_curs,
c4001d92 1275 off_t *curpos,
2b87c45b 1276 unsigned long *sizep)
a733cb60 1277{
03e79c88
SP
1278 unsigned char *base;
1279 unsigned int left;
ad1ed5ee 1280 unsigned long used;
2b87c45b 1281 enum object_type type;
a733cb60 1282
8d8a4ea5
SP
1283 /* use_pack() assures us we have [base, base + 20) available
1284 * as a range that we can look at at. (Its actually the hash
3dff5379 1285 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1286 * the maximum deflated object size is 2^137, which is just
1287 * insane, so we know won't exceed what we have been given.
1288 */
2b87c45b
NP
1289 base = use_pack(p, w_curs, *curpos, &left);
1290 used = unpack_object_header_gently(base, left, &type, sizep);
ad1ed5ee
JH
1291 if (!used)
1292 die("object offset outside of pack file");
2b87c45b 1293 *curpos += used;
ad1ed5ee 1294
2b87c45b 1295 return type;
a733cb60
LT
1296}
1297
21666f1a 1298const char *packed_object_info_detail(struct packed_git *p,
c4001d92 1299 off_t obj_offset,
21666f1a
NP
1300 unsigned long *size,
1301 unsigned long *store_size,
1302 unsigned int *delta_chain_length,
1303 unsigned char *base_sha1)
ad8c80a5 1304{
03e79c88 1305 struct pack_window *w_curs = NULL;
c4001d92
SP
1306 off_t curpos;
1307 unsigned long dummy;
43057304 1308 unsigned char *next_sha1;
21666f1a 1309 enum object_type type;
ad8c80a5 1310
43057304 1311 *delta_chain_length = 0;
2b87c45b 1312 curpos = obj_offset;
21666f1a 1313 type = unpack_object_header(p, &w_curs, &curpos, size);
43057304
NP
1314
1315 for (;;) {
21666f1a 1316 switch (type) {
43057304 1317 default:
08a19d87 1318 die("pack %s contains unknown object type %d",
21666f1a 1319 p->pack_name, type);
43057304
NP
1320 case OBJ_COMMIT:
1321 case OBJ_TREE:
1322 case OBJ_BLOB:
1323 case OBJ_TAG:
43057304 1324 *store_size = 0; /* notyet */
03e79c88 1325 unuse_pack(&w_curs);
21666f1a 1326 return typename(type);
eb32d236 1327 case OBJ_OFS_DELTA:
21666f1a 1328 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
eb32d236 1329 if (*delta_chain_length == 0) {
2b87c45b 1330 /* TODO: find base_sha1 as pointed by curpos */
30fee062 1331 hashclr(base_sha1);
eb32d236
NP
1332 }
1333 break;
1334 case OBJ_REF_DELTA:
2b87c45b 1335 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
43057304
NP
1336 if (*delta_chain_length == 0)
1337 hashcpy(base_sha1, next_sha1);
2b87c45b 1338 obj_offset = find_pack_entry_one(next_sha1, p);
43057304
NP
1339 break;
1340 }
43057304 1341 (*delta_chain_length)++;
2b87c45b 1342 curpos = obj_offset;
21666f1a 1343 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
ad8c80a5 1344 }
ad8c80a5
JH
1345}
1346
c4001d92 1347static int packed_object_info(struct packed_git *p, off_t obj_offset,
21666f1a 1348 unsigned long *sizep)
1f688557 1349{
03e79c88 1350 struct pack_window *w_curs = NULL;
c4001d92
SP
1351 unsigned long size;
1352 off_t curpos = obj_offset;
21666f1a 1353 enum object_type type;
5db47c2b 1354
21666f1a 1355 type = unpack_object_header(p, &w_curs, &curpos, &size);
5db47c2b 1356
21666f1a 1357 switch (type) {
eb32d236
NP
1358 case OBJ_OFS_DELTA:
1359 case OBJ_REF_DELTA:
21666f1a
NP
1360 type = packed_delta_info(p, &w_curs, curpos,
1361 type, obj_offset, sizep);
2b87c45b 1362 break;
a733cb60 1363 case OBJ_COMMIT:
a733cb60 1364 case OBJ_TREE:
a733cb60 1365 case OBJ_BLOB:
a733cb60 1366 case OBJ_TAG:
2b87c45b
NP
1367 if (sizep)
1368 *sizep = size;
a69d0943 1369 break;
1f688557 1370 default:
08a19d87 1371 die("pack %s contains unknown object type %d",
21666f1a 1372 p->pack_name, type);
1f688557 1373 }
2b87c45b 1374 unuse_pack(&w_curs);
21666f1a 1375 return type;
1f688557
JH
1376}
1377
eb950c19 1378static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1379 struct pack_window **w_curs,
c4001d92 1380 off_t curpos,
eb950c19 1381 unsigned long size)
de530aaa
SP
1382{
1383 int st;
1384 z_stream stream;
079afb18 1385 unsigned char *buffer, *in;
de530aaa
SP
1386
1387 buffer = xmalloc(size + 1);
1388 buffer[size] = 0;
1389 memset(&stream, 0, sizeof(stream));
de530aaa
SP
1390 stream.next_out = buffer;
1391 stream.avail_out = size;
1392
1393 inflateInit(&stream);
079afb18 1394 do {
2b87c45b 1395 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18
SP
1396 stream.next_in = in;
1397 st = inflate(&stream, Z_FINISH);
2b87c45b 1398 curpos += stream.next_in - in;
079afb18 1399 } while (st == Z_OK || st == Z_BUF_ERROR);
de530aaa
SP
1400 inflateEnd(&stream);
