]>
Commit | Line | Data |
---|---|---|
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 | */ | |
9 | #include "cache.h" | |
10 | #include "delta.h" | |
11 | #include "pack.h" | |
12 | #include "blob.h" | |
13 | #include "commit.h" | |
14 | #include "tag.h" | |
15 | #include "tree.h" | |
16 | ||
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 | ||
25 | #ifdef NO_C99_FORMAT | |
26 | #define SZ_FMT "lu" | |
27 | #else | |
28 | #define SZ_FMT "zu" | |
29 | #endif | |
30 | ||
31 | const unsigned char null_sha1[20]; | |
32 | ||
33 | static unsigned int sha1_file_open_flag = O_NOATIME; | |
34 | ||
35 | signed 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 | }; | |
69 | ||
70 | int 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 | ||
83 | int safe_create_leading_directories(char *path) | |
84 | { | |
85 | char *pos = path; | |
86 | struct stat st; | |
87 | ||
88 | if (*pos == '/') | |
89 | pos++; | |
90 | ||
91 | while (pos) { | |
92 | pos = strchr(pos, '/'); | |
93 | if (!pos) | |
94 | break; | |
95 | *pos = 0; | |
96 | if (!stat(path, &st)) { | |
97 | /* path exists */ | |
98 | if (!S_ISDIR(st.st_mode)) { | |
99 | *pos = '/'; | |
100 | return -3; | |
101 | } | |
102 | } | |
103 | else if (mkdir(path, 0777)) { | |
104 | *pos = '/'; | |
105 | return -1; | |
106 | } | |
107 | else if (adjust_shared_perm(path)) { | |
108 | *pos = '/'; | |
109 | return -2; | |
110 | } | |
111 | *pos++ = '/'; | |
112 | } | |
113 | return 0; | |
114 | } | |
115 | ||
116 | char * sha1_to_hex(const unsigned char *sha1) | |
117 | { | |
118 | static int bufno; | |
119 | static char hexbuffer[4][50]; | |
120 | static const char hex[] = "0123456789abcdef"; | |
121 | char *buffer = hexbuffer[3 & ++bufno], *buf = buffer; | |
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 | } | |
129 | *buf = '\0'; | |
130 | ||
131 | return buffer; | |
132 | } | |
133 | ||
134 | static 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 | ||
146 | /* | |
147 | * NOTE! This returns a statically allocated buffer, so you have to be | |
148 | * careful about using it. Do a "xstrdup()" if you need to save the | |
149 | * filename. | |
150 | * | |
151 | * Also note that this returns the location for creating. Reading | |
152 | * SHA1 file can happen from any alternate directory listed in the | |
153 | * DB_ENVIRONMENT environment variable if it is not found in | |
154 | * the primary object database. | |
155 | */ | |
156 | char *sha1_file_name(const unsigned char *sha1) | |
157 | { | |
158 | static char *name, *base; | |
159 | ||
160 | if (!base) { | |
161 | const char *sha1_file_directory = get_object_directory(); | |
162 | int len = strlen(sha1_file_directory); | |
163 | base = xmalloc(len + 60); | |
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 | } | |
170 | fill_sha1_path(name, sha1); | |
171 | return base; | |
172 | } | |
173 | ||
174 | char *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 | ||
199 | char *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 | ||
224 | struct alternate_object_database *alt_odb_list; | |
225 | static struct alternate_object_database **alt_odb_tail; | |
226 | ||
227 | static void read_info_alternates(const char * alternates, int depth); | |
228 | ||
229 | /* | |
230 | * Prepare alternate object database registry. | |
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, | |
236 | * whose contents is similar to that environment variable but can be | |
237 | * LF separated. Its base points at a statically allocated buffer that | |
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. | |
243 | */ | |
244 | static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth) | |
245 | { | |
246 | struct stat st; | |
247 | const char *objdir = get_object_directory(); | |
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; | |
253 | int base_len = -1; | |
254 | ||
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 | ||
312 | static 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 | ||
323 | last = alt; | |
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 | } | |
332 | while (cp < ep && *cp != sep) | |
333 | cp++; | |
334 | if (last != cp) { | |
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); | |
341 | } | |
342 | } | |
343 | while (cp < ep && *cp == sep) | |
344 | cp++; | |
345 | last = cp; | |
346 | } | |
347 | } | |
348 | ||
349 | static void read_info_alternates(const char * relative_base, int depth) | |
350 | { | |
351 | char *map; | |
352 | struct stat st; | |
353 | char path[PATH_MAX]; | |
354 | int fd; | |
355 | ||
356 | sprintf(path, "%s/info/alternates", relative_base); | |
357 | fd = open(path, O_RDONLY); | |
358 | if (fd < 0) | |
359 | return; | |
360 | if (fstat(fd, &st) || (st.st_size == 0)) { | |
361 | close(fd); | |
362 | return; | |
363 | } | |
364 | map = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0); | |
365 | close(fd); | |
366 | ||
367 | link_alt_odb_entries(map, map + st.st_size, '\n', relative_base, depth); | |
368 | ||
369 | munmap(map, st.st_size); | |
370 | } | |
371 | ||
372 | void prepare_alt_odb(void) | |
373 | { | |
374 | const char *alt; | |
375 | ||
376 | alt = getenv(ALTERNATE_DB_ENVIRONMENT); | |
377 | if (!alt) alt = ""; | |
378 | ||
379 | if (alt_odb_tail) | |
380 | return; | |
381 | alt_odb_tail = &alt_odb_list; | |
382 | link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0); | |
383 | ||
384 | read_info_alternates(get_object_directory(), 0); | |
385 | } | |
386 | ||
387 | static char *find_sha1_file(const unsigned char *sha1, struct stat *st) | |
388 | { | |
389 | char *name = sha1_file_name(sha1); | |
390 | struct alternate_object_database *alt; | |
391 | ||
392 | if (!stat(name, st)) | |
393 | return name; | |
394 | prepare_alt_odb(); | |
395 | for (alt = alt_odb_list; alt; alt = alt->next) { | |
396 | name = alt->name; | |
397 | fill_sha1_path(name, sha1); | |
398 | if (!stat(alt->base, st)) | |
399 | return alt->base; | |
400 | } | |
401 | return NULL; | |
402 | } | |
403 | ||
404 | static unsigned int pack_used_ctr; | |
405 | static unsigned int pack_mmap_calls; | |
406 | static unsigned int peak_pack_open_windows; | |
407 | static unsigned int pack_open_windows; | |
408 | static size_t peak_pack_mapped; | |
409 | static size_t pack_mapped; | |
410 | struct packed_git *packed_git; | |
411 | ||
412 | void pack_report() | |
413 | { | |
414 | fprintf(stderr, | |
415 | "pack_report: getpagesize() = %10" SZ_FMT "\n" | |
416 | "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n" | |
417 | "pack_report: core.packedGitLimit = %10" SZ_FMT "\n", | |
418 | (size_t) getpagesize(), | |
419 | packed_git_window_size, | |
420 | packed_git_limit); | |
421 | fprintf(stderr, | |
422 | "pack_report: pack_used_ctr = %10u\n" | |
423 | "pack_report: pack_mmap_calls = %10u\n" | |
424 | "pack_report: pack_open_windows = %10u / %10u\n" | |
425 | "pack_report: pack_mapped = " | |
426 | "%10" SZ_FMT " / %10" SZ_FMT "\n", | |
427 | pack_used_ctr, | |
428 | pack_mmap_calls, | |
429 | pack_open_windows, peak_pack_open_windows, | |
430 | pack_mapped, peak_pack_mapped); | |
431 | } | |
432 | ||
433 | static int check_packed_git_idx(const char *path, unsigned long *idx_size_, | |
434 | void **idx_map_) | |
435 | { | |
436 | void *idx_map; | |
437 | uint32_t *index; | |
438 | unsigned long idx_size; | |
439 | int nr, i; | |
440 | int fd = open(path, O_RDONLY); | |
441 | struct stat st; | |
442 | if (fd < 0) | |
443 | return -1; | |
444 | if (fstat(fd, &st)) { | |
445 | close(fd); | |
446 | return -1; | |
447 | } | |
448 | idx_size = st.st_size; | |
449 | idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0); | |
450 | close(fd); | |
451 | ||
452 | index = idx_map; | |
453 | *idx_map_ = idx_map; | |
454 | *idx_size_ = idx_size; | |
455 | ||
456 | /* check index map */ | |
457 | if (idx_size < 4*256 + 20 + 20) | |
458 | return error("index file %s is too small", path); | |
459 | ||
460 | /* a future index format would start with this, as older git | |
461 | * binaries would fail the non-monotonic index check below. | |
