don't ever allow SHA1 collisions to exist by fetching a pack
[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;
456cdf6e
LT
1033 if (bytes <= size) {
1034 /*
1035 * The above condition must be (bytes <= size), not
1036 * (bytes < size). In other words, even though we
1037 * expect no more output and set avail_out to zer0,
1038 * the input zlib stream may have bytes that express
1039 * "this concludes the stream", and we *do* want to
1040 * eat that input.
1041 *
1042 * Otherwise we would not be able to test that we
1043 * consumed all the input to reach the expected size;
1044 * we also want to check that zlib tells us that all
1045 * went well with status == Z_STREAM_END at the end.
1046 */
5180cacc
LT
1047 stream->next_out = buf + bytes;
1048 stream->avail_out = size - bytes;
7efbff75
JH
1049 while (status == Z_OK)
1050 status = inflate(stream, Z_FINISH);
5180cacc
LT
1051 }
1052 buf[size] = 0;
456cdf6e 1053 if (status == Z_STREAM_END && !stream->avail_in) {
7efbff75
JH
1054 inflateEnd(stream);
1055 return buf;
1056 }
1057
1058 if (status < 0)
1059 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1060 else if (stream->avail_in)
1061 error("garbage at end of loose object '%s'",
1062 sha1_to_hex(sha1));
1063 free(buf);
1064 return NULL;
5180cacc
LT
1065}
1066
1067/*
1068 * We used to just use "sscanf()", but that's actually way
1069 * too permissive for what we want to check. So do an anal
1070 * object header parse by hand.
1071 */
21666f1a 1072static int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1073{
21666f1a 1074 char type[10];
5180cacc
LT
1075 int i;
1076 unsigned long size;
1077
1078 /*
1079 * The type can be at most ten bytes (including the
1080 * terminating '\0' that we add), and is followed by
21666f1a 1081 * a space.
5180cacc 1082 */
21666f1a 1083 i = 0;
5180cacc
LT
1084 for (;;) {
1085 char c = *hdr++;
1086 if (c == ' ')
1087 break;
21666f1a
NP
1088 type[i++] = c;
1089 if (i >= sizeof(type))
5180cacc 1090 return -1;
5180cacc 1091 }
21666f1a 1092 type[i] = 0;
5180cacc
LT
1093
1094 /*
1095 * The length must follow immediately, and be in canonical
1096 * decimal format (ie "010" is not valid).
1097 */
1098 size = *hdr++ - '0';
1099 if (size > 9)
1100 return -1;
1101 if (size) {
1102 for (;;) {
1103 unsigned long c = *hdr - '0';
1104 if (c > 9)
1105 break;
1106 hdr++;
1107 size = size * 10 + c;
1108 }
1109 }
1110 *sizep = size;
1111
1112 /*
1113 * The length must be followed by a zero byte
1114 */
21666f1a 1115 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1116}
1117
7efbff75 1118static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1119{
5180cacc 1120 int ret;
0fcfd160 1121 z_stream stream;
5180cacc 1122 char hdr[8192];
0fcfd160 1123
5180cacc 1124 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1125 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1126 return NULL;
1127
7efbff75 1128 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1129}
1130
c4001d92 1131static off_t get_delta_base(struct packed_git *p,
03e79c88 1132 struct pack_window **w_curs,
c4001d92 1133 off_t *curpos,
21666f1a 1134 enum object_type type,
c4001d92 1135 off_t delta_obj_offset)
eb32d236 1136{
2b87c45b 1137 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1138 off_t base_offset;
eb32d236 1139
8d8a4ea5
SP
1140 /* use_pack() assured us we have [base_info, base_info + 20)
1141 * as a range that we can look at without walking off the
1142 * end of the mapped window. Its actually the hash size
1143 * that is assured. An OFS_DELTA longer than the hash size
1144 * is stupid, as then a REF_DELTA would be smaller to store.
1145 */
21666f1a 1146 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1147 unsigned used = 0;
1148 unsigned char c = base_info[used++];
1149 base_offset = c & 127;
1150 while (c & 128) {
1151 base_offset += 1;
1152 if (!base_offset || base_offset & ~(~0UL >> 7))
1153 die("offset value overflow for delta base object");
1154 c = base_info[used++];
1155 base_offset = (base_offset << 7) + (c & 127);
1156 }
1157 base_offset = delta_obj_offset - base_offset;
1158 if (base_offset >= delta_obj_offset)
1159 die("delta base offset out of bound");
2b87c45b 1160 *curpos += used;
21666f1a 1161 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1162 /* The base entry _must_ be in the same pack */
1163 base_offset = find_pack_entry_one(base_info, p);
1164 if (!base_offset)
1165 die("failed to find delta-pack base object %s",
1166 sha1_to_hex(base_info));
2b87c45b 1167 *curpos += 20;
eb32d236
NP
1168 } else
1169 die("I am totally screwed");
2b87c45b 1170 return base_offset;
eb32d236
NP
1171}
1172
f3bf9224 1173/* forward declaration for a mutually recursive function */
c4001d92 1174static int packed_object_info(struct packed_git *p, off_t offset,
21666f1a 1175 unsigned long *sizep);
f3bf9224 1176
43057304 1177static int packed_delta_info(struct packed_git *p,
03e79c88 1178 struct pack_window **w_curs,
c4001d92 1179 off_t curpos,
21666f1a 1180 enum object_type type,
c4001d92 1181 off_t obj_offset,
43057304 1182 unsigned long *sizep)
5db47c2b 1183{
c4001d92 1184 off_t base_offset;
f3bf9224 1185
21666f1a
NP
1186 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1187 type = packed_object_info(p, base_offset, NULL);
f3bf9224 1188
c62266f3
JH
1189 /* We choose to only get the type of the base object and
1190 * ignore potentially corrupt pack file that expects the delta
1191 * based on a base with a wrong size. This saves tons of
1192 * inflate() calls.
