call init_pack_revindex() lazily
[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"
70f5d5d3 17#include "pack-revindex.h"
628522ec 18#include "sha1-lookup.h"
0fcfd160 19
144bde78
LT
20#ifndef O_NOATIME
21#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
22#define O_NOATIME 01000000
23#else
24#define O_NOATIME 0
25#endif
26#endif
27
e05db0fd
PR
28#ifdef NO_C99_FORMAT
29#define SZ_FMT "lu"
9e42d6a1 30static unsigned long sz_fmt(size_t s) { return (unsigned long)s; }
e05db0fd
PR
31#else
32#define SZ_FMT "zu"
9e42d6a1 33static size_t sz_fmt(size_t s) { return s; }
e05db0fd
PR
34#endif
35
96f1e58f 36const unsigned char null_sha1[20];
88cd621d 37
192a6be2 38const signed char hexval_table[256] = {
e49521b5
JH
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
45 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
46 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
47 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
48 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
51 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
69 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
71};
0fcfd160
LT
72
73int get_sha1_hex(const char *hex, unsigned char *sha1)
74{
75 int i;
76 for (i = 0; i < 20; i++) {
77 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
78 if (val & ~0xff)
79 return -1;
80 *sha1++ = val;
81 hex += 2;
82 }
83 return 0;
84}
85
b2cb9425
JH
86int safe_create_leading_directories(char *path)
87{
88 char *pos = path;
67d42212
JR
89 struct stat st;
90
85dadc38 91 if (is_absolute_path(path))
67ffdf4c 92 pos++;
b2cb9425
JH
93
94 while (pos) {
95 pos = strchr(pos, '/');
96 if (!pos)
97 break;
98 *pos = 0;
67d42212
JR
99 if (!stat(path, &st)) {
100 /* path exists */
101 if (!S_ISDIR(st.st_mode)) {
b2cb9425 102 *pos = '/';
67d42212 103 return -3;
b2cb9425 104 }
457f06d6 105 }
67d42212
JR
106 else if (mkdir(path, 0777)) {
107 *pos = '/';
108 return -1;
109 }
457f06d6
JS
110 else if (adjust_shared_perm(path)) {
111 *pos = '/';
112 return -2;
113 }
b2cb9425
JH
114 *pos++ = '/';
115 }
116 return 0;
117}
723c31fe 118
3bfaf018 119char *sha1_to_hex(const unsigned char *sha1)
0fcfd160 120{
dcb3450f
LT
121 static int bufno;
122 static char hexbuffer[4][50];
0fcfd160 123 static const char hex[] = "0123456789abcdef";
dcb3450f 124 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
0fcfd160
LT
125 int i;
126
127 for (i = 0; i < 20; i++) {
128 unsigned int val = *sha1++;
129 *buf++ = hex[val >> 4];
130 *buf++ = hex[val & 0xf];
131 }
1e80e044
JS
132 *buf = '\0';
133
0fcfd160
LT
134 return buffer;
135}
136
ace1534d
JH
137static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
138{
139 int i;
140 for (i = 0; i < 20; i++) {
141 static char hex[] = "0123456789abcdef";
142 unsigned int val = sha1[i];
143 char *pos = pathbuf + i*2 + (i > 0);
144 *pos++ = hex[val >> 4];
145 *pos = hex[val & 0xf];
146 }
147}
148
0fcfd160
LT
149/*
150 * NOTE! This returns a statically allocated buffer, so you have to be
790296fd 151 * careful about using it. Do an "xstrdup()" if you need to save the
0fcfd160 152 * filename.
ace1534d
JH
153 *
154 * Also note that this returns the location for creating. Reading
155 * SHA1 file can happen from any alternate directory listed in the
d19938ab 156 * DB_ENVIRONMENT environment variable if it is not found in
ace1534d 157 * the primary object database.
0fcfd160
LT
158 */
159char *sha1_file_name(const unsigned char *sha1)
160{
0fcfd160
LT
161 static char *name, *base;
162
163 if (!base) {
d19938ab 164 const char *sha1_file_directory = get_object_directory();
0fcfd160 165 int len = strlen(sha1_file_directory);
812666c8 166 base = xmalloc(len + 60);
0fcfd160
LT
167 memcpy(base, sha1_file_directory, len);
168 memset(base+len, 0, 60);
169 base[len] = '/';
170 base[len+3] = '/';
171 name = base + len + 1;
172 }
ace1534d 173 fill_sha1_path(name, sha1);
0fcfd160
LT
174 return base;
175}
176
633f43e1 177static char *sha1_get_pack_name(const unsigned char *sha1,
6eec46bd 178 char **name, char **base, const char *which)
bf592c50
DB
179{
180 static const char hex[] = "0123456789abcdef";
633f43e1 181 char *buf;
bf592c50
DB
182 int i;
183
633f43e1 184 if (!*base) {
bf592c50
DB
185 const char *sha1_file_directory = get_object_directory();
186 int len = strlen(sha1_file_directory);
633f43e1 187 *base = xmalloc(len + 60);
6eec46bd
JH
188 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
189 sha1_file_directory, which);
633f43e1 190 *name = *base + len + 11;
bf592c50
DB
191 }
192
633f43e1 193 buf = *name;
bf592c50
DB
194
195 for (i = 0; i < 20; i++) {
196 unsigned int val = *sha1++;
197 *buf++ = hex[val >> 4];
198 *buf++ = hex[val & 0xf];
199 }
a6080a0a 200
633f43e1 201 return *base;
bf592c50
DB
202}
203
633f43e1 204char *sha1_pack_name(const unsigned char *sha1)
bf592c50 205{
633f43e1 206 static char *name, *base;
bf592c50 207
6eec46bd 208 return sha1_get_pack_name(sha1, &name, &base, "pack");
633f43e1 209}
bf592c50 210
633f43e1
HO
211char *sha1_pack_index_name(const unsigned char *sha1)
212{
213 static char *name, *base;
a6080a0a 214
6eec46bd 215 return sha1_get_pack_name(sha1, &name, &base, "idx");
bf592c50
DB
216}
217
d5a63b99
JH
218struct alternate_object_database *alt_odb_list;
219static struct alternate_object_database **alt_odb_tail;
ace1534d 220
c2f493a4
MW
221static void read_info_alternates(const char * alternates, int depth);
222
ddd5d056
JH
223/*
224 * Prepare alternate object database registry.
d5a63b99
JH
225 *
226 * The variable alt_odb_list points at the list of struct
227 * alternate_object_database. The elements on this list come from
228 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
229 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
230 * whose contents is similar to that environment variable but can be
231 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
232 * contains "/the/directory/corresponding/to/.git/objects/...", while
233 * its name points just after the slash at the end of ".git/objects/"
234 * in the example above, and has enough space to hold 40-byte hex
235 * SHA1, an extra slash for the first level indirection, and the
236 * terminating NUL.
ddd5d056 237 */
c2f493a4 238static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
ace1534d 239{
c2f493a4 240 struct stat st;
1494e038 241 const char *objdir = get_object_directory();
c2f493a4
MW
242 struct alternate_object_database *ent;
243 struct alternate_object_database *alt;
244 /* 43 = 40-byte + 2 '/' + terminating NUL */
245 int pfxlen = len;
246 int entlen = pfxlen + 43;
ccfd3e99 247 int base_len = -1;
d5a63b99 248
85dadc38 249 if (!is_absolute_path(entry) && relative_base) {
c2f493a4
MW
250 /* Relative alt-odb */
251 if (base_len < 0)
252 base_len = strlen(relative_base) + 1;
253 entlen += base_len;
254 pfxlen += base_len;
255 }
256 ent = xmalloc(sizeof(*ent) + entlen);
257
85dadc38 258 if (!is_absolute_path(entry) && relative_base) {
c2f493a4
MW
259 memcpy(ent->base, relative_base, base_len - 1);
260 ent->base[base_len - 1] = '/';
261 memcpy(ent->base + base_len, entry, len);
262 }
263 else
264 memcpy(ent->base, entry, pfxlen);
265
266 ent->name = ent->base + pfxlen + 1;
267 ent->base[pfxlen + 3] = '/';
268 ent->base[pfxlen] = ent->base[entlen-1] = 0;
269
270 /* Detect cases where alternate disappeared */
271 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
272 error("object directory %s does not exist; "
273 "check .git/objects/info/alternates.",
274 ent->base);
275 free(ent);
276 return -1;
277 }
278
279 /* Prevent the common mistake of listing the same
280 * thing twice, or object directory itself.
