clean up the delta base cache size a bit
[git/git.git] / sha1_file.c
CommitLineData
0fcfd160
LT
1/*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
0fcfd160 9#include "cache.h"
1f688557 10#include "delta.h"
a733cb60 11#include "pack.h"
8e440259
PE
12#include "blob.h"
13#include "commit.h"
14#include "tag.h"
15#include "tree.h"
0fcfd160 16
144bde78
LT
17#ifndef O_NOATIME
18#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
19#define O_NOATIME 01000000
20#else
21#define O_NOATIME 0
22#endif
23#endif
24
e05db0fd
PR
25#ifdef NO_C99_FORMAT
26#define SZ_FMT "lu"
27#else
28#define SZ_FMT "zu"
29#endif
30
96f1e58f 31const unsigned char null_sha1[20];
88cd621d 32
144bde78
LT
33static unsigned int sha1_file_open_flag = O_NOATIME;
34
e49521b5
JH
35signed char hexval_table[256] = {
36 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
37 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
38 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
42 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
43 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
44 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
45 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
47 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
48 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
68};
0fcfd160
LT
69
70int get_sha1_hex(const char *hex, unsigned char *sha1)
71{
72 int i;
73 for (i = 0; i < 20; i++) {
74 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
75 if (val & ~0xff)
76 return -1;
77 *sha1++ = val;
78 hex += 2;
79 }
80 return 0;
81}
82
b2cb9425
JH
83int safe_create_leading_directories(char *path)
84{
85 char *pos = path;
67d42212
JR
86 struct stat st;
87
67ffdf4c
JS
88 if (*pos == '/')
89 pos++;
b2cb9425
JH
90
91 while (pos) {
92 pos = strchr(pos, '/');
93 if (!pos)
94 break;
95 *pos = 0;
67d42212
JR
96 if (!stat(path, &st)) {
97 /* path exists */
98 if (!S_ISDIR(st.st_mode)) {
b2cb9425 99 *pos = '/';
67d42212 100 return -3;
b2cb9425 101 }
457f06d6 102 }
67d42212
JR
103 else if (mkdir(path, 0777)) {
104 *pos = '/';
105 return -1;
106 }
457f06d6
JS
107 else if (adjust_shared_perm(path)) {
108 *pos = '/';
109 return -2;
110 }
b2cb9425
JH
111 *pos++ = '/';
112 }
113 return 0;
114}
723c31fe 115
0fcfd160
LT
116char * sha1_to_hex(const unsigned char *sha1)
117{
dcb3450f
LT
118 static int bufno;
119 static char hexbuffer[4][50];
0fcfd160 120 static const char hex[] = "0123456789abcdef";
dcb3450f 121 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
0fcfd160
LT
122 int i;
123
124 for (i = 0; i < 20; i++) {
125 unsigned int val = *sha1++;
126 *buf++ = hex[val >> 4];
127 *buf++ = hex[val & 0xf];
128 }
1e80e044
JS
129 *buf = '\0';
130
0fcfd160
LT
131 return buffer;
132}
133
ace1534d
JH
134static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
135{
136 int i;
137 for (i = 0; i < 20; i++) {
138 static char hex[] = "0123456789abcdef";
139 unsigned int val = sha1[i];
140 char *pos = pathbuf + i*2 + (i > 0);
141 *pos++ = hex[val >> 4];
142 *pos = hex[val & 0xf];
143 }
144}
145
0fcfd160
LT
146/*
147 * NOTE! This returns a statically allocated buffer, so you have to be
9befac47 148 * careful about using it. Do a "xstrdup()" if you need to save the
0fcfd160 149 * filename.
ace1534d
JH
150 *
151 * Also note that this returns the location for creating. Reading
152 * SHA1 file can happen from any alternate directory listed in the
d19938ab 153 * DB_ENVIRONMENT environment variable if it is not found in
ace1534d 154 * the primary object database.
0fcfd160
LT
155 */
156char *sha1_file_name(const unsigned char *sha1)
157{
0fcfd160
LT
158 static char *name, *base;
159
160 if (!base) {
d19938ab 161 const char *sha1_file_directory = get_object_directory();
0fcfd160 162 int len = strlen(sha1_file_directory);
812666c8 163 base = xmalloc(len + 60);
0fcfd160
LT
164 memcpy(base, sha1_file_directory, len);
165 memset(base+len, 0, 60);
166 base[len] = '/';
167 base[len+3] = '/';
168 name = base + len + 1;
169 }
ace1534d 170 fill_sha1_path(name, sha1);
0fcfd160
LT
171 return base;
172}
173
bf592c50
DB
174char *sha1_pack_name(const unsigned char *sha1)
175{
176 static const char hex[] = "0123456789abcdef";
177 static char *name, *base, *buf;
178 int i;
179
180 if (!base) {
181 const char *sha1_file_directory = get_object_directory();
182 int len = strlen(sha1_file_directory);
183 base = xmalloc(len + 60);
184 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
185 name = base + len + 11;
186 }
187
188 buf = name;
189
190 for (i = 0; i < 20; i++) {
191 unsigned int val = *sha1++;
192 *buf++ = hex[val >> 4];
193 *buf++ = hex[val & 0xf];
194 }
195
196 return base;
197}
198
199char *sha1_pack_index_name(const unsigned char *sha1)
200{
201 static const char hex[] = "0123456789abcdef";
202 static char *name, *base, *buf;
203 int i;
204
205 if (!base) {
206 const char *sha1_file_directory = get_object_directory();
207 int len = strlen(sha1_file_directory);
208 base = xmalloc(len + 60);
209 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
210 name = base + len + 11;
211 }
212
213 buf = name;
214
215 for (i = 0; i < 20; i++) {
216 unsigned int val = *sha1++;
217 *buf++ = hex[val >> 4];
218 *buf++ = hex[val & 0xf];
219 }
220
221 return base;
222}
223
d5a63b99
JH
224struct alternate_object_database *alt_odb_list;
225static struct alternate_object_database **alt_odb_tail;
ace1534d 226
c2f493a4
MW
227static void read_info_alternates(const char * alternates, int depth);
228
ddd5d056
JH
229/*
230 * Prepare alternate object database registry.
d5a63b99
JH
231 *
232 * The variable alt_odb_list points at the list of struct
233 * alternate_object_database. The elements on this list come from
234 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
235 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
236 * whose contents is similar to that environment variable but can be
237 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
238 * contains "/the/directory/corresponding/to/.git/objects/...", while
239 * its name points just after the slash at the end of ".git/objects/"
240 * in the example above, and has enough space to hold 40-byte hex
241 * SHA1, an extra slash for the first level indirection, and the
242 * terminating NUL.
ddd5d056 243 */
c2f493a4 244static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
ace1534d 245{
c2f493a4 246 struct stat st;
1494e038 247 const char *objdir = get_object_directory();
c2f493a4
MW
248 struct alternate_object_database *ent;
249 struct alternate_object_database *alt;
250 /* 43 = 40-byte + 2 '/' + terminating NUL */
251 int pfxlen = len;
252 int entlen = pfxlen + 43;
ccfd3e99 253 int base_len = -1;
d5a63b99 254
c2f493a4
MW
255 if (*entry != '/' && relative_base) {
256 /* Relative alt-odb */
257 if (base_len < 0)
258 base_len = strlen(relative_base) + 1;
259 entlen += base_len;
260 pfxlen += base_len;
261 }
262 ent = xmalloc(sizeof(*ent) + entlen);
263
264 if (*entry != '/' && relative_base) {
265 memcpy(ent->base, relative_base, base_len - 1);
266 ent->base[base_len - 1] = '/';
267 memcpy(ent->base + base_len, entry, len);
268 }
269 else
270 memcpy(ent->base, entry, pfxlen);
271
272 ent->name = ent->base + pfxlen + 1;
273 ent->base[pfxlen + 3] = '/';
274 ent->base[pfxlen] = ent->base[entlen-1] = 0;
275
276 /* Detect cases where alternate disappeared */
277 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
278 error("object directory %s does not exist; "
279 "check .git/objects/info/alternates.",
280 ent->base);
281 free(ent);
282 return -1;
283 }
284
285 /* Prevent the common mistake of listing the same
286 * thing twice, or object directory itself.