1401 if ((st != Z_STREAM_END) || stream.total_out != size) {
1402 free(buffer);
1403 return NULL;
1404 }
1405
1406 return buffer;
1407}
1408
e5e01619
LT
1409#define MAX_DELTA_CACHE (256)
1410
18bdec11 1411static size_t delta_base_cached;
5e08ecbf
NP
1412
1413static struct delta_base_cache_lru_list {
1414 struct delta_base_cache_lru_list *prev;
1415 struct delta_base_cache_lru_list *next;
1416} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1417
e5e01619 1418static struct delta_base_cache_entry {
5e08ecbf
NP
1419 struct delta_base_cache_lru_list lru;
1420 void *data;
e5e01619
LT
1421 struct packed_git *p;
1422 off_t base_offset;
1423 unsigned long size;
e5e01619
LT
1424 enum object_type type;
1425} delta_base_cache[MAX_DELTA_CACHE];
1426
1427static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1428{
1429 unsigned long hash;
1430
1431 hash = (unsigned long)p + (unsigned long)base_offset;
1432 hash += (hash >> 8) + (hash >> 16);
3358004a 1433 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1434}
1435
62f255ad 1436static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 1437 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 1438{
e5e01619
LT
1439 void *ret;
1440 unsigned long hash = pack_entry_hash(p, base_offset);
1441 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1442
1443 ret = ent->data;
1444 if (ret && ent->p == p && ent->base_offset == base_offset)
1445 goto found_cache_entry;
62f255ad 1446 return unpack_entry(p, base_offset, type, base_size);
e5e01619
LT
1447
1448found_cache_entry:
18bdec11 1449 if (!keep_cache) {
a0cba108 1450 ent->data = NULL;
5e08ecbf
NP
1451 ent->lru.next->prev = ent->lru.prev;
1452 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
1453 delta_base_cached -= ent->size;
1454 }
a0cba108
NP
1455 else {
1456 ret = xmalloc(ent->size + 1);
1457 memcpy(ret, ent->data, ent->size);
1458 ((char *)ret)[ent->size] = 0;
1459 }
e5e01619
LT
1460 *type = ent->type;
1461 *base_size = ent->size;
1462 return ret;
62f255ad
LT
1463}
1464
18bdec11
SP
1465static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1466{
1467 if (ent->data) {
1468 free(ent->data);
1469 ent->data = NULL;
5e08ecbf
NP
1470 ent->lru.next->prev = ent->lru.prev;
1471 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
1472 delta_base_cached -= ent->size;
1473 }
1474}
1475
62f255ad
LT
1476static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1477 void *base, unsigned long base_size, enum object_type type)
1478{
5e08ecbf 1479 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 1480 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 1481 struct delta_base_cache_lru_list *lru;
e5e01619 1482
18bdec11
SP
1483 release_delta_base_cache(ent);
1484 delta_base_cached += base_size;
5e08ecbf
NP
1485
1486 for (lru = delta_base_cache_lru.next;
1487 delta_base_cached > delta_base_cache_limit
1488 && lru != &delta_base_cache_lru;
1489 lru = lru->next) {
1490 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
1491 if (f->type == OBJ_BLOB)
1492 release_delta_base_cache(f);
1493 }
5e08ecbf
NP
1494 for (lru = delta_base_cache_lru.next;
1495 delta_base_cached > delta_base_cache_limit
1496 && lru != &delta_base_cache_lru;
1497 lru = lru->next) {
1498 struct delta_base_cache_entry *f = (void *)lru;
1499 release_delta_base_cache(f);
1500 }
18bdec11 1501
e5e01619
LT
1502 ent->p = p;
1503 ent->base_offset = base_offset;
1504 ent->type = type;
1505 ent->data = base;
1506 ent->size = base_size;
5e08ecbf
NP
1507 ent->lru.next = &delta_base_cache_lru;
1508 ent->lru.prev = delta_base_cache_lru.prev;
1509 delta_base_cache_lru.prev->next = &ent->lru;
1510 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
1511}
1512
eb950c19 1513static void *unpack_delta_entry(struct packed_git *p,
03e79c88 1514 struct pack_window **w_curs,
c4001d92 1515 off_t curpos,
1f688557 1516 unsigned long delta_size,
c4001d92 1517 off_t obj_offset,
21666f1a 1518 enum object_type *type,
eb950c19 1519 unsigned long *sizep)
1f688557 1520{
7c3e8be3 1521 void *delta_data, *result, *base;
c4001d92
SP
1522 unsigned long base_size;
1523 off_t base_offset;
43057304 1524
21666f1a 1525 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
a0cba108 1526 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
67686d95 1527 if (!base)
c4001d92
SP
1528 die("failed to read delta base object"
1529 " at %"PRIuMAX" from %s",
1530 (uintmax_t)base_offset, p->pack_name);
67686d95 1531
