pack-objects: don't loosen objects available in alternate or kept packs
[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 close_pack_windows(p);
680 if (p->pack_fd != -1)
681 close(p->pack_fd);
682 if (p->index_data)
683 munmap((void *)p->index_data, p->index_size);
684 free(p->bad_object_sha1);
685 *pp = p->next;
686 free(p);
687 return;
688 }
689 pp = &p->next;
690 }
691 }
692
693 /*
694 * Do not call this directly as this leaks p->pack_fd on error return;
695 * call open_packed_git() instead.
696 */
697 static int open_packed_git_1(struct packed_git *p)
698 {
699 struct stat st;
700 struct pack_header hdr;
701 unsigned char sha1[20];
702 unsigned char *idx_sha1;
703 long fd_flag;
704
705 if (!p->index_data && open_pack_index(p))
706 return error("packfile %s index unavailable", p->pack_name);
707
708 p->pack_fd = open(p->pack_name, O_RDONLY);
709 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
710 return -1;
711
712 /* If we created the struct before we had the pack we lack size. */
713 if (!p->pack_size) {
714 if (!S_ISREG(st.st_mode))
715 return error("packfile %s not a regular file", p->pack_name);
716 p->pack_size = st.st_size;
717 } else if (p->pack_size != st.st_size)
718 return error("packfile %s size changed", p->pack_name);
719
720 /* We leave these file descriptors open with sliding mmap;
721 * there is no point keeping them open across exec(), though.
722 */
723 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
724 if (fd_flag < 0)
725 return error("cannot determine file descriptor flags");
726 fd_flag |= FD_CLOEXEC;
727 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
728 return error("cannot set FD_CLOEXEC");
729
730 /* Verify we recognize this pack file format. */
731 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
732 return error("file %s is far too short to be a packfile", p->pack_name);
733 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
734 return error("file %s is not a GIT packfile", p->pack_name);
735 if (!pack_version_ok(hdr.hdr_version))
736 return error("packfile %s is version %"PRIu32" and not"
737 " supported (try upgrading GIT to a newer version)",
738 p->pack_name, ntohl(hdr.hdr_version));
739
740 /* Verify the pack matches its index. */
741 if (p->num_objects != ntohl(hdr.hdr_entries))
742 return error("packfile %s claims to have %"PRIu32" objects"
743 " while index indicates %"PRIu32" objects",
744 p->pack_name, ntohl(hdr.hdr_entries),
745 p->num_objects);
746 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
747 return error("end of packfile %s is unavailable", p->pack_name);
748 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
749 return error("packfile %s signature is unavailable", p->pack_name);
750 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
751 if (hashcmp(sha1, idx_sha1))
752 return error("packfile %s does not match index", p->pack_name);
753 return 0;
754 }
755
756 static int open_packed_git(struct packed_git *p)
757 {
758 if (!open_packed_git_1(p))
759 return 0;
760 if (p->pack_fd != -1) {
761 close(p->pack_fd);
762 p->pack_fd = -1;
763 }
764 return -1;
765 }
766
767 static int in_window(struct pack_window *win, off_t offset)
768 {
769 /* We must promise at least 20 bytes (one hash) after the
770 * offset is available from this window, otherwise the offset
771 * is not actually in this window and a different window (which
772 * has that one hash excess) must be used. This is to support
773 * the object header and delta base parsing routines below.
774 */
775 off_t win_off = win->offset;
776 return win_off <= offset
777 && (offset + 20) <= (win_off + win->len);
778 }
779
780 unsigned char* use_pack(struct packed_git *p,
781 struct pack_window **w_cursor,
782 off_t offset,
783 unsigned int *left)
784 {
785 struct pack_window *win = *w_cursor;
786
787 if (p->pack_fd == -1 && open_packed_git(p))
788 die("packfile %s cannot be accessed", p->pack_name);
789
790 /* Since packfiles end in a hash of their content and its
791 * pointless to ask for an offset into the middle of that
792 * hash, and the in_window function above wouldn't match
793 * don't allow an offset too close to the end of the file.
794 */
795 if (offset > (p->pack_size - 20))
796 die("offset beyond end of packfile (truncated pack?)");
797
798 if (!win || !in_window(win, offset)) {
799 if (win)
800 win->inuse_cnt--;
801 for (win = p->windows; win; win = win->next) {
802 if (in_window(win, offset))
803 break;
804 }
805 if (!win) {
806 size_t window_align = packed_git_window_size / 2;
807 off_t len;
808 win = xcalloc(1, sizeof(*win));
809 win->offset = (offset / window_align) * window_align;
810 len = p->pack_size - win->offset;
811 if (len > packed_git_window_size)
812 len = packed_git_window_size;
813 win->len = (size_t)len;
814 pack_mapped += win->len;
815 while (packed_git_limit < pack_mapped
816 && unuse_one_window(p, p->pack_fd))
817 ; /* nothing */
818 win->base = xmmap(NULL, win->len,
819 PROT_READ, MAP_PRIVATE,
820 p->pack_fd, win->offset);
821 if (win->base == MAP_FAILED)
822 die("packfile %s cannot be mapped: %s",
823 p->pack_name,
824 strerror(errno));
825 pack_mmap_calls++;
826 pack_open_windows++;
827 if (pack_mapped > peak_pack_mapped)
828 peak_pack_mapped = pack_mapped;
829 if (pack_open_windows > peak_pack_open_windows)
830 peak_pack_open_windows = pack_open_windows;
831 win->next = p->windows;
832 p->windows = win;
833 }
834 }
835 if (win != *w_cursor) {
836 win->last_used = pack_used_ctr++;
837 win->inuse_cnt++;
838 *w_cursor = win;
839 }
840 offset -= win->offset;
841 if (left)
842 *left = win->len - xsize_t(offset);
843 return win->base + offset;
844 }
845
846 static struct packed_git *alloc_packed_git(int extra)
847 {
848 struct packed_git *p = xmalloc(sizeof(*p) + extra);
849 memset(p, 0, sizeof(*p));
850 p->pack_fd = -1;
851 return p;
852 }
853
854 struct packed_git *add_packed_git(const char *path, int path_len, int local)
855 {
856 struct stat st;
857 struct packed_git *p = alloc_packed_git(path_len + 2);
858
859 /*
860 * Make sure a corresponding .pack file exists and that
861 * the index looks sane.
862 */
863 path_len -= strlen(".idx");
864 if (path_len < 1) {
865 free(p);
866 return NULL;
867 }
868 memcpy(p->pack_name, path, path_len);
869
870 strcpy(p->pack_name + path_len, ".keep");
871 if (!access(p->pack_name, F_OK))
872 p->pack_keep = 1;
873
874 strcpy(p->pack_name + path_len, ".pack");
875 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
876 free(p);
877 return NULL;
878 }
879
880 /* ok, it looks sane as far as we can check without
881 * actually mapping the pack file.