287 */
288 for (alt = alt_odb_list; alt; alt = alt->next) {
289 if (!memcmp(ent->base, alt->base, pfxlen)) {
290 free(ent);
291 return -1;
292 }
293 }
294 if (!memcmp(ent->base, objdir, pfxlen)) {
295 free(ent);
296 return -1;
297 }
298
299 /* add the alternate entry */
300 *alt_odb_tail = ent;
301 alt_odb_tail = &(ent->next);
302 ent->next = NULL;
303
304 /* recursively add alternates */
305 read_info_alternates(ent->base, depth + 1);
306
307 ent->base[pfxlen] = '/';
308
309 return 0;
310}
311
312static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
313 const char *relative_base, int depth)
314{
315 const char *cp, *last;
316
317 if (depth > 5) {
318 error("%s: ignoring alternate object stores, nesting too deep.",
319 relative_base);
320 return;
321 }
322
d5a63b99 323 last = alt;
9577e7e3
JH
324 while (last < ep) {
325 cp = last;
326 if (cp < ep && *cp == '#') {
327 while (cp < ep && *cp != sep)
328 cp++;
329 last = cp + 1;
330 continue;
331 }
c2f493a4
MW
332 while (cp < ep && *cp != sep)
333 cp++;
d5a63b99 334 if (last != cp) {
c2f493a4
MW
335 if ((*last != '/') && depth) {
336 error("%s: ignoring relative alternate object store %s",
337 relative_base, last);
338 } else {
339 link_alt_odb_entry(last, cp - last,
340 relative_base, depth);
1494e038 341 }
d5a63b99 342 }
9577e7e3 343 while (cp < ep && *cp == sep)
d5a63b99
JH
344 cp++;
345 last = cp;
9577e7e3 346 }
d5a63b99
JH
347}
348
c2f493a4 349static void read_info_alternates(const char * relative_base, int depth)
d5a63b99 350{
9577e7e3 351 char *map;
dc49cd76 352 size_t mapsz;
d5a63b99 353 struct stat st;
c2f493a4
MW
354 char path[PATH_MAX];
355 int fd;
d5a63b99 356
c2f493a4 357 sprintf(path, "%s/info/alternates", relative_base);
d5a63b99
JH
358 fd = open(path, O_RDONLY);
359 if (fd < 0)
360 return;
361 if (fstat(fd, &st) || (st.st_size == 0)) {
362 close(fd);
9a217f2a 363 return;
ace1534d 364 }
dc49cd76
SP
365 mapsz = xsize_t(st.st_size);
366 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 367 close(fd);
d5a63b99 368
dc49cd76 369 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
c2f493a4 370
dc49cd76 371 munmap(map, mapsz);
ace1534d
JH
372}
373
c2f493a4
MW
374void prepare_alt_odb(void)
375{
554fe20d 376 const char *alt;
c2f493a4
MW
377
378 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
379 if (!alt) alt = "";
380
381 if (alt_odb_tail)
382 return;
383 alt_odb_tail = &alt_odb_list;
384 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
385
386 read_info_alternates(get_object_directory(), 0);
387}
388
ace1534d
JH
389static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
390{
ace1534d 391 char *name = sha1_file_name(sha1);
d5a63b99 392 struct alternate_object_database *alt;
ace1534d
JH
393
394 if (!stat(name, st))
395 return name;
9a217f2a 396 prepare_alt_odb();
d5a63b99
JH
397 for (alt = alt_odb_list; alt; alt = alt->next) {
398 name = alt->name;
ace1534d 399 fill_sha1_path(name, sha1);
d5a63b99
JH
400 if (!stat(alt->base, st))
401 return alt->base;
ace1534d
JH
402 }
403 return NULL;
404}
405
60bb8b14 406static unsigned int pack_used_ctr;
a53128b6
SP
407static unsigned int pack_mmap_calls;
408static unsigned int peak_pack_open_windows;
409static unsigned int pack_open_windows;
410static size_t peak_pack_mapped;
60bb8b14 411static size_t pack_mapped;
9a217f2a 412struct packed_git *packed_git;
1f688557 413
a53128b6
SP
414void pack_report()
415{
416 fprintf(stderr,
e05db0fd
PR
417 "pack_report: getpagesize() = %10" SZ_FMT "\n"
418 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
419 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
78a28df9 420 (size_t) getpagesize(),
a53128b6
SP
421 packed_git_window_size,
422 packed_git_limit);
423 fprintf(stderr,
424 "pack_report: pack_used_ctr = %10u\n"
425 "pack_report: pack_mmap_calls = %10u\n"
426 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
427 "pack_report: pack_mapped = "
428 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
429 pack_used_ctr,
430 pack_mmap_calls,
431 pack_open_windows, peak_pack_open_windows,
432 pack_mapped, peak_pack_mapped);
433}
434
42873078 435static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
436{
437 void *idx_map;
42873078 438 struct pack_idx_header *hdr;
dc49cd76 439 size_t idx_size;
42873078 440 uint32_t nr, i, *index;
e93ec6f9 441 int fd = open(path, O_RDONLY);
1f688557 442 struct stat st;
42873078 443
1f688557
JH
444 if (fd < 0)
445 return -1;
446 if (fstat(fd, &st)) {
447 close(fd);
448 return -1;
449 }
dc49cd76 450 idx_size = xsize_t(st.st_size);
2d88451b
SP
451 if (idx_size < 4 * 256 + 20 + 20) {
452 close(fd);
453 return error("index file %s is too small", path);
454 }
c4712e45 455 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 456 close(fd);
1f688557 457
df1b059d
SP
458 /* a future index format would start with this, as older git
459 * binaries would fail the non-monotonic index check below.
460 * give a nicer warning to the user if we can.
461 */
42873078
NP
462 hdr = idx_map;
463 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
2d88451b 464 munmap(idx_map, idx_size);
df1b059d
SP
465 return error("index file %s is a newer version"
466 " and is not supported by this binary"
467 " (try upgrading GIT to a newer version)",
468 path);
2d88451b 469 }
df1b059d 470
1f688557 471 nr = 0;
42873078 472 index = idx_map;
1f688557 473 for (i = 0; i < 256; i++) {
326bf396 474 uint32_t n = ntohl(index[i]);
2d88451b
SP
475 if (n < nr) {
476 munmap(idx_map, idx_size);
df1b059d 477 return error("non-monotonic index %s", path);
2d88451b 478 }
1f688557
JH
479 nr = n;
480 }
481
482 /*
483 * Total size:
484 * - 256 index entries 4 bytes each
485 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
486 * - 20-byte SHA1 of the packfile
487 * - 20-byte SHA1 file checksum
488 */
2d88451b
SP
489 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
490 munmap(idx_map, idx_size);
df1b059d 491 return error("wrong index file size in %s", path);
2d88451b 492 }
1f688557 493
42873078
NP
494 p->index_version = 1;
495 p->index_data = idx_map;
496 p->index_size = idx_size;
1f688557
JH
497 return 0;
498}
499
11daf39b
SP
500static void scan_windows(struct packed_git *p,
501 struct packed_git **lru_p,
502 struct pack_window **lru_w,
503 struct pack_window **lru_l)
1f688557 504{
11daf39b
SP
505 struct pack_window *w, *w_l;
506
507 for (w_l = NULL, w = p->windows; w; w = w->next) {
508 if (!w->inuse_cnt) {
509 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
510 *lru_p = p;
511 *lru_w = w;
512 *lru_l = w_l;
54044bf8 513 }
54044bf8 514 }
11daf39b 515 w_l = w;
f9253394 516 }
11daf39b
SP
517}
518
519static int unuse_one_window(struct packed_git *current)
520{
521 struct packed_git *p, *lru_p = NULL;
522 struct pack_window *lru_w = NULL, *lru_l = NULL;
523
524 if (current)
525 scan_windows(current, &lru_p, &lru_w, &lru_l);
526 for (p = packed_git; p; p = p->next)
527 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
528 if (lru_p) {
529 munmap(lru_w->base, lru_w->len);
530 pack_mapped -= lru_w->len;
531 if (lru_l)
532 lru_l->next = lru_w->next;
533 else {
534 lru_p->windows = lru_w->next;
11daf39b 535 if (!lru_p->windows && lru_p != current) {
54044bf8
SP
536 close(lru_p->pack_fd);
537 lru_p->pack_fd = -1;
538 }
539 }
540 free(lru_w);
a53128b6 541 pack_open_windows--;
54044bf8
SP
542 return 1;
543 }
544 return 0;
f9253394
JH
545}
546
97bfeb34
SP
547void release_pack_memory(size_t need)
548{
549 size_t cur = pack_mapped;
550 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
551 ; /* nothing */
552}
553
03e79c88 554void unuse_pack(struct pack_window **w_cursor)
f9253394 555{
03e79c88
SP
556 struct pack_window *w = *w_cursor;
557 if (w) {
558 w->inuse_cnt--;
559 *w_cursor = NULL;
560 }
1f688557
JH
561}
562
3cf8b462
SP
563/*
564 * Do not call this directly as this leaks p->pack_fd on error return;
565 * call open_packed_git() instead.
566 */
567static int open_packed_git_1(struct packed_git *p)
1f688557 568{
9bc879c1
SP
569 struct stat st;
570 struct pack_header hdr;
571 unsigned char sha1[20];
572 unsigned char *idx_sha1;
2c039da8 573 long fd_flag;
9bc879c1
SP
574
575 p->pack_fd = open(p->pack_name, O_RDONLY);
576 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 577 return -1;
9bc879c1
SP
578
579 /* If we created the struct before we had the pack we lack size. */
bf592c50 580 if (!p->pack_size) {
bf592c50 581 if (!S_ISREG(st.st_mode))
072db278 582 return error("packfile %s not a regular file", p->pack_name);
bf592c50 583 p->pack_size = st.st_size;
9bc879c1 584 } else if (p->pack_size != st.st_size)
072db278 585 return error("packfile %s size changed", p->pack_name);
9bc879c1 586
2c039da8
JH
587 /* We leave these file descriptors open with sliding mmap;
588 * there is no point keeping them open across exec(), though.
589 */
590 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
591 if (fd_flag < 0)
072db278 592 return error("cannot determine file descriptor flags");
2c039da8
JH
593 fd_flag |= FD_CLOEXEC;
594 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 595 return error("cannot set FD_CLOEXEC");
2c039da8 596
9bc879c1 597 /* Verify we recognize this pack file format. */
e6e2bd62 598 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 599 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 600 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 601 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 602 if (!pack_version_ok(hdr.hdr_version))
072db278 603 return error("packfile %s is version %u and not supported"
9bc879c1
SP
604 " (try upgrading GIT to a newer version)",
605 p->pack_name, ntohl(hdr.hdr_version));
606
607 /* Verify the pack matches its index. */
608 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
072db278 609 return error("packfile %s claims to have %u objects"
9bc879c1
SP
610 " while index size indicates %u objects",
611 p->pack_name, ntohl(hdr.hdr_entries),
612 num_packed_objects(p));
613 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 614 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 615 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 616 return error("packfile %s signature is unavailable", p->pack_name);
42873078 617 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 618 if (hashcmp(sha1, idx_sha1))
072db278
SP
619 return error("packfile %s does not match index", p->pack_name);
620 return 0;
9bc879c1
SP
621}
622
3cf8b462
SP
623static int open_packed_git(struct packed_git *p)
624{
625 if (!open_packed_git_1(p))
626 return 0;
627 if (p->pack_fd != -1) {
628 close(p->pack_fd);
629 p->pack_fd = -1;
630 }
631 return -1;
632}
633
c4001d92 634static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
635{
636 /* We must promise at least 20 bytes (one hash) after the
637 * offset is available from this window, otherwise the offset
638 * is not actually in this window and a different window (which
639 * has that one hash excess) must be used. This is to support
640 * the object header and delta base parsing routines below.
