Merge branch 'tr/unpack-entry-use-after-free-fix'
[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 "string-list.h"
11 #include "delta.h"
12 #include "pack.h"
13 #include "blob.h"
14 #include "commit.h"
15 #include "run-command.h"
16 #include "tag.h"
17 #include "tree.h"
18 #include "tree-walk.h"
19 #include "refs.h"
20 #include "pack-revindex.h"
21 #include "sha1-lookup.h"
22 #include "bulk-checkin.h"
23 #include "streaming.h"
24 #include "dir.h"
25
26 #ifndef O_NOATIME
27 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
28 #define O_NOATIME 01000000
29 #else
30 #define O_NOATIME 0
31 #endif
32 #endif
33
34 #define SZ_FMT PRIuMAX
35 static inline uintmax_t sz_fmt(size_t s) { return s; }
36
37 const unsigned char null_sha1[20];
38
39 /*
40 * This is meant to hold a *small* number of objects that you would
41 * want read_sha1_file() to be able to return, but yet you do not want
42 * to write them into the object store (e.g. a browse-only
43 * application).
44 */
45 static struct cached_object {
46 unsigned char sha1[20];
47 enum object_type type;
48 void *buf;
49 unsigned long size;
50 } *cached_objects;
51 static int cached_object_nr, cached_object_alloc;
52
53 static struct cached_object empty_tree = {
54 EMPTY_TREE_SHA1_BIN_LITERAL,
55 OBJ_TREE,
56 "",
57 0
58 };
59
60 static struct packed_git *last_found_pack;
61
62 static struct cached_object *find_cached_object(const unsigned char *sha1)
63 {
64 int i;
65 struct cached_object *co = cached_objects;
66
67 for (i = 0; i < cached_object_nr; i++, co++) {
68 if (!hashcmp(co->sha1, sha1))
69 return co;
70 }
71 if (!hashcmp(sha1, empty_tree.sha1))
72 return &empty_tree;
73 return NULL;
74 }
75
76 int mkdir_in_gitdir(const char *path)
77 {
78 if (mkdir(path, 0777)) {
79 int saved_errno = errno;
80 struct stat st;
81 struct strbuf sb = STRBUF_INIT;
82
83 if (errno != EEXIST)
84 return -1;
85 /*
86 * Are we looking at a path in a symlinked worktree
87 * whose original repository does not yet have it?
88 * e.g. .git/rr-cache pointing at its original
89 * repository in which the user hasn't performed any
90 * conflict resolution yet?
91 */
92 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
93 strbuf_readlink(&sb, path, st.st_size) ||
94 !is_absolute_path(sb.buf) ||
95 mkdir(sb.buf, 0777)) {
96 strbuf_release(&sb);
97 errno = saved_errno;
98 return -1;
99 }
100 strbuf_release(&sb);
101 }
102 return adjust_shared_perm(path);
103 }
104
105 int safe_create_leading_directories(char *path)
106 {
107 char *pos = path + offset_1st_component(path);
108 struct stat st;
109
110 while (pos) {
111 pos = strchr(pos, '/');
112 if (!pos)
113 break;
114 while (*++pos == '/')
115 ;
116 if (!*pos)
117 break;
118 *--pos = '\0';
119 if (!stat(path, &st)) {
120 /* path exists */
121 if (!S_ISDIR(st.st_mode)) {
122 *pos = '/';
123 return -3;
124 }
125 }
126 else if (mkdir(path, 0777)) {
127 if (errno == EEXIST &&
128 !stat(path, &st) && S_ISDIR(st.st_mode)) {
129 ; /* somebody created it since we checked */
130 } else {
131 *pos = '/';
132 return -1;
133 }
134 }
135 else if (adjust_shared_perm(path)) {
136 *pos = '/';
137 return -2;
138 }
139 *pos++ = '/';
140 }
141 return 0;
142 }
143
144 int safe_create_leading_directories_const(const char *path)
145 {
146 /* path points to cache entries, so xstrdup before messing with it */
147 char *buf = xstrdup(path);
148 int result = safe_create_leading_directories(buf);
149 free(buf);
150 return result;
151 }
152
153 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
154 {
155 int i;
156 for (i = 0; i < 20; i++) {
157 static char hex[] = "0123456789abcdef";
158 unsigned int val = sha1[i];
159 char *pos = pathbuf + i*2 + (i > 0);
160 *pos++ = hex[val >> 4];
161 *pos = hex[val & 0xf];
162 }
163 }
164
165 /*
166 * NOTE! This returns a statically allocated buffer, so you have to be
167 * careful about using it. Do an "xstrdup()" if you need to save the
168 * filename.