882 */
883 p->pack_size = st.st_size;
884 p->pack_local = local;
885 p->mtime = st.st_mtime;
886 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
887 hashclr(p->sha1);
888 return p;
889 }
890
891 struct packed_git *parse_pack_index(unsigned char *sha1)
892 {
893 const char *idx_path = sha1_pack_index_name(sha1);
894 const char *path = sha1_pack_name(sha1);
895 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
896
897 strcpy(p->pack_name, path);
898 hashcpy(p->sha1, sha1);
899 if (check_packed_git_idx(idx_path, p)) {
900 free(p);
901 return NULL;
902 }
903
904 return p;
905 }
906
907 void install_packed_git(struct packed_git *pack)
908 {
909 pack->next = packed_git;
910 packed_git = pack;
911 }
912
913 static void prepare_packed_git_one(char *objdir, int local)
914 {
915 /* Ensure that this buffer is large enough so that we can
916 append "/pack/" without clobbering the stack even if
917 strlen(objdir) were PATH_MAX. */
918 char path[PATH_MAX + 1 + 4 + 1 + 1];
919 int len;
920 DIR *dir;
921 struct dirent *de;
922
923 sprintf(path, "%s/pack", objdir);
924 len = strlen(path);
925 dir = opendir(path);
926 if (!dir) {
927 if (errno != ENOENT)
928 error("unable to open object pack directory: %s: %s",
929 path, strerror(errno));
930 return;
931 }
932 path[len++] = '/';
933 while ((de = readdir(dir)) != NULL) {
934 int namelen = strlen(de->d_name);
935 struct packed_git *p;
936
937 if (!has_extension(de->d_name, ".idx"))
938 continue;
939
940 if (len + namelen + 1 > sizeof(path))
941 continue;
942
943 /* Don't reopen a pack we already have. */
944 strcpy(path + len, de->d_name);
945 for (p = packed_git; p; p = p->next) {
946 if (!memcmp(path, p->pack_name, len + namelen - 4))
947 break;
948 }
949 if (p)
950 continue;
951 /* See if it really is a valid .idx file with corresponding
952 * .pack file that we can map.
953 */
954 p = add_packed_git(path, len + namelen, local);
955 if (!p)
956 continue;
957 install_packed_git(p);
958 }
959 closedir(dir);
960 }
961
962 static int sort_pack(const void *a_, const void *b_)
963 {
964 struct packed_git *a = *((struct packed_git **)a_);
965 struct packed_git *b = *((struct packed_git **)b_);
966 int st;
967
968 /*
969 * Local packs tend to contain objects specific to our
970 * variant of the project than remote ones. In addition,
971 * remote ones could be on a network mounted filesystem.
972 * Favor local ones for these reasons.
973 */
974 st = a->pack_local - b->pack_local;
975 if (st)
976 return -st;
977
978 /*
979 * Younger packs tend to contain more recent objects,
980 * and more recent objects tend to get accessed more
981 * often.
982 */
983 if (a->mtime < b->mtime)
984 return 1;
985 else if (a->mtime == b->mtime)
986 return 0;
987 return -1;
988 }
989
990 static void rearrange_packed_git(void)
991 {
992 struct packed_git **ary, *p;
993 int i, n;
994
995 for (n = 0, p = packed_git; p; p = p->next)
996 n++;
997 if (n < 2)
998 return;
999
1000 /* prepare an array of packed_git for easier sorting */
1001 ary = xcalloc(n, sizeof(struct packed_git *));
1002 for (n = 0, p = packed_git; p; p = p->next)
1003 ary[n++] = p;
1004
1005 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1006
1007 /* link them back again */
1008 for (i = 0; i < n - 1; i++)
1009 ary[i]->next = ary[i + 1];
1010 ary[n - 1]->next = NULL;
1011 packed_git = ary[0];
1012
1013 free(ary);
1014 }
1015
1016 static int prepare_packed_git_run_once = 0;
1017 void prepare_packed_git(void)
1018 {
1019 struct alternate_object_database *alt;
1020
1021 if (prepare_packed_git_run_once)
1022 return;
1023 prepare_packed_git_one(get_object_directory(), 1);
1024 prepare_alt_odb();
1025 for (alt = alt_odb_list; alt; alt = alt->next) {
1026 alt->name[-1] = 0;
1027 prepare_packed_git_one(alt->base, 0);
1028 alt->name[-1] = '/';
1029 }
1030 rearrange_packed_git();
1031 prepare_packed_git_run_once = 1;
1032 }
1033
1034 void reprepare_packed_git(void)
1035 {
1036 discard_revindex();
1037 prepare_packed_git_run_once = 0;
1038 prepare_packed_git();
1039 }
1040
1041 static void mark_bad_packed_object(struct packed_git *p,
1042 const unsigned char *sha1)
1043 {
1044 unsigned i;
1045 for (i = 0; i < p->num_bad_objects; i++)
1046 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1047 return;
1048 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1049 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1050 p->num_bad_objects++;
1051 }
1052
1053 static int has_packed_and_bad(const unsigned char *sha1)
1054 {
1055 struct packed_git *p;
1056 unsigned i;
1057
1058 for (p = packed_git; p; p = p->next)
1059 for (i = 0; i < p->num_bad_objects; i++)
1060 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1061 return 1;
1062 return 0;
1063 }
1064
1065 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1066 {
1067 unsigned char real_sha1[20];
1068 hash_sha1_file(map, size, type, real_sha1);
1069 return hashcmp(sha1, real_sha1) ? -1 : 0;
1070 }
1071
1072 static int git_open_noatime(const char *name)
1073 {
1074 static int sha1_file_open_flag = O_NOATIME;
1075 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1076
1077 /* Might the failure be due to O_NOATIME? */
1078 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1079 fd = open(name, O_RDONLY);
1080 if (fd >= 0)
1081 sha1_file_open_flag = 0;
1082 }
1083 return fd;
1084 }
1085
1086 static int open_sha1_file(const unsigned char *sha1)
1087 {
1088 int fd;
1089 char *name = sha1_file_name(sha1);
1090 struct alternate_object_database *alt;
1091
1092 fd = git_open_noatime(name);
1093 if (fd >= 0)
1094 return fd;
1095
1096 prepare_alt_odb();
1097 errno = ENOENT;
1098 for (alt = alt_odb_list; alt; alt = alt->next) {
1099 name = alt->name;
1100 fill_sha1_path(name, sha1);
1101 fd = git_open_noatime(alt->base);
1102 if (fd >= 0)
1103 return fd;
1104 }
1105 return -1;
1106 }
1107
1108 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1109 {
1110 void *map;
1111 int fd;
1112
1113 fd = open_sha1_file(sha1);
1114 map = NULL;
1115 if (fd >= 0) {
1116 struct stat st;
1117
1118 if (!fstat(fd, &st)) {
1119 *size = xsize_t(st.st_size);
1120 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1121 }
1122 close(fd);
1123 }
1124 return map;
1125 }
1126
1127 static int legacy_loose_object(unsigned char *map)
1128 {
1129 unsigned int word;
1130
1131 /*
1132 * Is it a zlib-compressed buffer? If so, the first byte
1133 * must be 0x78 (15-bit window size, deflated), and the
1134 * first 16-bit word is evenly divisible by 31
1135 */
1136 word = (map[0] << 8) + map[1];
1137 if (map[0] == 0x78 && !(word % 31))
