]> git.ipfire.org Git - thirdparty/git.git/blob - sha1_file.c
Merge branch 'jn/gitweb-show-size'
[thirdparty/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 while (p->pack_fd < 0 && errno == EMFILE && unuse_one_window(p, -1))
724 p->pack_fd = open(p->pack_name, O_RDONLY);
725 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
726 return -1;
727
728 /* If we created the struct before we had the pack we lack size. */
729 if (!p->pack_size) {
730 if (!S_ISREG(st.st_mode))
731 return error("packfile %s not a regular file", p->pack_name);
732 p->pack_size = st.st_size;
733 } else if (p->pack_size != st.st_size)
734 return error("packfile %s size changed", p->pack_name);
735
736 /* We leave these file descriptors open with sliding mmap;
737 * there is no point keeping them open across exec(), though.
738 */
739 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
740 if (fd_flag < 0)
741 return error("cannot determine file descriptor flags");
742 fd_flag |= FD_CLOEXEC;
743 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
744 return error("cannot set FD_CLOEXEC");
745
746 /* Verify we recognize this pack file format. */
747 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
748 return error("file %s is far too short to be a packfile", p->pack_name);
749 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
750 return error("file %s is not a GIT packfile", p->pack_name);
751 if (!pack_version_ok(hdr.hdr_version))
752 return error("packfile %s is version %"PRIu32" and not"
753 " supported (try upgrading GIT to a newer version)",
754 p->pack_name, ntohl(hdr.hdr_version));
755
756 /* Verify the pack matches its index. */
757 if (p->num_objects != ntohl(hdr.hdr_entries))
758 return error("packfile %s claims to have %"PRIu32" objects"
759 " while index indicates %"PRIu32" objects",
760 p->pack_name, ntohl(hdr.hdr_entries),
761 p->num_objects);
762 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
763 return error("end of packfile %s is unavailable", p->pack_name);
764 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
765 return error("packfile %s signature is unavailable", p->pack_name);
766 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
767 if (hashcmp(sha1, idx_sha1))
768 return error("packfile %s does not match index", p->pack_name);
769 return 0;
770 }
771
772 static int open_packed_git(struct packed_git *p)
773 {
774 if (!open_packed_git_1(p))
775 return 0;
776 if (p->pack_fd != -1) {
777 close(p->pack_fd);
778 p->pack_fd = -1;
779 }
780 return -1;
781 }
782
783 static int in_window(struct pack_window *win, off_t offset)
784 {
785 /* We must promise at least 20 bytes (one hash) after the
786 * offset is available from this window, otherwise the offset
787 * is not actually in this window and a different window (which
788 * has that one hash excess) must be used. This is to support
789 * the object header and delta base parsing routines below.
790 */
791 off_t win_off = win->offset;
792 return win_off <= offset
793 && (offset + 20) <= (win_off + win->len);
794 }
795
796 unsigned char *use_pack(struct packed_git *p,
797 struct pack_window **w_cursor,
798 off_t offset,
799 unsigned int *left)
800 {
801 struct pack_window *win = *w_cursor;
802
803 if (p->pack_fd == -1 && open_packed_git(p))
804 die("packfile %s cannot be accessed", p->pack_name);
805
806 /* Since packfiles end in a hash of their content and it's
807 * pointless to ask for an offset into the middle of that
808 * hash, and the in_window function above wouldn't match
809 * don't allow an offset too close to the end of the file.
810 */
811 if (offset > (p->pack_size - 20))
812 die("offset beyond end of packfile (truncated pack?)");
813
814 if (!win || !in_window(win, offset)) {
815 if (win)
816 win->inuse_cnt--;
817 for (win = p->windows; win; win = win->next) {
818 if (in_window(win, offset))
819 break;
820 }
821 if (!win) {
822 size_t window_align = packed_git_window_size / 2;
823 off_t len;
824 win = xcalloc(1, sizeof(*win));
825 win->offset = (offset / window_align) * window_align;
826 len = p->pack_size - win->offset;
827 if (len > packed_git_window_size)
828 len = packed_git_window_size;
829 win->len = (size_t)len;
830 pack_mapped += win->len;
831 while (packed_git_limit < pack_mapped
832 && unuse_one_window(p, p->pack_fd))
833 ; /* nothing */
834 win->base = xmmap(NULL, win->len,
835 PROT_READ, MAP_PRIVATE,
836 p->pack_fd, win->offset);
837 if (win->base == MAP_FAILED)
838 die("packfile %s cannot be mapped: %s",
839 p->pack_name,
840 strerror(errno));
841 pack_mmap_calls++;
842 pack_open_windows++;
843 if (pack_mapped > peak_pack_mapped)
844 peak_pack_mapped = pack_mapped;
845 if (pack_open_windows > peak_pack_open_windows)
846 peak_pack_open_windows = pack_open_windows;
847 win->next = p->windows;
848 p->windows = win;
849 }
850 }
851 if (win != *w_cursor) {
852 win->last_used = pack_used_ctr++;
853 win->inuse_cnt++;
854 *w_cursor = win;
855 }
856 offset -= win->offset;
857 if (left)
858 *left = win->len - xsize_t(offset);
859 return win->base + offset;
860 }
861
862 static struct packed_git *alloc_packed_git(int extra)
863 {
864 struct packed_git *p = xmalloc(sizeof(*p) + extra);
865 memset(p, 0, sizeof(*p));
866 p->pack_fd = -1;
867 return p;
868 }
869
870 struct packed_git *add_packed_git(const char *path, int path_len, int local)
871 {
872 struct stat st;
873 struct packed_git *p = alloc_packed_git(path_len + 2);
874
875 /*
876 * Make sure a corresponding .pack file exists and that
877 * the index looks sane.
878 */
879 path_len -= strlen(".idx");
880 if (path_len < 1) {
881 free(p);
882 return NULL;
883 }
884 memcpy(p->pack_name, path, path_len);
885
886 strcpy(p->pack_name + path_len, ".keep");
887 if (!access(p->pack_name, F_OK))
888 p->pack_keep = 1;
889
890 strcpy(p->pack_name + path_len, ".pack");
891 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
892 free(p);
893 return NULL;
894 }
895
896 /* ok, it looks sane as far as we can check without
897 * actually mapping the pack file.
