Merge branch 'jc/maint-1.6.0-split-diff-metainfo' into maint-1.6.0
[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 *pos = 0;
103 if (!stat(path, &st)) {
104 /* path exists */
105 if (!S_ISDIR(st.st_mode)) {
106 *pos = '/';
107 return -3;
108 }
109 }
110 else if (mkdir(path, 0777)) {
111 *pos = '/';
112 return -1;
113 }
114 else if (adjust_shared_perm(path)) {
115 *pos = '/';
116 return -2;
117 }
118 *pos++ = '/';
119 }
120 return 0;
121 }
122
123 int safe_create_leading_directories_const(const char *path)
124 {
125 /* path points to cache entries, so xstrdup before messing with it */
126 char *buf = xstrdup(path);
127 int result = safe_create_leading_directories(buf);
128 free(buf);
129 return result;
130 }
131
132 char *sha1_to_hex(const unsigned char *sha1)
133 {
134 static int bufno;
135 static char hexbuffer[4][50];
136 static const char hex[] = "0123456789abcdef";
137 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
138 int i;
139
140 for (i = 0; i < 20; i++) {
141 unsigned int val = *sha1++;
142 *buf++ = hex[val >> 4];
143 *buf++ = hex[val & 0xf];
144 }
145 *buf = '\0';
146
147 return buffer;
148 }
149
150 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
151 {
152 int i;
153 for (i = 0; i < 20; i++) {
154 static char hex[] = "0123456789abcdef";
155 unsigned int val = sha1[i];
156 char *pos = pathbuf + i*2 + (i > 0);
157 *pos++ = hex[val >> 4];
158 *pos = hex[val & 0xf];
159 }
160 }
161
162 /*
163 * NOTE! This returns a statically allocated buffer, so you have to be
164 * careful about using it. Do an "xstrdup()" if you need to save the
165 * filename.
166 *
167 * Also note that this returns the location for creating. Reading
168 * SHA1 file can happen from any alternate directory listed in the
169 * DB_ENVIRONMENT environment variable if it is not found in
170 * the primary object database.
171 */
172 char *sha1_file_name(const unsigned char *sha1)
173 {
174 static char *name, *base;
175
176 if (!base) {
177 const char *sha1_file_directory = get_object_directory();
178 int len = strlen(sha1_file_directory);
179 base = xmalloc(len + 60);
180 memcpy(base, sha1_file_directory, len);
181 memset(base+len, 0, 60);
182 base[len] = '/';
183 base[len+3] = '/';
184 name = base + len + 1;
185 }
186 fill_sha1_path(name, sha1);
187 return base;
188 }
189
190 static char *sha1_get_pack_name(const unsigned char *sha1,
191 char **name, char **base, const char *which)
192 {
193 static const char hex[] = "0123456789abcdef";
194 char *buf;
195 int i;
196
197 if (!*base) {
198 const char *sha1_file_directory = get_object_directory();
199 int len = strlen(sha1_file_directory);
200 *base = xmalloc(len + 60);
201 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
202 sha1_file_directory, which);
203 *name = *base + len + 11;
204 }
205
206 buf = *name;
207
208 for (i = 0; i < 20; i++) {
209 unsigned int val = *sha1++;
210 *buf++ = hex[val >> 4];
211 *buf++ = hex[val & 0xf];
212 }
213
214 return *base;
215 }
216
217 char *sha1_pack_name(const unsigned char *sha1)
218 {
219 static char *name, *base;
220
221 return sha1_get_pack_name(sha1, &name, &base, "pack");
222 }
223
224 char *sha1_pack_index_name(const unsigned char *sha1)
225 {
226 static char *name, *base;
227
228 return sha1_get_pack_name(sha1, &name, &base, "idx");
229 }
230
231 struct alternate_object_database *alt_odb_list;
232 static struct alternate_object_database **alt_odb_tail;
233
234 static void read_info_alternates(const char * alternates, int depth);
235
236 /*
237 * Prepare alternate object database registry.
238 *
239 * The variable alt_odb_list points at the list of struct
240 * alternate_object_database. The elements on this list come from
241 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
242 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
243 * whose contents is similar to that environment variable but can be
244 * LF separated. Its base points at a statically allocated buffer that
245 * contains "/the/directory/corresponding/to/.git/objects/...", while
246 * its name points just after the slash at the end of ".git/objects/"
247 * in the example above, and has enough space to hold 40-byte hex
248 * SHA1, an extra slash for the first level indirection, and the
249 * terminating NUL.
250 */
251 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
252 {
253 struct stat st;
254 const char *objdir = get_object_directory();
255 struct alternate_object_database *ent;
256 struct alternate_object_database *alt;
257 /* 43 = 40-byte + 2 '/' + terminating NUL */
258 int pfxlen = len;
259 int entlen = pfxlen + 43;
260 int base_len = -1;
261
262 if (!is_absolute_path(entry) && relative_base) {
263 /* Relative alt-odb */
264 if (base_len < 0)
265 base_len = strlen(relative_base) + 1;
266 entlen += base_len;
267 pfxlen += base_len;
268 }
269 ent = xmalloc(sizeof(*ent) + entlen);
270
271 if (!is_absolute_path(entry) && relative_base) {
272 memcpy(ent->base, relative_base, base_len - 1);
273 ent->base[base_len - 1] = '/';
274 memcpy(ent->base + base_len, entry, len);
275 }
276 else
277 memcpy(ent->base, entry, pfxlen);
278
279 ent->name = ent->base + pfxlen + 1;
280 ent->base[pfxlen + 3] = '/';
281 ent->base[pfxlen] = ent->base[entlen-1] = 0;
282
283 /* Detect cases where alternate disappeared */
284 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
285 error("object directory %s does not exist; "
286 "check .git/objects/info/alternates.",
287 ent->base);
288 free(ent);
289 return -1;
290 }
291
292 /* Prevent the common mistake of listing the same
293 * thing twice, or object directory itself.
294 */
295 for (alt = alt_odb_list; alt; alt = alt->next) {
296 if (!memcmp(ent->base, alt->base, pfxlen)) {
297 free(ent);
298 return -1;
299 }
300 }
301 if (!memcmp(ent->base, objdir, pfxlen)) {
302 free(ent);
303 return -1;
304 }
305
306 /* add the alternate entry */
307 *alt_odb_tail = ent;
308 alt_odb_tail = &(ent->next);
309 ent->next = NULL;
310
311 /* recursively add alternates */
312 read_info_alternates(ent->base, depth + 1);
313
314 ent->base[pfxlen] = '/';
315
316 return 0;
317 }
318
319 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
320 const char *relative_base, int depth)
321 {
322 const char *cp, *last;
323
324 if (depth > 5) {
325 error("%s: ignoring alternate object stores, nesting too deep.",
326 relative_base);
327 return;
328 }
329
330 last = alt;
331 while (last < ep) {
332 cp = last;
333 if (cp < ep && *cp == '#') {
334 while (cp < ep && *cp != sep)
335 cp++;
336 last = cp + 1;
337 continue;
338 }
339 while (cp < ep && *cp != sep)
340 cp++;
341 if (last != cp) {
342 if (!is_absolute_path(last) && depth) {
343 error("%s: ignoring relative alternate object store %s",
344 relative_base, last);
345 } else {
346 link_alt_odb_entry(last, cp - last,
347 relative_base, depth);
348 }
349 }
350 while (cp < ep && *cp == sep)
351 cp++;
352 last = cp;
353 }
354 }
355
356 static void read_info_alternates(const char * relative_base, int depth)
357 {
358 char *map;
359 size_t mapsz;
360 struct stat st;
361 const char alt_file_name[] = "info/alternates";
362 /* Given that relative_base is no longer than PATH_MAX,
363 ensure that "path" has enough space to append "/", the
364 file name, "info/alternates", and a trailing NUL. */
365 char path[PATH_MAX + 1 + sizeof alt_file_name];
366 int fd;
367
368 sprintf(path, "%s/%s", relative_base, alt_file_name);
369 fd = open(path, O_RDONLY);
370 if (fd < 0)
371 return;
372 if (fstat(fd, &st) || (st.st_size == 0)) {
373 close(fd);
374 return;
375 }
376 mapsz = xsize_t(st.st_size);
377 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
378 close(fd);
379
380 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
381
382 munmap(map, mapsz);
383 }
384
385 void add_to_alternates_file(const char *reference)
386 {
387 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
388 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
389 char *alt = mkpath("%s/objects\n", reference);
390 write_or_die(fd, alt, strlen(alt));
391 if (commit_lock_file(lock))
392 die("could not close alternates file");
393 if (alt_odb_tail)
394 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
395 }
396
397 void prepare_alt_odb(void)
398 {
399 const char *alt;
400
401 if (alt_odb_tail)
402 return;
403
404 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
405 if (!alt) alt = "";
406
407 alt_odb_tail = &alt_odb_list;
408 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
409
410 read_info_alternates(get_object_directory(), 0);
411 }
412
413 static int has_loose_object_local(const unsigned char *sha1)
