A loose object is not corrupt if it cannot be read due to EMFILE
[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 int safe_create_leading_directories(char *path)
39 {
40 char *pos = path + offset_1st_component(path);
41 struct stat st;
42
43 while (pos) {
44 pos = strchr(pos, '/');
45 if (!pos)
46 break;
47 while (*++pos == '/')
48 ;
49 if (!*pos)
50 break;
51 *--pos = '\0';
52 if (!stat(path, &st)) {
53 /* path exists */
54 if (!S_ISDIR(st.st_mode)) {
55 *pos = '/';
56 return -3;
57 }
58 }
59 else if (mkdir(path, 0777)) {
60 *pos = '/';
61 return -1;
62 }
63 else if (adjust_shared_perm(path)) {
64 *pos = '/';
65 return -2;
66 }
67 *pos++ = '/';
68 }
69 return 0;
70 }
71
72 int safe_create_leading_directories_const(const char *path)
73 {
74 /* path points to cache entries, so xstrdup before messing with it */
75 char *buf = xstrdup(path);
76 int result = safe_create_leading_directories(buf);
77 free(buf);
78 return result;
79 }
80
81 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
82 {
83 int i;
84 for (i = 0; i < 20; i++) {
85 static char hex[] = "0123456789abcdef";
86 unsigned int val = sha1[i];
87 char *pos = pathbuf + i*2 + (i > 0);
88 *pos++ = hex[val >> 4];
89 *pos = hex[val & 0xf];
90 }
91 }
92
93 /*
94 * NOTE! This returns a statically allocated buffer, so you have to be
95 * careful about using it. Do an "xstrdup()" if you need to save the
96 * filename.
97 *
98 * Also note that this returns the location for creating. Reading
99 * SHA1 file can happen from any alternate directory listed in the
100 * DB_ENVIRONMENT environment variable if it is not found in
101 * the primary object database.
102 */
103 char *sha1_file_name(const unsigned char *sha1)
104 {
105 static char buf[PATH_MAX];
106 const char *objdir;
107 int len;
108
109 objdir = get_object_directory();
110 len = strlen(objdir);
111
112 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
113 if (len + 43 > PATH_MAX)
114 die("insanely long object directory %s", objdir);
115 memcpy(buf, objdir, len);
116 buf[len] = '/';
117 buf[len+3] = '/';
118 buf[len+42] = '\0';
119 fill_sha1_path(buf + len + 1, sha1);
120 return buf;
121 }
122
123 static char *sha1_get_pack_name(const unsigned char *sha1,
124 char **name, char **base, const char *which)
125 {
126 static const char hex[] = "0123456789abcdef";
127 char *buf;
128 int i;
129
130 if (!*base) {
131 const char *sha1_file_directory = get_object_directory();
132 int len = strlen(sha1_file_directory);
133 *base = xmalloc(len + 60);
134 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
135 sha1_file_directory, which);
136 *name = *base + len + 11;
137 }
138
139 buf = *name;
140
141 for (i = 0; i < 20; i++) {
142 unsigned int val = *sha1++;
143 *buf++ = hex[val >> 4];
144 *buf++ = hex[val & 0xf];
145 }
146
147 return *base;
148 }
149
150 char *sha1_pack_name(const unsigned char *sha1)
151 {
152 static char *name, *base;
153
154 return sha1_get_pack_name(sha1, &name, &base, "pack");
155 }
156
157 char *sha1_pack_index_name(const unsigned char *sha1)
158 {
159 static char *name, *base;
160
161 return sha1_get_pack_name(sha1, &name, &base, "idx");
162 }
163
164 struct alternate_object_database *alt_odb_list;
165 static struct alternate_object_database **alt_odb_tail;
166
167 static void read_info_alternates(const char * alternates, int depth);
168
169 /*
170 * Prepare alternate object database registry.
171 *
172 * The variable alt_odb_list points at the list of struct
173 * alternate_object_database. The elements on this list come from
174 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
175 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
176 * whose contents is similar to that environment variable but can be
177 * LF separated. Its base points at a statically allocated buffer that
178 * contains "/the/directory/corresponding/to/.git/objects/...", while
179 * its name points just after the slash at the end of ".git/objects/"
180 * in the example above, and has enough space to hold 40-byte hex
181 * SHA1, an extra slash for the first level indirection, and the
182 * terminating NUL.
183 */
184 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
185 {
186 const char *objdir = get_object_directory();
187 struct alternate_object_database *ent;
188 struct alternate_object_database *alt;
189 /* 43 = 40-byte + 2 '/' + terminating NUL */
190 int pfxlen = len;
191 int entlen = pfxlen + 43;
192 int base_len = -1;
193
194 if (!is_absolute_path(entry) && relative_base) {
195 /* Relative alt-odb */
196 if (base_len < 0)
197 base_len = strlen(relative_base) + 1;
198 entlen += base_len;
199 pfxlen += base_len;
200 }
201 ent = xmalloc(sizeof(*ent) + entlen);
202
203 if (!is_absolute_path(entry) && relative_base) {
204 memcpy(ent->base, relative_base, base_len - 1);
205 ent->base[base_len - 1] = '/';
206 memcpy(ent->base + base_len, entry, len);
207 }
208 else
209 memcpy(ent->base, entry, pfxlen);
210
211 ent->name = ent->base + pfxlen + 1;
212 ent->base[pfxlen + 3] = '/';
213 ent->base[pfxlen] = ent->base[entlen-1] = 0;
214
215 /* Detect cases where alternate disappeared */
216 if (!is_directory(ent->base)) {
217 error("object directory %s does not exist; "
218 "check .git/objects/info/alternates.",
219 ent->base);
220 free(ent);
221 return -1;
222 }
223
224 /* Prevent the common mistake of listing the same
225 * thing twice, or object directory itself.
226 */
227 for (alt = alt_odb_list; alt; alt = alt->next) {
228 if (!memcmp(ent->base, alt->base, pfxlen)) {
229 free(ent);
230 return -1;
231 }
232 }
233 if (!memcmp(ent->base, objdir, pfxlen)) {
234 free(ent);
235 return -1;
236 }
237
238 /* add the alternate entry */
239 *alt_odb_tail = ent;
240 alt_odb_tail = &(ent->next);
241 ent->next = NULL;
242
243 /* recursively add alternates */
244 read_info_alternates(ent->base, depth + 1);
245
246 ent->base[pfxlen] = '/';
247
248 return 0;
249 }
250
251 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
252 const char *relative_base, int depth)
253 {
254 const char *cp, *last;
255
256 if (depth > 5) {
257 error("%s: ignoring alternate object stores, nesting too deep.",
258 relative_base);
259 return;
260 }
261
262 last = alt;
263 while (last < ep) {
264 cp = last;
265 if (cp < ep && *cp == '#') {
266 while (cp < ep && *cp != sep)
267 cp++;
268 last = cp + 1;
269 continue;
270 }
271 while (cp < ep && *cp != sep)
272 cp++;
273 if (last != cp) {
274 if (!is_absolute_path(last) && depth) {
275 error("%s: ignoring relative alternate object store %s",
276 relative_base, last);
277 } else {
278 link_alt_odb_entry(last, cp - last,
279 relative_base, depth);
280 }
281 }
282 while (cp < ep && *cp == sep)
283 cp++;
284 last = cp;
285 }
286 }
287
288 static void read_info_alternates(const char * relative_base, int depth)
289 {
290 char *map;
291 size_t mapsz;
292 struct stat st;
293 const char alt_file_name[] = "info/alternates";
294 /* Given that relative_base is no longer than PATH_MAX,
295 ensure that "path" has enough space to append "/", the
296 file name, "info/alternates", and a trailing NUL. */
297 char path[PATH_MAX + 1 + sizeof alt_file_name];
298 int fd;
299
300 sprintf(path, "%s/%s", relative_base, alt_file_name);
301 fd = open(path, O_RDONLY);
302 if (fd < 0)
303 return;
304 if (fstat(fd, &st) || (st.st_size == 0)) {
305 close(fd);
306 return;
307 }
308 mapsz = xsize_t(st.st_size);
309 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
310 close(fd);
311
312 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
313
314 munmap(map, mapsz);
315 }
316
317 void add_to_alternates_file(const char *reference)
318 {
319 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
320 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
321 char *alt = mkpath("%s/objects\n", reference);
322 write_or_die(fd, alt, strlen(alt));
323 if (commit_lock_file(lock))
324 die("could not close alternates file");
325 if (alt_odb_tail)
326 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
327 }
328
329 void foreach_alt_odb(alt_odb_fn fn, void *cb)
330 {
331 struct alternate_object_database *ent;
332
333 prepare_alt_odb();
334 for (ent = alt_odb_list; ent; ent = ent->next)
335 if (fn(ent, cb))
336 return;
337 }
338
339 void prepare_alt_odb(void)
340 {
341 const char *alt;
342
343 if (alt_odb_tail)
344 return;
345
346 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
347 if (!alt) alt = "";
348
349 alt_odb_tail = &alt_odb_list;
350 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
351
352 read_info_alternates(get_object_directory(), 0);
353 }
354
355 static int has_loose_object_local(const unsigned char *sha1)
356 {
357 char *name = sha1_file_name(sha1);
358 return !access(name, F_OK);
359 }
360