462 | * give a nicer warning to the user if we can. | |
463 | */ | |
464 | if (index[0] == htonl(PACK_IDX_SIGNATURE)) | |
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); | |
469 | ||
470 | nr = 0; | |
471 | for (i = 0; i < 256; i++) { | |
472 | unsigned int n = ntohl(index[i]); | |
473 | if (n < nr) | |
474 | return error("non-monotonic index %s", path); | |
475 | nr = n; | |
476 | } | |
477 | ||
478 | /* | |
479 | * Total size: | |
480 | * - 256 index entries 4 bytes each | |
481 | * - 24-byte entries * nr (20-byte sha1 + 4-byte offset) | |
482 | * - 20-byte SHA1 of the packfile | |
483 | * - 20-byte SHA1 file checksum | |
484 | */ | |
485 | if (idx_size != 4*256 + nr * 24 + 20 + 20) | |
486 | return error("wrong index file size in %s", path); | |
487 | ||
488 | return 0; | |
489 | } | |
490 | ||
491 | static void scan_windows(struct packed_git *p, | |
492 | struct packed_git **lru_p, | |
493 | struct pack_window **lru_w, | |
494 | struct pack_window **lru_l) | |
495 | { | |
496 | struct pack_window *w, *w_l; | |
497 | ||
498 | for (w_l = NULL, w = p->windows; w; w = w->next) { | |
499 | if (!w->inuse_cnt) { | |
500 | if (!*lru_w || w->last_used < (*lru_w)->last_used) { | |
501 | *lru_p = p; | |
502 | *lru_w = w; | |
503 | *lru_l = w_l; | |
504 | } | |
505 | } | |
506 | w_l = w; | |
507 | } | |
508 | } | |
509 | ||
510 | static int unuse_one_window(struct packed_git *current) | |
511 | { | |
512 | struct packed_git *p, *lru_p = NULL; | |
513 | struct pack_window *lru_w = NULL, *lru_l = NULL; | |
514 | ||
515 | if (current) | |
516 | scan_windows(current, &lru_p, &lru_w, &lru_l); | |
517 | for (p = packed_git; p; p = p->next) | |
518 | scan_windows(p, &lru_p, &lru_w, &lru_l); | |
519 | if (lru_p) { | |
520 | munmap(lru_w->base, lru_w->len); | |
521 | pack_mapped -= lru_w->len; | |
522 | if (lru_l) | |
523 | lru_l->next = lru_w->next; | |
524 | else { | |
525 | lru_p->windows = lru_w->next; | |
526 | if (!lru_p->windows && lru_p != current) { | |
527 | close(lru_p->pack_fd); | |
528 | lru_p->pack_fd = -1; | |
529 | } | |
530 | } | |
531 | free(lru_w); | |
532 | pack_open_windows--; | |
533 | return 1; | |
534 | } | |
535 | return 0; | |
536 | } | |
537 | ||
538 | void release_pack_memory(size_t need) | |
539 | { | |
540 | size_t cur = pack_mapped; | |
541 | while (need >= (cur - pack_mapped) && unuse_one_window(NULL)) | |
542 | ; /* nothing */ | |
543 | } | |
544 | ||
545 | void unuse_pack(struct pack_window **w_cursor) | |
546 | { | |
547 | struct pack_window *w = *w_cursor; | |
548 | if (w) { | |
549 | w->inuse_cnt--; | |
550 | *w_cursor = NULL; | |
551 | } | |
552 | } | |
553 | ||
554 | /* | |
555 | * Do not call this directly as this leaks p->pack_fd on error return; | |
556 | * call open_packed_git() instead. | |
557 | */ | |
558 | static int open_packed_git_1(struct packed_git *p) | |
559 | { | |
560 | struct stat st; | |
561 | struct pack_header hdr; | |
562 | unsigned char sha1[20]; | |
563 | unsigned char *idx_sha1; | |
564 | long fd_flag; | |
565 | ||
566 | p->pack_fd = open(p->pack_name, O_RDONLY); | |
567 | if (p->pack_fd < 0 || fstat(p->pack_fd, &st)) | |
568 | return -1; | |
569 | ||
570 | /* If we created the struct before we had the pack we lack size. */ | |
571 | if (!p->pack_size) { | |
572 | if (!S_ISREG(st.st_mode)) | |
573 | return error("packfile %s not a regular file", p->pack_name); | |
574 | p->pack_size = st.st_size; | |
575 | } else if (p->pack_size != st.st_size) | |
576 | return error("packfile %s size changed", p->pack_name); | |
577 | ||
578 | /* We leave these file descriptors open with sliding mmap; | |
579 | * there is no point keeping them open across exec(), though. | |
580 | */ | |
581 | fd_flag = fcntl(p->pack_fd, F_GETFD, 0); | |
582 | if (fd_flag < 0) | |
583 | return error("cannot determine file descriptor flags"); | |
584 | fd_flag |= FD_CLOEXEC; | |
585 | if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1) | |
586 | return error("cannot set FD_CLOEXEC"); | |
587 | ||
588 | /* Verify we recognize this pack file format. */ | |
589 | if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr)) | |
590 | return error("file %s is far too short to be a packfile", p->pack_name); | |
591 | if (hdr.hdr_signature != htonl(PACK_SIGNATURE)) | |
592 | return error("file %s is not a GIT packfile", p->pack_name); | |
593 | if (!pack_version_ok(hdr.hdr_version)) | |
594 | return error("packfile %s is version %u and not supported" | |
595 | " (try upgrading GIT to a newer version)", | |
596 | p->pack_name, ntohl(hdr.hdr_version)); | |
597 | ||
598 | /* Verify the pack matches its index. */ | |
599 | if (num_packed_objects(p) != ntohl(hdr.hdr_entries)) | |
600 | return error("packfile %s claims to have %u objects" | |
601 | " while index size indicates %u objects", | |
602 | p->pack_name, ntohl(hdr.hdr_entries), | |
603 | num_packed_objects(p)); | |
604 | if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1) | |
605 | return error("end of packfile %s is unavailable", p->pack_name); | |
606 | if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1)) | |
607 | return error("packfile %s signature is unavailable", p->pack_name); | |
608 | idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40; | |
609 | if (hashcmp(sha1, idx_sha1)) | |
610 | return error("packfile %s does not match index", p->pack_name); | |
611 | return 0; | |
612 | } | |
613 | ||
614 | static int open_packed_git(struct packed_git *p) | |
615 | { | |
616 | if (!open_packed_git_1(p)) | |
617 | return 0; | |
618 | if (p->pack_fd != -1) { | |
619 | close(p->pack_fd); | |
620 | p->pack_fd = -1; | |
621 | } | |
622 | return -1; | |
623 | } | |
624 | ||
625 | static int in_window(struct pack_window *win, unsigned long offset) | |
626 | { | |
627 | /* We must promise at least 20 bytes (one hash) after the | |
628 | * offset is available from this window, otherwise the offset | |
629 | * is not actually in this window and a different window (which | |
630 | * has that one hash excess) must be used. This is to support | |
631 | * the object header and delta base parsing routines below. | |
632 | */ | |
633 | off_t win_off = win->offset; | |
634 | return win_off <= offset | |
635 | && (offset + 20) <= (win_off + win->len); | |
636 | } | |
637 | ||
638 | unsigned char* use_pack(struct packed_git *p, | |
639 | struct pack_window **w_cursor, | |
640 | unsigned long offset, | |
641 | unsigned int *left) | |
642 | { | |
643 | struct pack_window *win = *w_cursor; | |
644 | ||
645 | if (p->pack_fd == -1 && open_packed_git(p)) | |
646 | die("packfile %s cannot be accessed", p->pack_name); | |
647 | ||
648 | /* Since packfiles end in a hash of their content and its | |
649 | * pointless to ask for an offset into the middle of that | |
650 | * hash, and the in_window function above wouldn't match | |
651 | * don't allow an offset too close to the end of the file. | |
652 | */ | |
653 | if (offset > (p->pack_size - 20)) | |
654 | die("offset beyond end of packfile (truncated pack?)"); | |
655 | ||
656 | if (!win || !in_window(win, offset)) { | |
657 | if (win) | |
658 | win->inuse_cnt--; | |
659 | for (win = p->windows; win; win = win->next) { | |
660 | if (in_window(win, offset)) | |
661 | break; | |
662 | } | |
663 | if (!win) { | |
664 | size_t window_align = packed_git_window_size / 2; | |
665 | win = xcalloc(1, sizeof(*win)); | |
666 | win->offset = (offset / window_align) * window_align; | |
667 | win->len = p->pack_size - win->offset; | |
668 | if (win->len > packed_git_window_size) | |
669 | win->len = packed_git_window_size; | |
670 | pack_mapped += win->len; | |
671 | while (packed_git_limit < pack_mapped | |
672 | && unuse_one_window(p)) | |
673 | ; /* nothing */ | |
674 | win->base = xmmap(NULL, win->len, | |
675 | PROT_READ, MAP_PRIVATE, | |
676 | p->pack_fd, win->offset); | |
677 | if (win->base == MAP_FAILED) | |
678 | die("packfile %s cannot be mapped: %s", | |
679 | p->pack_name, | |
680 | strerror(errno)); | |
681 | pack_mmap_calls++; | |