414 {
415 char *name = sha1_file_name(sha1);
416 return !access(name, F_OK);
417 }
418
419 int has_loose_object_nonlocal(const unsigned char *sha1)
420 {
421 struct alternate_object_database *alt;
422 prepare_alt_odb();
423 for (alt = alt_odb_list; alt; alt = alt->next) {
424 fill_sha1_path(alt->name, sha1);
425 if (!access(alt->base, F_OK))
426 return 1;
427 }
428 return 0;
429 }
430
431 static int has_loose_object(const unsigned char *sha1)
432 {
433 return has_loose_object_local(sha1) ||
434 has_loose_object_nonlocal(sha1);
435 }
436
437 static unsigned int pack_used_ctr;
438 static unsigned int pack_mmap_calls;
439 static unsigned int peak_pack_open_windows;
440 static unsigned int pack_open_windows;
441 static size_t peak_pack_mapped;
442 static size_t pack_mapped;
443 struct packed_git *packed_git;
444
445 void pack_report(void)
446 {
447 fprintf(stderr,
448 "pack_report: getpagesize() = %10" SZ_FMT "\n"
449 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
450 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
451 sz_fmt(getpagesize()),
452 sz_fmt(packed_git_window_size),
453 sz_fmt(packed_git_limit));
454 fprintf(stderr,
455 "pack_report: pack_used_ctr = %10u\n"
456 "pack_report: pack_mmap_calls = %10u\n"
457 "pack_report: pack_open_windows = %10u / %10u\n"
458 "pack_report: pack_mapped = "
459 "%10" SZ_FMT " / %10" SZ_FMT "\n",
460 pack_used_ctr,
461 pack_mmap_calls,
462 pack_open_windows, peak_pack_open_windows,
463 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
464 }
465
466 static int check_packed_git_idx(const char *path, struct packed_git *p)
467 {
468 void *idx_map;
469 struct pack_idx_header *hdr;
470 size_t idx_size;
471 uint32_t version, nr, i, *index;
472 int fd = open(path, O_RDONLY);
473 struct stat st;
474
475 if (fd < 0)
476 return -1;
477 if (fstat(fd, &st)) {
478 close(fd);
479 return -1;
480 }
481 idx_size = xsize_t(st.st_size);
482 if (idx_size < 4 * 256 + 20 + 20) {
483 close(fd);
484 return error("index file %s is too small", path);
485 }
486 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
487 close(fd);
488
489 hdr = idx_map;
490 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
491 version = ntohl(hdr->idx_version);
492 if (version < 2 || version > 2) {
493 munmap(idx_map, idx_size);
494 return error("index file %s is version %"PRIu32
495 " and is not supported by this binary"
496 " (try upgrading GIT to a newer version)",
497 path, version);
498 }
499 } else
500 version = 1;
501
502 nr = 0;
503 index = idx_map;
504 if (version > 1)
505 index += 2; /* skip index header */
506 for (i = 0; i < 256; i++) {
507 uint32_t n = ntohl(index[i]);
508 if (n < nr) {
509 munmap(idx_map, idx_size);
510 return error("non-monotonic index %s", path);
511 }
512 nr = n;
513 }
514
515 if (version == 1) {
516 /*
517 * Total size:
518 * - 256 index entries 4 bytes each
519 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
520 * - 20-byte SHA1 of the packfile
521 * - 20-byte SHA1 file checksum
522 */
523 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
524 munmap(idx_map, idx_size);
525 return error("wrong index v1 file size in %s", path);
526 }
527 } else if (version == 2) {
528 /*
529 * Minimum size:
530 * - 8 bytes of header
531 * - 256 index entries 4 bytes each
532 * - 20-byte sha1 entry * nr
533 * - 4-byte crc entry * nr
534 * - 4-byte offset entry * nr
535 * - 20-byte SHA1 of the packfile
536 * - 20-byte SHA1 file checksum
537 * And after the 4-byte offset table might be a
538 * variable sized table containing 8-byte entries
539 * for offsets larger than 2^31.
540 */
541 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
542 unsigned long max_size = min_size;
543 if (nr)
544 max_size += (nr - 1)*8;
545 if (idx_size < min_size || idx_size > max_size) {
546 munmap(idx_map, idx_size);
547 return error("wrong index v2 file size in %s", path);
548 }
549 if (idx_size != min_size &&
550 /*
551 * make sure we can deal with large pack offsets.
552 * 31-bit signed offset won't be enough, neither
553 * 32-bit unsigned one will be.
554 */
555 (sizeof(off_t) <= 4)) {
556 munmap(idx_map, idx_size);
557 return error("pack too large for current definition of off_t in %s", path);
558 }
559 }
560
561 p->index_version = version;
562 p->index_data = idx_map;
563 p->index_size = idx_size;
564 p->num_objects = nr;
565 return 0;
566 }
567
568 int open_pack_index(struct packed_git *p)
569 {
570 char *idx_name;
571 int ret;
572
573 if (p->index_data)
574 return 0;
575
576 idx_name = xstrdup(p->pack_name);
577 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
578 ret = check_packed_git_idx(idx_name, p);
579 free(idx_name);
580 return ret;
581 }
582
583 static void scan_windows(struct packed_git *p,
584 struct packed_git **lru_p,
585 struct pack_window **lru_w,
586 struct pack_window **lru_l)
587 {
588 struct pack_window *w, *w_l;
589
590 for (w_l = NULL, w = p->windows; w; w = w->next) {
591 if (!w->inuse_cnt) {
592 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
593 *lru_p = p;
594 *lru_w = w;
595 *lru_l = w_l;
596 }
597 }
598 w_l = w;
599 }
600 }
601
602 static int unuse_one_window(struct packed_git *current, int keep_fd)
603 {
604 struct packed_git *p, *lru_p = NULL;
605 struct pack_window *lru_w = NULL, *lru_l = NULL;
606
607 if (current)
608 scan_windows(current, &lru_p, &lru_w, &lru_l);
609 for (p = packed_git; p; p = p->next)
610 scan_windows(p, &lru_p, &lru_w, &lru_l);
611 if (lru_p) {
612 munmap(lru_w->base, lru_w->len);
613 pack_mapped -= lru_w->len;
614 if (lru_l)
615 lru_l->next = lru_w->next;
616 else {
617 lru_p->windows = lru_w->next;
618 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
619 close(lru_p->pack_fd);
620 lru_p->pack_fd = -1;
621 }
622 }
623 free(lru_w);
624 pack_open_windows--;
625 return 1;
626 }
627 return 0;
628 }
629
630 void release_pack_memory(size_t need, int fd)
631 {
632 size_t cur = pack_mapped;
633 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
634 ; /* nothing */
635 }
636
637 void close_pack_windows(struct packed_git *p)
638 {
639 while (p->windows) {
640 struct pack_window *w = p->windows;
641
642 if (w->inuse_cnt)
643 die("pack '%s' still has open windows to it",
644 p->pack_name);
645 munmap(w->base, w->len);
646 pack_mapped -= w->len;
647 pack_open_windows--;
648 p->windows = w->next;
649 free(w);
650 }
651 }
652
653 void unuse_pack(struct pack_window **w_cursor)
654 {
655 struct pack_window *w = *w_cursor;
656 if (w) {
657 w->inuse_cnt--;
658 *w_cursor = NULL;
659 }
660 }
661
662 /*
663 * This is used by git-repack in case a newly created pack happens to
664 * contain the same set of objects as an existing one. In that case
665 * the resulting file might be different even if its name would be the
666 * same. It is best to close any reference to the old pack before it is
667 * replaced on disk. Of course no index pointers nor windows for given pack
668 * must subsist at this point. If ever objects from this pack are requested
669 * again, the new version of the pack will be reinitialized through
670 * reprepare_packed_git().
671 */
672 void free_pack_by_name(const char *pack_name)
673 {
674 struct packed_git *p, **pp = &packed_git;
675
676 while (*pp) {
677 p = *pp;
678 if (strcmp(pack_name, p->pack_name) == 0) {
679 clear_delta_base_cache();
680 close_pack_windows(p);
681 if (p->pack_fd != -1)
682 close(p->pack_fd);
683 if (p->index_data)
684 munmap((void *)p->index_data, p->index_size);
685 free(p->bad_object_sha1);
686 *pp = p->next;
687 free(p);
688 return;
689 }
690 pp = &p->next;
691 }
692 }
693
694 /*
695 * Do not call this directly as this leaks p->pack_fd on error return;
696 * call open_packed_git() instead.
697 */
698 static int open_packed_git_1(struct packed_git *p)
699 {
700 struct stat st;
701 struct pack_header hdr;
702 unsigned char sha1[20];
703 unsigned char *idx_sha1;
704 long fd_flag;
705
706 if (!p->index_data && open_pack_index(p))
707 return error("packfile %s index unavailable", p->pack_name);
708
709 p->pack_fd = open(p->pack_name, O_RDONLY);
710 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
711 return -1;
712
713 /* If we created the struct before we had the pack we lack size. */
714 if (!p->pack_size) {
715 if (!S_ISREG(st.st_mode))
716 return error("packfile %s not a regular file", p->pack_name);
717 p->pack_size = st.st_size;
718 } else if (p->pack_size != st.st_size)
719 return error("packfile %s size changed", p->pack_name);
720
721 /* We leave these file descriptors open with sliding mmap;
722 * there is no point keeping them open across exec(), though.