361 int has_loose_object_nonlocal(const unsigned char *sha1)
362 {
363 struct alternate_object_database *alt;
364 prepare_alt_odb();
365 for (alt = alt_odb_list; alt; alt = alt->next) {
366 fill_sha1_path(alt->name, sha1);
367 if (!access(alt->base, F_OK))
368 return 1;
369 }
370 return 0;
371 }
372
373 static int has_loose_object(const unsigned char *sha1)
374 {
375 return has_loose_object_local(sha1) ||
376 has_loose_object_nonlocal(sha1);
377 }
378
379 static unsigned int pack_used_ctr;
380 static unsigned int pack_mmap_calls;
381 static unsigned int peak_pack_open_windows;
382 static unsigned int pack_open_windows;
383 static size_t peak_pack_mapped;
384 static size_t pack_mapped;
385 struct packed_git *packed_git;
386
387 void pack_report(void)
388 {
389 fprintf(stderr,
390 "pack_report: getpagesize() = %10" SZ_FMT "\n"
391 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
392 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
393 sz_fmt(getpagesize()),
394 sz_fmt(packed_git_window_size),
395 sz_fmt(packed_git_limit));
396 fprintf(stderr,
397 "pack_report: pack_used_ctr = %10u\n"
398 "pack_report: pack_mmap_calls = %10u\n"
399 "pack_report: pack_open_windows = %10u / %10u\n"
400 "pack_report: pack_mapped = "
401 "%10" SZ_FMT " / %10" SZ_FMT "\n",
402 pack_used_ctr,
403 pack_mmap_calls,
404 pack_open_windows, peak_pack_open_windows,
405 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
406 }
407
408 static int check_packed_git_idx(const char *path, struct packed_git *p)
409 {
410 void *idx_map;
411 struct pack_idx_header *hdr;
412 size_t idx_size;
413 uint32_t version, nr, i, *index;
414 int fd = open(path, O_RDONLY);
415 struct stat st;
416
417 if (fd < 0)
418 return -1;
419 if (fstat(fd, &st)) {
420 close(fd);
421 return -1;
422 }
423 idx_size = xsize_t(st.st_size);
424 if (idx_size < 4 * 256 + 20 + 20) {
425 close(fd);
426 return error("index file %s is too small", path);
427 }
428 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
429 close(fd);
430
431 hdr = idx_map;
432 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
433 version = ntohl(hdr->idx_version);
434 if (version < 2 || version > 2) {
435 munmap(idx_map, idx_size);
436 return error("index file %s is version %"PRIu32
437 " and is not supported by this binary"
438 " (try upgrading GIT to a newer version)",
439 path, version);
440 }
441 } else
442 version = 1;
443
444 nr = 0;
445 index = idx_map;
446 if (version > 1)
447 index += 2; /* skip index header */
448 for (i = 0; i < 256; i++) {
449 uint32_t n = ntohl(index[i]);
450 if (n < nr) {
451 munmap(idx_map, idx_size);
452 return error("non-monotonic index %s", path);
453 }
454 nr = n;
455 }
456
457 if (version == 1) {
458 /*
459 * Total size:
460 * - 256 index entries 4 bytes each
461 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
462 * - 20-byte SHA1 of the packfile
463 * - 20-byte SHA1 file checksum
464 */
465 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
466 munmap(idx_map, idx_size);
467 return error("wrong index v1 file size in %s", path);
468 }
469 } else if (version == 2) {
470 /*
471 * Minimum size:
472 * - 8 bytes of header
473 * - 256 index entries 4 bytes each
474 * - 20-byte sha1 entry * nr
475 * - 4-byte crc entry * nr
476 * - 4-byte offset entry * nr
477 * - 20-byte SHA1 of the packfile
478 * - 20-byte SHA1 file checksum
479 * And after the 4-byte offset table might be a
480 * variable sized table containing 8-byte entries
481 * for offsets larger than 2^31.
482 */
483 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
484 unsigned long max_size = min_size;
485 if (nr)
486 max_size += (nr - 1)*8;
487 if (idx_size < min_size || idx_size > max_size) {
488 munmap(idx_map, idx_size);
489 return error("wrong index v2 file size in %s", path);
490 }
491 if (idx_size != min_size &&
492 /*
493 * make sure we can deal with large pack offsets.
494 * 31-bit signed offset won't be enough, neither
495 * 32-bit unsigned one will be.
496 */
497 (sizeof(off_t) <= 4)) {
498 munmap(idx_map, idx_size);
499 return error("pack too large for current definition of off_t in %s", path);
500 }
501 }
502
503 p->index_version = version;
504 p->index_data = idx_map;
505 p->index_size = idx_size;
506 p->num_objects = nr;
507 return 0;
508 }
509
510 int open_pack_index(struct packed_git *p)
511 {
512 char *idx_name;
513 int ret;
514
515 if (p->index_data)
516 return 0;
517
518 idx_name = xstrdup(p->pack_name);
519 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
520 ret = check_packed_git_idx(idx_name, p);
521 free(idx_name);
522 return ret;
523 }
524
525 static void scan_windows(struct packed_git *p,
526 struct packed_git **lru_p,
527 struct pack_window **lru_w,
528 struct pack_window **lru_l)
529 {
530 struct pack_window *w, *w_l;
531
532 for (w_l = NULL, w = p->windows; w; w = w->next) {
533 if (!w->inuse_cnt) {
534 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
535 *lru_p = p;
536 *lru_w = w;
537 *lru_l = w_l;
538 }
539 }
540 w_l = w;
541 }
542 }
543
544 static int unuse_one_window(struct packed_git *current, int keep_fd)
545 {
546 struct packed_git *p, *lru_p = NULL;
547 struct pack_window *lru_w = NULL, *lru_l = NULL;
548
549 if (current)
550 scan_windows(current, &lru_p, &lru_w, &lru_l);
551 for (p = packed_git; p; p = p->next)
552 scan_windows(p, &lru_p, &lru_w, &lru_l);
553 if (lru_p) {
554 munmap(lru_w->base, lru_w->len);
555 pack_mapped -= lru_w->len;
556 if (lru_l)
557 lru_l->next = lru_w->next;
558 else {
559 lru_p->windows = lru_w->next;
560 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
561 close(lru_p->pack_fd);
562 lru_p->pack_fd = -1;
563 }
564 }
565 free(lru_w);
566 pack_open_windows--;
567 return 1;
568 }
569 return 0;
570 }
571
572 void release_pack_memory(size_t need, int fd)
573 {
574 size_t cur = pack_mapped;
575 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
576 ; /* nothing */
577 }
578
579 void close_pack_windows(struct packed_git *p)
580 {
581 while (p->windows) {
582 struct pack_window *w = p->windows;
583
584 if (w->inuse_cnt)
585 die("pack '%s' still has open windows to it",
586 p->pack_name);
587 munmap(w->base, w->len);
588 pack_mapped -= w->len;
589 pack_open_windows--;
590 p->windows = w->next;
591 free(w);
592 }
593 }
594
595 void unuse_pack(struct pack_window **w_cursor)
596 {
597 struct pack_window *w = *w_cursor;
598 if (w) {
599 w->inuse_cnt--;
600 *w_cursor = NULL;
601 }
602 }
603
604 void close_pack_index(struct packed_git *p)
605 {
606 if (p->index_data) {
607 munmap((void *)p->index_data, p->index_size);
608 p->index_data = NULL;
609 }
610 }
611
612 /*
613 * This is used by git-repack in case a newly created pack happens to
614 * contain the same set of objects as an existing one. In that case
615 * the resulting file might be different even if its name would be the
616 * same. It is best to close any reference to the old pack before it is
617 * replaced on disk. Of course no index pointers nor windows for given pack
618 * must subsist at this point. If ever objects from this pack are requested
619 * again, the new version of the pack will be reinitialized through
620 * reprepare_packed_git().
621 */
622 void free_pack_by_name(const char *pack_name)
623 {
624 struct packed_git *p, **pp = &packed_git;
625
626 while (*pp) {
627 p = *pp;
628 if (strcmp(pack_name, p->pack_name) == 0) {
629 clear_delta_base_cache();
630 close_pack_windows(p);
631 if (p->pack_fd != -1)
632 close(p->pack_fd);
633 close_pack_index(p);
634 free(p->bad_object_sha1);
635 *pp = p->next;
636 free(p);
637 return;
638 }
639 pp = &p->next;
640 }
641 }
642
643 /*
644 * Do not call this directly as this leaks p->pack_fd on error return;
645 * call open_packed_git() instead.
646 */
647 static int open_packed_git_1(struct packed_git *p)
648 {
649 struct stat st;
650 struct pack_header hdr;
651 unsigned char sha1[20];
652 unsigned char *idx_sha1;
653 long fd_flag;
654
655 if (!p->index_data && open_pack_index(p))
656 return error("packfile %s index unavailable", p->pack_name);
657
658 p->pack_fd = open(p->pack_name, O_RDONLY);
659 while (p->pack_fd < 0 && errno == EMFILE && unuse_one_window(p, -1))
660 p->pack_fd = open(p->pack_name, O_RDONLY);
661 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
662 return -1;
663
664 /* If we created the struct before we had the pack we lack size. */
665 if (!p->pack_size) {
666 if (!S_ISREG(st.st_mode))
667 return error("packfile %s not a regular file", p->pack_name);
668 p->pack_size = st.st_size;
669 } else if (p->pack_size != st.st_size)
670 return error("packfile %s size changed", p->pack_name);
671
672 /* We leave these file descriptors open with sliding mmap;
673 * there is no point keeping them open across exec(), though.
