Merge branch 'jc/maint-1.6.0-diff-borrow-carefully' into maint-1.6.1
[git/git.git] / sha1_file.c
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 #include "refs.h"
17 #include "pack-revindex.h"
18 #include "sha1-lookup.h"
19
20 #ifndef O_NOATIME
21 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
22 #define O_NOATIME 01000000
23 #else
24 #define O_NOATIME 0
25 #endif
26 #endif
27
28 #ifdef NO_C99_FORMAT
29 #define SZ_FMT "lu"
30 static unsigned long sz_fmt(size_t s) { return (unsigned long)s; }
31 #else
32 #define SZ_FMT "zu"
33 static size_t sz_fmt(size_t s) { return s; }
34 #endif
35
36 const unsigned char null_sha1[20];
37
38 const signed char hexval_table[256] = {
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
45 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
46 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
47 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
48 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
51 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
69 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
71 };
72
73 int get_sha1_hex(const char *hex, unsigned char *sha1)
74 {
75 int i;
76 for (i = 0; i < 20; i++) {
77 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
78 if (val & ~0xff)
79 return -1;
80 *sha1++ = val;
81 hex += 2;
82 }
83 return 0;
84 }
85
86 static inline int offset_1st_component(const char *path)
87 {
88 if (has_dos_drive_prefix(path))
89 return 2 + (path[2] == '/');
90 return *path == '/';
91 }
92
93 int safe_create_leading_directories(char *path)
94 {
95 char *pos = path + offset_1st_component(path);
96 struct stat st;
97
98 while (pos) {
99 pos = strchr(pos, '/');
100 if (!pos)
101 break;
102 while (*++pos == '/')
103 ;
104 if (!*pos)
105 break;
106 *--pos = '\0';
107 if (!stat(path, &st)) {
108 /* path exists */
109 if (!S_ISDIR(st.st_mode)) {
110 *pos = '/';
111 return -3;
112 }
113 }
114 else if (mkdir(path, 0777)) {
115 *pos = '/';
116 return -1;
117 }
118 else if (adjust_shared_perm(path)) {
119 *pos = '/';
120 return -2;
121 }
122 *pos++ = '/';
123 }
124 return 0;
125 }
126
127 int safe_create_leading_directories_const(const char *path)
128 {
129 /* path points to cache entries, so xstrdup before messing with it */
130 char *buf = xstrdup(path);
131 int result = safe_create_leading_directories(buf);
132 free(buf);
133 return result;
134 }
135
136 char *sha1_to_hex(const unsigned char *sha1)
137 {
138 static int bufno;
139 static char hexbuffer[4][50];
140 static const char hex[] = "0123456789abcdef";
141 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
142 int i;
143
144 for (i = 0; i < 20; i++) {
145 unsigned int val = *sha1++;
146 *buf++ = hex[val >> 4];
147 *buf++ = hex[val & 0xf];
148 }
149 *buf = '\0';
150
151 return buffer;
152 }
153
154 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
155 {
156 int i;
157 for (i = 0; i < 20; i++) {
158 static char hex[] = "0123456789abcdef";
159 unsigned int val = sha1[i];
160 char *pos = pathbuf + i*2 + (i > 0);
161 *pos++ = hex[val >> 4];
162 *pos = hex[val & 0xf];
163 }
164 }
165
166 /*
167 * NOTE! This returns a statically allocated buffer, so you have to be
168 * careful about using it. Do an "xstrdup()" if you need to save the
169 * filename.
170 *
171 * Also note that this returns the location for creating. Reading
172 * SHA1 file can happen from any alternate directory listed in the
173 * DB_ENVIRONMENT environment variable if it is not found in
174 * the primary object database.
175 */
176 char *sha1_file_name(const unsigned char *sha1)
177 {
178 static char *name, *base;
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 memcpy(base, sha1_file_directory, len);
185 memset(base+len, 0, 60);
186 base[len] = '/';
187 base[len+3] = '/';
188 name = base + len + 1;
189 }
190 fill_sha1_path(name, sha1);
191 return base;
192 }
193
194 static char *sha1_get_pack_name(const unsigned char *sha1,
195 char **name, char **base, const char *which)
196 {
197 static const char hex[] = "0123456789abcdef";
198 char *buf;
199 int i;
200
201 if (!*base) {
202 const char *sha1_file_directory = get_object_directory();
203 int len = strlen(sha1_file_directory);
204 *base = xmalloc(len + 60);
205 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
206 sha1_file_directory, which);
207 *name = *base + len + 11;
208 }
209
210 buf = *name;
211
212 for (i = 0; i < 20; i++) {
213 unsigned int val = *sha1++;
214 *buf++ = hex[val >> 4];
215 *buf++ = hex[val & 0xf];
216 }
217
218 return *base;
219 }
220
221 char *sha1_pack_name(const unsigned char *sha1)
222 {
223 static char *name, *base;
224
225 return sha1_get_pack_name(sha1, &name, &base, "pack");
226 }
227
228 char *sha1_pack_index_name(const unsigned char *sha1)
229 {
230 static char *name, *base;
231
232 return sha1_get_pack_name(sha1, &name, &base, "idx");
233 }
234
235 struct alternate_object_database *alt_odb_list;
236 static struct alternate_object_database **alt_odb_tail;
237
238 static void read_info_alternates(const char * alternates, int depth);
239
240 /*
241 * Prepare alternate object database registry.
242 *
243 * The variable alt_odb_list points at the list of struct
244 * alternate_object_database. The elements on this list come from
245 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
246 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
247 * whose contents is similar to that environment variable but can be
248 * LF separated. Its base points at a statically allocated buffer that
249 * contains "/the/directory/corresponding/to/.git/objects/...", while
250 * its name points just after the slash at the end of ".git/objects/"
251 * in the example above, and has enough space to hold 40-byte hex
252 * SHA1, an extra slash for the first level indirection, and the
253 * terminating NUL.
254 */
255 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
256 {
257 const char *objdir = get_object_directory();
258 struct alternate_object_database *ent;
259 struct alternate_object_database *alt;
260 /* 43 = 40-byte + 2 '/' + terminating NUL */
261 int pfxlen = len;
262 int entlen = pfxlen + 43;
263 int base_len = -1;
264
265 if (!is_absolute_path(entry) && relative_base) {
266 /* Relative alt-odb */
267 if (base_len < 0)
268 base_len = strlen(relative_base) + 1;
269 entlen += base_len;
270 pfxlen += base_len;
271 }
272 ent = xmalloc(sizeof(*ent) + entlen);
273
274 if (!is_absolute_path(entry) && relative_base) {
275 memcpy(ent->base, relative_base, base_len - 1);
276 ent->base[base_len - 1] = '/';
277 memcpy(ent->base + base_len, entry, len);
278 }
279 else
280 memcpy(ent->base, entry, pfxlen);
281
282 ent->name = ent->base + pfxlen + 1;
283 ent->base[pfxlen + 3] = '/';
284 ent->base[pfxlen] = ent->base[entlen-1] = 0;
285
286 /* Detect cases where alternate disappeared */
287 if (!is_directory(ent->base)) {
288 error("object directory %s does not exist; "
289 "check .git/objects/info/alternates.",
290 ent->base);
291 free(ent);
292 return -1;
293 }
294
295 /* Prevent the common mistake of listing the same
296 * thing twice, or object directory itself.