2b87c45b 1532 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1f688557
JH
1533 result = patch_delta(base, base_size,
1534 delta_data, delta_size,
21666f1a 1535 sizep);
1f688557
JH
1536 if (!result)
1537 die("failed to apply delta");
1538 free(delta_data);
62f255ad 1539 add_delta_base_cache(p, base_offset, base, base_size, *type);
1f688557
JH
1540 return result;
1541}
1542
c4001d92 1543void *unpack_entry(struct packed_git *p, off_t obj_offset,
21666f1a 1544 enum object_type *type, unsigned long *sizep)
1f688557 1545{
03e79c88 1546 struct pack_window *w_curs = NULL;
c4001d92 1547 off_t curpos = obj_offset;
21666f1a 1548 void *data;
1f688557 1549
21666f1a
NP
1550 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1551 switch (*type) {
eb32d236
NP
1552 case OBJ_OFS_DELTA:
1553 case OBJ_REF_DELTA:
21666f1a
NP
1554 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1555 obj_offset, type, sizep);
4d703a1a 1556 break;
a733cb60 1557 case OBJ_COMMIT:
a733cb60 1558 case OBJ_TREE:
a733cb60 1559 case OBJ_BLOB:
a733cb60 1560 case OBJ_TAG:
21666f1a 1561 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
4d703a1a 1562 break;
1f688557 1563 default:
21666f1a 1564 die("unknown object type %i in %s", *type, p->pack_name);
1f688557 1565 }
03e79c88 1566 unuse_pack(&w_curs);
21666f1a 1567 return data;
1f688557
JH
1568}
1569
d72308e0
NP
1570const unsigned char *nth_packed_object_sha1(const struct packed_git *p,
1571 uint32_t n)
9a217f2a 1572{
42873078 1573 const unsigned char *index = p->index_data;
57059091 1574 if (n >= p->num_objects)
d72308e0 1575 return NULL;
74e34e1f
NP
1576 index += 4 * 256;
1577 if (p->index_version == 1) {
1578 return index + 24 * n + 4;
1579 } else {
1580 index += 8;
1581 return index + 20 * n;
1582 }
1583}
1584
1585static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1586{
1587 const unsigned char *index = p->index_data;
1588 index += 4 * 256;
1589 if (p->index_version == 1) {
1590 return ntohl(*((uint32_t *)(index + 24 * n)));
1591 } else {
1592 uint32_t off;
1593 index += 8 + p->num_objects * (20 + 4);
1594 off = ntohl(*((uint32_t *)(index + 4 * n)));
1595 if (!(off & 0x80000000))
1596 return off;
1597 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1598 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1599 ntohl(*((uint32_t *)(index + 4)));
1600 }
9a217f2a
JH
1601}
1602
c4001d92 1603off_t find_pack_entry_one(const unsigned char *sha1,
43057304 1604 struct packed_git *p)
1f688557 1605{
42873078 1606 const uint32_t *level1_ofs = p->index_data;
42873078 1607 const unsigned char *index = p->index_data;
74e34e1f 1608 unsigned hi, lo;
42873078 1609
74e34e1f
NP
1610 if (p->index_version > 1) {
1611 level1_ofs += 2;
1612 index += 8;
1613 }
42873078 1614 index += 4 * 256;
74e34e1f
NP
1615 hi = ntohl(level1_ofs[*sha1]);
1616 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1f688557
JH
1617
1618 do {
74e34e1f
NP
1619 unsigned mi = (lo + hi) / 2;
1620 unsigned x = (p->index_version > 1) ? (mi * 20) : (mi * 24 + 4);
1621 int cmp = hashcmp(index + x, sha1);
43057304 1622 if (!cmp)
74e34e1f 1623 return nth_packed_object_offset(p, mi);
1f688557
JH
1624 if (cmp > 0)
1625 hi = mi;
1626 else
1627 lo = mi+1;
1628 } while (lo < hi);
1629 return 0;
1630}
1631
d4ff6d92
SP
1632static int matches_pack_name(struct packed_git *p, const char *ig)
1633{
1634 const char *last_c, *c;
1635
1636 if (!strcmp(p->pack_name, ig))
1637 return 0;
1638
1639 for (c = p->pack_name, last_c = c; *c;)
1640 if (*c == '/')
1641 last_c = ++c;
1642 else
1643 ++c;
1644 if (!strcmp(last_c, ig))
1645 return 0;
1646
1647 return 1;
1648}
1649
106d710b 1650static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1f688557
JH
1651{
1652 struct packed_git *p;
c4001d92 1653 off_t offset;
43057304 1654
1f688557
JH
1655 prepare_packed_git();
1656
1657 for (p = packed_git; p; p = p->next) {
106d710b
JH
1658 if (ignore_packed) {
1659 const char **ig;
1660 for (ig = ignore_packed; *ig; ig++)
d4ff6d92 1661 if (!matches_pack_name(p, *ig))
106d710b
JH
1662 break;
1663 if (*ig)
1664 continue;
1665 }
43057304
NP
1666 offset = find_pack_entry_one(sha1, p);
1667 if (offset) {
c715f783
SP
1668 /*
1669 * We are about to tell the caller where they can
1670 * locate the requested object. We better make
1671 * sure the packfile is still here and can be
1672 * accessed before supplying that answer, as
1673 * it may have been deleted since the index
1674 * was loaded!