674 */
675 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
676 if (fd_flag < 0)
677 return error("cannot determine file descriptor flags");
678 fd_flag |= FD_CLOEXEC;
679 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
680 return error("cannot set FD_CLOEXEC");
681
682 /* Verify we recognize this pack file format. */
683 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
684 return error("file %s is far too short to be a packfile", p->pack_name);
685 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
686 return error("file %s is not a GIT packfile", p->pack_name);
687 if (!pack_version_ok(hdr.hdr_version))
688 return error("packfile %s is version %"PRIu32" and not"
689 " supported (try upgrading GIT to a newer version)",
690 p->pack_name, ntohl(hdr.hdr_version));
691
692 /* Verify the pack matches its index. */
693 if (p->num_objects != ntohl(hdr.hdr_entries))
694 return error("packfile %s claims to have %"PRIu32" objects"
695 " while index indicates %"PRIu32" objects",
696 p->pack_name, ntohl(hdr.hdr_entries),
697 p->num_objects);
698 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
699 return error("end of packfile %s is unavailable", p->pack_name);
700 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
701 return error("packfile %s signature is unavailable", p->pack_name);
702 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
703 if (hashcmp(sha1, idx_sha1))
704 return error("packfile %s does not match index", p->pack_name);
705 return 0;
706 }
707
708 static int open_packed_git(struct packed_git *p)
709 {
710 if (!open_packed_git_1(p))
711 return 0;
712 if (p->pack_fd != -1) {
713 close(p->pack_fd);
714 p->pack_fd = -1;
715 }
716 return -1;
717 }
718
719 static int in_window(struct pack_window *win, off_t offset)
720 {
721 /* We must promise at least 20 bytes (one hash) after the
722 * offset is available from this window, otherwise the offset
723 * is not actually in this window and a different window (which
724 * has that one hash excess) must be used. This is to support
725 * the object header and delta base parsing routines below.
726 */
727 off_t win_off = win->offset;
728 return win_off <= offset
729 && (offset + 20) <= (win_off + win->len);
730 }
731
732 unsigned char *use_pack(struct packed_git *p,
733 struct pack_window **w_cursor,
734 off_t offset,
735 unsigned int *left)
736 {
737 struct pack_window *win = *w_cursor;
738
739 if (p->pack_fd == -1 && open_packed_git(p))
740 die("packfile %s cannot be accessed", p->pack_name);
741
742 /* Since packfiles end in a hash of their content and it's
743 * pointless to ask for an offset into the middle of that
744 * hash, and the in_window function above wouldn't match
745 * don't allow an offset too close to the end of the file.
746 */
747 if (offset > (p->pack_size - 20))
748 die("offset beyond end of packfile (truncated pack?)");
749
750 if (!win || !in_window(win, offset)) {
751 if (win)
752 win->inuse_cnt--;
753 for (win = p->windows; win; win = win->next) {
754 if (in_window(win, offset))
755 break;
756 }
757 if (!win) {
758 size_t window_align = packed_git_window_size / 2;
759 off_t len;
760 win = xcalloc(1, sizeof(*win));
761 win->offset = (offset / window_align) * window_align;
762 len = p->pack_size - win->offset;
763 if (len > packed_git_window_size)
764 len = packed_git_window_size;
765 win->len = (size_t)len;
766 pack_mapped += win->len;
767 while (packed_git_limit < pack_mapped
768 && unuse_one_window(p, p->pack_fd))
769 ; /* nothing */
770 win->base = xmmap(NULL, win->len,
771 PROT_READ, MAP_PRIVATE,
772 p->pack_fd, win->offset);
773 if (win->base == MAP_FAILED)
774 die("packfile %s cannot be mapped: %s",
775 p->pack_name,
776 strerror(errno));
777 pack_mmap_calls++;
778 pack_open_windows++;
779 if (pack_mapped > peak_pack_mapped)
780 peak_pack_mapped = pack_mapped;
781 if (pack_open_windows > peak_pack_open_windows)
782 peak_pack_open_windows = pack_open_windows;
783 win->next = p->windows;
784 p->windows = win;
785 }
786 }
787 if (win != *w_cursor) {
788 win->last_used = pack_used_ctr++;
789 win->inuse_cnt++;
790 *w_cursor = win;
791 }
792 offset -= win->offset;
793 if (left)
794 *left = win->len - xsize_t(offset);
795 return win->base + offset;
796 }
797
798 static struct packed_git *alloc_packed_git(int extra)
799 {
800 struct packed_git *p = xmalloc(sizeof(*p) + extra);
801 memset(p, 0, sizeof(*p));
802 p->pack_fd = -1;
803 return p;
804 }
805
806 struct packed_git *add_packed_git(const char *path, int path_len, int local)
807 {
808 struct stat st;
809 struct packed_git *p = alloc_packed_git(path_len + 2);
810
811 /*
812 * Make sure a corresponding .pack file exists and that
813 * the index looks sane.
814 */
815 path_len -= strlen(".idx");
816 if (path_len < 1) {
817 free(p);
818 return NULL;
819 }
820 memcpy(p->pack_name, path, path_len);
821
822 strcpy(p->pack_name + path_len, ".keep");
823 if (!access(p->pack_name, F_OK))
824 p->pack_keep = 1;
825
826 strcpy(p->pack_name + path_len, ".pack");
827 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
828 free(p);
829 return NULL;
830 }
831
832 /* ok, it looks sane as far as we can check without
833 * actually mapping the pack file.
834 */
835 p->pack_size = st.st_size;
836 p->pack_local = local;
837 p->mtime = st.st_mtime;
838 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
839 hashclr(p->sha1);
840 return p;
841 }
842
843 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
844 {
845 const char *path = sha1_pack_name(sha1);
846 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
847
848 strcpy(p->pack_name, path);
849 hashcpy(p->sha1, sha1);
850 if (check_packed_git_idx(idx_path, p)) {
851 free(p);
852 return NULL;
853 }
854
855 return p;
856 }
857
858 void install_packed_git(struct packed_git *pack)
859 {
860 pack->next = packed_git;
861 packed_git = pack;
862 }
863
864 static void prepare_packed_git_one(char *objdir, int local)
865 {
866 /* Ensure that this buffer is large enough so that we can
867 append "/pack/" without clobbering the stack even if
868 strlen(objdir) were PATH_MAX. */
869 char path[PATH_MAX + 1 + 4 + 1 + 1];
870 int len;
871 DIR *dir;
872 struct dirent *de;
873
874 sprintf(path, "%s/pack", objdir);
875 len = strlen(path);
876 dir = opendir(path);
877 while (!dir && errno == EMFILE && unuse_one_window(packed_git, -1))
878 dir = opendir(path);
879 if (!dir) {
880 if (errno != ENOENT)
881 error("unable to open object pack directory: %s: %s",
882 path, strerror(errno));
883 return;
884 }
885 path[len++] = '/';
886 while ((de = readdir(dir)) != NULL) {
887 int namelen = strlen(de->d_name);
888 struct packed_git *p;
889
890 if (!has_extension(de->d_name, ".idx"))
891 continue;
892
893 if (len + namelen + 1 > sizeof(path))
894 continue;
895
896 /* Don't reopen a pack we already have. */
897 strcpy(path + len, de->d_name);
898 for (p = packed_git; p; p = p->next) {
899 if (!memcmp(path, p->pack_name, len + namelen - 4))
900 break;
901 }
902 if (p)
903 continue;
904 /* See if it really is a valid .idx file with corresponding
905 * .pack file that we can map.
906 */
907 p = add_packed_git(path, len + namelen, local);
908 if (!p)
909 continue;
910 install_packed_git(p);
911 }
912 closedir(dir);
913 }
914
915 static int sort_pack(const void *a_, const void *b_)
916 {
917 struct packed_git *a = *((struct packed_git **)a_);
918 struct packed_git *b = *((struct packed_git **)b_);
919 int st;
920
921 /*
922 * Local packs tend to contain objects specific to our
923 * variant of the project than remote ones. In addition,
924 * remote ones could be on a network mounted filesystem.
925 * Favor local ones for these reasons.
926 */
927 st = a->pack_local - b->pack_local;
928 if (st)
929 return -st;
930
931 /*
932 * Younger packs tend to contain more recent objects,
933 * and more recent objects tend to get accessed more
934 * often.