1193 */
c62266f3
JH
1194 if (sizep) {
1195 const unsigned char *data;
079afb18 1196 unsigned char delta_head[20], *in;
c62266f3
JH
1197 z_stream stream;
1198 int st;
1199
1200 memset(&stream, 0, sizeof(stream));
c62266f3
JH
1201 stream.next_out = delta_head;
1202 stream.avail_out = sizeof(delta_head);
1203
1204 inflateInit(&stream);
079afb18 1205 do {
2b87c45b 1206 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18
SP
1207 stream.next_in = in;
1208 st = inflate(&stream, Z_FINISH);
2b87c45b 1209 curpos += stream.next_in - in;
079afb18
SP
1210 } while ((st == Z_OK || st == Z_BUF_ERROR)
1211 && stream.total_out < sizeof(delta_head));
c62266f3
JH
1212 inflateEnd(&stream);
1213 if ((st != Z_STREAM_END) &&
1214 stream.total_out != sizeof(delta_head))
1215 die("delta data unpack-initial failed");
1216
1217 /* Examine the initial part of the delta to figure out
1218 * the result size.
1219 */
1220 data = delta_head;
8960844a
NP
1221
1222 /* ignore base size */
1223 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
5db47c2b 1224
c62266f3 1225 /* Read the result size */
2b87c45b 1226 *sizep = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
c62266f3 1227 }
21666f1a
NP
1228
1229 return type;
5db47c2b
JH
1230}
1231
2b87c45b
NP
1232static int unpack_object_header(struct packed_git *p,
1233 struct pack_window **w_curs,
c4001d92 1234 off_t *curpos,
2b87c45b 1235 unsigned long *sizep)
a733cb60 1236{
03e79c88
SP
1237 unsigned char *base;
1238 unsigned int left;
ad1ed5ee 1239 unsigned long used;
2b87c45b 1240 enum object_type type;
a733cb60 1241
8d8a4ea5
SP
1242 /* use_pack() assures us we have [base, base + 20) available
1243 * as a range that we can look at at. (Its actually the hash
3dff5379 1244 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1245 * the maximum deflated object size is 2^137, which is just
1246 * insane, so we know won't exceed what we have been given.
1247 */
2b87c45b
NP
1248 base = use_pack(p, w_curs, *curpos, &left);
1249 used = unpack_object_header_gently(base, left, &type, sizep);
ad1ed5ee
JH
1250 if (!used)
1251 die("object offset outside of pack file");
2b87c45b 1252 *curpos += used;
ad1ed5ee 1253
2b87c45b 1254 return type;
a733cb60
LT
1255}
1256
21666f1a 1257const char *packed_object_info_detail(struct packed_git *p,
c4001d92 1258 off_t obj_offset,
21666f1a
NP
1259 unsigned long *size,
1260 unsigned long *store_size,
1261 unsigned int *delta_chain_length,
1262 unsigned char *base_sha1)
ad8c80a5 1263{
03e79c88 1264 struct pack_window *w_curs = NULL;
c4001d92
SP
1265 off_t curpos;
1266 unsigned long dummy;
43057304 1267 unsigned char *next_sha1;
21666f1a 1268 enum object_type type;
ad8c80a5 1269
43057304 1270 *delta_chain_length = 0;
2b87c45b 1271 curpos = obj_offset;
21666f1a 1272 type = unpack_object_header(p, &w_curs, &curpos, size);
43057304
NP
1273
1274 for (;;) {
21666f1a 1275 switch (type) {
43057304 1276 default:
08a19d87 1277 die("pack %s contains unknown object type %d",
21666f1a 1278 p->pack_name, type);
43057304
NP
1279 case OBJ_COMMIT:
1280 case OBJ_TREE:
1281 case OBJ_BLOB:
1282 case OBJ_TAG:
43057304 1283 *store_size = 0; /* notyet */
03e79c88 1284 unuse_pack(&w_curs);
21666f1a 1285 return typename(type);
eb32d236 1286 case OBJ_OFS_DELTA:
21666f1a 1287 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
eb32d236 1288 if (*delta_chain_length == 0) {
2b87c45b 1289 /* TODO: find base_sha1 as pointed by curpos */
30fee062 1290 hashclr(base_sha1);
eb32d236
NP
1291 }
1292 break;
1293 case OBJ_REF_DELTA:
2b87c45b 1294 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
43057304
NP
1295 if (*delta_chain_length == 0)
1296 hashcpy(base_sha1, next_sha1);
2b87c45b 1297 obj_offset = find_pack_entry_one(next_sha1, p);
43057304
NP
1298 break;
1299 }
43057304 1300 (*delta_chain_length)++;
2b87c45b 1301 curpos = obj_offset;
21666f1a 1302 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
ad8c80a5 1303 }
ad8c80a5
JH
1304}
1305
c4001d92 1306static int packed_object_info(struct packed_git *p, off_t obj_offset,
21666f1a 1307 unsigned long *sizep)
1f688557 1308{
03e79c88 1309 struct pack_window *w_curs = NULL;
c4001d92
SP
1310 unsigned long size;
1311 off_t curpos = obj_offset;
21666f1a 1312 enum object_type type;
5db47c2b 1313
21666f1a 1314 type = unpack_object_header(p, &w_curs, &curpos, &size);
5db47c2b 1315
21666f1a 1316 switch (type) {
eb32d236
NP
1317 case OBJ_OFS_DELTA:
1318 case OBJ_REF_DELTA:
21666f1a
NP
1319 type = packed_delta_info(p, &w_curs, curpos,
1320 type, obj_offset, sizep);
2b87c45b 1321 break;
a733cb60 1322 case OBJ_COMMIT:
a733cb60 1323 case OBJ_TREE:
a733cb60 1324 case OBJ_BLOB:
a733cb60 1325 case OBJ_TAG:
2b87c45b
NP
1326 if (sizep)
1327 *sizep = size;
a69d0943 1328 break;
1f688557 1329 default:
08a19d87 1330 die("pack %s contains unknown object type %d",
21666f1a 1331 p->pack_name, type);
1f688557 1332 }
2b87c45b 1333 unuse_pack(&w_curs);
21666f1a 1334 return type;
1f688557
JH
1335}
1336
eb950c19 1337static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1338 struct pack_window **w_curs,
c4001d92 1339 off_t curpos,
eb950c19 1340 unsigned long size)
de530aaa
SP
1341{
1342 int st;
1343 z_stream stream;
079afb18 1344 unsigned char *buffer, *in;
de530aaa
SP
1345
1346 buffer = xmalloc(size + 1);
1347 buffer[size] = 0;
1348 memset(&stream, 0, sizeof(stream));
de530aaa
SP
1349 stream.next_out = buffer;
1350 stream.avail_out = size;
1351
1352 inflateInit(&stream);
079afb18 1353 do {
2b87c45b 1354 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18