1675 */
1676 if (p->pack_fd == -1 && open_packed_git(p)) {
1677 error("packfile %s cannot be accessed", p->pack_name);
1678 continue;
1679 }
43057304
NP
1680 e->offset = offset;
1681 e->p = p;
1682 hashcpy(e->sha1, sha1);
1f688557 1683 return 1;
43057304 1684 }
1f688557
JH
1685 }
1686 return 0;
1687}
1688
bf592c50
DB
1689struct packed_git *find_sha1_pack(const unsigned char *sha1,
1690 struct packed_git *packs)
1691{
1692 struct packed_git *p;
bf592c50
DB
1693
1694 for (p = packs; p; p = p->next) {
43057304 1695 if (find_pack_entry_one(sha1, p))
bf592c50
DB
1696 return p;
1697 }
1698 return NULL;
21666f1a 1699
bf592c50
DB
1700}
1701
21666f1a 1702static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
65c2e0c3 1703{
36e4d74a 1704 int status;
65c2e0c3
JH
1705 unsigned long mapsize, size;
1706 void *map;
1707 z_stream stream;
d65a16f6 1708 char hdr[32];
65c2e0c3 1709
bb6b8e4f 1710 map = map_sha1_file(sha1, &mapsize);
f0df4ed5
JS
1711 if (!map)
1712 return error("unable to find %s", sha1_to_hex(sha1));
36e4d74a
JH
1713 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1714 status = error("unable to unpack %s header",
1715 sha1_to_hex(sha1));
21666f1a 1716 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 1717 status = error("unable to parse %s header", sha1_to_hex(sha1));
21666f1a
NP
1718 else if (sizep)
1719 *sizep = size;
65c2e0c3
JH
1720 inflateEnd(&stream);
1721 munmap(map, mapsize);
1722 return status;
1723}
1724
21666f1a 1725int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
f0df4ed5 1726{
f0df4ed5
JS
1727 struct pack_entry e;
1728
1729 if (!find_pack_entry(sha1, &e, NULL)) {
1730 reprepare_packed_git();
1731 if (!find_pack_entry(sha1, &e, NULL))
21666f1a 1732 return sha1_loose_object_info(sha1, sizep);
f0df4ed5 1733 }
21666f1a 1734 return packed_object_info(e.p, e.offset, sizep);
f0df4ed5
JS
1735}
1736
21666f1a
NP
1737static void *read_packed_sha1(const unsigned char *sha1,
1738 enum object_type *type, unsigned long *size)
1f688557
JH
1739{
1740 struct pack_entry e;
1741
8276c007 1742 if (!find_pack_entry(sha1, &e, NULL))
1f688557 1743 return NULL;
8276c007 1744 else
a0cba108 1745 return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1f688557
JH
1746}
1747
d66b37bb
JH
1748/*
1749 * This is meant to hold a *small* number of objects that you would
1750 * want read_sha1_file() to be able to return, but yet you do not want
1751 * to write them into the object store (e.g. a browse-only
1752 * application).
1753 */
1754static struct cached_object {
1755 unsigned char sha1[20];
21666f1a 1756 enum object_type type;
d66b37bb
JH
1757 void *buf;
1758 unsigned long size;
1759} *cached_objects;
1760static int cached_object_nr, cached_object_alloc;
1761
1762static struct cached_object *find_cached_object(const unsigned char *sha1)
1763{
1764 int i;
1765 struct cached_object *co = cached_objects;
1766
1767 for (i = 0; i < cached_object_nr; i++, co++) {
1768 if (!hashcmp(co->sha1, sha1))
1769 return co;
1770 }
1771 return NULL;
1772}
1773
21666f1a
NP
1774int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1775 unsigned char *sha1)
d66b37bb
JH
1776{
1777 struct cached_object *co;
1778
21666f1a 1779 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
1780 if (has_sha1_file(sha1) || find_cached_object(sha1))
1781 return 0;
1782 if (cached_object_alloc <= cached_object_nr) {
1783 cached_object_alloc = alloc_nr(cached_object_alloc);
1784 cached_objects = xrealloc(cached_objects,
1785 sizeof(*cached_objects) *
1786 cached_object_alloc);
1787 }
1788 co = &cached_objects[cached_object_nr++];
1789 co->size = len;
21666f1a 1790 co->type = type;
efa13f7b
JH
1791 co->buf = xmalloc(len);
1792 memcpy(co->buf, buf, len);
d66b37bb
JH
1793 hashcpy(co->sha1, sha1);
1794 return 0;
1795}
1796
21666f1a
NP
1797void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1798 unsigned long *size)
0fcfd160
LT
1799{
1800 unsigned long mapsize;
1801 void *map, *buf;
d66b37bb
JH
1802 struct cached_object *co;
1803
1804 co = find_cached_object(sha1);
1805 if (co) {
1806 buf = xmalloc(co->size + 1);
1807 memcpy(buf, co->buf, co->size);
1808 ((char*)buf)[co->size] = 0;
21666f1a 1809 *type = co->type;
d66b37bb
JH
1810 *size = co->size;
1811 return buf;
1812 }
0fcfd160 1813
8276c007
PE
1814 buf = read_packed_sha1(sha1, type, size);
1815 if (buf)
1816 return buf;
bb6b8e4f 1817 map = map_sha1_file(sha1, &mapsize);
0fcfd160 1818 if (map) {
7efbff75 1819 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
1820 munmap(map, mapsize);
1821 return buf;
1822 }
637cdd9d 1823 reprepare_packed_git();
8276c007 1824 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
1825}
1826
40469ee9 1827void *read_object_with_reference(const unsigned char *sha1,
21666f1a 1828 const char *required_type_name,
40469ee9
JH
1829 unsigned long *size,
1830 unsigned char *actual_sha1_return)
f4913f91 1831{
21666f1a 1832 enum object_type type, required_type;
f4913f91
JH
1833 void *buffer;
1834 unsigned long isize;
40469ee9 1835 unsigned char actual_sha1[20];
f4913f91 1836
21666f1a 1837 required_type = type_from_string(required_type_name);
e702496e 1838 hashcpy(actual_sha1, sha1);
40469ee9
JH
1839 while (1) {
1840 int ref_length = -1;
1841 const char *ref_type = NULL;