1138 return 1;
1139 else
1140 return 0;
1141 }
1142
1143 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1144 {
1145 unsigned shift;
1146 unsigned char c;
1147 unsigned long size;
1148 unsigned long used = 0;
1149
1150 c = buf[used++];
1151 *type = (c >> 4) & 7;
1152 size = c & 15;
1153 shift = 4;
1154 while (c & 0x80) {
1155 if (len <= used)
1156 return 0;
1157 if (sizeof(long) * 8 <= shift)
1158 return 0;
1159 c = buf[used++];
1160 size += (c & 0x7f) << shift;
1161 shift += 7;
1162 }
1163 *sizep = size;
1164 return used;
1165 }
1166
1167 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1168 {
1169 unsigned long size, used;
1170 static const char valid_loose_object_type[8] = {
1171 0, /* OBJ_EXT */
1172 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1173 0, /* "delta" and others are invalid in a loose object */
1174 };
1175 enum object_type type;
1176
1177 /* Get the data stream */
1178 memset(stream, 0, sizeof(*stream));
1179 stream->next_in = map;
1180 stream->avail_in = mapsize;
1181 stream->next_out = buffer;
1182 stream->avail_out = bufsiz;
1183
1184 if (legacy_loose_object(map)) {
1185 inflateInit(stream);
1186 return inflate(stream, 0);
1187 }
1188
1189
1190 /*
1191 * There used to be a second loose object header format which
1192 * was meant to mimic the in-pack format, allowing for direct
1193 * copy of the object data. This format turned up not to be
1194 * really worth it and we don't write it any longer. But we
1195 * can still read it.
1196 */
1197 used = unpack_object_header_gently(map, mapsize, &type, &size);
1198 if (!used || !valid_loose_object_type[type])
1199 return -1;
1200 map += used;
1201 mapsize -= used;
1202
1203 /* Set up the stream for the rest.. */
1204 stream->next_in = map;
1205 stream->avail_in = mapsize;
1206 inflateInit(stream);
1207
1208 /* And generate the fake traditional header */
1209 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1210 typename(type), size);
1211 return 0;
1212 }
1213
1214 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1215 {
1216 int bytes = strlen(buffer) + 1;
1217 unsigned char *buf = xmalloc(1+size);
1218 unsigned long n;
1219 int status = Z_OK;
1220
1221 n = stream->total_out - bytes;
1222 if (n > size)
1223 n = size;
1224 memcpy(buf, (char *) buffer + bytes, n);
1225 bytes = n;
1226 if (bytes <= size) {
1227 /*
1228 * The above condition must be (bytes <= size), not
1229 * (bytes < size). In other words, even though we
1230 * expect no more output and set avail_out to zer0,
1231 * the input zlib stream may have bytes that express
1232 * "this concludes the stream", and we *do* want to
1233 * eat that input.
1234 *
1235 * Otherwise we would not be able to test that we
1236 * consumed all the input to reach the expected size;
1237 * we also want to check that zlib tells us that all
1238 * went well with status == Z_STREAM_END at the end.
1239 */
1240 stream->next_out = buf + bytes;
1241 stream->avail_out = size - bytes;
1242 while (status == Z_OK)
1243 status = inflate(stream, Z_FINISH);
1244 }
1245 buf[size] = 0;
1246 if (status == Z_STREAM_END && !stream->avail_in) {
1247 inflateEnd(stream);
1248 return buf;
1249 }
1250
1251 if (status < 0)
1252 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1253 else if (stream->avail_in)
1254 error("garbage at end of loose object '%s'",
1255 sha1_to_hex(sha1));
1256 free(buf);
1257 return NULL;
1258 }
1259
1260 /*
1261 * We used to just use "sscanf()", but that's actually way
1262 * too permissive for what we want to check. So do an anal
1263 * object header parse by hand.
1264 */
1265 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1266 {
1267 char type[10];
1268 int i;
1269 unsigned long size;
1270
1271 /*
1272 * The type can be at most ten bytes (including the
1273 * terminating '\0' that we add), and is followed by
1274 * a space.
1275 */
1276 i = 0;
1277 for (;;) {
1278 char c = *hdr++;
1279 if (c == ' ')
1280 break;
1281 type[i++] = c;
1282 if (i >= sizeof(type))
1283 return -1;
1284 }
1285 type[i] = 0;
1286
1287 /*
1288 * The length must follow immediately, and be in canonical
1289 * decimal format (ie "010" is not valid).
1290 */
1291 size = *hdr++ - '0';
1292 if (size > 9)
1293 return -1;
1294 if (size) {
1295 for (;;) {
1296 unsigned long c = *hdr - '0';
1297 if (c > 9)
1298 break;
1299 hdr++;
1300 size = size * 10 + c;
1301 }
1302 }
1303 *sizep = size;
1304
1305 /*
1306 * The length must be followed by a zero byte
1307 */
1308 return *hdr ? -1 : type_from_string(type);
1309 }
1310
1311 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1312 {
1313 int ret;
1314 z_stream stream;
1315 char hdr[8192];
1316
1317 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1318 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1319 return NULL;
1320
1321 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1322 }
1323
1324 unsigned long get_size_from_delta(struct packed_git *p,
1325 struct pack_window **w_curs,
1326 off_t curpos)
1327 {
1328 const unsigned char *data;
1329 unsigned char delta_head[20], *in;
1330 z_stream stream;
1331 int st;
1332
1333 memset(&stream, 0, sizeof(stream));
1334 stream.next_out = delta_head;
1335 stream.avail_out = sizeof(delta_head);
1336
1337 inflateInit(&stream);
1338 do {
1339 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1340 stream.next_in = in;
1341 st = inflate(&stream, Z_FINISH);
1342 curpos += stream.next_in - in;
1343 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1344 stream.total_out < sizeof(delta_head));
1345 inflateEnd(&stream);
1346 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1347 die("delta data unpack-initial failed");
1348
1349 /* Examine the initial part of the delta to figure out
1350 * the result size.
1351 */
1352 data = delta_head;
1353
1354 /* ignore base size */
1355 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1356
1357 /* Read the result size */
1358 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1359 }
1360
1361 static off_t get_delta_base(struct packed_git *p,
1362 struct pack_window **w_curs,
1363 off_t *curpos,
1364 enum object_type type,
1365 off_t delta_obj_offset)
1366 {
1367 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1368 off_t base_offset;
1369
1370 /* use_pack() assured us we have [base_info, base_info + 20)
1371 * as a range that we can look at without walking off the
1372 * end of the mapped window. Its actually the hash size
1373 * that is assured. An OFS_DELTA longer than the hash size
1374 * is stupid, as then a REF_DELTA would be smaller to store.