935 */
936 if (a->mtime < b->mtime)
937 return 1;
938 else if (a->mtime == b->mtime)
939 return 0;
940 return -1;
941 }
942
943 static void rearrange_packed_git(void)
944 {
945 struct packed_git **ary, *p;
946 int i, n;
947
948 for (n = 0, p = packed_git; p; p = p->next)
949 n++;
950 if (n < 2)
951 return;
952
953 /* prepare an array of packed_git for easier sorting */
954 ary = xcalloc(n, sizeof(struct packed_git *));
955 for (n = 0, p = packed_git; p; p = p->next)
956 ary[n++] = p;
957
958 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
959
960 /* link them back again */
961 for (i = 0; i < n - 1; i++)
962 ary[i]->next = ary[i + 1];
963 ary[n - 1]->next = NULL;
964 packed_git = ary[0];
965
966 free(ary);
967 }
968
969 static int prepare_packed_git_run_once = 0;
970 void prepare_packed_git(void)
971 {
972 struct alternate_object_database *alt;
973
974 if (prepare_packed_git_run_once)
975 return;
976 prepare_packed_git_one(get_object_directory(), 1);
977 prepare_alt_odb();
978 for (alt = alt_odb_list; alt; alt = alt->next) {
979 alt->name[-1] = 0;
980 prepare_packed_git_one(alt->base, 0);
981 alt->name[-1] = '/';
982 }
983 rearrange_packed_git();
984 prepare_packed_git_run_once = 1;
985 }
986
987 void reprepare_packed_git(void)
988 {
989 discard_revindex();
990 prepare_packed_git_run_once = 0;
991 prepare_packed_git();
992 }
993
994 static void mark_bad_packed_object(struct packed_git *p,
995 const unsigned char *sha1)
996 {
997 unsigned i;
998 for (i = 0; i < p->num_bad_objects; i++)
999 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1000 return;
1001 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1002 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1003 p->num_bad_objects++;
1004 }
1005
1006 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1007 {
1008 struct packed_git *p;
1009 unsigned i;
1010
1011 for (p = packed_git; p; p = p->next)
1012 for (i = 0; i < p->num_bad_objects; i++)
1013 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1014 return p;
1015 return NULL;
1016 }
1017
1018 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1019 {
1020 unsigned char real_sha1[20];
1021 hash_sha1_file(map, size, type, real_sha1);
1022 return hashcmp(sha1, real_sha1) ? -1 : 0;
1023 }
1024
1025 static int git_open_noatime(const char *name)
1026 {
1027 static int sha1_file_open_flag = O_NOATIME;
1028 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1029
1030 /* Might the failure be due to O_NOATIME? */
1031 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1032 fd = open(name, O_RDONLY);
1033 if (fd >= 0)
1034 sha1_file_open_flag = 0;
1035 }
1036 return fd;
1037 }
1038
1039 static int open_sha1_file(const unsigned char *sha1)
1040 {
1041 int fd;
1042 char *name = sha1_file_name(sha1);
1043 struct alternate_object_database *alt;
1044
1045 fd = git_open_noatime(name);
1046 if (fd >= 0)
1047 return fd;
1048
1049 prepare_alt_odb();
1050 errno = ENOENT;
1051 for (alt = alt_odb_list; alt; alt = alt->next) {
1052 name = alt->name;
1053 fill_sha1_path(name, sha1);
1054 fd = git_open_noatime(alt->base);
1055 if (fd >= 0)
1056 return fd;
1057 }
1058 return -1;
1059 }
1060
1061 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1062 {
1063 void *map;
1064 int fd;
1065
1066 fd = open_sha1_file(sha1);
1067 map = NULL;
1068 if (fd >= 0) {
1069 struct stat st;
1070
1071 if (!fstat(fd, &st)) {
1072 *size = xsize_t(st.st_size);
1073 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1074 }
1075 close(fd);
1076 }
1077 return map;
1078 }
1079
1080 static int legacy_loose_object(unsigned char *map)
1081 {
1082 unsigned int word;
1083
1084 /*
1085 * Is it a zlib-compressed buffer? If so, the first byte
1086 * must be 0x78 (15-bit window size, deflated), and the
1087 * first 16-bit word is evenly divisible by 31
1088 */
1089 word = (map[0] << 8) + map[1];
1090 if (map[0] == 0x78 && !(word % 31))
1091 return 1;
1092 else
1093 return 0;
1094 }
1095
1096 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1097 unsigned long len, enum object_type *type, unsigned long *sizep)
1098 {
1099 unsigned shift;
1100 unsigned long size, c;
1101 unsigned long used = 0;
1102
1103 c = buf[used++];
1104 *type = (c >> 4) & 7;
1105 size = c & 15;
1106 shift = 4;
1107 while (c & 0x80) {
1108 if (len <= used || bitsizeof(long) <= shift) {
1109 error("bad object header");
1110 return 0;
1111 }
1112 c = buf[used++];
1113 size += (c & 0x7f) << shift;
1114 shift += 7;
1115 }
1116 *sizep = size;
1117 return used;
1118 }
1119
1120 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1121 {
1122 unsigned long size, used;
1123 static const char valid_loose_object_type[8] = {
1124 0, /* OBJ_EXT */
1125 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1126 0, /* "delta" and others are invalid in a loose object */
1127 };
1128 enum object_type type;
1129
1130 /* Get the data stream */
1131 memset(stream, 0, sizeof(*stream));
1132 stream->next_in = map;
1133 stream->avail_in = mapsize;
1134 stream->next_out = buffer;
1135 stream->avail_out = bufsiz;
1136
1137 if (legacy_loose_object(map)) {
1138 git_inflate_init(stream);
1139 return git_inflate(stream, 0);
1140 }
1141
1142
1143 /*
1144 * There used to be a second loose object header format which
1145 * was meant to mimic the in-pack format, allowing for direct
1146 * copy of the object data. This format turned up not to be
1147 * really worth it and we don't write it any longer. But we
1148 * can still read it.
1149 */
1150 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1151 if (!used || !valid_loose_object_type[type])
1152 return -1;
1153 map += used;
1154 mapsize -= used;
1155
1156 /* Set up the stream for the rest.. */
1157 stream->next_in = map;
1158 stream->avail_in = mapsize;
1159 git_inflate_init(stream);
1160
1161 /* And generate the fake traditional header */
1162 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1163 typename(type), size);
1164 return 0;
1165 }
1166
1167 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1168 {
1169 int bytes = strlen(buffer) + 1;
1170 unsigned char *buf = xmallocz(size);
1171 unsigned long n;
1172 int status = Z_OK;
1173
1174 n = stream->total_out - bytes;
1175 if (n > size)
1176 n = size;
1177 memcpy(buf, (char *) buffer + bytes, n);
1178 bytes = n;
1179 if (bytes <= size) {
1180 /*
1181 * The above condition must be (bytes <= size), not
1182 * (bytes < size). In other words, even though we
1183 * expect no more output and set avail_out to zer0,
1184 * the input zlib stream may have bytes that express
1185 * "this concludes the stream", and we *do* want to
1186 * eat that input.
1187 *
1188 * Otherwise we would not be able to test that we
1189 * consumed all the input to reach the expected size;
1190 * we also want to check that zlib tells us that all
1191 * went well with status == Z_STREAM_END at the end.
1192 */
1193 stream->next_out = buf + bytes;
1194 stream->avail_out = size - bytes;
1195 while (status == Z_OK)
1196 status = git_inflate(stream, Z_FINISH);
1197 }
1198 if (status == Z_STREAM_END && !stream->avail_in) {
1199 git_inflate_end(stream);
1200 return buf;
1201 }
1202
1203 if (status < 0)
1204 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1205 else if (stream->avail_in)
1206 error("garbage at end of loose object '%s'",
1207 sha1_to_hex(sha1));
1208 free(buf);
1209 return NULL;
1210 }
1211
1212 /*
1213 * We used to just use "sscanf()", but that's actually way
1214 * too permissive for what we want to check. So do an anal
1215 * object header parse by hand.
1216 */
1217 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1218 {
1219 char type[10];
1220 int i;
1221 unsigned long size;
1222
1223 /*
1224 * The type can be at most ten bytes (including the
1225 * terminating '\0' that we add), and is followed by
1226 * a space.
1227 */
1228 i = 0;
1229 for (;;) {
1230 char c = *hdr++;
1231 if (c == ' ')
1232 break;
1233 type[i++] = c;
1234 if (i >= sizeof(type))
1235 return -1;
1236 }
1237 type[i] = 0;
1238
1239 /*
1240 * The length must follow immediately, and be in canonical
1241 * decimal format (ie "010" is not valid).
1242 */
1243 size = *hdr++ - '0';
1244 if (size > 9)
1245 return -1;
1246 if (size) {
1247 for (;;) {
1248 unsigned long c = *hdr - '0';
1249 if (c > 9)
1250 break;
1251 hdr++;
1252 size = size * 10 + c;
1253 }
1254 }
1255 *sizep = size;
1256
1257 /*
1258 * The length must be followed by a zero byte
1259 */
1260 return *hdr ? -1 : type_from_string(type);
1261 }
1262
1263 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1264 {
1265 int ret;
1266 z_stream stream;
1267 char hdr[8192];
1268
1269 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1270 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1271 return NULL;
1272
1273 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1274 }
1275
1276 unsigned long get_size_from_delta(struct packed_git *p,
1277 struct pack_window **w_curs,
1278 off_t curpos)
1279 {
1280 const unsigned char *data;
1281 unsigned char delta_head[20], *in;
1282 z_stream stream;
1283 int st;
1284
1285 memset(&stream, 0, sizeof(stream));
1286 stream.next_out = delta_head;
1287 stream.avail_out = sizeof(delta_head);
1288
1289 git_inflate_init(&stream);
1290 do {
1291 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1292 stream.next_in = in;
1293 st = git_inflate(&stream, Z_FINISH);
1294 curpos += stream.next_in - in;
1295 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1296 stream.total_out < sizeof(delta_head));
1297 git_inflate_end(&stream);
1298 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1299 error("delta data unpack-initial failed");
1300 return 0;
1301 }
1302
1303 /* Examine the initial part of the delta to figure out
1304 * the result size.
1305 */
1306 data = delta_head;
1307
1308 /* ignore base size */
1309 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1310
1311 /* Read the result size */
1312 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1313 }
1314
1315 static off_t get_delta_base(struct packed_git *p,
1316 struct pack_window **w_curs,
1317 off_t *curpos,
1318 enum object_type type,
1319 off_t delta_obj_offset)
1320 {
1321 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1322 off_t base_offset;
1323
1324 /* use_pack() assured us we have [base_info, base_info + 20)
1325 * as a range that we can look at without walking off the
1326 * end of the mapped window. Its actually the hash size
1327 * that is assured. An OFS_DELTA longer than the hash size
1328 * is stupid, as then a REF_DELTA would be smaller to store.