281 */
282 for (alt = alt_odb_list; alt; alt = alt->next) {
283 if (!memcmp(ent->base, alt->base, pfxlen)) {
284 free(ent);
285 return -1;
286 }
287 }
288 if (!memcmp(ent->base, objdir, pfxlen)) {
289 free(ent);
290 return -1;
291 }
292
293 /* add the alternate entry */
294 *alt_odb_tail = ent;
295 alt_odb_tail = &(ent->next);
296 ent->next = NULL;
297
298 /* recursively add alternates */
299 read_info_alternates(ent->base, depth + 1);
300
301 ent->base[pfxlen] = '/';
302
303 return 0;
304}
305
306static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
307 const char *relative_base, int depth)
308{
309 const char *cp, *last;
310
311 if (depth > 5) {
312 error("%s: ignoring alternate object stores, nesting too deep.",
313 relative_base);
314 return;
315 }
316
d5a63b99 317 last = alt;
9577e7e3
JH
318 while (last < ep) {
319 cp = last;
320 if (cp < ep && *cp == '#') {
321 while (cp < ep && *cp != sep)
322 cp++;
323 last = cp + 1;
324 continue;
325 }
c2f493a4
MW
326 while (cp < ep && *cp != sep)
327 cp++;
d5a63b99 328 if (last != cp) {
85dadc38 329 if (!is_absolute_path(last) && depth) {
c2f493a4
MW
330 error("%s: ignoring relative alternate object store %s",
331 relative_base, last);
332 } else {
333 link_alt_odb_entry(last, cp - last,
334 relative_base, depth);
1494e038 335 }
d5a63b99 336 }
9577e7e3 337 while (cp < ep && *cp == sep)
d5a63b99
JH
338 cp++;
339 last = cp;
9577e7e3 340 }
d5a63b99
JH
341}
342
c2f493a4 343static void read_info_alternates(const char * relative_base, int depth)
d5a63b99 344{
9577e7e3 345 char *map;
dc49cd76 346 size_t mapsz;
d5a63b99 347 struct stat st;
9cb18f56
JM
348 const char alt_file_name[] = "info/alternates";
349 /* Given that relative_base is no longer than PATH_MAX,
350 ensure that "path" has enough space to append "/", the
351 file name, "info/alternates", and a trailing NUL. */
352 char path[PATH_MAX + 1 + sizeof alt_file_name];
c2f493a4 353 int fd;
d5a63b99 354
9cb18f56 355 sprintf(path, "%s/%s", relative_base, alt_file_name);
d5a63b99
JH
356 fd = open(path, O_RDONLY);
357 if (fd < 0)
358 return;
359 if (fstat(fd, &st) || (st.st_size == 0)) {
360 close(fd);
9a217f2a 361 return;
ace1534d 362 }
dc49cd76
SP
363 mapsz = xsize_t(st.st_size);
364 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 365 close(fd);
d5a63b99 366
dc49cd76 367 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
c2f493a4 368
dc49cd76 369 munmap(map, mapsz);
ace1534d
JH
370}
371
bef70b22
DB
372void add_to_alternates_file(const char *reference)
373{
374 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
375 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), 1);
376 char *alt = mkpath("%s/objects\n", reference);
377 write_or_die(fd, alt, strlen(alt));
378 if (commit_lock_file(lock))
379 die("could not close alternates file");
380 if (alt_odb_tail)
381 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
382}
383
c2f493a4
MW
384void prepare_alt_odb(void)
385{
554fe20d 386 const char *alt;
c2f493a4 387
7dc24aa5
SP
388 if (alt_odb_tail)
389 return;
390
c2f493a4
MW
391 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
392 if (!alt) alt = "";
393
c2f493a4
MW
394 alt_odb_tail = &alt_odb_list;
395 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
396
397 read_info_alternates(get_object_directory(), 0);
398}
399
c529d75a 400static int has_loose_object(const unsigned char *sha1)
ace1534d 401{
ace1534d 402 char *name = sha1_file_name(sha1);
d5a63b99 403 struct alternate_object_database *alt;
ace1534d 404
c529d75a
LT
405 if (!access(name, F_OK))
406 return 1;
9a217f2a 407 prepare_alt_odb();
d5a63b99
JH
408 for (alt = alt_odb_list; alt; alt = alt->next) {
409 name = alt->name;
ace1534d 410 fill_sha1_path(name, sha1);
c529d75a
LT
411 if (!access(alt->base, F_OK))
412 return 1;
ace1534d 413 }
c529d75a 414 return 0;
ace1534d
JH
415}
416
60bb8b14 417static unsigned int pack_used_ctr;
a53128b6
SP
418static unsigned int pack_mmap_calls;
419static unsigned int peak_pack_open_windows;
420static unsigned int pack_open_windows;
421static size_t peak_pack_mapped;
60bb8b14 422static size_t pack_mapped;
9a217f2a 423struct packed_git *packed_git;
1f688557 424
b79d18c9 425void pack_report(void)
a53128b6
SP
426{
427 fprintf(stderr,
e05db0fd
PR
428 "pack_report: getpagesize() = %10" SZ_FMT "\n"
429 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
430 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
431 sz_fmt(getpagesize()),
432 sz_fmt(packed_git_window_size),
433 sz_fmt(packed_git_limit));
a53128b6
SP
434 fprintf(stderr,
435 "pack_report: pack_used_ctr = %10u\n"
436 "pack_report: pack_mmap_calls = %10u\n"
437 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
438 "pack_report: pack_mapped = "
439 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
440 pack_used_ctr,
441 pack_mmap_calls,
442 pack_open_windows, peak_pack_open_windows,
9e42d6a1 443 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
444}
445
42873078 446static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
447{
448 void *idx_map;
42873078 449 struct pack_idx_header *hdr;
dc49cd76 450 size_t idx_size;
74e34e1f 451 uint32_t version, nr, i, *index;
e93ec6f9 452 int fd = open(path, O_RDONLY);
1f688557 453 struct stat st;
42873078 454
1f688557
JH
455 if (fd < 0)
456 return -1;
457 if (fstat(fd, &st)) {
458 close(fd);
459 return -1;
460 }
dc49cd76 461 idx_size = xsize_t(st.st_size);
2d88451b
SP
462 if (idx_size < 4 * 256 + 20 + 20) {
463 close(fd);
464 return error("index file %s is too small", path);
465 }
c4712e45 466 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 467 close(fd);
1f688557 468
42873078
NP
469 hdr = idx_map;
470 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
471 version = ntohl(hdr->idx_version);
472 if (version < 2 || version > 2) {
473 munmap(idx_map, idx_size);
474 return error("index file %s is version %d"
475 " and is not supported by this binary"
476 " (try upgrading GIT to a newer version)",
477 path, version);
478 }
479 } else
480 version = 1;
df1b059d 481
1f688557 482 nr = 0;
42873078 483 index = idx_map;
74e34e1f
NP
484 if (version > 1)
485 index += 2; /* skip index header */
1f688557 486 for (i = 0; i < 256; i++) {
326bf396 487 uint32_t n = ntohl(index[i]);
2d88451b
SP
488 if (n < nr) {
489 munmap(idx_map, idx_size);
df1b059d 490 return error("non-monotonic index %s", path);
2d88451b 491 }
1f688557
JH
492 nr = n;
493 }
494
74e34e1f
NP
495 if (version == 1) {
496 /*
497 * Total size:
498 * - 256 index entries 4 bytes each
499 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
500 * - 20-byte SHA1 of the packfile
501 * - 20-byte SHA1 file checksum
502 */
503 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
504 munmap(idx_map, idx_size);
eef427a0 505 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
506 }
507 } else if (version == 2) {
508 /*
509 * Minimum size:
510 * - 8 bytes of header
511 * - 256 index entries 4 bytes each
512 * - 20-byte sha1 entry * nr
513 * - 4-byte crc entry * nr
514 * - 4-byte offset entry * nr
515 * - 20-byte SHA1 of the packfile
516 * - 20-byte SHA1 file checksum
517 * And after the 4-byte offset table might be a
518 * variable sized table containing 8-byte entries
519 * for offsets larger than 2^31.
520 */
521 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
522 unsigned long max_size = min_size;
523 if (nr)
524 max_size += (nr - 1)*8;
525 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 526 munmap(idx_map, idx_size);
eef427a0 527 return error("wrong index v2 file size in %s", path);
74e34e1f 528 }
7109c889
JH
529 if (idx_size != min_size &&
530 /*
531 * make sure we can deal with large pack offsets.
532 * 31-bit signed offset won't be enough, neither
533 * 32-bit unsigned one will be.
534 */
535 (sizeof(off_t) <= 4)) {
536 munmap(idx_map, idx_size);
537 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 538 }
2d88451b 539 }
1f688557 540
74e34e1f 541 p->index_version = version;
42873078
NP
542 p->index_data = idx_map;
543 p->index_size = idx_size;
57059091 544 p->num_objects = nr;
1f688557
JH
545 return 0;
546}
547
bc8e478a 548int open_pack_index(struct packed_git *p)
d079837e
SP
549{
550 char *idx_name;
551 int ret;
552
553 if (p->index_data)
554 return 0;
555
556 idx_name = xstrdup(p->pack_name);
557 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
558 ret = check_packed_git_idx(idx_name, p);
559 free(idx_name);
560 return ret;
561}
562
11daf39b
SP
563static void scan_windows(struct packed_git *p,
564 struct packed_git **lru_p,
565 struct pack_window **lru_w,
566 struct pack_window **lru_l)
1f688557 567{
11daf39b
SP
568 struct pack_window *w, *w_l;
569
570 for (w_l = NULL, w = p->windows; w; w = w->next) {
571 if (!w->inuse_cnt) {
572 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
573 *lru_p = p;
574 *lru_w = w;
575 *lru_l = w_l;
54044bf8 576 }
54044bf8 577 }
11daf39b 578 w_l = w;
f9253394 579 }
11daf39b
SP
580}
581
d1efefa4 582static int unuse_one_window(struct packed_git *current, int keep_fd)
11daf39b
SP
583{
584 struct packed_git *p, *lru_p = NULL;
585 struct pack_window *lru_w = NULL, *lru_l = NULL;
586
587 if (current)
588 scan_windows(current, &lru_p, &lru_w, &lru_l);
589 for (p = packed_git; p; p = p->next)
590 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
591 if (lru_p) {
592 munmap(lru_w->base, lru_w->len);
593 pack_mapped -= lru_w->len;
594 if (lru_l)
595 lru_l->next = lru_w->next;
596 else {
597 lru_p->windows = lru_w->next;
d1efefa4 598 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
54044bf8
SP
599 close(lru_p->pack_fd);
600 lru_p->pack_fd = -1;
601 }
602 }
603 free(lru_w);
a53128b6 604 pack_open_windows--;
54044bf8
SP
605 return 1;
606 }
607 return 0;
f9253394
JH
608}
609
d1efefa4 610void release_pack_memory(size_t need, int fd)
97bfeb34
SP
611{
612 size_t cur = pack_mapped;
d1efefa4 613 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
97bfeb34
SP
614 ; /* nothing */
615}
616
c9ced051
SP
617void close_pack_windows(struct packed_git *p)
618{
619 while (p->windows) {
620 struct pack_window *w = p->windows;
621
622 if (w->inuse_cnt)
623 die("pack '%s' still has open windows to it",
624 p->pack_name);
625 munmap(w->base, w->len);
626 pack_mapped -= w->len;
627 pack_open_windows--;
628 p->windows = w->next;
629 free(w);
630 }
631}
632
03e79c88 633void unuse_pack(struct pack_window **w_cursor)
f9253394 634{
03e79c88
SP
635 struct pack_window *w = *w_cursor;
636 if (w) {
637 w->inuse_cnt--;
638 *w_cursor = NULL;
639 }
1f688557
JH
640}
641
3cf8b462
SP
642/*
643 * Do not call this directly as this leaks p->pack_fd on error return;
644 * call open_packed_git() instead.