297 */
298 for (alt = alt_odb_list; alt; alt = alt->next) {
299 if (!memcmp(ent->base, alt->base, pfxlen)) {
300 free(ent);
301 return -1;
302 }
303 }
304 if (!memcmp(ent->base, objdir, pfxlen)) {
305 free(ent);
306 return -1;
307 }
308
309 /* add the alternate entry */
310 *alt_odb_tail = ent;
311 alt_odb_tail = &(ent->next);
312 ent->next = NULL;
313
314 /* recursively add alternates */
315 read_info_alternates(ent->base, depth + 1);
316
317 ent->base[pfxlen] = '/';
318
319 return 0;
320 }
321
322 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
323 const char *relative_base, int depth)
324 {
325 const char *cp, *last;
326
327 if (depth > 5) {
328 error("%s: ignoring alternate object stores, nesting too deep.",
329 relative_base);
330 return;
331 }
332
333 last = alt;
334 while (last < ep) {
335 cp = last;
336 if (cp < ep && *cp == '#') {
337 while (cp < ep && *cp != sep)
338 cp++;
339 last = cp + 1;
340 continue;
341 }
342 while (cp < ep && *cp != sep)
343 cp++;
344 if (last != cp) {
345 if (!is_absolute_path(last) && depth) {
346 error("%s: ignoring relative alternate object store %s",
347 relative_base, last);
348 } else {
349 link_alt_odb_entry(last, cp - last,
350 relative_base, depth);
351 }
352 }
353 while (cp < ep && *cp == sep)
354 cp++;
355 last = cp;
356 }
357 }
358
359 static void read_info_alternates(const char * relative_base, int depth)
360 {
361 char *map;
362 size_t mapsz;
363 struct stat st;
364 const char alt_file_name[] = "info/alternates";
365 /* Given that relative_base is no longer than PATH_MAX,
366 ensure that "path" has enough space to append "/", the
367 file name, "info/alternates", and a trailing NUL. */
368 char path[PATH_MAX + 1 + sizeof alt_file_name];
369 int fd;
370
371 sprintf(path, "%s/%s", relative_base, alt_file_name);
372 fd = open(path, O_RDONLY);
373 if (fd < 0)
374 return;
375 if (fstat(fd, &st) || (st.st_size == 0)) {
376 close(fd);
377 return;
378 }
379 mapsz = xsize_t(st.st_size);
380 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
381 close(fd);
382
383 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
384
385 munmap(map, mapsz);
386 }
387
388 void add_to_alternates_file(const char *reference)
389 {
390 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
391 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
392 char *alt = mkpath("%s/objects\n", reference);
393 write_or_die(fd, alt, strlen(alt));
394 if (commit_lock_file(lock))
395 die("could not close alternates file");
396 if (alt_odb_tail)
397 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
398 }
399
400 void foreach_alt_odb(alt_odb_fn fn, void *cb)
401 {
402 struct alternate_object_database *ent;
403
404 prepare_alt_odb();
405 for (ent = alt_odb_list; ent; ent = ent->next)
406 if (fn(ent, cb))
407 return;
408 }
409
410 void prepare_alt_odb(void)
411 {
412 const char *alt;
413
414 if (alt_odb_tail)
415 return;
416
417 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
418 if (!alt) alt = "";
419
420 alt_odb_tail = &alt_odb_list;
421 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
422
423 read_info_alternates(get_object_directory(), 0);
424 }
425
426 static int has_loose_object_local(const unsigned char *sha1)
427 {
428 char *name = sha1_file_name(sha1);
429 return !access(name, F_OK);
430 }
431
432 int has_loose_object_nonlocal(const unsigned char *sha1)
433 {
434 struct alternate_object_database *alt;
435 prepare_alt_odb();
436 for (alt = alt_odb_list; alt; alt = alt->next) {
437 fill_sha1_path(alt->name, sha1);
438 if (!access(alt->base, F_OK))
439 return 1;
440 }
441 return 0;
442 }
443
444 static int has_loose_object(const unsigned char *sha1)
445 {
446 return has_loose_object_local(sha1) ||
447 has_loose_object_nonlocal(sha1);
448 }
449
450 static unsigned int pack_used_ctr;
451 static unsigned int pack_mmap_calls;
452 static unsigned int peak_pack_open_windows;
453 static unsigned int pack_open_windows;
454 static size_t peak_pack_mapped;
455 static size_t pack_mapped;
456 struct packed_git *packed_git;
457
458 void pack_report(void)
459 {
460 fprintf(stderr,
461 "pack_report: getpagesize() = %10" SZ_FMT "\n"
462 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
463 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
464 sz_fmt(getpagesize()),
465 sz_fmt(packed_git_window_size),
466 sz_fmt(packed_git_limit));
467 fprintf(stderr,
468 "pack_report: pack_used_ctr = %10u\n"
469 "pack_report: pack_mmap_calls = %10u\n"
470 "pack_report: pack_open_windows = %10u / %10u\n"
471 "pack_report: pack_mapped = "
472 "%10" SZ_FMT " / %10" SZ_FMT "\n",
473 pack_used_ctr,
474 pack_mmap_calls,
475 pack_open_windows, peak_pack_open_windows,
476 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
477 }
478
479 static int check_packed_git_idx(const char *path, struct packed_git *p)
480 {
481 void *idx_map;
482 struct pack_idx_header *hdr;
483 size_t idx_size;
484 uint32_t version, nr, i, *index;
485 int fd = open(path, O_RDONLY);
486 struct stat st;
487
488 if (fd < 0)
489 return -1;
490 if (fstat(fd, &st)) {
491 close(fd);
492 return -1;
493 }
494 idx_size = xsize_t(st.st_size);
495 if (idx_size < 4 * 256 + 20 + 20) {
496 close(fd);
497 return error("index file %s is too small", path);
498 }
499 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
500 close(fd);
501
502 hdr = idx_map;
503 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
504 version = ntohl(hdr->idx_version);
505 if (version < 2 || version > 2) {
506 munmap(idx_map, idx_size);
507 return error("index file %s is version %"PRIu32
508 " and is not supported by this binary"
509 " (try upgrading GIT to a newer version)",
510 path, version);
511 }
512 } else
513 version = 1;
514
515 nr = 0;
516 index = idx_map;
517 if (version > 1)
518 index += 2; /* skip index header */
519 for (i = 0; i < 256; i++) {
520 uint32_t n = ntohl(index[i]);
521 if (n < nr) {
522 munmap(idx_map, idx_size);
523 return error("non-monotonic index %s", path);
524 }
525 nr = n;
526 }
527
528 if (version == 1) {
529 /*
530 * Total size:
531 * - 256 index entries 4 bytes each
532 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
533 * - 20-byte SHA1 of the packfile
534 * - 20-byte SHA1 file checksum
535 */
536 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
537 munmap(idx_map, idx_size);
538 return error("wrong index v1 file size in %s", path);
539 }
540 } else if (version == 2) {
541 /*
542 * Minimum size:
543 * - 8 bytes of header
544 * - 256 index entries 4 bytes each
545 * - 20-byte sha1 entry * nr
546 * - 4-byte crc entry * nr
547 * - 4-byte offset entry * nr
548 * - 20-byte SHA1 of the packfile
549 * - 20-byte SHA1 file checksum
550 * And after the 4-byte offset table might be a
551 * variable sized table containing 8-byte entries
552 * for offsets larger than 2^31.
553 */
554 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
555 unsigned long max_size = min_size;
556 if (nr)
557 max_size += (nr - 1)*8;
558 if (idx_size < min_size || idx_size > max_size) {
559 munmap(idx_map, idx_size);
560 return error("wrong index v2 file size in %s", path);
561 }
562 if (idx_size != min_size &&
563 /*
564 * make sure we can deal with large pack offsets.
565 * 31-bit signed offset won't be enough, neither
566 * 32-bit unsigned one will be.
567 */
568 (sizeof(off_t) <= 4)) {
569 munmap(idx_map, idx_size);
570 return error("pack too large for current definition of off_t in %s", path);
571 }
572 }
573
574 p->index_version = version;
575 p->index_data = idx_map;
576 p->index_size = idx_size;
577 p->num_objects = nr;
578 return 0;
579 }
580
581 int open_pack_index(struct packed_git *p)
582 {
583 char *idx_name;
584 int ret;
585
586 if (p->index_data)
587 return 0;
588
589 idx_name = xstrdup(p->pack_name);
590 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
591 ret = check_packed_git_idx(idx_name, p);
592 free(idx_name);
593 return ret;
594 }
595
596 static void scan_windows(struct packed_git *p,
597 struct packed_git **lru_p,
598 struct pack_window **lru_w,
599 struct pack_window **lru_l)
600 {
601 struct pack_window *w, *w_l;
602
603 for (w_l = NULL, w = p->windows; w; w = w->next) {
604 if (!w->inuse_cnt) {
605 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
606 *lru_p = p;
607 *lru_w = w;
608 *lru_l = w_l;
609 }
610 }
611 w_l = w;
612 }
613 }
614
615 static int unuse_one_window(struct packed_git *current, int keep_fd)
616 {
617 struct packed_git *p, *lru_p = NULL;
618 struct pack_window *lru_w = NULL, *lru_l = NULL;
619
620 if (current)
621 scan_windows(current, &lru_p, &lru_w, &lru_l);
622 for (p = packed_git; p; p = p->next)
623 scan_windows(p, &lru_p, &lru_w, &lru_l);
624 if (lru_p) {
625 munmap(lru_w->base, lru_w->len);
626 pack_mapped -= lru_w->len;
627 if (lru_l)
628 lru_l->next = lru_w->next;
629 else {
630 lru_p->windows = lru_w->next;
631 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
632 close(lru_p->pack_fd);
633 lru_p->pack_fd = -1;
634 }
635 }
636 free(lru_w);
637 pack_open_windows--;
638 return 1;
639 }
640 return 0;
641 }
642
643 void release_pack_memory(size_t need, int fd)
644 {
645 size_t cur = pack_mapped;
646 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
647 ; /* nothing */
648 }
649
650 void close_pack_windows(struct packed_git *p)
651 {
652 while (p->windows) {
653 struct pack_window *w = p->windows;
654
655 if (w->inuse_cnt)
656 die("pack '%s' still has open windows to it",
657 p->pack_name);
658 munmap(w->base, w->len);
659 pack_mapped -= w->len;
660 pack_open_windows--;
661 p->windows = w->next;
662 free(w);
663 }
664 }
665
666 void unuse_pack(struct pack_window **w_cursor)
667 {
668 struct pack_window *w = *w_cursor;
669 if (w) {
670 w->inuse_cnt--;
671 *w_cursor = NULL;
672 }
673 }
674
675 /*
676 * This is used by git-repack in case a newly created pack happens to
677 * contain the same set of objects as an existing one. In that case
678 * the resulting file might be different even if its name would be the
679 * same. It is best to close any reference to the old pack before it is
680 * replaced on disk. Of course no index pointers nor windows for given pack
681 * must subsist at this point. If ever objects from this pack are requested
682 * again, the new version of the pack will be reinitialized through
683 * reprepare_packed_git().
684 */
685 void free_pack_by_name(const char *pack_name)
686 {
687 struct packed_git *p, **pp = &packed_git;
688
689 while (*pp) {
690 p = *pp;
691 if (strcmp(pack_name, p->pack_name) == 0) {
692 clear_delta_base_cache();
693 close_pack_windows(p);
694 if (p->pack_fd != -1)
695 close(p->pack_fd);
696 if (p->index_data)
697 munmap((void *)p->index_data, p->index_size);
698 free(p->bad_object_sha1);
699 *pp = p->next;
700 free(p);
701 return;
702 }
703 pp = &p->next;
704 }
705 }
706
707 /*
708 * Do not call this directly as this leaks p->pack_fd on error return;
709 * call open_packed_git() instead.