SP
1355 stream.next_in = in;
1356 st = inflate(&stream, Z_FINISH);
2b87c45b 1357 curpos += stream.next_in - in;
079afb18 1358 } while (st == Z_OK || st == Z_BUF_ERROR);
de530aaa
SP
1359 inflateEnd(&stream);
1360 if ((st != Z_STREAM_END) || stream.total_out != size) {
1361 free(buffer);
1362 return NULL;
1363 }
1364
1365 return buffer;
1366}
1367
e5e01619
LT
1368#define MAX_DELTA_CACHE (256)
1369
18bdec11 1370static size_t delta_base_cached;
5e08ecbf
NP
1371
1372static struct delta_base_cache_lru_list {
1373 struct delta_base_cache_lru_list *prev;
1374 struct delta_base_cache_lru_list *next;
1375} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1376
e5e01619 1377static struct delta_base_cache_entry {
5e08ecbf
NP
1378 struct delta_base_cache_lru_list lru;
1379 void *data;
e5e01619
LT
1380 struct packed_git *p;
1381 off_t base_offset;
1382 unsigned long size;
e5e01619
LT
1383 enum object_type type;
1384} delta_base_cache[MAX_DELTA_CACHE];
1385
1386static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1387{
1388 unsigned long hash;
1389
1390 hash = (unsigned long)p + (unsigned long)base_offset;
1391 hash += (hash >> 8) + (hash >> 16);
3358004a 1392 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1393}
1394
62f255ad 1395static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 1396 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 1397{
e5e01619
LT
1398 void *ret;
1399 unsigned long hash = pack_entry_hash(p, base_offset);
1400 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1401
1402 ret = ent->data;
1403 if (ret && ent->p == p && ent->base_offset == base_offset)
1404 goto found_cache_entry;
62f255ad 1405 return unpack_entry(p, base_offset, type, base_size);
e5e01619
LT
1406
1407found_cache_entry:
18bdec11 1408 if (!keep_cache) {
a0cba108 1409 ent->data = NULL;
5e08ecbf
NP
1410 ent->lru.next->prev = ent->lru.prev;
1411 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
1412 delta_base_cached -= ent->size;
1413 }
a0cba108
NP
1414 else {
1415 ret = xmalloc(ent->size + 1);
1416 memcpy(ret, ent->data, ent->size);
1417 ((char *)ret)[ent->size] = 0;
1418 }
e5e01619
LT
1419 *type = ent->type;
1420 *base_size = ent->size;
1421 return ret;
62f255ad
LT
1422}
1423
18bdec11
SP
1424static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1425{
1426 if (ent->data) {
1427 free(ent->data);
1428 ent->data = NULL;
5e08ecbf
NP
1429 ent->lru.next->prev = ent->lru.prev;
1430 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
1431 delta_base_cached -= ent->size;
1432 }
1433}
1434
62f255ad
LT
1435static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1436 void *base, unsigned long base_size, enum object_type type)
1437{
5e08ecbf 1438 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 1439 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 1440 struct delta_base_cache_lru_list *lru;
e5e01619 1441
18bdec11
SP
1442 release_delta_base_cache(ent);
1443 delta_base_cached += base_size;
5e08ecbf
NP
1444
1445 for (lru = delta_base_cache_lru.next;
1446 delta_base_cached > delta_base_cache_limit
1447 && lru != &delta_base_cache_lru;
1448 lru = lru->next) {
1449 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
1450 if (f->type == OBJ_BLOB)
1451 release_delta_base_cache(f);
1452 }
5e08ecbf
NP
1453 for (lru = delta_base_cache_lru.next;
1454 delta_base_cached > delta_base_cache_limit
1455 && lru != &delta_base_cache_lru;
1456 lru = lru->next) {
1457 struct delta_base_cache_entry *f = (void *)lru;
1458 release_delta_base_cache(f);
1459 }
18bdec11 1460
e5e01619
LT
1461 ent->p = p;
1462 ent->base_offset = base_offset;
1463 ent->type = type;
1464 ent->data = base;
1465 ent->size = base_size;
5e08ecbf
NP
1466 ent->lru.next = &delta_base_cache_lru;
1467 ent->lru.prev = delta_base_cache_lru.prev;
1468 delta_base_cache_lru.prev->next = &ent->lru;
1469 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
1470}
1471
eb950c19 1472static void *unpack_delta_entry(struct packed_git *p,
03e79c88 1473 struct pack_window **w_curs,
c4001d92 1474 off_t curpos,
1f688557 1475 unsigned long delta_size,
c4001d92 1476 off_t obj_offset,
21666f1a 1477 enum object_type *type,
eb950c19 1478 unsigned long *sizep)
1f688557 1479{
7c3e8be3 1480 void *delta_data, *result, *base;
c4001d92
SP
1481 unsigned long base_size;
1482 off_t base_offset;
43057304 1483
21666f1a 1484 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
a0cba108 1485 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
67686d95 1486 if (!base)
c4001d92
SP
1487 die("failed to read delta base object"
1488 " at %"PRIuMAX" from %s",
1489 (uintmax_t)base_offset, p->pack_name);
67686d95 1490
2b87c45b 1491 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1f688557
JH
1492 result = patch_delta(base, base_size,
1493 delta_data, delta_size,
21666f1a 1494 sizep);
1f688557
JH
1495 if (!result)
1496 die("failed to apply delta");
1497 free(delta_data);
62f255ad 1498 add_delta_base_cache(p, base_offset, base, base_size, *type);
1f688557
JH
1499 return result;
1500}
1501
c4001d92 1502void *unpack_entry(struct packed_git *p, off_t obj_offset,
21666f1a 1503 enum object_type *type, unsigned long *sizep)
1f688557 1504{
03e79c88 1505 struct pack_window *w_curs = NULL;
c4001d92 1506 off_t curpos = obj_offset;
21666f1a 1507 void *data;
1f688557 1508
21666f1a
NP
1509 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1510 switch (*type) {
eb32d236
NP
1511 case OBJ_OFS_DELTA:
1512 case OBJ_REF_DELTA:
21666f1a