682 | pack_open_windows++; | |
683 | if (pack_mapped > peak_pack_mapped) | |
684 | peak_pack_mapped = pack_mapped; | |
685 | if (pack_open_windows > peak_pack_open_windows) | |
686 | peak_pack_open_windows = pack_open_windows; | |
687 | win->next = p->windows; | |
688 | p->windows = win; | |
689 | } | |
690 | } | |
691 | if (win != *w_cursor) { | |
692 | win->last_used = pack_used_ctr++; | |
693 | win->inuse_cnt++; | |
694 | *w_cursor = win; | |
695 | } | |
696 | offset -= win->offset; | |
697 | if (left) | |
698 | *left = win->len - offset; | |
699 | return win->base + offset; | |
700 | } | |
701 | ||
702 | struct packed_git *add_packed_git(char *path, int path_len, int local) | |
703 | { | |
704 | struct stat st; | |
705 | struct packed_git *p; | |
706 | unsigned long idx_size; | |
707 | void *idx_map; | |
708 | unsigned char sha1[20]; | |
709 | ||
710 | if (check_packed_git_idx(path, &idx_size, &idx_map)) | |
711 | return NULL; | |
712 | ||
713 | /* do we have a corresponding .pack file? */ | |
714 | strcpy(path + path_len - 4, ".pack"); | |
715 | if (stat(path, &st) || !S_ISREG(st.st_mode)) { | |
716 | munmap(idx_map, idx_size); | |
717 | return NULL; | |
718 | } | |
719 | /* ok, it looks sane as far as we can check without | |
720 | * actually mapping the pack file. | |
721 | */ | |
722 | p = xmalloc(sizeof(*p) + path_len + 2); | |
723 | strcpy(p->pack_name, path); | |
724 | p->index_size = idx_size; | |
725 | p->pack_size = st.st_size; | |
726 | p->index_base = idx_map; | |
727 | p->next = NULL; | |
728 | p->windows = NULL; | |
729 | p->pack_fd = -1; | |
730 | p->pack_local = local; | |
731 | if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1)) | |
732 | hashcpy(p->sha1, sha1); | |
733 | return p; | |
734 | } | |
735 | ||
736 | struct packed_git *parse_pack_index(unsigned char *sha1) | |
737 | { | |
738 | char *path = sha1_pack_index_name(sha1); | |
739 | return parse_pack_index_file(sha1, path); | |
740 | } | |
741 | ||
742 | struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path) | |
743 | { | |
744 | struct packed_git *p; | |
745 | unsigned long idx_size; | |
746 | void *idx_map; | |
747 | char *path; | |
748 | ||
749 | if (check_packed_git_idx(idx_path, &idx_size, &idx_map)) | |
750 | return NULL; | |
751 | ||
752 | path = sha1_pack_name(sha1); | |
753 | ||
754 | p = xmalloc(sizeof(*p) + strlen(path) + 2); | |
755 | strcpy(p->pack_name, path); | |
756 | p->index_size = idx_size; | |
757 | p->pack_size = 0; | |
758 | p->index_base = idx_map; | |
759 | p->next = NULL; | |
760 | p->windows = NULL; | |
761 | p->pack_fd = -1; | |
762 | hashcpy(p->sha1, sha1); | |
763 | return p; | |
764 | } | |
765 | ||
766 | void install_packed_git(struct packed_git *pack) | |
767 | { | |
768 | pack->next = packed_git; | |
769 | packed_git = pack; | |
770 | } | |
771 | ||
772 | static void prepare_packed_git_one(char *objdir, int local) | |
773 | { | |
774 | char path[PATH_MAX]; | |
775 | int len; | |
776 | DIR *dir; | |
777 | struct dirent *de; | |
778 | ||
779 | sprintf(path, "%s/pack", objdir); | |
780 | len = strlen(path); | |
781 | dir = opendir(path); | |
782 | if (!dir) { | |
783 | if (errno != ENOENT) | |
784 | error("unable to open object pack directory: %s: %s", | |
785 | path, strerror(errno)); | |
786 | return; | |
787 | } | |
788 | path[len++] = '/'; | |
789 | while ((de = readdir(dir)) != NULL) { | |
790 | int namelen = strlen(de->d_name); | |
791 | struct packed_git *p; | |
792 | ||
793 | if (!has_extension(de->d_name, ".idx")) | |
794 | continue; | |
795 | ||
796 | /* Don't reopen a pack we already have. */ | |
797 | strcpy(path + len, de->d_name); | |
798 | for (p = packed_git; p; p = p->next) { | |
799 | if (!memcmp(path, p->pack_name, len + namelen - 4)) | |
800 | break; | |
801 | } | |
802 | if (p) | |
803 | continue; | |
804 | /* See if it really is a valid .idx file with corresponding | |
805 | * .pack file that we can map. | |
806 | */ | |
807 | p = add_packed_git(path, len + namelen, local); | |
808 | if (!p) | |
809 | continue; | |
810 | install_packed_git(p); | |
811 | } | |
812 | closedir(dir); | |
813 | } | |
814 | ||
815 | static int prepare_packed_git_run_once = 0; | |
816 | void prepare_packed_git(void) | |
817 | { | |
818 | struct alternate_object_database *alt; | |
819 | ||
820 | if (prepare_packed_git_run_once) | |
821 | return; | |
822 | prepare_packed_git_one(get_object_directory(), 1); | |
823 | prepare_alt_odb(); | |
824 | for (alt = alt_odb_list; alt; alt = alt->next) { | |
825 | alt->name[-1] = 0; | |
826 | prepare_packed_git_one(alt->base, 0); | |
827 | alt->name[-1] = '/'; | |
828 | } | |
829 | prepare_packed_git_run_once = 1; | |
830 | } | |
831 | ||
832 | void reprepare_packed_git(void) | |
833 | { | |
834 | prepare_packed_git_run_once = 0; | |
835 | prepare_packed_git(); | |
836 | } | |
837 | ||
838 | int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type) | |
839 | { | |
840 | unsigned char real_sha1[20]; | |
841 | hash_sha1_file(map, size, type, real_sha1); | |
842 | return hashcmp(sha1, real_sha1) ? -1 : 0; | |
843 | } | |
844 | ||
845 | void *map_sha1_file(const unsigned char *sha1, unsigned long *size) | |
846 | { | |
847 | struct stat st; | |
848 | void *map; | |
849 | int fd; | |
850 | char *filename = find_sha1_file(sha1, &st); | |
851 | ||
852 | if (!filename) { | |
853 | return NULL; | |
854 | } | |
855 | ||
856 | fd = open(filename, O_RDONLY | sha1_file_open_flag); | |
857 | if (fd < 0) { | |
858 | /* See if it works without O_NOATIME */ | |
859 | switch (sha1_file_open_flag) { | |
860 | default: | |
861 | fd = open(filename, O_RDONLY); | |
862 | if (fd >= 0) | |
863 | break; | |
864 | /* Fallthrough */ | |
865 | case 0: | |
866 | return NULL; | |
867 | } | |
868 | ||
869 | /* If it failed once, it will probably fail again. | |
870 | * Stop using O_NOATIME | |
871 | */ | |
872 | sha1_file_open_flag = 0; | |
873 | } | |
874 | map = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0); | |
875 | close(fd); | |
876 | *size = st.st_size; | |
877 | return map; | |
878 | } | |
879 | ||
880 | int legacy_loose_object(unsigned char *map) | |
881 | { | |
882 | unsigned int word; | |
883 | ||
884 | /* | |
885 | * Is it a zlib-compressed buffer? If so, the first byte | |
886 | * must be 0x78 (15-bit window size, deflated), and the | |
887 | * first 16-bit word is evenly divisible by 31 | |
888 | */ | |
889 | word = (map[0] << 8) + map[1]; | |
890 | if (map[0] == 0x78 && !(word % 31)) | |
891 | return 1; | |
892 | else | |
893 | return 0; | |
894 | } | |
895 | ||
896 | unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep) | |
897 | { | |
898 | unsigned shift; | |
899 | unsigned char c; | |
900 | unsigned long size; | |
901 | unsigned long used = 0; | |
902 | ||
903 | c = buf[used++]; | |
904 | *type = (c >> 4) & 7; | |
905 | size = c & 15; | |
906 | shift = 4; | |
907 | while (c & 0x80) { | |
908 | if (len <= used) | |
909 | return 0; | |
910 | if (sizeof(long) * 8 <= shift) | |
911 | return 0; | |
912 | c = buf[used++]; | |
913 | size += (c & 0x7f) << shift; | |
914 | shift += 7; | |
915 | } | |
916 | *sizep = size; | |
917 | return used; | |
918 | } | |
919 | ||
920 | static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz) | |
921 | { | |
922 | unsigned long size, used; | |
923 | static const char valid_loose_object_type[8] = { | |
924 | 0, /* OBJ_EXT */ | |
925 | 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */ | |
926 | 0, /* "delta" and others are invalid in a loose object */ | |
927 | }; | |
928 | enum object_type type; | |
929 | ||
930 | /* Get the data stream */ | |
931 | memset(stream, 0, sizeof(*stream)); | |
932 | stream->next_in = map; | |
933 | stream->avail_in = mapsize; | |
934 | stream->next_out = buffer; | |
935 | stream->avail_out = bufsiz; | |
936 | ||
937 | if (legacy_loose_object(map)) { | |
938 | inflateInit(stream); | |
939 | return inflate(stream, 0); | |
940 | } | |
941 | ||
942 | used = unpack_object_header_gently(map, mapsize, &type, &size); | |