710 */
711 static int open_packed_git_1(struct packed_git *p)
712 {
713 struct stat st;
714 struct pack_header hdr;
715 unsigned char sha1[20];
716 unsigned char *idx_sha1;
717 long fd_flag;
718
719 if (!p->index_data && open_pack_index(p))
720 return error("packfile %s index unavailable", p->pack_name);
721
722 p->pack_fd = open(p->pack_name, O_RDONLY);
723 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
724 return -1;
725
726 /* If we created the struct before we had the pack we lack size. */
727 if (!p->pack_size) {
728 if (!S_ISREG(st.st_mode))
729 return error("packfile %s not a regular file", p->pack_name);
730 p->pack_size = st.st_size;
731 } else if (p->pack_size != st.st_size)
732 return error("packfile %s size changed", p->pack_name);
733
734 /* We leave these file descriptors open with sliding mmap;
735 * there is no point keeping them open across exec(), though.
736 */
737 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
738 if (fd_flag < 0)
739 return error("cannot determine file descriptor flags");
740 fd_flag |= FD_CLOEXEC;
741 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
742 return error("cannot set FD_CLOEXEC");
743
744 /* Verify we recognize this pack file format. */
745 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
746 return error("file %s is far too short to be a packfile", p->pack_name);
747 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
748 return error("file %s is not a GIT packfile", p->pack_name);
749 if (!pack_version_ok(hdr.hdr_version))
750 return error("packfile %s is version %"PRIu32" and not"
751 " supported (try upgrading GIT to a newer version)",
752 p->pack_name, ntohl(hdr.hdr_version));
753
754 /* Verify the pack matches its index. */
755 if (p->num_objects != ntohl(hdr.hdr_entries))
756 return error("packfile %s claims to have %"PRIu32" objects"
757 " while index indicates %"PRIu32" objects",
758 p->pack_name, ntohl(hdr.hdr_entries),
759 p->num_objects);
760 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
761 return error("end of packfile %s is unavailable", p->pack_name);
762 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
763 return error("packfile %s signature is unavailable", p->pack_name);
764 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
765 if (hashcmp(sha1, idx_sha1))
766 return error("packfile %s does not match index", p->pack_name);
767 return 0;
768 }
769
770 static int open_packed_git(struct packed_git *p)
771 {
772 if (!open_packed_git_1(p))
773 return 0;
774 if (p->pack_fd != -1) {
775 close(p->pack_fd);
776 p->pack_fd = -1;
777 }
778 return -1;
779 }
780
781 static int in_window(struct pack_window *win, off_t offset)
782 {
783 /* We must promise at least 20 bytes (one hash) after the
784 * offset is available from this window, otherwise the offset
785 * is not actually in this window and a different window (which
786 * has that one hash excess) must be used. This is to support
787 * the object header and delta base parsing routines below.
788 */
789 off_t win_off = win->offset;
790 return win_off <= offset
791 && (offset + 20) <= (win_off + win->len);
792 }
793
794 unsigned char* use_pack(struct packed_git *p,
795 struct pack_window **w_cursor,
796 off_t offset,
797 unsigned int *left)
798 {
799 struct pack_window *win = *w_cursor;
800
801 if (p->pack_fd == -1 && open_packed_git(p))
802 die("packfile %s cannot be accessed", p->pack_name);
803
804 /* Since packfiles end in a hash of their content and its
805 * pointless to ask for an offset into the middle of that
806 * hash, and the in_window function above wouldn't match
807 * don't allow an offset too close to the end of the file.
808 */
809 if (offset > (p->pack_size - 20))
810 die("offset beyond end of packfile (truncated pack?)");
811
812 if (!win || !in_window(win, offset)) {
813 if (win)
814 win->inuse_cnt--;
815 for (win = p->windows; win; win = win->next) {
816 if (in_window(win, offset))
817 break;
818 }
819 if (!win) {
820 size_t window_align = packed_git_window_size / 2;
821 off_t len;
822 win = xcalloc(1, sizeof(*win));
823 win->offset = (offset / window_align) * window_align;
824 len = p->pack_size - win->offset;
825 if (len > packed_git_window_size)
826 len = packed_git_window_size;
827 win->len = (size_t)len;
828 pack_mapped += win->len;
829 while (packed_git_limit < pack_mapped
830 && unuse_one_window(p, p->pack_fd))
831 ; /* nothing */
832 win->base = xmmap(NULL, win->len,
833 PROT_READ, MAP_PRIVATE,
834 p->pack_fd, win->offset);
835 if (win->base == MAP_FAILED)
836 die("packfile %s cannot be mapped: %s",
837 p->pack_name,
838 strerror(errno));
839 pack_mmap_calls++;
840 pack_open_windows++;
841 if (pack_mapped > peak_pack_mapped)
842 peak_pack_mapped = pack_mapped;
843 if (pack_open_windows > peak_pack_open_windows)
844 peak_pack_open_windows = pack_open_windows;
845 win->next = p->windows;
846 p->windows = win;
847 }
848 }
849 if (win != *w_cursor) {
850 win->last_used = pack_used_ctr++;
851 win->inuse_cnt++;
852 *w_cursor = win;
853 }
854 offset -= win->offset;
855 if (left)
856 *left = win->len - xsize_t(offset);
857 return win->base + offset;
858 }
859
860 static struct packed_git *alloc_packed_git(int extra)
861 {
862 struct packed_git *p = xmalloc(sizeof(*p) + extra);
863 memset(p, 0, sizeof(*p));
864 p->pack_fd = -1;
865 return p;
866 }
867
868 struct packed_git *add_packed_git(const char *path, int path_len, int local)
869 {
870 struct stat st;
871 struct packed_git *p = alloc_packed_git(path_len + 2);
872
873 /*
874 * Make sure a corresponding .pack file exists and that
875 * the index looks sane.
876 */
877 path_len -= strlen(".idx");
878 if (path_len < 1) {
879 free(p);
880 return NULL;
881 }
882 memcpy(p->pack_name, path, path_len);
883
884 strcpy(p->pack_name + path_len, ".keep");
885 if (!access(p->pack_name, F_OK))
886 p->pack_keep = 1;
887
888 strcpy(p->pack_name + path_len, ".pack");
889 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
890 free(p);
891 return NULL;
892 }
893
894 /* ok, it looks sane as far as we can check without
895 * actually mapping the pack file.
896 */
897 p->pack_size = st.st_size;
898 p->pack_local = local;
899 p->mtime = st.st_mtime;
900 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
901 hashclr(p->sha1);
902 return p;
903 }
904
905 struct packed_git *parse_pack_index(unsigned char *sha1)
906 {
907 const char *idx_path = sha1_pack_index_name(sha1);
908 const char *path = sha1_pack_name(sha1);
909 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
910
911 strcpy(p->pack_name, path);
912 hashcpy(p->sha1, sha1);
913 if (check_packed_git_idx(idx_path, p)) {
914 free(p);
915 return NULL;
916 }
917
918 return p;
919 }
920
921 void install_packed_git(struct packed_git *pack)
922 {
923 pack->next = packed_git;
924 packed_git = pack;
925 }
926
927 static void prepare_packed_git_one(char *objdir, int local)
928 {
929 /* Ensure that this buffer is large enough so that we can
930 append "/pack/" without clobbering the stack even if
931 strlen(objdir) were PATH_MAX. */
932 char path[PATH_MAX + 1 + 4 + 1 + 1];
933 int len;
934 DIR *dir;
935 struct dirent *de;
936
937 sprintf(path, "%s/pack", objdir);
938 len = strlen(path);
939 dir = opendir(path);
940 if (!dir) {
941 if (errno != ENOENT)
942 error("unable to open object pack directory: %s: %s",
943 path, strerror(errno));
944 return;
945 }
946 path[len++] = '/';
947 while ((de = readdir(dir)) != NULL) {
948 int namelen = strlen(de->d_name);
949 struct packed_git *p;
950
951 if (!has_extension(de->d_name, ".idx"))
952 continue;
953
954 if (len + namelen + 1 > sizeof(path))
955 continue;
956
957 /* Don't reopen a pack we already have. */
958 strcpy(path + len, de->d_name);
959 for (p = packed_git; p; p = p->next) {
960 if (!memcmp(path, p->pack_name, len + namelen - 4))
961 break;
962 }
963 if (p)
964 continue;
965 /* See if it really is a valid .idx file with corresponding
966 * .pack file that we can map.
967 */
968 p = add_packed_git(path, len + namelen, local);
969 if (!p)
970 continue;
971 install_packed_git(p);
972 }
973 closedir(dir);
974 }
975
976 static int sort_pack(const void *a_, const void *b_)
977 {
978 struct packed_git *a = *((struct packed_git **)a_);
979 struct packed_git *b = *((struct packed_git **)b_);
980 int st;
981
982 /*
983 * Local packs tend to contain objects specific to our
984 * variant of the project than remote ones. In addition,
985 * remote ones could be on a network mounted filesystem.
986 * Favor local ones for these reasons.
987 */
988 st = a->pack_local - b->pack_local;
989 if (st)
990 return -st;
991
992 /*
993 * Younger packs tend to contain more recent objects,
994 * and more recent objects tend to get accessed more
995 * often.