898 */
899 p->pack_size = st.st_size;
900 p->pack_local = local;
901 p->mtime = st.st_mtime;
902 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
903 hashclr(p->sha1);
904 return p;
905 }
906
907 struct packed_git *parse_pack_index(unsigned char *sha1)
908 {
909 const char *idx_path = sha1_pack_index_name(sha1);
910 const char *path = sha1_pack_name(sha1);
911 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
912
913 strcpy(p->pack_name, path);
914 hashcpy(p->sha1, sha1);
915 if (check_packed_git_idx(idx_path, p)) {
916 free(p);
917 return NULL;
918 }
919
920 return p;
921 }
922
923 void install_packed_git(struct packed_git *pack)
924 {
925 pack->next = packed_git;
926 packed_git = pack;
927 }
928
929 static void prepare_packed_git_one(char *objdir, int local)
930 {
931 /* Ensure that this buffer is large enough so that we can
932 append "/pack/" without clobbering the stack even if
933 strlen(objdir) were PATH_MAX. */
934 char path[PATH_MAX + 1 + 4 + 1 + 1];
935 int len;
936 DIR *dir;
937 struct dirent *de;
938
939 sprintf(path, "%s/pack", objdir);
940 len = strlen(path);
941 dir = opendir(path);
942 while (!dir && errno == EMFILE && unuse_one_window(packed_git, -1))
943 dir = opendir(path);
944 if (!dir) {
945 if (errno != ENOENT)
946 error("unable to open object pack directory: %s: %s",
947 path, strerror(errno));
948 return;
949 }
950 path[len++] = '/';
951 while ((de = readdir(dir)) != NULL) {
952 int namelen = strlen(de->d_name);
953 struct packed_git *p;
954
955 if (!has_extension(de->d_name, ".idx"))
956 continue;
957
958 if (len + namelen + 1 > sizeof(path))
959 continue;
960
961 /* Don't reopen a pack we already have. */
962 strcpy(path + len, de->d_name);
963 for (p = packed_git; p; p = p->next) {
964 if (!memcmp(path, p->pack_name, len + namelen - 4))
965 break;
966 }
967 if (p)
968 continue;
969 /* See if it really is a valid .idx file with corresponding
970 * .pack file that we can map.
971 */
972 p = add_packed_git(path, len + namelen, local);
973 if (!p)
974 continue;
975 install_packed_git(p);
976 }
977 closedir(dir);
978 }
979
980 static int sort_pack(const void *a_, const void *b_)
981 {
982 struct packed_git *a = *((struct packed_git **)a_);
983 struct packed_git *b = *((struct packed_git **)b_);
984 int st;
985
986 /*
987 * Local packs tend to contain objects specific to our
988 * variant of the project than remote ones. In addition,
989 * remote ones could be on a network mounted filesystem.
990 * Favor local ones for these reasons.
991 */
992 st = a->pack_local - b->pack_local;
993 if (st)
994 return -st;
995
996 /*
997 * Younger packs tend to contain more recent objects,
998 * and more recent objects tend to get accessed more
999 * often.
1000 */
1001 if (a->mtime < b->mtime)
1002 return 1;
1003 else if (a->mtime == b->mtime)
1004 return 0;
1005 return -1;
1006 }
1007
1008 static void rearrange_packed_git(void)
1009 {
1010 struct packed_git **ary, *p;
1011 int i, n;
1012
1013 for (n = 0, p = packed_git; p; p = p->next)
1014 n++;
1015 if (n < 2)
1016 return;
1017
1018 /* prepare an array of packed_git for easier sorting */
1019 ary = xcalloc(n, sizeof(struct packed_git *));
1020 for (n = 0, p = packed_git; p; p = p->next)
1021 ary[n++] = p;
1022
1023 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1024
1025 /* link them back again */
1026 for (i = 0; i < n - 1; i++)
1027 ary[i]->next = ary[i + 1];
1028 ary[n - 1]->next = NULL;
1029 packed_git = ary[0];
1030
1031 free(ary);
1032 }
1033
1034 static int prepare_packed_git_run_once = 0;
1035 void prepare_packed_git(void)
1036 {
1037 struct alternate_object_database *alt;
1038
1039 if (prepare_packed_git_run_once)
1040 return;
1041 prepare_packed_git_one(get_object_directory(), 1);
1042 prepare_alt_odb();
1043 for (alt = alt_odb_list; alt; alt = alt->next) {
1044 alt->name[-1] = 0;
1045 prepare_packed_git_one(alt->base, 0);
1046 alt->name[-1] = '/';
1047 }
1048 rearrange_packed_git();
1049 prepare_packed_git_run_once = 1;
1050 }
1051
1052 void reprepare_packed_git(void)
1053 {
1054 discard_revindex();
1055 prepare_packed_git_run_once = 0;
1056 prepare_packed_git();
1057 }
1058
1059 static void mark_bad_packed_object(struct packed_git *p,
1060 const unsigned char *sha1)
1061 {
1062 unsigned i;
1063 for (i = 0; i < p->num_bad_objects; i++)
1064 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1065 return;
1066 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1067 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1068 p->num_bad_objects++;
1069 }
1070
1071 static int has_packed_and_bad(const unsigned char *sha1)
1072 {
1073 struct packed_git *p;
1074 unsigned i;
1075
1076 for (p = packed_git; p; p = p->next)
1077 for (i = 0; i < p->num_bad_objects; i++)
1078 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1079 return 1;
1080 return 0;
1081 }
1082
1083 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1084 {
1085 unsigned char real_sha1[20];
1086 hash_sha1_file(map, size, type, real_sha1);
1087 return hashcmp(sha1, real_sha1) ? -1 : 0;
1088 }
1089
1090 static int git_open_noatime(const char *name)
1091 {
1092 static int sha1_file_open_flag = O_NOATIME;
1093 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1094
1095 /* Might the failure be due to O_NOATIME? */
1096 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1097 fd = open(name, O_RDONLY);
1098 if (fd >= 0)
1099 sha1_file_open_flag = 0;
1100 }
1101 return fd;
1102 }
1103
1104 static int open_sha1_file(const unsigned char *sha1)
1105 {
1106 int fd;
1107 char *name = sha1_file_name(sha1);
1108 struct alternate_object_database *alt;
1109
1110 fd = git_open_noatime(name);
1111 if (fd >= 0)
1112 return fd;
1113
1114 prepare_alt_odb();
1115 errno = ENOENT;
1116 for (alt = alt_odb_list; alt; alt = alt->next) {
1117 name = alt->name;
1118 fill_sha1_path(name, sha1);
1119 fd = git_open_noatime(alt->base);
1120 if (fd >= 0)
1121 return fd;
1122 }
1123 return -1;
1124 }
1125
1126 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1127 {
1128 void *map;
1129 int fd;
1130
1131 fd = open_sha1_file(sha1);
1132 map = NULL;
1133 if (fd >= 0) {
1134 struct stat st;
1135
1136 if (!fstat(fd, &st)) {
1137 *size = xsize_t(st.st_size);
1138 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1139 }
1140 close(fd);
1141 }
1142 return map;
1143 }
1144
1145 static int legacy_loose_object(unsigned char *map)
1146 {
1147 unsigned int word;
1148
1149 /*
1150 * Is it a zlib-compressed buffer? If so, the first byte
1151 * must be 0x78 (15-bit window size, deflated), and the
1152 * first 16-bit word is evenly divisible by 31
1153 */
1154 word = (map[0] << 8) + map[1];
1155 if (map[0] == 0x78 && !(word % 31))
1156 return 1;
1157 else
1158 return 0;
1159 }
1160
1161 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1162 unsigned long len, enum object_type *type, unsigned long *sizep)
1163 {
1164 unsigned shift;
1165 unsigned long size, c;
1166 unsigned long used = 0;
1167
1168 c = buf[used++];
1169 *type = (c >> 4) & 7;
1170 size = c & 15;
1171 shift = 4;
1172 while (c & 0x80) {
1173 if (len <= used || bitsizeof(long) <= shift) {
1174 error("bad object header");
1175 return 0;
1176 }
1177 c = buf[used++];
1178 size += (c & 0x7f) << shift;
1179 shift += 7;
1180 }
1181 *sizep = size;
1182 return used;
1183 }
1184
1185 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1186 {
1187 unsigned long size, used;
1188 static const char valid_loose_object_type[8] = {
1189 0, /* OBJ_EXT */
1190 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1191 0, /* "delta" and others are invalid in a loose object */
1192 };
1193 enum object_type type;
1194
1195 /* Get the data stream */
1196 memset(stream, 0, sizeof(*stream));
1197 stream->next_in = map;
1198 stream->avail_in = mapsize;
1199 stream->next_out = buffer;
1200 stream->avail_out = bufsiz;
1201
1202 if (legacy_loose_object(map)) {
1203 git_inflate_init(stream);
1204 return git_inflate(stream, 0);
1205 }
1206
1207
1208 /*
1209 * There used to be a second loose object header format which
1210 * was meant to mimic the in-pack format, allowing for direct
1211 * copy of the object data. This format turned up not to be
1212 * really worth it and we don't write it any longer. But we
1213 * can still read it.