NP
1513 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1514 obj_offset, type, sizep);
4d703a1a 1515 break;
a733cb60 1516 case OBJ_COMMIT:
a733cb60 1517 case OBJ_TREE:
a733cb60 1518 case OBJ_BLOB:
a733cb60 1519 case OBJ_TAG:
21666f1a 1520 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
4d703a1a 1521 break;
1f688557 1522 default:
21666f1a 1523 die("unknown object type %i in %s", *type, p->pack_name);
1f688557 1524 }
03e79c88 1525 unuse_pack(&w_curs);
21666f1a 1526 return data;
1f688557
JH
1527}
1528
326bf396 1529uint32_t num_packed_objects(const struct packed_git *p)
9a217f2a 1530{
f9253394 1531 /* See check_packed_git_idx() */
dc49cd76 1532 return (uint32_t)((p->index_size - 20 - 20 - 4*256) / 24);
9a217f2a
JH
1533}
1534
326bf396 1535int nth_packed_object_sha1(const struct packed_git *p, uint32_t n,
9a217f2a
JH
1536 unsigned char* sha1)
1537{
42873078
NP
1538 const unsigned char *index = p->index_data;
1539 index += 4 * 256;
326bf396 1540 if (num_packed_objects(p) <= n)
9a217f2a 1541 return -1;
42873078 1542 hashcpy(sha1, index + 24 * n + 4);
9a217f2a
JH
1543 return 0;
1544}
1545
c4001d92 1546off_t find_pack_entry_one(const unsigned char *sha1,
43057304 1547 struct packed_git *p)
1f688557 1548{
42873078 1549 const uint32_t *level1_ofs = p->index_data;
1f688557
JH
1550 int hi = ntohl(level1_ofs[*sha1]);
1551 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
42873078
NP
1552 const unsigned char *index = p->index_data;
1553
1554 index += 4 * 256;
1f688557
JH
1555
1556 do {
1557 int mi = (lo + hi) / 2;
42873078 1558 int cmp = hashcmp(index + 24 * mi + 4, sha1);
43057304 1559 if (!cmp)
b18b00a6 1560 return ntohl(*((uint32_t *)((char *)index + (24 * mi))));
1f688557
JH
1561 if (cmp > 0)
1562 hi = mi;
1563 else
1564 lo = mi+1;
1565 } while (lo < hi);
1566 return 0;
1567}
1568
d4ff6d92
SP
1569static int matches_pack_name(struct packed_git *p, const char *ig)
1570{
1571 const char *last_c, *c;
1572
1573 if (!strcmp(p->pack_name, ig))
1574 return 0;
1575
1576 for (c = p->pack_name, last_c = c; *c;)
1577 if (*c == '/')
1578 last_c = ++c;
1579 else
1580 ++c;
1581 if (!strcmp(last_c, ig))
1582 return 0;
1583
1584 return 1;
1585}
1586
106d710b 1587static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1f688557
JH
1588{
1589 struct packed_git *p;
c4001d92 1590 off_t offset;
43057304 1591
1f688557
JH
1592 prepare_packed_git();
1593
1594 for (p = packed_git; p; p = p->next) {
106d710b
JH
1595 if (ignore_packed) {
1596 const char **ig;
1597 for (ig = ignore_packed; *ig; ig++)
d4ff6d92 1598 if (!matches_pack_name(p, *ig))
106d710b
JH
1599 break;
1600 if (*ig)
1601 continue;
1602 }
43057304
NP
1603 offset = find_pack_entry_one(sha1, p);
1604 if (offset) {
c715f783
SP
1605 /*
1606 * We are about to tell the caller where they can
1607 * locate the requested object. We better make
1608 * sure the packfile is still here and can be
1609 * accessed before supplying that answer, as
1610 * it may have been deleted since the index
1611 * was loaded!
1612 */
1613 if (p->pack_fd == -1 && open_packed_git(p)) {
1614 error("packfile %s cannot be accessed", p->pack_name);
1615 continue;
1616 }
43057304
NP
1617 e->offset = offset;
1618 e->p = p;
1619 hashcpy(e->sha1, sha1);
1f688557 1620 return 1;
43057304 1621 }
1f688557
JH
1622 }
1623 return 0;
1624}
1625
bf592c50
DB
1626struct packed_git *find_sha1_pack(const unsigned char *sha1,
1627 struct packed_git *packs)
1628{
1629 struct packed_git *p;
bf592c50
DB
1630
1631 for (p = packs; p; p = p->next) {
43057304 1632 if (find_pack_entry_one(sha1, p))
bf592c50
DB
1633 return p;
1634 }
1635 return NULL;
21666f1a 1636
bf592c50
DB
1637}
1638
21666f1a 1639static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
65c2e0c3 1640{
36e4d74a 1641 int status;
65c2e0c3
JH
1642 unsigned long mapsize, size;
1643 void *map;
1644 z_stream stream;
d65a16f6 1645 char hdr[32];
65c2e0c3 1646
bb6b8e4f 1647 map = map_sha1_file(sha1, &mapsize);
f0df4ed5
JS
1648 if (!map)
1649 return error("unable to find %s", sha1_to_hex(sha1));
36e4d74a
JH
1650 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1651 status = error("unable to unpack %s header",
1652 sha1_to_hex(sha1));
21666f1a 1653 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 1654 status = error("unable to parse %s header", sha1_to_hex(sha1));
21666f1a
NP
1655 else if (sizep)
1656 *sizep = size;
65c2e0c3
JH
1657 inflateEnd(&stream);
1658 munmap(map, mapsize);
1659 return status;
1660}
1661
21666f1a 1662int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
f0df4ed5 1663{
f0df4ed5
JS
1664 struct pack_entry e;
1665
1666 if (!find_pack_entry(sha1, &e, NULL)) {
1667 reprepare_packed_git();
1668 if (!find_pack_entry(sha1, &e, NULL))
21666f1a 1669 return sha1_loose_object_info(sha1, sizep);
f0df4ed5 1670 }
21666f1a 1671 return packed_object_info(e.p, e.offset, sizep);
f0df4ed5
JS
1672}
1673
21666f1a
NP
1674static void *read_packed_sha1(const unsigned char *sha1,
1675 enum object_type *type, unsigned long *size)
1f688557
JH
1676{
1677 struct pack_entry e;
1678
8276c007 1679 if (!find_pack_entry(sha1, &e, NULL))
1f688557 1680 return NULL;
8276c007 1681 else
a0cba108 1682 return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1f688557
JH
1683}
1684
d66b37bb
JH
1685/*
1686 * This is meant to hold a *small* number of objects that you would
1687 * want read_sha1_file() to be able to return, but yet you do not want
1688 * to write them into the object store (e.g. a browse-only
1689 * application).