1375 */
1376 if (type == OBJ_OFS_DELTA) {
1377 unsigned used = 0;
1378 unsigned char c = base_info[used++];
1379 base_offset = c & 127;
1380 while (c & 128) {
1381 base_offset += 1;
1382 if (!base_offset || MSB(base_offset, 7))
1383 return 0; /* overflow */
1384 c = base_info[used++];
1385 base_offset = (base_offset << 7) + (c & 127);
1386 }
1387 base_offset = delta_obj_offset - base_offset;
1388 if (base_offset >= delta_obj_offset)
1389 return 0; /* out of bound */
1390 *curpos += used;
1391 } else if (type == OBJ_REF_DELTA) {
1392 /* The base entry _must_ be in the same pack */
1393 base_offset = find_pack_entry_one(base_info, p);
1394 *curpos += 20;
1395 } else
1396 die("I am totally screwed");
1397 return base_offset;
1398 }
1399
1400 /* forward declaration for a mutually recursive function */
1401 static int packed_object_info(struct packed_git *p, off_t offset,
1402 unsigned long *sizep);
1403
1404 static int packed_delta_info(struct packed_git *p,
1405 struct pack_window **w_curs,
1406 off_t curpos,
1407 enum object_type type,
1408 off_t obj_offset,
1409 unsigned long *sizep)
1410 {
1411 off_t base_offset;
1412
1413 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1414 type = packed_object_info(p, base_offset, NULL);
1415
1416 /* We choose to only get the type of the base object and
1417 * ignore potentially corrupt pack file that expects the delta
1418 * based on a base with a wrong size. This saves tons of
1419 * inflate() calls.
1420 */
1421 if (sizep)
1422 *sizep = get_size_from_delta(p, w_curs, curpos);
1423
1424 return type;
1425 }
1426
1427 static int unpack_object_header(struct packed_git *p,
1428 struct pack_window **w_curs,
1429 off_t *curpos,
1430 unsigned long *sizep)
1431 {
1432 unsigned char *base;
1433 unsigned int left;
1434 unsigned long used;
1435 enum object_type type;
1436
1437 /* use_pack() assures us we have [base, base + 20) available
1438 * as a range that we can look at at. (Its actually the hash
1439 * size that is assured.) With our object header encoding
1440 * the maximum deflated object size is 2^137, which is just
1441 * insane, so we know won't exceed what we have been given.
1442 */
1443 base = use_pack(p, w_curs, *curpos, &left);
1444 used = unpack_object_header_gently(base, left, &type, sizep);
1445 if (!used)
1446 die("object offset outside of pack file");
1447 *curpos += used;
1448
1449 return type;
1450 }
1451
1452 const char *packed_object_info_detail(struct packed_git *p,
1453 off_t obj_offset,
1454 unsigned long *size,
1455 unsigned long *store_size,
1456 unsigned int *delta_chain_length,
1457 unsigned char *base_sha1)
1458 {
1459 struct pack_window *w_curs = NULL;
1460 off_t curpos;
1461 unsigned long dummy;
1462 unsigned char *next_sha1;
1463 enum object_type type;
1464 struct revindex_entry *revidx;
1465
1466 *delta_chain_length = 0;
1467 curpos = obj_offset;
1468 type = unpack_object_header(p, &w_curs, &curpos, size);
1469
1470 revidx = find_pack_revindex(p, obj_offset);
1471 *store_size = revidx[1].offset - obj_offset;
1472
1473 for (;;) {
1474 switch (type) {
1475 default:
1476 die("pack %s contains unknown object type %d",
1477 p->pack_name, type);
1478 case OBJ_COMMIT:
1479 case OBJ_TREE:
1480 case OBJ_BLOB:
1481 case OBJ_TAG:
1482 unuse_pack(&w_curs);
1483 return typename(type);
1484 case OBJ_OFS_DELTA:
1485 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1486 if (!obj_offset)
1487 die("pack %s contains bad delta base reference of type %s",
1488 p->pack_name, typename(type));
1489 if (*delta_chain_length == 0) {
1490 revidx = find_pack_revindex(p, obj_offset);
1491 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1492 }
1493 break;
1494 case OBJ_REF_DELTA:
1495 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1496 if (*delta_chain_length == 0)
1497 hashcpy(base_sha1, next_sha1);
1498 obj_offset = find_pack_entry_one(next_sha1, p);
1499 break;
1500 }
1501 (*delta_chain_length)++;
1502 curpos = obj_offset;
1503 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1504 }
1505 }
1506
1507 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1508 unsigned long *sizep)
1509 {
1510 struct pack_window *w_curs = NULL;
1511 unsigned long size;
1512 off_t curpos = obj_offset;
1513 enum object_type type;
1514
1515 type = unpack_object_header(p, &w_curs, &curpos, &size);
1516
1517 switch (type) {
1518 case OBJ_OFS_DELTA:
1519 case OBJ_REF_DELTA:
1520 type = packed_delta_info(p, &w_curs, curpos,
1521 type, obj_offset, sizep);
1522 break;
1523 case OBJ_COMMIT:
1524 case OBJ_TREE:
1525 case OBJ_BLOB:
1526 case OBJ_TAG:
1527 if (sizep)
1528 *sizep = size;
1529 break;
1530 default:
1531 die("pack %s contains unknown object type %d",
1532 p->pack_name, type);
1533 }
1534 unuse_pack(&w_curs);
1535 return type;
1536 }
1537
1538 static void *unpack_compressed_entry(struct packed_git *p,
1539 struct pack_window **w_curs,
1540 off_t curpos,
1541 unsigned long size)
1542 {
1543 int st;
1544 z_stream stream;
1545 unsigned char *buffer, *in;
1546
1547 buffer = xmalloc(size + 1);
1548 buffer[size] = 0;
1549 memset(&stream, 0, sizeof(stream));
1550 stream.next_out = buffer;
1551 stream.avail_out = size;
1552
1553 inflateInit(&stream);
1554 do {
1555 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1556 stream.next_in = in;
1557 st = inflate(&stream, Z_FINISH);
1558 curpos += stream.next_in - in;
1559 } while (st == Z_OK || st == Z_BUF_ERROR);
1560 inflateEnd(&stream);
1561 if ((st != Z_STREAM_END) || stream.total_out != size) {
1562 free(buffer);
1563 return NULL;
1564 }
1565
1566 return buffer;
1567 }
1568
1569 #define MAX_DELTA_CACHE (256)
1570
1571 static size_t delta_base_cached;
1572
1573 static struct delta_base_cache_lru_list {
1574 struct delta_base_cache_lru_list *prev;
1575 struct delta_base_cache_lru_list *next;
1576 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1577