1214 */
1215 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1216 if (!used || !valid_loose_object_type[type])
1217 return -1;
1218 map += used;
1219 mapsize -= used;
1220
1221 /* Set up the stream for the rest.. */
1222 stream->next_in = map;
1223 stream->avail_in = mapsize;
1224 git_inflate_init(stream);
1225
1226 /* And generate the fake traditional header */
1227 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1228 typename(type), size);
1229 return 0;
1230 }
1231
1232 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1233 {
1234 int bytes = strlen(buffer) + 1;
1235 unsigned char *buf = xmalloc(1+size);
1236 unsigned long n;
1237 int status = Z_OK;
1238
1239 n = stream->total_out - bytes;
1240 if (n > size)
1241 n = size;
1242 memcpy(buf, (char *) buffer + bytes, n);
1243 bytes = n;
1244 if (bytes <= size) {
1245 /*
1246 * The above condition must be (bytes <= size), not
1247 * (bytes < size). In other words, even though we
1248 * expect no more output and set avail_out to zer0,
1249 * the input zlib stream may have bytes that express
1250 * "this concludes the stream", and we *do* want to
1251 * eat that input.
1252 *
1253 * Otherwise we would not be able to test that we
1254 * consumed all the input to reach the expected size;
1255 * we also want to check that zlib tells us that all
1256 * went well with status == Z_STREAM_END at the end.
1257 */
1258 stream->next_out = buf + bytes;
1259 stream->avail_out = size - bytes;
1260 while (status == Z_OK)
1261 status = git_inflate(stream, Z_FINISH);
1262 }
1263 buf[size] = 0;
1264 if (status == Z_STREAM_END && !stream->avail_in) {
1265 git_inflate_end(stream);
1266 return buf;
1267 }
1268
1269 if (status < 0)
1270 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1271 else if (stream->avail_in)
1272 error("garbage at end of loose object '%s'",
1273 sha1_to_hex(sha1));
1274 free(buf);
1275 return NULL;
1276 }
1277
1278 /*
1279 * We used to just use "sscanf()", but that's actually way
1280 * too permissive for what we want to check. So do an anal
1281 * object header parse by hand.
1282 */
1283 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1284 {
1285 char type[10];
1286 int i;
1287 unsigned long size;
1288
1289 /*
1290 * The type can be at most ten bytes (including the
1291 * terminating '\0' that we add), and is followed by
1292 * a space.
1293 */
1294 i = 0;
1295 for (;;) {
1296 char c = *hdr++;
1297 if (c == ' ')
1298 break;
1299 type[i++] = c;
1300 if (i >= sizeof(type))
1301 return -1;
1302 }
1303 type[i] = 0;
1304
1305 /*
1306 * The length must follow immediately, and be in canonical
1307 * decimal format (ie "010" is not valid).
1308 */
1309 size = *hdr++ - '0';
1310 if (size > 9)
1311 return -1;
1312 if (size) {
1313 for (;;) {
1314 unsigned long c = *hdr - '0';
1315 if (c > 9)
1316 break;
1317 hdr++;
1318 size = size * 10 + c;
1319 }
1320 }
1321 *sizep = size;
1322
1323 /*
1324 * The length must be followed by a zero byte
1325 */
1326 return *hdr ? -1 : type_from_string(type);
1327 }
1328
1329 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1330 {
1331 int ret;
1332 z_stream stream;
1333 char hdr[8192];
1334
1335 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1336 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1337 return NULL;
1338
1339 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1340 }
1341
1342 unsigned long get_size_from_delta(struct packed_git *p,
1343 struct pack_window **w_curs,
1344 off_t curpos)
1345 {
1346 const unsigned char *data;
1347 unsigned char delta_head[20], *in;
1348 z_stream stream;
1349 int st;
1350
1351 memset(&stream, 0, sizeof(stream));
1352 stream.next_out = delta_head;
1353 stream.avail_out = sizeof(delta_head);
1354
1355 git_inflate_init(&stream);
1356 do {
1357 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1358 stream.next_in = in;
1359 st = git_inflate(&stream, Z_FINISH);
1360 if (st == Z_BUF_ERROR && (stream.avail_in || !stream.avail_out))
1361 break;
1362 curpos += stream.next_in - in;
1363 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1364 stream.total_out < sizeof(delta_head));
1365 git_inflate_end(&stream);
1366 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1367 error("delta data unpack-initial failed");
1368 return 0;
1369 }
1370
1371 /* Examine the initial part of the delta to figure out
1372 * the result size.
1373 */
1374 data = delta_head;
1375
1376 /* ignore base size */
1377 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1378
1379 /* Read the result size */
1380 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1381 }
1382
1383 static off_t get_delta_base(struct packed_git *p,
1384 struct pack_window **w_curs,
1385 off_t *curpos,
1386 enum object_type type,
1387 off_t delta_obj_offset)
1388 {
1389 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1390 off_t base_offset;
1391
1392 /* use_pack() assured us we have [base_info, base_info + 20)
1393 * as a range that we can look at without walking off the
1394 * end of the mapped window. Its actually the hash size
1395 * that is assured. An OFS_DELTA longer than the hash size
1396 * is stupid, as then a REF_DELTA would be smaller to store.