1690 */
1691static struct cached_object {
1692 unsigned char sha1[20];
21666f1a 1693 enum object_type type;
d66b37bb
JH
1694 void *buf;
1695 unsigned long size;
1696} *cached_objects;
1697static int cached_object_nr, cached_object_alloc;
1698
1699static struct cached_object *find_cached_object(const unsigned char *sha1)
1700{
1701 int i;
1702 struct cached_object *co = cached_objects;
1703
1704 for (i = 0; i < cached_object_nr; i++, co++) {
1705 if (!hashcmp(co->sha1, sha1))
1706 return co;
1707 }
1708 return NULL;
1709}
1710
21666f1a
NP
1711int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1712 unsigned char *sha1)
d66b37bb
JH
1713{
1714 struct cached_object *co;
1715
21666f1a 1716 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
1717 if (has_sha1_file(sha1) || find_cached_object(sha1))
1718 return 0;
1719 if (cached_object_alloc <= cached_object_nr) {
1720 cached_object_alloc = alloc_nr(cached_object_alloc);
1721 cached_objects = xrealloc(cached_objects,
1722 sizeof(*cached_objects) *
1723 cached_object_alloc);
1724 }
1725 co = &cached_objects[cached_object_nr++];
1726 co->size = len;
21666f1a 1727 co->type = type;
efa13f7b
JH
1728 co->buf = xmalloc(len);
1729 memcpy(co->buf, buf, len);
d66b37bb
JH
1730 hashcpy(co->sha1, sha1);
1731 return 0;
1732}
1733
21666f1a
NP
1734void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1735 unsigned long *size)
0fcfd160
LT
1736{
1737 unsigned long mapsize;
1738 void *map, *buf;
d66b37bb
JH
1739 struct cached_object *co;
1740
1741 co = find_cached_object(sha1);
1742 if (co) {
1743 buf = xmalloc(co->size + 1);
1744 memcpy(buf, co->buf, co->size);
1745 ((char*)buf)[co->size] = 0;
21666f1a 1746 *type = co->type;
d66b37bb
JH
1747 *size = co->size;
1748 return buf;
1749 }
0fcfd160 1750
8276c007
PE
1751 buf = read_packed_sha1(sha1, type, size);
1752 if (buf)
1753 return buf;
bb6b8e4f 1754 map = map_sha1_file(sha1, &mapsize);
0fcfd160 1755 if (map) {
7efbff75 1756 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
1757 munmap(map, mapsize);
1758 return buf;
1759 }
637cdd9d 1760 reprepare_packed_git();
8276c007 1761 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
1762}
1763
40469ee9 1764void *read_object_with_reference(const unsigned char *sha1,
21666f1a 1765 const char *required_type_name,
40469ee9
JH
1766 unsigned long *size,
1767 unsigned char *actual_sha1_return)
f4913f91 1768{
21666f1a 1769 enum object_type type, required_type;
f4913f91
JH
1770 void *buffer;
1771 unsigned long isize;
40469ee9 1772 unsigned char actual_sha1[20];
f4913f91 1773
21666f1a 1774 required_type = type_from_string(required_type_name);
e702496e 1775 hashcpy(actual_sha1, sha1);
40469ee9
JH
1776 while (1) {
1777 int ref_length = -1;
1778 const char *ref_type = NULL;
f4913f91 1779
21666f1a 1780 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
1781 if (!buffer)
1782 return NULL;
21666f1a 1783 if (type == required_type) {
40469ee9
JH
1784 *size = isize;
1785 if (actual_sha1_return)
e702496e 1786 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
1787 return buffer;
1788 }
1789 /* Handle references */
21666f1a 1790 else if (type == OBJ_COMMIT)
40469ee9 1791 ref_type = "tree ";
21666f1a 1792 else if (type == OBJ_TAG)
40469ee9
JH
1793 ref_type = "object ";
1794 else {
1795 free(buffer);
1796 return NULL;
1797 }
1798 ref_length = strlen(ref_type);
f4913f91 1799
40469ee9 1800 if (memcmp(buffer, ref_type, ref_length) ||
1d7f171c 1801 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
1802 free(buffer);
1803 return NULL;
1804 }
1cf58e72 1805 free(buffer);
40469ee9
JH
1806 /* Now we have the ID of the referred-to object in
1807 * actual_sha1. Check again. */
f4913f91 1808 }
f4913f91
JH
1809}
1810
972a9155
RS
1811static void write_sha1_file_prepare(void *buf, unsigned long len,
1812 const char *type, unsigned char *sha1,
d65a16f6 1813 char *hdr, int *hdrlen)
d410c0f5
JH
1814{
1815 SHA_CTX c;
1816
1817 /* Generate the header */
d65a16f6 1818 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
1819
1820 /* Sha1.. */
1821 SHA1_Init(&c);
1822 SHA1_Update(&c, hdr, *hdrlen);
1823 SHA1_Update(&c, buf, len);
1824 SHA1_Final(sha1, &c);
d410c0f5
JH
1825}
1826
230f1322
LT
1827/*
1828 * Link the tempfile to the final place, possibly creating the
1829 * last directory level as you do so.
1830 *
1831 * Returns the errno on failure, 0 on success.