f4913f91 1842
21666f1a 1843 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
1844 if (!buffer)
1845 return NULL;
21666f1a 1846 if (type == required_type) {
40469ee9
JH
1847 *size = isize;
1848 if (actual_sha1_return)
e702496e 1849 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
1850 return buffer;
1851 }
1852 /* Handle references */
21666f1a 1853 else if (type == OBJ_COMMIT)
40469ee9 1854 ref_type = "tree ";
21666f1a 1855 else if (type == OBJ_TAG)
40469ee9
JH
1856 ref_type = "object ";
1857 else {
1858 free(buffer);
1859 return NULL;
1860 }
1861 ref_length = strlen(ref_type);
f4913f91 1862
40469ee9 1863 if (memcmp(buffer, ref_type, ref_length) ||
1d7f171c 1864 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
1865 free(buffer);
1866 return NULL;
1867 }
1cf58e72 1868 free(buffer);
40469ee9
JH
1869 /* Now we have the ID of the referred-to object in
1870 * actual_sha1. Check again. */
f4913f91 1871 }
f4913f91
JH
1872}
1873
ce9fbf16 1874static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 1875 const char *type, unsigned char *sha1,
d65a16f6 1876 char *hdr, int *hdrlen)
d410c0f5
JH
1877{
1878 SHA_CTX c;
1879
1880 /* Generate the header */
d65a16f6 1881 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
1882
1883 /* Sha1.. */
1884 SHA1_Init(&c);
1885 SHA1_Update(&c, hdr, *hdrlen);
1886 SHA1_Update(&c, buf, len);
1887 SHA1_Final(sha1, &c);
d410c0f5
JH
1888}
1889
230f1322
LT
1890/*
1891 * Link the tempfile to the final place, possibly creating the
1892 * last directory level as you do so.
1893 *
1894 * Returns the errno on failure, 0 on success.
1895 */
839837b9 1896static int link_temp_to_file(const char *tmpfile, const char *filename)
230f1322
LT
1897{
1898 int ret;
756aaf4a 1899 char *dir;
230f1322
LT
1900
1901 if (!link(tmpfile, filename))
1902 return 0;
1903
1904 /*
756aaf4a 1905 * Try to mkdir the last path component if that failed.
230f1322
LT
1906 *
1907 * Re-try the "link()" regardless of whether the mkdir
1908 * succeeds, since a race might mean that somebody
1909 * else succeeded.
1910 */
1911 ret = errno;
756aaf4a
SP
1912 dir = strrchr(filename, '/');
1913 if (dir) {
1914 *dir = 0;
866cae0d 1915 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
91c23e48 1916 *dir = '/';
756aaf4a 1917 return -2;
91c23e48 1918 }
756aaf4a
SP
1919 *dir = '/';
1920 if (!link(tmpfile, filename))
1921 return 0;
1922 ret = errno;
230f1322
LT
1923 }
1924 return ret;
1925}
1926
1927/*
1928 * Move the just written object into its final resting place
1929 */
839837b9 1930int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322
LT
1931{
1932 int ret = link_temp_to_file(tmpfile, filename);
7ebb6fca
LT
1933
1934 /*
1935 * Coda hack - coda doesn't like cross-directory links,
1936 * so we fall back to a rename, which will mean that it
1937 * won't be able to check collisions, but that's not a
1938 * big deal.
1939 *
1940 * The same holds for FAT formatted media.
1941 *
1942 * When this succeeds, we just return 0. We have nothing
1943 * left to unlink.
1944 */
1945 if (ret && ret != EEXIST) {
1946 if (!rename(tmpfile, filename))
230f1322 1947 return 0;
9e48b389 1948 ret = errno;
230f1322
LT
1949 }
1950 unlink(tmpfile);
1951 if (ret) {
1952 if (ret != EEXIST) {
916d081b 1953 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
230f1322
LT
1954 }
1955 /* FIXME!!! Collision check here ? */
1956 }
1957
1958 return 0;
1959}
1960
4d548150
LT
1961static int write_buffer(int fd, const void *buf, size_t len)
1962{
d34cf19b 1963 if (write_in_full(fd, buf, len) < 0)
93822c22 1964 return error("file write error (%s)", strerror(errno));
4d548150
LT
1965 return 0;
1966}
1967
93821bd9
LT
1968static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1969{
1970 int hdr_len;
1971 unsigned char c;
1972
1973 c = (type << 4) | (len & 15);
1974 len >>= 4;
1975 hdr_len = 1;
1976 while (len) {
1977 *hdr++ = c | 0x80;
1978 hdr_len++;
1979 c = (len & 0x7f);
1980 len >>= 7;
1981 }
1982 *hdr = c;
1983 return hdr_len;
1984}
1985
1986static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1987{
d65a16f6 1988 int obj_type, hdrlen;
93821bd9
LT
1989
1990 if (use_legacy_headers) {
1991 while (deflate(stream, 0) == Z_OK)
1992 /* nothing */;
1993 return;
1994 }
df843662 1995 obj_type = type_from_string(type);
d65a16f6
NP
1996 hdrlen = write_binary_header(stream->next_out, obj_type, len);
1997 stream->total_out = hdrlen;
1998 stream->next_out += hdrlen;
1999 stream->avail_out -= hdrlen;
93821bd9
LT
2000}
2001
ce9fbf16 2002int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2003 unsigned char *sha1)
2004{
d65a16f6 2005 char hdr[32];
abdc3fc8
RS
2006 int hdrlen;
2007 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2008 return 0;
2009}
2010
bf0f910d 2011int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
0fcfd160 2012{
ac54c277 2013 int size, ret;
bf0f910d 2014 unsigned char *compressed;
0fcfd160
LT
2015 z_stream stream;
2016 unsigned char sha1[20];
706bc531 2017 char *filename;
aac17941 2018 static char tmpfile[PATH_MAX];
d65a16f6 2019 char hdr[32];
230f1322 2020 int fd, hdrlen;
a44c9a5e 2021
d410c0f5
JH
2022 /* Normally if we have it in the pack then we do not bother writing
2023 * it out into .git/objects/??/?{38} file.