1578 static struct delta_base_cache_entry {
1579 struct delta_base_cache_lru_list lru;
1580 void *data;
1581 struct packed_git *p;
1582 off_t base_offset;
1583 unsigned long size;
1584 enum object_type type;
1585 } delta_base_cache[MAX_DELTA_CACHE];
1586
1587 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1588 {
1589 unsigned long hash;
1590
1591 hash = (unsigned long)p + (unsigned long)base_offset;
1592 hash += (hash >> 8) + (hash >> 16);
1593 return hash % MAX_DELTA_CACHE;
1594 }
1595
1596 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1597 unsigned long *base_size, enum object_type *type, int keep_cache)
1598 {
1599 void *ret;
1600 unsigned long hash = pack_entry_hash(p, base_offset);
1601 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1602
1603 ret = ent->data;
1604 if (ret && ent->p == p && ent->base_offset == base_offset)
1605 goto found_cache_entry;
1606 return unpack_entry(p, base_offset, type, base_size);
1607
1608 found_cache_entry:
1609 if (!keep_cache) {
1610 ent->data = NULL;
1611 ent->lru.next->prev = ent->lru.prev;
1612 ent->lru.prev->next = ent->lru.next;
1613 delta_base_cached -= ent->size;
1614 } else {
1615 ret = xmemdupz(ent->data, ent->size);
1616 }
1617 *type = ent->type;
1618 *base_size = ent->size;
1619 return ret;
1620 }
1621
1622 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1623 {
1624 if (ent->data) {
1625 free(ent->data);
1626 ent->data = NULL;
1627 ent->lru.next->prev = ent->lru.prev;
1628 ent->lru.prev->next = ent->lru.next;
1629 delta_base_cached -= ent->size;
1630 }
1631 }
1632
1633 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1634 void *base, unsigned long base_size, enum object_type type)
1635 {
1636 unsigned long hash = pack_entry_hash(p, base_offset);
1637 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1638 struct delta_base_cache_lru_list *lru;
1639
1640 release_delta_base_cache(ent);
1641 delta_base_cached += base_size;
1642
1643 for (lru = delta_base_cache_lru.next;
1644 delta_base_cached > delta_base_cache_limit
1645 && lru != &delta_base_cache_lru;
1646 lru = lru->next) {
1647 struct delta_base_cache_entry *f = (void *)lru;
1648 if (f->type == OBJ_BLOB)
1649 release_delta_base_cache(f);
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 release_delta_base_cache(f);
1657 }
1658
1659 ent->p = p;
1660 ent->base_offset = base_offset;
1661 ent->type = type;
1662 ent->data = base;
1663 ent->size = base_size;
1664 ent->lru.next = &delta_base_cache_lru;
1665 ent->lru.prev = delta_base_cache_lru.prev;
1666 delta_base_cache_lru.prev->next = &ent->lru;
1667 delta_base_cache_lru.prev = &ent->lru;
1668 }
1669
1670 static void *unpack_delta_entry(struct packed_git *p,
1671 struct pack_window **w_curs,
1672 off_t curpos,
1673 unsigned long delta_size,
1674 off_t obj_offset,
1675 enum object_type *type,
1676 unsigned long *sizep)
1677 {
1678 void *delta_data, *result, *base;
1679 unsigned long base_size;
1680 off_t base_offset;
1681
1682 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1683 if (!base_offset) {
1684 error("failed to validate delta base reference "
1685 "at offset %"PRIuMAX" from %s",
1686 (uintmax_t)curpos, p->pack_name);
1687 return NULL;
1688 }
1689 unuse_pack(w_curs);
1690 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1691 if (!base) {
1692 /*
1693 * We're probably in deep shit, but let's try to fetch
1694 * the required base anyway from another pack or loose.
1695 * This is costly but should happen only in the presence
1696 * of a corrupted pack, and is better than failing outright.
1697 */
1698 struct revindex_entry *revidx = find_pack_revindex(p, base_offset);
1699 const unsigned char *base_sha1 =
1700 nth_packed_object_sha1(p, revidx->nr);
1701 error("failed to read delta base object %s"
1702 " at offset %"PRIuMAX" from %s",
1703 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1704 p->pack_name);
1705 mark_bad_packed_object(p, base_sha1);
1706 base = read_object(base_sha1, type, &base_size);
1707 if (!base)
1708 return NULL;
1709 }
1710
1711 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1712 if (!delta_data) {
1713 error("failed to unpack compressed delta "
1714 "at offset %"PRIuMAX" from %s",
1715 (uintmax_t)curpos, p->pack_name);
1716 free(base);
1717 return NULL;
1718 }
1719 result = patch_delta(base, base_size,
1720 delta_data, delta_size,
1721 sizep);
1722 if (!result)
1723 die("failed to apply delta");
1724 free(delta_data);
1725 add_delta_base_cache(p, base_offset, base, base_size, *type);
1726 return result;
1727 }
1728
1729 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1730 enum object_type *type, unsigned long *sizep)
1731 {
1732 struct pack_window *w_curs = NULL;
1733 off_t curpos = obj_offset;
1734 void *data;
1735
1736 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1737 switch (*type) {
1738 case OBJ_OFS_DELTA:
1739 case OBJ_REF_DELTA:
1740 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1741 obj_offset, type, sizep);
1742 break;
1743 case OBJ_COMMIT:
1744 case OBJ_TREE:
1745 case OBJ_BLOB:
1746 case OBJ_TAG:
1747 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1748 break;
1749 default:
1750 data = NULL;
1751 error("unknown object type %i at offset %"PRIuMAX" in %s",
1752 *type, (uintmax_t)obj_offset, p->pack_name);
1753 }
1754 unuse_pack(&w_curs);
1755 return data;
1756 }
1757
1758 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1759 uint32_t n)
1760 {
1761 const unsigned char *index = p->index_data;
1762 if (!index) {
1763 if (open_pack_index(p))
1764 return NULL;
1765 index = p->index_data;
1766 }
1767 if (n >= p->num_objects)
1768 return NULL;
1769 index += 4 * 256;
1770 if (p->index_version == 1) {
1771 return index + 24 * n + 4;
1772 } else {
1773 index += 8;
1774 return index + 20 * n;
1775 }
1776 }
1777
1778 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1779 {