1832 */
839837b9 1833static int link_temp_to_file(const char *tmpfile, const char *filename)
230f1322
LT
1834{
1835 int ret;
756aaf4a 1836 char *dir;
230f1322
LT
1837
1838 if (!link(tmpfile, filename))
1839 return 0;
1840
1841 /*
756aaf4a 1842 * Try to mkdir the last path component if that failed.
230f1322
LT
1843 *
1844 * Re-try the "link()" regardless of whether the mkdir
1845 * succeeds, since a race might mean that somebody
1846 * else succeeded.
1847 */
1848 ret = errno;
756aaf4a
SP
1849 dir = strrchr(filename, '/');
1850 if (dir) {
1851 *dir = 0;
866cae0d 1852 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
91c23e48 1853 *dir = '/';
756aaf4a 1854 return -2;
91c23e48 1855 }
756aaf4a
SP
1856 *dir = '/';
1857 if (!link(tmpfile, filename))
1858 return 0;
1859 ret = errno;
230f1322
LT
1860 }
1861 return ret;
1862}
1863
1864/*
1865 * Move the just written object into its final resting place
1866 */
839837b9 1867int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322
LT
1868{
1869 int ret = link_temp_to_file(tmpfile, filename);
7ebb6fca
LT
1870
1871 /*
1872 * Coda hack - coda doesn't like cross-directory links,
1873 * so we fall back to a rename, which will mean that it
1874 * won't be able to check collisions, but that's not a
1875 * big deal.
1876 *
1877 * The same holds for FAT formatted media.
1878 *
1879 * When this succeeds, we just return 0. We have nothing
1880 * left to unlink.
1881 */
1882 if (ret && ret != EEXIST) {
1883 if (!rename(tmpfile, filename))
230f1322 1884 return 0;
9e48b389 1885 ret = errno;
230f1322
LT
1886 }
1887 unlink(tmpfile);
1888 if (ret) {
1889 if (ret != EEXIST) {
916d081b 1890 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
230f1322
LT
1891 }
1892 /* FIXME!!! Collision check here ? */
1893 }
1894
1895 return 0;
1896}
1897
4d548150
LT
1898static int write_buffer(int fd, const void *buf, size_t len)
1899{
d34cf19b 1900 if (write_in_full(fd, buf, len) < 0)
93822c22 1901 return error("file write error (%s)", strerror(errno));
4d548150
LT
1902 return 0;
1903}
1904
93821bd9
LT
1905static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1906{
1907 int hdr_len;
1908 unsigned char c;
1909
1910 c = (type << 4) | (len & 15);
1911 len >>= 4;
1912 hdr_len = 1;
1913 while (len) {
1914 *hdr++ = c | 0x80;
1915 hdr_len++;
1916 c = (len & 0x7f);
1917 len >>= 7;
1918 }
1919 *hdr = c;
1920 return hdr_len;
1921}
1922
1923static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1924{
d65a16f6 1925 int obj_type, hdrlen;
93821bd9
LT
1926
1927 if (use_legacy_headers) {
1928 while (deflate(stream, 0) == Z_OK)
1929 /* nothing */;
1930 return;
1931 }
df843662 1932 obj_type = type_from_string(type);
d65a16f6
NP
1933 hdrlen = write_binary_header(stream->next_out, obj_type, len);
1934 stream->total_out = hdrlen;
1935 stream->next_out += hdrlen;
1936 stream->avail_out -= hdrlen;
93821bd9
LT
1937}
1938
abdc3fc8
RS
1939int hash_sha1_file(void *buf, unsigned long len, const char *type,
1940 unsigned char *sha1)
1941{
d65a16f6 1942 char hdr[32];
abdc3fc8
RS
1943 int hdrlen;
1944 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1945 return 0;
1946}
1947
bf0f910d 1948int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
0fcfd160
LT
1949{
1950 int size;
bf0f910d 1951 unsigned char *compressed;
0fcfd160
LT
1952 z_stream stream;
1953 unsigned char sha1[20];
706bc531 1954 char *filename;
aac17941 1955 static char tmpfile[PATH_MAX];
d65a16f6 1956 char hdr[32];
230f1322 1957 int fd, hdrlen;
a44c9a5e 1958
d410c0f5
JH
1959 /* Normally if we have it in the pack then we do not bother writing
1960 * it out into .git/objects/??/?{38} file.
1961 */
972a9155
RS
1962 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1963 filename = sha1_file_name(sha1);
706bc531 1964 if (returnsha1)
e702496e 1965 hashcpy(returnsha1, sha1);
d410c0f5
JH
1966 if (has_sha1_file(sha1))
1967 return 0;
aac17941
LT
1968 fd = open(filename, O_RDONLY);
1969 if (fd >= 0) {
706bc531 1970 /*
aac17941
LT
1971 * FIXME!!! We might do collision checking here, but we'd
1972 * need to uncompress the old file and check it. Later.
706bc531 1973 */
aac17941 1974 close(fd);
706bc531
LT
1975 return 0;
1976 }
1977
aac17941 1978 if (errno != ENOENT) {
916d081b 1979 return error("sha1 file %s: %s\n", filename, strerror(errno));
aac17941
LT
1980 }
1981
1982 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
ace1534d 1983
aac17941
LT
1984 fd = mkstemp(tmpfile);
1985 if (fd < 0) {
916d081b
PB
1986 if (errno == EPERM)
1987 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1988 else
1989 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
aac17941
LT
1990 }
1991
0fcfd160
LT
1992 /* Set it up */
1993 memset(&stream, 0, sizeof(stream));
12f6c308 1994 deflateInit(&stream, zlib_compression_level);
93821bd9 1995 size = 8 + deflateBound(&stream, len+hdrlen);
812666c8 1996 compressed = xmalloc(size);
0fcfd160
LT
1997
1998 /* Compress it */
0fcfd160
LT
1999 stream.next_out = compressed;
2000 stream.avail_out = size;
a44c9a5e
LT
2001
2002 /* First header.. */
d65a16f6 2003 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2004 stream.avail_in = hdrlen;
93821bd9 2005 setup_object_header(&stream, type, len);
a44c9a5e
LT
2006
2007 /* Then the data itself.. */
2008 stream.next_in = buf;
2009 stream.avail_in = len;
0fcfd160
LT
2010 while (deflate(&stream, Z_FINISH) == Z_OK)
2011 /* nothing */;
2012 deflateEnd(&stream);
2013 size = stream.total_out;
2014
4d548150
LT
2015 if (write_buffer(fd, compressed, size) < 0)
2016 die("unable to write sha1 file");
aac17941 2017 fchmod(fd, 0444);
706bc531 2018 close(fd);
383f85b7 2019 free(compressed);
0fcfd160 2020
230f1322 2021 return move_temp_to_file(tmpfile, filename);
0fcfd160 2022}
8237b185 2023
4d548150
LT
2024/*
2025 * We need to unpack and recompress the object for writing
2026 * it out to a different file.