723 */
724 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
725 if (fd_flag < 0)
726 return error("cannot determine file descriptor flags");
727 fd_flag |= FD_CLOEXEC;
728 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
729 return error("cannot set FD_CLOEXEC");
730
731 /* Verify we recognize this pack file format. */
732 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
733 return error("file %s is far too short to be a packfile", p->pack_name);
734 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
735 return error("file %s is not a GIT packfile", p->pack_name);
736 if (!pack_version_ok(hdr.hdr_version))
737 return error("packfile %s is version %"PRIu32" and not"
738 " supported (try upgrading GIT to a newer version)",
739 p->pack_name, ntohl(hdr.hdr_version));
740
741 /* Verify the pack matches its index. */
742 if (p->num_objects != ntohl(hdr.hdr_entries))
743 return error("packfile %s claims to have %"PRIu32" objects"
744 " while index indicates %"PRIu32" objects",
745 p->pack_name, ntohl(hdr.hdr_entries),
746 p->num_objects);
747 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
748 return error("end of packfile %s is unavailable", p->pack_name);
749 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
750 return error("packfile %s signature is unavailable", p->pack_name);
751 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
752 if (hashcmp(sha1, idx_sha1))
753 return error("packfile %s does not match index", p->pack_name);
754 return 0;
755 }
756
757 static int open_packed_git(struct packed_git *p)
758 {
759 if (!open_packed_git_1(p))
760 return 0;
761 if (p->pack_fd != -1) {
762 close(p->pack_fd);
763 p->pack_fd = -1;
764 }
765 return -1;
766 }
767
768 static int in_window(struct pack_window *win, off_t offset)
769 {
770 /* We must promise at least 20 bytes (one hash) after the
771 * offset is available from this window, otherwise the offset
772 * is not actually in this window and a different window (which
773 * has that one hash excess) must be used. This is to support
774 * the object header and delta base parsing routines below.
775 */
776 off_t win_off = win->offset;
777 return win_off <= offset
778 && (offset + 20) <= (win_off + win->len);
779 }
780
781 unsigned char* use_pack(struct packed_git *p,
782 struct pack_window **w_cursor,
783 off_t offset,
784 unsigned int *left)
785 {
786 struct pack_window *win = *w_cursor;
787
788 if (p->pack_fd == -1 && open_packed_git(p))
789 die("packfile %s cannot be accessed", p->pack_name);
790
791 /* Since packfiles end in a hash of their content and its
792 * pointless to ask for an offset into the middle of that
793 * hash, and the in_window function above wouldn't match
794 * don't allow an offset too close to the end of the file.
795 */
796 if (offset > (p->pack_size - 20))
797 die("offset beyond end of packfile (truncated pack?)");
798
799 if (!win || !in_window(win, offset)) {
800 if (win)
801 win->inuse_cnt--;
802 for (win = p->windows; win; win = win->next) {
803 if (in_window(win, offset))
804 break;
805 }
806 if (!win) {
807 size_t window_align = packed_git_window_size / 2;
808 off_t len;
809 win = xcalloc(1, sizeof(*win));
810 win->offset = (offset / window_align) * window_align;
811 len = p->pack_size - win->offset;
812 if (len > packed_git_window_size)
813 len = packed_git_window_size;
814 win->len = (size_t)len;
815 pack_mapped += win->len;
816 while (packed_git_limit < pack_mapped
817 && unuse_one_window(p, p->pack_fd))
818 ; /* nothing */
819 win->base = xmmap(NULL, win->len,
820 PROT_READ, MAP_PRIVATE,
821 p->pack_fd, win->offset);
822 if (win->base == MAP_FAILED)
823 die("packfile %s cannot be mapped: %s",
824 p->pack_name,
825 strerror(errno));
826 pack_mmap_calls++;
827 pack_open_windows++;
828 if (pack_mapped > peak_pack_mapped)
829 peak_pack_mapped = pack_mapped;
830 if (pack_open_windows > peak_pack_open_windows)
831 peak_pack_open_windows = pack_open_windows;
832 win->next = p->windows;
833 p->windows = win;
834 }
835 }
836 if (win != *w_cursor) {
837 win->last_used = pack_used_ctr++;
838 win->inuse_cnt++;
839 *w_cursor = win;
840 }
841 offset -= win->offset;
842 if (left)
843 *left = win->len - xsize_t(offset);
844 return win->base + offset;
845 }
846
847 static struct packed_git *alloc_packed_git(int extra)
848 {
849 struct packed_git *p = xmalloc(sizeof(*p) + extra);
850 memset(p, 0, sizeof(*p));
851 p->pack_fd = -1;
852 return p;
853 }
854
855 struct packed_git *add_packed_git(const char *path, int path_len, int local)
856 {
857 struct stat st;
858 struct packed_git *p = alloc_packed_git(path_len + 2);
859
860 /*
861 * Make sure a corresponding .pack file exists and that
862 * the index looks sane.
863 */
864 path_len -= strlen(".idx");
865 if (path_len < 1) {
866 free(p);
867 return NULL;
868 }
869 memcpy(p->pack_name, path, path_len);
870
871 strcpy(p->pack_name + path_len, ".keep");
872 if (!access(p->pack_name, F_OK))
873 p->pack_keep = 1;
874
875 strcpy(p->pack_name + path_len, ".pack");
876 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
877 free(p);
878 return NULL;
879 }
880
881 /* ok, it looks sane as far as we can check without
882 * actually mapping the pack file.
883 */
884 p->pack_size = st.st_size;
885 p->pack_local = local;
886 p->mtime = st.st_mtime;
887 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
888 hashclr(p->sha1);
889 return p;
890 }
891
892 struct packed_git *parse_pack_index(unsigned char *sha1)
893 {
894 const char *idx_path = sha1_pack_index_name(sha1);
895 const char *path = sha1_pack_name(sha1);
896 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
897
898 strcpy(p->pack_name, path);
899 hashcpy(p->sha1, sha1);
900 if (check_packed_git_idx(idx_path, p)) {
901 free(p);
902 return NULL;
903 }
904
905 return p;
906 }
907
908 void install_packed_git(struct packed_git *pack)
909 {
910 pack->next = packed_git;
911 packed_git = pack;
912 }
913
914 static void prepare_packed_git_one(char *objdir, int local)
915 {
916 /* Ensure that this buffer is large enough so that we can
917 append "/pack/" without clobbering the stack even if
918 strlen(objdir) were PATH_MAX. */
919 char path[PATH_MAX + 1 + 4 + 1 + 1];
920 int len;
921 DIR *dir;
922 struct dirent *de;
923
924 sprintf(path, "%s/pack", objdir);
925 len = strlen(path);
926 dir = opendir(path);
927 if (!dir) {
928 if (errno != ENOENT)
929 error("unable to open object pack directory: %s: %s",
930 path, strerror(errno));
931 return;
932 }
933 path[len++] = '/';
934 while ((de = readdir(dir)) != NULL) {
935 int namelen = strlen(de->d_name);
936 struct packed_git *p;
937
938 if (!has_extension(de->d_name, ".idx"))
939 continue;
940
941 if (len + namelen + 1 > sizeof(path))
942 continue;
943
944 /* Don't reopen a pack we already have. */
945 strcpy(path + len, de->d_name);
946 for (p = packed_git; p; p = p->next) {
947 if (!memcmp(path, p->pack_name, len + namelen - 4))
948 break;
949 }
950 if (p)
951 continue;
952 /* See if it really is a valid .idx file with corresponding
953 * .pack file that we can map.
954 */
955 p = add_packed_git(path, len + namelen, local);
956 if (!p)
957 continue;
958 install_packed_git(p);
959 }
960 closedir(dir);
961 }
962
963 static int sort_pack(const void *a_, const void *b_)
964 {
965 struct packed_git *a = *((struct packed_git **)a_);
966 struct packed_git *b = *((struct packed_git **)b_);
967 int st;
968
969 /*
970 * Local packs tend to contain objects specific to our
971 * variant of the project than remote ones. In addition,
972 * remote ones could be on a network mounted filesystem.
973 * Favor local ones for these reasons.
974 */
975 st = a->pack_local - b->pack_local;
976 if (st)
977 return -st;
978
979 /*
980 * Younger packs tend to contain more recent objects,
981 * and more recent objects tend to get accessed more
982 * often.
