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