645 */
646static int open_packed_git_1(struct packed_git *p)
1f688557 647{
9bc879c1
SP
648 struct stat st;
649 struct pack_header hdr;
650 unsigned char sha1[20];
651 unsigned char *idx_sha1;
2c039da8 652 long fd_flag;
9bc879c1 653
d079837e
SP
654 if (!p->index_data && open_pack_index(p))
655 return error("packfile %s index unavailable", p->pack_name);
656
9bc879c1
SP
657 p->pack_fd = open(p->pack_name, O_RDONLY);
658 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 659 return -1;
9bc879c1
SP
660
661 /* If we created the struct before we had the pack we lack size. */
bf592c50 662 if (!p->pack_size) {
bf592c50 663 if (!S_ISREG(st.st_mode))
072db278 664 return error("packfile %s not a regular file", p->pack_name);
bf592c50 665 p->pack_size = st.st_size;
9bc879c1 666 } else if (p->pack_size != st.st_size)
072db278 667 return error("packfile %s size changed", p->pack_name);
9bc879c1 668
2c039da8
JH
669 /* We leave these file descriptors open with sliding mmap;
670 * there is no point keeping them open across exec(), though.
671 */
672 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
673 if (fd_flag < 0)
072db278 674 return error("cannot determine file descriptor flags");
2c039da8
JH
675 fd_flag |= FD_CLOEXEC;
676 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 677 return error("cannot set FD_CLOEXEC");
2c039da8 678
9bc879c1 679 /* Verify we recognize this pack file format. */
e6e2bd62 680 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 681 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 682 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 683 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 684 if (!pack_version_ok(hdr.hdr_version))
072db278 685 return error("packfile %s is version %u and not supported"
9bc879c1
SP
686 " (try upgrading GIT to a newer version)",
687 p->pack_name, ntohl(hdr.hdr_version));
688
689 /* Verify the pack matches its index. */
57059091 690 if (p->num_objects != ntohl(hdr.hdr_entries))
072db278 691 return error("packfile %s claims to have %u objects"
57059091
NP
692 " while index indicates %u objects",
693 p->pack_name, ntohl(hdr.hdr_entries),
694 p->num_objects);
9bc879c1 695 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 696 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 697 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 698 return error("packfile %s signature is unavailable", p->pack_name);
42873078 699 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 700 if (hashcmp(sha1, idx_sha1))
072db278
SP
701 return error("packfile %s does not match index", p->pack_name);
702 return 0;
9bc879c1
SP
703}
704
3cf8b462
SP
705static int open_packed_git(struct packed_git *p)
706{
707 if (!open_packed_git_1(p))
708 return 0;
709 if (p->pack_fd != -1) {
710 close(p->pack_fd);
711 p->pack_fd = -1;
712 }
713 return -1;
714}
715
c4001d92 716static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
717{
718 /* We must promise at least 20 bytes (one hash) after the
719 * offset is available from this window, otherwise the offset
720 * is not actually in this window and a different window (which
721 * has that one hash excess) must be used. This is to support
722 * the object header and delta base parsing routines below.
723 */
724 off_t win_off = win->offset;
725 return win_off <= offset
726 && (offset + 20) <= (win_off + win->len);
727}
728
03e79c88
SP
729unsigned char* use_pack(struct packed_git *p,
730 struct pack_window **w_cursor,
c4001d92 731 off_t offset,
03e79c88 732 unsigned int *left)
9bc879c1 733{
60bb8b14 734 struct pack_window *win = *w_cursor;
03e79c88 735
072db278
SP
736 if (p->pack_fd == -1 && open_packed_git(p))
737 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
738
739 /* Since packfiles end in a hash of their content and its
740 * pointless to ask for an offset into the middle of that
741 * hash, and the in_window function above wouldn't match
742 * don't allow an offset too close to the end of the file.
743 */
744 if (offset > (p->pack_size - 20))
745 die("offset beyond end of packfile (truncated pack?)");
746
747 if (!win || !in_window(win, offset)) {
748 if (win)
749 win->inuse_cnt--;
750 for (win = p->windows; win; win = win->next) {
751 if (in_window(win, offset))
752 break;
753 }
754 if (!win) {
78a28df9 755 size_t window_align = packed_git_window_size / 2;
dc49cd76 756 off_t len;
60bb8b14 757 win = xcalloc(1, sizeof(*win));
78a28df9 758 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
759 len = p->pack_size - win->offset;
760 if (len > packed_git_window_size)
761 len = packed_git_window_size;
762 win->len = (size_t)len;
60bb8b14 763 pack_mapped += win->len;
11daf39b 764 while (packed_git_limit < pack_mapped
d1efefa4 765 && unuse_one_window(p, p->pack_fd))
60bb8b14 766 ; /* nothing */
c4712e45 767 win->base = xmmap(NULL, win->len,
60bb8b14
SP
768 PROT_READ, MAP_PRIVATE,
769 p->pack_fd, win->offset);
770 if (win->base == MAP_FAILED)
73b4e4be
SP
771 die("packfile %s cannot be mapped: %s",
772 p->pack_name,
773 strerror(errno));
a53128b6
SP
774 pack_mmap_calls++;
775 pack_open_windows++;
776 if (pack_mapped > peak_pack_mapped)
777 peak_pack_mapped = pack_mapped;
778 if (pack_open_windows > peak_pack_open_windows)
779 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
780 win->next = p->windows;
781 p->windows = win;
782 }
1f688557 783 }
03e79c88
SP
784 if (win != *w_cursor) {
785 win->last_used = pack_used_ctr++;
786 win->inuse_cnt++;
787 *w_cursor = win;
788 }
60bb8b14 789 offset -= win->offset;
03e79c88 790 if (left)
dc49cd76 791 *left = win->len - xsize_t(offset);
03e79c88 792 return win->base + offset;
1f688557
JH
793}
794
42873078 795struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557
JH
796{
797 struct stat st;
42873078 798 struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
1f688557 799
42873078
NP
800 /*
801 * Make sure a corresponding .pack file exists and that
802 * the index looks sane.
803 */
804 path_len -= strlen(".idx");
805 if (path_len < 1)
1f688557 806 return NULL;
42873078
NP
807 memcpy(p->pack_name, path, path_len);
808 strcpy(p->pack_name + path_len, ".pack");
d079837e 809 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 810 free(p);
1f688557
JH
811 return NULL;
812 }
42873078 813
1f688557
JH
814 /* ok, it looks sane as far as we can check without
815 * actually mapping the pack file.
816 */
d079837e
SP
817 p->index_version = 0;
818 p->index_data = NULL;
819 p->index_size = 0;
820 p->num_objects = 0;
1f688557 821 p->pack_size = st.st_size;
1f688557 822 p->next = NULL;
c41ee586 823 p->windows = NULL;
9bc879c1 824 p->pack_fd = -1;
9d835df2 825 p->pack_local = local;
b867092f 826 p->mtime = st.st_mtime;
42873078
NP
827 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
828 hashclr(p->sha1);
1f688557
JH
829 return p;
830}
831
bf592c50 832struct packed_git *parse_pack_index(unsigned char *sha1)
c508df5e 833{
79c6dca4 834 const char *idx_path = sha1_pack_index_name(sha1);
42873078
NP
835 const char *path = sha1_pack_name(sha1);
836 struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
bf592c50 837
42873078
NP
838 if (check_packed_git_idx(idx_path, p)) {
839 free(p);
bf592c50 840 return NULL;
42873078 841 }
bf592c50 842
bf592c50 843 strcpy(p->pack_name, path);
bf592c50 844 p->pack_size = 0;
bf592c50 845 p->next = NULL;
c41ee586 846 p->windows = NULL;
9bc879c1 847 p->pack_fd = -1;
e702496e 848 hashcpy(p->sha1, sha1);
bf592c50
DB
849 return p;
850}
851
852void install_packed_git(struct packed_git *pack)
853{
854 pack->next = packed_git;
855 packed_git = pack;
856}
857
9d835df2 858static void prepare_packed_git_one(char *objdir, int local)
1f688557 859{
9cb18f56
JM
860 /* Ensure that this buffer is large enough so that we can
861 append "/pack/" without clobbering the stack even if
862 strlen(objdir) were PATH_MAX. */
863 char path[PATH_MAX + 1 + 4 + 1 + 1];
1f688557
JH
864 int len;
865 DIR *dir;
866 struct dirent *de;
867
868 sprintf(path, "%s/pack", objdir);
869 len = strlen(path);
870 dir = opendir(path);
b5b16990 871 if (!dir) {
26125f6b 872 if (errno != ENOENT)
bd2afde8 873 error("unable to open object pack directory: %s: %s",
26125f6b 874 path, strerror(errno));
1f688557 875 return;
b5b16990 876 }
1f688557
JH
877 path[len++] = '/';
878 while ((de = readdir(dir)) != NULL) {
879 int namelen = strlen(de->d_name);
880 struct packed_git *p;
881
5bb1cda5 882 if (!has_extension(de->d_name, ".idx"))
1f688557
JH
883 continue;
884
9cb18f56
JM
885 if (len + namelen + 1 > sizeof(path))
886 continue;
887
54a15a8d 888 /* Don't reopen a pack we already have. */
1f688557 889 strcpy(path + len, de->d_name);
86f7780c
JK
890 for (p = packed_git; p; p = p->next) {
891 if (!memcmp(path, p->pack_name, len + namelen - 4))
892 break;
893 }
894 if (p)
895 continue;
54a15a8d
SP
896 /* See if it really is a valid .idx file with corresponding
897 * .pack file that we can map.
898 */
9d835df2 899 p = add_packed_git(path, len + namelen, local);
1f688557
JH
900 if (!p)
901 continue;
625e9421 902 install_packed_git(p);
1f688557 903 }
5b35bcd5 904 closedir(dir);
1f688557
JH
905}
906
b867092f
JH
907static int sort_pack(const void *a_, const void *b_)
908{
909 struct packed_git *a = *((struct packed_git **)a_);
910 struct packed_git *b = *((struct packed_git **)b_);
911 int st;
912
913 /*
914 * Local packs tend to contain objects specific to our
915 * variant of the project than remote ones. In addition,
916 * remote ones could be on a network mounted filesystem.
917 * Favor local ones for these reasons.
918 */
919 st = a->pack_local - b->pack_local;
920 if (st)
921 return -st;
922
923 /*
924 * Younger packs tend to contain more recent objects,
925 * and more recent objects tend to get accessed more
926 * often.