2024 */
972a9155
RS
2025 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2026 filename = sha1_file_name(sha1);
706bc531 2027 if (returnsha1)
e702496e 2028 hashcpy(returnsha1, sha1);
d410c0f5
JH
2029 if (has_sha1_file(sha1))
2030 return 0;
aac17941
LT
2031 fd = open(filename, O_RDONLY);
2032 if (fd >= 0) {
706bc531 2033 /*
aac17941
LT
2034 * FIXME!!! We might do collision checking here, but we'd
2035 * need to uncompress the old file and check it. Later.
706bc531 2036 */
aac17941 2037 close(fd);
706bc531
LT
2038 return 0;
2039 }
2040
aac17941 2041 if (errno != ENOENT) {
916d081b 2042 return error("sha1 file %s: %s\n", filename, strerror(errno));
aac17941
LT
2043 }
2044
0e55181f 2045 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
ace1534d 2046
aac17941
LT
2047 fd = mkstemp(tmpfile);
2048 if (fd < 0) {
916d081b
PB
2049 if (errno == EPERM)
2050 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2051 else
2052 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
aac17941
LT
2053 }
2054
0fcfd160
LT
2055 /* Set it up */
2056 memset(&stream, 0, sizeof(stream));
12f6c308 2057 deflateInit(&stream, zlib_compression_level);
93821bd9 2058 size = 8 + deflateBound(&stream, len+hdrlen);
812666c8 2059 compressed = xmalloc(size);
0fcfd160
LT
2060
2061 /* Compress it */
0fcfd160
LT
2062 stream.next_out = compressed;
2063 stream.avail_out = size;
a44c9a5e
LT
2064
2065 /* First header.. */
d65a16f6 2066 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2067 stream.avail_in = hdrlen;
93821bd9 2068 setup_object_header(&stream, type, len);
a44c9a5e
LT
2069
2070 /* Then the data itself.. */
2071 stream.next_in = buf;
2072 stream.avail_in = len;
ac54c277
LT
2073 ret = deflate(&stream, Z_FINISH);
2074 if (ret != Z_STREAM_END)
2075 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2076
2077 ret = deflateEnd(&stream);
2078 if (ret != Z_OK)
2079 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2080
0fcfd160
LT
2081 size = stream.total_out;
2082
4d548150
LT
2083 if (write_buffer(fd, compressed, size) < 0)
2084 die("unable to write sha1 file");
aac17941 2085 fchmod(fd, 0444);
e82973cf
JH
2086 if (close(fd))
2087 die("unable to write sha1 file");
383f85b7 2088 free(compressed);
0fcfd160 2089
230f1322 2090 return move_temp_to_file(tmpfile, filename);
0fcfd160 2091}
8237b185 2092
4d548150
LT
2093/*
2094 * We need to unpack and recompress the object for writing
2095 * it out to a different file.