996 */
997 if (a->mtime < b->mtime)
998 return 1;
999 else if (a->mtime == b->mtime)
1000 return 0;
1001 return -1;
1002 }
1003
1004 static void rearrange_packed_git(void)
1005 {
1006 struct packed_git **ary, *p;
1007 int i, n;
1008
1009 for (n = 0, p = packed_git; p; p = p->next)
1010 n++;
1011 if (n < 2)
1012 return;
1013
1014 /* prepare an array of packed_git for easier sorting */
1015 ary = xcalloc(n, sizeof(struct packed_git *));
1016 for (n = 0, p = packed_git; p; p = p->next)
1017 ary[n++] = p;
1018
1019 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1020
1021 /* link them back again */
1022 for (i = 0; i < n - 1; i++)
1023 ary[i]->next = ary[i + 1];
1024 ary[n - 1]->next = NULL;
1025 packed_git = ary[0];
1026
1027 free(ary);
1028 }
1029
1030 static int prepare_packed_git_run_once = 0;
1031 void prepare_packed_git(void)
1032 {
1033 struct alternate_object_database *alt;
1034
1035 if (prepare_packed_git_run_once)
1036 return;
1037 prepare_packed_git_one(get_object_directory(), 1);
1038 prepare_alt_odb();
1039 for (alt = alt_odb_list; alt; alt = alt->next) {
1040 alt->name[-1] = 0;
1041 prepare_packed_git_one(alt->base, 0);
1042 alt->name[-1] = '/';
1043 }
1044 rearrange_packed_git();
1045 prepare_packed_git_run_once = 1;
1046 }
1047
1048 void reprepare_packed_git(void)
1049 {
1050 discard_revindex();
1051 prepare_packed_git_run_once = 0;
1052 prepare_packed_git();
1053 }
1054
1055 static void mark_bad_packed_object(struct packed_git *p,
1056 const unsigned char *sha1)
1057 {
1058 unsigned i;
1059 for (i = 0; i < p->num_bad_objects; i++)
1060 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1061 return;
1062 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1063 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1064 p->num_bad_objects++;
1065 }
1066
1067 static int has_packed_and_bad(const unsigned char *sha1)
1068 {
1069 struct packed_git *p;
1070 unsigned i;
1071
1072 for (p = packed_git; p; p = p->next)
1073 for (i = 0; i < p->num_bad_objects; i++)
1074 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1075 return 1;
1076 return 0;
1077 }
1078
1079 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1080 {
1081 unsigned char real_sha1[20];
1082 hash_sha1_file(map, size, type, real_sha1);
1083 return hashcmp(sha1, real_sha1) ? -1 : 0;
1084 }
1085
1086 static int git_open_noatime(const char *name)
1087 {
1088 static int sha1_file_open_flag = O_NOATIME;
1089 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1090
1091 /* Might the failure be due to O_NOATIME? */
1092 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1093 fd = open(name, O_RDONLY);
1094 if (fd >= 0)
1095 sha1_file_open_flag = 0;
1096 }
1097 return fd;
1098 }
1099
1100 static int open_sha1_file(const unsigned char *sha1)
1101 {
1102 int fd;
1103 char *name = sha1_file_name(sha1);
1104 struct alternate_object_database *alt;
1105
1106 fd = git_open_noatime(name);
1107 if (fd >= 0)
1108 return fd;
1109
1110 prepare_alt_odb();
1111 errno = ENOENT;
1112 for (alt = alt_odb_list; alt; alt = alt->next) {
1113 name = alt->name;
1114 fill_sha1_path(name, sha1);
1115 fd = git_open_noatime(alt->base);
1116 if (fd >= 0)
1117 return fd;
1118 }
1119 return -1;
1120 }
1121
1122 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1123 {
1124 void *map;
1125 int fd;
1126
1127 fd = open_sha1_file(sha1);
1128 map = NULL;
1129 if (fd >= 0) {
1130 struct stat st;
1131
1132 if (!fstat(fd, &st)) {
1133 *size = xsize_t(st.st_size);
1134 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1135 }
1136 close(fd);
1137 }
1138 return map;
1139 }
1140
1141 static int legacy_loose_object(unsigned char *map)
1142 {
1143 unsigned int word;
1144
1145 /*
1146 * Is it a zlib-compressed buffer? If so, the first byte
1147 * must be 0x78 (15-bit window size, deflated), and the
1148 * first 16-bit word is evenly divisible by 31
1149 */
1150 word = (map[0] << 8) + map[1];
1151 if (map[0] == 0x78 && !(word % 31))
1152 return 1;
1153 else
1154 return 0;
1155 }
1156
1157 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1158 unsigned long len, enum object_type *type, unsigned long *sizep)
1159 {
1160 unsigned shift;
1161 unsigned char c;
1162 unsigned long size;
1163 unsigned long used = 0;
1164
1165 c = buf[used++];
1166 *type = (c >> 4) & 7;
1167 size = c & 15;
1168 shift = 4;
1169 while (c & 0x80) {
1170 if (len <= used || sizeof(long) * 8 <= shift) {
1171 error("bad object header");
1172 return 0;
1173 }
1174 c = buf[used++];
1175 size += (c & 0x7f) << shift;
1176 shift += 7;
1177 }
1178 *sizep = size;
1179 return used;
1180 }
1181
1182 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1183 {
1184 unsigned long size, used;
1185 static const char valid_loose_object_type[8] = {
1186 0, /* OBJ_EXT */
1187 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1188 0, /* "delta" and others are invalid in a loose object */
1189 };
1190 enum object_type type;
1191
1192 /* Get the data stream */
1193 memset(stream, 0, sizeof(*stream));
1194 stream->next_in = map;
1195 stream->avail_in = mapsize;
1196 stream->next_out = buffer;
1197 stream->avail_out = bufsiz;
1198
1199 if (legacy_loose_object(map)) {
1200 git_inflate_init(stream);
1201 return git_inflate(stream, 0);
1202 }
1203
1204
1205 /*
1206 * There used to be a second loose object header format which
1207 * was meant to mimic the in-pack format, allowing for direct
1208 * copy of the object data. This format turned up not to be
1209 * really worth it and we don't write it any longer. But we
1210 * can still read it.
1211 */
1212 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1213 if (!used || !valid_loose_object_type[type])
1214 return -1;
1215 map += used;
1216 mapsize -= used;
1217
1218 /* Set up the stream for the rest.. */
1219 stream->next_in = map;
1220 stream->avail_in = mapsize;
1221 git_inflate_init(stream);
1222
1223 /* And generate the fake traditional header */
1224 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1225 typename(type), size);
1226 return 0;
1227 }
1228
1229 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1230 {
1231 int bytes = strlen(buffer) + 1;
1232 unsigned char *buf = xmalloc(1+size);
1233 unsigned long n;
1234 int status = Z_OK;
1235
1236 n = stream->total_out - bytes;
1237 if (n > size)
1238 n = size;
1239 memcpy(buf, (char *) buffer + bytes, n);
1240 bytes = n;
1241 if (bytes <= size) {
1242 /*
1243 * The above condition must be (bytes <= size), not
1244 * (bytes < size). In other words, even though we
1245 * expect no more output and set avail_out to zer0,
1246 * the input zlib stream may have bytes that express
1247 * "this concludes the stream", and we *do* want to
1248 * eat that input.
1249 *
1250 * Otherwise we would not be able to test that we
1251 * consumed all the input to reach the expected size;
1252 * we also want to check that zlib tells us that all
1253 * went well with status == Z_STREAM_END at the end.
1254 */
1255 stream->next_out = buf + bytes;
1256 stream->avail_out = size - bytes;
1257 while (status == Z_OK)
1258 status = git_inflate(stream, Z_FINISH);
1259 }
1260 buf[size] = 0;
1261 if (status == Z_STREAM_END && !stream->avail_in) {
1262 git_inflate_end(stream);
1263 return buf;
1264 }
1265
1266 if (status < 0)
1267 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1268 else if (stream->avail_in)
1269 error("garbage at end of loose object '%s'",
1270 sha1_to_hex(sha1));
1271 free(buf);
1272 return NULL;
1273 }
1274
1275 /*
1276 * We used to just use "sscanf()", but that's actually way
1277 * too permissive for what we want to check. So do an anal
1278 * object header parse by hand.
1279 */
1280 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1281 {
1282 char type[10];
1283 int i;
1284 unsigned long size;
1285
1286 /*
1287 * The type can be at most ten bytes (including the
1288 * terminating '\0' that we add), and is followed by
1289 * a space.
1290 */
1291 i = 0;
1292 for (;;) {
1293 char c = *hdr++;
1294 if (c == ' ')
1295 break;
1296 type[i++] = c;
1297 if (i >= sizeof(type))
1298 return -1;
1299 }
1300 type[i] = 0;
1301
1302 /*
1303 * The length must follow immediately, and be in canonical
1304 * decimal format (ie "010" is not valid).