1329 */
1330 if (type == OBJ_OFS_DELTA) {
1331 unsigned used = 0;
1332 unsigned char c = base_info[used++];
1333 base_offset = c & 127;
1334 while (c & 128) {
1335 base_offset += 1;
1336 if (!base_offset || MSB(base_offset, 7))
1337 return 0; /* overflow */
1338 c = base_info[used++];
1339 base_offset = (base_offset << 7) + (c & 127);
1340 }
1341 base_offset = delta_obj_offset - base_offset;
1342 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1343 return 0; /* out of bound */
1344 *curpos += used;
1345 } else if (type == OBJ_REF_DELTA) {
1346 /* The base entry _must_ be in the same pack */
1347 base_offset = find_pack_entry_one(base_info, p);
1348 *curpos += 20;
1349 } else
1350 die("I am totally screwed");
1351 return base_offset;
1352 }
1353
1354 /* forward declaration for a mutually recursive function */
1355 static int packed_object_info(struct packed_git *p, off_t offset,
1356 unsigned long *sizep);
1357
1358 static int packed_delta_info(struct packed_git *p,
1359 struct pack_window **w_curs,
1360 off_t curpos,
1361 enum object_type type,
1362 off_t obj_offset,
1363 unsigned long *sizep)
1364 {
1365 off_t base_offset;
1366
1367 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1368 if (!base_offset)
1369 return OBJ_BAD;
1370 type = packed_object_info(p, base_offset, NULL);
1371 if (type <= OBJ_NONE) {
1372 struct revindex_entry *revidx;
1373 const unsigned char *base_sha1;
1374 revidx = find_pack_revindex(p, base_offset);
1375 if (!revidx)
1376 return OBJ_BAD;
1377 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1378 mark_bad_packed_object(p, base_sha1);
1379 type = sha1_object_info(base_sha1, NULL);
1380 if (type <= OBJ_NONE)
1381 return OBJ_BAD;
1382 }
1383
1384 /* We choose to only get the type of the base object and
1385 * ignore potentially corrupt pack file that expects the delta
1386 * based on a base with a wrong size. This saves tons of
1387 * inflate() calls.
1388 */
1389 if (sizep) {
1390 *sizep = get_size_from_delta(p, w_curs, curpos);
1391 if (*sizep == 0)
1392 type = OBJ_BAD;
1393 }
1394
1395 return type;
1396 }
1397
1398 static int unpack_object_header(struct packed_git *p,
1399 struct pack_window **w_curs,
1400 off_t *curpos,
1401 unsigned long *sizep)
1402 {
1403 unsigned char *base;
1404 unsigned int left;
1405 unsigned long used;
1406 enum object_type type;
1407
1408 /* use_pack() assures us we have [base, base + 20) available
1409 * as a range that we can look at at. (Its actually the hash
1410 * size that is assured.) With our object header encoding
1411 * the maximum deflated object size is 2^137, which is just
1412 * insane, so we know won't exceed what we have been given.
1413 */
1414 base = use_pack(p, w_curs, *curpos, &left);
1415 used = unpack_object_header_buffer(base, left, &type, sizep);
1416 if (!used) {
1417 type = OBJ_BAD;
1418 } else
1419 *curpos += used;
1420
1421 return type;
1422 }
1423
1424 const char *packed_object_info_detail(struct packed_git *p,
1425 off_t obj_offset,
1426 unsigned long *size,
1427 unsigned long *store_size,
1428 unsigned int *delta_chain_length,
1429 unsigned char *base_sha1)
1430 {
1431 struct pack_window *w_curs = NULL;
1432 off_t curpos;
1433 unsigned long dummy;
1434 unsigned char *next_sha1;
1435 enum object_type type;
1436 struct revindex_entry *revidx;
1437
1438 *delta_chain_length = 0;
1439 curpos = obj_offset;
1440 type = unpack_object_header(p, &w_curs, &curpos, size);
1441
1442 revidx = find_pack_revindex(p, obj_offset);
1443 *store_size = revidx[1].offset - obj_offset;
1444
1445 for (;;) {
1446 switch (type) {
1447 default:
1448 die("pack %s contains unknown object type %d",
1449 p->pack_name, type);
1450 case OBJ_COMMIT:
1451 case OBJ_TREE:
1452 case OBJ_BLOB:
1453 case OBJ_TAG:
1454 unuse_pack(&w_curs);
1455 return typename(type);
1456 case OBJ_OFS_DELTA:
1457 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1458 if (!obj_offset)
1459 die("pack %s contains bad delta base reference of type %s",
1460 p->pack_name, typename(type));
1461 if (*delta_chain_length == 0) {
1462 revidx = find_pack_revindex(p, obj_offset);
1463 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1464 }
1465 break;
1466 case OBJ_REF_DELTA:
1467 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1468 if (*delta_chain_length == 0)
1469 hashcpy(base_sha1, next_sha1);
1470 obj_offset = find_pack_entry_one(next_sha1, p);
1471 break;
1472 }
1473 (*delta_chain_length)++;
1474 curpos = obj_offset;
1475 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1476 }
1477 }
1478
1479 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1480 unsigned long *sizep)
1481 {
1482 struct pack_window *w_curs = NULL;
1483 unsigned long size;
1484 off_t curpos = obj_offset;
1485 enum object_type type;
1486
1487 type = unpack_object_header(p, &w_curs, &curpos, &size);
1488
1489 switch (type) {
1490 case OBJ_OFS_DELTA:
1491 case OBJ_REF_DELTA:
1492 type = packed_delta_info(p, &w_curs, curpos,
1493 type, obj_offset, sizep);
1494 break;
1495 case OBJ_COMMIT:
1496 case OBJ_TREE:
1497 case OBJ_BLOB:
1498 case OBJ_TAG:
1499 if (sizep)
1500 *sizep = size;
1501 break;
1502 default:
1503 error("unknown object type %i at offset %"PRIuMAX" in %s",
1504 type, (uintmax_t)obj_offset, p->pack_name);
1505 type = OBJ_BAD;
1506 }
1507 unuse_pack(&w_curs);
1508 return type;
1509 }
1510
1511 static void *unpack_compressed_entry(struct packed_git *p,
1512 struct pack_window **w_curs,
1513 off_t curpos,
1514 unsigned long size)
1515 {
1516 int st;
1517 z_stream stream;
1518 unsigned char *buffer, *in;
1519
1520 buffer = xmallocz(size);
1521 memset(&stream, 0, sizeof(stream));
1522 stream.next_out = buffer;
1523 stream.avail_out = size + 1;
1524
1525 git_inflate_init(&stream);
1526 do {
1527 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1528 stream.next_in = in;
1529 st = git_inflate(&stream, Z_FINISH);
1530 if (!stream.avail_out)
1531 break; /* the payload is larger than it should be */
1532 curpos += stream.next_in - in;
1533 } while (st == Z_OK || st == Z_BUF_ERROR);
1534 git_inflate_end(&stream);
1535 if ((st != Z_STREAM_END) || stream.total_out != size) {
1536 free(buffer);
1537 return NULL;
1538 }
1539
1540 return buffer;
1541 }
1542
1543 #define MAX_DELTA_CACHE (256)
1544
1545 static size_t delta_base_cached;
1546
1547 static struct delta_base_cache_lru_list {
1548 struct delta_base_cache_lru_list *prev;
1549 struct delta_base_cache_lru_list *next;
1550 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1551
1552 static struct delta_base_cache_entry {
1553 struct delta_base_cache_lru_list lru;
1554 void *data;
1555 struct packed_git *p;
1556 off_t base_offset;
1557 unsigned long size;
1558 enum object_type type;
1559 } delta_base_cache[MAX_DELTA_CACHE];
1560
1561 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1562 {
1563 unsigned long hash;
1564
1565 hash = (unsigned long)p + (unsigned long)base_offset;
1566 hash += (hash >> 8) + (hash >> 16);
1567 return hash % MAX_DELTA_CACHE;
1568 }
1569
1570 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1571 unsigned long *base_size, enum object_type *type, int keep_cache)
1572 {
1573 void *ret;
1574 unsigned long hash = pack_entry_hash(p, base_offset);
1575 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1576
1577 ret = ent->data;
1578 if (!ret || ent->p != p || ent->base_offset != base_offset)
1579 return unpack_entry(p, base_offset, type, base_size);
1580
1581 if (!keep_cache) {
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 } else {
1587 ret = xmemdupz(ent->data, ent->size);
1588 }
1589 *type = ent->type;
1590 *base_size = ent->size;
1591 return ret;
1592 }
1593
1594 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1595 {
1596 if (ent->data) {
1597 free(ent->data);
1598 ent->data = NULL;
1599 ent->lru.next->prev = ent->lru.prev;
1600 ent->lru.prev->next = ent->lru.next;
1601 delta_base_cached -= ent->size;
1602 }
1603 }
1604
1605 void clear_delta_base_cache(void)
1606 {
1607 unsigned long p;
1608 for (p = 0; p < MAX_DELTA_CACHE; p++)
1609 release_delta_base_cache(&delta_base_cache[p]);
1610 }
1611
1612 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1613 void *base, unsigned long base_size, enum object_type type)
1614 {
1615 unsigned long hash = pack_entry_hash(p, base_offset);
1616 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1617 struct delta_base_cache_lru_list *lru;
1618
1619 release_delta_base_cache(ent);
1620 delta_base_cached += base_size;
1621
1622 for (lru = delta_base_cache_lru.next;
1623 delta_base_cached > delta_base_cache_limit
1624 && lru != &delta_base_cache_lru;
1625 lru = lru->next) {
1626 struct delta_base_cache_entry *f = (void *)lru;
1627 if (f->type == OBJ_BLOB)
1628 release_delta_base_cache(f);
1629 }
1630 for (lru = delta_base_cache_lru.next;
1631 delta_base_cached > delta_base_cache_limit
1632 && lru != &delta_base_cache_lru;
1633 lru = lru->next) {
1634 struct delta_base_cache_entry *f = (void *)lru;
1635 release_delta_base_cache(f);
1636 }
1637
1638 ent->p = p;
1639 ent->base_offset = base_offset;
1640 ent->type = type;
1641 ent->data = base;
1642 ent->size = base_size;
1643 ent->lru.next = &delta_base_cache_lru;
1644 ent->lru.prev = delta_base_cache_lru.prev;
1645 delta_base_cache_lru.prev->next = &ent->lru;
1646 delta_base_cache_lru.prev = &ent->lru;
1647 }
1648
1649 static void *read_object(const unsigned char *sha1, enum object_type *type,
1650 unsigned long *size);
1651
1652 static void *unpack_delta_entry(struct packed_git *p,
1653 struct pack_window **w_curs,
1654 off_t curpos,
1655 unsigned long delta_size,
1656 off_t obj_offset,
1657 enum object_type *type,
1658 unsigned long *sizep)
1659 {
1660 void *delta_data, *result, *base;
1661 unsigned long base_size;
1662 off_t base_offset;
1663
1664 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1665 if (!base_offset) {
1666 error("failed to validate delta base reference "
1667 "at offset %"PRIuMAX" from %s",
1668 (uintmax_t)curpos, p->pack_name);
1669 return NULL;
1670 }
1671 unuse_pack(w_curs);
1672 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1673 if (!base) {
1674 /*
1675 * We're probably in deep shit, but let's try to fetch
1676 * the required base anyway from another pack or loose.