2096 */
2097static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
a5eda52b 2098{
4d548150 2099 size_t size;
a5eda52b 2100 z_stream stream;
4d548150
LT
2101 unsigned char *unpacked;
2102 unsigned long len;
21666f1a 2103 enum object_type type;
d65a16f6 2104 char hdr[32];
4d548150
LT
2105 int hdrlen;
2106 void *buf;
bfc66daf 2107
a9486b02 2108 /* need to unpack and recompress it by itself */
21666f1a 2109 unpacked = read_packed_sha1(sha1, &type, &len);
8276c007
PE
2110 if (!unpacked)
2111 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
a5eda52b 2112
21666f1a 2113 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
a5eda52b 2114
4d548150
LT
2115 /* Set it up */
2116 memset(&stream, 0, sizeof(stream));
12f6c308 2117 deflateInit(&stream, zlib_compression_level);
4d548150
LT
2118 size = deflateBound(&stream, len + hdrlen);
2119 buf = xmalloc(size);
a5eda52b 2120
4d548150
LT
2121 /* Compress it */
2122 stream.next_out = buf;
2123 stream.avail_out = size;
a5eda52b 2124
4d548150
LT
2125 /* First header.. */
2126 stream.next_in = (void *)hdr;
2127 stream.avail_in = hdrlen;
2128 while (deflate(&stream, 0) == Z_OK)
2129 /* nothing */;
bfc66daf 2130
4d548150
LT
2131 /* Then the data itself.. */
2132 stream.next_in = unpacked;
2133 stream.avail_in = len;
2134 while (deflate(&stream, Z_FINISH) == Z_OK)
2135 /* nothing */;
2136 deflateEnd(&stream);
2137 free(unpacked);
bfc66daf 2138
4d548150
LT
2139 *objsize = stream.total_out;
2140 return buf;
2141}
2142
2143int write_sha1_to_fd(int fd, const unsigned char *sha1)
2144{
2145 int retval;
2146 unsigned long objsize;
bb6b8e4f 2147 void *buf = map_sha1_file(sha1, &objsize);
4d548150
LT
2148
2149 if (buf) {
2150 retval = write_buffer(fd, buf, objsize);
2151 munmap(buf, objsize);
2152 return retval;
2153 }
2154
2155 buf = repack_object(sha1, &objsize);
2156 retval = write_buffer(fd, buf, objsize);
2157 free(buf);
2158 return retval;
a5eda52b
DB
2159}
2160
70b9829e
DB
2161int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2162 size_t bufsize, size_t *bufposn)
8237b185 2163{
230f1322 2164 char tmpfile[PATH_MAX];
8237b185
DB
2165 int local;
2166 z_stream stream;
2167 unsigned char real_sha1[20];
bf0f910d 2168 unsigned char discard[4096];
8237b185
DB
2169 int ret;
2170 SHA_CTX c;
2171
0e55181f 2172 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
8237b185 2173
230f1322 2174 local = mkstemp(tmpfile);
916d081b
PB
2175 if (local < 0) {
2176 if (errno == EPERM)
2177 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2178 else
2179 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2180 }
8237b185
DB
2181
2182 memset(&stream, 0, sizeof(stream));
2183
2184 inflateInit(&stream);
2185
2186 SHA1_Init(&c);
2187
2188 do {
2189 ssize_t size;
70b9829e
DB
2190 if (*bufposn) {
2191 stream.avail_in = *bufposn;
96ad15ae 2192 stream.next_in = (unsigned char *) buffer;
70b9829e
DB
2193 do {
2194 stream.next_out = discard;
2195 stream.avail_out = sizeof(discard);
2196 ret = inflate(&stream, Z_SYNC_FLUSH);
2197 SHA1_Update(&c, discard, sizeof(discard) -
2198 stream.avail_out);
2199 } while (stream.avail_in && ret == Z_OK);
4d548150
LT
2200 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2201 die("unable to write sha1 file");
70b9829e
DB
2202 memmove(buffer, buffer + *bufposn - stream.avail_in,
2203 stream.avail_in);
2204 *bufposn = stream.avail_in;
2205 if (ret != Z_OK)
2206 break;
2207 }
93d26e4c 2208 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
8237b185
DB
2209 if (size <= 0) {
2210 close(local);
230f1322 2211 unlink(tmpfile);
8237b185
DB
2212 if (!size)
2213 return error("Connection closed?");
2214 perror("Reading from connection");
2215 return -1;
2216 }
70b9829e
DB
2217 *bufposn += size;
2218 } while (1);
8237b185
DB
2219 inflateEnd(&stream);
2220
b5b8d814 2221 fchmod(local, 0444);
0d315468
JM
2222 if (close(local) != 0)
2223 die("unable to write sha1 file");
8237b185
DB
2224 SHA1_Final(real_sha1, &c);
2225 if (ret != Z_STREAM_END) {
230f1322 2226 unlink(tmpfile);
8237b185
DB
2227 return error("File %s corrupted", sha1_to_hex(sha1));
2228 }
a89fccd2 2229 if (hashcmp(sha1, real_sha1)) {
230f1322 2230 unlink(tmpfile);
bd2afde8 2231 return error("File %s has bad hash", sha1_to_hex(sha1));
8237b185 2232 }
230f1322
LT
2233
2234 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
8237b185
DB
2235}
2236
bf592c50
DB
2237int has_pack_index(const unsigned char *sha1)
2238{
2239 struct stat st;
2240 if (stat(sha1_pack_index_name(sha1), &st))
2241 return 0;
2242 return 1;
2243}
2244
2245int has_pack_file(const unsigned char *sha1)
2246{
2247 struct stat st;
2248 if (stat(sha1_pack_name(sha1), &st))
2249 return 0;
2250 return 1;
2251}
2252
106d710b 2253int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
dade09c2
LT
2254{
2255 struct pack_entry e;
106d710b 2256 return find_pack_entry(sha1, &e, ignore_packed);
dade09c2
LT
2257}
2258
8237b185
DB
2259int has_sha1_file(const unsigned char *sha1)
2260{
8237b185 2261 struct stat st;
1f688557
JH
2262 struct pack_entry e;
2263
106d710b 2264 if (find_pack_entry(sha1, &e, NULL))
1f688557 2265 return 1;
ab90ea5d 2266 return find_sha1_file(sha1, &st) ? 1 : 0;
8237b185 2267}
74400e71 2268
e7332f96
BE
2269/*
2270 * reads from fd as long as possible into a supplied buffer of size bytes.
82e5a82f 2271 * If necessary the buffer's size is increased using realloc()
e7332f96
BE
2272 *
2273 * returns 0 if anything went fine and -1 otherwise
2274 *
2275 * NOTE: both buf and size may change, but even when -1 is returned
2276 * you still have to free() it yourself.