1305 */
1306 size = *hdr++ - '0';
1307 if (size > 9)
1308 return -1;
1309 if (size) {
1310 for (;;) {
1311 unsigned long c = *hdr - '0';
1312 if (c > 9)
1313 break;
1314 hdr++;
1315 size = size * 10 + c;
1316 }
1317 }
1318 *sizep = size;
1319
1320 /*
1321 * The length must be followed by a zero byte
1322 */
1323 return *hdr ? -1 : type_from_string(type);
1324 }
1325
1326 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1327 {
1328 int ret;
1329 z_stream stream;
1330 char hdr[8192];
1331
1332 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1333 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1334 return NULL;
1335
1336 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1337 }
1338
1339 unsigned long get_size_from_delta(struct packed_git *p,
1340 struct pack_window **w_curs,
1341 off_t curpos)
1342 {
1343 const unsigned char *data;
1344 unsigned char delta_head[20], *in;
1345 z_stream stream;
1346 int st;
1347
1348 memset(&stream, 0, sizeof(stream));
1349 stream.next_out = delta_head;
1350 stream.avail_out = sizeof(delta_head);
1351
1352 git_inflate_init(&stream);
1353 do {
1354 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1355 stream.next_in = in;
1356 st = git_inflate(&stream, Z_FINISH);
1357 curpos += stream.next_in - in;
1358 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1359 stream.total_out < sizeof(delta_head));
1360 git_inflate_end(&stream);
1361 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1362 error("delta data unpack-initial failed");
1363 return 0;
1364 }
1365
1366 /* Examine the initial part of the delta to figure out
1367 * the result size.
1368 */
1369 data = delta_head;
1370
1371 /* ignore base size */
1372 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1373
1374 /* Read the result size */
1375 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1376 }
1377
1378 static off_t get_delta_base(struct packed_git *p,
1379 struct pack_window **w_curs,
1380 off_t *curpos,
1381 enum object_type type,
1382 off_t delta_obj_offset)
1383 {
1384 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1385 off_t base_offset;
1386
1387 /* use_pack() assured us we have [base_info, base_info + 20)
1388 * as a range that we can look at without walking off the
1389 * end of the mapped window. Its actually the hash size
1390 * that is assured. An OFS_DELTA longer than the hash size
1391 * is stupid, as then a REF_DELTA would be smaller to store.
1392 */
1393 if (type == OBJ_OFS_DELTA) {
1394 unsigned used = 0;
1395 unsigned char c = base_info[used++];
1396 base_offset = c & 127;
1397 while (c & 128) {
1398 base_offset += 1;
1399 if (!base_offset || MSB(base_offset, 7))
1400 return 0; /* overflow */
1401 c = base_info[used++];
1402 base_offset = (base_offset << 7) + (c & 127);
1403 }
1404 base_offset = delta_obj_offset - base_offset;
1405 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1406 return 0; /* out of bound */
1407 *curpos += used;
1408 } else if (type == OBJ_REF_DELTA) {
1409 /* The base entry _must_ be in the same pack */
1410 base_offset = find_pack_entry_one(base_info, p);
1411 *curpos += 20;
1412 } else
1413 die("I am totally screwed");
1414 return base_offset;
1415 }
1416
1417 /* forward declaration for a mutually recursive function */
1418 static int packed_object_info(struct packed_git *p, off_t offset,
1419 unsigned long *sizep);
1420
1421 static int packed_delta_info(struct packed_git *p,
1422 struct pack_window **w_curs,
1423 off_t curpos,
1424 enum object_type type,
1425 off_t obj_offset,
1426 unsigned long *sizep)
1427 {
1428 off_t base_offset;
1429
1430 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1431 if (!base_offset)
1432 return OBJ_BAD;
1433 type = packed_object_info(p, base_offset, NULL);
1434 if (type <= OBJ_NONE) {
1435 struct revindex_entry *revidx;
1436 const unsigned char *base_sha1;
1437 revidx = find_pack_revindex(p, base_offset);
1438 if (!revidx)
1439 return OBJ_BAD;
1440 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1441 mark_bad_packed_object(p, base_sha1);
1442 type = sha1_object_info(base_sha1, NULL);
1443 if (type <= OBJ_NONE)
1444 return OBJ_BAD;
1445 }
1446
1447 /* We choose to only get the type of the base object and
1448 * ignore potentially corrupt pack file that expects the delta
1449 * based on a base with a wrong size. This saves tons of
1450 * inflate() calls.
1451 */
1452 if (sizep) {
1453 *sizep = get_size_from_delta(p, w_curs, curpos);
1454 if (*sizep == 0)
1455 type = OBJ_BAD;
1456 }
1457
1458 return type;
1459 }
1460
1461 static int unpack_object_header(struct packed_git *p,
1462 struct pack_window **w_curs,
1463 off_t *curpos,
1464 unsigned long *sizep)
1465 {
1466 unsigned char *base;
1467 unsigned int left;
1468 unsigned long used;
1469 enum object_type type;
1470
1471 /* use_pack() assures us we have [base, base + 20) available
1472 * as a range that we can look at at. (Its actually the hash
1473 * size that is assured.) With our object header encoding
1474 * the maximum deflated object size is 2^137, which is just
1475 * insane, so we know won't exceed what we have been given.
1476 */
1477 base = use_pack(p, w_curs, *curpos, &left);
1478 used = unpack_object_header_buffer(base, left, &type, sizep);
1479 if (!used) {
1480 type = OBJ_BAD;
1481 } else
1482 *curpos += used;
1483
1484 return type;
1485 }
1486
1487 const char *packed_object_info_detail(struct packed_git *p,
1488 off_t obj_offset,
1489 unsigned long *size,
1490 unsigned long *store_size,
1491 unsigned int *delta_chain_length,
1492 unsigned char *base_sha1)
1493 {
1494 struct pack_window *w_curs = NULL;
1495 off_t curpos;
1496 unsigned long dummy;
1497 unsigned char *next_sha1;
1498 enum object_type type;
1499 struct revindex_entry *revidx;
1500
1501 *delta_chain_length = 0;
1502 curpos = obj_offset;
1503 type = unpack_object_header(p, &w_curs, &curpos, size);
1504
1505 revidx = find_pack_revindex(p, obj_offset);
1506 *store_size = revidx[1].offset - obj_offset;
1507
1508 for (;;) {
1509 switch (type) {
1510 default:
1511 die("pack %s contains unknown object type %d",
1512 p->pack_name, type);
1513 case OBJ_COMMIT:
1514 case OBJ_TREE:
1515 case OBJ_BLOB:
1516 case OBJ_TAG:
1517 unuse_pack(&w_curs);
1518 return typename(type);
1519 case OBJ_OFS_DELTA:
1520 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1521 if (!obj_offset)
1522 die("pack %s contains bad delta base reference of type %s",
1523 p->pack_name, typename(type));
1524 if (*delta_chain_length == 0) {
1525 revidx = find_pack_revindex(p, obj_offset);
1526 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1527 }
1528 break;
1529 case OBJ_REF_DELTA:
1530 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1531 if (*delta_chain_length == 0)
1532 hashcpy(base_sha1, next_sha1);
1533 obj_offset = find_pack_entry_one(next_sha1, p);
1534 break;
1535 }
1536 (*delta_chain_length)++;
1537 curpos = obj_offset;
1538 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1539 }
1540 }
1541
1542 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1543 unsigned long *sizep)
1544 {
1545 struct pack_window *w_curs = NULL;
1546 unsigned long size;
1547 off_t curpos = obj_offset;
1548 enum object_type type;
1549
1550 type = unpack_object_header(p, &w_curs, &curpos, &size);
1551
1552 switch (type) {
1553 case OBJ_OFS_DELTA:
1554 case OBJ_REF_DELTA:
1555 type = packed_delta_info(p, &w_curs, curpos,
1556 type, obj_offset, sizep);
1557 break;
1558 case OBJ_COMMIT:
1559 case OBJ_TREE:
1560 case OBJ_BLOB:
1561 case OBJ_TAG:
1562 if (sizep)
1563 *sizep = size;
1564 break;
1565 default:
1566 error("unknown object type %i at offset %"PRIuMAX" in %s",
1567 type, (uintmax_t)obj_offset, p->pack_name);
1568 type = OBJ_BAD;
1569 }
1570 unuse_pack(&w_curs);
1571 return type;
1572 }
1573
1574 static void *unpack_compressed_entry(struct packed_git *p,
1575 struct pack_window **w_curs,
1576 off_t curpos,
1577 unsigned long size)
1578 {
1579 int st;
1580 z_stream stream;
1581 unsigned char *buffer, *in;
1582
1583 buffer = xmalloc(size + 1);
1584 buffer[size] = 0;
1585 memset(&stream, 0, sizeof(stream));
1586 stream.next_out = buffer;
1587 stream.avail_out = size;
1588
1589 git_inflate_init(&stream);
1590 do {
1591 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1592 stream.next_in = in;
1593 st = git_inflate(&stream, Z_FINISH);
1594 curpos += stream.next_in - in;
1595 } while (st == Z_OK || st == Z_BUF_ERROR);
1596 git_inflate_end(&stream);
1597 if ((st != Z_STREAM_END) || stream.total_out != size) {
1598 free(buffer);
1599 return NULL;
1600 }
1601
1602 return buffer;
1603 }
1604
1605 #define MAX_DELTA_CACHE (256)
1606
1607 static size_t delta_base_cached;
1608
1609 static struct delta_base_cache_lru_list {
1610 struct delta_base_cache_lru_list *prev;
1611 struct delta_base_cache_lru_list *next;
1612 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1613
1614 static struct delta_base_cache_entry {
1615 struct delta_base_cache_lru_list lru;
1616 void *data;
1617 struct packed_git *p;
1618 off_t base_offset;
1619 unsigned long size;
1620 enum object_type type;
1621 } delta_base_cache[MAX_DELTA_CACHE];
1622
1623 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1624 {
1625 unsigned long hash;
1626
1627 hash = (unsigned long)p + (unsigned long)base_offset;
1628 hash += (hash >> 8) + (hash >> 16);
1629 return hash % MAX_DELTA_CACHE;
1630 }
1631
1632 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1633 unsigned long *base_size, enum object_type *type, int keep_cache)
1634 {
1635 void *ret;