1677 * This is costly but should happen only in the presence
1678 * of a corrupted pack, and is better than failing outright.
1679 */
1680 struct revindex_entry *revidx;
1681 const unsigned char *base_sha1;
1682 revidx = find_pack_revindex(p, base_offset);
1683 if (!revidx)
1684 return NULL;
1685 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1686 error("failed to read delta base object %s"
1687 " at offset %"PRIuMAX" from %s",
1688 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1689 p->pack_name);
1690 mark_bad_packed_object(p, base_sha1);
1691 base = read_object(base_sha1, type, &base_size);
1692 if (!base)
1693 return NULL;
1694 }
1695
1696 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1697 if (!delta_data) {
1698 error("failed to unpack compressed delta "
1699 "at offset %"PRIuMAX" from %s",
1700 (uintmax_t)curpos, p->pack_name);
1701 free(base);
1702 return NULL;
1703 }
1704 result = patch_delta(base, base_size,
1705 delta_data, delta_size,
1706 sizep);
1707 if (!result)
1708 die("failed to apply delta");
1709 free(delta_data);
1710 add_delta_base_cache(p, base_offset, base, base_size, *type);
1711 return result;
1712 }
1713
1714 int do_check_packed_object_crc;
1715
1716 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1717 enum object_type *type, unsigned long *sizep)
1718 {
1719 struct pack_window *w_curs = NULL;
1720 off_t curpos = obj_offset;
1721 void *data;
1722
1723 if (do_check_packed_object_crc && p->index_version > 1) {
1724 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1725 unsigned long len = revidx[1].offset - obj_offset;
1726 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1727 const unsigned char *sha1 =
1728 nth_packed_object_sha1(p, revidx->nr);
1729 error("bad packed object CRC for %s",
1730 sha1_to_hex(sha1));
1731 mark_bad_packed_object(p, sha1);
1732 unuse_pack(&w_curs);
1733 return NULL;
1734 }
1735 }
1736
1737 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1738 switch (*type) {
1739 case OBJ_OFS_DELTA:
1740 case OBJ_REF_DELTA:
1741 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1742 obj_offset, type, sizep);
1743 break;
1744 case OBJ_COMMIT:
1745 case OBJ_TREE:
1746 case OBJ_BLOB:
1747 case OBJ_TAG:
1748 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1749 break;
1750 default:
1751 data = NULL;
1752 error("unknown object type %i at offset %"PRIuMAX" in %s",
1753 *type, (uintmax_t)obj_offset, p->pack_name);
1754 }
1755 unuse_pack(&w_curs);
1756 return data;
1757 }
1758
1759 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1760 uint32_t n)
1761 {
1762 const unsigned char *index = p->index_data;
1763 if (!index) {
1764 if (open_pack_index(p))
1765 return NULL;
1766 index = p->index_data;
1767 }
1768 if (n >= p->num_objects)
1769 return NULL;
1770 index += 4 * 256;
1771 if (p->index_version == 1) {
1772 return index + 24 * n + 4;
1773 } else {
1774 index += 8;
1775 return index + 20 * n;
1776 }
1777 }
1778
1779 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1780 {
1781 const unsigned char *index = p->index_data;
1782 index += 4 * 256;
1783 if (p->index_version == 1) {
1784 return ntohl(*((uint32_t *)(index + 24 * n)));
1785 } else {
1786 uint32_t off;
1787 index += 8 + p->num_objects * (20 + 4);
1788 off = ntohl(*((uint32_t *)(index + 4 * n)));
1789 if (!(off & 0x80000000))
1790 return off;
1791 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1792 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1793 ntohl(*((uint32_t *)(index + 4)));
1794 }
1795 }
1796
1797 off_t find_pack_entry_one(const unsigned char *sha1,
1798 struct packed_git *p)
1799 {
1800 const uint32_t *level1_ofs = p->index_data;
1801 const unsigned char *index = p->index_data;
1802 unsigned hi, lo, stride;
1803 static int use_lookup = -1;
1804 static int debug_lookup = -1;
1805
1806 if (debug_lookup < 0)
1807 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1808
1809 if (!index) {
1810 if (open_pack_index(p))
1811 return 0;
1812 level1_ofs = p->index_data;
1813 index = p->index_data;
1814 }
1815 if (p->index_version > 1) {
1816 level1_ofs += 2;
1817 index += 8;
1818 }
1819 index += 4 * 256;
1820 hi = ntohl(level1_ofs[*sha1]);
1821 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1822 if (p->index_version > 1) {
1823 stride = 20;
1824 } else {
1825 stride = 24;
1826 index += 4;
1827 }
1828
1829 if (debug_lookup)
1830 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1831 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1832
1833 if (use_lookup < 0)
1834 use_lookup = !!getenv("GIT_USE_LOOKUP");
1835 if (use_lookup) {
1836 int pos = sha1_entry_pos(index, stride, 0,
1837 lo, hi, p->num_objects, sha1);
1838 if (pos < 0)
1839 return 0;
1840 return nth_packed_object_offset(p, pos);
1841 }
1842
1843 do {
1844 unsigned mi = (lo + hi) / 2;
1845 int cmp = hashcmp(index + mi * stride, sha1);
1846
1847 if (debug_lookup)
1848 printf("lo %u hi %u rg %u mi %u\n",
1849 lo, hi, hi - lo, mi);
1850 if (!cmp)
1851 return nth_packed_object_offset(p, mi);
1852 if (cmp > 0)
1853 hi = mi;
1854 else
1855 lo = mi+1;
1856 } while (lo < hi);
1857 return 0;
1858 }
1859
1860 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1861 {
1862 static struct packed_git *last_found = (void *)1;
1863 struct packed_git *p;
1864 off_t offset;
1865
1866 prepare_packed_git();
1867 if (!packed_git)
1868 return 0;
1869 p = (last_found == (void *)1) ? packed_git : last_found;
1870
1871 do {
1872 if (p->num_bad_objects) {
1873 unsigned i;
1874 for (i = 0; i < p->num_bad_objects; i++)
1875 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1876 goto next;
1877 }
1878
1879 offset = find_pack_entry_one(sha1, p);
1880 if (offset) {
1881 /*
1882 * We are about to tell the caller where they can
1883 * locate the requested object. We better make
1884 * sure the packfile is still here and can be
1885 * accessed before supplying that answer, as
1886 * it may have been deleted since the index
1887 * was loaded!