1397 */
1398 if (type == OBJ_OFS_DELTA) {
1399 unsigned used = 0;
1400 unsigned char c = base_info[used++];
1401 base_offset = c & 127;
1402 while (c & 128) {
1403 base_offset += 1;
1404 if (!base_offset || MSB(base_offset, 7))
1405 return 0; /* overflow */
1406 c = base_info[used++];
1407 base_offset = (base_offset << 7) + (c & 127);
1408 }
1409 base_offset = delta_obj_offset - base_offset;
1410 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1411 return 0; /* out of bound */
1412 *curpos += used;
1413 } else if (type == OBJ_REF_DELTA) {
1414 /* The base entry _must_ be in the same pack */
1415 base_offset = find_pack_entry_one(base_info, p);
1416 *curpos += 20;
1417 } else
1418 die("I am totally screwed");
1419 return base_offset;
1420 }
1421
1422 /* forward declaration for a mutually recursive function */
1423 static int packed_object_info(struct packed_git *p, off_t offset,
1424 unsigned long *sizep);
1425
1426 static int packed_delta_info(struct packed_git *p,
1427 struct pack_window **w_curs,
1428 off_t curpos,
1429 enum object_type type,
1430 off_t obj_offset,
1431 unsigned long *sizep)
1432 {
1433 off_t base_offset;
1434
1435 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1436 if (!base_offset)
1437 return OBJ_BAD;
1438 type = packed_object_info(p, base_offset, NULL);
1439 if (type <= OBJ_NONE) {
1440 struct revindex_entry *revidx;
1441 const unsigned char *base_sha1;
1442 revidx = find_pack_revindex(p, base_offset);
1443 if (!revidx)
1444 return OBJ_BAD;
1445 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1446 mark_bad_packed_object(p, base_sha1);
1447 type = sha1_object_info(base_sha1, NULL);
1448 if (type <= OBJ_NONE)
1449 return OBJ_BAD;
1450 }
1451
1452 /* We choose to only get the type of the base object and
1453 * ignore potentially corrupt pack file that expects the delta
1454 * based on a base with a wrong size. This saves tons of
1455 * inflate() calls.
1456 */
1457 if (sizep) {
1458 *sizep = get_size_from_delta(p, w_curs, curpos);
1459 if (*sizep == 0)
1460 type = OBJ_BAD;
1461 }
1462
1463 return type;
1464 }
1465
1466 static int unpack_object_header(struct packed_git *p,
1467 struct pack_window **w_curs,
1468 off_t *curpos,
1469 unsigned long *sizep)
1470 {
1471 unsigned char *base;
1472 unsigned int left;
1473 unsigned long used;
1474 enum object_type type;
1475
1476 /* use_pack() assures us we have [base, base + 20) available
1477 * as a range that we can look at at. (Its actually the hash
1478 * size that is assured.) With our object header encoding
1479 * the maximum deflated object size is 2^137, which is just
1480 * insane, so we know won't exceed what we have been given.
1481 */
1482 base = use_pack(p, w_curs, *curpos, &left);
1483 used = unpack_object_header_buffer(base, left, &type, sizep);
1484 if (!used) {
1485 type = OBJ_BAD;
1486 } else
1487 *curpos += used;
1488
1489 return type;
1490 }
1491
1492 const char *packed_object_info_detail(struct packed_git *p,
1493 off_t obj_offset,
1494 unsigned long *size,
1495 unsigned long *store_size,
1496 unsigned int *delta_chain_length,
1497 unsigned char *base_sha1)
1498 {
1499 struct pack_window *w_curs = NULL;
1500 off_t curpos;
1501 unsigned long dummy;
1502 unsigned char *next_sha1;
1503 enum object_type type;
1504 struct revindex_entry *revidx;
1505
1506 *delta_chain_length = 0;
1507 curpos = obj_offset;
1508 type = unpack_object_header(p, &w_curs, &curpos, size);
1509
1510 revidx = find_pack_revindex(p, obj_offset);
1511 *store_size = revidx[1].offset - obj_offset;
1512
1513 for (;;) {
1514 switch (type) {
1515 default:
1516 die("pack %s contains unknown object type %d",
1517 p->pack_name, type);
1518 case OBJ_COMMIT:
1519 case OBJ_TREE:
1520 case OBJ_BLOB:
1521 case OBJ_TAG:
1522 unuse_pack(&w_curs);
1523 return typename(type);
1524 case OBJ_OFS_DELTA:
1525 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1526 if (!obj_offset)
1527 die("pack %s contains bad delta base reference of type %s",
1528 p->pack_name, typename(type));
1529 if (*delta_chain_length == 0) {
1530 revidx = find_pack_revindex(p, obj_offset);
1531 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1532 }
1533 break;
1534 case OBJ_REF_DELTA:
1535 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1536 if (*delta_chain_length == 0)
1537 hashcpy(base_sha1, next_sha1);
1538 obj_offset = find_pack_entry_one(next_sha1, p);
1539 break;
1540 }
1541 (*delta_chain_length)++;
1542 curpos = obj_offset;
1543 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1544 }
1545 }
1546
1547 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1548 unsigned long *sizep)
1549 {
1550 struct pack_window *w_curs = NULL;
1551 unsigned long size;
1552 off_t curpos = obj_offset;
1553 enum object_type type;
1554
1555 type = unpack_object_header(p, &w_curs, &curpos, &size);
1556
1557 switch (type) {
1558 case OBJ_OFS_DELTA:
1559 case OBJ_REF_DELTA:
1560 type = packed_delta_info(p, &w_curs, curpos,
1561 type, obj_offset, sizep);
1562 break;
1563 case OBJ_COMMIT:
1564 case OBJ_TREE:
1565 case OBJ_BLOB:
1566 case OBJ_TAG:
1567 if (sizep)
1568 *sizep = size;
1569 break;
1570 default:
1571 error("unknown object type %i at offset %"PRIuMAX" in %s",
1572 type, (uintmax_t)obj_offset, p->pack_name);
1573 type = OBJ_BAD;
1574 }
1575 unuse_pack(&w_curs);
1576 return type;
1577 }
1578
1579 static void *unpack_compressed_entry(struct packed_git *p,
1580 struct pack_window **w_curs,
1581 off_t curpos,
1582 unsigned long size)
1583 {
1584 int st;
1585 z_stream stream;
1586 unsigned char *buffer, *in;
1587
1588 buffer = xmalloc(size + 1);
1589 buffer[size] = 0;
1590 memset(&stream, 0, sizeof(stream));
1591 stream.next_out = buffer;
1592 stream.avail_out = size;
1593
1594 git_inflate_init(&stream);
1595 do {
1596 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1597 stream.next_in = in;
1598 st = git_inflate(&stream, Z_FINISH);
1599 if (st == Z_BUF_ERROR && (stream.avail_in || !stream.avail_out))
1600 break;
1601 curpos += stream.next_in - in;
1602 } while (st == Z_OK || st == Z_BUF_ERROR);
1603 git_inflate_end(&stream);
1604 if ((st != Z_STREAM_END) || stream.total_out != size) {
1605 free(buffer);
1606 return NULL;
1607 }
1608
1609 return buffer;
1610 }
1611
1612 #define MAX_DELTA_CACHE (256)
1613
1614 static size_t delta_base_cached;
1615
1616 static struct delta_base_cache_lru_list {
1617 struct delta_base_cache_lru_list *prev;
1618 struct delta_base_cache_lru_list *next;