169 *
170 * Also note that this returns the location for creating. Reading
171 * SHA1 file can happen from any alternate directory listed in the
172 * DB_ENVIRONMENT environment variable if it is not found in
173 * the primary object database.
174 */
175 char *sha1_file_name(const unsigned char *sha1)
176 {
177 static char buf[PATH_MAX];
178 const char *objdir;
179 int len;
180
181 objdir = get_object_directory();
182 len = strlen(objdir);
183
184 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
185 if (len + 43 > PATH_MAX)
186 die("insanely long object directory %s", objdir);
187 memcpy(buf, objdir, len);
188 buf[len] = '/';
189 buf[len+3] = '/';
190 buf[len+42] = '\0';
191 fill_sha1_path(buf + len + 1, sha1);
192 return buf;
193 }
194
195 static char *sha1_get_pack_name(const unsigned char *sha1,
196 char **name, char **base, const char *which)
197 {
198 static const char hex[] = "0123456789abcdef";
199 char *buf;
200 int i;
201
202 if (!*base) {
203 const char *sha1_file_directory = get_object_directory();
204 int len = strlen(sha1_file_directory);
205 *base = xmalloc(len + 60);
206 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
207 sha1_file_directory, which);
208 *name = *base + len + 11;
209 }
210
211 buf = *name;
212
213 for (i = 0; i < 20; i++) {
214 unsigned int val = *sha1++;
215 *buf++ = hex[val >> 4];
216 *buf++ = hex[val & 0xf];
217 }
218
219 return *base;
220 }
221
222 char *sha1_pack_name(const unsigned char *sha1)
223 {
224 static char *name, *base;
225
226 return sha1_get_pack_name(sha1, &name, &base, "pack");
227 }
228
229 char *sha1_pack_index_name(const unsigned char *sha1)
230 {
231 static char *name, *base;
232
233 return sha1_get_pack_name(sha1, &name, &base, "idx");
234 }
235
236 struct alternate_object_database *alt_odb_list;
237 static struct alternate_object_database **alt_odb_tail;
238
239 static int git_open_noatime(const char *name);
240
241 /*
242 * Prepare alternate object database registry.
243 *
244 * The variable alt_odb_list points at the list of struct
245 * alternate_object_database. The elements on this list come from
246 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
247 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
248 * whose contents is similar to that environment variable but can be
249 * LF separated. Its base points at a statically allocated buffer that
250 * contains "/the/directory/corresponding/to/.git/objects/...", while
251 * its name points just after the slash at the end of ".git/objects/"
252 * in the example above, and has enough space to hold 40-byte hex
253 * SHA1, an extra slash for the first level indirection, and the
254 * terminating NUL.
255 */
256 static int link_alt_odb_entry(const char *entry, const char *relative_base, int depth)
257 {
258 const char *objdir = get_object_directory();
259 struct alternate_object_database *ent;
260 struct alternate_object_database *alt;
261 int pfxlen, entlen;
262 struct strbuf pathbuf = STRBUF_INIT;
263
264 if (!is_absolute_path(entry) && relative_base) {
265 strbuf_addstr(&pathbuf, real_path(relative_base));
266 strbuf_addch(&pathbuf, '/');
267 }
268 strbuf_addstr(&pathbuf, entry);
269
270 normalize_path_copy(pathbuf.buf, pathbuf.buf);
271
272 pfxlen = strlen(pathbuf.buf);
273
274 /*
275 * The trailing slash after the directory name is given by
276 * this function at the end. Remove duplicates.
277 */
278 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
279 pfxlen -= 1;
280
281 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
282 ent = xmalloc(sizeof(*ent) + entlen);
283 memcpy(ent->base, pathbuf.buf, pfxlen);
284 strbuf_release(&pathbuf);
285
286 ent->name = ent->base + pfxlen + 1;
287 ent->base[pfxlen + 3] = '/';
288 ent->base[pfxlen] = ent->base[entlen-1] = 0;
289
290 /* Detect cases where alternate disappeared */
291 if (!is_directory(ent->base)) {
292 error("object directory %s does not exist; "
293 "check .git/objects/info/alternates.",
294 ent->base);
295 free(ent);
296 return -1;
297 }
298
299 /* Prevent the common mistake of listing the same
300 * thing twice, or object directory itself.