1888 */
1889 if (p->pack_fd == -1 && open_packed_git(p)) {
1890 error("packfile %s cannot be accessed", p->pack_name);
1891 goto next;
1892 }
1893 e->offset = offset;
1894 e->p = p;
1895 hashcpy(e->sha1, sha1);
1896 last_found = p;
1897 return 1;
1898 }
1899
1900 next:
1901 if (p == last_found)
1902 p = packed_git;
1903 else
1904 p = p->next;
1905 if (p == last_found)
1906 p = p->next;
1907 } while (p);
1908 return 0;
1909 }
1910
1911 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1912 struct packed_git *packs)
1913 {
1914 struct packed_git *p;
1915
1916 for (p = packs; p; p = p->next) {
1917 if (find_pack_entry_one(sha1, p))
1918 return p;
1919 }
1920 return NULL;
1921
1922 }
1923
1924 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1925 {
1926 int status;
1927 unsigned long mapsize, size;
1928 void *map;
1929 z_stream stream;
1930 char hdr[32];
1931
1932 map = map_sha1_file(sha1, &mapsize);
1933 if (!map)
1934 return error("unable to find %s", sha1_to_hex(sha1));
1935 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1936 status = error("unable to unpack %s header",
1937 sha1_to_hex(sha1));
1938 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1939 status = error("unable to parse %s header", sha1_to_hex(sha1));
1940 else if (sizep)
1941 *sizep = size;
1942 git_inflate_end(&stream);
1943 munmap(map, mapsize);
1944 return status;
1945 }
1946
1947 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1948 {
1949 struct pack_entry e;
1950 int status;
1951
1952 if (!find_pack_entry(sha1, &e)) {
1953 /* Most likely it's a loose object. */
1954 status = sha1_loose_object_info(sha1, sizep);
1955 if (status >= 0)
1956 return status;
1957
1958 /* Not a loose object; someone else may have just packed it. */
1959 reprepare_packed_git();
1960 if (!find_pack_entry(sha1, &e))
1961 return status;
1962 }
1963
1964 status = packed_object_info(e.p, e.offset, sizep);
1965 if (status < 0) {
1966 mark_bad_packed_object(e.p, sha1);
1967 status = sha1_object_info(sha1, sizep);
1968 }
1969
1970 return status;
1971 }
1972
1973 static void *read_packed_sha1(const unsigned char *sha1,
1974 enum object_type *type, unsigned long *size)
1975 {
1976 struct pack_entry e;
1977 void *data;
1978
1979 if (!find_pack_entry(sha1, &e))
1980 return NULL;
1981 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1982 if (!data) {
1983 /*
1984 * We're probably in deep shit, but let's try to fetch
1985 * the required object anyway from another pack or loose.
1986 * This should happen only in the presence of a corrupted
1987 * pack, and is better than failing outright.
1988 */
1989 error("failed to read object %s at offset %"PRIuMAX" from %s",
1990 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
1991 mark_bad_packed_object(e.p, sha1);
1992 data = read_object(sha1, type, size);
1993 }
1994 return data;
1995 }
1996
1997 /*
1998 * This is meant to hold a *small* number of objects that you would
1999 * want read_sha1_file() to be able to return, but yet you do not want
2000 * to write them into the object store (e.g. a browse-only
2001 * application).
2002 */
2003 static struct cached_object {
2004 unsigned char sha1[20];
2005 enum object_type type;
2006 void *buf;
2007 unsigned long size;
2008 } *cached_objects;
2009 static int cached_object_nr, cached_object_alloc;
2010
2011 static struct cached_object empty_tree = {
2012 EMPTY_TREE_SHA1_BIN,
2013 OBJ_TREE,
2014 "",
2015 0
2016 };
2017
2018 static struct cached_object *find_cached_object(const unsigned char *sha1)
2019 {
2020 int i;
2021 struct cached_object *co = cached_objects;
2022
2023 for (i = 0; i < cached_object_nr; i++, co++) {
2024 if (!hashcmp(co->sha1, sha1))
2025 return co;
2026 }
2027 if (!hashcmp(sha1, empty_tree.sha1))
2028 return &empty_tree;
2029 return NULL;
2030 }
2031
2032 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2033 unsigned char *sha1)
2034 {
2035 struct cached_object *co;
2036
2037 hash_sha1_file(buf, len, typename(type), sha1);
2038 if (has_sha1_file(sha1) || find_cached_object(sha1))
2039 return 0;
2040 if (cached_object_alloc <= cached_object_nr) {
2041 cached_object_alloc = alloc_nr(cached_object_alloc);
2042 cached_objects = xrealloc(cached_objects,
2043 sizeof(*cached_objects) *
2044 cached_object_alloc);
2045 }
2046 co = &cached_objects[cached_object_nr++];
2047 co->size = len;
2048 co->type = type;
2049 co->buf = xmalloc(len);
2050 memcpy(co->buf, buf, len);
2051 hashcpy(co->sha1, sha1);
2052 return 0;
2053 }
2054
2055 static void *read_object(const unsigned char *sha1, enum object_type *type,
2056 unsigned long *size)
2057 {
2058 unsigned long mapsize;
2059 void *map, *buf;
2060 struct cached_object *co;
2061
2062 co = find_cached_object(sha1);
2063 if (co) {
2064 *type = co->type;
2065 *size = co->size;
2066 return xmemdupz(co->buf, co->size);
2067 }
2068
2069 buf = read_packed_sha1(sha1, type, size);
2070 if (buf)
2071 return buf;
2072 map = map_sha1_file(sha1, &mapsize);
2073 if (map) {
2074 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2075 munmap(map, mapsize);
2076 return buf;
2077 }
2078 reprepare_packed_git();
2079 return read_packed_sha1(sha1, type, size);
2080 }
2081
2082 /*
2083 * This function dies on corrupt objects; the callers who want to
2084 * deal with them should arrange to call read_object() and give error
2085 * messages themselves.
2086 */
2087 void *read_sha1_file_repl(const unsigned char *sha1,
2088 enum object_type *type,
2089 unsigned long *size,
2090 const unsigned char **replacement)
2091 {
2092 const unsigned char *repl = lookup_replace_object(sha1);
2093 void *data;
2094 char *path;
2095 const struct packed_git *p;
2096
2097 errno = 0;
2098 data = read_object(repl, type, size);
2099 if (data) {
2100 if (replacement)
2101 *replacement = repl;
2102 return data;
2103 }
2104
2105 if (errno != ENOENT)
2106 die_errno("failed to read object %s", sha1_to_hex(sha1));
2107
2108 /* die if we replaced an object with one that does not exist */
2109 if (repl != sha1)
2110 die("replacement %s not found for %s",
2111 sha1_to_hex(repl), sha1_to_hex(sha1));
2112
2113 if (has_loose_object(repl)) {
2114 path = sha1_file_name(sha1);
2115 die("loose object %s (stored in %s) is corrupt",
2116 sha1_to_hex(repl), path);
2117 }
2118
2119 if ((p = has_packed_and_bad(repl)) != NULL)
2120 die("packed object %s (stored in %s) is corrupt",
2121 sha1_to_hex(repl), p->pack_name);
2122
2123 return NULL;
2124 }
2125
2126 void *read_object_with_reference(const unsigned char *sha1,
2127 const char *required_type_name,
2128 unsigned long *size,
2129 unsigned char *actual_sha1_return)
2130 {
2131 enum object_type type, required_type;
2132 void *buffer;
2133 unsigned long isize;
2134 unsigned char actual_sha1[20];
2135
2136 required_type = type_from_string(required_type_name);
2137 hashcpy(actual_sha1, sha1);
2138 while (1) {
2139 int ref_length = -1;
2140 const char *ref_type = NULL;
2141
2142 buffer = read_sha1_file(actual_sha1, &type, &isize);
2143 if (!buffer)
2144 return NULL;
2145 if (type == required_type) {
2146 *size = isize;
2147 if (actual_sha1_return)
2148 hashcpy(actual_sha1_return, actual_sha1);
2149 return buffer;
2150 }
2151 /* Handle references */
2152 else if (type == OBJ_COMMIT)
2153 ref_type = "tree ";
2154 else if (type == OBJ_TAG)
2155 ref_type = "object ";
2156 else {
2157 free(buffer);
2158 return NULL;
2159 }
2160 ref_length = strlen(ref_type);
2161
2162 if (ref_length + 40 > isize ||
2163 memcmp(buffer, ref_type, ref_length) ||
2164 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2165 free(buffer);
2166 return NULL;
2167 }
2168 free(buffer);
2169 /* Now we have the ID of the referred-to object in
2170 * actual_sha1. Check again. */
2171 }
2172 }
2173
2174 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2175 const char *type, unsigned char *sha1,
2176 char *hdr, int *hdrlen)
2177 {
2178 git_SHA_CTX c;
2179
2180 /* Generate the header */
2181 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2182
2183 /* Sha1.. */
2184 git_SHA1_Init(&c);
2185 git_SHA1_Update(&c, hdr, *hdrlen);
2186 git_SHA1_Update(&c, buf, len);
2187 git_SHA1_Final(sha1, &c);
2188 }
2189
2190 /*
2191 * Move the just written object into its final resting place.
2192 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2193 * "moving" is only a part of what it does, when no patch between
2194 * master to pu changes the call sites of this function.
2195 */
2196 int move_temp_to_file(const char *tmpfile, const char *filename)
2197 {
2198 int ret = 0;
2199
2200 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2201 goto try_rename;
2202 else if (link(tmpfile, filename))
2203 ret = errno;
2204
2205 /*
2206 * Coda hack - coda doesn't like cross-directory links,
2207 * so we fall back to a rename, which will mean that it
2208 * won't be able to check collisions, but that's not a
2209 * big deal.
2210 *
2211 * The same holds for FAT formatted media.
2212 *
2213 * When this succeeds, we just return. We have nothing
2214 * left to unlink.