1636 unsigned long hash = pack_entry_hash(p, base_offset);
1637 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1638
1639 ret = ent->data;
1640 if (!ret || ent->p != p || ent->base_offset != base_offset)
1641 return unpack_entry(p, base_offset, type, base_size);
1642
1643 if (!keep_cache) {
1644 ent->data = NULL;
1645 ent->lru.next->prev = ent->lru.prev;
1646 ent->lru.prev->next = ent->lru.next;
1647 delta_base_cached -= ent->size;
1648 } else {
1649 ret = xmemdupz(ent->data, ent->size);
1650 }
1651 *type = ent->type;
1652 *base_size = ent->size;
1653 return ret;
1654 }
1655
1656 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1657 {
1658 if (ent->data) {
1659 free(ent->data);
1660 ent->data = NULL;
1661 ent->lru.next->prev = ent->lru.prev;
1662 ent->lru.prev->next = ent->lru.next;
1663 delta_base_cached -= ent->size;
1664 }
1665 }
1666
1667 void clear_delta_base_cache(void)
1668 {
1669 unsigned long p;
1670 for (p = 0; p < MAX_DELTA_CACHE; p++)
1671 release_delta_base_cache(&delta_base_cache[p]);
1672 }
1673
1674 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1675 void *base, unsigned long base_size, enum object_type type)
1676 {
1677 unsigned long hash = pack_entry_hash(p, base_offset);
1678 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1679 struct delta_base_cache_lru_list *lru;
1680
1681 release_delta_base_cache(ent);
1682 delta_base_cached += base_size;
1683
1684 for (lru = delta_base_cache_lru.next;
1685 delta_base_cached > delta_base_cache_limit
1686 && lru != &delta_base_cache_lru;
1687 lru = lru->next) {
1688 struct delta_base_cache_entry *f = (void *)lru;
1689 if (f->type == OBJ_BLOB)
1690 release_delta_base_cache(f);
1691 }
1692 for (lru = delta_base_cache_lru.next;
1693 delta_base_cached > delta_base_cache_limit
1694 && lru != &delta_base_cache_lru;
1695 lru = lru->next) {
1696 struct delta_base_cache_entry *f = (void *)lru;
1697 release_delta_base_cache(f);
1698 }
1699
1700 ent->p = p;
1701 ent->base_offset = base_offset;
1702 ent->type = type;
1703 ent->data = base;
1704 ent->size = base_size;
1705 ent->lru.next = &delta_base_cache_lru;
1706 ent->lru.prev = delta_base_cache_lru.prev;
1707 delta_base_cache_lru.prev->next = &ent->lru;
1708 delta_base_cache_lru.prev = &ent->lru;
1709 }
1710
1711 static void *unpack_delta_entry(struct packed_git *p,
1712 struct pack_window **w_curs,
1713 off_t curpos,
1714 unsigned long delta_size,
1715 off_t obj_offset,
1716 enum object_type *type,
1717 unsigned long *sizep)
1718 {
1719 void *delta_data, *result, *base;
1720 unsigned long base_size;
1721 off_t base_offset;
1722
1723 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1724 if (!base_offset) {
1725 error("failed to validate delta base reference "
1726 "at offset %"PRIuMAX" from %s",
1727 (uintmax_t)curpos, p->pack_name);
1728 return NULL;
1729 }
1730 unuse_pack(w_curs);
1731 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1732 if (!base) {
1733 /*
1734 * We're probably in deep shit, but let's try to fetch
1735 * the required base anyway from another pack or loose.
1736 * This is costly but should happen only in the presence
1737 * of a corrupted pack, and is better than failing outright.
1738 */
1739 struct revindex_entry *revidx;
1740 const unsigned char *base_sha1;
1741 revidx = find_pack_revindex(p, base_offset);
1742 if (!revidx)
1743 return NULL;
1744 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1745 error("failed to read delta base object %s"
1746 " at offset %"PRIuMAX" from %s",
1747 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1748 p->pack_name);
1749 mark_bad_packed_object(p, base_sha1);
1750 base = read_object(base_sha1, type, &base_size);
1751 if (!base)
1752 return NULL;
1753 }
1754
1755 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1756 if (!delta_data) {
1757 error("failed to unpack compressed delta "
1758 "at offset %"PRIuMAX" from %s",
1759 (uintmax_t)curpos, p->pack_name);
1760 free(base);
1761 return NULL;
1762 }
1763 result = patch_delta(base, base_size,
1764 delta_data, delta_size,
1765 sizep);
1766 if (!result)
1767 die("failed to apply delta");
1768 free(delta_data);
1769 add_delta_base_cache(p, base_offset, base, base_size, *type);
1770 return result;
1771 }
1772
1773 int do_check_packed_object_crc;
1774
1775 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1776 enum object_type *type, unsigned long *sizep)
1777 {
1778 struct pack_window *w_curs = NULL;
1779 off_t curpos = obj_offset;
1780 void *data;
1781
1782 if (do_check_packed_object_crc && p->index_version > 1) {
1783 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1784 unsigned long len = revidx[1].offset - obj_offset;
1785 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1786 const unsigned char *sha1 =
1787 nth_packed_object_sha1(p, revidx->nr);
1788 error("bad packed object CRC for %s",
1789 sha1_to_hex(sha1));
1790 mark_bad_packed_object(p, sha1);
1791 unuse_pack(&w_curs);
1792 return NULL;
1793 }
1794 }
1795
1796 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1797 switch (*type) {
1798 case OBJ_OFS_DELTA:
1799 case OBJ_REF_DELTA:
1800 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1801 obj_offset, type, sizep);
1802 break;
1803 case OBJ_COMMIT:
1804 case OBJ_TREE:
1805 case OBJ_BLOB:
1806 case OBJ_TAG:
1807 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1808 break;
1809 default:
1810 data = NULL;
1811 error("unknown object type %i at offset %"PRIuMAX" in %s",
1812 *type, (uintmax_t)obj_offset, p->pack_name);
1813 }
1814 unuse_pack(&w_curs);
1815 return data;
1816 }
1817
1818 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1819 uint32_t n)
1820 {
1821 const unsigned char *index = p->index_data;
1822 if (!index) {
1823 if (open_pack_index(p))
1824 return NULL;
1825 index = p->index_data;
1826 }
1827 if (n >= p->num_objects)
1828 return NULL;
1829 index += 4 * 256;
1830 if (p->index_version == 1) {
1831 return index + 24 * n + 4;
1832 } else {
1833 index += 8;
1834 return index + 20 * n;
1835 }
1836 }
1837
1838 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1839 {
1840 const unsigned char *index = p->index_data;
1841 index += 4 * 256;
1842 if (p->index_version == 1) {
1843 return ntohl(*((uint32_t *)(index + 24 * n)));
1844 } else {
1845 uint32_t off;
1846 index += 8 + p->num_objects * (20 + 4);
1847 off = ntohl(*((uint32_t *)(index + 4 * n)));
1848 if (!(off & 0x80000000))
1849 return off;
1850 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1851 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1852 ntohl(*((uint32_t *)(index + 4)));
1853 }
1854 }
1855
1856 off_t find_pack_entry_one(const unsigned char *sha1,
1857 struct packed_git *p)
1858 {
1859 const uint32_t *level1_ofs = p->index_data;
1860 const unsigned char *index = p->index_data;
1861 unsigned hi, lo, stride;
1862 static int use_lookup = -1;
1863 static int debug_lookup = -1;
1864
1865 if (debug_lookup < 0)
1866 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1867
1868 if (!index) {
1869 if (open_pack_index(p))
1870 return 0;
1871 level1_ofs = p->index_data;
1872 index = p->index_data;
1873 }
1874 if (p->index_version > 1) {
1875 level1_ofs += 2;
1876 index += 8;
1877 }
1878 index += 4 * 256;
1879 hi = ntohl(level1_ofs[*sha1]);
1880 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1881 if (p->index_version > 1) {
1882 stride = 20;
1883 } else {
1884 stride = 24;
1885 index += 4;
1886 }
1887
1888 if (debug_lookup)
1889 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1890 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1891
1892 if (use_lookup < 0)
1893 use_lookup = !!getenv("GIT_USE_LOOKUP");
1894 if (use_lookup) {
1895 int pos = sha1_entry_pos(index, stride, 0,
1896 lo, hi, p->num_objects, sha1);
1897 if (pos < 0)
1898 return 0;
1899 return nth_packed_object_offset(p, pos);
1900 }
1901
1902 do {
1903 unsigned mi = (lo + hi) / 2;
1904 int cmp = hashcmp(index + mi * stride, sha1);
1905
1906 if (debug_lookup)
1907 printf("lo %u hi %u rg %u mi %u\n",
1908 lo, hi, hi - lo, mi);
1909 if (!cmp)
1910 return nth_packed_object_offset(p, mi);
1911 if (cmp > 0)
1912 hi = mi;
1913 else
1914 lo = mi+1;
1915 } while (lo < hi);
1916 return 0;
1917 }
1918
1919 int matches_pack_name(struct packed_git *p, const char *name)
1920 {
1921 const char *last_c, *c;
1922
1923 if (!strcmp(p->pack_name, name))
1924 return 1;
1925
1926 for (c = p->pack_name, last_c = c; *c;)
1927 if (*c == '/')
1928 last_c = ++c;
1929 else
1930 ++c;
1931 if (!strcmp(last_c, name))
1932 return 1;
1933
1934 return 0;
1935 }
1936
1937 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1938 {
1939 static struct packed_git *last_found = (void *)1;
1940 struct packed_git *p;
1941 off_t offset;
1942
1943 prepare_packed_git();
1944 if (!packed_git)
1945 return 0;
1946 p = (last_found == (void *)1) ? packed_git : last_found;
1947
1948 do {
1949 if (ignore_packed) {
1950 const char **ig;
1951 for (ig = ignore_packed; *ig; ig++)
1952 if (matches_pack_name(p, *ig))
1953 break;
1954 if (*ig)
1955 goto next;
1956 }
1957
1958 if (p->num_bad_objects) {
1959 unsigned i;
1960 for (i = 0; i < p->num_bad_objects; i++)
1961 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1962 goto next;
1963 }
1964
1965 offset = find_pack_entry_one(sha1, p);
1966 if (offset) {
1967 /*
1968 * We are about to tell the caller where they can
1969 * locate the requested object. We better make
1970 * sure the packfile is still here and can be
1971 * accessed before supplying that answer, as
1972 * it may have been deleted since the index
1973 * was loaded!