2027 */
2028static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
a5eda52b 2029{
4d548150 2030 size_t size;
a5eda52b 2031 z_stream stream;
4d548150
LT
2032 unsigned char *unpacked;
2033 unsigned long len;
21666f1a 2034 enum object_type type;
d65a16f6 2035 char hdr[32];
4d548150
LT
2036 int hdrlen;
2037 void *buf;
bfc66daf 2038
a9486b02 2039 /* need to unpack and recompress it by itself */
21666f1a 2040 unpacked = read_packed_sha1(sha1, &type, &len);
8276c007
PE
2041 if (!unpacked)
2042 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
a5eda52b 2043
21666f1a 2044 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
a5eda52b 2045
4d548150
LT
2046 /* Set it up */
2047 memset(&stream, 0, sizeof(stream));
12f6c308 2048 deflateInit(&stream, zlib_compression_level);
4d548150
LT
2049 size = deflateBound(&stream, len + hdrlen);
2050 buf = xmalloc(size);
a5eda52b 2051
4d548150
LT
2052 /* Compress it */
2053 stream.next_out = buf;
2054 stream.avail_out = size;
a5eda52b 2055
4d548150
LT
2056 /* First header.. */
2057 stream.next_in = (void *)hdr;
2058 stream.avail_in = hdrlen;
2059 while (deflate(&stream, 0) == Z_OK)
2060 /* nothing */;
bfc66daf 2061
4d548150
LT
2062 /* Then the data itself.. */
2063 stream.next_in = unpacked;
2064 stream.avail_in = len;
2065 while (deflate(&stream, Z_FINISH) == Z_OK)
2066 /* nothing */;
2067 deflateEnd(&stream);
2068 free(unpacked);
bfc66daf 2069
4d548150
LT
2070 *objsize = stream.total_out;
2071 return buf;
2072}
2073
2074int write_sha1_to_fd(int fd, const unsigned char *sha1)
2075{
2076 int retval;
2077 unsigned long objsize;
bb6b8e4f 2078 void *buf = map_sha1_file(sha1, &objsize);
4d548150
LT
2079
2080 if (buf) {
2081 retval = write_buffer(fd, buf, objsize);
2082 munmap(buf, objsize);
2083 return retval;
2084 }
2085
2086 buf = repack_object(sha1, &objsize);
2087 retval = write_buffer(fd, buf, objsize);
2088 free(buf);
2089 return retval;
a5eda52b
DB
2090}
2091
70b9829e
DB
2092int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2093 size_t bufsize, size_t *bufposn)
8237b185 2094{
230f1322 2095 char tmpfile[PATH_MAX];
8237b185
DB
2096 int local;
2097 z_stream stream;
2098 unsigned char real_sha1[20];
bf0f910d 2099 unsigned char discard[4096];
8237b185
DB
2100 int ret;
2101 SHA_CTX c;
2102
230f1322 2103 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
8237b185 2104
230f1322 2105 local = mkstemp(tmpfile);
916d081b
PB
2106 if (local < 0) {
2107 if (errno == EPERM)
2108 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2109 else
2110 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2111 }
8237b185
DB
2112
2113 memset(&stream, 0, sizeof(stream));
2114
2115 inflateInit(&stream);
2116
2117 SHA1_Init(&c);
2118
2119 do {
2120 ssize_t size;
70b9829e
DB
2121 if (*bufposn) {
2122 stream.avail_in = *bufposn;
96ad15ae 2123 stream.next_in = (unsigned char *) buffer;
70b9829e
DB
2124 do {
2125 stream.next_out = discard;
2126 stream.avail_out = sizeof(discard);
2127 ret = inflate(&stream, Z_SYNC_FLUSH);
2128 SHA1_Update(&c, discard, sizeof(discard) -
2129 stream.avail_out);
2130 } while (stream.avail_in && ret == Z_OK);
4d548150
LT
2131 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2132 die("unable to write sha1 file");
70b9829e
DB
2133 memmove(buffer, buffer + *bufposn - stream.avail_in,
2134 stream.avail_in);
2135 *bufposn = stream.avail_in;
2136 if (ret != Z_OK)
2137 break;
2138 }
93d26e4c 2139 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
8237b185
DB
2140 if (size <= 0) {
2141 close(local);
230f1322 2142 unlink(tmpfile);
8237b185
DB
2143 if (!size)
2144 return error("Connection closed?");
2145 perror("Reading from connection");
2146 return -1;
2147 }
70b9829e
DB
2148 *bufposn += size;
2149 } while (1);
8237b185
DB
2150 inflateEnd(&stream);
2151
2152 close(local);
2153 SHA1_Final(real_sha1, &c);
2154 if (ret != Z_STREAM_END) {
230f1322 2155 unlink(tmpfile);
8237b185
DB
2156 return error("File %s corrupted", sha1_to_hex(sha1));
2157 }
a89fccd2 2158 if (hashcmp(sha1, real_sha1)) {
230f1322 2159 unlink(tmpfile);
bd2afde8 2160 return error("File %s has bad hash", sha1_to_hex(sha1));
8237b185 2161 }
230f1322
LT
2162
2163 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
8237b185
DB
2164}
2165
bf592c50
DB
2166int has_pack_index(const unsigned char *sha1)
2167{
2168 struct stat st;
2169 if (stat(sha1_pack_index_name(sha1), &st))
2170 return 0;
2171 return 1;
2172}
2173
2174int has_pack_file(const unsigned char *sha1)
2175{
2176 struct stat st;
2177 if (stat(sha1_pack_name(sha1), &st))
2178 return 0;
2179 return 1;
2180}
2181
106d710b 2182int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
dade09c2
LT
2183{
2184 struct pack_entry e;
106d710b 2185 return find_pack_entry(sha1, &e, ignore_packed);
dade09c2
LT
2186}
2187
8237b185
DB
2188int has_sha1_file(const unsigned char *sha1)
2189{
8237b185 2190 struct stat st;
1f688557
JH
2191 struct pack_entry e;
2192
106d710b 2193 if (find_pack_entry(sha1, &e, NULL))
1f688557 2194 return 1;
ab90ea5d 2195 return find_sha1_file(sha1, &st) ? 1 : 0;
8237b185 2196}
74400e71 2197
e7332f96
BE
2198/*
2199 * reads from fd as long as possible into a supplied buffer of size bytes.