983 */
984 if (a->mtime < b->mtime)
985 return 1;
986 else if (a->mtime == b->mtime)
987 return 0;
988 return -1;
989 }
990
991 static void rearrange_packed_git(void)
992 {
993 struct packed_git **ary, *p;
994 int i, n;
995
996 for (n = 0, p = packed_git; p; p = p->next)
997 n++;
998 if (n < 2)
999 return;
1000
1001 /* prepare an array of packed_git for easier sorting */
1002 ary = xcalloc(n, sizeof(struct packed_git *));
1003 for (n = 0, p = packed_git; p; p = p->next)
1004 ary[n++] = p;
1005
1006 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1007
1008 /* link them back again */
1009 for (i = 0; i < n - 1; i++)
1010 ary[i]->next = ary[i + 1];
1011 ary[n - 1]->next = NULL;
1012 packed_git = ary[0];
1013
1014 free(ary);
1015 }
1016
1017 static int prepare_packed_git_run_once = 0;
1018 void prepare_packed_git(void)
1019 {
1020 struct alternate_object_database *alt;
1021
1022 if (prepare_packed_git_run_once)
1023 return;
1024 prepare_packed_git_one(get_object_directory(), 1);
1025 prepare_alt_odb();
1026 for (alt = alt_odb_list; alt; alt = alt->next) {
1027 alt->name[-1] = 0;
1028 prepare_packed_git_one(alt->base, 0);
1029 alt->name[-1] = '/';
1030 }
1031 rearrange_packed_git();
1032 prepare_packed_git_run_once = 1;
1033 }
1034
1035 void reprepare_packed_git(void)
1036 {
1037 discard_revindex();
1038 prepare_packed_git_run_once = 0;
1039 prepare_packed_git();
1040 }
1041
1042 static void mark_bad_packed_object(struct packed_git *p,
1043 const unsigned char *sha1)
1044 {
1045 unsigned i;
1046 for (i = 0; i < p->num_bad_objects; i++)
1047 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1048 return;
1049 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1050 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1051 p->num_bad_objects++;
1052 }
1053
1054 static int has_packed_and_bad(const unsigned char *sha1)
1055 {
1056 struct packed_git *p;
1057 unsigned i;
1058
1059 for (p = packed_git; p; p = p->next)
1060 for (i = 0; i < p->num_bad_objects; i++)
1061 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1062 return 1;
1063 return 0;
1064 }
1065
1066 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1067 {
1068 unsigned char real_sha1[20];
1069 hash_sha1_file(map, size, type, real_sha1);
1070 return hashcmp(sha1, real_sha1) ? -1 : 0;
1071 }
1072
1073 static int git_open_noatime(const char *name)
1074 {
1075 static int sha1_file_open_flag = O_NOATIME;
1076 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1077
1078 /* Might the failure be due to O_NOATIME? */
1079 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1080 fd = open(name, O_RDONLY);
1081 if (fd >= 0)
1082 sha1_file_open_flag = 0;
1083 }
1084 return fd;
1085 }
1086
1087 static int open_sha1_file(const unsigned char *sha1)
1088 {
1089 int fd;
1090 char *name = sha1_file_name(sha1);
1091 struct alternate_object_database *alt;
1092
1093 fd = git_open_noatime(name);
1094 if (fd >= 0)
1095 return fd;
1096
1097 prepare_alt_odb();
1098 errno = ENOENT;
1099 for (alt = alt_odb_list; alt; alt = alt->next) {
1100 name = alt->name;
1101 fill_sha1_path(name, sha1);
1102 fd = git_open_noatime(alt->base);
1103 if (fd >= 0)
1104 return fd;
1105 }
1106 return -1;
1107 }
1108
1109 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1110 {
1111 void *map;
1112 int fd;
1113
1114 fd = open_sha1_file(sha1);
1115 map = NULL;
1116 if (fd >= 0) {
1117 struct stat st;
1118
1119 if (!fstat(fd, &st)) {
1120 *size = xsize_t(st.st_size);
1121 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1122 }
1123 close(fd);
1124 }
1125 return map;
1126 }
1127
1128 static int legacy_loose_object(unsigned char *map)
1129 {
1130 unsigned int word;
1131
1132 /*
1133 * Is it a zlib-compressed buffer? If so, the first byte
1134 * must be 0x78 (15-bit window size, deflated), and the
1135 * first 16-bit word is evenly divisible by 31
1136 */
1137 word = (map[0] << 8) + map[1];
1138 if (map[0] == 0x78 && !(word % 31))
1139 return 1;
1140 else
1141 return 0;
1142 }
1143
1144 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1145 {
1146 unsigned shift;
1147 unsigned char c;
1148 unsigned long size;
1149 unsigned long used = 0;
1150
1151 c = buf[used++];
1152 *type = (c >> 4) & 7;
1153 size = c & 15;
1154 shift = 4;
1155 while (c & 0x80) {
1156 if (len <= used)
1157 return 0;
1158 if (sizeof(long) * 8 <= shift)
1159 return 0;
1160 c = buf[used++];
1161 size += (c & 0x7f) << shift;
1162 shift += 7;
1163 }
1164 *sizep = size;
1165 return used;
1166 }
1167
1168 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1169 {
1170 unsigned long size, used;
1171 static const char valid_loose_object_type[8] = {
1172 0, /* OBJ_EXT */
1173 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1174 0, /* "delta" and others are invalid in a loose object */
1175 };
1176 enum object_type type;
1177
1178 /* Get the data stream */
1179 memset(stream, 0, sizeof(*stream));
1180 stream->next_in = map;
1181 stream->avail_in = mapsize;
1182 stream->next_out = buffer;
1183 stream->avail_out = bufsiz;
1184
1185 if (legacy_loose_object(map)) {
1186 inflateInit(stream);
1187 return inflate(stream, 0);
1188 }
1189
1190
1191 /*
1192 * There used to be a second loose object header format which
1193 * was meant to mimic the in-pack format, allowing for direct
1194 * copy of the object data. This format turned up not to be
1195 * really worth it and we don't write it any longer. But we
1196 * can still read it.
1197 */
1198 used = unpack_object_header_gently(map, mapsize, &type, &size);
1199 if (!used || !valid_loose_object_type[type])
1200 return -1;
1201 map += used;
1202 mapsize -= used;
1203
1204 /* Set up the stream for the rest.. */
1205 stream->next_in = map;
1206 stream->avail_in = mapsize;
1207 inflateInit(stream);
1208
1209 /* And generate the fake traditional header */
1210 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1211 typename(type), size);
1212 return 0;
1213 }
1214
1215 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1216 {
1217 int bytes = strlen(buffer) + 1;
1218 unsigned char *buf = xmalloc(1+size);
1219 unsigned long n;
1220 int status = Z_OK;
1221
1222 n = stream->total_out - bytes;
1223 if (n > size)
1224 n = size;
1225 memcpy(buf, (char *) buffer + bytes, n);
1226 bytes = n;
1227 if (bytes <= size) {
1228 /*
1229 * The above condition must be (bytes <= size), not
1230 * (bytes < size). In other words, even though we
1231 * expect no more output and set avail_out to zer0,
1232 * the input zlib stream may have bytes that express
1233 * "this concludes the stream", and we *do* want to
1234 * eat that input.
1235 *
1236 * Otherwise we would not be able to test that we
1237 * consumed all the input to reach the expected size;
1238 * we also want to check that zlib tells us that all
1239 * went well with status == Z_STREAM_END at the end.
1240 */
1241 stream->next_out = buf + bytes;
1242 stream->avail_out = size - bytes;
1243 while (status == Z_OK)
1244 status = inflate(stream, Z_FINISH);
1245 }
1246 buf[size] = 0;
1247 if (status == Z_STREAM_END && !stream->avail_in) {
1248 inflateEnd(stream);
1249 return buf;
1250 }
1251
1252 if (status < 0)
1253 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1254 else if (stream->avail_in)
1255 error("garbage at end of loose object '%s'",
1256 sha1_to_hex(sha1));
1257 free(buf);
1258 return NULL;
1259 }
1260
1261 /*
1262 * We used to just use "sscanf()", but that's actually way
1263 * too permissive for what we want to check. So do an anal
1264 * object header parse by hand.
1265 */
1266 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1267 {
1268 char type[10];
1269 int i;
1270 unsigned long size;
1271
1272 /*
1273 * The type can be at most ten bytes (including the
1274 * terminating '\0' that we add), and is followed by
1275 * a space.
1276 */
1277 i = 0;
1278 for (;;) {
1279 char c = *hdr++;
1280 if (c == ' ')
1281 break;
1282 type[i++] = c;
1283 if (i >= sizeof(type))
1284 return -1;
1285 }
1286 type[i] = 0;
1287
1288 /*
1289 * The length must follow immediately, and be in canonical
1290 * decimal format (ie "010" is not valid).
1291 */
1292 size = *hdr++ - '0';
1293 if (size > 9)
1294 return -1;
1295 if (size) {
1296 for (;;) {
1297 unsigned long c = *hdr - '0';
1298 if (c > 9)
1299 break;
1300 hdr++;
1301 size = size * 10 + c;
1302 }
1303 }
1304 *sizep = size;
1305
1306 /*
1307 * The length must be followed by a zero byte
1308 */
1309 return *hdr ? -1 : type_from_string(type);
1310 }
1311
1312 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1313 {
1314 int ret;
1315 z_stream stream;
1316 char hdr[8192];
1317
1318 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1319 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1320 return NULL;
1321
1322 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1323 }
1324
1325 unsigned long get_size_from_delta(struct packed_git *p,
1326 struct pack_window **w_curs,
1327 off_t curpos)
1328 {
1329 const unsigned char *data;
1330 unsigned char delta_head[20], *in;
1331 z_stream stream;
1332 int st;
1333
1334 memset(&stream, 0, sizeof(stream));
1335 stream.next_out = delta_head;
1336 stream.avail_out = sizeof(delta_head);
1337
1338 inflateInit(&stream);
1339 do {
1340 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1341 stream.next_in = in;
1342 st = inflate(&stream, Z_FINISH);
1343 curpos += stream.next_in - in;
1344 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1345 stream.total_out < sizeof(delta_head));
1346 inflateEnd(&stream);
1347 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1348 die("delta data unpack-initial failed");
1349
1350 /* Examine the initial part of the delta to figure out
1351 * the result size.