641 */
642 off_t win_off = win->offset;
643 return win_off <= offset
644 && (offset + 20) <= (win_off + win->len);
645}
646
03e79c88
SP
647unsigned char* use_pack(struct packed_git *p,
648 struct pack_window **w_cursor,
c4001d92 649 off_t offset,
03e79c88 650 unsigned int *left)
9bc879c1 651{
60bb8b14 652 struct pack_window *win = *w_cursor;
03e79c88 653
072db278
SP
654 if (p->pack_fd == -1 && open_packed_git(p))
655 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
656
657 /* Since packfiles end in a hash of their content and its
658 * pointless to ask for an offset into the middle of that
659 * hash, and the in_window function above wouldn't match
660 * don't allow an offset too close to the end of the file.
661 */
662 if (offset > (p->pack_size - 20))
663 die("offset beyond end of packfile (truncated pack?)");
664
665 if (!win || !in_window(win, offset)) {
666 if (win)
667 win->inuse_cnt--;
668 for (win = p->windows; win; win = win->next) {
669 if (in_window(win, offset))
670 break;
671 }
672 if (!win) {
78a28df9 673 size_t window_align = packed_git_window_size / 2;
dc49cd76 674 off_t len;
60bb8b14 675 win = xcalloc(1, sizeof(*win));
78a28df9 676 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
677 len = p->pack_size - win->offset;
678 if (len > packed_git_window_size)
679 len = packed_git_window_size;
680 win->len = (size_t)len;
60bb8b14 681 pack_mapped += win->len;
11daf39b
SP
682 while (packed_git_limit < pack_mapped
683 && unuse_one_window(p))
60bb8b14 684 ; /* nothing */
c4712e45 685 win->base = xmmap(NULL, win->len,
60bb8b14
SP
686 PROT_READ, MAP_PRIVATE,
687 p->pack_fd, win->offset);
688 if (win->base == MAP_FAILED)
73b4e4be
SP
689 die("packfile %s cannot be mapped: %s",
690 p->pack_name,
691 strerror(errno));
a53128b6
SP
692 pack_mmap_calls++;
693 pack_open_windows++;
694 if (pack_mapped > peak_pack_mapped)
695 peak_pack_mapped = pack_mapped;
696 if (pack_open_windows > peak_pack_open_windows)
697 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
698 win->next = p->windows;
699 p->windows = win;
700 }
1f688557 701 }
03e79c88
SP
702 if (win != *w_cursor) {
703 win->last_used = pack_used_ctr++;
704 win->inuse_cnt++;
705 *w_cursor = win;
706 }
60bb8b14 707 offset -= win->offset;
03e79c88 708 if (left)
dc49cd76 709 *left = win->len - xsize_t(offset);
03e79c88 710 return win->base + offset;
1f688557
JH
711}
712
42873078 713struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557
JH
714{
715 struct stat st;
42873078 716 struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
1f688557 717
42873078
NP
718 /*
719 * Make sure a corresponding .pack file exists and that
720 * the index looks sane.
721 */
722 path_len -= strlen(".idx");
723 if (path_len < 1)
1f688557 724 return NULL;
42873078
NP
725 memcpy(p->pack_name, path, path_len);
726 strcpy(p->pack_name + path_len, ".pack");
727 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode) ||
728 check_packed_git_idx(path, p)) {
729 free(p);
1f688557
JH
730 return NULL;
731 }
42873078 732
1f688557
JH
733 /* ok, it looks sane as far as we can check without
734 * actually mapping the pack file.
735 */
1f688557 736 p->pack_size = st.st_size;
1f688557 737 p->next = NULL;
c41ee586 738 p->windows = NULL;
9bc879c1 739 p->pack_fd = -1;
9d835df2 740 p->pack_local = local;
b867092f 741 p->mtime = st.st_mtime;
42873078
NP
742 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
743 hashclr(p->sha1);
1f688557
JH
744 return p;
745}
746
bf592c50 747struct packed_git *parse_pack_index(unsigned char *sha1)
c508df5e
DB
748{
749 char *path = sha1_pack_index_name(sha1);
750 return parse_pack_index_file(sha1, path);
751}
752
42873078
NP
753struct packed_git *parse_pack_index_file(const unsigned char *sha1,
754 const char *idx_path)
bf592c50 755{
42873078
NP
756 const char *path = sha1_pack_name(sha1);
757 struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
bf592c50 758
42873078
NP
759 if (check_packed_git_idx(idx_path, p)) {
760 free(p);
bf592c50 761 return NULL;
42873078 762 }
bf592c50 763
bf592c50 764 strcpy(p->pack_name, path);
bf592c50 765 p->pack_size = 0;
bf592c50 766 p->next = NULL;
c41ee586 767 p->windows = NULL;
9bc879c1 768 p->pack_fd = -1;
e702496e 769 hashcpy(p->sha1, sha1);
bf592c50
DB
770 return p;
771}
772
773void install_packed_git(struct packed_git *pack)
774{
775 pack->next = packed_git;
776 packed_git = pack;
777}
778
9d835df2 779static void prepare_packed_git_one(char *objdir, int local)
1f688557
JH
780{
781 char path[PATH_MAX];
782 int len;
783 DIR *dir;
784 struct dirent *de;
785
786 sprintf(path, "%s/pack", objdir);
787 len = strlen(path);
788 dir = opendir(path);
b5b16990 789 if (!dir) {
26125f6b 790 if (errno != ENOENT)
bd2afde8 791 error("unable to open object pack directory: %s: %s",
26125f6b 792 path, strerror(errno));
1f688557 793 return;
b5b16990 794 }
1f688557
JH
795 path[len++] = '/';
796 while ((de = readdir(dir)) != NULL) {
797 int namelen = strlen(de->d_name);
798 struct packed_git *p;
799
5bb1cda5 800 if (!has_extension(de->d_name, ".idx"))
1f688557
JH
801 continue;
802
54a15a8d 803 /* Don't reopen a pack we already have. */
1f688557 804 strcpy(path + len, de->d_name);
86f7780c
JK
805 for (p = packed_git; p; p = p->next) {
806 if (!memcmp(path, p->pack_name, len + namelen - 4))
807 break;
808 }
809 if (p)
810 continue;
54a15a8d
SP
811 /* See if it really is a valid .idx file with corresponding
812 * .pack file that we can map.
813 */
9d835df2 814 p = add_packed_git(path, len + namelen, local);
1f688557
JH
815 if (!p)
816 continue;
625e9421 817 install_packed_git(p);
1f688557 818 }
5b35bcd5 819 closedir(dir);
1f688557
JH
820}
821
b867092f
JH
822static int sort_pack(const void *a_, const void *b_)
823{
824 struct packed_git *a = *((struct packed_git **)a_);
825 struct packed_git *b = *((struct packed_git **)b_);
826 int st;
827
828 /*
829 * Local packs tend to contain objects specific to our
830 * variant of the project than remote ones. In addition,
831 * remote ones could be on a network mounted filesystem.
832 * Favor local ones for these reasons.
833 */
834 st = a->pack_local - b->pack_local;
835 if (st)
836 return -st;
837
838 /*
839 * Younger packs tend to contain more recent objects,
840 * and more recent objects tend to get accessed more
841 * often.
842 */
843 if (a->mtime < b->mtime)
844 return 1;
845 else if (a->mtime == b->mtime)
846 return 0;
847 return -1;
848}
849
850static void rearrange_packed_git(void)
851{
852 struct packed_git **ary, *p;
853 int i, n;
854
855 for (n = 0, p = packed_git; p; p = p->next)
856 n++;
857 if (n < 2)
858 return;
859
860 /* prepare an array of packed_git for easier sorting */
861 ary = xcalloc(n, sizeof(struct packed_git *));
862 for (n = 0, p = packed_git; p; p = p->next)
863 ary[n++] = p;
864
865 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
866
867 /* link them back again */
868 for (i = 0; i < n - 1; i++)
869 ary[i]->next = ary[i + 1];
870 ary[n - 1]->next = NULL;
871 packed_git = ary[0];
872
873 free(ary);
874}
875
637cdd9d 876static int prepare_packed_git_run_once = 0;
9a217f2a 877void prepare_packed_git(void)
1f688557 878{
d5a63b99 879 struct alternate_object_database *alt;
1f688557 880
637cdd9d 881 if (prepare_packed_git_run_once)
1f688557 882 return;
9d835df2 883 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 884 prepare_alt_odb();
d5a63b99 885 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 886 alt->name[-1] = 0;
9d835df2 887 prepare_packed_git_one(alt->base, 0);
1494e038 888 alt->name[-1] = '/';
1f688557 889 }
b867092f 890 rearrange_packed_git();
637cdd9d
JK
891 prepare_packed_git_run_once = 1;
892}
893
fc04c412 894void reprepare_packed_git(void)
637cdd9d
JK
895{
896 prepare_packed_git_run_once = 0;
897 prepare_packed_git();
1f688557
JH
898}
899
5d6ccf5c 900int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
0fcfd160
LT
901{
902 unsigned char real_sha1[20];
7cfb5f36 903 hash_sha1_file(map, size, type, real_sha1);
a89fccd2 904 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
905}
906
bb6b8e4f 907void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 908{
0fcfd160
LT
909 struct stat st;
910 void *map;
144bde78 911 int fd;
ace1534d
JH
912 char *filename = find_sha1_file(sha1, &st);
913
914 if (!filename) {
ace1534d
JH
915 return NULL;
916 }
0fcfd160 917
144bde78 918 fd = open(filename, O_RDONLY | sha1_file_open_flag);
0fcfd160 919 if (fd < 0) {
144bde78
LT
920 /* See if it works without O_NOATIME */
921 switch (sha1_file_open_flag) {
922 default:
923 fd = open(filename, O_RDONLY);
924 if (fd >= 0)
925 break;
926 /* Fallthrough */
927 case 0:
144bde78
LT
928 return NULL;
929 }
930
1f688557
JH
931 /* If it failed once, it will probably fail again.