301 */
302 for (alt = alt_odb_list; alt; alt = alt->next) {
303 if (!memcmp(ent->base, alt->base, pfxlen)) {
304 free(ent);
305 return -1;
306 }
307 }
308 if (!strcmp(ent->base, objdir)) {
309 free(ent);
310 return -1;
311 }
312
313 /* add the alternate entry */
314 *alt_odb_tail = ent;
315 alt_odb_tail = &(ent->next);
316 ent->next = NULL;
317
318 /* recursively add alternates */
319 read_info_alternates(ent->base, depth + 1);
320
321 ent->base[pfxlen] = '/';
322
323 return 0;
324 }
325
326 static void link_alt_odb_entries(const char *alt, int len, int sep,
327 const char *relative_base, int depth)
328 {
329 struct string_list entries = STRING_LIST_INIT_NODUP;
330 char *alt_copy;
331 int i;
332
333 if (depth > 5) {
334 error("%s: ignoring alternate object stores, nesting too deep.",
335 relative_base);
336 return;
337 }
338
339 alt_copy = xmemdupz(alt, len);
340 string_list_split_in_place(&entries, alt_copy, sep, -1);
341 for (i = 0; i < entries.nr; i++) {
342 const char *entry = entries.items[i].string;
343 if (entry[0] == '\0' || entry[0] == '#')
344 continue;
345 if (!is_absolute_path(entry) && depth) {
346 error("%s: ignoring relative alternate object store %s",
347 relative_base, entry);
348 } else {
349 link_alt_odb_entry(entry, relative_base, depth);
350 }
351 }
352 string_list_clear(&entries, 0);
353 free(alt_copy);
354 }
355
356 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 = git_open_noatime(path);
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, mapsz, '\n', relative_base, depth);
381
382 munmap(map, mapsz);
383 }
384
385 void add_to_alternates_file(const char *reference)
386 {
387 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
388 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
389 char *alt = mkpath("%s\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, strlen(alt), '\n', NULL, 0);
395 }
396
397 void foreach_alt_odb(alt_odb_fn fn, void *cb)
398 {
399 struct alternate_object_database *ent;
400
401 prepare_alt_odb();
402 for (ent = alt_odb_list; ent; ent = ent->next)
403 if (fn(ent, cb))
404 return;
405 }
406
407 void prepare_alt_odb(void)
408 {
409 const char *alt;
410
411 if (alt_odb_tail)
412 return;
413
414 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
415 if (!alt) alt = "";
416
417 alt_odb_tail = &alt_odb_list;
418 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
419
420 read_info_alternates(get_object_directory(), 0);
421 }
422
423 static int has_loose_object_local(const unsigned char *sha1)
424 {
425 char *name = sha1_file_name(sha1);
426 return !access(name, F_OK);
427 }
428
429 int has_loose_object_nonlocal(const unsigned char *sha1)
430 {
431 struct alternate_object_database *alt;
432 prepare_alt_odb();
433 for (alt = alt_odb_list; alt; alt = alt->next) {
434 fill_sha1_path(alt->name, sha1);
435 if (!access(alt->base, F_OK))
436 return 1;
437 }
438 return 0;
439 }
440
441 static int has_loose_object(const unsigned char *sha1)
442 {
443 return has_loose_object_local(sha1) ||
444 has_loose_object_nonlocal(sha1);
445 }
446
447 static unsigned int pack_used_ctr;
448 static unsigned int pack_mmap_calls;
449 static unsigned int peak_pack_open_windows;
450 static unsigned int pack_open_windows;
451 static unsigned int pack_open_fds;
452 static unsigned int pack_max_fds;
453 static size_t peak_pack_mapped;
454 static size_t pack_mapped;
455 struct packed_git *packed_git;
456
457 void pack_report(void)
458 {
459 fprintf(stderr,
460 "pack_report: getpagesize() = %10" SZ_FMT "\n"
461 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
462 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
463 sz_fmt(getpagesize()),
464 sz_fmt(packed_git_window_size),
465 sz_fmt(packed_git_limit));
466 fprintf(stderr,
467 "pack_report: pack_used_ctr = %10u\n"
468 "pack_report: pack_mmap_calls = %10u\n"
469 "pack_report: pack_open_windows = %10u / %10u\n"
470 "pack_report: pack_mapped = "
471 "%10" SZ_FMT " / %10" SZ_FMT "\n",