927 */
928 if (a->mtime < b->mtime)
929 return 1;
930 else if (a->mtime == b->mtime)
931 return 0;
932 return -1;
933}
934
935static void rearrange_packed_git(void)
936{
937 struct packed_git **ary, *p;
938 int i, n;
939
940 for (n = 0, p = packed_git; p; p = p->next)
941 n++;
942 if (n < 2)
943 return;
944
945 /* prepare an array of packed_git for easier sorting */
946 ary = xcalloc(n, sizeof(struct packed_git *));
947 for (n = 0, p = packed_git; p; p = p->next)
948 ary[n++] = p;
949
950 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
951
952 /* link them back again */
953 for (i = 0; i < n - 1; i++)
954 ary[i]->next = ary[i + 1];
955 ary[n - 1]->next = NULL;
956 packed_git = ary[0];
957
958 free(ary);
959}
960
637cdd9d 961static int prepare_packed_git_run_once = 0;
9a217f2a 962void prepare_packed_git(void)
1f688557 963{
d5a63b99 964 struct alternate_object_database *alt;
1f688557 965
637cdd9d 966 if (prepare_packed_git_run_once)
1f688557 967 return;
9d835df2 968 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 969 prepare_alt_odb();
d5a63b99 970 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 971 alt->name[-1] = 0;
9d835df2 972 prepare_packed_git_one(alt->base, 0);
1494e038 973 alt->name[-1] = '/';
1f688557 974 }
b867092f 975 rearrange_packed_git();
637cdd9d
JK
976 prepare_packed_git_run_once = 1;
977}
978
fc04c412 979void reprepare_packed_git(void)
637cdd9d
JK
980{
981 prepare_packed_git_run_once = 0;
982 prepare_packed_git();
1f688557
JH
983}
984
5d6ccf5c 985int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
0fcfd160
LT
986{
987 unsigned char real_sha1[20];
7cfb5f36 988 hash_sha1_file(map, size, type, real_sha1);
a89fccd2 989 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
990}
991
44d1c19e
LT
992static int git_open_noatime(const char *name)
993{
994 static int sha1_file_open_flag = O_NOATIME;
995 int fd = open(name, O_RDONLY | sha1_file_open_flag);
996
997 /* Might the failure be due to O_NOATIME? */
998 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
999 fd = open(name, O_RDONLY);
1000 if (fd >= 0)
1001 sha1_file_open_flag = 0;
1002 }
1003 return fd;
1004}
1005
1006static int open_sha1_file(const unsigned char *sha1)
1007{
1008 int fd;
1009 char *name = sha1_file_name(sha1);
1010 struct alternate_object_database *alt;
1011
1012 fd = git_open_noatime(name);
1013 if (fd >= 0)
1014 return fd;
1015
1016 prepare_alt_odb();
1017 errno = ENOENT;
1018 for (alt = alt_odb_list; alt; alt = alt->next) {
1019 name = alt->name;
1020 fill_sha1_path(name, sha1);
1021 fd = git_open_noatime(alt->base);
1022 if (fd >= 0)
1023 return fd;
1024 }
1025 return -1;
1026}
1027
4175e9e3 1028static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1029{
0fcfd160 1030 void *map;
144bde78 1031 int fd;
ace1534d 1032
44d1c19e
LT
1033 fd = open_sha1_file(sha1);
1034 map = NULL;
1035 if (fd >= 0) {
1036 struct stat st;
0fcfd160 1037
44d1c19e
LT
1038 if (!fstat(fd, &st)) {
1039 *size = xsize_t(st.st_size);
1040 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1041 }
44d1c19e 1042 close(fd);
0fcfd160 1043 }
0fcfd160
LT
1044 return map;
1045}
1046
726f852b 1047static int legacy_loose_object(unsigned char *map)
bb6b8e4f
JH
1048{
1049 unsigned int word;
1050
1051 /*
1052 * Is it a zlib-compressed buffer? If so, the first byte
1053 * must be 0x78 (15-bit window size, deflated), and the
1054 * first 16-bit word is evenly divisible by 31
1055 */
1056 word = (map[0] << 8) + map[1];
1057 if (map[0] == 0x78 && !(word % 31))
1058 return 1;
1059 else
1060 return 0;
1061}
1062
72518e9c 1063unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1064{
ad1ed5ee 1065 unsigned shift;
93821bd9 1066 unsigned char c;
93821bd9 1067 unsigned long size;
ad1ed5ee
JH
1068 unsigned long used = 0;
1069
1070 c = buf[used++];
1071 *type = (c >> 4) & 7;
1072 size = c & 15;
1073 shift = 4;
1074 while (c & 0x80) {
1075 if (len <= used)
1076 return 0;
1077 if (sizeof(long) * 8 <= shift)
1078 return 0;
1079 c = buf[used++];
1080 size += (c & 0x7f) << shift;
1081 shift += 7;
1082 }
1083 *sizep = size;
1084 return used;
1085}
1086
1087static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1088{
1089 unsigned long size, used;
1090 static const char valid_loose_object_type[8] = {
1091 0, /* OBJ_EXT */
1092 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1093 0, /* "delta" and others are invalid in a loose object */
93821bd9 1094 };
ad1ed5ee 1095 enum object_type type;
93821bd9 1096
c4483576
LT
1097 /* Get the data stream */
1098 memset(stream, 0, sizeof(*stream));
1099 stream->next_in = map;
1100 stream->avail_in = mapsize;
1101 stream->next_out = buffer;
93821bd9
LT
1102 stream->avail_out = bufsiz;
1103
bb6b8e4f 1104 if (legacy_loose_object(map)) {
93821bd9
LT
1105 inflateInit(stream);
1106 return inflate(stream, 0);
1107 }
1108
726f852b
NP
1109
1110 /*
1111 * There used to be a second loose object header format which
1112 * was meant to mimic the in-pack format, allowing for direct
1113 * copy of the object data. This format turned up not to be
1114 * really worth it and we don't write it any longer. But we
1115 * can still read it.
1116 */
ad1ed5ee
JH
1117 used = unpack_object_header_gently(map, mapsize, &type, &size);
1118 if (!used || !valid_loose_object_type[type])
93821bd9 1119 return -1;
ad1ed5ee
JH
1120 map += used;
1121 mapsize -= used;
93821bd9
LT
1122
1123 /* Set up the stream for the rest.. */
1124 stream->next_in = map;
1125 stream->avail_in = mapsize;
c4483576 1126 inflateInit(stream);
93821bd9
LT
1127
1128 /* And generate the fake traditional header */
ad1ed5ee 1129 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
df843662 1130 typename(type), size);
93821bd9 1131 return 0;
c4483576
LT
1132}
1133
7efbff75 1134static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1135{
1136 int bytes = strlen(buffer) + 1;
d565b341 1137 unsigned char *buf = xmalloc(1+size);
93821bd9 1138 unsigned long n;
7efbff75 1139 int status = Z_OK;
5180cacc 1140
93821bd9
LT
1141 n = stream->total_out - bytes;
1142 if (n > size)
1143 n = size;
1144 memcpy(buf, (char *) buffer + bytes, n);
1145 bytes = n;
456cdf6e
LT
1146 if (bytes <= size) {
1147 /*
1148 * The above condition must be (bytes <= size), not
1149 * (bytes < size). In other words, even though we
1150 * expect no more output and set avail_out to zer0,
1151 * the input zlib stream may have bytes that express
1152 * "this concludes the stream", and we *do* want to
1153 * eat that input.
1154 *
1155 * Otherwise we would not be able to test that we
1156 * consumed all the input to reach the expected size;
1157 * we also want to check that zlib tells us that all
1158 * went well with status == Z_STREAM_END at the end.
1159 */
5180cacc
LT
1160 stream->next_out = buf + bytes;
1161 stream->avail_out = size - bytes;
7efbff75
JH
1162 while (status == Z_OK)
1163 status = inflate(stream, Z_FINISH);
5180cacc
LT
1164 }
1165 buf[size] = 0;
456cdf6e 1166 if (status == Z_STREAM_END && !stream->avail_in) {
7efbff75
JH
1167 inflateEnd(stream);
1168 return buf;
1169 }
1170
1171 if (status < 0)
1172 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1173 else if (stream->avail_in)
1174 error("garbage at end of loose object '%s'",
1175 sha1_to_hex(sha1));
1176 free(buf);
1177 return NULL;
5180cacc
LT
1178}
1179
1180/*
1181 * We used to just use "sscanf()", but that's actually way
1182 * too permissive for what we want to check. So do an anal
1183 * object header parse by hand.
1184 */
21666f1a 1185static int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1186{
21666f1a 1187 char type[10];
5180cacc
LT
1188 int i;
1189 unsigned long size;
1190
1191 /*
a6080a0a 1192 * The type can be at most ten bytes (including the
5180cacc 1193 * terminating '\0' that we add), and is followed by
21666f1a 1194 * a space.
5180cacc 1195 */
21666f1a 1196 i = 0;
5180cacc
LT
1197 for (;;) {
1198 char c = *hdr++;
1199 if (c == ' ')
1200 break;
21666f1a
NP
1201 type[i++] = c;
1202 if (i >= sizeof(type))
5180cacc 1203 return -1;
5180cacc 1204 }
21666f1a 1205 type[i] = 0;
5180cacc
LT
1206
1207 /*
1208 * The length must follow immediately, and be in canonical
1209 * decimal format (ie "010" is not valid).
1210 */
1211 size = *hdr++ - '0';
1212 if (size > 9)
1213 return -1;
1214 if (size) {
1215 for (;;) {
1216 unsigned long c = *hdr - '0';
1217 if (c > 9)
1218 break;
1219 hdr++;
1220 size = size * 10 + c;
1221 }
1222 }
1223 *sizep = size;
1224
1225 /*
1226 * The length must be followed by a zero byte
1227 */
21666f1a 1228 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1229}
1230
7efbff75 1231static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1232{
5180cacc 1233 int ret;
0fcfd160 1234 z_stream stream;
5180cacc 1235 char hdr[8192];
0fcfd160 1236
5180cacc 1237 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1238 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1239 return NULL;
1240
7efbff75 1241 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1242}
1243
54dab52a
NP
1244unsigned long get_size_from_delta(struct packed_git *p,
1245 struct pack_window **w_curs,
1246 off_t curpos)
1247{
1248 const unsigned char *data;
1249 unsigned char delta_head[20], *in;
1250 z_stream stream;
1251 int st;
1252
1253 memset(&stream, 0, sizeof(stream));
1254 stream.next_out = delta_head;
1255 stream.avail_out = sizeof(delta_head);
1256
1257 inflateInit(&stream);
1258 do {
1259 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1260 stream.next_in = in;
1261 st = inflate(&stream, Z_FINISH);
1262 curpos += stream.next_in - in;
1263 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1264 stream.total_out < sizeof(delta_head));
1265 inflateEnd(&stream);
1266 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1267 die("delta data unpack-initial failed");
1268
1269 /* Examine the initial part of the delta to figure out
1270 * the result size.