943 | if (!used || !valid_loose_object_type[type]) | |
944 | return -1; | |
945 | map += used; | |
946 | mapsize -= used; | |
947 | ||
948 | /* Set up the stream for the rest.. */ | |
949 | stream->next_in = map; | |
950 | stream->avail_in = mapsize; | |
951 | inflateInit(stream); | |
952 | ||
953 | /* And generate the fake traditional header */ | |
954 | stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu", | |
955 | type_names[type], size); | |
956 | return 0; | |
957 | } | |
958 | ||
959 | static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size) | |
960 | { | |
961 | int bytes = strlen(buffer) + 1; | |
962 | unsigned char *buf = xmalloc(1+size); | |
963 | unsigned long n; | |
964 | ||
965 | n = stream->total_out - bytes; | |
966 | if (n > size) | |
967 | n = size; | |
968 | memcpy(buf, (char *) buffer + bytes, n); | |
969 | bytes = n; | |
970 | if (bytes < size) { | |
971 | stream->next_out = buf + bytes; | |
972 | stream->avail_out = size - bytes; | |
973 | while (inflate(stream, Z_FINISH) == Z_OK) | |
974 | /* nothing */; | |
975 | } | |
976 | buf[size] = 0; | |
977 | inflateEnd(stream); | |
978 | return buf; | |
979 | } | |
980 | ||
981 | /* | |
982 | * We used to just use "sscanf()", but that's actually way | |
983 | * too permissive for what we want to check. So do an anal | |
984 | * object header parse by hand. | |
985 | */ | |
986 | static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep) | |
987 | { | |
988 | int i; | |
989 | unsigned long size; | |
990 | ||
991 | /* | |
992 | * The type can be at most ten bytes (including the | |
993 | * terminating '\0' that we add), and is followed by | |
994 | * a space. | |
995 | */ | |
996 | i = 10; | |
997 | for (;;) { | |
998 | char c = *hdr++; | |
999 | if (c == ' ') | |
1000 | break; | |
1001 | if (!--i) | |
1002 | return -1; | |
1003 | *type++ = c; | |
1004 | } | |
1005 | *type = 0; | |
1006 | ||
1007 | /* | |
1008 | * The length must follow immediately, and be in canonical | |
1009 | * decimal format (ie "010" is not valid). | |
1010 | */ | |
1011 | size = *hdr++ - '0'; | |
1012 | if (size > 9) | |
1013 | return -1; | |
1014 | if (size) { | |
1015 | for (;;) { | |
1016 | unsigned long c = *hdr - '0'; | |
1017 | if (c > 9) | |
1018 | break; | |
1019 | hdr++; | |
1020 | size = size * 10 + c; | |
1021 | } | |
1022 | } | |
1023 | *sizep = size; | |
1024 | ||
1025 | /* | |
1026 | * The length must be followed by a zero byte | |
1027 | */ | |
1028 | return *hdr ? -1 : 0; | |
1029 | } | |
1030 | ||
1031 | void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size) | |
1032 | { | |
1033 | int ret; | |
1034 | z_stream stream; | |
1035 | char hdr[8192]; | |
1036 | ||
1037 | ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)); | |
1038 | if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0) | |
1039 | return NULL; | |
1040 | ||
1041 | return unpack_sha1_rest(&stream, hdr, *size); | |
1042 | } | |
1043 | ||
1044 | static unsigned long get_delta_base(struct packed_git *p, | |
1045 | struct pack_window **w_curs, | |
1046 | unsigned long offset, | |
1047 | enum object_type kind, | |
1048 | unsigned long delta_obj_offset, | |
1049 | unsigned long *base_obj_offset) | |
1050 | { | |
1051 | unsigned char *base_info = use_pack(p, w_curs, offset, NULL); | |
1052 | unsigned long base_offset; | |
1053 | ||
1054 | /* use_pack() assured us we have [base_info, base_info + 20) | |
1055 | * as a range that we can look at without walking off the | |
1056 | * end of the mapped window. Its actually the hash size | |
1057 | * that is assured. An OFS_DELTA longer than the hash size | |
1058 | * is stupid, as then a REF_DELTA would be smaller to store. | |
1059 | */ | |
1060 | if (kind == OBJ_OFS_DELTA) { | |
1061 | unsigned used = 0; | |
1062 | unsigned char c = base_info[used++]; | |
1063 | base_offset = c & 127; | |
1064 | while (c & 128) { | |
1065 | base_offset += 1; | |
1066 | if (!base_offset || base_offset & ~(~0UL >> 7)) | |
1067 | die("offset value overflow for delta base object"); | |
1068 | c = base_info[used++]; | |
1069 | base_offset = (base_offset << 7) + (c & 127); | |
1070 | } | |
1071 | base_offset = delta_obj_offset - base_offset; | |
1072 | if (base_offset >= delta_obj_offset) | |
1073 | die("delta base offset out of bound"); | |
1074 | offset += used; | |
1075 | } else if (kind == OBJ_REF_DELTA) { | |
1076 | /* The base entry _must_ be in the same pack */ | |
1077 | base_offset = find_pack_entry_one(base_info, p); | |
1078 | if (!base_offset) | |
1079 | die("failed to find delta-pack base object %s", | |
1080 | sha1_to_hex(base_info)); | |
1081 | offset += 20; | |
1082 | } else | |
1083 | die("I am totally screwed"); | |
1084 | *base_obj_offset = base_offset; | |
1085 | return offset; | |
1086 | } | |
1087 | ||
1088 | /* forward declaration for a mutually recursive function */ | |
1089 | static int packed_object_info(struct packed_git *p, unsigned long offset, | |
1090 | char *type, unsigned long *sizep); | |
1091 | ||
1092 | static int packed_delta_info(struct packed_git *p, | |
1093 | struct pack_window **w_curs, | |
1094 | unsigned long offset, | |
1095 | enum object_type kind, | |
1096 | unsigned long obj_offset, | |
1097 | char *type, | |
1098 | unsigned long *sizep) | |
1099 | { | |
1100 | unsigned long base_offset; | |
1101 | ||
1102 | offset = get_delta_base(p, w_curs, offset, kind, | |
1103 | obj_offset, &base_offset); | |
1104 | ||
1105 | /* We choose to only get the type of the base object and | |
1106 | * ignore potentially corrupt pack file that expects the delta | |
1107 | * based on a base with a wrong size. This saves tons of | |
1108 | * inflate() calls. | |
1109 | */ | |
1110 | if (packed_object_info(p, base_offset, type, NULL)) | |
1111 | die("cannot get info for delta-pack base"); | |
1112 | ||
1113 | if (sizep) { | |
1114 | const unsigned char *data; | |
1115 | unsigned char delta_head[20], *in; | |
1116 | unsigned long result_size; | |
1117 | z_stream stream; | |
1118 | int st; | |
1119 | ||
1120 | memset(&stream, 0, sizeof(stream)); | |
1121 | stream.next_out = delta_head; | |
1122 | stream.avail_out = sizeof(delta_head); | |
1123 | ||
1124 | inflateInit(&stream); | |
1125 | do { | |
1126 | in = use_pack(p, w_curs, offset, &stream.avail_in); | |
1127 | stream.next_in = in; | |
1128 | st = inflate(&stream, Z_FINISH); | |
1129 | offset += stream.next_in - in; | |
1130 | } while ((st == Z_OK || st == Z_BUF_ERROR) | |
1131 | && stream.total_out < sizeof(delta_head)); | |
1132 | inflateEnd(&stream); | |
1133 | if ((st != Z_STREAM_END) && | |
1134 | stream.total_out != sizeof(delta_head)) | |
1135 | die("delta data unpack-initial failed"); | |
1136 | ||
1137 | /* Examine the initial part of the delta to figure out | |
1138 | * the result size. | |
1139 | */ | |
1140 | data = delta_head; | |
1141 | ||
1142 | /* ignore base size */ | |
1143 | get_delta_hdr_size(&data, delta_head+sizeof(delta_head)); | |
1144 | ||
1145 | /* Read the result size */ | |
1146 | result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head)); | |
1147 | *sizep = result_size; | |
1148 | } | |
1149 | return 0; | |
1150 | } | |
1151 | ||
1152 | static unsigned long unpack_object_header(struct packed_git *p, | |
1153 | struct pack_window **w_curs, | |
1154 | unsigned long offset, | |
1155 | enum object_type *type, | |
1156 | unsigned long *sizep) | |
1157 | { | |
1158 | unsigned char *base; | |
1159 | unsigned int left; | |
1160 | unsigned long used; | |
1161 | ||
1162 | /* use_pack() assures us we have [base, base + 20) available | |
1163 | * as a range that we can look at at. (Its actually the hash | |
1164 | * size that is assured.) With our object header encoding | |
1165 | * the maximum deflated object size is 2^137, which is just | |
1166 | * insane, so we know won't exceed what we have been given. | |
1167 | */ | |
1168 | base = use_pack(p, w_curs, offset, &left); | |
1169 | used = unpack_object_header_gently(base, left, type, sizep); | |
1170 | if (!used) | |
1171 | die("object offset outside of pack file"); | |