1352 */
1353 data = delta_head;
1354
1355 /* ignore base size */
1356 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1357
1358 /* Read the result size */
1359 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1360 }
1361
1362 static off_t get_delta_base(struct packed_git *p,
1363 struct pack_window **w_curs,
1364 off_t *curpos,
1365 enum object_type type,
1366 off_t delta_obj_offset)
1367 {
1368 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1369 off_t base_offset;
1370
1371 /* use_pack() assured us we have [base_info, base_info + 20)
1372 * as a range that we can look at without walking off the
1373 * end of the mapped window. Its actually the hash size
1374 * that is assured. An OFS_DELTA longer than the hash size
1375 * is stupid, as then a REF_DELTA would be smaller to store.
1376 */
1377 if (type == OBJ_OFS_DELTA) {
1378 unsigned used = 0;
1379 unsigned char c = base_info[used++];
1380 base_offset = c & 127;
1381 while (c & 128) {
1382 base_offset += 1;
1383 if (!base_offset || MSB(base_offset, 7))
1384 return 0; /* overflow */
1385 c = base_info[used++];
1386 base_offset = (base_offset << 7) + (c & 127);
1387 }
1388 base_offset = delta_obj_offset - base_offset;
1389 if (base_offset >= delta_obj_offset)
1390 return 0; /* out of bound */
1391 *curpos += used;
1392 } else if (type == OBJ_REF_DELTA) {
1393 /* The base entry _must_ be in the same pack */
1394 base_offset = find_pack_entry_one(base_info, p);
1395 *curpos += 20;
1396 } else
1397 die("I am totally screwed");
1398 return base_offset;
1399 }
1400
1401 /* forward declaration for a mutually recursive function */
1402 static int packed_object_info(struct packed_git *p, off_t offset,
1403 unsigned long *sizep);
1404
1405 static int packed_delta_info(struct packed_git *p,
1406 struct pack_window **w_curs,
1407 off_t curpos,
1408 enum object_type type,
1409 off_t obj_offset,
1410 unsigned long *sizep)
1411 {
1412 off_t base_offset;
1413
1414 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1415 type = packed_object_info(p, base_offset, NULL);
1416
1417 /* We choose to only get the type of the base object and
1418 * ignore potentially corrupt pack file that expects the delta
1419 * based on a base with a wrong size. This saves tons of
1420 * inflate() calls.
1421 */
1422 if (sizep)
1423 *sizep = get_size_from_delta(p, w_curs, curpos);
1424
1425 return type;
1426 }
1427
1428 static int unpack_object_header(struct packed_git *p,
1429 struct pack_window **w_curs,
1430 off_t *curpos,
1431 unsigned long *sizep)
1432 {
1433 unsigned char *base;
1434 unsigned int left;
1435 unsigned long used;
1436 enum object_type type;
1437
1438 /* use_pack() assures us we have [base, base + 20) available
1439 * as a range that we can look at at. (Its actually the hash
1440 * size that is assured.) With our object header encoding
1441 * the maximum deflated object size is 2^137, which is just
1442 * insane, so we know won't exceed what we have been given.
1443 */
1444 base = use_pack(p, w_curs, *curpos, &left);
1445 used = unpack_object_header_gently(base, left, &type, sizep);
1446 if (!used)
1447 die("object offset outside of pack file");
1448 *curpos += used;
1449
1450 return type;
1451 }
1452
1453 const char *packed_object_info_detail(struct packed_git *p,
1454 off_t obj_offset,
1455 unsigned long *size,
1456 unsigned long *store_size,
1457 unsigned int *delta_chain_length,
1458 unsigned char *base_sha1)
1459 {
1460 struct pack_window *w_curs = NULL;
1461 off_t curpos;
1462 unsigned long dummy;
1463 unsigned char *next_sha1;
1464 enum object_type type;
1465 struct revindex_entry *revidx;
1466
1467 *delta_chain_length = 0;
1468 curpos = obj_offset;
1469 type = unpack_object_header(p, &w_curs, &curpos, size);
1470
1471 revidx = find_pack_revindex(p, obj_offset);
1472 *store_size = revidx[1].offset - obj_offset;
1473
1474 for (;;) {
1475 switch (type) {
1476 default:
1477 die("pack %s contains unknown object type %d",
1478 p->pack_name, type);
1479 case OBJ_COMMIT:
1480 case OBJ_TREE:
1481 case OBJ_BLOB:
1482 case OBJ_TAG:
1483 unuse_pack(&w_curs);
1484 return typename(type);
1485 case OBJ_OFS_DELTA:
1486 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1487 if (!obj_offset)
1488 die("pack %s contains bad delta base reference of type %s",
1489 p->pack_name, typename(type));
1490 if (*delta_chain_length == 0) {
1491 revidx = find_pack_revindex(p, obj_offset);
1492 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1493 }
1494 break;
1495 case OBJ_REF_DELTA:
1496 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1497 if (*delta_chain_length == 0)
1498 hashcpy(base_sha1, next_sha1);
1499 obj_offset = find_pack_entry_one(next_sha1, p);
1500 break;
1501 }
1502 (*delta_chain_length)++;
1503 curpos = obj_offset;
1504 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1505 }
1506 }
1507
1508 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1509 unsigned long *sizep)
1510 {
1511 struct pack_window *w_curs = NULL;
1512 unsigned long size;
1513 off_t curpos = obj_offset;
1514 enum object_type type;
1515
1516 type = unpack_object_header(p, &w_curs, &curpos, &size);
1517
1518 switch (type) {
1519 case OBJ_OFS_DELTA:
1520 case OBJ_REF_DELTA:
1521 type = packed_delta_info(p, &w_curs, curpos,
1522 type, obj_offset, sizep);
1523 break;
1524 case OBJ_COMMIT:
1525 case OBJ_TREE:
1526 case OBJ_BLOB:
1527 case OBJ_TAG:
1528 if (sizep)
1529 *sizep = size;
1530 break;
1531 default:
1532 die("pack %s contains unknown object type %d",
1533 p->pack_name, type);
1534 }
1535 unuse_pack(&w_curs);
1536 return type;
1537 }
1538
1539 static void *unpack_compressed_entry(struct packed_git *p,
1540 struct pack_window **w_curs,
1541 off_t curpos,
1542 unsigned long size)
1543 {
1544 int st;
1545 z_stream stream;
1546 unsigned char *buffer, *in;
1547
1548 buffer = xmalloc(size + 1);
1549 buffer[size] = 0;
1550 memset(&stream, 0, sizeof(stream));
1551 stream.next_out = buffer;
1552 stream.avail_out = size;
1553
1554 inflateInit(&stream);
1555 do {
1556 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1557 stream.next_in = in;
1558 st = inflate(&stream, Z_FINISH);
1559 curpos += stream.next_in - in;
1560 } while (st == Z_OK || st == Z_BUF_ERROR);
1561 inflateEnd(&stream);
1562 if ((st != Z_STREAM_END) || stream.total_out != size) {
1563 free(buffer);
1564 return NULL;
1565 }
1566
1567 return buffer;
1568 }
1569
1570 #define MAX_DELTA_CACHE (256)
1571
1572 static size_t delta_base_cached;
1573
1574 static struct delta_base_cache_lru_list {
1575 struct delta_base_cache_lru_list *prev;
1576 struct delta_base_cache_lru_list *next;
1577 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1578
1579 static struct delta_base_cache_entry {
1580 struct delta_base_cache_lru_list lru;
1581 void *data;
1582 struct packed_git *p;
1583 off_t base_offset;
1584 unsigned long size;
1585 enum object_type type;
1586 } delta_base_cache[MAX_DELTA_CACHE];
1587
1588 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1589 {
1590 unsigned long hash;
1591
1592 hash = (unsigned long)p + (unsigned long)base_offset;
1593 hash += (hash >> 8) + (hash >> 16);
1594 return hash % MAX_DELTA_CACHE;
1595 }
1596
1597 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1598 unsigned long *base_size, enum object_type *type, int keep_cache)
1599 {
1600 void *ret;
1601 unsigned long hash = pack_entry_hash(p, base_offset);
1602 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1603
1604 ret = ent->data;
1605 if (ret && ent->p == p && ent->base_offset == base_offset)
1606 goto found_cache_entry;
1607 return unpack_entry(p, base_offset, type, base_size);
1608
1609 found_cache_entry:
1610 if (!keep_cache) {
1611 ent->data = NULL;
1612 ent->lru.next->prev = ent->lru.prev;
1613 ent->lru.prev->next = ent->lru.next;
1614 delta_base_cached -= ent->size;
1615 } else {
1616 ret = xmemdupz(ent->data, ent->size);
1617 }
1618 *type = ent->type;
1619 *base_size = ent->size;
1620 return ret;
1621 }
1622
1623 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1624 {
1625 if (ent->data) {
1626 free(ent->data);
1627 ent->data = NULL;
1628 ent->lru.next->prev = ent->lru.prev;
1629 ent->lru.prev->next = ent->lru.next;