1619 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1620
1621 static struct delta_base_cache_entry {
1622 struct delta_base_cache_lru_list lru;
1623 void *data;
1624 struct packed_git *p;
1625 off_t base_offset;
1626 unsigned long size;
1627 enum object_type type;
1628 } delta_base_cache[MAX_DELTA_CACHE];
1629
1630 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1631 {
1632 unsigned long hash;
1633
1634 hash = (unsigned long)p + (unsigned long)base_offset;
1635 hash += (hash >> 8) + (hash >> 16);
1636 return hash % MAX_DELTA_CACHE;
1637 }
1638
1639 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1640 unsigned long *base_size, enum object_type *type, int keep_cache)
1641 {
1642 void *ret;
1643 unsigned long hash = pack_entry_hash(p, base_offset);
1644 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1645
1646 ret = ent->data;
1647 if (!ret || ent->p != p || ent->base_offset != base_offset)
1648 return unpack_entry(p, base_offset, type, base_size);
1649
1650 if (!keep_cache) {
1651 ent->data = NULL;
1652 ent->lru.next->prev = ent->lru.prev;
1653 ent->lru.prev->next = ent->lru.next;
1654 delta_base_cached -= ent->size;
1655 } else {
1656 ret = xmemdupz(ent->data, ent->size);
1657 }
1658 *type = ent->type;
1659 *base_size = ent->size;
1660 return ret;
1661 }
1662
1663 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1664 {
1665 if (ent->data) {
1666 free(ent->data);
1667 ent->data = NULL;
1668 ent->lru.next->prev = ent->lru.prev;
1669 ent->lru.prev->next = ent->lru.next;
1670 delta_base_cached -= ent->size;
1671 }
1672 }
1673
1674 void clear_delta_base_cache(void)
1675 {
1676 unsigned long p;
1677 for (p = 0; p < MAX_DELTA_CACHE; p++)
1678 release_delta_base_cache(&delta_base_cache[p]);
1679 }
1680
1681 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1682 void *base, unsigned long base_size, enum object_type type)
1683 {
1684 unsigned long hash = pack_entry_hash(p, base_offset);
1685 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1686 struct delta_base_cache_lru_list *lru;
1687
1688 release_delta_base_cache(ent);
1689 delta_base_cached += base_size;
1690
1691 for (lru = delta_base_cache_lru.next;
1692 delta_base_cached > delta_base_cache_limit
1693 && lru != &delta_base_cache_lru;
1694 lru = lru->next) {
1695 struct delta_base_cache_entry *f = (void *)lru;
1696 if (f->type == OBJ_BLOB)
1697 release_delta_base_cache(f);
1698 }
1699 for (lru = delta_base_cache_lru.next;
1700 delta_base_cached > delta_base_cache_limit
1701 && lru != &delta_base_cache_lru;
1702 lru = lru->next) {
1703 struct delta_base_cache_entry *f = (void *)lru;
1704 release_delta_base_cache(f);
1705 }
1706
1707 ent->p = p;
1708 ent->base_offset = base_offset;
1709 ent->type = type;
1710 ent->data = base;
1711 ent->size = base_size;
1712 ent->lru.next = &delta_base_cache_lru;
1713 ent->lru.prev = delta_base_cache_lru.prev;
1714 delta_base_cache_lru.prev->next = &ent->lru;
1715 delta_base_cache_lru.prev = &ent->lru;
1716 }
1717
1718 static void *read_object(const unsigned char *sha1, enum object_type *type,
1719 unsigned long *size);
1720
1721 static void *unpack_delta_entry(struct packed_git *p,
1722 struct pack_window **w_curs,
1723 off_t curpos,
1724 unsigned long delta_size,
1725 off_t obj_offset,
1726 enum object_type *type,
1727 unsigned long *sizep)
1728 {
1729 void *delta_data, *result, *base;
1730 unsigned long base_size;
1731 off_t base_offset;
1732
1733 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1734 if (!base_offset) {
1735 error("failed to validate delta base reference "
1736 "at offset %"PRIuMAX" from %s",
1737 (uintmax_t)curpos, p->pack_name);
1738 return NULL;
1739 }
1740 unuse_pack(w_curs);
1741 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1742 if (!base) {
1743 /*
1744 * We're probably in deep shit, but let's try to fetch
1745 * the required base anyway from another pack or loose.
1746 * This is costly but should happen only in the presence
1747 * of a corrupted pack, and is better than failing outright.
1748 */
1749 struct revindex_entry *revidx;
1750 const unsigned char *base_sha1;
1751 revidx = find_pack_revindex(p, base_offset);
1752 if (!revidx)
1753 return NULL;
1754 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1755 error("failed to read delta base object %s"
1756 " at offset %"PRIuMAX" from %s",
1757 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1758 p->pack_name);
1759 mark_bad_packed_object(p, base_sha1);
1760 base = read_object(base_sha1, type, &base_size);
1761 if (!base)
1762 return NULL;
1763 }
1764
1765 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1766 if (!delta_data) {
1767 error("failed to unpack compressed delta "
1768 "at offset %"PRIuMAX" from %s",
1769 (uintmax_t)curpos, p->pack_name);
1770 free(base);
1771 return NULL;
1772 }
1773 result = patch_delta(base, base_size,
1774 delta_data, delta_size,
1775 sizep);
1776 if (!result)
1777 die("failed to apply delta");
1778 free(delta_data);
1779 add_delta_base_cache(p, base_offset, base, base_size, *type);
1780 return result;
1781 }
1782
1783 int do_check_packed_object_crc;
1784
1785 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1786 enum object_type *type, unsigned long *sizep)
1787 {
1788 struct pack_window *w_curs = NULL;
1789 off_t curpos = obj_offset;
1790 void *data;
1791
1792 if (do_check_packed_object_crc && p->index_version > 1) {
1793 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1794 unsigned long len = revidx[1].offset - obj_offset;
1795 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1796 const unsigned char *sha1 =
1797 nth_packed_object_sha1(p, revidx->nr);
1798 error("bad packed object CRC for %s",
1799 sha1_to_hex(sha1));
1800 mark_bad_packed_object(p, sha1);
1801 unuse_pack(&w_curs);
1802 return NULL;
1803 }
1804 }
1805
1806 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1807 switch (*type) {
1808 case OBJ_OFS_DELTA:
1809 case OBJ_REF_DELTA:
1810 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1811 obj_offset, type, sizep);
1812 break;
1813 case OBJ_COMMIT:
1814 case OBJ_TREE:
1815 case OBJ_BLOB:
1816 case OBJ_TAG:
1817 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1818 break;
1819 default:
1820 data = NULL;
1821 error("unknown object type %i at offset %"PRIuMAX" in %s",
1822 *type, (uintmax_t)obj_offset, p->pack_name);
1823 }
1824 unuse_pack(&w_curs);
1825 return data;
1826 }
1827
1828 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1829 uint32_t n)
1830 {