1172 | ||
1173 | return offset + used; | |
1174 | } | |
1175 | ||
1176 | void packed_object_info_detail(struct packed_git *p, | |
1177 | unsigned long offset, | |
1178 | char *type, | |
1179 | unsigned long *size, | |
1180 | unsigned long *store_size, | |
1181 | unsigned int *delta_chain_length, | |
1182 | unsigned char *base_sha1) | |
1183 | { | |
1184 | struct pack_window *w_curs = NULL; | |
1185 | unsigned long obj_offset, val; | |
1186 | unsigned char *next_sha1; | |
1187 | enum object_type kind; | |
1188 | ||
1189 | *delta_chain_length = 0; | |
1190 | obj_offset = offset; | |
1191 | offset = unpack_object_header(p, &w_curs, offset, &kind, size); | |
1192 | ||
1193 | for (;;) { | |
1194 | switch (kind) { | |
1195 | default: | |
1196 | die("pack %s contains unknown object type %d", | |
1197 | p->pack_name, kind); | |
1198 | case OBJ_COMMIT: | |
1199 | case OBJ_TREE: | |
1200 | case OBJ_BLOB: | |
1201 | case OBJ_TAG: | |
1202 | strcpy(type, type_names[kind]); | |
1203 | *store_size = 0; /* notyet */ | |
1204 | unuse_pack(&w_curs); | |
1205 | return; | |
1206 | case OBJ_OFS_DELTA: | |
1207 | get_delta_base(p, &w_curs, offset, kind, | |
1208 | obj_offset, &offset); | |
1209 | if (*delta_chain_length == 0) { | |
1210 | /* TODO: find base_sha1 as pointed by offset */ | |
1211 | } | |
1212 | break; | |
1213 | case OBJ_REF_DELTA: | |
1214 | next_sha1 = use_pack(p, &w_curs, offset, NULL); | |
1215 | if (*delta_chain_length == 0) | |
1216 | hashcpy(base_sha1, next_sha1); | |
1217 | offset = find_pack_entry_one(next_sha1, p); | |
1218 | break; | |
1219 | } | |
1220 | obj_offset = offset; | |
1221 | offset = unpack_object_header(p, &w_curs, offset, &kind, &val); | |
1222 | (*delta_chain_length)++; | |
1223 | } | |
1224 | } | |
1225 | ||
1226 | static int packed_object_info(struct packed_git *p, unsigned long offset, | |
1227 | char *type, unsigned long *sizep) | |
1228 | { | |
1229 | struct pack_window *w_curs = NULL; | |
1230 | unsigned long size, obj_offset = offset; | |
1231 | enum object_type kind; | |
1232 | int r; | |
1233 | ||
1234 | offset = unpack_object_header(p, &w_curs, offset, &kind, &size); | |
1235 | ||
1236 | switch (kind) { | |
1237 | case OBJ_OFS_DELTA: | |
1238 | case OBJ_REF_DELTA: | |
1239 | r = packed_delta_info(p, &w_curs, offset, kind, | |
1240 | obj_offset, type, sizep); | |
1241 | unuse_pack(&w_curs); | |
1242 | return r; | |
1243 | case OBJ_COMMIT: | |
1244 | case OBJ_TREE: | |
1245 | case OBJ_BLOB: | |
1246 | case OBJ_TAG: | |
1247 | strcpy(type, type_names[kind]); | |
1248 | unuse_pack(&w_curs); | |
1249 | break; | |
1250 | default: | |
1251 | die("pack %s contains unknown object type %d", | |
1252 | p->pack_name, kind); | |
1253 | } | |
1254 | if (sizep) | |
1255 | *sizep = size; | |
1256 | return 0; | |
1257 | } | |
1258 | ||
1259 | static void *unpack_compressed_entry(struct packed_git *p, | |
1260 | struct pack_window **w_curs, | |
1261 | unsigned long offset, | |
1262 | unsigned long size) | |
1263 | { | |
1264 | int st; | |
1265 | z_stream stream; | |
1266 | unsigned char *buffer, *in; | |
1267 | ||
1268 | buffer = xmalloc(size + 1); | |
1269 | buffer[size] = 0; | |
1270 | memset(&stream, 0, sizeof(stream)); | |
1271 | stream.next_out = buffer; | |
1272 | stream.avail_out = size; | |
1273 | ||
1274 | inflateInit(&stream); | |
1275 | do { | |
1276 | in = use_pack(p, w_curs, offset, &stream.avail_in); | |
1277 | stream.next_in = in; | |
1278 | st = inflate(&stream, Z_FINISH); | |
1279 | offset += stream.next_in - in; | |
1280 | } while (st == Z_OK || st == Z_BUF_ERROR); | |
1281 | inflateEnd(&stream); | |
1282 | if ((st != Z_STREAM_END) || stream.total_out != size) { | |
1283 | free(buffer); | |
1284 | return NULL; | |
1285 | } | |
1286 | ||
1287 | return buffer; | |
1288 | } | |
1289 | ||
1290 | static void *unpack_delta_entry(struct packed_git *p, | |
1291 | struct pack_window **w_curs, | |
1292 | unsigned long offset, | |
1293 | unsigned long delta_size, | |
1294 | enum object_type kind, | |
1295 | unsigned long obj_offset, | |
1296 | char *type, | |
1297 | unsigned long *sizep) | |
1298 | { | |
1299 | void *delta_data, *result, *base; | |
1300 | unsigned long result_size, base_size, base_offset; | |
1301 | ||
1302 | offset = get_delta_base(p, w_curs, offset, kind, | |
1303 | obj_offset, &base_offset); | |
1304 | base = unpack_entry(p, base_offset, type, &base_size); | |
1305 | if (!base) | |
1306 | die("failed to read delta base object at %lu from %s", | |
1307 | base_offset, p->pack_name); | |
1308 | ||
1309 | delta_data = unpack_compressed_entry(p, w_curs, offset, delta_size); | |
1310 | result = patch_delta(base, base_size, | |
1311 | delta_data, delta_size, | |
1312 | &result_size); | |
1313 | if (!result) | |
1314 | die("failed to apply delta"); | |
1315 | free(delta_data); | |
1316 | free(base); | |
1317 | *sizep = result_size; | |
1318 | return result; | |
1319 | } | |
1320 | ||
1321 | void *unpack_entry(struct packed_git *p, unsigned long offset, | |
1322 | char *type, unsigned long *sizep) | |
1323 | { | |
1324 | struct pack_window *w_curs = NULL; | |
1325 | unsigned long size, obj_offset = offset; | |
1326 | enum object_type kind; | |
1327 | void *retval; | |
1328 | ||
1329 | offset = unpack_object_header(p, &w_curs, offset, &kind, &size); | |
1330 | switch (kind) { | |
1331 | case OBJ_OFS_DELTA: | |
1332 | case OBJ_REF_DELTA: | |
1333 | retval = unpack_delta_entry(p, &w_curs, offset, size, | |
1334 | kind, obj_offset, type, sizep); | |
1335 | break; | |
1336 | case OBJ_COMMIT: | |
1337 | case OBJ_TREE: | |
1338 | case OBJ_BLOB: | |
1339 | case OBJ_TAG: | |
1340 | strcpy(type, type_names[kind]); | |
1341 | *sizep = size; | |
1342 | retval = unpack_compressed_entry(p, &w_curs, offset, size); | |
1343 | break; | |
1344 | default: | |
1345 | die("unknown object type %i in %s", kind, p->pack_name); | |
1346 | } | |
1347 | unuse_pack(&w_curs); | |
1348 | return retval; | |
1349 | } | |
1350 | ||
1351 | int num_packed_objects(const struct packed_git *p) | |
1352 | { | |
1353 | /* See check_packed_git_idx() */ | |
1354 | return (p->index_size - 20 - 20 - 4*256) / 24; | |
1355 | } | |
1356 | ||
1357 | int nth_packed_object_sha1(const struct packed_git *p, int n, | |
1358 | unsigned char* sha1) | |
1359 | { | |
1360 | void *index = p->index_base + 256; | |
1361 | if (n < 0 || num_packed_objects(p) <= n) | |
1362 | return -1; | |
1363 | hashcpy(sha1, (unsigned char *) index + (24 * n) + 4); | |
1364 | return 0; | |
1365 | } | |
1366 | ||
1367 | unsigned long find_pack_entry_one(const unsigned char *sha1, | |
1368 | struct packed_git *p) | |
1369 | { | |
1370 | uint32_t *level1_ofs = p->index_base; | |
1371 | int hi = ntohl(level1_ofs[*sha1]); | |
1372 | int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1])); | |
1373 | void *index = p->index_base + 256; | |
1374 | ||
1375 | do { | |
1376 | int mi = (lo + hi) / 2; | |
1377 | int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1); | |
1378 | if (!cmp) | |
1379 | return ntohl(*((uint32_t *)((char *)index + (24 * mi)))); | |
1380 | if (cmp > 0) | |
1381 | hi = mi; | |
1382 | else | |
1383 | lo = mi+1; | |
1384 | } while (lo < hi); | |
1385 | return 0; | |
1386 | } | |
1387 | ||
1388 | static int matches_pack_name(struct packed_git *p, const char *ig) | |
1389 | { | |
1390 | const char *last_c, *c; | |
1391 | ||
1392 | if (!strcmp(p->pack_name, ig)) | |
1393 | return 0; | |
1394 | ||
1395 | for (c = p->pack_name, last_c = c; *c;) | |
1396 | if (*c == '/') | |
1397 | last_c = ++c; | |
1398 | else | |
1399 | ++c; | |
1400 | if (!strcmp(last_c, ig)) | |
1401 | return 0; | |
1402 | ||
1403 | return 1; | |
1404 | } | |
1405 | ||
1406 | static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed) | |
1407 | { | |
1408 | struct packed_git *p; | |
1409 | unsigned long offset; | |
1410 | ||
1411 | prepare_packed_git(); | |
1412 | ||
1413 | for (p = packed_git; p; p = p->next) { | |
1414 | if (ignore_packed) { | |
1415 | const char **ig; | |
1416 | for (ig = ignore_packed; *ig; ig++) | |