2277 */
2278int read_pipe(int fd, char** return_buf, unsigned long* return_size)
024510c8 2279{
e7332f96
BE
2280 char* buf = *return_buf;
2281 unsigned long size = *return_size;
8a912bcb 2282 ssize_t iret;
024510c8 2283 unsigned long off = 0;
e7332f96 2284
024510c8 2285 do {
e7332f96 2286 iret = xread(fd, buf + off, size - off);
024510c8
DB
2287 if (iret > 0) {
2288 off += iret;
2289 if (off == size) {
2290 size *= 2;
83572c1a 2291 buf = xrealloc(buf, size);
024510c8
DB
2292 }
2293 }
2294 } while (iret > 0);
e7332f96
BE
2295
2296 *return_buf = buf;
2297 *return_size = off;
2298
2299 if (iret < 0)
2300 return -1;
2301 return 0;
2302}
2303
2304int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2305{
2306 unsigned long size = 4096;
2d7320d0 2307 char *buf = xmalloc(size);
e7332f96 2308 int ret;
e7332f96
BE
2309
2310 if (read_pipe(fd, &buf, &size)) {
024510c8
DB
2311 free(buf);
2312 return -1;
2313 }
e7332f96 2314
024510c8 2315 if (!type)
8e440259 2316 type = blob_type;
024510c8 2317 if (write_object)
e7332f96 2318 ret = write_sha1_file(buf, size, type, sha1);
abdc3fc8
RS
2319 else
2320 ret = hash_sha1_file(buf, size, type, sha1);
024510c8
DB
2321 free(buf);
2322 return ret;
2323}
2324
edaec3fb 2325int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
53bca91a 2326 enum object_type type, const char *path)
74400e71 2327{
dc49cd76 2328 size_t size = xsize_t(st->st_size);
3a55602e 2329 void *buf = NULL;
6c510bee 2330 int ret, re_allocated = 0;
74400e71 2331
74400e71 2332 if (size)
c4712e45 2333 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
74400e71 2334 close(fd);
74400e71 2335
7672db20 2336 if (!type)
edaec3fb 2337 type = OBJ_BLOB;
6c510bee
LT
2338
2339 /*
2340 * Convert blobs to git internal format
2341 */
d0d8e14d 2342 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
6c510bee 2343 unsigned long nsize = size;
ac78e548
AR
2344 char *nbuf = convert_to_git(path, buf, &nsize);
2345 if (nbuf) {
2346 munmap(buf, size);
6c510bee
LT
2347 size = nsize;
2348 buf = nbuf;
2349 re_allocated = 1;
2350 }
2351 }
2352
7672db20 2353 if (write_object)
edaec3fb 2354 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 2355 else
edaec3fb 2356 ret = hash_sha1_file(buf, size, typename(type), sha1);
6c510bee
LT
2357 if (re_allocated) {
2358 free(buf);
2359 return ret;
2360 }
aac17941
LT
2361 if (size)
2362 munmap(buf, size);
2363 return ret;
74400e71 2364}
ec1fcc16
JH
2365
2366int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2367{
2368 int fd;
2369 char *target;
dc49cd76 2370 size_t len;
ec1fcc16
JH
2371
2372 switch (st->st_mode & S_IFMT) {
2373 case S_IFREG:
2374 fd = open(path, O_RDONLY);
2375 if (fd < 0)
2376 return error("open(\"%s\"): %s", path,
2377 strerror(errno));
53bca91a 2378 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
ec1fcc16
JH
2379 return error("%s: failed to insert into database",
2380 path);
2381 break;
2382 case S_IFLNK:
dc49cd76
SP
2383 len = xsize_t(st->st_size);
2384 target = xmalloc(len + 1);
2385 if (readlink(path, target, len + 1) != st->st_size) {
ec1fcc16
JH
2386 char *errstr = strerror(errno);
2387 free(target);
2388 return error("readlink(\"%s\"): %s", path,
2389 errstr);
2390 }
abdc3fc8 2391 if (!write_object)
dc49cd76
SP
2392 hash_sha1_file(target, len, blob_type, sha1);
2393 else if (write_sha1_file(target, len, blob_type, sha1))
ec1fcc16
JH
2394 return error("%s: failed to insert into database",
2395 path);
2396 free(target);
2397 break;
f35a6d3b
LT
2398 case S_IFDIR:
2399 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
2400 default:
2401 return error("%s: unsupported file type", path);
2402 }
2403 return 0;
2404}
a69e5429
JH
2405
2406int read_pack_header(int fd, struct pack_header *header)
2407{
2408 char *c = (char*)header;
2409 ssize_t remaining = sizeof(struct pack_header);
2410 do {
2411 ssize_t r = xread(fd, c, remaining);
2412 if (r <= 0)
2413 /* "eof before pack header was fully read" */
2414 return PH_ERROR_EOF;
2415 remaining -= r;
2416 c += r;
2417 } while (remaining > 0);
2418 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2419 /* "protocol error (pack signature mismatch detected)" */
2420 return PH_ERROR_PACK_SIGNATURE;
2421 if (!pack_version_ok(header->hdr_version))
2422 /* "protocol error (pack version unsupported)" */
2423 return PH_ERROR_PROTOCOL;
2424 return 0;
2425}