932 * Stop using O_NOATIME
933 */
144bde78 934 sha1_file_open_flag = 0;
0fcfd160 935 }
dc49cd76
SP
936 *size = xsize_t(st.st_size);
937 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
0fcfd160 938 close(fd);
0fcfd160
LT
939 return map;
940}
941
bb6b8e4f
JH
942int legacy_loose_object(unsigned char *map)
943{
944 unsigned int word;
945
946 /*
947 * Is it a zlib-compressed buffer? If so, the first byte
948 * must be 0x78 (15-bit window size, deflated), and the
949 * first 16-bit word is evenly divisible by 31
950 */
951 word = (map[0] << 8) + map[1];
952 if (map[0] == 0x78 && !(word % 31))
953 return 1;
954 else
955 return 0;
956}
957
72518e9c 958unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 959{
ad1ed5ee 960 unsigned shift;
93821bd9 961 unsigned char c;
93821bd9 962 unsigned long size;
ad1ed5ee
JH
963 unsigned long used = 0;
964
965 c = buf[used++];
966 *type = (c >> 4) & 7;
967 size = c & 15;
968 shift = 4;
969 while (c & 0x80) {
970 if (len <= used)
971 return 0;
972 if (sizeof(long) * 8 <= shift)
973 return 0;
974 c = buf[used++];
975 size += (c & 0x7f) << shift;
976 shift += 7;
977 }
978 *sizep = size;
979 return used;
980}
981
982static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
983{
984 unsigned long size, used;
985 static const char valid_loose_object_type[8] = {
986 0, /* OBJ_EXT */
987 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
988 0, /* "delta" and others are invalid in a loose object */
93821bd9 989 };
ad1ed5ee 990 enum object_type type;
93821bd9 991
c4483576
LT
992 /* Get the data stream */
993 memset(stream, 0, sizeof(*stream));
994 stream->next_in = map;
995 stream->avail_in = mapsize;
996 stream->next_out = buffer;
93821bd9
LT
997 stream->avail_out = bufsiz;
998
bb6b8e4f 999 if (legacy_loose_object(map)) {
93821bd9
LT
1000 inflateInit(stream);
1001 return inflate(stream, 0);
1002 }
1003
ad1ed5ee
JH
1004 used = unpack_object_header_gently(map, mapsize, &type, &size);
1005 if (!used || !valid_loose_object_type[type])
93821bd9 1006 return -1;
ad1ed5ee
JH
1007 map += used;
1008 mapsize -= used;
93821bd9
LT
1009
1010 /* Set up the stream for the rest.. */
1011 stream->next_in = map;
1012 stream->avail_in = mapsize;
c4483576 1013 inflateInit(stream);
93821bd9
LT
1014
1015 /* And generate the fake traditional header */
ad1ed5ee 1016 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
df843662 1017 typename(type), size);
93821bd9 1018 return 0;
c4483576
LT
1019}
1020
7efbff75 1021static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1022{
1023 int bytes = strlen(buffer) + 1;
d565b341 1024 unsigned char *buf = xmalloc(1+size);
93821bd9 1025 unsigned long n;
7efbff75 1026 int status = Z_OK;
5180cacc 1027
93821bd9
LT
1028 n = stream->total_out - bytes;
1029 if (n > size)
1030 n = size;
1031 memcpy(buf, (char *) buffer + bytes, n);
1032 bytes = n;
5180cacc
LT
1033 if (bytes < size) {
1034 stream->next_out = buf + bytes;
1035 stream->avail_out = size - bytes;
7efbff75
JH
1036 while (status == Z_OK)
1037 status = inflate(stream, Z_FINISH);
5180cacc
LT
1038 }
1039 buf[size] = 0;
7efbff75
JH
1040 if ((status == Z_OK || status == Z_STREAM_END) && !stream->avail_in) {
1041 inflateEnd(stream);
1042 return buf;
1043 }
1044
1045 if (status < 0)
1046 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1047 else if (stream->avail_in)
1048 error("garbage at end of loose object '%s'",
1049 sha1_to_hex(sha1));
1050 free(buf);
1051 return NULL;
5180cacc
LT
1052}
1053
1054/*
1055 * We used to just use "sscanf()", but that's actually way
1056 * too permissive for what we want to check. So do an anal
1057 * object header parse by hand.
1058 */
21666f1a 1059static int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1060{
21666f1a 1061 char type[10];
5180cacc
LT
1062 int i;
1063 unsigned long size;
1064
1065 /*
1066 * The type can be at most ten bytes (including the
1067 * terminating '\0' that we add), and is followed by
21666f1a 1068 * a space.
5180cacc 1069 */
21666f1a 1070 i = 0;
5180cacc
LT
1071 for (;;) {
1072 char c = *hdr++;
1073 if (c == ' ')
1074 break;
21666f1a
NP
1075 type[i++] = c;
1076 if (i >= sizeof(type))
5180cacc 1077 return -1;
5180cacc 1078 }
21666f1a 1079 type[i] = 0;
5180cacc
LT
1080
1081 /*
1082 * The length must follow immediately, and be in canonical
1083 * decimal format (ie "010" is not valid).
1084 */
1085 size = *hdr++ - '0';
1086 if (size > 9)
1087 return -1;
1088 if (size) {
1089 for (;;) {
1090 unsigned long c = *hdr - '0';
1091 if (c > 9)
1092 break;
1093 hdr++;
1094 size = size * 10 + c;
1095 }
1096 }
1097 *sizep = size;
1098
1099 /*
1100 * The length must be followed by a zero byte
1101 */
21666f1a 1102 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1103}
1104
7efbff75 1105static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1106{
5180cacc 1107 int ret;
0fcfd160 1108 z_stream stream;
5180cacc 1109 char hdr[8192];
0fcfd160 1110
5180cacc 1111 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1112 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1113 return NULL;
1114
7efbff75 1115 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1116}
1117
c4001d92 1118static off_t get_delta_base(struct packed_git *p,
03e79c88 1119 struct pack_window **w_curs,
c4001d92 1120 off_t *curpos,
21666f1a 1121 enum object_type type,
c4001d92 1122 off_t delta_obj_offset)
eb32d236 1123{
2b87c45b 1124 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1125 off_t base_offset;
eb32d236 1126
8d8a4ea5
SP
1127 /* use_pack() assured us we have [base_info, base_info + 20)
1128 * as a range that we can look at without walking off the
1129 * end of the mapped window. Its actually the hash size
1130 * that is assured. An OFS_DELTA longer than the hash size
1131 * is stupid, as then a REF_DELTA would be smaller to store.
1132 */
21666f1a 1133 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1134 unsigned used = 0;
1135 unsigned char c = base_info[used++];
1136 base_offset = c & 127;
1137 while (c & 128) {
1138 base_offset += 1;
1139 if (!base_offset || base_offset & ~(~0UL >> 7))
1140 die("offset value overflow for delta base object");
1141 c = base_info[used++];
1142 base_offset = (base_offset << 7) + (c & 127);
1143 }
1144 base_offset = delta_obj_offset - base_offset;
1145 if (base_offset >= delta_obj_offset)
1146 die("delta base offset out of bound");
2b87c45b 1147 *curpos += used;
21666f1a 1148 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1149 /* The base entry _must_ be in the same pack */
1150 base_offset = find_pack_entry_one(base_info, p);
1151 if (!base_offset)
1152 die("failed to find delta-pack base object %s",
1153 sha1_to_hex(base_info));
2b87c45b 1154 *curpos += 20;
eb32d236
NP
1155 } else
1156 die("I am totally screwed");
2b87c45b 1157 return base_offset;
eb32d236
NP
1158}
1159
f3bf9224 1160/* forward declaration for a mutually recursive function */
c4001d92 1161static int packed_object_info(struct packed_git *p, off_t offset,
21666f1a 1162 unsigned long *sizep);
f3bf9224 1163
43057304 1164static int packed_delta_info(struct packed_git *p,
03e79c88 1165 struct pack_window **w_curs,
c4001d92 1166 off_t curpos,
21666f1a 1167 enum object_type type,
c4001d92 1168 off_t obj_offset,
43057304 1169 unsigned long *sizep)
5db47c2b 1170{
c4001d92 1171 off_t base_offset;
f3bf9224 1172
21666f1a
NP
1173 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1174 type = packed_object_info(p, base_offset, NULL);
f3bf9224 1175
c62266f3
JH
1176 /* We choose to only get the type of the base object and
1177 * ignore potentially corrupt pack file that expects the delta
1178 * based on a base with a wrong size. This saves tons of
1179 * inflate() calls.
1180 */
c62266f3
JH
1181 if (sizep) {
1182 const unsigned char *data;
079afb18 1183 unsigned char delta_head[20], *in;
c62266f3
JH
1184 z_stream stream;
1185 int st;
1186
1187 memset(&stream, 0, sizeof(stream));
c62266f3
JH
1188 stream.next_out = delta_head;
1189 stream.avail_out = sizeof(delta_head);
1190
1191 inflateInit(&stream);
079afb18 1192 do {
2b87c45b 1193 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18
SP
1194 stream.next_in = in;
1195 st = inflate(&stream, Z_FINISH);
2b87c45b 1196 curpos += stream.next_in - in;
079afb18
SP
1197 } while ((st == Z_OK || st == Z_BUF_ERROR)
1198 && stream.total_out < sizeof(delta_head));
c62266f3
JH
1199 inflateEnd(&stream);
1200 if ((st != Z_STREAM_END) &&
1201 stream.total_out != sizeof(delta_head))
1202 die("delta data unpack-initial failed");
1203
1204 /* Examine the initial part of the delta to figure out
1205 * the result size.
1206 */
1207 data = delta_head;
8960844a
NP
1208
1209 /* ignore base size */
1210 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
5db47c2b 1211
c62266f3 1212 /* Read the result size */
2b87c45b 1213 *sizep = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
c62266f3 1214 }
21666f1a
NP
1215
1216 return type;
5db47c2b
JH
1217}
1218
2b87c45b
NP
1219static int unpack_object_header(struct packed_git *p,
1220 struct pack_window **w_curs,
c4001d92 1221 off_t *curpos,
2b87c45b 1222 unsigned long *sizep)
a733cb60 1223{
03e79c88
SP
1224 unsigned char *base;
1225 unsigned int left;
ad1ed5ee 1226 unsigned long used;
2b87c45b 1227 enum object_type type;
a733cb60 1228
8d8a4ea5
SP
1229 /* use_pack() assures us we have [base, base + 20) available
1230 * as a range that we can look at at. (Its actually the hash
3dff5379 1231 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1232 * the maximum deflated object size is 2^137, which is just
1233 * insane, so we know won't exceed what we have been given.