2215 */
2216 if (ret && ret != EEXIST) {
2217 try_rename:
2218 if (!rename(tmpfile, filename))
2219 goto out;
2220 ret = errno;
2221 }
2222 unlink_or_warn(tmpfile);
2223 if (ret) {
2224 if (ret != EEXIST) {
2225 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2226 }
2227 /* FIXME!!! Collision check here ? */
2228 }
2229
2230 out:
2231 if (adjust_shared_perm(filename))
2232 return error("unable to set permission to '%s'", filename);
2233 return 0;
2234 }
2235
2236 static int write_buffer(int fd, const void *buf, size_t len)
2237 {
2238 if (write_in_full(fd, buf, len) < 0)
2239 return error("file write error (%s)", strerror(errno));
2240 return 0;
2241 }
2242
2243 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2244 unsigned char *sha1)
2245 {
2246 char hdr[32];
2247 int hdrlen;
2248 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2249 return 0;
2250 }
2251
2252 /* Finalize a file on disk, and close it. */
2253 static void close_sha1_file(int fd)
2254 {
2255 if (fsync_object_files)
2256 fsync_or_die(fd, "sha1 file");
2257 if (close(fd) != 0)
2258 die_errno("error when closing sha1 file");
2259 }
2260
2261 /* Size of directory component, including the ending '/' */
2262 static inline int directory_size(const char *filename)
2263 {
2264 const char *s = strrchr(filename, '/');
2265 if (!s)
2266 return 0;
2267 return s - filename + 1;
2268 }
2269
2270 /*
2271 * This creates a temporary file in the same directory as the final
2272 * 'filename'
2273 *
2274 * We want to avoid cross-directory filename renames, because those
2275 * can have problems on various filesystems (FAT, NFS, Coda).
2276 */
2277 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2278 {
2279 int fd, dirlen = directory_size(filename);
2280
2281 if (dirlen + 20 > bufsiz) {
2282 errno = ENAMETOOLONG;
2283 return -1;
2284 }
2285 memcpy(buffer, filename, dirlen);
2286 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2287 fd = git_mkstemp_mode(buffer, 0444);
2288 if (fd < 0 && dirlen && errno == ENOENT) {
2289 /* Make sure the directory exists */
2290 memcpy(buffer, filename, dirlen);
2291 buffer[dirlen-1] = 0;
2292 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2293 return -1;
2294
2295 /* Try again */
2296 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2297 fd = git_mkstemp_mode(buffer, 0444);
2298 }
2299 return fd;
2300 }
2301
2302 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2303 const void *buf, unsigned long len, time_t mtime)
2304 {
2305 int fd, ret;
2306 unsigned char compressed[4096];
2307 z_stream stream;
2308 git_SHA_CTX c;
2309 unsigned char parano_sha1[20];
2310 char *filename;
2311 static char tmpfile[PATH_MAX];
2312
2313 filename = sha1_file_name(sha1);
2314 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2315 while (fd < 0 && errno == EMFILE && unuse_one_window(packed_git, -1))
2316 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2317 if (fd < 0) {
2318 if (errno == EACCES)
2319 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2320 else
2321 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2322 }
2323
2324 /* Set it up */
2325 memset(&stream, 0, sizeof(stream));
2326 deflateInit(&stream, zlib_compression_level);
2327 stream.next_out = compressed;
2328 stream.avail_out = sizeof(compressed);
2329 git_SHA1_Init(&c);
2330
2331 /* First header.. */
2332 stream.next_in = (unsigned char *)hdr;
2333 stream.avail_in = hdrlen;
2334 while (deflate(&stream, 0) == Z_OK)
2335 /* nothing */;
2336 git_SHA1_Update(&c, hdr, hdrlen);
2337
2338 /* Then the data itself.. */
2339 stream.next_in = (void *)buf;
2340 stream.avail_in = len;
2341 do {
2342 unsigned char *in0 = stream.next_in;
2343 ret = deflate(&stream, Z_FINISH);
2344 git_SHA1_Update(&c, in0, stream.next_in - in0);
2345 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2346 die("unable to write sha1 file");
2347 stream.next_out = compressed;
2348 stream.avail_out = sizeof(compressed);
2349 } while (ret == Z_OK);
2350
2351 if (ret != Z_STREAM_END)
2352 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2353 ret = deflateEnd(&stream);
2354 if (ret != Z_OK)
2355 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2356 git_SHA1_Final(parano_sha1, &c);
2357 if (hashcmp(sha1, parano_sha1) != 0)
2358 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2359
2360 close_sha1_file(fd);
2361
2362 if (mtime) {
2363 struct utimbuf utb;
2364 utb.actime = mtime;
2365 utb.modtime = mtime;
2366 if (utime(tmpfile, &utb) < 0)
2367 warning("failed utime() on %s: %s",
2368 tmpfile, strerror(errno));
2369 }
2370
2371 return move_temp_to_file(tmpfile, filename);
2372 }
2373
2374 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2375 {
2376 unsigned char sha1[20];
2377 char hdr[32];
2378 int hdrlen;
2379
2380 /* Normally if we have it in the pack then we do not bother writing
2381 * it out into .git/objects/??/?{38} file.
2382 */
2383 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2384 if (returnsha1)
2385 hashcpy(returnsha1, sha1);
2386 if (has_sha1_file(sha1))
2387 return 0;
2388 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2389 }
2390
2391 int force_object_loose(const unsigned char *sha1, time_t mtime)
2392 {
2393 void *buf;
2394 unsigned long len;
2395 enum object_type type;
2396 char hdr[32];
2397 int hdrlen;
2398 int ret;
2399
2400 if (has_loose_object(sha1))
2401 return 0;
2402 buf = read_packed_sha1(sha1, &type, &len);
2403 if (!buf)
2404 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2405 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2406 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2407 free(buf);
2408
2409 return ret;
2410 }
2411
2412 int has_pack_index(const unsigned char *sha1)
2413 {
2414 struct stat st;
2415 if (stat(sha1_pack_index_name(sha1), &st))
2416 return 0;
2417 return 1;
2418 }
2419
2420 int has_sha1_pack(const unsigned char *sha1)
2421 {
2422 struct pack_entry e;
2423 return find_pack_entry(sha1, &e);
2424 }
2425
2426 int has_sha1_file(const unsigned char *sha1)
2427 {
2428 struct pack_entry e;
2429
2430 if (find_pack_entry(sha1, &e))
2431 return 1;
2432 return has_loose_object(sha1);
2433 }
2434
2435 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2436 int write_object, enum object_type type, const char *path)
2437 {
2438 int ret, re_allocated = 0;
2439
2440 if (!type)
2441 type = OBJ_BLOB;
2442
2443 /*
2444 * Convert blobs to git internal format
2445 */
2446 if ((type == OBJ_BLOB) && path) {
2447 struct strbuf nbuf = STRBUF_INIT;
2448 if (convert_to_git(path, buf, size, &nbuf,
2449 write_object ? safe_crlf : 0)) {
2450 buf = strbuf_detach(&nbuf, &size);
2451 re_allocated = 1;
2452 }
2453 }
2454
2455 if (write_object)
2456 ret = write_sha1_file(buf, size, typename(type), sha1);
2457 else
2458 ret = hash_sha1_file(buf, size, typename(type), sha1);
2459 if (re_allocated)
2460 free(buf);
2461 return ret;
2462 }
2463
2464 #define SMALL_FILE_SIZE (32*1024)
2465
2466 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2467 enum object_type type, const char *path)
2468 {
2469 int ret;
2470 size_t size = xsize_t(st->st_size);
2471
2472 if (!S_ISREG(st->st_mode)) {
2473 struct strbuf sbuf = STRBUF_INIT;
2474 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2475 ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2476 type, path);
2477 else
2478 ret = -1;
2479 strbuf_release(&sbuf);
2480 } else if (!size) {
2481 ret = index_mem(sha1, NULL, size, write_object, type, path);
2482 } else if (size <= SMALL_FILE_SIZE) {
2483 char *buf = xmalloc(size);
2484 if (size == read_in_full(fd, buf, size))
2485 ret = index_mem(sha1, buf, size, write_object, type,
2486 path);
2487 else
2488 ret = error("short read %s", strerror(errno));
2489 free(buf);
2490 } else {
2491 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2492 ret = index_mem(sha1, buf, size, write_object, type, path);
2493 munmap(buf, size);
2494 }
2495 close(fd);
2496 return ret;
2497 }
2498
2499 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2500 {
2501 int fd;
2502 struct strbuf sb = STRBUF_INIT;
2503
2504 switch (st->st_mode & S_IFMT) {
2505 case S_IFREG:
2506 fd = open(path, O_RDONLY);
2507 if (fd < 0)
2508 return error("open(\"%s\"): %s", path,
2509 strerror(errno));
2510 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2511 return error("%s: failed to insert into database",
2512 path);
2513 break;
2514 case S_IFLNK:
2515 if (strbuf_readlink(&sb, path, st->st_size)) {
2516 char *errstr = strerror(errno);
2517 return error("readlink(\"%s\"): %s", path,
2518 errstr);
2519 }
2520 if (!write_object)
2521 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2522 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2523 return error("%s: failed to insert into database",
2524 path);
2525 strbuf_release(&sb);
2526 break;
2527 case S_IFDIR:
2528 return resolve_gitlink_ref(path, "HEAD", sha1);
2529 default:
2530 return error("%s: unsupported file type", path);
2531 }
2532 return 0;
2533 }
2534
2535 int read_pack_header(int fd, struct pack_header *header)
2536 {
2537 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2538 /* "eof before pack header was fully read" */
2539 return PH_ERROR_EOF;
2540
2541 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2542 /* "protocol error (pack signature mismatch detected)" */
2543 return PH_ERROR_PACK_SIGNATURE;
2544 if (!pack_version_ok(header->hdr_version))
2545 /* "protocol error (pack version unsupported)" */
2546 return PH_ERROR_PROTOCOL;
2547 return 0;
2548 }
2549
2550 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2551 {
2552 enum object_type type = sha1_object_info(sha1, NULL);
2553 if (type < 0)
2554 die("%s is not a valid object", sha1_to_hex(sha1));
2555 if (type != expect)
2556 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2557 typename(expect));
2558 }