1974 */
1975 if (p->pack_fd == -1 && open_packed_git(p)) {
1976 error("packfile %s cannot be accessed", p->pack_name);
1977 goto next;
1978 }
1979 e->offset = offset;
1980 e->p = p;
1981 hashcpy(e->sha1, sha1);
1982 last_found = p;
1983 return 1;
1984 }
1985
1986 next:
1987 if (p == last_found)
1988 p = packed_git;
1989 else
1990 p = p->next;
1991 if (p == last_found)
1992 p = p->next;
1993 } while (p);
1994 return 0;
1995 }
1996
1997 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1998 struct packed_git *packs)
1999 {
2000 struct packed_git *p;
2001
2002 for (p = packs; p; p = p->next) {
2003 if (find_pack_entry_one(sha1, p))
2004 return p;
2005 }
2006 return NULL;
2007
2008 }
2009
2010 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
2011 {
2012 int status;
2013 unsigned long mapsize, size;
2014 void *map;
2015 z_stream stream;
2016 char hdr[32];
2017
2018 map = map_sha1_file(sha1, &mapsize);
2019 if (!map)
2020 return error("unable to find %s", sha1_to_hex(sha1));
2021 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2022 status = error("unable to unpack %s header",
2023 sha1_to_hex(sha1));
2024 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2025 status = error("unable to parse %s header", sha1_to_hex(sha1));
2026 else if (sizep)
2027 *sizep = size;
2028 git_inflate_end(&stream);
2029 munmap(map, mapsize);
2030 return status;
2031 }
2032
2033 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2034 {
2035 struct pack_entry e;
2036 int status;
2037
2038 if (!find_pack_entry(sha1, &e, NULL)) {
2039 /* Most likely it's a loose object. */
2040 status = sha1_loose_object_info(sha1, sizep);
2041 if (status >= 0)
2042 return status;
2043
2044 /* Not a loose object; someone else may have just packed it. */
2045 reprepare_packed_git();
2046 if (!find_pack_entry(sha1, &e, NULL))
2047 return status;
2048 }
2049
2050 status = packed_object_info(e.p, e.offset, sizep);
2051 if (status < 0) {
2052 mark_bad_packed_object(e.p, sha1);
2053 status = sha1_object_info(sha1, sizep);
2054 }
2055
2056 return status;
2057 }
2058
2059 static void *read_packed_sha1(const unsigned char *sha1,
2060 enum object_type *type, unsigned long *size)
2061 {
2062 struct pack_entry e;
2063 void *data;
2064
2065 if (!find_pack_entry(sha1, &e, NULL))
2066 return NULL;
2067 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2068 if (!data) {
2069 /*
2070 * We're probably in deep shit, but let's try to fetch
2071 * the required object anyway from another pack or loose.
2072 * This should happen only in the presence of a corrupted
2073 * pack, and is better than failing outright.
2074 */
2075 error("failed to read object %s at offset %"PRIuMAX" from %s",
2076 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2077 mark_bad_packed_object(e.p, sha1);
2078 data = read_object(sha1, type, size);
2079 }
2080 return data;
2081 }
2082
2083 /*
2084 * This is meant to hold a *small* number of objects that you would
2085 * want read_sha1_file() to be able to return, but yet you do not want
2086 * to write them into the object store (e.g. a browse-only
2087 * application).
2088 */
2089 static struct cached_object {
2090 unsigned char sha1[20];
2091 enum object_type type;
2092 void *buf;
2093 unsigned long size;
2094 } *cached_objects;
2095 static int cached_object_nr, cached_object_alloc;
2096
2097 static struct cached_object empty_tree = {
2098 EMPTY_TREE_SHA1_BIN,
2099 OBJ_TREE,
2100 "",
2101 0
2102 };
2103
2104 static struct cached_object *find_cached_object(const unsigned char *sha1)
2105 {
2106 int i;
2107 struct cached_object *co = cached_objects;
2108
2109 for (i = 0; i < cached_object_nr; i++, co++) {
2110 if (!hashcmp(co->sha1, sha1))
2111 return co;
2112 }
2113 if (!hashcmp(sha1, empty_tree.sha1))
2114 return &empty_tree;
2115 return NULL;
2116 }
2117
2118 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2119 unsigned char *sha1)
2120 {
2121 struct cached_object *co;
2122
2123 hash_sha1_file(buf, len, typename(type), sha1);
2124 if (has_sha1_file(sha1) || find_cached_object(sha1))
2125 return 0;
2126 if (cached_object_alloc <= cached_object_nr) {
2127 cached_object_alloc = alloc_nr(cached_object_alloc);
2128 cached_objects = xrealloc(cached_objects,
2129 sizeof(*cached_objects) *
2130 cached_object_alloc);
2131 }
2132 co = &cached_objects[cached_object_nr++];
2133 co->size = len;
2134 co->type = type;
2135 co->buf = xmalloc(len);
2136 memcpy(co->buf, buf, len);
2137 hashcpy(co->sha1, sha1);
2138 return 0;
2139 }
2140
2141 void *read_object(const unsigned char *sha1, enum object_type *type,
2142 unsigned long *size)
2143 {
2144 unsigned long mapsize;
2145 void *map, *buf;
2146 struct cached_object *co;
2147
2148 co = find_cached_object(sha1);
2149 if (co) {
2150 *type = co->type;
2151 *size = co->size;
2152 return xmemdupz(co->buf, co->size);
2153 }
2154
2155 buf = read_packed_sha1(sha1, type, size);
2156 if (buf)
2157 return buf;
2158 map = map_sha1_file(sha1, &mapsize);
2159 if (map) {
2160 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2161 munmap(map, mapsize);
2162 return buf;
2163 }
2164 reprepare_packed_git();
2165 return read_packed_sha1(sha1, type, size);
2166 }
2167
2168 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
2169 unsigned long *size)
2170 {
2171 void *data = read_object(sha1, type, size);
2172 /* legacy behavior is to die on corrupted objects */
2173 if (!data && (has_loose_object(sha1) || has_packed_and_bad(sha1)))
2174 die("object %s is corrupted", sha1_to_hex(sha1));
2175 return data;
2176 }
2177
2178 void *read_object_with_reference(const unsigned char *sha1,
2179 const char *required_type_name,
2180 unsigned long *size,
2181 unsigned char *actual_sha1_return)
2182 {
2183 enum object_type type, required_type;
2184 void *buffer;
2185 unsigned long isize;
2186 unsigned char actual_sha1[20];
2187
2188 required_type = type_from_string(required_type_name);
2189 hashcpy(actual_sha1, sha1);
2190 while (1) {
2191 int ref_length = -1;
2192 const char *ref_type = NULL;
2193
2194 buffer = read_sha1_file(actual_sha1, &type, &isize);
2195 if (!buffer)
2196 return NULL;
2197 if (type == required_type) {
2198 *size = isize;
2199 if (actual_sha1_return)
2200 hashcpy(actual_sha1_return, actual_sha1);
2201 return buffer;
2202 }
2203 /* Handle references */
2204 else if (type == OBJ_COMMIT)
2205 ref_type = "tree ";
2206 else if (type == OBJ_TAG)
2207 ref_type = "object ";
2208 else {
2209 free(buffer);
2210 return NULL;
2211 }
2212 ref_length = strlen(ref_type);
2213
2214 if (ref_length + 40 > isize ||
2215 memcmp(buffer, ref_type, ref_length) ||
2216 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2217 free(buffer);
2218 return NULL;
2219 }
2220 free(buffer);
2221 /* Now we have the ID of the referred-to object in
2222 * actual_sha1. Check again. */
2223 }
2224 }
2225
2226 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2227 const char *type, unsigned char *sha1,
2228 char *hdr, int *hdrlen)
2229 {
2230 git_SHA_CTX c;
2231
2232 /* Generate the header */
2233 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2234
2235 /* Sha1.. */
2236 git_SHA1_Init(&c);
2237 git_SHA1_Update(&c, hdr, *hdrlen);
2238 git_SHA1_Update(&c, buf, len);
2239 git_SHA1_Final(sha1, &c);
2240 }
2241
2242 /*
2243 * Move the just written object into its final resting place
2244 */
2245 int move_temp_to_file(const char *tmpfile, const char *filename)
2246 {
2247 int ret = 0;
2248 if (link(tmpfile, filename))
2249 ret = errno;
2250
2251 /*
2252 * Coda hack - coda doesn't like cross-directory links,
2253 * so we fall back to a rename, which will mean that it
2254 * won't be able to check collisions, but that's not a
2255 * big deal.