1234 */
2b87c45b
NP
1235 base = use_pack(p, w_curs, *curpos, &left);
1236 used = unpack_object_header_gently(base, left, &type, sizep);
ad1ed5ee
JH
1237 if (!used)
1238 die("object offset outside of pack file");
2b87c45b 1239 *curpos += used;
ad1ed5ee 1240
2b87c45b 1241 return type;
a733cb60
LT
1242}
1243
21666f1a 1244const char *packed_object_info_detail(struct packed_git *p,
c4001d92 1245 off_t obj_offset,
21666f1a
NP
1246 unsigned long *size,
1247 unsigned long *store_size,
1248 unsigned int *delta_chain_length,
1249 unsigned char *base_sha1)
ad8c80a5 1250{
03e79c88 1251 struct pack_window *w_curs = NULL;
c4001d92
SP
1252 off_t curpos;
1253 unsigned long dummy;
43057304 1254 unsigned char *next_sha1;
21666f1a 1255 enum object_type type;
ad8c80a5 1256
43057304 1257 *delta_chain_length = 0;
2b87c45b 1258 curpos = obj_offset;
21666f1a 1259 type = unpack_object_header(p, &w_curs, &curpos, size);
43057304
NP
1260
1261 for (;;) {
21666f1a 1262 switch (type) {
43057304 1263 default:
08a19d87 1264 die("pack %s contains unknown object type %d",
21666f1a 1265 p->pack_name, type);
43057304
NP
1266 case OBJ_COMMIT:
1267 case OBJ_TREE:
1268 case OBJ_BLOB:
1269 case OBJ_TAG:
43057304 1270 *store_size = 0; /* notyet */
03e79c88 1271 unuse_pack(&w_curs);
21666f1a 1272 return typename(type);
eb32d236 1273 case OBJ_OFS_DELTA:
21666f1a 1274 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
eb32d236 1275 if (*delta_chain_length == 0) {
2b87c45b 1276 /* TODO: find base_sha1 as pointed by curpos */
30fee062 1277 hashclr(base_sha1);
eb32d236
NP
1278 }
1279 break;
1280 case OBJ_REF_DELTA:
2b87c45b 1281 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
43057304
NP
1282 if (*delta_chain_length == 0)
1283 hashcpy(base_sha1, next_sha1);
2b87c45b 1284 obj_offset = find_pack_entry_one(next_sha1, p);
43057304
NP
1285 break;
1286 }
43057304 1287 (*delta_chain_length)++;
2b87c45b 1288 curpos = obj_offset;
21666f1a 1289 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
ad8c80a5 1290 }
ad8c80a5
JH
1291}
1292
c4001d92 1293static int packed_object_info(struct packed_git *p, off_t obj_offset,
21666f1a 1294 unsigned long *sizep)
1f688557 1295{
03e79c88 1296 struct pack_window *w_curs = NULL;
c4001d92
SP
1297 unsigned long size;
1298 off_t curpos = obj_offset;
21666f1a 1299 enum object_type type;
5db47c2b 1300
21666f1a 1301 type = unpack_object_header(p, &w_curs, &curpos, &size);
5db47c2b 1302
21666f1a 1303 switch (type) {
eb32d236
NP
1304 case OBJ_OFS_DELTA:
1305 case OBJ_REF_DELTA:
21666f1a
NP
1306 type = packed_delta_info(p, &w_curs, curpos,
1307 type, obj_offset, sizep);
2b87c45b 1308 break;
a733cb60 1309 case OBJ_COMMIT:
a733cb60 1310 case OBJ_TREE:
a733cb60 1311 case OBJ_BLOB:
a733cb60 1312 case OBJ_TAG:
2b87c45b
NP
1313 if (sizep)
1314 *sizep = size;
a69d0943 1315 break;
1f688557 1316 default:
08a19d87 1317 die("pack %s contains unknown object type %d",
21666f1a 1318 p->pack_name, type);
1f688557 1319 }
2b87c45b 1320 unuse_pack(&w_curs);
21666f1a 1321 return type;
1f688557
JH
1322}
1323
eb950c19 1324static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1325 struct pack_window **w_curs,
c4001d92 1326 off_t curpos,
eb950c19 1327 unsigned long size)
de530aaa
SP
1328{
1329 int st;
1330 z_stream stream;
079afb18 1331 unsigned char *buffer, *in;
de530aaa
SP
1332
1333 buffer = xmalloc(size + 1);
1334 buffer[size] = 0;
1335 memset(&stream, 0, sizeof(stream));
de530aaa
SP
1336 stream.next_out = buffer;
1337 stream.avail_out = size;
1338
1339 inflateInit(&stream);
079afb18 1340 do {
2b87c45b 1341 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18
SP
1342 stream.next_in = in;
1343 st = inflate(&stream, Z_FINISH);
2b87c45b 1344 curpos += stream.next_in - in;
079afb18 1345 } while (st == Z_OK || st == Z_BUF_ERROR);
de530aaa
SP
1346 inflateEnd(&stream);
1347 if ((st != Z_STREAM_END) || stream.total_out != size) {
1348 free(buffer);
1349 return NULL;
1350 }
1351
1352 return buffer;
1353}
1354
e5e01619
LT
1355#define MAX_DELTA_CACHE (256)
1356
18bdec11 1357static size_t delta_base_cached;
e5e01619
LT
1358static struct delta_base_cache_entry {
1359 struct packed_git *p;
1360 off_t base_offset;
1361 unsigned long size;
1362 void *data;
1363 enum object_type type;
1364} delta_base_cache[MAX_DELTA_CACHE];
1365
1366static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1367{
1368 unsigned long hash;
1369
1370 hash = (unsigned long)p + (unsigned long)base_offset;
1371 hash += (hash >> 8) + (hash >> 16);
3358004a 1372 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1373}
1374
62f255ad 1375static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 1376 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 1377{
e5e01619
LT
1378 void *ret;
1379 unsigned long hash = pack_entry_hash(p, base_offset);
1380 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1381
1382 ret = ent->data;
1383 if (ret && ent->p == p && ent->base_offset == base_offset)
1384 goto found_cache_entry;
62f255ad 1385 return unpack_entry(p, base_offset, type, base_size);
e5e01619
LT
1386
1387found_cache_entry:
18bdec11 1388 if (!keep_cache) {
a0cba108 1389 ent->data = NULL;
18bdec11
SP
1390 delta_base_cached -= ent->size;
1391 }
a0cba108
NP
1392 else {
1393 ret = xmalloc(ent->size + 1);
1394 memcpy(ret, ent->data, ent->size);
1395 ((char *)ret)[ent->size] = 0;
1396 }
e5e01619
LT
1397 *type = ent->type;
1398 *base_size = ent->size;
1399 return ret;
62f255ad
LT
1400}
1401
18bdec11
SP
1402static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1403{
1404 if (ent->data) {
1405 free(ent->data);
1406 ent->data = NULL;
1407 delta_base_cached -= ent->size;
1408 }
1409}
1410
62f255ad
LT
1411static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1412 void *base, unsigned long base_size, enum object_type type)
1413{
18bdec11 1414 unsigned long i, hash = pack_entry_hash(p, base_offset);
e5e01619
LT
1415 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1416
18bdec11
SP
1417 release_delta_base_cache(ent);
1418 delta_base_cached += base_size;
1419 for (i = 0; delta_base_cached > delta_base_cache_limit
3358004a 1420 && i < MAX_DELTA_CACHE ; i++) {
18bdec11
SP
1421 struct delta_base_cache_entry *f = delta_base_cache + i;
1422 if (f->type == OBJ_BLOB)
1423 release_delta_base_cache(f);
1424 }
1425 for (i = 0; delta_base_cached > delta_base_cache_limit
3358004a 1426 && i < MAX_DELTA_CACHE ; i++)
18bdec11
SP
1427 release_delta_base_cache(delta_base_cache + i);
1428
e5e01619
LT
1429 ent->p = p;
1430 ent->base_offset = base_offset;
1431 ent->type = type;
1432 ent->data = base;
1433 ent->size = base_size;
62f255ad
LT
1434}
1435
eb950c19 1436static void *unpack_delta_entry(struct packed_git *p,
03e79c88 1437 struct pack_window **w_curs,
c4001d92 1438 off_t curpos,
1f688557 1439 unsigned long delta_size,
c4001d92 1440 off_t obj_offset,
21666f1a 1441 enum object_type *type,
eb950c19 1442 unsigned long *sizep)
1f688557 1443{
7c3e8be3 1444 void *delta_data, *result, *base;
c4001d92
SP
1445 unsigned long base_size;
1446 off_t base_offset;
43057304 1447
21666f1a 1448 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
a0cba108 1449 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
67686d95 1450 if (!base)
c4001d92
SP
1451 die("failed to read delta base object"
1452 " at %"PRIuMAX" from %s",
1453 (uintmax_t)base_offset, p->pack_name);
67686d95 1454
2b87c45b 1455 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1f688557
JH
1456 result = patch_delta(base, base_size,
1457 delta_data, delta_size,
21666f1a 1458 sizep);
1f688557
JH
1459 if (!result)
1460 die("failed to apply delta");
1461 free(delta_data);
62f255ad 1462 add_delta_base_cache(p, base_offset, base, base_size, *type);
1f688557
JH
1463 return result;
1464}
1465
c4001d92 1466void *unpack_entry(struct packed_git *p, off_t obj_offset,
21666f1a 1467 enum object_type *type, unsigned long *sizep)
1f688557 1468{
03e79c88 1469 struct pack_window *w_curs = NULL;
c4001d92 1470 off_t curpos = obj_offset;
21666f1a 1471 void *data;
1f688557 1472
21666f1a
NP
1473 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1474 switch (*type) {
eb32d236
NP
1475 case OBJ_OFS_DELTA:
1476 case OBJ_REF_DELTA:
21666f1a
NP
1477 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1478 obj_offset, type, sizep);
4d703a1a 1479 break;
a733cb60 1480 case OBJ_COMMIT:
a733cb60 1481 case OBJ_TREE:
a733cb60 1482 case OBJ_BLOB:
a733cb60 1483 case OBJ_TAG:
21666f1a 1484 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
4d703a1a 1485 break;
1f688557 1486 default:
21666f1a 1487 die("unknown object type %i in %s", *type, p->pack_name);
1f688557 1488 }
03e79c88 1489 unuse_pack(&w_curs);
21666f1a 1490 return data;
1f688557
JH
1491}
1492
326bf396 1493uint32_t num_packed_objects(const struct packed_git *p)
9a217f2a 1494{
f9253394 1495 /* See check_packed_git_idx() */
dc49cd76 1496 return (uint32_t)((p->index_size - 20 - 20 - 4*256) / 24);
9a217f2a
JH
1497}
1498
326bf396 1499int nth_packed_object_sha1(const struct packed_git *p, uint32_t n,