1630 delta_base_cached -= ent->size;
1631 }
1632 }
1633
1634 void clear_delta_base_cache(void)
1635 {
1636 unsigned long p;
1637 for (p = 0; p < MAX_DELTA_CACHE; p++)
1638 release_delta_base_cache(&delta_base_cache[p]);
1639 }
1640
1641 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1642 void *base, unsigned long base_size, enum object_type type)
1643 {
1644 unsigned long hash = pack_entry_hash(p, base_offset);
1645 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1646 struct delta_base_cache_lru_list *lru;
1647
1648 release_delta_base_cache(ent);
1649 delta_base_cached += base_size;
1650
1651 for (lru = delta_base_cache_lru.next;
1652 delta_base_cached > delta_base_cache_limit
1653 && lru != &delta_base_cache_lru;
1654 lru = lru->next) {
1655 struct delta_base_cache_entry *f = (void *)lru;
1656 if (f->type == OBJ_BLOB)
1657 release_delta_base_cache(f);
1658 }
1659 for (lru = delta_base_cache_lru.next;
1660 delta_base_cached > delta_base_cache_limit
1661 && lru != &delta_base_cache_lru;
1662 lru = lru->next) {
1663 struct delta_base_cache_entry *f = (void *)lru;
1664 release_delta_base_cache(f);
1665 }
1666
1667 ent->p = p;
1668 ent->base_offset = base_offset;
1669 ent->type = type;
1670 ent->data = base;
1671 ent->size = base_size;
1672 ent->lru.next = &delta_base_cache_lru;
1673 ent->lru.prev = delta_base_cache_lru.prev;
1674 delta_base_cache_lru.prev->next = &ent->lru;
1675 delta_base_cache_lru.prev = &ent->lru;
1676 }
1677
1678 static void *unpack_delta_entry(struct packed_git *p,
1679 struct pack_window **w_curs,
1680 off_t curpos,
1681 unsigned long delta_size,
1682 off_t obj_offset,
1683 enum object_type *type,
1684 unsigned long *sizep)
1685 {
1686 void *delta_data, *result, *base;
1687 unsigned long base_size;
1688 off_t base_offset;
1689
1690 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1691 if (!base_offset) {
1692 error("failed to validate delta base reference "
1693 "at offset %"PRIuMAX" from %s",
1694 (uintmax_t)curpos, p->pack_name);
1695 return NULL;
1696 }
1697 unuse_pack(w_curs);
1698 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1699 if (!base) {
1700 /*
1701 * We're probably in deep shit, but let's try to fetch
1702 * the required base anyway from another pack or loose.
1703 * This is costly but should happen only in the presence
1704 * of a corrupted pack, and is better than failing outright.
1705 */
1706 struct revindex_entry *revidx = find_pack_revindex(p, base_offset);
1707 const unsigned char *base_sha1 =
1708 nth_packed_object_sha1(p, revidx->nr);
1709 error("failed to read delta base object %s"
1710 " at offset %"PRIuMAX" from %s",
1711 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1712 p->pack_name);
1713 mark_bad_packed_object(p, base_sha1);
1714 base = read_object(base_sha1, type, &base_size);
1715 if (!base)
1716 return NULL;
1717 }
1718
1719 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1720 if (!delta_data) {
1721 error("failed to unpack compressed delta "
1722 "at offset %"PRIuMAX" from %s",
1723 (uintmax_t)curpos, p->pack_name);
1724 free(base);
1725 return NULL;
1726 }
1727 result = patch_delta(base, base_size,
1728 delta_data, delta_size,
1729 sizep);
1730 if (!result)
1731 die("failed to apply delta");
1732 free(delta_data);
1733 add_delta_base_cache(p, base_offset, base, base_size, *type);
1734 return result;
1735 }
1736
1737 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1738 enum object_type *type, unsigned long *sizep)
1739 {
1740 struct pack_window *w_curs = NULL;
1741 off_t curpos = obj_offset;
1742 void *data;
1743
1744 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1745 switch (*type) {
1746 case OBJ_OFS_DELTA:
1747 case OBJ_REF_DELTA:
1748 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1749 obj_offset, type, sizep);
1750 break;
1751 case OBJ_COMMIT:
1752 case OBJ_TREE:
1753 case OBJ_BLOB:
1754 case OBJ_TAG:
1755 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1756 break;
1757 default:
1758 data = NULL;
1759 error("unknown object type %i at offset %"PRIuMAX" in %s",
1760 *type, (uintmax_t)obj_offset, p->pack_name);
1761 }
1762 unuse_pack(&w_curs);
1763 return data;
1764 }
1765
1766 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1767 uint32_t n)
1768 {
1769 const unsigned char *index = p->index_data;
1770 if (!index) {
1771 if (open_pack_index(p))
1772 return NULL;
1773 index = p->index_data;
1774 }
1775 if (n >= p->num_objects)
1776 return NULL;
1777 index += 4 * 256;
1778 if (p->index_version == 1) {
1779 return index + 24 * n + 4;
1780 } else {
1781 index += 8;
1782 return index + 20 * n;
1783 }
1784 }
1785
1786 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1787 {
1788 const unsigned char *index = p->index_data;
1789 index += 4 * 256;
1790 if (p->index_version == 1) {
1791 return ntohl(*((uint32_t *)(index + 24 * n)));
1792 } else {
1793 uint32_t off;
1794 index += 8 + p->num_objects * (20 + 4);
1795 off = ntohl(*((uint32_t *)(index + 4 * n)));
1796 if (!(off & 0x80000000))
1797 return off;
1798 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1799 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1800 ntohl(*((uint32_t *)(index + 4)));
1801 }
1802 }
1803
1804 off_t find_pack_entry_one(const unsigned char *sha1,
1805 struct packed_git *p)
1806 {
1807 const uint32_t *level1_ofs = p->index_data;
1808 const unsigned char *index = p->index_data;
1809 unsigned hi, lo, stride;
1810 static int use_lookup = -1;
1811 static int debug_lookup = -1;
1812
1813 if (debug_lookup < 0)
1814 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1815
1816 if (!index) {
1817 if (open_pack_index(p))
1818 return 0;
1819 level1_ofs = p->index_data;
1820 index = p->index_data;
1821 }
1822 if (p->index_version > 1) {
1823 level1_ofs += 2;
1824 index += 8;
1825 }
1826 index += 4 * 256;
1827 hi = ntohl(level1_ofs[*sha1]);
1828 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1829 if (p->index_version > 1) {
1830 stride = 20;
1831 } else {
1832 stride = 24;
1833 index += 4;
1834 }
1835
1836 if (debug_lookup)
1837 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1838 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1839
1840 if (use_lookup < 0)
1841 use_lookup = !!getenv("GIT_USE_LOOKUP");
1842 if (use_lookup) {
1843 int pos = sha1_entry_pos(index, stride, 0,
1844 lo, hi, p->num_objects, sha1);
1845 if (pos < 0)
1846 return 0;
1847 return nth_packed_object_offset(p, pos);
1848 }
1849
1850 do {
1851 unsigned mi = (lo + hi) / 2;
1852 int cmp = hashcmp(index + mi * stride, sha1);
1853
1854 if (debug_lookup)
1855 printf("lo %u hi %u rg %u mi %u\n",
1856 lo, hi, hi - lo, mi);
1857 if (!cmp)
1858 return nth_packed_object_offset(p, mi);
1859 if (cmp > 0)
1860 hi = mi;
1861 else
1862 lo = mi+1;
1863 } while (lo < hi);
1864 return 0;
1865 }
1866
1867 int matches_pack_name(struct packed_git *p, const char *name)
1868 {
1869 const char *last_c, *c;
1870
1871 if (!strcmp(p->pack_name, name))
1872 return 1;
1873
1874 for (c = p->pack_name, last_c = c; *c;)
1875 if (*c == '/')
1876 last_c = ++c;
1877 else
1878 ++c;
1879 if (!strcmp(last_c, name))
1880 return 1;
1881
1882 return 0;
1883 }
1884
1885 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1886 {
1887 static struct packed_git *last_found = (void *)1;
1888 struct packed_git *p;
1889 off_t offset;
1890
1891 prepare_packed_git();
1892 if (!packed_git)
1893 return 0;
1894 p = (last_found == (void *)1) ? packed_git : last_found;
1895
1896 do {
1897 if (ignore_packed) {
1898 const char **ig;
1899 for (ig = ignore_packed; *ig; ig++)
1900 if (matches_pack_name(p, *ig))
1901 break;
1902 if (*ig)
1903 goto next;
1904 }
1905
1906 if (p->num_bad_objects) {
1907 unsigned i;
1908 for (i = 0; i < p->num_bad_objects; i++)
1909 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1910 goto next;
1911 }
1912
1913 offset = find_pack_entry_one(sha1, p);
1914 if (offset) {
1915 /*
1916 * We are about to tell the caller where they can
1917 * locate the requested object. We better make
1918 * sure the packfile is still here and can be
1919 * accessed before supplying that answer, as
1920 * it may have been deleted since the index
1921 * was loaded!