1831 const unsigned char *index = p->index_data;
1832 if (!index) {
1833 if (open_pack_index(p))
1834 return NULL;
1835 index = p->index_data;
1836 }
1837 if (n >= p->num_objects)
1838 return NULL;
1839 index += 4 * 256;
1840 if (p->index_version == 1) {
1841 return index + 24 * n + 4;
1842 } else {
1843 index += 8;
1844 return index + 20 * n;
1845 }
1846 }
1847
1848 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1849 {
1850 const unsigned char *index = p->index_data;
1851 index += 4 * 256;
1852 if (p->index_version == 1) {
1853 return ntohl(*((uint32_t *)(index + 24 * n)));
1854 } else {
1855 uint32_t off;
1856 index += 8 + p->num_objects * (20 + 4);
1857 off = ntohl(*((uint32_t *)(index + 4 * n)));
1858 if (!(off & 0x80000000))
1859 return off;
1860 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1861 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1862 ntohl(*((uint32_t *)(index + 4)));
1863 }
1864 }
1865
1866 off_t find_pack_entry_one(const unsigned char *sha1,
1867 struct packed_git *p)
1868 {
1869 const uint32_t *level1_ofs = p->index_data;
1870 const unsigned char *index = p->index_data;
1871 unsigned hi, lo, stride;
1872 static int use_lookup = -1;
1873 static int debug_lookup = -1;
1874
1875 if (debug_lookup < 0)
1876 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1877
1878 if (!index) {
1879 if (open_pack_index(p))
1880 return 0;
1881 level1_ofs = p->index_data;
1882 index = p->index_data;
1883 }
1884 if (p->index_version > 1) {
1885 level1_ofs += 2;
1886 index += 8;
1887 }
1888 index += 4 * 256;
1889 hi = ntohl(level1_ofs[*sha1]);
1890 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1891 if (p->index_version > 1) {
1892 stride = 20;
1893 } else {
1894 stride = 24;
1895 index += 4;
1896 }
1897
1898 if (debug_lookup)
1899 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1900 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1901
1902 if (use_lookup < 0)
1903 use_lookup = !!getenv("GIT_USE_LOOKUP");
1904 if (use_lookup) {
1905 int pos = sha1_entry_pos(index, stride, 0,
1906 lo, hi, p->num_objects, sha1);
1907 if (pos < 0)
1908 return 0;
1909 return nth_packed_object_offset(p, pos);
1910 }
1911
1912 do {
1913 unsigned mi = (lo + hi) / 2;
1914 int cmp = hashcmp(index + mi * stride, sha1);
1915
1916 if (debug_lookup)
1917 printf("lo %u hi %u rg %u mi %u\n",
1918 lo, hi, hi - lo, mi);
1919 if (!cmp)
1920 return nth_packed_object_offset(p, mi);
1921 if (cmp > 0)
1922 hi = mi;
1923 else
1924 lo = mi+1;
1925 } while (lo < hi);
1926 return 0;
1927 }
1928
1929 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1930 {
1931 static struct packed_git *last_found = (void *)1;
1932 struct packed_git *p;
1933 off_t offset;
1934
1935 prepare_packed_git();
1936 if (!packed_git)
1937 return 0;
1938 p = (last_found == (void *)1) ? packed_git : last_found;
1939
1940 do {
1941 if (p->num_bad_objects) {
1942 unsigned i;
1943 for (i = 0; i < p->num_bad_objects; i++)
1944 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1945 goto next;
1946 }
1947
1948 offset = find_pack_entry_one(sha1, p);
1949 if (offset) {
1950 /*
1951 * We are about to tell the caller where they can
1952 * locate the requested object. We better make
1953 * sure the packfile is still here and can be
1954 * accessed before supplying that answer, as
1955 * it may have been deleted since the index
1956 * was loaded!
1957 */
1958 if (p->pack_fd == -1 && open_packed_git(p)) {
1959 error("packfile %s cannot be accessed", p->pack_name);
1960 goto next;
1961 }
1962 e->offset = offset;
1963 e->p = p;
1964 hashcpy(e->sha1, sha1);
1965 last_found = p;
1966 return 1;
1967 }
1968
1969 next:
1970 if (p == last_found)
1971 p = packed_git;
1972 else
1973 p = p->next;
1974 if (p == last_found)
1975 p = p->next;
1976 } while (p);
1977 return 0;
1978 }
1979
1980 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1981 struct packed_git *packs)
1982 {
1983 struct packed_git *p;
1984
1985 for (p = packs; p; p = p->next) {
1986 if (find_pack_entry_one(sha1, p))
1987 return p;
1988 }
1989 return NULL;
1990
1991 }
1992
1993 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1994 {
1995 int status;
1996 unsigned long mapsize, size;
1997 void *map;
1998 z_stream stream;
1999 char hdr[32];
2000
2001 map = map_sha1_file(sha1, &mapsize);
2002 if (!map)
2003 return error("unable to find %s", sha1_to_hex(sha1));
2004 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2005 status = error("unable to unpack %s header",
2006 sha1_to_hex(sha1));
2007 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2008 status = error("unable to parse %s header", sha1_to_hex(sha1));
2009 else if (sizep)
2010 *sizep = size;
2011 git_inflate_end(&stream);
2012 munmap(map, mapsize);
2013 return status;
2014 }
2015
2016 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2017 {
2018 struct pack_entry e;
2019 int status;
2020
2021 if (!find_pack_entry(sha1, &e)) {
2022 /* Most likely it's a loose object. */
2023 status = sha1_loose_object_info(sha1, sizep);
2024 if (status >= 0)
2025 return status;
2026
2027 /* Not a loose object; someone else may have just packed it. */
2028 reprepare_packed_git();
2029 if (!find_pack_entry(sha1, &e))
2030 return status;
2031 }
2032
2033 status = packed_object_info(e.p, e.offset, sizep);
2034 if (status < 0) {
2035 mark_bad_packed_object(e.p, sha1);
2036 status = sha1_object_info(sha1, sizep);
2037 }
2038
2039 return status;
2040 }
2041
2042 static void *read_packed_sha1(const unsigned char *sha1,
2043 enum object_type *type, unsigned long *size)
2044 {
2045 struct pack_entry e;
2046 void *data;
2047
2048 if (!find_pack_entry(sha1, &e))
2049 return NULL;
2050 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2051 if (!data) {
2052 /*
2053 * We're probably in deep shit, but let's try to fetch
2054 * the required object anyway from another pack or loose.
2055 * This should happen only in the presence of a corrupted
2056 * pack, and is better than failing outright.
2057 */
2058 error("failed to read object %s at offset %"PRIuMAX" from %s",
2059 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2060 mark_bad_packed_object(e.p, sha1);
2061 data = read_object(sha1, type, size);
2062 }
2063 return data;
2064 }
2065
2066 /*
2067 * This is meant to hold a *small* number of objects that you would
2068 * want read_sha1_file() to be able to return, but yet you do not want
2069 * to write them into the object store (e.g. a browse-only
2070 * application).