9a217f2a
JH
1500 unsigned char* sha1)
1501{
42873078
NP
1502 const unsigned char *index = p->index_data;
1503 index += 4 * 256;
326bf396 1504 if (num_packed_objects(p) <= n)
9a217f2a 1505 return -1;
42873078 1506 hashcpy(sha1, index + 24 * n + 4);
9a217f2a
JH
1507 return 0;
1508}
1509
c4001d92 1510off_t find_pack_entry_one(const unsigned char *sha1,
43057304 1511 struct packed_git *p)
1f688557 1512{
42873078 1513 const uint32_t *level1_ofs = p->index_data;
1f688557
JH
1514 int hi = ntohl(level1_ofs[*sha1]);
1515 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
42873078
NP
1516 const unsigned char *index = p->index_data;
1517
1518 index += 4 * 256;
1f688557
JH
1519
1520 do {
1521 int mi = (lo + hi) / 2;
42873078 1522 int cmp = hashcmp(index + 24 * mi + 4, sha1);
43057304 1523 if (!cmp)
b18b00a6 1524 return ntohl(*((uint32_t *)((char *)index + (24 * mi))));
1f688557
JH
1525 if (cmp > 0)
1526 hi = mi;
1527 else
1528 lo = mi+1;
1529 } while (lo < hi);
1530 return 0;
1531}
1532
d4ff6d92
SP
1533static int matches_pack_name(struct packed_git *p, const char *ig)
1534{
1535 const char *last_c, *c;
1536
1537 if (!strcmp(p->pack_name, ig))
1538 return 0;
1539
1540 for (c = p->pack_name, last_c = c; *c;)
1541 if (*c == '/')
1542 last_c = ++c;
1543 else
1544 ++c;
1545 if (!strcmp(last_c, ig))
1546 return 0;
1547
1548 return 1;
1549}
1550
106d710b 1551static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1f688557
JH
1552{
1553 struct packed_git *p;
c4001d92 1554 off_t offset;
43057304 1555
1f688557
JH
1556 prepare_packed_git();
1557
1558 for (p = packed_git; p; p = p->next) {
106d710b
JH
1559 if (ignore_packed) {
1560 const char **ig;
1561 for (ig = ignore_packed; *ig; ig++)
d4ff6d92 1562 if (!matches_pack_name(p, *ig))
106d710b
JH
1563 break;
1564 if (*ig)
1565 continue;
1566 }
43057304
NP
1567 offset = find_pack_entry_one(sha1, p);
1568 if (offset) {
c715f783
SP
1569 /*
1570 * We are about to tell the caller where they can
1571 * locate the requested object. We better make
1572 * sure the packfile is still here and can be
1573 * accessed before supplying that answer, as
1574 * it may have been deleted since the index
1575 * was loaded!
1576 */
1577 if (p->pack_fd == -1 && open_packed_git(p)) {
1578 error("packfile %s cannot be accessed", p->pack_name);
1579 continue;
1580 }
43057304
NP
1581 e->offset = offset;
1582 e->p = p;
1583 hashcpy(e->sha1, sha1);
1f688557 1584 return 1;
43057304 1585 }
1f688557
JH
1586 }
1587 return 0;
1588}
1589
bf592c50
DB
1590struct packed_git *find_sha1_pack(const unsigned char *sha1,
1591 struct packed_git *packs)
1592{
1593 struct packed_git *p;
bf592c50
DB
1594
1595 for (p = packs; p; p = p->next) {
43057304 1596 if (find_pack_entry_one(sha1, p))
bf592c50
DB
1597 return p;
1598 }
1599 return NULL;
21666f1a 1600
bf592c50
DB
1601}
1602
21666f1a 1603static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
65c2e0c3 1604{
36e4d74a 1605 int status;
65c2e0c3
JH
1606 unsigned long mapsize, size;
1607 void *map;
1608 z_stream stream;
d65a16f6 1609 char hdr[32];
65c2e0c3 1610
bb6b8e4f 1611 map = map_sha1_file(sha1, &mapsize);
f0df4ed5
JS
1612 if (!map)
1613 return error("unable to find %s", sha1_to_hex(sha1));
36e4d74a
JH
1614 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1615 status = error("unable to unpack %s header",
1616 sha1_to_hex(sha1));
21666f1a 1617 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 1618 status = error("unable to parse %s header", sha1_to_hex(sha1));
21666f1a
NP
1619 else if (sizep)
1620 *sizep = size;
65c2e0c3
JH
1621 inflateEnd(&stream);
1622 munmap(map, mapsize);
1623 return status;
1624}
1625
21666f1a 1626int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
f0df4ed5 1627{
f0df4ed5
JS
1628 struct pack_entry e;
1629
1630 if (!find_pack_entry(sha1, &e, NULL)) {
1631 reprepare_packed_git();
1632 if (!find_pack_entry(sha1, &e, NULL))
21666f1a 1633 return sha1_loose_object_info(sha1, sizep);
f0df4ed5 1634 }
21666f1a 1635 return packed_object_info(e.p, e.offset, sizep);
f0df4ed5
JS
1636}
1637
21666f1a
NP
1638static void *read_packed_sha1(const unsigned char *sha1,
1639 enum object_type *type, unsigned long *size)
1f688557
JH
1640{
1641 struct pack_entry e;
1642
8276c007 1643 if (!find_pack_entry(sha1, &e, NULL))
1f688557 1644 return NULL;
8276c007 1645 else
a0cba108 1646 return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1f688557
JH
1647}
1648
d66b37bb
JH
1649/*
1650 * This is meant to hold a *small* number of objects that you would
1651 * want read_sha1_file() to be able to return, but yet you do not want
1652 * to write them into the object store (e.g. a browse-only
1653 * application).
1654 */
1655static struct cached_object {
1656 unsigned char sha1[20];
21666f1a 1657 enum object_type type;
d66b37bb
JH
1658 void *buf;
1659 unsigned long size;
1660} *cached_objects;
1661static int cached_object_nr, cached_object_alloc;
1662
1663static struct cached_object *find_cached_object(const unsigned char *sha1)
1664{
1665 int i;
1666 struct cached_object *co = cached_objects;
1667
1668 for (i = 0; i < cached_object_nr; i++, co++) {
1669 if (!hashcmp(co->sha1, sha1))
1670 return co;
1671 }
1672 return NULL;
1673}
1674
21666f1a
NP
1675int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1676 unsigned char *sha1)
d66b37bb
JH
1677{
1678 struct cached_object *co;
1679
21666f1a 1680 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
1681 if (has_sha1_file(sha1) || find_cached_object(sha1))
1682 return 0;
1683 if (cached_object_alloc <= cached_object_nr) {
1684 cached_object_alloc = alloc_nr(cached_object_alloc);
1685 cached_objects = xrealloc(cached_objects,
1686 sizeof(*cached_objects) *
1687 cached_object_alloc);
1688 }
1689 co = &cached_objects[cached_object_nr++];
1690 co->size = len;
21666f1a 1691 co->type = type;
efa13f7b
JH
1692 co->buf = xmalloc(len);
1693 memcpy(co->buf, buf, len);
d66b37bb
JH
1694 hashcpy(co->sha1, sha1);
1695 return 0;
1696}
1697
21666f1a
NP
1698void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1699 unsigned long *size)
0fcfd160
LT
1700{
1701 unsigned long mapsize;
1702 void *map, *buf;
d66b37bb
JH
1703 struct cached_object *co;
1704
1705 co = find_cached_object(sha1);
1706 if (co) {
1707 buf = xmalloc(co->size + 1);
1708 memcpy(buf, co->buf, co->size);
1709 ((char*)buf)[co->size] = 0;
21666f1a 1710 *type = co->type;
d66b37bb
JH
1711 *size = co->size;
1712 return buf;
1713 }
0fcfd160 1714
8276c007
PE
1715 buf = read_packed_sha1(sha1, type, size);
1716 if (buf)
1717 return buf;
bb6b8e4f 1718 map = map_sha1_file(sha1, &mapsize);
0fcfd160 1719 if (map) {
7efbff75 1720 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
1721 munmap(map, mapsize);
1722 return buf;
1723 }
637cdd9d 1724 reprepare_packed_git();
8276c007 1725 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
1726}
1727
40469ee9 1728void *read_object_with_reference(const unsigned char *sha1,
21666f1a 1729 const char *required_type_name,
40469ee9
JH
1730 unsigned long *size,
1731 unsigned char *actual_sha1_return)
f4913f91 1732{
21666f1a 1733 enum object_type type, required_type;
f4913f91
JH
1734 void *buffer;
1735 unsigned long isize;
40469ee9 1736 unsigned char actual_sha1[20];
f4913f91 1737
21666f1a 1738 required_type = type_from_string(required_type_name);
e702496e 1739 hashcpy(actual_sha1, sha1);
40469ee9
JH
1740 while (1) {
1741 int ref_length = -1;
1742 const char *ref_type = NULL;
f4913f91 1743
21666f1a 1744 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
1745 if (!buffer)
1746 return NULL;
21666f1a 1747 if (type == required_type) {
40469ee9
JH
1748 *size = isize;
1749 if (actual_sha1_return)
e702496e 1750 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
1751 return buffer;
1752 }
1753 /* Handle references */
21666f1a 1754 else if (type == OBJ_COMMIT)
40469ee9 1755 ref_type = "tree ";
21666f1a 1756 else if (type == OBJ_TAG)
40469ee9
JH
1757 ref_type = "object ";
1758 else {
1759 free(buffer);
1760 return NULL;
1761 }
1762 ref_length = strlen(ref_type);
f4913f91 1763
40469ee9 1764 if (memcmp(buffer, ref_type, ref_length) ||
1d7f171c 1765 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
1766 free(buffer);
1767 return NULL;
1768 }
1cf58e72 1769 free(buffer);
40469ee9
JH
1770 /* Now we have the ID of the referred-to object in
1771 * actual_sha1. Check again. */
f4913f91 1772 }
f4913f91
JH
1773}
1774
972a9155
RS
1775static void write_sha1_file_prepare(void *buf, unsigned long len,
1776 const char *type, unsigned char *sha1,
d65a16f6 1777 char *hdr, int *hdrlen)
d410c0f5
JH
1778{
1779 SHA_CTX c;
1780
1781 /* Generate the header */
d65a16f6 1782 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
1783
1784 /* Sha1.. */
1785 SHA1_Init(&c);
1786 SHA1_Update(&c, hdr, *hdrlen);
1787 SHA1_Update(&c, buf, len);
1788 SHA1_Final(sha1, &c);
d410c0f5
JH
1789}
1790
230f1322
LT
1791/*
1792 * Link the tempfile to the final place, possibly creating the
1793 * last directory level as you do so.