2256 *
2257 * The same holds for FAT formatted media.
2258 *
2259 * When this succeeds, we just return 0. We have nothing
2260 * left to unlink.
2261 */
2262 if (ret && ret != EEXIST) {
2263 if (!rename(tmpfile, filename))
2264 return 0;
2265 ret = errno;
2266 }
2267 unlink(tmpfile);
2268 if (ret) {
2269 if (ret != EEXIST) {
2270 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2271 }
2272 /* FIXME!!! Collision check here ? */
2273 }
2274
2275 return 0;
2276 }
2277
2278 static int write_buffer(int fd, const void *buf, size_t len)
2279 {
2280 if (write_in_full(fd, buf, len) < 0)
2281 return error("file write error (%s)", strerror(errno));
2282 return 0;
2283 }
2284
2285 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2286 unsigned char *sha1)
2287 {
2288 char hdr[32];
2289 int hdrlen;
2290 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2291 return 0;
2292 }
2293
2294 /* Finalize a file on disk, and close it. */
2295 static void close_sha1_file(int fd)
2296 {
2297 if (fsync_object_files)
2298 fsync_or_die(fd, "sha1 file");
2299 fchmod(fd, 0444);
2300 if (close(fd) != 0)
2301 die("error when closing sha1 file (%s)", strerror(errno));
2302 }
2303
2304 /* Size of directory component, including the ending '/' */
2305 static inline int directory_size(const char *filename)
2306 {
2307 const char *s = strrchr(filename, '/');
2308 if (!s)
2309 return 0;
2310 return s - filename + 1;
2311 }
2312
2313 /*
2314 * This creates a temporary file in the same directory as the final
2315 * 'filename'
2316 *
2317 * We want to avoid cross-directory filename renames, because those
2318 * can have problems on various filesystems (FAT, NFS, Coda).
2319 */
2320 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2321 {
2322 int fd, dirlen = directory_size(filename);
2323
2324 if (dirlen + 20 > bufsiz) {
2325 errno = ENAMETOOLONG;
2326 return -1;
2327 }
2328 memcpy(buffer, filename, dirlen);
2329 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2330 fd = mkstemp(buffer);
2331 if (fd < 0 && dirlen && errno == ENOENT) {
2332 /* Make sure the directory exists */
2333 memcpy(buffer, filename, dirlen);
2334 buffer[dirlen-1] = 0;
2335 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2336 return -1;
2337
2338 /* Try again */
2339 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2340 fd = mkstemp(buffer);
2341 }
2342 return fd;
2343 }
2344
2345 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2346 void *buf, unsigned long len, time_t mtime)
2347 {
2348 int fd, ret;
2349 size_t size;
2350 unsigned char *compressed;
2351 z_stream stream;
2352 char *filename;
2353 static char tmpfile[PATH_MAX];
2354
2355 filename = sha1_file_name(sha1);
2356 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2357 if (fd < 0) {
2358 if (errno == EACCES)
2359 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2360 else
2361 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2362 }
2363
2364 /* Set it up */
2365 memset(&stream, 0, sizeof(stream));
2366 deflateInit(&stream, zlib_compression_level);
2367 size = 8 + deflateBound(&stream, len+hdrlen);
2368 compressed = xmalloc(size);
2369
2370 /* Compress it */
2371 stream.next_out = compressed;
2372 stream.avail_out = size;
2373
2374 /* First header.. */
2375 stream.next_in = (unsigned char *)hdr;
2376 stream.avail_in = hdrlen;
2377 while (deflate(&stream, 0) == Z_OK)
2378 /* nothing */;
2379
2380 /* Then the data itself.. */
2381 stream.next_in = buf;
2382 stream.avail_in = len;
2383 ret = deflate(&stream, Z_FINISH);
2384 if (ret != Z_STREAM_END)
2385 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2386
2387 ret = deflateEnd(&stream);
2388 if (ret != Z_OK)
2389 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2390
2391 size = stream.total_out;
2392
2393 if (write_buffer(fd, compressed, size) < 0)
2394 die("unable to write sha1 file");
2395 close_sha1_file(fd);
2396 free(compressed);
2397
2398 if (mtime) {
2399 struct utimbuf utb;
2400 utb.actime = mtime;
2401 utb.modtime = mtime;
2402 if (utime(tmpfile, &utb) < 0)
2403 warning("failed utime() on %s: %s",
2404 tmpfile, strerror(errno));
2405 }
2406
2407 return move_temp_to_file(tmpfile, filename);
2408 }
2409
2410 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2411 {
2412 unsigned char sha1[20];
2413 char hdr[32];
2414 int hdrlen;
2415
2416 /* Normally if we have it in the pack then we do not bother writing
2417 * it out into .git/objects/??/?{38} file.
2418 */
2419 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2420 if (returnsha1)
2421 hashcpy(returnsha1, sha1);
2422 if (has_sha1_file(sha1))
2423 return 0;
2424 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2425 }
2426
2427 int force_object_loose(const unsigned char *sha1, time_t mtime)
2428 {
2429 void *buf;
2430 unsigned long len;
2431 enum object_type type;
2432 char hdr[32];
2433 int hdrlen;
2434 int ret;
2435
2436 if (has_loose_object(sha1))
2437 return 0;
2438 buf = read_packed_sha1(sha1, &type, &len);
2439 if (!buf)
2440 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2441 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2442 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2443 free(buf);
2444
2445 return ret;
2446 }
2447
2448 int has_pack_index(const unsigned char *sha1)
2449 {
2450 struct stat st;
2451 if (stat(sha1_pack_index_name(sha1), &st))
2452 return 0;
2453 return 1;
2454 }
2455
2456 int has_pack_file(const unsigned char *sha1)
2457 {
2458 struct stat st;
2459 if (stat(sha1_pack_name(sha1), &st))
2460 return 0;
2461 return 1;
2462 }
2463
2464 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2465 {
2466 struct pack_entry e;
2467 return find_pack_entry(sha1, &e, ignore_packed);
2468 }
2469
2470 int has_sha1_file(const unsigned char *sha1)
2471 {
2472 struct pack_entry e;
2473
2474 if (find_pack_entry(sha1, &e, NULL))
2475 return 1;
2476 return has_loose_object(sha1);
2477 }
2478
2479 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2480 int write_object, enum object_type type, const char *path)
2481 {
2482 int ret, re_allocated = 0;
2483
2484 if (!type)
2485 type = OBJ_BLOB;
2486
2487 /*
2488 * Convert blobs to git internal format
2489 */
2490 if ((type == OBJ_BLOB) && path) {
2491 struct strbuf nbuf = STRBUF_INIT;
2492 if (convert_to_git(path, buf, size, &nbuf,
2493 write_object ? safe_crlf : 0)) {
2494 buf = strbuf_detach(&nbuf, &size);
2495 re_allocated = 1;
2496 }
2497 }
2498
2499 if (write_object)
2500 ret = write_sha1_file(buf, size, typename(type), sha1);
2501 else
2502 ret = hash_sha1_file(buf, size, typename(type), sha1);
2503 if (re_allocated)
2504 free(buf);
2505 return ret;
2506 }
2507
2508 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2509 enum object_type type, const char *path)
2510 {
2511 int ret;
2512 size_t size = xsize_t(st->st_size);
2513
2514 if (!S_ISREG(st->st_mode)) {
2515 struct strbuf sbuf = STRBUF_INIT;
2516 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2517 ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2518 type, path);
2519 else
2520 ret = -1;
2521 strbuf_release(&sbuf);
2522 } else if (size) {
2523 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2524 ret = index_mem(sha1, buf, size, write_object, type, path);
2525 munmap(buf, size);
2526 } else
2527 ret = index_mem(sha1, NULL, size, write_object, type, path);
2528 close(fd);
2529 return ret;
2530 }
2531
2532 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2533 {
2534 int fd;
2535 struct strbuf sb = STRBUF_INIT;
2536
2537 switch (st->st_mode & S_IFMT) {
2538 case S_IFREG:
2539 fd = open(path, O_RDONLY);
2540 if (fd < 0)
2541 return error("open(\"%s\"): %s", path,
2542 strerror(errno));
2543 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2544 return error("%s: failed to insert into database",
2545 path);
2546 break;
2547 case S_IFLNK:
2548 if (strbuf_readlink(&sb, path, st->st_size)) {
2549 char *errstr = strerror(errno);
2550 return error("readlink(\"%s\"): %s", path,
2551 errstr);
2552 }
2553 if (!write_object)
2554 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2555 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2556 return error("%s: failed to insert into database",
2557 path);
2558 strbuf_release(&sb);
2559 break;
2560 case S_IFDIR:
2561 return resolve_gitlink_ref(path, "HEAD", sha1);
2562 default:
2563 return error("%s: unsupported file type", path);
2564 }
2565 return 0;
2566 }
2567
2568 int read_pack_header(int fd, struct pack_header *header)
2569 {
2570 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2571 /* "eof before pack header was fully read" */
2572 return PH_ERROR_EOF;
2573
2574 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2575 /* "protocol error (pack signature mismatch detected)" */
2576 return PH_ERROR_PACK_SIGNATURE;
2577 if (!pack_version_ok(header->hdr_version))
2578 /* "protocol error (pack version unsupported)" */
2579 return PH_ERROR_PROTOCOL;
2580 return 0;
2581 }