1417 | if (!matches_pack_name(p, *ig)) | |
1418 | break; | |
1419 | if (*ig) | |
1420 | continue; | |
1421 | } | |
1422 | offset = find_pack_entry_one(sha1, p); | |
1423 | if (offset) { | |
1424 | /* | |
1425 | * We are about to tell the caller where they can | |
1426 | * locate the requested object. We better make | |
1427 | * sure the packfile is still here and can be | |
1428 | * accessed before supplying that answer, as | |
1429 | * it may have been deleted since the index | |
1430 | * was loaded! | |
1431 | */ | |
1432 | if (p->pack_fd == -1 && open_packed_git(p)) { | |
1433 | error("packfile %s cannot be accessed", p->pack_name); | |
1434 | continue; | |
1435 | } | |
1436 | e->offset = offset; | |
1437 | e->p = p; | |
1438 | hashcpy(e->sha1, sha1); | |
1439 | return 1; | |
1440 | } | |
1441 | } | |
1442 | return 0; | |
1443 | } | |
1444 | ||
1445 | struct packed_git *find_sha1_pack(const unsigned char *sha1, | |
1446 | struct packed_git *packs) | |
1447 | { | |
1448 | struct packed_git *p; | |
1449 | ||
1450 | for (p = packs; p; p = p->next) { | |
1451 | if (find_pack_entry_one(sha1, p)) | |
1452 | return p; | |
1453 | } | |
1454 | return NULL; | |
1455 | ||
1456 | } | |
1457 | ||
1458 | static int sha1_loose_object_info(const unsigned char *sha1, char *type, unsigned long *sizep) | |
1459 | { | |
1460 | int status; | |
1461 | unsigned long mapsize, size; | |
1462 | void *map; | |
1463 | z_stream stream; | |
1464 | char hdr[128]; | |
1465 | ||
1466 | map = map_sha1_file(sha1, &mapsize); | |
1467 | if (!map) | |
1468 | return error("unable to find %s", sha1_to_hex(sha1)); | |
1469 | if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) | |
1470 | status = error("unable to unpack %s header", | |
1471 | sha1_to_hex(sha1)); | |
1472 | if (parse_sha1_header(hdr, type, &size) < 0) | |
1473 | status = error("unable to parse %s header", sha1_to_hex(sha1)); | |
1474 | else { | |
1475 | status = 0; | |
1476 | if (sizep) | |
1477 | *sizep = size; | |
1478 | } | |
1479 | inflateEnd(&stream); | |
1480 | munmap(map, mapsize); | |
1481 | return status; | |
1482 | } | |
1483 | ||
1484 | int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep) | |
1485 | { | |
1486 | struct pack_entry e; | |
1487 | ||
1488 | if (!find_pack_entry(sha1, &e, NULL)) { | |
1489 | reprepare_packed_git(); | |
1490 | if (!find_pack_entry(sha1, &e, NULL)) | |
1491 | return sha1_loose_object_info(sha1, type, sizep); | |
1492 | } | |
1493 | return packed_object_info(e.p, e.offset, type, sizep); | |
1494 | } | |
1495 | ||
1496 | static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size) | |
1497 | { | |
1498 | struct pack_entry e; | |
1499 | ||
1500 | if (!find_pack_entry(sha1, &e, NULL)) | |
1501 | return NULL; | |
1502 | else | |
1503 | return unpack_entry(e.p, e.offset, type, size); | |
1504 | } | |
1505 | ||
1506 | /* | |
1507 | * This is meant to hold a *small* number of objects that you would | |
1508 | * want read_sha1_file() to be able to return, but yet you do not want | |
1509 | * to write them into the object store (e.g. a browse-only | |
1510 | * application). | |
1511 | */ | |
1512 | static struct cached_object { | |
1513 | unsigned char sha1[20]; | |
1514 | const char *type; | |
1515 | void *buf; | |
1516 | unsigned long size; | |
1517 | } *cached_objects; | |
1518 | static int cached_object_nr, cached_object_alloc; | |
1519 | ||
1520 | static struct cached_object *find_cached_object(const unsigned char *sha1) | |
1521 | { | |
1522 | int i; | |
1523 | struct cached_object *co = cached_objects; | |
1524 | ||
1525 | for (i = 0; i < cached_object_nr; i++, co++) { | |
1526 | if (!hashcmp(co->sha1, sha1)) | |
1527 | return co; | |
1528 | } | |
1529 | return NULL; | |
1530 | } | |
1531 | ||
1532 | int pretend_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *sha1) | |
1533 | { | |
1534 | struct cached_object *co; | |
1535 | ||
1536 | hash_sha1_file(buf, len, type, sha1); | |
1537 | if (has_sha1_file(sha1) || find_cached_object(sha1)) | |
1538 | return 0; | |
1539 | if (cached_object_alloc <= cached_object_nr) { | |
1540 | cached_object_alloc = alloc_nr(cached_object_alloc); | |
1541 | cached_objects = xrealloc(cached_objects, | |
1542 | sizeof(*cached_objects) * | |
1543 | cached_object_alloc); | |
1544 | } | |
1545 | co = &cached_objects[cached_object_nr++]; | |
1546 | co->size = len; | |
1547 | co->type = strdup(type); | |
1548 | hashcpy(co->sha1, sha1); | |
1549 | return 0; | |
1550 | } | |
1551 | ||
1552 | void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size) | |
1553 | { | |
1554 | unsigned long mapsize; | |
1555 | void *map, *buf; | |
1556 | struct cached_object *co; | |
1557 | ||
1558 | co = find_cached_object(sha1); | |
1559 | if (co) { | |
1560 | buf = xmalloc(co->size + 1); | |
1561 | memcpy(buf, co->buf, co->size); | |
1562 | ((char*)buf)[co->size] = 0; | |
1563 | strcpy(type, co->type); | |
1564 | *size = co->size; | |
1565 | return buf; | |
1566 | } | |
1567 | ||
1568 | buf = read_packed_sha1(sha1, type, size); | |
1569 | if (buf) | |
1570 | return buf; | |
1571 | map = map_sha1_file(sha1, &mapsize); | |
1572 | if (map) { | |
1573 | buf = unpack_sha1_file(map, mapsize, type, size); | |
1574 | munmap(map, mapsize); | |
1575 | return buf; | |
1576 | } | |
1577 | reprepare_packed_git(); | |
1578 | return read_packed_sha1(sha1, type, size); | |
1579 | } | |
1580 | ||
1581 | void *read_object_with_reference(const unsigned char *sha1, | |
1582 | const char *required_type, | |
1583 | unsigned long *size, | |
1584 | unsigned char *actual_sha1_return) | |
1585 | { | |
1586 | char type[20]; | |
1587 | void *buffer; | |
1588 | unsigned long isize; | |
1589 | unsigned char actual_sha1[20]; | |
1590 | ||
1591 | hashcpy(actual_sha1, sha1); | |
1592 | while (1) { | |
1593 | int ref_length = -1; | |
1594 | const char *ref_type = NULL; | |
1595 | ||
1596 | buffer = read_sha1_file(actual_sha1, type, &isize); | |
1597 | if (!buffer) | |
1598 | return NULL; | |
1599 | if (!strcmp(type, required_type)) { | |
1600 | *size = isize; | |
1601 | if (actual_sha1_return) | |
1602 | hashcpy(actual_sha1_return, actual_sha1); | |
1603 | return buffer; | |
1604 | } | |
1605 | /* Handle references */ | |
1606 | else if (!strcmp(type, commit_type)) | |
1607 | ref_type = "tree "; | |
1608 | else if (!strcmp(type, tag_type)) | |
1609 | ref_type = "object "; | |
1610 | else { | |
1611 | free(buffer); | |
1612 | return NULL; | |
1613 | } | |
1614 | ref_length = strlen(ref_type); | |
1615 | ||
1616 | if (memcmp(buffer, ref_type, ref_length) || | |
1617 | get_sha1_hex((char *) buffer + ref_length, actual_sha1)) { | |
1618 | free(buffer); | |
1619 | return NULL; | |
1620 | } | |
1621 | free(buffer); | |
1622 | /* Now we have the ID of the referred-to object in | |
1623 | * actual_sha1. Check again. */ | |
1624 | } | |
1625 | } | |
1626 | ||
1627 | static void write_sha1_file_prepare(void *buf, unsigned long len, | |
1628 | const char *type, unsigned char *sha1, | |
1629 | unsigned char *hdr, int *hdrlen) | |
1630 | { | |
1631 | SHA_CTX c; | |
1632 | ||
1633 | /* Generate the header */ | |
1634 | *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1; | |
1635 | ||
1636 | /* Sha1.. */ | |
1637 | SHA1_Init(&c); | |
1638 | SHA1_Update(&c, hdr, *hdrlen); | |
1639 | SHA1_Update(&c, buf, len); | |
1640 | SHA1_Final(sha1, &c); | |
1641 | } | |
1642 | ||
1643 | /* | |
1644 | * Link the tempfile to the final place, possibly creating the | |
1645 | * last directory level as you do so. | |
1646 | * | |
1647 | * Returns the errno on failure, 0 on success. | |
1648 | */ | |
1649 | static int link_temp_to_file(const char *tmpfile, const char *filename) | |
1650 | { | |
1651 | int ret; | |
1652 | char *dir; | |
1653 | ||
1654 | if (!link(tmpfile, filename)) | |
1655 | return 0; | |
1656 | ||
1657 | /* | |
1658 | * Try to mkdir the last path component if that failed. | |
1659 | * | |
1660 | * Re-try the "link()" regardless of whether the mkdir | |
1661 | * succeeds, since a race might mean that somebody | |