1780 const unsigned char *index = p->index_data;
1781 index += 4 * 256;
1782 if (p->index_version == 1) {
1783 return ntohl(*((uint32_t *)(index + 24 * n)));
1784 } else {
1785 uint32_t off;
1786 index += 8 + p->num_objects * (20 + 4);
1787 off = ntohl(*((uint32_t *)(index + 4 * n)));
1788 if (!(off & 0x80000000))
1789 return off;
1790 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1791 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1792 ntohl(*((uint32_t *)(index + 4)));
1793 }
1794 }
1795
1796 off_t find_pack_entry_one(const unsigned char *sha1,
1797 struct packed_git *p)
1798 {
1799 const uint32_t *level1_ofs = p->index_data;
1800 const unsigned char *index = p->index_data;
1801 unsigned hi, lo, stride;
1802 static int use_lookup = -1;
1803 static int debug_lookup = -1;
1804
1805 if (debug_lookup < 0)
1806 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1807
1808 if (!index) {
1809 if (open_pack_index(p))
1810 return 0;
1811 level1_ofs = p->index_data;
1812 index = p->index_data;
1813 }
1814 if (p->index_version > 1) {
1815 level1_ofs += 2;
1816 index += 8;
1817 }
1818 index += 4 * 256;
1819 hi = ntohl(level1_ofs[*sha1]);
1820 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1821 if (p->index_version > 1) {
1822 stride = 20;
1823 } else {
1824 stride = 24;
1825 index += 4;
1826 }
1827
1828 if (debug_lookup)
1829 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1830 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1831
1832 if (use_lookup < 0)
1833 use_lookup = !!getenv("GIT_USE_LOOKUP");
1834 if (use_lookup) {
1835 int pos = sha1_entry_pos(index, stride, 0,
1836 lo, hi, p->num_objects, sha1);
1837 if (pos < 0)
1838 return 0;
1839 return nth_packed_object_offset(p, pos);
1840 }
1841
1842 do {
1843 unsigned mi = (lo + hi) / 2;
1844 int cmp = hashcmp(index + mi * stride, sha1);
1845
1846 if (debug_lookup)
1847 printf("lo %u hi %u rg %u mi %u\n",
1848 lo, hi, hi - lo, mi);
1849 if (!cmp)
1850 return nth_packed_object_offset(p, mi);
1851 if (cmp > 0)
1852 hi = mi;
1853 else
1854 lo = mi+1;
1855 } while (lo < hi);
1856 return 0;
1857 }
1858
1859 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1860 {
1861 static struct packed_git *last_found = (void *)1;
1862 struct packed_git *p;
1863 off_t offset;
1864
1865 prepare_packed_git();
1866 if (!packed_git)
1867 return 0;
1868 p = (last_found == (void *)1) ? packed_git : last_found;
1869
1870 do {
1871 if (p->num_bad_objects) {
1872 unsigned i;
1873 for (i = 0; i < p->num_bad_objects; i++)
1874 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1875 goto next;
1876 }
1877
1878 offset = find_pack_entry_one(sha1, p);
1879 if (offset) {
1880 /*
1881 * We are about to tell the caller where they can
1882 * locate the requested object. We better make
1883 * sure the packfile is still here and can be
1884 * accessed before supplying that answer, as
1885 * it may have been deleted since the index
1886 * was loaded!
1887 */
1888 if (p->pack_fd == -1 && open_packed_git(p)) {
1889 error("packfile %s cannot be accessed", p->pack_name);
1890 goto next;
1891 }
1892 e->offset = offset;
1893 e->p = p;
1894 hashcpy(e->sha1, sha1);
1895 last_found = p;
1896 return 1;
1897 }
1898
1899 next:
1900 if (p == last_found)
1901 p = packed_git;
1902 else
1903 p = p->next;
1904 if (p == last_found)
1905 p = p->next;
1906 } while (p);
1907 return 0;
1908 }
1909
1910 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1911 struct packed_git *packs)
1912 {
1913 struct packed_git *p;
1914
1915 for (p = packs; p; p = p->next) {
1916 if (find_pack_entry_one(sha1, p))
1917 return p;
1918 }
1919 return NULL;
1920
1921 }
1922
1923 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1924 {
1925 int status;
1926 unsigned long mapsize, size;
1927 void *map;
1928 z_stream stream;
1929 char hdr[32];
1930
1931 map = map_sha1_file(sha1, &mapsize);
1932 if (!map)
1933 return error("unable to find %s", sha1_to_hex(sha1));
1934 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1935 status = error("unable to unpack %s header",
1936 sha1_to_hex(sha1));
1937 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1938 status = error("unable to parse %s header", sha1_to_hex(sha1));
1939 else if (sizep)
1940 *sizep = size;
1941 inflateEnd(&stream);
1942 munmap(map, mapsize);
1943 return status;
1944 }
1945
1946 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1947 {
1948 struct pack_entry e;
1949 int status;
1950
1951 if (!find_pack_entry(sha1, &e)) {
1952 /* Most likely it's a loose object. */
1953 status = sha1_loose_object_info(sha1, sizep);
1954 if (status >= 0)
1955 return status;
1956
1957 /* Not a loose object; someone else may have just packed it. */
1958 reprepare_packed_git();
1959 if (!find_pack_entry(sha1, &e))
1960 return status;
1961 }
1962 return packed_object_info(e.p, e.offset, sizep);
1963 }
1964
1965 static void *read_packed_sha1(const unsigned char *sha1,
1966 enum object_type *type, unsigned long *size)
1967 {
1968 struct pack_entry e;
1969 void *data;
1970
1971 if (!find_pack_entry(sha1, &e))
1972 return NULL;
1973 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1974 if (!data) {
1975 /*
1976 * We're probably in deep shit, but let's try to fetch
1977 * the required object anyway from another pack or loose.
1978 * This should happen only in the presence of a corrupted
1979 * pack, and is better than failing outright.
1980 */
1981 error("failed to read object %s at offset %"PRIuMAX" from %s",
1982 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
1983 mark_bad_packed_object(e.p, sha1);
1984 data = read_object(sha1, type, size);
1985 }
1986 return data;
1987 }
1988
1989 /*
1990 * This is meant to hold a *small* number of objects that you would
1991 * want read_sha1_file() to be able to return, but yet you do not want
1992 * to write them into the object store (e.g. a browse-only
1993 * application).