1271 */
1272 data = delta_head;
1273
1274 /* ignore base size */
1275 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1276
1277 /* Read the result size */
1278 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1279}
1280
c4001d92 1281static off_t get_delta_base(struct packed_git *p,
03e79c88 1282 struct pack_window **w_curs,
c4001d92 1283 off_t *curpos,
21666f1a 1284 enum object_type type,
c4001d92 1285 off_t delta_obj_offset)
eb32d236 1286{
2b87c45b 1287 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1288 off_t base_offset;
eb32d236 1289
8d8a4ea5
SP
1290 /* use_pack() assured us we have [base_info, base_info + 20)
1291 * as a range that we can look at without walking off the
1292 * end of the mapped window. Its actually the hash size
1293 * that is assured. An OFS_DELTA longer than the hash size
1294 * is stupid, as then a REF_DELTA would be smaller to store.
1295 */
21666f1a 1296 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1297 unsigned used = 0;
1298 unsigned char c = base_info[used++];
1299 base_offset = c & 127;
1300 while (c & 128) {
1301 base_offset += 1;
8723f216 1302 if (!base_offset || MSB(base_offset, 7))
eb32d236
NP
1303 die("offset value overflow for delta base object");
1304 c = base_info[used++];
1305 base_offset = (base_offset << 7) + (c & 127);
1306 }
1307 base_offset = delta_obj_offset - base_offset;
1308 if (base_offset >= delta_obj_offset)
1309 die("delta base offset out of bound");
2b87c45b 1310 *curpos += used;
21666f1a 1311 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1312 /* The base entry _must_ be in the same pack */
1313 base_offset = find_pack_entry_one(base_info, p);
1314 if (!base_offset)
1315 die("failed to find delta-pack base object %s",
1316 sha1_to_hex(base_info));
2b87c45b 1317 *curpos += 20;
eb32d236
NP
1318 } else
1319 die("I am totally screwed");
2b87c45b 1320 return base_offset;
eb32d236
NP
1321}
1322
f3bf9224 1323/* forward declaration for a mutually recursive function */
c4001d92 1324static int packed_object_info(struct packed_git *p, off_t offset,
21666f1a 1325 unsigned long *sizep);
f3bf9224 1326
43057304 1327static int packed_delta_info(struct packed_git *p,
03e79c88 1328 struct pack_window **w_curs,
c4001d92 1329 off_t curpos,
21666f1a 1330 enum object_type type,
c4001d92 1331 off_t obj_offset,
43057304 1332 unsigned long *sizep)
5db47c2b 1333{
c4001d92 1334 off_t base_offset;
f3bf9224 1335
21666f1a
NP
1336 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1337 type = packed_object_info(p, base_offset, NULL);
f3bf9224 1338
c62266f3
JH
1339 /* We choose to only get the type of the base object and
1340 * ignore potentially corrupt pack file that expects the delta
1341 * based on a base with a wrong size. This saves tons of
1342 * inflate() calls.
1343 */
54dab52a
NP
1344 if (sizep)
1345 *sizep = get_size_from_delta(p, w_curs, curpos);
21666f1a
NP
1346
1347 return type;
5db47c2b
JH
1348}
1349
2b87c45b
NP
1350static int unpack_object_header(struct packed_git *p,
1351 struct pack_window **w_curs,
c4001d92 1352 off_t *curpos,
2b87c45b 1353 unsigned long *sizep)
a733cb60 1354{
03e79c88
SP
1355 unsigned char *base;
1356 unsigned int left;
ad1ed5ee 1357 unsigned long used;
2b87c45b 1358 enum object_type type;
a733cb60 1359
8d8a4ea5
SP
1360 /* use_pack() assures us we have [base, base + 20) available
1361 * as a range that we can look at at. (Its actually the hash
3dff5379 1362 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1363 * the maximum deflated object size is 2^137, which is just
1364 * insane, so we know won't exceed what we have been given.
1365 */
2b87c45b
NP
1366 base = use_pack(p, w_curs, *curpos, &left);
1367 used = unpack_object_header_gently(base, left, &type, sizep);
ad1ed5ee
JH
1368 if (!used)
1369 die("object offset outside of pack file");
2b87c45b 1370 *curpos += used;
ad1ed5ee 1371
2b87c45b 1372 return type;
a733cb60
LT
1373}
1374
21666f1a 1375const char *packed_object_info_detail(struct packed_git *p,
c4001d92 1376 off_t obj_offset,
21666f1a
NP
1377 unsigned long *size,
1378 unsigned long *store_size,
1379 unsigned int *delta_chain_length,
1380 unsigned char *base_sha1)
ad8c80a5 1381{
03e79c88 1382 struct pack_window *w_curs = NULL;
c4001d92
SP
1383 off_t curpos;
1384 unsigned long dummy;
43057304 1385 unsigned char *next_sha1;
21666f1a 1386 enum object_type type;
70f5d5d3 1387 struct revindex_entry *revidx;
ad8c80a5 1388
43057304 1389 *delta_chain_length = 0;
2b87c45b 1390 curpos = obj_offset;
21666f1a 1391 type = unpack_object_header(p, &w_curs, &curpos, size);
43057304 1392
70f5d5d3
NP
1393 revidx = find_pack_revindex(p, obj_offset);
1394 *store_size = revidx[1].offset - obj_offset;
1395
43057304 1396 for (;;) {
21666f1a 1397 switch (type) {
43057304 1398 default:
08a19d87 1399 die("pack %s contains unknown object type %d",
21666f1a 1400 p->pack_name, type);
43057304
NP
1401 case OBJ_COMMIT:
1402 case OBJ_TREE:
1403 case OBJ_BLOB:
1404 case OBJ_TAG:
03e79c88 1405 unuse_pack(&w_curs);
21666f1a 1406 return typename(type);
eb32d236 1407 case OBJ_OFS_DELTA:
21666f1a 1408 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
eb32d236 1409 if (*delta_chain_length == 0) {
70f5d5d3
NP
1410 revidx = find_pack_revindex(p, obj_offset);
1411 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
eb32d236
NP
1412 }
1413 break;
1414 case OBJ_REF_DELTA:
2b87c45b 1415 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
43057304
NP
1416 if (*delta_chain_length == 0)
1417 hashcpy(base_sha1, next_sha1);
2b87c45b 1418 obj_offset = find_pack_entry_one(next_sha1, p);
43057304
NP
1419 break;
1420 }
43057304 1421 (*delta_chain_length)++;
2b87c45b 1422 curpos = obj_offset;
21666f1a 1423 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
ad8c80a5 1424 }
ad8c80a5
JH
1425}
1426
c4001d92 1427static int packed_object_info(struct packed_git *p, off_t obj_offset,
21666f1a 1428 unsigned long *sizep)
1f688557 1429{
03e79c88 1430 struct pack_window *w_curs = NULL;
c4001d92
SP
1431 unsigned long size;
1432 off_t curpos = obj_offset;
21666f1a 1433 enum object_type type;
5db47c2b 1434
21666f1a 1435 type = unpack_object_header(p, &w_curs, &curpos, &size);
5db47c2b 1436
21666f1a 1437 switch (type) {
eb32d236
NP
1438 case OBJ_OFS_DELTA:
1439 case OBJ_REF_DELTA:
21666f1a
NP
1440 type = packed_delta_info(p, &w_curs, curpos,
1441 type, obj_offset, sizep);
2b87c45b 1442 break;
a733cb60 1443 case OBJ_COMMIT:
a733cb60 1444 case OBJ_TREE:
a733cb60 1445 case OBJ_BLOB:
a733cb60 1446 case OBJ_TAG:
2b87c45b
NP
1447 if (sizep)
1448 *sizep = size;
a69d0943 1449 break;
1f688557 1450 default:
08a19d87 1451 die("pack %s contains unknown object type %d",
21666f1a 1452 p->pack_name, type);
1f688557 1453 }
2b87c45b 1454 unuse_pack(&w_curs);
21666f1a 1455 return type;
1f688557
JH
1456}
1457
eb950c19 1458static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1459 struct pack_window **w_curs,
c4001d92 1460 off_t curpos,
eb950c19 1461 unsigned long size)
de530aaa
SP
1462{
1463 int st;
1464 z_stream stream;
079afb18 1465 unsigned char *buffer, *in;
de530aaa
SP
1466
1467 buffer = xmalloc(size + 1);
1468 buffer[size] = 0;
1469 memset(&stream, 0, sizeof(stream));
de530aaa
SP
1470 stream.next_out = buffer;
1471 stream.avail_out = size;
1472
1473 inflateInit(&stream);
079afb18 1474 do {
2b87c45b 1475 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18
SP
1476 stream.next_in = in;
1477 st = inflate(&stream, Z_FINISH);
2b87c45b 1478 curpos += stream.next_in - in;
079afb18 1479 } while (st == Z_OK || st == Z_BUF_ERROR);
de530aaa
SP
1480 inflateEnd(&stream);
1481 if ((st != Z_STREAM_END) || stream.total_out != size) {
1482 free(buffer);
1483 return NULL;
1484 }
1485
1486 return buffer;
1487}
1488
e5e01619
LT
1489#define MAX_DELTA_CACHE (256)
1490
18bdec11 1491static size_t delta_base_cached;
5e08ecbf
NP
1492
1493static struct delta_base_cache_lru_list {
1494 struct delta_base_cache_lru_list *prev;
1495 struct delta_base_cache_lru_list *next;
1496} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1497
e5e01619 1498static struct delta_base_cache_entry {
5e08ecbf
NP
1499 struct delta_base_cache_lru_list lru;
1500 void *data;
e5e01619
LT
1501 struct packed_git *p;
1502 off_t base_offset;
1503 unsigned long size;
e5e01619
LT
1504 enum object_type type;
1505} delta_base_cache[MAX_DELTA_CACHE];
1506
1507static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1508{
1509 unsigned long hash;
1510
1511 hash = (unsigned long)p + (unsigned long)base_offset;
1512 hash += (hash >> 8) + (hash >> 16);
3358004a 1513 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1514}
1515
62f255ad 1516static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 1517 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 1518{
e5e01619
LT
1519 void *ret;
1520 unsigned long hash = pack_entry_hash(p, base_offset);
1521 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1522