2071 */
2072 static struct cached_object {
2073 unsigned char sha1[20];
2074 enum object_type type;
2075 void *buf;
2076 unsigned long size;
2077 } *cached_objects;
2078 static int cached_object_nr, cached_object_alloc;
2079
2080 static struct cached_object empty_tree = {
2081 EMPTY_TREE_SHA1_BIN,
2082 OBJ_TREE,
2083 "",
2084 0
2085 };
2086
2087 static struct cached_object *find_cached_object(const unsigned char *sha1)
2088 {
2089 int i;
2090 struct cached_object *co = cached_objects;
2091
2092 for (i = 0; i < cached_object_nr; i++, co++) {
2093 if (!hashcmp(co->sha1, sha1))
2094 return co;
2095 }
2096 if (!hashcmp(sha1, empty_tree.sha1))
2097 return &empty_tree;
2098 return NULL;
2099 }
2100
2101 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2102 unsigned char *sha1)
2103 {
2104 struct cached_object *co;
2105
2106 hash_sha1_file(buf, len, typename(type), sha1);
2107 if (has_sha1_file(sha1) || find_cached_object(sha1))
2108 return 0;
2109 if (cached_object_alloc <= cached_object_nr) {
2110 cached_object_alloc = alloc_nr(cached_object_alloc);
2111 cached_objects = xrealloc(cached_objects,
2112 sizeof(*cached_objects) *
2113 cached_object_alloc);
2114 }
2115 co = &cached_objects[cached_object_nr++];
2116 co->size = len;
2117 co->type = type;
2118 co->buf = xmalloc(len);
2119 memcpy(co->buf, buf, len);
2120 hashcpy(co->sha1, sha1);
2121 return 0;
2122 }
2123
2124 static void *read_object(const unsigned char *sha1, enum object_type *type,
2125 unsigned long *size)
2126 {
2127 unsigned long mapsize;
2128 void *map, *buf;
2129 struct cached_object *co;
2130
2131 co = find_cached_object(sha1);
2132 if (co) {
2133 *type = co->type;
2134 *size = co->size;
2135 return xmemdupz(co->buf, co->size);
2136 }
2137
2138 buf = read_packed_sha1(sha1, type, size);
2139 if (buf)
2140 return buf;
2141 map = map_sha1_file(sha1, &mapsize);
2142 if (map) {
2143 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2144 munmap(map, mapsize);
2145 return buf;
2146 }
2147 reprepare_packed_git();
2148 return read_packed_sha1(sha1, type, size);
2149 }
2150
2151 void *read_sha1_file_repl(const unsigned char *sha1,
2152 enum object_type *type,
2153 unsigned long *size,
2154 const unsigned char **replacement)
2155 {
2156 const unsigned char *repl = lookup_replace_object(sha1);
2157 void *data = read_object(repl, type, size);
2158
2159 /* die if we replaced an object with one that does not exist */
2160 if (!data && repl != sha1)
2161 die("replacement %s not found for %s",
2162 sha1_to_hex(repl), sha1_to_hex(sha1));
2163
2164 /* legacy behavior is to die on corrupted objects */
2165 if (!data && (has_loose_object(repl) || has_packed_and_bad(repl)))
2166 die("object %s is corrupted", sha1_to_hex(repl));
2167
2168 if (replacement)
2169 *replacement = repl;
2170
2171 return data;
2172 }
2173
2174 void *read_object_with_reference(const unsigned char *sha1,
2175 const char *required_type_name,
2176 unsigned long *size,
2177 unsigned char *actual_sha1_return)
2178 {
2179 enum object_type type, required_type;
2180 void *buffer;
2181 unsigned long isize;
2182 unsigned char actual_sha1[20];
2183
2184 required_type = type_from_string(required_type_name);
2185 hashcpy(actual_sha1, sha1);
2186 while (1) {
2187 int ref_length = -1;
2188 const char *ref_type = NULL;
2189
2190 buffer = read_sha1_file(actual_sha1, &type, &isize);
2191 if (!buffer)
2192 return NULL;
2193 if (type == required_type) {
2194 *size = isize;
2195 if (actual_sha1_return)
2196 hashcpy(actual_sha1_return, actual_sha1);
2197 return buffer;
2198 }
2199 /* Handle references */
2200 else if (type == OBJ_COMMIT)
2201 ref_type = "tree ";
2202 else if (type == OBJ_TAG)
2203 ref_type = "object ";
2204 else {
2205 free(buffer);
2206 return NULL;
2207 }
2208 ref_length = strlen(ref_type);
2209
2210 if (ref_length + 40 > isize ||
2211 memcmp(buffer, ref_type, ref_length) ||
2212 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2213 free(buffer);
2214 return NULL;
2215 }
2216 free(buffer);
2217 /* Now we have the ID of the referred-to object in
2218 * actual_sha1. Check again. */
2219 }
2220 }
2221
2222 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2223 const char *type, unsigned char *sha1,
2224 char *hdr, int *hdrlen)
2225 {
2226 git_SHA_CTX c;
2227
2228 /* Generate the header */
2229 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2230
2231 /* Sha1.. */
2232 git_SHA1_Init(&c);
2233 git_SHA1_Update(&c, hdr, *hdrlen);
2234 git_SHA1_Update(&c, buf, len);
2235 git_SHA1_Final(sha1, &c);
2236 }
2237
2238 /*
2239 * Move the just written object into its final resting place.
2240 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2241 * "moving" is only a part of what it does, when no patch between
2242 * master to pu changes the call sites of this function.
2243 */
2244 int move_temp_to_file(const char *tmpfile, const char *filename)
2245 {
2246 int ret = 0;
2247
2248 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2249 goto try_rename;
2250 else if (link(tmpfile, filename))
2251 ret = errno;
2252
2253 /*
2254 * Coda hack - coda doesn't like cross-directory links,
2255 * so we fall back to a rename, which will mean that it
2256 * won't be able to check collisions, but that's not a
2257 * big deal.
2258 *
2259 * The same holds for FAT formatted media.
2260 *
2261 * When this succeeds, we just return. We have nothing
2262 * left to unlink.
2263 */
2264 if (ret && ret != EEXIST) {
2265 try_rename:
2266 if (!rename(tmpfile, filename))
2267 goto out;
2268 ret = errno;
2269 }
2270 unlink_or_warn(tmpfile);
2271 if (ret) {
2272 if (ret != EEXIST) {
2273 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2274 }
2275 /* FIXME!!! Collision check here ? */
2276 }
2277
2278 out:
2279 if (set_shared_perm(filename, (S_IFREG|0444)))
2280 return error("unable to set permission to '%s'", filename);
2281 return 0;
2282 }
2283
2284 static int write_buffer(int fd, const void *buf, size_t len)
2285 {
2286 if (write_in_full(fd, buf, len) < 0)
2287 return error("file write error (%s)", strerror(errno));
2288 return 0;
2289 }
2290
2291 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2292 unsigned char *sha1)
2293 {
2294 char hdr[32];
2295 int hdrlen;
2296 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2297 return 0;
2298 }
2299
2300 /* Finalize a file on disk, and close it. */
2301 static void close_sha1_file(int fd)
2302 {
2303 if (fsync_object_files)
2304 fsync_or_die(fd, "sha1 file");
2305 if (close(fd) != 0)
2306 die_errno("error when closing sha1 file");
2307 }
2308
2309 /* Size of directory component, including the ending '/' */
2310 static inline int directory_size(const char *filename)
2311 {
2312 const char *s = strrchr(filename, '/');
2313 if (!s)
2314 return 0;
2315 return s - filename + 1;
2316 }
2317
2318 /*
2319 * This creates a temporary file in the same directory as the final
2320 * 'filename'
2321 *
2322 * We want to avoid cross-directory filename renames, because those
2323 * can have problems on various filesystems (FAT, NFS, Coda).