1994 */
1995 static struct cached_object {
1996 unsigned char sha1[20];
1997 enum object_type type;
1998 void *buf;
1999 unsigned long size;
2000 } *cached_objects;
2001 static int cached_object_nr, cached_object_alloc;
2002
2003 static struct cached_object empty_tree = {
2004 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
2005 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
2006 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
2007 OBJ_TREE,
2008 "",
2009 0
2010 };
2011
2012 static struct cached_object *find_cached_object(const unsigned char *sha1)
2013 {
2014 int i;
2015 struct cached_object *co = cached_objects;
2016
2017 for (i = 0; i < cached_object_nr; i++, co++) {
2018 if (!hashcmp(co->sha1, sha1))
2019 return co;
2020 }
2021 if (!hashcmp(sha1, empty_tree.sha1))
2022 return &empty_tree;
2023 return NULL;
2024 }
2025
2026 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2027 unsigned char *sha1)
2028 {
2029 struct cached_object *co;
2030
2031 hash_sha1_file(buf, len, typename(type), sha1);
2032 if (has_sha1_file(sha1) || find_cached_object(sha1))
2033 return 0;
2034 if (cached_object_alloc <= cached_object_nr) {
2035 cached_object_alloc = alloc_nr(cached_object_alloc);
2036 cached_objects = xrealloc(cached_objects,
2037 sizeof(*cached_objects) *
2038 cached_object_alloc);
2039 }
2040 co = &cached_objects[cached_object_nr++];
2041 co->size = len;
2042 co->type = type;
2043 co->buf = xmalloc(len);
2044 memcpy(co->buf, buf, len);
2045 hashcpy(co->sha1, sha1);
2046 return 0;
2047 }
2048
2049 void *read_object(const unsigned char *sha1, enum object_type *type,
2050 unsigned long *size)
2051 {
2052 unsigned long mapsize;
2053 void *map, *buf;
2054 struct cached_object *co;
2055
2056 co = find_cached_object(sha1);
2057 if (co) {
2058 *type = co->type;
2059 *size = co->size;
2060 return xmemdupz(co->buf, co->size);
2061 }
2062
2063 buf = read_packed_sha1(sha1, type, size);
2064 if (buf)
2065 return buf;
2066 map = map_sha1_file(sha1, &mapsize);
2067 if (map) {
2068 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2069 munmap(map, mapsize);
2070 return buf;
2071 }
2072 reprepare_packed_git();
2073 return read_packed_sha1(sha1, type, size);
2074 }
2075
2076 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
2077 unsigned long *size)
2078 {
2079 void *data = read_object(sha1, type, size);
2080 /* legacy behavior is to die on corrupted objects */
2081 if (!data && (has_loose_object(sha1) || has_packed_and_bad(sha1)))
2082 die("object %s is corrupted", sha1_to_hex(sha1));
2083 return data;
2084 }
2085
2086 void *read_object_with_reference(const unsigned char *sha1,
2087 const char *required_type_name,
2088 unsigned long *size,
2089 unsigned char *actual_sha1_return)
2090 {
2091 enum object_type type, required_type;
2092 void *buffer;
2093 unsigned long isize;
2094 unsigned char actual_sha1[20];
2095
2096 required_type = type_from_string(required_type_name);
2097 hashcpy(actual_sha1, sha1);
2098 while (1) {
2099 int ref_length = -1;
2100 const char *ref_type = NULL;
2101
2102 buffer = read_sha1_file(actual_sha1, &type, &isize);
2103 if (!buffer)
2104 return NULL;
2105 if (type == required_type) {
2106 *size = isize;
2107 if (actual_sha1_return)
2108 hashcpy(actual_sha1_return, actual_sha1);
2109 return buffer;
2110 }
2111 /* Handle references */
2112 else if (type == OBJ_COMMIT)
2113 ref_type = "tree ";
2114 else if (type == OBJ_TAG)
2115 ref_type = "object ";
2116 else {
2117 free(buffer);
2118 return NULL;
2119 }
2120 ref_length = strlen(ref_type);
2121
2122 if (ref_length + 40 > isize ||
2123 memcmp(buffer, ref_type, ref_length) ||
2124 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2125 free(buffer);
2126 return NULL;
2127 }
2128 free(buffer);
2129 /* Now we have the ID of the referred-to object in
2130 * actual_sha1. Check again. */
2131 }
2132 }
2133
2134 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2135 const char *type, unsigned char *sha1,
2136 char *hdr, int *hdrlen)
2137 {
2138 SHA_CTX c;
2139
2140 /* Generate the header */
2141 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2142
2143 /* Sha1.. */
2144 SHA1_Init(&c);
2145 SHA1_Update(&c, hdr, *hdrlen);
2146 SHA1_Update(&c, buf, len);
2147 SHA1_Final(sha1, &c);
2148 }
2149
2150 /*
2151 * Move the just written object into its final resting place
2152 */
2153 int move_temp_to_file(const char *tmpfile, const char *filename)
2154 {
2155 int ret = 0;
2156 if (link(tmpfile, filename))
2157 ret = errno;
2158
2159 /*
2160 * Coda hack - coda doesn't like cross-directory links,
2161 * so we fall back to a rename, which will mean that it
2162 * won't be able to check collisions, but that's not a
2163 * big deal.
2164 *
2165 * The same holds for FAT formatted media.
2166 *
2167 * When this succeeds, we just return 0. We have nothing
2168 * left to unlink.
2169 */
2170 if (ret && ret != EEXIST) {
2171 if (!rename(tmpfile, filename))
2172 return 0;
2173 ret = errno;
2174 }
2175 unlink(tmpfile);
2176 if (ret) {
2177 if (ret != EEXIST) {
2178 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2179 }
2180 /* FIXME!!! Collision check here ? */
2181 }
2182
2183 return 0;
2184 }
2185
2186 static int write_buffer(int fd, const void *buf, size_t len)
2187 {
2188 if (write_in_full(fd, buf, len) < 0)
2189 return error("file write error (%s)", strerror(errno));
2190 return 0;
2191 }
2192
2193 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2194 unsigned char *sha1)
2195 {
2196 char hdr[32];
2197 int hdrlen;
2198 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2199 return 0;
2200 }
2201
2202 /* Finalize a file on disk, and close it. */
2203 static void close_sha1_file(int fd)
2204 {
2205 if (fsync_object_files)
2206 fsync_or_die(fd, "sha1 file");
2207 fchmod(fd, 0444);
2208 if (close(fd) != 0)
2209 die("unable to write sha1 file");
2210 }
2211
2212 /* Size of directory component, including the ending '/' */
2213 static inline int directory_size(const char *filename)
2214 {
2215 const char *s = strrchr(filename, '/');
2216 if (!s)
2217 return 0;
2218 return s - filename + 1;
2219 }
2220
2221 /*
2222 * This creates a temporary file in the same directory as the final
2223 * 'filename'
2224 *
2225 * We want to avoid cross-directory filename renames, because those
2226 * can have problems on various filesystems (FAT, NFS, Coda).