1662 | * else succeeded. | |
1663 | */ | |
1664 | ret = errno; | |
1665 | dir = strrchr(filename, '/'); | |
1666 | if (dir) { | |
1667 | *dir = 0; | |
1668 | if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) { | |
1669 | *dir = '/'; | |
1670 | return -2; | |
1671 | } | |
1672 | *dir = '/'; | |
1673 | if (!link(tmpfile, filename)) | |
1674 | return 0; | |
1675 | ret = errno; | |
1676 | } | |
1677 | return ret; | |
1678 | } | |
1679 | ||
1680 | /* | |
1681 | * Move the just written object into its final resting place | |
1682 | */ | |
1683 | int move_temp_to_file(const char *tmpfile, const char *filename) | |
1684 | { | |
1685 | int ret = link_temp_to_file(tmpfile, filename); | |
1686 | ||
1687 | /* | |
1688 | * Coda hack - coda doesn't like cross-directory links, | |
1689 | * so we fall back to a rename, which will mean that it | |
1690 | * won't be able to check collisions, but that's not a | |
1691 | * big deal. | |
1692 | * | |
1693 | * The same holds for FAT formatted media. | |
1694 | * | |
1695 | * When this succeeds, we just return 0. We have nothing | |
1696 | * left to unlink. | |
1697 | */ | |
1698 | if (ret && ret != EEXIST) { | |
1699 | if (!rename(tmpfile, filename)) | |
1700 | return 0; | |
1701 | ret = errno; | |
1702 | } | |
1703 | unlink(tmpfile); | |
1704 | if (ret) { | |
1705 | if (ret != EEXIST) { | |
1706 | return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret)); | |
1707 | } | |
1708 | /* FIXME!!! Collision check here ? */ | |
1709 | } | |
1710 | ||
1711 | return 0; | |
1712 | } | |
1713 | ||
1714 | static int write_buffer(int fd, const void *buf, size_t len) | |
1715 | { | |
1716 | if (write_in_full(fd, buf, len) < 0) | |
1717 | return error("file write error (%s)", strerror(errno)); | |
1718 | return 0; | |
1719 | } | |
1720 | ||
1721 | static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len) | |
1722 | { | |
1723 | int hdr_len; | |
1724 | unsigned char c; | |
1725 | ||
1726 | c = (type << 4) | (len & 15); | |
1727 | len >>= 4; | |
1728 | hdr_len = 1; | |
1729 | while (len) { | |
1730 | *hdr++ = c | 0x80; | |
1731 | hdr_len++; | |
1732 | c = (len & 0x7f); | |
1733 | len >>= 7; | |
1734 | } | |
1735 | *hdr = c; | |
1736 | return hdr_len; | |
1737 | } | |
1738 | ||
1739 | static void setup_object_header(z_stream *stream, const char *type, unsigned long len) | |
1740 | { | |
1741 | int obj_type, hdr; | |
1742 | ||
1743 | if (use_legacy_headers) { | |
1744 | while (deflate(stream, 0) == Z_OK) | |
1745 | /* nothing */; | |
1746 | return; | |
1747 | } | |
1748 | if (!strcmp(type, blob_type)) | |
1749 | obj_type = OBJ_BLOB; | |
1750 | else if (!strcmp(type, tree_type)) | |
1751 | obj_type = OBJ_TREE; | |
1752 | else if (!strcmp(type, commit_type)) | |
1753 | obj_type = OBJ_COMMIT; | |
1754 | else if (!strcmp(type, tag_type)) | |
1755 | obj_type = OBJ_TAG; | |
1756 | else | |
1757 | die("trying to generate bogus object of type '%s'", type); | |
1758 | hdr = write_binary_header(stream->next_out, obj_type, len); | |
1759 | stream->total_out = hdr; | |
1760 | stream->next_out += hdr; | |
1761 | stream->avail_out -= hdr; | |
1762 | } | |
1763 | ||
1764 | int hash_sha1_file(void *buf, unsigned long len, const char *type, | |
1765 | unsigned char *sha1) | |
1766 | { | |
1767 | unsigned char hdr[50]; | |
1768 | int hdrlen; | |
1769 | write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen); | |
1770 | return 0; | |
1771 | } | |
1772 | ||
1773 | int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1) | |
1774 | { | |
1775 | int size; | |
1776 | unsigned char *compressed; | |
1777 | z_stream stream; | |
1778 | unsigned char sha1[20]; | |
1779 | char *filename; | |
1780 | static char tmpfile[PATH_MAX]; | |
1781 | unsigned char hdr[50]; | |
1782 | int fd, hdrlen; | |
1783 | ||
1784 | /* Normally if we have it in the pack then we do not bother writing | |
1785 | * it out into .git/objects/??/?{38} file. | |
1786 | */ | |
1787 | write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen); | |
1788 | filename = sha1_file_name(sha1); | |
1789 | if (returnsha1) | |
1790 | hashcpy(returnsha1, sha1); | |
1791 | if (has_sha1_file(sha1)) | |
1792 | return 0; | |
1793 | fd = open(filename, O_RDONLY); | |
1794 | if (fd >= 0) { | |
1795 | /* | |
1796 | * FIXME!!! We might do collision checking here, but we'd | |
1797 | * need to uncompress the old file and check it. Later. | |
1798 | */ | |
1799 | close(fd); | |
1800 | return 0; | |
1801 | } | |
1802 | ||
1803 | if (errno != ENOENT) { | |
1804 | return error("sha1 file %s: %s\n", filename, strerror(errno)); | |
1805 | } | |
1806 | ||
1807 | snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory()); | |
1808 | ||
1809 | fd = mkstemp(tmpfile); | |
1810 | if (fd < 0) { | |
1811 | if (errno == EPERM) | |
1812 | return error("insufficient permission for adding an object to repository database %s\n", get_object_directory()); | |
1813 | else | |
1814 | return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno)); | |
1815 | } | |
1816 | ||
1817 | /* Set it up */ | |
1818 | memset(&stream, 0, sizeof(stream)); | |
1819 | deflateInit(&stream, zlib_compression_level); | |
1820 | size = 8 + deflateBound(&stream, len+hdrlen); | |
1821 | compressed = xmalloc(size); | |
1822 | ||
1823 | /* Compress it */ | |
1824 | stream.next_out = compressed; | |
1825 | stream.avail_out = size; | |
1826 | ||
1827 | /* First header.. */ | |
1828 | stream.next_in = hdr; | |
1829 | stream.avail_in = hdrlen; | |
1830 | setup_object_header(&stream, type, len); | |
1831 | ||
1832 | /* Then the data itself.. */ | |
1833 | stream.next_in = buf; | |
1834 | stream.avail_in = len; | |
1835 | while (deflate(&stream, Z_FINISH) == Z_OK) | |
1836 | /* nothing */; | |
1837 | deflateEnd(&stream); | |
1838 | size = stream.total_out; | |
1839 | ||
1840 | if (write_buffer(fd, compressed, size) < 0) | |
1841 | die("unable to write sha1 file"); | |
1842 | fchmod(fd, 0444); | |
1843 | close(fd); | |
1844 | free(compressed); | |
1845 | ||
1846 | return move_temp_to_file(tmpfile, filename); | |
1847 | } | |
1848 | ||
1849 | /* | |
1850 | * We need to unpack and recompress the object for writing | |
1851 | * it out to a different file. | |
1852 | */ | |
1853 | static void *repack_object(const unsigned char *sha1, unsigned long *objsize) | |
1854 | { | |
1855 | size_t size; | |
1856 | z_stream stream; | |
1857 | unsigned char *unpacked; | |
1858 | unsigned long len; | |
1859 | char type[20]; | |
1860 | char hdr[50]; | |
1861 | int hdrlen; | |
1862 | void *buf; | |
1863 | ||
1864 | /* need to unpack and recompress it by itself */ | |
1865 | unpacked = read_packed_sha1(sha1, type, &len); | |
1866 | if (!unpacked) | |
1867 | error("cannot read sha1_file for %s", sha1_to_hex(sha1)); | |
1868 | ||
1869 | hdrlen = sprintf(hdr, "%s %lu", type, len) + 1; | |
1870 | ||
1871 | /* Set it up */ | |
1872 | memset(&stream, 0, sizeof(stream)); | |
1873 | deflateInit(&stream, zlib_compression_level); | |
1874 | size = deflateBound(&stream, len + hdrlen); | |
1875 | buf = xmalloc(size); | |
1876 | ||
1877 | /* Compress it */ | |
1878 | stream.next_out = buf; | |
1879 | stream.avail_out = size; | |
1880 | ||
1881 | /* First header.. */ | |
1882 | stream.next_in = (void *)hdr; | |
1883 | stream.avail_in = hdrlen; | |
1884 | while (deflate(&stream, 0) == Z_OK) | |
1885 | /* nothing */; | |
1886 | ||
1887 | /* Then the data itself.. */ | |
1888 | stream.next_in = unpacked; | |
1889 | stream.avail_in = len; | |
1890 | while (deflate(&stream, Z_FINISH) == Z_OK) | |
1891 | /* nothing */; | |
1892 | deflateEnd(&stream); | |
1893 | free(unpacked); | |
1894 | ||
1895 | *objsize = stream.total_out; | |
1896 | return buf; | |
1897 | } | |
1898 | ||
1899 | int write_sha1_to_fd(int fd, const unsigned char *sha1) | |
1900 | { | |
1901 | int retval; | |
1902 | unsigned long objsize; | |
1903 | void *buf = map_sha1_file(sha1, &objsize); | |
1904 | ||
1905 | if (buf) { | |
1906 | retval = write_buffer(fd, buf, objsize); | |
1907 | munmap(buf, objsize); | |
1908 | return retval; | |