82e5a82f 2200 * If necessary the buffer's size is increased using realloc()
e7332f96
BE
2201 *
2202 * returns 0 if anything went fine and -1 otherwise
2203 *
2204 * NOTE: both buf and size may change, but even when -1 is returned
2205 * you still have to free() it yourself.
2206 */
2207int read_pipe(int fd, char** return_buf, unsigned long* return_size)
024510c8 2208{
e7332f96
BE
2209 char* buf = *return_buf;
2210 unsigned long size = *return_size;
2211 int iret;
024510c8 2212 unsigned long off = 0;
e7332f96 2213
024510c8 2214 do {
e7332f96 2215 iret = xread(fd, buf + off, size - off);
024510c8
DB
2216 if (iret > 0) {
2217 off += iret;
2218 if (off == size) {
2219 size *= 2;
83572c1a 2220 buf = xrealloc(buf, size);
024510c8
DB
2221 }
2222 }
2223 } while (iret > 0);
e7332f96
BE
2224
2225 *return_buf = buf;
2226 *return_size = off;
2227
2228 if (iret < 0)
2229 return -1;
2230 return 0;
2231}
2232
2233int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2234{
2235 unsigned long size = 4096;
2d7320d0 2236 char *buf = xmalloc(size);
e7332f96 2237 int ret;
e7332f96
BE
2238
2239 if (read_pipe(fd, &buf, &size)) {
024510c8
DB
2240 free(buf);
2241 return -1;
2242 }
e7332f96 2243
024510c8 2244 if (!type)
8e440259 2245 type = blob_type;
024510c8 2246 if (write_object)
e7332f96 2247 ret = write_sha1_file(buf, size, type, sha1);
abdc3fc8
RS
2248 else
2249 ret = hash_sha1_file(buf, size, type, sha1);
024510c8
DB
2250 free(buf);
2251 return ret;
2252}
2253
edaec3fb 2254int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
53bca91a 2255 enum object_type type, const char *path)
74400e71 2256{
dc49cd76 2257 size_t size = xsize_t(st->st_size);
3a55602e 2258 void *buf = NULL;
6c510bee 2259 int ret, re_allocated = 0;
74400e71 2260
74400e71 2261 if (size)
c4712e45 2262 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
74400e71 2263 close(fd);
74400e71 2264
7672db20 2265 if (!type)
edaec3fb 2266 type = OBJ_BLOB;
6c510bee
LT
2267
2268 /*
2269 * Convert blobs to git internal format
2270 */
d0d8e14d 2271 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
6c510bee
LT
2272 unsigned long nsize = size;
2273 char *nbuf = buf;
53bca91a 2274 if (convert_to_git(path, &nbuf, &nsize)) {
6c510bee
LT
2275 if (size)
2276 munmap(buf, size);
2277 size = nsize;
2278 buf = nbuf;
2279 re_allocated = 1;
2280 }
2281 }
2282
7672db20 2283 if (write_object)
edaec3fb 2284 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 2285 else
edaec3fb 2286 ret = hash_sha1_file(buf, size, typename(type), sha1);
6c510bee
LT
2287 if (re_allocated) {
2288 free(buf);
2289 return ret;
2290 }
aac17941
LT
2291 if (size)
2292 munmap(buf, size);
2293 return ret;
74400e71 2294}
ec1fcc16
JH
2295
2296int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2297{
2298 int fd;
2299 char *target;
dc49cd76 2300 size_t len;
ec1fcc16
JH
2301
2302 switch (st->st_mode & S_IFMT) {
2303 case S_IFREG:
2304 fd = open(path, O_RDONLY);
2305 if (fd < 0)
2306 return error("open(\"%s\"): %s", path,
2307 strerror(errno));
53bca91a 2308 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
ec1fcc16
JH
2309 return error("%s: failed to insert into database",
2310 path);
2311 break;
2312 case S_IFLNK:
dc49cd76
SP
2313 len = xsize_t(st->st_size);
2314 target = xmalloc(len + 1);
2315 if (readlink(path, target, len + 1) != st->st_size) {
ec1fcc16
JH
2316 char *errstr = strerror(errno);
2317 free(target);
2318 return error("readlink(\"%s\"): %s", path,
2319 errstr);
2320 }
abdc3fc8 2321 if (!write_object)
dc49cd76
SP
2322 hash_sha1_file(target, len, blob_type, sha1);
2323 else if (write_sha1_file(target, len, blob_type, sha1))
ec1fcc16
JH
2324 return error("%s: failed to insert into database",
2325 path);
2326 free(target);
2327 break;
2328 default:
2329 return error("%s: unsupported file type", path);
2330 }
2331 return 0;
2332}
a69e5429
JH
2333
2334int read_pack_header(int fd, struct pack_header *header)
2335{
2336 char *c = (char*)header;
2337 ssize_t remaining = sizeof(struct pack_header);
2338 do {
2339 ssize_t r = xread(fd, c, remaining);
2340 if (r <= 0)
2341 /* "eof before pack header was fully read" */
2342 return PH_ERROR_EOF;
2343 remaining -= r;
2344 c += r;
2345 } while (remaining > 0);
2346 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2347 /* "protocol error (pack signature mismatch detected)" */
2348 return PH_ERROR_PACK_SIGNATURE;
2349 if (!pack_version_ok(header->hdr_version))
2350 /* "protocol error (pack version unsupported)" */
2351 return PH_ERROR_PROTOCOL;
2352 return 0;
2353}