1922 */
1923 if (p->pack_fd == -1 && open_packed_git(p)) {
1924 error("packfile %s cannot be accessed", p->pack_name);
1925 goto next;
1926 }
1927 e->offset = offset;
1928 e->p = p;
1929 hashcpy(e->sha1, sha1);
1930 last_found = p;
1931 return 1;
1932 }
1933
1934 next:
1935 if (p == last_found)
1936 p = packed_git;
1937 else
1938 p = p->next;
1939 if (p == last_found)
1940 p = p->next;
1941 } while (p);
1942 return 0;
1943 }
1944
1945 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1946 struct packed_git *packs)
1947 {
1948 struct packed_git *p;
1949
1950 for (p = packs; p; p = p->next) {
1951 if (find_pack_entry_one(sha1, p))
1952 return p;
1953 }
1954 return NULL;
1955
1956 }
1957
1958 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1959 {
1960 int status;
1961 unsigned long mapsize, size;
1962 void *map;
1963 z_stream stream;
1964 char hdr[32];
1965
1966 map = map_sha1_file(sha1, &mapsize);
1967 if (!map)
1968 return error("unable to find %s", sha1_to_hex(sha1));
1969 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1970 status = error("unable to unpack %s header",
1971 sha1_to_hex(sha1));
1972 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1973 status = error("unable to parse %s header", sha1_to_hex(sha1));
1974 else if (sizep)
1975 *sizep = size;
1976 inflateEnd(&stream);
1977 munmap(map, mapsize);
1978 return status;
1979 }
1980
1981 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1982 {
1983 struct pack_entry e;
1984 int status;
1985
1986 if (!find_pack_entry(sha1, &e, NULL)) {
1987 /* Most likely it's a loose object. */
1988 status = sha1_loose_object_info(sha1, sizep);
1989 if (status >= 0)
1990 return status;
1991
1992 /* Not a loose object; someone else may have just packed it. */
1993 reprepare_packed_git();
1994 if (!find_pack_entry(sha1, &e, NULL))
1995 return status;
1996 }
1997 return packed_object_info(e.p, e.offset, sizep);
1998 }
1999
2000 static void *read_packed_sha1(const unsigned char *sha1,
2001 enum object_type *type, unsigned long *size)
2002 {
2003 struct pack_entry e;
2004 void *data;
2005
2006 if (!find_pack_entry(sha1, &e, NULL))
2007 return NULL;
2008 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2009 if (!data) {
2010 /*
2011 * We're probably in deep shit, but let's try to fetch
2012 * the required object anyway from another pack or loose.
2013 * This should happen only in the presence of a corrupted
2014 * pack, and is better than failing outright.
2015 */
2016 error("failed to read object %s at offset %"PRIuMAX" from %s",
2017 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2018 mark_bad_packed_object(e.p, sha1);
2019 data = read_object(sha1, type, size);
2020 }
2021 return data;
2022 }
2023
2024 /*
2025 * This is meant to hold a *small* number of objects that you would
2026 * want read_sha1_file() to be able to return, but yet you do not want
2027 * to write them into the object store (e.g. a browse-only
2028 * application).
2029 */
2030 static struct cached_object {
2031 unsigned char sha1[20];
2032 enum object_type type;
2033 void *buf;
2034 unsigned long size;
2035 } *cached_objects;
2036 static int cached_object_nr, cached_object_alloc;
2037
2038 static struct cached_object empty_tree = {
2039 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
2040 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
2041 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
2042 OBJ_TREE,
2043 "",
2044 0
2045 };
2046
2047 static struct cached_object *find_cached_object(const unsigned char *sha1)
2048 {
2049 int i;
2050 struct cached_object *co = cached_objects;
2051
2052 for (i = 0; i < cached_object_nr; i++, co++) {
2053 if (!hashcmp(co->sha1, sha1))
2054 return co;
2055 }
2056 if (!hashcmp(sha1, empty_tree.sha1))
2057 return &empty_tree;
2058 return NULL;
2059 }
2060
2061 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2062 unsigned char *sha1)
2063 {
2064 struct cached_object *co;
2065
2066 hash_sha1_file(buf, len, typename(type), sha1);
2067 if (has_sha1_file(sha1) || find_cached_object(sha1))
2068 return 0;
2069 if (cached_object_alloc <= cached_object_nr) {
2070 cached_object_alloc = alloc_nr(cached_object_alloc);
2071 cached_objects = xrealloc(cached_objects,
2072 sizeof(*cached_objects) *
2073 cached_object_alloc);
2074 }
2075 co = &cached_objects[cached_object_nr++];
2076 co->size = len;
2077 co->type = type;
2078 co->buf = xmalloc(len);
2079 memcpy(co->buf, buf, len);
2080 hashcpy(co->sha1, sha1);
2081 return 0;
2082 }
2083
2084 void *read_object(const unsigned char *sha1, enum object_type *type,
2085 unsigned long *size)
2086 {
2087 unsigned long mapsize;
2088 void *map, *buf;
2089 struct cached_object *co;
2090
2091 co = find_cached_object(sha1);
2092 if (co) {
2093 *type = co->type;
2094 *size = co->size;
2095 return xmemdupz(co->buf, co->size);
2096 }
2097
2098 buf = read_packed_sha1(sha1, type, size);
2099 if (buf)
2100 return buf;
2101 map = map_sha1_file(sha1, &mapsize);
2102 if (map) {
2103 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2104 munmap(map, mapsize);
2105 return buf;
2106 }
2107 reprepare_packed_git();
2108 return read_packed_sha1(sha1, type, size);
2109 }
2110
2111 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
2112 unsigned long *size)
2113 {
2114 void *data = read_object(sha1, type, size);
2115 /* legacy behavior is to die on corrupted objects */
2116 if (!data && (has_loose_object(sha1) || has_packed_and_bad(sha1)))
2117 die("object %s is corrupted", sha1_to_hex(sha1));
2118 return data;
2119 }
2120
2121 void *read_object_with_reference(const unsigned char *sha1,
2122 const char *required_type_name,
2123 unsigned long *size,
2124 unsigned char *actual_sha1_return)
2125 {
2126 enum object_type type, required_type;
2127 void *buffer;
2128 unsigned long isize;
2129 unsigned char actual_sha1[20];
2130
2131 required_type = type_from_string(required_type_name);
2132 hashcpy(actual_sha1, sha1);
2133 while (1) {
2134 int ref_length = -1;
2135 const char *ref_type = NULL;
2136
2137 buffer = read_sha1_file(actual_sha1, &type, &isize);
2138 if (!buffer)
2139 return NULL;
2140 if (type == required_type) {
2141 *size = isize;
2142 if (actual_sha1_return)
2143 hashcpy(actual_sha1_return, actual_sha1);
2144 return buffer;
2145 }
2146 /* Handle references */
2147 else if (type == OBJ_COMMIT)
2148 ref_type = "tree ";
2149 else if (type == OBJ_TAG)
2150 ref_type = "object ";
2151 else {
2152 free(buffer);
2153 return NULL;
2154 }
2155 ref_length = strlen(ref_type);
2156
2157 if (ref_length + 40 > isize ||
2158 memcmp(buffer, ref_type, ref_length) ||
2159 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2160 free(buffer);
2161 return NULL;
2162 }
2163 free(buffer);
2164 /* Now we have the ID of the referred-to object in
2165 * actual_sha1. Check again. */
2166 }
2167 }
2168
2169 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2170 const char *type, unsigned char *sha1,
2171 char *hdr, int *hdrlen)
2172 {
2173 SHA_CTX c;
2174
2175 /* Generate the header */
2176 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2177
2178 /* Sha1.. */
2179 SHA1_Init(&c);
2180 SHA1_Update(&c, hdr, *hdrlen);
2181 SHA1_Update(&c, buf, len);
2182 SHA1_Final(sha1, &c);
2183 }
2184
2185 /*
2186 * Move the just written object into its final resting place
2187 */
2188 int move_temp_to_file(const char *tmpfile, const char *filename)
2189 {
2190 int ret = 0;
2191 if (link(tmpfile, filename))
2192 ret = errno;
2193
2194 /*
2195 * Coda hack - coda doesn't like cross-directory links,
2196 * so we fall back to a rename, which will mean that it
2197 * won't be able to check collisions, but that's not a
2198 * big deal.
2199 *
2200 * The same holds for FAT formatted media.
2201 *
2202 * When this succeeds, we just return 0. We have nothing
2203 * left to unlink.