1523 ret = ent->data;
1524 if (ret && ent->p == p && ent->base_offset == base_offset)
1525 goto found_cache_entry;
62f255ad 1526 return unpack_entry(p, base_offset, type, base_size);
e5e01619
LT
1527
1528found_cache_entry:
18bdec11 1529 if (!keep_cache) {
a0cba108 1530 ent->data = NULL;
5e08ecbf
NP
1531 ent->lru.next->prev = ent->lru.prev;
1532 ent->lru.prev->next = ent->lru.next;
18bdec11 1533 delta_base_cached -= ent->size;
182af834
PH
1534 } else {
1535 ret = xmemdupz(ent->data, ent->size);
a0cba108 1536 }
e5e01619
LT
1537 *type = ent->type;
1538 *base_size = ent->size;
1539 return ret;
62f255ad
LT
1540}
1541
18bdec11
SP
1542static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1543{
1544 if (ent->data) {
1545 free(ent->data);
1546 ent->data = NULL;
5e08ecbf
NP
1547 ent->lru.next->prev = ent->lru.prev;
1548 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
1549 delta_base_cached -= ent->size;
1550 }
1551}
1552
62f255ad
LT
1553static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1554 void *base, unsigned long base_size, enum object_type type)
1555{
5e08ecbf 1556 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 1557 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 1558 struct delta_base_cache_lru_list *lru;
e5e01619 1559
18bdec11
SP
1560 release_delta_base_cache(ent);
1561 delta_base_cached += base_size;
5e08ecbf
NP
1562
1563 for (lru = delta_base_cache_lru.next;
1564 delta_base_cached > delta_base_cache_limit
1565 && lru != &delta_base_cache_lru;
1566 lru = lru->next) {
1567 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
1568 if (f->type == OBJ_BLOB)
1569 release_delta_base_cache(f);
1570 }
5e08ecbf
NP
1571 for (lru = delta_base_cache_lru.next;
1572 delta_base_cached > delta_base_cache_limit
1573 && lru != &delta_base_cache_lru;
1574 lru = lru->next) {
1575 struct delta_base_cache_entry *f = (void *)lru;
1576 release_delta_base_cache(f);
1577 }
18bdec11 1578
e5e01619
LT
1579 ent->p = p;
1580 ent->base_offset = base_offset;
1581 ent->type = type;
1582 ent->data = base;
1583 ent->size = base_size;
5e08ecbf
NP
1584 ent->lru.next = &delta_base_cache_lru;
1585 ent->lru.prev = delta_base_cache_lru.prev;
1586 delta_base_cache_lru.prev->next = &ent->lru;
1587 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
1588}
1589
eb950c19 1590static void *unpack_delta_entry(struct packed_git *p,
03e79c88 1591 struct pack_window **w_curs,
c4001d92 1592 off_t curpos,
1f688557 1593 unsigned long delta_size,
c4001d92 1594 off_t obj_offset,
21666f1a 1595 enum object_type *type,
eb950c19 1596 unsigned long *sizep)
1f688557 1597{
7c3e8be3 1598 void *delta_data, *result, *base;
c4001d92
SP
1599 unsigned long base_size;
1600 off_t base_offset;
43057304 1601
21666f1a 1602 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
a0cba108 1603 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
67686d95 1604 if (!base)
c4001d92
SP
1605 die("failed to read delta base object"
1606 " at %"PRIuMAX" from %s",
1607 (uintmax_t)base_offset, p->pack_name);
67686d95 1608
2b87c45b 1609 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
9064d87b
SP
1610 if (!delta_data)
1611 die("failed to unpack compressed delta"
1612 " at %"PRIuMAX" from %s",
1613 (uintmax_t)curpos, p->pack_name);
1f688557
JH
1614 result = patch_delta(base, base_size,
1615 delta_data, delta_size,
21666f1a 1616 sizep);
1f688557
JH
1617 if (!result)
1618 die("failed to apply delta");
1619 free(delta_data);
62f255ad 1620 add_delta_base_cache(p, base_offset, base, base_size, *type);
1f688557
JH
1621 return result;
1622}
1623
c4001d92 1624void *unpack_entry(struct packed_git *p, off_t obj_offset,
21666f1a 1625 enum object_type *type, unsigned long *sizep)
1f688557 1626{
03e79c88 1627 struct pack_window *w_curs = NULL;
c4001d92 1628 off_t curpos = obj_offset;
21666f1a 1629 void *data;
1f688557 1630
21666f1a
NP
1631 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1632 switch (*type) {
eb32d236
NP
1633 case OBJ_OFS_DELTA:
1634 case OBJ_REF_DELTA:
21666f1a
NP
1635 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1636 obj_offset, type, sizep);
4d703a1a 1637 break;
a733cb60 1638 case OBJ_COMMIT:
a733cb60 1639 case OBJ_TREE:
a733cb60 1640 case OBJ_BLOB:
a733cb60 1641 case OBJ_TAG:
21666f1a 1642 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
4d703a1a 1643 break;
1f688557 1644 default:
21666f1a 1645 die("unknown object type %i in %s", *type, p->pack_name);
1f688557 1646 }
03e79c88 1647 unuse_pack(&w_curs);
21666f1a 1648 return data;
1f688557
JH
1649}
1650
d079837e 1651const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 1652 uint32_t n)
9a217f2a 1653{
42873078 1654 const unsigned char *index = p->index_data;
d079837e
SP
1655 if (!index) {
1656 if (open_pack_index(p))
1657 return NULL;
1658 index = p->index_data;
1659 }
57059091 1660 if (n >= p->num_objects)
d72308e0 1661 return NULL;
74e34e1f
NP
1662 index += 4 * 256;
1663 if (p->index_version == 1) {
1664 return index + 24 * n + 4;
1665 } else {
1666 index += 8;
1667 return index + 20 * n;
1668 }
1669}
1670
1671static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1672{
1673 const unsigned char *index = p->index_data;
1674 index += 4 * 256;
1675 if (p->index_version == 1) {
1676 return ntohl(*((uint32_t *)(index + 24 * n)));
1677 } else {
1678 uint32_t off;
1679 index += 8 + p->num_objects * (20 + 4);
1680 off = ntohl(*((uint32_t *)(index + 4 * n)));
1681 if (!(off & 0x80000000))
1682 return off;
1683 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1684 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1685 ntohl(*((uint32_t *)(index + 4)));
1686 }
9a217f2a
JH
1687}
1688
c4001d92 1689off_t find_pack_entry_one(const unsigned char *sha1,
43057304 1690 struct packed_git *p)
1f688557 1691{
42873078 1692 const uint32_t *level1_ofs = p->index_data;
42873078 1693 const unsigned char *index = p->index_data;
628522ec
JH
1694 unsigned hi, lo, stride;
1695 static int use_lookup = -1;
1696 static int debug_lookup = -1;
1697
1698 if (debug_lookup < 0)
1699 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 1700
d079837e
SP
1701 if (!index) {
1702 if (open_pack_index(p))
1703 return 0;
1704 level1_ofs = p->index_data;
1705 index = p->index_data;
1706 }
74e34e1f
NP
1707 if (p->index_version > 1) {
1708 level1_ofs += 2;
1709 index += 8;
1710 }
42873078 1711 index += 4 * 256;
74e34e1f
NP
1712 hi = ntohl(level1_ofs[*sha1]);
1713 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
1714 if (p->index_version > 1) {
1715 stride = 20;
1716 } else {
1717 stride = 24;
1718 index += 4;
1719 }
1720
1721 if (debug_lookup)
1722 printf("%02x%02x%02x... lo %u hi %u nr %u\n",
1723 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1724
1725 if (use_lookup < 0)
1726 use_lookup = !!getenv("GIT_USE_LOOKUP");
1727 if (use_lookup) {
1728 int pos = sha1_entry_pos(index, stride, 0,
1729 lo, hi, p->num_objects, sha1);
1730 if (pos < 0)
1731 return 0;
1732 return nth_packed_object_offset(p, pos);
1733 }
1f688557
JH
1734
1735 do {
74e34e1f 1736 unsigned mi = (lo + hi) / 2;
628522ec
JH
1737 int cmp = hashcmp(index + mi * stride, sha1);
1738
1739 if (debug_lookup)
1740 printf("lo %u hi %u rg %u mi %u\n",
1741 lo, hi, hi - lo, mi);
43057304 1742 if (!cmp)
74e34e1f 1743 return nth_packed_object_offset(p, mi);
1f688557
JH
1744 if (cmp > 0)
1745 hi = mi;
1746 else
1747 lo = mi+1;
1748 } while (lo < hi);
1749 return 0;
1750}
1751
000dfd3f 1752int matches_pack_name(struct packed_git *p, const char *name)
d4ff6d92
SP
1753{
1754 const char *last_c, *c;
1755
000dfd3f
JH
1756 if (!strcmp(p->pack_name, name))
1757 return 1;
d4ff6d92
SP
1758
1759 for (c = p->pack_name, last_c = c; *c;)
1760 if (*c == '/')
1761 last_c = ++c;
1762 else
1763 ++c;
000dfd3f
JH
1764 if (!strcmp(last_c, name))
1765 return 1;
d4ff6d92 1766
000dfd3f 1767 return 0;
d4ff6d92
SP
1768}
1769
106d710b 1770static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1f688557 1771{
f7c22cc6 1772 static struct packed_git *last_found = (void *)1;
1f688557 1773 struct packed_git *p;
c4001d92 1774 off_t offset;
43057304 1775
1f688557 1776 prepare_packed_git();
f7c22cc6
NP
1777 if (!packed_git)
1778 return 0;
1779 p = (last_found == (void *)1) ? packed_git : last_found;
1f688557 1780
f7c22cc6 1781 do {
106d710b
JH
1782 if (ignore_packed) {
1783 const char **ig;
1784 for (ig = ignore_packed; *ig; ig++)
000dfd3f 1785 if (matches_pack_name(p, *ig))
106d710b
JH
1786 break;
1787 if (*ig)
f7c22cc6 1788 goto next;
106d710b 1789 }
f7c22cc6 1790
43057304
NP
1791 offset = find_pack_entry_one(sha1, p);
1792 if (offset) {
c715f783
SP
1793 /*
1794 * We are about to tell the caller where they can
1795 * locate the requested object. We better make
1796 * sure the packfile is still here and can be
1797 * accessed before supplying that answer, as
1798 * it may have been deleted since the index
1799 * was loaded!