2227 */
2228 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2229 {
2230 int fd, dirlen = directory_size(filename);
2231
2232 if (dirlen + 20 > bufsiz) {
2233 errno = ENAMETOOLONG;
2234 return -1;
2235 }
2236 memcpy(buffer, filename, dirlen);
2237 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2238 fd = mkstemp(buffer);
2239 if (fd < 0 && dirlen && errno == ENOENT) {
2240 /* Make sure the directory exists */
2241 memcpy(buffer, filename, dirlen);
2242 buffer[dirlen-1] = 0;
2243 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2244 return -1;
2245
2246 /* Try again */
2247 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2248 fd = mkstemp(buffer);
2249 }
2250 return fd;
2251 }
2252
2253 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2254 void *buf, unsigned long len, time_t mtime)
2255 {
2256 int fd, size, ret;
2257 unsigned char *compressed;
2258 z_stream stream;
2259 char *filename;
2260 static char tmpfile[PATH_MAX];
2261
2262 filename = sha1_file_name(sha1);
2263 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2264 if (fd < 0) {
2265 if (errno == EACCES)
2266 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2267 else
2268 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2269 }
2270
2271 /* Set it up */
2272 memset(&stream, 0, sizeof(stream));
2273 deflateInit(&stream, zlib_compression_level);
2274 size = 8 + deflateBound(&stream, len+hdrlen);
2275 compressed = xmalloc(size);
2276
2277 /* Compress it */
2278 stream.next_out = compressed;
2279 stream.avail_out = size;
2280
2281 /* First header.. */
2282 stream.next_in = (unsigned char *)hdr;
2283 stream.avail_in = hdrlen;
2284 while (deflate(&stream, 0) == Z_OK)
2285 /* nothing */;
2286
2287 /* Then the data itself.. */
2288 stream.next_in = buf;
2289 stream.avail_in = len;
2290 ret = deflate(&stream, Z_FINISH);
2291 if (ret != Z_STREAM_END)
2292 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2293
2294 ret = deflateEnd(&stream);
2295 if (ret != Z_OK)
2296 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2297
2298 size = stream.total_out;
2299
2300 if (write_buffer(fd, compressed, size) < 0)
2301 die("unable to write sha1 file");
2302 close_sha1_file(fd);
2303 free(compressed);
2304
2305 if (mtime) {
2306 struct utimbuf utb;
2307 utb.actime = mtime;
2308 utb.modtime = mtime;
2309 if (utime(tmpfile, &utb) < 0)
2310 warning("failed utime() on %s: %s",
2311 tmpfile, strerror(errno));
2312 }
2313
2314 return move_temp_to_file(tmpfile, filename);
2315 }
2316
2317 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2318 {
2319 unsigned char sha1[20];
2320 char hdr[32];
2321 int hdrlen;
2322
2323 /* Normally if we have it in the pack then we do not bother writing
2324 * it out into .git/objects/??/?{38} file.
2325 */
2326 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2327 if (returnsha1)
2328 hashcpy(returnsha1, sha1);
2329 if (has_sha1_file(sha1))
2330 return 0;
2331 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2332 }
2333
2334 int force_object_loose(const unsigned char *sha1, time_t mtime)
2335 {
2336 void *buf;
2337 unsigned long len;
2338 enum object_type type;
2339 char hdr[32];
2340 int hdrlen;
2341 int ret;
2342
2343 if (has_loose_object(sha1))
2344 return 0;
2345 buf = read_packed_sha1(sha1, &type, &len);
2346 if (!buf)
2347 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2348 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2349 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2350 free(buf);
2351
2352 return ret;
2353 }
2354
2355 int has_pack_index(const unsigned char *sha1)
2356 {
2357 struct stat st;
2358 if (stat(sha1_pack_index_name(sha1), &st))
2359 return 0;
2360 return 1;
2361 }
2362
2363 int has_pack_file(const unsigned char *sha1)
2364 {
2365 struct stat st;
2366 if (stat(sha1_pack_name(sha1), &st))
2367 return 0;
2368 return 1;
2369 }
2370
2371 int has_sha1_pack(const unsigned char *sha1)
2372 {
2373 struct pack_entry e;
2374 return find_pack_entry(sha1, &e);
2375 }
2376
2377 int has_sha1_file(const unsigned char *sha1)
2378 {
2379 struct pack_entry e;
2380
2381 if (find_pack_entry(sha1, &e))
2382 return 1;
2383 return has_loose_object(sha1);
2384 }
2385
2386 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2387 {
2388 struct strbuf buf;
2389 int ret;
2390
2391 strbuf_init(&buf, 0);
2392 if (strbuf_read(&buf, fd, 4096) < 0) {
2393 strbuf_release(&buf);
2394 return -1;
2395 }
2396
2397 if (!type)
2398 type = blob_type;
2399 if (write_object)
2400 ret = write_sha1_file(buf.buf, buf.len, type, sha1);
2401 else
2402 ret = hash_sha1_file(buf.buf, buf.len, type, sha1);
2403 strbuf_release(&buf);
2404
2405 return ret;
2406 }
2407
2408 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2409 enum object_type type, const char *path)
2410 {
2411 size_t size = xsize_t(st->st_size);
2412 void *buf = NULL;
2413 int ret, re_allocated = 0;
2414
2415 if (size)
2416 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2417 close(fd);
2418
2419 if (!type)
2420 type = OBJ_BLOB;
2421
2422 /*
2423 * Convert blobs to git internal format
2424 */
2425 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2426 struct strbuf nbuf;
2427 strbuf_init(&nbuf, 0);
2428 if (convert_to_git(path, buf, size, &nbuf,
2429 write_object ? safe_crlf : 0)) {
2430 munmap(buf, size);
2431 buf = strbuf_detach(&nbuf, &size);
2432 re_allocated = 1;
2433 }
2434 }
2435
2436 if (write_object)
2437 ret = write_sha1_file(buf, size, typename(type), sha1);
2438 else
2439 ret = hash_sha1_file(buf, size, typename(type), sha1);
2440 if (re_allocated) {
2441 free(buf);
2442 return ret;
2443 }
2444 if (size)
2445 munmap(buf, size);
2446 return ret;
2447 }
2448
2449 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2450 {
2451 int fd;
2452 char *target;
2453 size_t len;
2454
2455 switch (st->st_mode & S_IFMT) {
2456 case S_IFREG:
2457 fd = open(path, O_RDONLY);
2458 if (fd < 0)
2459 return error("open(\"%s\"): %s", path,
2460 strerror(errno));
2461 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2462 return error("%s: failed to insert into database",
2463 path);
2464 break;
2465 case S_IFLNK:
2466 len = xsize_t(st->st_size);
2467 target = xmalloc(len + 1);
2468 if (readlink(path, target, len + 1) != st->st_size) {
2469 char *errstr = strerror(errno);
2470 free(target);
2471 return error("readlink(\"%s\"): %s", path,
2472 errstr);
2473 }
2474 if (!write_object)
2475 hash_sha1_file(target, len, blob_type, sha1);
2476 else if (write_sha1_file(target, len, blob_type, sha1))
2477 return error("%s: failed to insert into database",
2478 path);
2479 free(target);
2480 break;
2481 case S_IFDIR:
2482 return resolve_gitlink_ref(path, "HEAD", sha1);
2483 default:
2484 return error("%s: unsupported file type", path);
2485 }
2486 return 0;
2487 }
2488
2489 int read_pack_header(int fd, struct pack_header *header)
2490 {
2491 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2492 /* "eof before pack header was fully read" */
2493 return PH_ERROR_EOF;
2494
2495 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2496 /* "protocol error (pack signature mismatch detected)" */
2497 return PH_ERROR_PACK_SIGNATURE;
2498 if (!pack_version_ok(header->hdr_version))
2499 /* "protocol error (pack version unsupported)" */
2500 return PH_ERROR_PROTOCOL;
2501 return 0;
2502 }