1794 *
1795 * Returns the errno on failure, 0 on success.
1796 */
839837b9 1797static int link_temp_to_file(const char *tmpfile, const char *filename)
230f1322
LT
1798{
1799 int ret;
756aaf4a 1800 char *dir;
230f1322
LT
1801
1802 if (!link(tmpfile, filename))
1803 return 0;
1804
1805 /*
756aaf4a 1806 * Try to mkdir the last path component if that failed.
230f1322
LT
1807 *
1808 * Re-try the "link()" regardless of whether the mkdir
1809 * succeeds, since a race might mean that somebody
1810 * else succeeded.
1811 */
1812 ret = errno;
756aaf4a
SP
1813 dir = strrchr(filename, '/');
1814 if (dir) {
1815 *dir = 0;
866cae0d 1816 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
91c23e48 1817 *dir = '/';
756aaf4a 1818 return -2;
91c23e48 1819 }
756aaf4a
SP
1820 *dir = '/';
1821 if (!link(tmpfile, filename))
1822 return 0;
1823 ret = errno;
230f1322
LT
1824 }
1825 return ret;
1826}
1827
1828/*
1829 * Move the just written object into its final resting place
1830 */
839837b9 1831int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322
LT
1832{
1833 int ret = link_temp_to_file(tmpfile, filename);
7ebb6fca
LT
1834
1835 /*
1836 * Coda hack - coda doesn't like cross-directory links,
1837 * so we fall back to a rename, which will mean that it
1838 * won't be able to check collisions, but that's not a
1839 * big deal.
1840 *
1841 * The same holds for FAT formatted media.
1842 *
1843 * When this succeeds, we just return 0. We have nothing
1844 * left to unlink.
1845 */
1846 if (ret && ret != EEXIST) {
1847 if (!rename(tmpfile, filename))
230f1322 1848 return 0;
9e48b389 1849 ret = errno;
230f1322
LT
1850 }
1851 unlink(tmpfile);
1852 if (ret) {
1853 if (ret != EEXIST) {
916d081b 1854 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
230f1322
LT
1855 }
1856 /* FIXME!!! Collision check here ? */
1857 }
1858
1859 return 0;
1860}
1861
4d548150
LT
1862static int write_buffer(int fd, const void *buf, size_t len)
1863{
d34cf19b 1864 if (write_in_full(fd, buf, len) < 0)
93822c22 1865 return error("file write error (%s)", strerror(errno));
4d548150
LT
1866 return 0;
1867}
1868
93821bd9
LT
1869static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1870{
1871 int hdr_len;
1872 unsigned char c;
1873
1874 c = (type << 4) | (len & 15);
1875 len >>= 4;
1876 hdr_len = 1;
1877 while (len) {
1878 *hdr++ = c | 0x80;
1879 hdr_len++;
1880 c = (len & 0x7f);
1881 len >>= 7;
1882 }
1883 *hdr = c;
1884 return hdr_len;
1885}
1886
1887static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1888{
d65a16f6 1889 int obj_type, hdrlen;
93821bd9
LT
1890
1891 if (use_legacy_headers) {
1892 while (deflate(stream, 0) == Z_OK)
1893 /* nothing */;
1894 return;
1895 }
df843662 1896 obj_type = type_from_string(type);
d65a16f6
NP
1897 hdrlen = write_binary_header(stream->next_out, obj_type, len);
1898 stream->total_out = hdrlen;
1899 stream->next_out += hdrlen;
1900 stream->avail_out -= hdrlen;
93821bd9
LT
1901}
1902
abdc3fc8
RS
1903int hash_sha1_file(void *buf, unsigned long len, const char *type,
1904 unsigned char *sha1)
1905{
d65a16f6 1906 char hdr[32];
abdc3fc8
RS
1907 int hdrlen;
1908 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1909 return 0;
1910}
1911
bf0f910d 1912int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
0fcfd160
LT
1913{
1914 int size;
bf0f910d 1915 unsigned char *compressed;
0fcfd160
LT
1916 z_stream stream;
1917 unsigned char sha1[20];
706bc531 1918 char *filename;
aac17941 1919 static char tmpfile[PATH_MAX];
d65a16f6 1920 char hdr[32];
230f1322 1921 int fd, hdrlen;
a44c9a5e 1922
d410c0f5
JH
1923 /* Normally if we have it in the pack then we do not bother writing
1924 * it out into .git/objects/??/?{38} file.
1925 */
972a9155
RS
1926 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1927 filename = sha1_file_name(sha1);
706bc531 1928 if (returnsha1)
e702496e 1929 hashcpy(returnsha1, sha1);
d410c0f5
JH
1930 if (has_sha1_file(sha1))
1931 return 0;
aac17941
LT
1932 fd = open(filename, O_RDONLY);
1933 if (fd >= 0) {
706bc531 1934 /*
aac17941
LT
1935 * FIXME!!! We might do collision checking here, but we'd
1936 * need to uncompress the old file and check it. Later.
706bc531 1937 */
aac17941 1938 close(fd);
706bc531
LT
1939 return 0;
1940 }
1941
aac17941 1942 if (errno != ENOENT) {
916d081b 1943 return error("sha1 file %s: %s\n", filename, strerror(errno));
aac17941
LT
1944 }
1945
1946 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
ace1534d 1947
aac17941
LT
1948 fd = mkstemp(tmpfile);
1949 if (fd < 0) {
916d081b
PB
1950 if (errno == EPERM)
1951 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1952 else
1953 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
aac17941
LT
1954 }
1955
0fcfd160
LT
1956 /* Set it up */
1957 memset(&stream, 0, sizeof(stream));
12f6c308 1958 deflateInit(&stream, zlib_compression_level);
93821bd9 1959 size = 8 + deflateBound(&stream, len+hdrlen);
812666c8 1960 compressed = xmalloc(size);
0fcfd160
LT
1961
1962 /* Compress it */
0fcfd160
LT
1963 stream.next_out = compressed;
1964 stream.avail_out = size;
a44c9a5e
LT
1965
1966 /* First header.. */
d65a16f6 1967 stream.next_in = (unsigned char *)hdr;
a44c9a5e 1968 stream.avail_in = hdrlen;
93821bd9 1969 setup_object_header(&stream, type, len);
a44c9a5e
LT
1970
1971 /* Then the data itself.. */
1972 stream.next_in = buf;
1973 stream.avail_in = len;
0fcfd160
LT
1974 while (deflate(&stream, Z_FINISH) == Z_OK)
1975 /* nothing */;
1976 deflateEnd(&stream);
1977 size = stream.total_out;
1978
4d548150
LT
1979 if (write_buffer(fd, compressed, size) < 0)
1980 die("unable to write sha1 file");
aac17941 1981 fchmod(fd, 0444);
706bc531 1982 close(fd);
383f85b7 1983 free(compressed);
0fcfd160 1984
230f1322 1985 return move_temp_to_file(tmpfile, filename);
0fcfd160 1986}
8237b185 1987
4d548150
LT
1988/*
1989 * We need to unpack and recompress the object for writing
1990 * it out to a different file.