1800 */
1801 if (p->pack_fd == -1 && open_packed_git(p)) {
1802 error("packfile %s cannot be accessed", p->pack_name);
f7c22cc6 1803 goto next;
c715f783 1804 }
43057304
NP
1805 e->offset = offset;
1806 e->p = p;
1807 hashcpy(e->sha1, sha1);
f7c22cc6 1808 last_found = p;
1f688557 1809 return 1;
43057304 1810 }
f7c22cc6
NP
1811
1812 next:
1813 if (p == last_found)
1814 p = packed_git;
1815 else
1816 p = p->next;
1817 if (p == last_found)
1818 p = p->next;
1819 } while (p);
1f688557
JH
1820 return 0;
1821}
1822
a6080a0a 1823struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
1824 struct packed_git *packs)
1825{
1826 struct packed_git *p;
bf592c50
DB
1827
1828 for (p = packs; p; p = p->next) {
43057304 1829 if (find_pack_entry_one(sha1, p))
bf592c50
DB
1830 return p;
1831 }
1832 return NULL;
21666f1a 1833
bf592c50
DB
1834}
1835
21666f1a 1836static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
65c2e0c3 1837{
36e4d74a 1838 int status;
65c2e0c3
JH
1839 unsigned long mapsize, size;
1840 void *map;
1841 z_stream stream;
d65a16f6 1842 char hdr[32];
65c2e0c3 1843
bb6b8e4f 1844 map = map_sha1_file(sha1, &mapsize);
f0df4ed5
JS
1845 if (!map)
1846 return error("unable to find %s", sha1_to_hex(sha1));
36e4d74a
JH
1847 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1848 status = error("unable to unpack %s header",
1849 sha1_to_hex(sha1));
21666f1a 1850 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 1851 status = error("unable to parse %s header", sha1_to_hex(sha1));
21666f1a
NP
1852 else if (sizep)
1853 *sizep = size;
65c2e0c3
JH
1854 inflateEnd(&stream);
1855 munmap(map, mapsize);
1856 return status;
1857}
1858
21666f1a 1859int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
f0df4ed5 1860{
f0df4ed5
JS
1861 struct pack_entry e;
1862
1863 if (!find_pack_entry(sha1, &e, NULL)) {
1864 reprepare_packed_git();
1865 if (!find_pack_entry(sha1, &e, NULL))
21666f1a 1866 return sha1_loose_object_info(sha1, sizep);
f0df4ed5 1867 }
21666f1a 1868 return packed_object_info(e.p, e.offset, sizep);
f0df4ed5
JS
1869}
1870
21666f1a
NP
1871static void *read_packed_sha1(const unsigned char *sha1,
1872 enum object_type *type, unsigned long *size)
1f688557
JH
1873{
1874 struct pack_entry e;
1875
8276c007 1876 if (!find_pack_entry(sha1, &e, NULL))
1f688557 1877 return NULL;
8276c007 1878 else
a0cba108 1879 return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1f688557
JH
1880}
1881
d66b37bb
JH
1882/*
1883 * This is meant to hold a *small* number of objects that you would
1884 * want read_sha1_file() to be able to return, but yet you do not want
1885 * to write them into the object store (e.g. a browse-only
1886 * application).
1887 */
1888static struct cached_object {
1889 unsigned char sha1[20];
21666f1a 1890 enum object_type type;
d66b37bb
JH
1891 void *buf;
1892 unsigned long size;
1893} *cached_objects;
1894static int cached_object_nr, cached_object_alloc;
1895
346245a1
JK
1896static struct cached_object empty_tree = {
1897 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
1898 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
1899 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
1900 OBJ_TREE,
1901 "",
1902 0
1903};
1904
d66b37bb
JH
1905static struct cached_object *find_cached_object(const unsigned char *sha1)
1906{
1907 int i;
1908 struct cached_object *co = cached_objects;
1909
1910 for (i = 0; i < cached_object_nr; i++, co++) {
1911 if (!hashcmp(co->sha1, sha1))
1912 return co;
1913 }
346245a1
JK
1914 if (!hashcmp(sha1, empty_tree.sha1))
1915 return &empty_tree;
d66b37bb
JH
1916 return NULL;
1917}
1918
21666f1a
NP
1919int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1920 unsigned char *sha1)
d66b37bb
JH
1921{
1922 struct cached_object *co;
1923
21666f1a 1924 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
1925 if (has_sha1_file(sha1) || find_cached_object(sha1))
1926 return 0;
1927 if (cached_object_alloc <= cached_object_nr) {
1928 cached_object_alloc = alloc_nr(cached_object_alloc);
1929 cached_objects = xrealloc(cached_objects,
1930 sizeof(*cached_objects) *
1931 cached_object_alloc);
1932 }
1933 co = &cached_objects[cached_object_nr++];
1934 co->size = len;
21666f1a 1935 co->type = type;
efa13f7b
JH
1936 co->buf = xmalloc(len);
1937 memcpy(co->buf, buf, len);
d66b37bb
JH
1938 hashcpy(co->sha1, sha1);
1939 return 0;
1940}
1941
21666f1a
NP
1942void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1943 unsigned long *size)
0fcfd160
LT
1944{
1945 unsigned long mapsize;
1946 void *map, *buf;
d66b37bb
JH
1947 struct cached_object *co;
1948
1949 co = find_cached_object(sha1);
1950 if (co) {
21666f1a 1951 *type = co->type;
d66b37bb 1952 *size = co->size;
182af834 1953 return xmemdupz(co->buf, co->size);
d66b37bb 1954 }
0fcfd160 1955
8276c007
PE
1956 buf = read_packed_sha1(sha1, type, size);
1957 if (buf)
1958 return buf;
bb6b8e4f 1959 map = map_sha1_file(sha1, &mapsize);
0fcfd160 1960 if (map) {
7efbff75 1961 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
1962 munmap(map, mapsize);
1963 return buf;
1964 }
637cdd9d 1965 reprepare_packed_git();
8276c007 1966 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
1967}
1968
40469ee9 1969void *read_object_with_reference(const unsigned char *sha1,
21666f1a 1970 const char *required_type_name,
40469ee9
JH
1971 unsigned long *size,
1972 unsigned char *actual_sha1_return)
f4913f91 1973{
21666f1a 1974 enum object_type type, required_type;
f4913f91
JH
1975 void *buffer;
1976 unsigned long isize;
40469ee9 1977 unsigned char actual_sha1[20];
f4913f91 1978
21666f1a 1979 required_type = type_from_string(required_type_name);
e702496e 1980 hashcpy(actual_sha1, sha1);
40469ee9
JH
1981 while (1) {
1982 int ref_length = -1;
1983 const char *ref_type = NULL;
f4913f91 1984
21666f1a 1985 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
1986 if (!buffer)
1987 return NULL;
21666f1a 1988 if (type == required_type) {
40469ee9
JH
1989 *size = isize;
1990 if (actual_sha1_return)
e702496e 1991 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
1992 return buffer;
1993 }
1994 /* Handle references */
21666f1a 1995 else if (type == OBJ_COMMIT)
40469ee9 1996 ref_type = "tree ";
21666f1a 1997 else if (type == OBJ_TAG)
40469ee9
JH
1998 ref_type = "object ";
1999 else {
2000 free(buffer);
2001 return NULL;
2002 }
2003 ref_length = strlen(ref_type);
f4913f91 2004
50974ec9
MK
2005 if (ref_length + 40 > isize ||
2006 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2007 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2008 free(buffer);
2009 return NULL;
2010 }
1cf58e72 2011 free(buffer);
40469ee9
JH
2012 /* Now we have the ID of the referred-to object in
2013 * actual_sha1. Check again. */
f4913f91 2014 }
f4913f91
JH
2015}
2016
ce9fbf16 2017static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2018 const char *type, unsigned char *sha1,
d65a16f6 2019 char *hdr, int *hdrlen)
d410c0f5
JH
2020{
2021 SHA_CTX c;
2022
2023 /* Generate the header */
d65a16f6 2024 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
2025
2026 /* Sha1.. */
2027 SHA1_Init(&c);
2028 SHA1_Update(&c, hdr, *hdrlen);
2029 SHA1_Update(&c, buf, len);
2030 SHA1_Final(sha1, &c);
d410c0f5
JH
2031}
2032
230f1322
LT
2033/*
2034 * Move the just written object into its final resting place
2035 */
839837b9 2036int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322 2037{
5723fe7e 2038 int ret = link(tmpfile, filename);
7ebb6fca
LT
2039
2040 /*
2041 * Coda hack - coda doesn't like cross-directory links,
2042 * so we fall back to a rename, which will mean that it
2043 * won't be able to check collisions, but that's not a
2044 * big deal.
2045 *
2046 * The same holds for FAT formatted media.
2047 *
2048 * When this succeeds, we just return 0. We have nothing
2049 * left to unlink.
2050 */
2051 if (ret && ret != EEXIST) {
2052 if (!rename(tmpfile, filename))
230f1322 2053 return 0;
9e48b389 2054 ret = errno;
230f1322
LT
2055 }
2056 unlink(tmpfile);
2057 if (ret) {
2058 if (ret != EEXIST) {
916d081b 2059 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
230f1322
LT
2060 }
2061 /* FIXME!!! Collision check here ? */
2062 }
2063
2064 return 0;
2065}
2066
4d548150
LT
2067static int write_buffer(int fd, const void *buf, size_t len)
2068{
d34cf19b 2069 if (write_in_full(fd, buf, len) < 0)
93822c22 2070 return error("file write error (%s)", strerror(errno));
4d548150
LT
2071 return 0;
2072}
2073
ce9fbf16 2074int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2075 unsigned char *sha1)
2076{
d65a16f6 2077 char hdr[32];
abdc3fc8
RS
2078 int hdrlen;
2079 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2080 return 0;
2081}
2082
e9039dd3
LT
2083/* Finalize a file on disk, and close it. */
2084static void close_sha1_file(int fd)
2085{
2086 /* For safe-mode, we could fsync_or_die(fd, "sha1 file") here */
2087 fchmod(fd, 0444);
2088 if (close(fd) != 0)
2089 die("unable to write sha1 file");
2090}
2091
5723fe7e
LT
2092/* Size of directory component, including the ending '/' */
2093static inline int directory_size(const char *filename)
2094{
2095 const char *s = strrchr(filename, '/');
2096 if (!s)
2097 return 0;
2098 return s - filename + 1;
2099}
2100
2101/*
2102 * This creates a temporary file in the same directory as the final
2103 * 'filename'
2104 *
2105 * We want to avoid cross-directory filename renames, because those
2106 * can have problems on various filesystems (FAT, NFS, Coda).