2324 */
2325 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2326 {
2327 int fd, dirlen = directory_size(filename);
2328
2329 if (dirlen + 20 > bufsiz) {
2330 errno = ENAMETOOLONG;
2331 return -1;
2332 }
2333 memcpy(buffer, filename, dirlen);
2334 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2335 fd = mkstemp(buffer);
2336 if (fd < 0 && dirlen && errno == ENOENT) {
2337 /* Make sure the directory exists */
2338 memcpy(buffer, filename, dirlen);
2339 buffer[dirlen-1] = 0;
2340 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2341 return -1;
2342
2343 /* Try again */
2344 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2345 fd = mkstemp(buffer);
2346 }
2347 return fd;
2348 }
2349
2350 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2351 void *buf, unsigned long len, time_t mtime)
2352 {
2353 int fd, ret;
2354 size_t size;
2355 unsigned char *compressed;
2356 z_stream stream;
2357 char *filename;
2358 static char tmpfile[PATH_MAX];
2359
2360 filename = sha1_file_name(sha1);
2361 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2362 while (fd < 0 && errno == EMFILE && unuse_one_window(packed_git, -1))
2363 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2364 if (fd < 0) {
2365 if (errno == EACCES)
2366 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2367 else
2368 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2369 }
2370
2371 /* Set it up */
2372 memset(&stream, 0, sizeof(stream));
2373 deflateInit(&stream, zlib_compression_level);
2374 size = 8 + deflateBound(&stream, len+hdrlen);
2375 compressed = xmalloc(size);
2376
2377 /* Compress it */
2378 stream.next_out = compressed;
2379 stream.avail_out = size;
2380
2381 /* First header.. */
2382 stream.next_in = (unsigned char *)hdr;
2383 stream.avail_in = hdrlen;
2384 while (deflate(&stream, 0) == Z_OK)
2385 /* nothing */;
2386
2387 /* Then the data itself.. */
2388 stream.next_in = buf;
2389 stream.avail_in = len;
2390 ret = deflate(&stream, Z_FINISH);
2391 if (ret != Z_STREAM_END)
2392 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2393
2394 ret = deflateEnd(&stream);
2395 if (ret != Z_OK)
2396 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2397
2398 size = stream.total_out;
2399
2400 if (write_buffer(fd, compressed, size) < 0)
2401 die("unable to write sha1 file");
2402 close_sha1_file(fd);
2403 free(compressed);
2404
2405 if (mtime) {
2406 struct utimbuf utb;
2407 utb.actime = mtime;
2408 utb.modtime = mtime;
2409 if (utime(tmpfile, &utb) < 0)
2410 warning("failed utime() on %s: %s",
2411 tmpfile, strerror(errno));
2412 }
2413
2414 return move_temp_to_file(tmpfile, filename);
2415 }
2416
2417 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2418 {
2419 unsigned char sha1[20];
2420 char hdr[32];
2421 int hdrlen;
2422
2423 /* Normally if we have it in the pack then we do not bother writing
2424 * it out into .git/objects/??/?{38} file.
2425 */
2426 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2427 if (returnsha1)
2428 hashcpy(returnsha1, sha1);
2429 if (has_sha1_file(sha1))
2430 return 0;
2431 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2432 }
2433
2434 int force_object_loose(const unsigned char *sha1, time_t mtime)
2435 {
2436 void *buf;
2437 unsigned long len;
2438 enum object_type type;
2439 char hdr[32];
2440 int hdrlen;
2441 int ret;
2442
2443 if (has_loose_object(sha1))
2444 return 0;
2445 buf = read_packed_sha1(sha1, &type, &len);
2446 if (!buf)
2447 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2448 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2449 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2450 free(buf);
2451
2452 return ret;
2453 }
2454
2455 int has_pack_index(const unsigned char *sha1)
2456 {
2457 struct stat st;
2458 if (stat(sha1_pack_index_name(sha1), &st))
2459 return 0;
2460 return 1;
2461 }
2462
2463 int has_pack_file(const unsigned char *sha1)
2464 {
2465 struct stat st;
2466 if (stat(sha1_pack_name(sha1), &st))
2467 return 0;
2468 return 1;
2469 }
2470
2471 int has_sha1_pack(const unsigned char *sha1)
2472 {
2473 struct pack_entry e;
2474 return find_pack_entry(sha1, &e);
2475 }
2476
2477 int has_sha1_file(const unsigned char *sha1)
2478 {
2479 struct pack_entry e;
2480
2481 if (find_pack_entry(sha1, &e))
2482 return 1;
2483 return has_loose_object(sha1);
2484 }
2485
2486 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2487 int write_object, enum object_type type, const char *path)
2488 {
2489 int ret, re_allocated = 0;
2490
2491 if (!type)
2492 type = OBJ_BLOB;
2493
2494 /*
2495 * Convert blobs to git internal format
2496 */
2497 if ((type == OBJ_BLOB) && path) {
2498 struct strbuf nbuf = STRBUF_INIT;
2499 if (convert_to_git(path, buf, size, &nbuf,
2500 write_object ? safe_crlf : 0)) {
2501 buf = strbuf_detach(&nbuf, &size);
2502 re_allocated = 1;
2503 }
2504 }
2505
2506 if (write_object)
2507 ret = write_sha1_file(buf, size, typename(type), sha1);
2508 else
2509 ret = hash_sha1_file(buf, size, typename(type), sha1);
2510 if (re_allocated)
2511 free(buf);
2512 return ret;
2513 }
2514
2515 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2516 enum object_type type, const char *path)
2517 {
2518 int ret;
2519 size_t size = xsize_t(st->st_size);
2520
2521 if (!S_ISREG(st->st_mode)) {
2522 struct strbuf sbuf = STRBUF_INIT;
2523 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2524 ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2525 type, path);
2526 else
2527 ret = -1;
2528 strbuf_release(&sbuf);
2529 } else if (size) {
2530 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2531 ret = index_mem(sha1, buf, size, write_object, type, path);
2532 munmap(buf, size);
2533 } else
2534 ret = index_mem(sha1, NULL, size, write_object, type, path);
2535 close(fd);
2536 return ret;
2537 }
2538
2539 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2540 {
2541 int fd;
2542 struct strbuf sb = STRBUF_INIT;
2543
2544 switch (st->st_mode & S_IFMT) {
2545 case S_IFREG:
2546 fd = open(path, O_RDONLY);
2547 if (fd < 0)
2548 return error("open(\"%s\"): %s", path,
2549 strerror(errno));
2550 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2551 return error("%s: failed to insert into database",
2552 path);
2553 break;
2554 case S_IFLNK:
2555 if (strbuf_readlink(&sb, path, st->st_size)) {
2556 char *errstr = strerror(errno);
2557 return error("readlink(\"%s\"): %s", path,
2558 errstr);
2559 }
2560 if (!write_object)
2561 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2562 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2563 return error("%s: failed to insert into database",
2564 path);
2565 strbuf_release(&sb);
2566 break;
2567 case S_IFDIR:
2568 return resolve_gitlink_ref(path, "HEAD", sha1);
2569 default:
2570 return error("%s: unsupported file type", path);
2571 }
2572 return 0;
2573 }
2574
2575 int read_pack_header(int fd, struct pack_header *header)
2576 {
2577 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2578 /* "eof before pack header was fully read" */
2579 return PH_ERROR_EOF;
2580
2581 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2582 /* "protocol error (pack signature mismatch detected)" */
2583 return PH_ERROR_PACK_SIGNATURE;
2584 if (!pack_version_ok(header->hdr_version))
2585 /* "protocol error (pack version unsupported)" */
2586 return PH_ERROR_PROTOCOL;
2587 return 0;
2588 }