1991 */
1992static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
a5eda52b 1993{
4d548150 1994 size_t size;
a5eda52b 1995 z_stream stream;
4d548150
LT
1996 unsigned char *unpacked;
1997 unsigned long len;
21666f1a 1998 enum object_type type;
d65a16f6 1999 char hdr[32];
4d548150
LT
2000 int hdrlen;
2001 void *buf;
bfc66daf 2002
a9486b02 2003 /* need to unpack and recompress it by itself */
21666f1a 2004 unpacked = read_packed_sha1(sha1, &type, &len);
8276c007
PE
2005 if (!unpacked)
2006 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
a5eda52b 2007
21666f1a 2008 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
a5eda52b 2009
4d548150
LT
2010 /* Set it up */
2011 memset(&stream, 0, sizeof(stream));
12f6c308 2012 deflateInit(&stream, zlib_compression_level);
4d548150
LT
2013 size = deflateBound(&stream, len + hdrlen);
2014 buf = xmalloc(size);
a5eda52b 2015
4d548150
LT
2016 /* Compress it */
2017 stream.next_out = buf;
2018 stream.avail_out = size;
a5eda52b 2019
4d548150
LT
2020 /* First header.. */
2021 stream.next_in = (void *)hdr;
2022 stream.avail_in = hdrlen;
2023 while (deflate(&stream, 0) == Z_OK)
2024 /* nothing */;
bfc66daf 2025
4d548150
LT
2026 /* Then the data itself.. */
2027 stream.next_in = unpacked;
2028 stream.avail_in = len;
2029 while (deflate(&stream, Z_FINISH) == Z_OK)
2030 /* nothing */;
2031 deflateEnd(&stream);
2032 free(unpacked);
bfc66daf 2033
4d548150
LT
2034 *objsize = stream.total_out;
2035 return buf;
2036}
2037
2038int write_sha1_to_fd(int fd, const unsigned char *sha1)
2039{
2040 int retval;
2041 unsigned long objsize;
bb6b8e4f 2042 void *buf = map_sha1_file(sha1, &objsize);
4d548150
LT
2043
2044 if (buf) {
2045 retval = write_buffer(fd, buf, objsize);
2046 munmap(buf, objsize);
2047 return retval;
2048 }
2049
2050 buf = repack_object(sha1, &objsize);
2051 retval = write_buffer(fd, buf, objsize);
2052 free(buf);
2053 return retval;
a5eda52b
DB
2054}
2055
70b9829e
DB
2056int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2057 size_t bufsize, size_t *bufposn)
8237b185 2058{
230f1322 2059 char tmpfile[PATH_MAX];
8237b185
DB
2060 int local;
2061 z_stream stream;
2062 unsigned char real_sha1[20];
bf0f910d 2063 unsigned char discard[4096];
8237b185
DB
2064 int ret;
2065 SHA_CTX c;
2066
230f1322 2067 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
8237b185 2068
230f1322 2069 local = mkstemp(tmpfile);
916d081b
PB
2070 if (local < 0) {
2071 if (errno == EPERM)
2072 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2073 else
2074 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2075 }
8237b185
DB
2076
2077 memset(&stream, 0, sizeof(stream));
2078
2079 inflateInit(&stream);
2080
2081 SHA1_Init(&c);
2082
2083 do {
2084 ssize_t size;
70b9829e
DB
2085 if (*bufposn) {
2086 stream.avail_in = *bufposn;
96ad15ae 2087 stream.next_in = (unsigned char *) buffer;
70b9829e
DB
2088 do {
2089 stream.next_out = discard;
2090 stream.avail_out = sizeof(discard);
2091 ret = inflate(&stream, Z_SYNC_FLUSH);
2092 SHA1_Update(&c, discard, sizeof(discard) -
2093 stream.avail_out);
2094 } while (stream.avail_in && ret == Z_OK);
4d548150
LT
2095 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2096 die("unable to write sha1 file");
70b9829e
DB
2097 memmove(buffer, buffer + *bufposn - stream.avail_in,
2098 stream.avail_in);
2099 *bufposn = stream.avail_in;
2100 if (ret != Z_OK)
2101 break;
2102 }
93d26e4c 2103 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
8237b185
DB
2104 if (size <= 0) {
2105 close(local);
230f1322 2106 unlink(tmpfile);
8237b185
DB
2107 if (!size)
2108 return error("Connection closed?");
2109 perror("Reading from connection");
2110 return -1;
2111 }
70b9829e
DB
2112 *bufposn += size;
2113 } while (1);
8237b185
DB
2114 inflateEnd(&stream);
2115
2116 close(local);
2117 SHA1_Final(real_sha1, &c);
2118 if (ret != Z_STREAM_END) {
230f1322 2119 unlink(tmpfile);
8237b185
DB
2120 return error("File %s corrupted", sha1_to_hex(sha1));
2121 }
a89fccd2 2122 if (hashcmp(sha1, real_sha1)) {
230f1322 2123 unlink(tmpfile);
bd2afde8 2124 return error("File %s has bad hash", sha1_to_hex(sha1));
8237b185 2125 }
230f1322
LT
2126
2127 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
8237b185
DB
2128}
2129
bf592c50
DB
2130int has_pack_index(const unsigned char *sha1)
2131{
2132 struct stat st;
2133 if (stat(sha1_pack_index_name(sha1), &st))
2134 return 0;
2135 return 1;
2136}
2137
2138int has_pack_file(const unsigned char *sha1)
2139{
2140 struct stat st;
2141 if (stat(sha1_pack_name(sha1), &st))
2142 return 0;
2143 return 1;
2144}
2145
106d710b 2146int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
dade09c2
LT
2147{
2148 struct pack_entry e;
106d710b 2149 return find_pack_entry(sha1, &e, ignore_packed);
dade09c2
LT
2150}
2151
8237b185
DB
2152int has_sha1_file(const unsigned char *sha1)
2153{
8237b185 2154 struct stat st;
1f688557
JH
2155 struct pack_entry e;
2156
106d710b 2157 if (find_pack_entry(sha1, &e, NULL))
1f688557 2158 return 1;
ab90ea5d 2159 return find_sha1_file(sha1, &st) ? 1 : 0;
8237b185 2160}
74400e71 2161
e7332f96
BE
2162/*
2163 * reads from fd as long as possible into a supplied buffer of size bytes.
82e5a82f 2164 * If necessary the buffer's size is increased using realloc()
e7332f96
BE
2165 *
2166 * returns 0 if anything went fine and -1 otherwise
2167 *
2168 * NOTE: both buf and size may change, but even when -1 is returned
2169 * you still have to free() it yourself.
2170 */
2171int read_pipe(int fd, char** return_buf, unsigned long* return_size)
024510c8 2172{
e7332f96
BE
2173 char* buf = *return_buf;
2174 unsigned long size = *return_size;
2175 int iret;
024510c8 2176 unsigned long off = 0;
e7332f96 2177
024510c8 2178 do {
e7332f96 2179 iret = xread(fd, buf + off, size - off);
024510c8
DB
2180 if (iret > 0) {
2181 off += iret;
2182 if (off == size) {
2183 size *= 2;
83572c1a 2184 buf = xrealloc(buf, size);
024510c8
DB
2185 }
2186 }
2187 } while (iret > 0);
e7332f96
BE
2188
2189 *return_buf = buf;
2190 *return_size = off;
2191
2192 if (iret < 0)
2193 return -1;
2194 return 0;
2195}
2196
2197int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2198{
2199 unsigned long size = 4096;
2d7320d0 2200 char *buf = xmalloc(size);
e7332f96 2201 int ret;
e7332f96
BE
2202
2203 if (read_pipe(fd, &buf, &size)) {
024510c8
DB
2204 free(buf);
2205 return -1;
2206 }
e7332f96 2207
024510c8 2208 if (!type)
8e440259 2209 type = blob_type;
024510c8 2210 if (write_object)
e7332f96 2211 ret = write_sha1_file(buf, size, type, sha1);
abdc3fc8
RS
2212 else
2213 ret = hash_sha1_file(buf, size, type, sha1);
024510c8
DB
2214 free(buf);
2215 return ret;
2216}
2217
edaec3fb 2218int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
53bca91a 2219 enum object_type type, const char *path)
74400e71 2220{
dc49cd76 2221 size_t size = xsize_t(st->st_size);
3a55602e 2222 void *buf = NULL;
6c510bee 2223 int ret, re_allocated = 0;
74400e71 2224
74400e71 2225 if (size)
c4712e45 2226 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
74400e71 2227 close(fd);
74400e71 2228
7672db20 2229 if (!type)
edaec3fb 2230 type = OBJ_BLOB;
6c510bee
LT
2231
2232 /*
2233 * Convert blobs to git internal format
2234 */
d0d8e14d 2235 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
6c510bee
LT
2236 unsigned long nsize = size;
2237 char *nbuf = buf;
53bca91a 2238 if (convert_to_git(path, &nbuf, &nsize)) {
6c510bee
LT
2239 if (size)
2240 munmap(buf, size);
2241 size = nsize;
2242 buf = nbuf;
2243 re_allocated = 1;
2244 }
2245 }
2246
7672db20 2247 if (write_object)
edaec3fb 2248 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 2249 else
edaec3fb 2250 ret = hash_sha1_file(buf, size, typename(type), sha1);
6c510bee
LT
2251 if (re_allocated) {
2252 free(buf);
2253 return ret;
2254 }
aac17941
LT
2255 if (size)
2256 munmap(buf, size);
2257 return ret;
74400e71 2258}
ec1fcc16
JH
2259
2260int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2261{
2262 int fd;
2263 char *target;
dc49cd76 2264 size_t len;
ec1fcc16
JH
2265
2266 switch (st->st_mode & S_IFMT) {
2267 case S_IFREG:
2268 fd = open(path, O_RDONLY);
2269 if (fd < 0)
2270 return error("open(\"%s\"): %s", path,
2271 strerror(errno));
53bca91a 2272 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
ec1fcc16
JH
2273 return error("%s: failed to insert into database",
2274 path);
2275 break;
2276 case S_IFLNK:
dc49cd76
SP
2277 len = xsize_t(st->st_size);
2278 target = xmalloc(len + 1);
2279 if (readlink(path, target, len + 1) != st->st_size) {
ec1fcc16
JH
2280 char *errstr = strerror(errno);
2281 free(target);
2282 return error("readlink(\"%s\"): %s", path,
2283 errstr);
2284 }
abdc3fc8 2285 if (!write_object)
dc49cd76
SP
2286 hash_sha1_file(target, len, blob_type, sha1);
2287 else if (write_sha1_file(target, len, blob_type, sha1))
ec1fcc16
JH
2288 return error("%s: failed to insert into database",
2289 path);
2290 free(target);
2291 break;
2292 default:
2293 return error("%s: unsupported file type", path);
2294 }
2295 return 0;
2296}
a69e5429
JH
2297
2298int read_pack_header(int fd, struct pack_header *header)
2299{
2300 char *c = (char*)header;
2301 ssize_t remaining = sizeof(struct pack_header);
2302 do {
2303 ssize_t r = xread(fd, c, remaining);
2304 if (r <= 0)
2305 /* "eof before pack header was fully read" */
2306 return PH_ERROR_EOF;
2307 remaining -= r;
2308 c += r;
2309 } while (remaining > 0);
2310 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2311 /* "protocol error (pack signature mismatch detected)" */
2312 return PH_ERROR_PACK_SIGNATURE;
2313 if (!pack_version_ok(header->hdr_version))
2314 /* "protocol error (pack version unsupported)" */
2315 return PH_ERROR_PROTOCOL;
2316 return 0;
2317}