2107 */
2108static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2109{
2110 int fd, dirlen = directory_size(filename);
2111
2112 if (dirlen + 20 > bufsiz) {
2113 errno = ENAMETOOLONG;
2114 return -1;
2115 }
2116 memcpy(buffer, filename, dirlen);
2117 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2118 fd = mkstemp(buffer);
2119 if (fd < 0 && dirlen) {
2120 /* Make sure the directory exists */
6ff6af62 2121 memcpy(buffer, filename, dirlen);
5723fe7e 2122 buffer[dirlen-1] = 0;
3bfaf018 2123 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
5723fe7e
LT
2124 return -1;
2125
2126 /* Try again */
2127 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2128 fd = mkstemp(buffer);
2129 }
2130 return fd;
2131}
2132
bbac7311
NP
2133static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2134 void *buf, unsigned long len, time_t mtime)
0fcfd160 2135{
bbac7311 2136 int fd, size, ret;
bf0f910d 2137 unsigned char *compressed;
0fcfd160 2138 z_stream stream;
706bc531 2139 char *filename;
aac17941 2140 static char tmpfile[PATH_MAX];
a44c9a5e 2141
972a9155 2142 filename = sha1_file_name(sha1);
5723fe7e 2143 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
aac17941 2144 if (fd < 0) {
916d081b
PB
2145 if (errno == EPERM)
2146 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2147 else
2148 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
aac17941
LT
2149 }
2150
0fcfd160
LT
2151 /* Set it up */
2152 memset(&stream, 0, sizeof(stream));
12f6c308 2153 deflateInit(&stream, zlib_compression_level);
93821bd9 2154 size = 8 + deflateBound(&stream, len+hdrlen);
812666c8 2155 compressed = xmalloc(size);
0fcfd160
LT
2156
2157 /* Compress it */
0fcfd160
LT
2158 stream.next_out = compressed;
2159 stream.avail_out = size;
a44c9a5e
LT
2160
2161 /* First header.. */
d65a16f6 2162 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2163 stream.avail_in = hdrlen;
726f852b
NP
2164 while (deflate(&stream, 0) == Z_OK)
2165 /* nothing */;
a44c9a5e
LT
2166
2167 /* Then the data itself.. */
2168 stream.next_in = buf;
2169 stream.avail_in = len;
ac54c277
LT
2170 ret = deflate(&stream, Z_FINISH);
2171 if (ret != Z_STREAM_END)
2172 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2173
2174 ret = deflateEnd(&stream);
2175 if (ret != Z_OK)
2176 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2177
0fcfd160
LT
2178 size = stream.total_out;
2179
4d548150
LT
2180 if (write_buffer(fd, compressed, size) < 0)
2181 die("unable to write sha1 file");
e9039dd3 2182 close_sha1_file(fd);
383f85b7 2183 free(compressed);
0fcfd160 2184
bbac7311
NP
2185 if (mtime) {
2186 struct utimbuf utb;
2187 utb.actime = mtime;
2188 utb.modtime = mtime;
2189 if (utime(tmpfile, &utb) < 0)
2190 warning("failed utime() on %s: %s",
2191 tmpfile, strerror(errno));
2192 }
2193
230f1322 2194 return move_temp_to_file(tmpfile, filename);
0fcfd160 2195}
8237b185 2196
bbac7311
NP
2197int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2198{
2199 unsigned char sha1[20];
2200 char hdr[32];
2201 int hdrlen;
2202
2203 /* Normally if we have it in the pack then we do not bother writing
2204 * it out into .git/objects/??/?{38} file.
2205 */
2206 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2207 if (returnsha1)
2208 hashcpy(returnsha1, sha1);
2209 if (has_sha1_file(sha1))
2210 return 0;
2211 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2212}
2213
2214int force_object_loose(const unsigned char *sha1, time_t mtime)
2215{
bbac7311
NP
2216 void *buf;
2217 unsigned long len;
2218 enum object_type type;
2219 char hdr[32];
2220 int hdrlen;
2221
c529d75a 2222 if (has_loose_object(sha1))
bbac7311
NP
2223 return 0;
2224 buf = read_packed_sha1(sha1, &type, &len);
2225 if (!buf)
2226 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2227 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2228 return write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2229}
2230
bf592c50
DB
2231int has_pack_index(const unsigned char *sha1)
2232{
2233 struct stat st;
2234 if (stat(sha1_pack_index_name(sha1), &st))
2235 return 0;
2236 return 1;
2237}
2238
2239int has_pack_file(const unsigned char *sha1)
2240{
2241 struct stat st;
2242 if (stat(sha1_pack_name(sha1), &st))
2243 return 0;
2244 return 1;
2245}
2246
106d710b 2247int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
dade09c2
LT
2248{
2249 struct pack_entry e;
106d710b 2250 return find_pack_entry(sha1, &e, ignore_packed);
dade09c2
LT
2251}
2252
8237b185
DB
2253int has_sha1_file(const unsigned char *sha1)
2254{
1f688557
JH
2255 struct pack_entry e;
2256
106d710b 2257 if (find_pack_entry(sha1, &e, NULL))
1f688557 2258 return 1;
c529d75a 2259 return has_loose_object(sha1);
8237b185 2260}
74400e71 2261
e7332f96
BE
2262int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2263{
fd17f5b5 2264 struct strbuf buf;
e7332f96 2265 int ret;
e7332f96 2266
fd17f5b5
PH
2267 strbuf_init(&buf, 0);
2268 if (strbuf_read(&buf, fd, 4096) < 0) {
2269 strbuf_release(&buf);
024510c8
DB
2270 return -1;
2271 }
e7332f96 2272
024510c8 2273 if (!type)
8e440259 2274 type = blob_type;
024510c8 2275 if (write_object)
fd17f5b5 2276 ret = write_sha1_file(buf.buf, buf.len, type, sha1);
abdc3fc8 2277 else
fd17f5b5
PH
2278 ret = hash_sha1_file(buf.buf, buf.len, type, sha1);
2279 strbuf_release(&buf);
2280
024510c8
DB
2281 return ret;
2282}
2283
edaec3fb 2284int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
53bca91a 2285 enum object_type type, const char *path)
74400e71 2286{
dc49cd76 2287 size_t size = xsize_t(st->st_size);
3a55602e 2288 void *buf = NULL;
6c510bee 2289 int ret, re_allocated = 0;
74400e71 2290
74400e71 2291 if (size)
c4712e45 2292 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
74400e71 2293 close(fd);
74400e71 2294
7672db20 2295 if (!type)
edaec3fb 2296 type = OBJ_BLOB;
6c510bee
LT
2297
2298 /*
2299 * Convert blobs to git internal format
2300 */
d0d8e14d 2301 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
5ecd293d
PH
2302 struct strbuf nbuf;
2303 strbuf_init(&nbuf, 0);
21e5ad50
SP
2304 if (convert_to_git(path, buf, size, &nbuf,
2305 write_object ? safe_crlf : 0)) {
ac78e548 2306 munmap(buf, size);
b315c5c0 2307 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
2308 re_allocated = 1;
2309 }
2310 }
2311
7672db20 2312 if (write_object)
edaec3fb 2313 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 2314 else
edaec3fb 2315 ret = hash_sha1_file(buf, size, typename(type), sha1);
6c510bee
LT
2316 if (re_allocated) {
2317 free(buf);
2318 return ret;
2319 }
aac17941
LT
2320 if (size)
2321 munmap(buf, size);
2322 return ret;
74400e71 2323}
ec1fcc16
JH
2324
2325int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2326{
2327 int fd;
2328 char *target;
dc49cd76 2329 size_t len;
ec1fcc16
JH
2330
2331 switch (st->st_mode & S_IFMT) {
2332 case S_IFREG:
2333 fd = open(path, O_RDONLY);
2334 if (fd < 0)
2335 return error("open(\"%s\"): %s", path,
2336 strerror(errno));
53bca91a 2337 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
ec1fcc16
JH
2338 return error("%s: failed to insert into database",
2339 path);
2340 break;
2341 case S_IFLNK:
dc49cd76
SP
2342 len = xsize_t(st->st_size);
2343 target = xmalloc(len + 1);
2344 if (readlink(path, target, len + 1) != st->st_size) {
ec1fcc16
JH
2345 char *errstr = strerror(errno);
2346 free(target);
2347 return error("readlink(\"%s\"): %s", path,
2348 errstr);
2349 }
abdc3fc8 2350 if (!write_object)
dc49cd76
SP
2351 hash_sha1_file(target, len, blob_type, sha1);
2352 else if (write_sha1_file(target, len, blob_type, sha1))
ec1fcc16
JH
2353 return error("%s: failed to insert into database",
2354 path);
2355 free(target);
2356 break;
f35a6d3b
LT
2357 case S_IFDIR:
2358 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
2359 default:
2360 return error("%s: unsupported file type", path);
2361 }
2362 return 0;
2363}
a69e5429
JH
2364
2365int read_pack_header(int fd, struct pack_header *header)
2366{
c697ad14
HO
2367 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2368 /* "eof before pack header was fully read" */
2369 return PH_ERROR_EOF;
2370
a69e5429
JH
2371 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2372 /* "protocol error (pack signature mismatch detected)" */
2373 return PH_ERROR_PACK_SIGNATURE;
2374 if (!pack_version_ok(header->hdr_version))
2375 /* "protocol error (pack version unsupported)" */
2376 return PH_ERROR_PROTOCOL;
2377 return 0;
2378}