2204 */
2205 if (ret && ret != EEXIST) {
2206 if (!rename(tmpfile, filename))
2207 return 0;
2208 ret = errno;
2209 }
2210 unlink(tmpfile);
2211 if (ret) {
2212 if (ret != EEXIST) {
2213 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2214 }
2215 /* FIXME!!! Collision check here ? */
2216 }
2217
2218 return 0;
2219 }
2220
2221 static int write_buffer(int fd, const void *buf, size_t len)
2222 {
2223 if (write_in_full(fd, buf, len) < 0)
2224 return error("file write error (%s)", strerror(errno));
2225 return 0;
2226 }
2227
2228 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2229 unsigned char *sha1)
2230 {
2231 char hdr[32];
2232 int hdrlen;
2233 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2234 return 0;
2235 }
2236
2237 /* Finalize a file on disk, and close it. */
2238 static void close_sha1_file(int fd)
2239 {
2240 if (fsync_object_files)
2241 fsync_or_die(fd, "sha1 file");
2242 fchmod(fd, 0444);
2243 if (close(fd) != 0)
2244 die("unable to write sha1 file");
2245 }
2246
2247 /* Size of directory component, including the ending '/' */
2248 static inline int directory_size(const char *filename)
2249 {
2250 const char *s = strrchr(filename, '/');
2251 if (!s)
2252 return 0;
2253 return s - filename + 1;
2254 }
2255
2256 /*
2257 * This creates a temporary file in the same directory as the final
2258 * 'filename'
2259 *
2260 * We want to avoid cross-directory filename renames, because those
2261 * can have problems on various filesystems (FAT, NFS, Coda).
2262 */
2263 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2264 {
2265 int fd, dirlen = directory_size(filename);
2266
2267 if (dirlen + 20 > bufsiz) {
2268 errno = ENAMETOOLONG;
2269 return -1;
2270 }
2271 memcpy(buffer, filename, dirlen);
2272 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2273 fd = mkstemp(buffer);
2274 if (fd < 0 && dirlen && errno == ENOENT) {
2275 /* Make sure the directory exists */
2276 memcpy(buffer, filename, dirlen);
2277 buffer[dirlen-1] = 0;
2278 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2279 return -1;
2280
2281 /* Try again */
2282 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2283 fd = mkstemp(buffer);
2284 }
2285 return fd;
2286 }
2287
2288 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2289 void *buf, unsigned long len, time_t mtime)
2290 {
2291 int fd, ret;
2292 size_t size;
2293 unsigned char *compressed;
2294 z_stream stream;
2295 char *filename;
2296 static char tmpfile[PATH_MAX];
2297
2298 filename = sha1_file_name(sha1);
2299 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2300 if (fd < 0) {
2301 if (errno == EACCES)
2302 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2303 else
2304 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2305 }
2306
2307 /* Set it up */
2308 memset(&stream, 0, sizeof(stream));
2309 deflateInit(&stream, zlib_compression_level);
2310 size = 8 + deflateBound(&stream, len+hdrlen);
2311 compressed = xmalloc(size);
2312
2313 /* Compress it */
2314 stream.next_out = compressed;
2315 stream.avail_out = size;
2316
2317 /* First header.. */
2318 stream.next_in = (unsigned char *)hdr;
2319 stream.avail_in = hdrlen;
2320 while (deflate(&stream, 0) == Z_OK)
2321 /* nothing */;
2322
2323 /* Then the data itself.. */
2324 stream.next_in = buf;
2325 stream.avail_in = len;
2326 ret = deflate(&stream, Z_FINISH);
2327 if (ret != Z_STREAM_END)
2328 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2329
2330 ret = deflateEnd(&stream);
2331 if (ret != Z_OK)
2332 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2333
2334 size = stream.total_out;
2335
2336 if (write_buffer(fd, compressed, size) < 0)
2337 die("unable to write sha1 file");
2338 close_sha1_file(fd);
2339 free(compressed);
2340
2341 if (mtime) {
2342 struct utimbuf utb;
2343 utb.actime = mtime;
2344 utb.modtime = mtime;
2345 if (utime(tmpfile, &utb) < 0)
2346 warning("failed utime() on %s: %s",
2347 tmpfile, strerror(errno));
2348 }
2349
2350 return move_temp_to_file(tmpfile, filename);
2351 }
2352
2353 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2354 {
2355 unsigned char sha1[20];
2356 char hdr[32];
2357 int hdrlen;
2358
2359 /* Normally if we have it in the pack then we do not bother writing
2360 * it out into .git/objects/??/?{38} file.
2361 */
2362 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2363 if (returnsha1)
2364 hashcpy(returnsha1, sha1);
2365 if (has_sha1_file(sha1))
2366 return 0;
2367 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2368 }
2369
2370 int force_object_loose(const unsigned char *sha1, time_t mtime)
2371 {
2372 void *buf;
2373 unsigned long len;
2374 enum object_type type;
2375 char hdr[32];
2376 int hdrlen;
2377 int ret;
2378
2379 if (has_loose_object(sha1))
2380 return 0;
2381 buf = read_packed_sha1(sha1, &type, &len);
2382 if (!buf)
2383 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2384 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2385 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2386 free(buf);
2387
2388 return ret;
2389 }
2390
2391 int has_pack_index(const unsigned char *sha1)
2392 {
2393 struct stat st;
2394 if (stat(sha1_pack_index_name(sha1), &st))
2395 return 0;
2396 return 1;
2397 }
2398
2399 int has_pack_file(const unsigned char *sha1)
2400 {
2401 struct stat st;
2402 if (stat(sha1_pack_name(sha1), &st))
2403 return 0;
2404 return 1;
2405 }
2406
2407 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2408 {
2409 struct pack_entry e;
2410 return find_pack_entry(sha1, &e, ignore_packed);
2411 }
2412
2413 int has_sha1_file(const unsigned char *sha1)
2414 {
2415 struct pack_entry e;
2416
2417 if (find_pack_entry(sha1, &e, NULL))
2418 return 1;
2419 return has_loose_object(sha1);
2420 }
2421
2422 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2423 {
2424 struct strbuf buf;
2425 int ret;
2426
2427 strbuf_init(&buf, 0);
2428 if (strbuf_read(&buf, fd, 4096) < 0) {
2429 strbuf_release(&buf);
2430 return -1;
2431 }
2432
2433 if (!type)
2434 type = blob_type;
2435 if (write_object)
2436 ret = write_sha1_file(buf.buf, buf.len, type, sha1);
2437 else
2438 ret = hash_sha1_file(buf.buf, buf.len, type, sha1);
2439 strbuf_release(&buf);
2440
2441 return ret;
2442 }
2443
2444 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2445 enum object_type type, const char *path)
2446 {
2447 size_t size = xsize_t(st->st_size);
2448 void *buf = NULL;
2449 int ret, re_allocated = 0;
2450
2451 if (size)
2452 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2453 close(fd);
2454
2455 if (!type)
2456 type = OBJ_BLOB;
2457
2458 /*
2459 * Convert blobs to git internal format
2460 */
2461 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2462 struct strbuf nbuf;
2463 strbuf_init(&nbuf, 0);
2464 if (convert_to_git(path, buf, size, &nbuf,
2465 write_object ? safe_crlf : 0)) {
2466 munmap(buf, size);
2467 buf = strbuf_detach(&nbuf, &size);
2468 re_allocated = 1;
2469 }
2470 }
2471
2472 if (write_object)
2473 ret = write_sha1_file(buf, size, typename(type), sha1);
2474 else
2475 ret = hash_sha1_file(buf, size, typename(type), sha1);
2476 if (re_allocated) {
2477 free(buf);
2478 return ret;
2479 }
2480 if (size)
2481 munmap(buf, size);
2482 return ret;
2483 }
2484
2485 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2486 {
2487 int fd;
2488 char *target;
2489 size_t len;
2490
2491 switch (st->st_mode & S_IFMT) {
2492 case S_IFREG:
2493 fd = open(path, O_RDONLY);
2494 if (fd < 0)
2495 return error("open(\"%s\"): %s", path,
2496 strerror(errno));
2497 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2498 return error("%s: failed to insert into database",
2499 path);
2500 break;
2501 case S_IFLNK:
2502 len = xsize_t(st->st_size);
2503 target = xmalloc(len + 1);
2504 if (readlink(path, target, len + 1) != st->st_size) {
2505 char *errstr = strerror(errno);
2506 free(target);
2507 return error("readlink(\"%s\"): %s", path,
2508 errstr);
2509 }
2510 if (!write_object)
2511 hash_sha1_file(target, len, blob_type, sha1);
2512 else if (write_sha1_file(target, len, blob_type, sha1))
2513 return error("%s: failed to insert into database",
2514 path);
2515 free(target);
2516 break;
2517 case S_IFDIR:
2518 return resolve_gitlink_ref(path, "HEAD", sha1);
2519 default:
2520 return error("%s: unsupported file type", path);
2521 }
2522 return 0;
2523 }
2524
2525 int read_pack_header(int fd, struct pack_header *header)
2526 {
2527 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2528 /* "eof before pack header was fully read" */
2529 return PH_ERROR_EOF;
2530
2531 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2532 /* "protocol error (pack signature mismatch detected)" */
2533 return PH_ERROR_PACK_SIGNATURE;
2534 if (!pack_version_ok(header->hdr_version))
2535 /* "protocol error (pack version unsupported)" */
2536 return PH_ERROR_PROTOCOL;
2537 return 0;
2538 }