1909 | } | |
1910 | ||
1911 | buf = repack_object(sha1, &objsize); | |
1912 | retval = write_buffer(fd, buf, objsize); | |
1913 | free(buf); | |
1914 | return retval; | |
1915 | } | |
1916 | ||
1917 | int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer, | |
1918 | size_t bufsize, size_t *bufposn) | |
1919 | { | |
1920 | char tmpfile[PATH_MAX]; | |
1921 | int local; | |
1922 | z_stream stream; | |
1923 | unsigned char real_sha1[20]; | |
1924 | unsigned char discard[4096]; | |
1925 | int ret; | |
1926 | SHA_CTX c; | |
1927 | ||
1928 | snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory()); | |
1929 | ||
1930 | local = mkstemp(tmpfile); | |
1931 | if (local < 0) { | |
1932 | if (errno == EPERM) | |
1933 | return error("insufficient permission for adding an object to repository database %s\n", get_object_directory()); | |
1934 | else | |
1935 | return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno)); | |
1936 | } | |
1937 | ||
1938 | memset(&stream, 0, sizeof(stream)); | |
1939 | ||
1940 | inflateInit(&stream); | |
1941 | ||
1942 | SHA1_Init(&c); | |
1943 | ||
1944 | do { | |
1945 | ssize_t size; | |
1946 | if (*bufposn) { | |
1947 | stream.avail_in = *bufposn; | |
1948 | stream.next_in = (unsigned char *) buffer; | |
1949 | do { | |
1950 | stream.next_out = discard; | |
1951 | stream.avail_out = sizeof(discard); | |
1952 | ret = inflate(&stream, Z_SYNC_FLUSH); | |
1953 | SHA1_Update(&c, discard, sizeof(discard) - | |
1954 | stream.avail_out); | |
1955 | } while (stream.avail_in && ret == Z_OK); | |
1956 | if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0) | |
1957 | die("unable to write sha1 file"); | |
1958 | memmove(buffer, buffer + *bufposn - stream.avail_in, | |
1959 | stream.avail_in); | |
1960 | *bufposn = stream.avail_in; | |
1961 | if (ret != Z_OK) | |
1962 | break; | |
1963 | } | |
1964 | size = xread(fd, buffer + *bufposn, bufsize - *bufposn); | |
1965 | if (size <= 0) { | |
1966 | close(local); | |
1967 | unlink(tmpfile); | |
1968 | if (!size) | |
1969 | return error("Connection closed?"); | |
1970 | perror("Reading from connection"); | |
1971 | return -1; | |
1972 | } | |
1973 | *bufposn += size; | |
1974 | } while (1); | |
1975 | inflateEnd(&stream); | |
1976 | ||
1977 | close(local); | |
1978 | SHA1_Final(real_sha1, &c); | |
1979 | if (ret != Z_STREAM_END) { | |
1980 | unlink(tmpfile); | |
1981 | return error("File %s corrupted", sha1_to_hex(sha1)); | |
1982 | } | |
1983 | if (hashcmp(sha1, real_sha1)) { | |
1984 | unlink(tmpfile); | |
1985 | return error("File %s has bad hash", sha1_to_hex(sha1)); | |
1986 | } | |
1987 | ||
1988 | return move_temp_to_file(tmpfile, sha1_file_name(sha1)); | |
1989 | } | |
1990 | ||
1991 | int has_pack_index(const unsigned char *sha1) | |
1992 | { | |
1993 | struct stat st; | |
1994 | if (stat(sha1_pack_index_name(sha1), &st)) | |
1995 | return 0; | |
1996 | return 1; | |
1997 | } | |
1998 | ||
1999 | int has_pack_file(const unsigned char *sha1) | |
2000 | { | |
2001 | struct stat st; | |
2002 | if (stat(sha1_pack_name(sha1), &st)) | |
2003 | return 0; | |
2004 | return 1; | |
2005 | } | |
2006 | ||
2007 | int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed) | |
2008 | { | |
2009 | struct pack_entry e; | |
2010 | return find_pack_entry(sha1, &e, ignore_packed); | |
2011 | } | |
2012 | ||
2013 | int has_sha1_file(const unsigned char *sha1) | |
2014 | { | |
2015 | struct stat st; | |
2016 | struct pack_entry e; | |
2017 | ||
2018 | if (find_pack_entry(sha1, &e, NULL)) | |
2019 | return 1; | |
2020 | return find_sha1_file(sha1, &st) ? 1 : 0; | |
2021 | } | |
2022 | ||
2023 | /* | |
2024 | * reads from fd as long as possible into a supplied buffer of size bytes. | |
2025 | * If necessary the buffer's size is increased using realloc() | |
2026 | * | |
2027 | * returns 0 if anything went fine and -1 otherwise | |
2028 | * | |
2029 | * NOTE: both buf and size may change, but even when -1 is returned | |
2030 | * you still have to free() it yourself. | |
2031 | */ | |
2032 | int read_pipe(int fd, char** return_buf, unsigned long* return_size) | |
2033 | { | |
2034 | char* buf = *return_buf; | |
2035 | unsigned long size = *return_size; | |
2036 | int iret; | |
2037 | unsigned long off = 0; | |
2038 | ||
2039 | do { | |
2040 | iret = xread(fd, buf + off, size - off); | |
2041 | if (iret > 0) { | |
2042 | off += iret; | |
2043 | if (off == size) { | |
2044 | size *= 2; | |
2045 | buf = xrealloc(buf, size); | |
2046 | } | |
2047 | } | |
2048 | } while (iret > 0); | |
2049 | ||
2050 | *return_buf = buf; | |
2051 | *return_size = off; | |
2052 | ||
2053 | if (iret < 0) | |
2054 | return -1; | |
2055 | return 0; | |
2056 | } | |
2057 | ||
2058 | int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object) | |
2059 | { | |
2060 | unsigned long size = 4096; | |
2061 | char *buf = xmalloc(size); | |
2062 | int ret; | |
2063 | ||
2064 | if (read_pipe(fd, &buf, &size)) { | |
2065 | free(buf); | |
2066 | return -1; | |
2067 | } | |
2068 | ||
2069 | if (!type) | |
2070 | type = blob_type; | |
2071 | if (write_object) | |
2072 | ret = write_sha1_file(buf, size, type, sha1); | |
2073 | else | |
2074 | ret = hash_sha1_file(buf, size, type, sha1); | |
2075 | free(buf); | |
2076 | return ret; | |
2077 | } | |
2078 | ||
2079 | int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type) | |
2080 | { | |
2081 | unsigned long size = st->st_size; | |
2082 | void *buf; | |
2083 | int ret; | |
2084 | ||
2085 | buf = ""; | |
2086 | if (size) | |
2087 | buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0); | |
2088 | close(fd); | |
2089 | ||
2090 | if (!type) | |
2091 | type = blob_type; | |
2092 | /* FIXME: CRLF -> LF conversion here for blobs! We'll need the path! */ | |
2093 | if (write_object) | |
2094 | ret = write_sha1_file(buf, size, type, sha1); | |
2095 | else | |
2096 | ret = hash_sha1_file(buf, size, type, sha1); | |
2097 | if (size) | |
2098 | munmap(buf, size); | |
2099 | return ret; | |
2100 | } | |
2101 | ||
2102 | int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object) | |
2103 | { | |
2104 | int fd; | |
2105 | char *target; | |
2106 | ||
2107 | switch (st->st_mode & S_IFMT) { | |
2108 | case S_IFREG: | |
2109 | fd = open(path, O_RDONLY); | |
2110 | if (fd < 0) | |
2111 | return error("open(\"%s\"): %s", path, | |
2112 | strerror(errno)); | |
2113 | if (index_fd(sha1, fd, st, write_object, NULL) < 0) | |
2114 | return error("%s: failed to insert into database", | |
2115 | path); | |
2116 | break; | |
2117 | case S_IFLNK: | |
2118 | target = xmalloc(st->st_size+1); | |
2119 | if (readlink(path, target, st->st_size+1) != st->st_size) { | |
2120 | char *errstr = strerror(errno); | |
2121 | free(target); | |
2122 | return error("readlink(\"%s\"): %s", path, | |
2123 | errstr); | |
2124 | } | |
2125 | if (!write_object) | |
2126 | hash_sha1_file(target, st->st_size, blob_type, sha1); | |
2127 | else if (write_sha1_file(target, st->st_size, blob_type, sha1)) | |
2128 | return error("%s: failed to insert into database", | |
2129 | path); | |
2130 | free(target); | |
2131 | break; | |
2132 | default: | |
2133 | return error("%s: unsupported file type", path); | |
2134 | } | |
2135 | return 0; | |
2136 | } | |
2137 | ||
2138 | int read_pack_header(int fd, struct pack_header *header) | |
2139 | { | |
2140 | char *c = (char*)header; | |
2141 | ssize_t remaining = sizeof(struct pack_header); | |
2142 | do { | |
2143 | ssize_t r = xread(fd, c, remaining); | |
2144 | if (r <= 0) | |
2145 | /* "eof before pack header was fully read" */ | |
2146 | return PH_ERROR_EOF; | |
2147 | remaining -= r; | |
2148 | c += r; | |
2149 | } while (remaining > 0); | |
2150 | if (header->hdr_signature != htonl(PACK_SIGNATURE)) | |
2151 | /* "protocol error (pack signature mismatch detected)" */ | |
2152 | return PH_ERROR_PACK_SIGNATURE; | |
2153 | if (!pack_version_ok(header->hdr_version)) | |
2154 | /* "protocol error (pack version unsupported)" */ | |
2155 | return PH_ERROR_PROTOCOL; | |
2156 | return 0; | |
2157 | } |