472 pack_used_ctr,
473 pack_mmap_calls,
474 pack_open_windows, peak_pack_open_windows,
475 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
476 }
477
478 static int check_packed_git_idx(const char *path, struct packed_git *p)
479 {
480 void *idx_map;
481 struct pack_idx_header *hdr;
482 size_t idx_size;
483 uint32_t version, nr, i, *index;
484 int fd = git_open_noatime(path);
485 struct stat st;
486
487 if (fd < 0)
488 return -1;
489 if (fstat(fd, &st)) {
490 close(fd);
491 return -1;
492 }
493 idx_size = xsize_t(st.st_size);
494 if (idx_size < 4 * 256 + 20 + 20) {
495 close(fd);
496 return error("index file %s is too small", path);
497 }
498 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
499 close(fd);
500
501 hdr = idx_map;
502 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
503 version = ntohl(hdr->idx_version);
504 if (version < 2 || version > 2) {
505 munmap(idx_map, idx_size);
506 return error("index file %s is version %"PRIu32
507 " and is not supported by this binary"
508 " (try upgrading GIT to a newer version)",
509 path, version);
510 }
511 } else
512 version = 1;
513
514 nr = 0;
515 index = idx_map;
516 if (version > 1)
517 index += 2; /* skip index header */
518 for (i = 0; i < 256; i++) {
519 uint32_t n = ntohl(index[i]);
520 if (n < nr) {
521 munmap(idx_map, idx_size);
522 return error("non-monotonic index %s", path);
523 }
524 nr = n;
525 }
526
527 if (version == 1) {
528 /*
529 * Total size:
530 * - 256 index entries 4 bytes each
531 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
532 * - 20-byte SHA1 of the packfile
533 * - 20-byte SHA1 file checksum
534 */
535 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
536 munmap(idx_map, idx_size);
537 return error("wrong index v1 file size in %s", path);
538 }
539 } else if (version == 2) {
540 /*
541 * Minimum size:
542 * - 8 bytes of header
543 * - 256 index entries 4 bytes each
544 * - 20-byte sha1 entry * nr
545 * - 4-byte crc entry * nr
546 * - 4-byte offset entry * nr
547 * - 20-byte SHA1 of the packfile
548 * - 20-byte SHA1 file checksum
549 * And after the 4-byte offset table might be a
550 * variable sized table containing 8-byte entries
551 * for offsets larger than 2^31.
552 */
553 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
554 unsigned long max_size = min_size;
555 if (nr)
556 max_size += (nr - 1)*8;
557 if (idx_size < min_size || idx_size > max_size) {
558 munmap(idx_map, idx_size);
559 return error("wrong index v2 file size in %s", path);
560 }
561 if (idx_size != min_size &&
562 /*
563 * make sure we can deal with large pack offsets.
564 * 31-bit signed offset won't be enough, neither
565 * 32-bit unsigned one will be.
566 */
567 (sizeof(off_t) <= 4)) {
568 munmap(idx_map, idx_size);
569 return error("pack too large for current definition of off_t in %s", path);
570 }
571 }
572
573 p->index_version = version;
574 p->index_data = idx_map;
575 p->index_size = idx_size;
576 p->num_objects = nr;
577 return 0;
578 }
579
580 int open_pack_index(struct packed_git *p)
581 {
582 char *idx_name;
583 int ret;
584
585 if (p->index_data)
586 return 0;
587
588 idx_name = xstrdup(p->pack_name);
589 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
590 ret = check_packed_git_idx(idx_name, p);
591 free(idx_name);
592 return ret;
593 }
594
595 static void scan_windows(struct packed_git *p,
596 struct packed_git **lru_p,
597 struct pack_window **lru_w,
598 struct pack_window **lru_l)
599 {
600 struct pack_window *w, *w_l;
601
602 for (w_l = NULL, w = p->windows; w; w = w->next) {
603 if (!w->inuse_cnt) {
604 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
605 *lru_p = p;
606 *lru_w = w;
607 *lru_l = w_l;
608 }
609 }
610 w_l = w;
611 }
612 }
613
614 static int unuse_one_window(struct packed_git *current, int keep_fd)
615 {
616 struct packed_git *p, *lru_p = NULL;
617 struct pack_window *lru_w = NULL, *lru_l = NULL;
618
619 if (current)
620 scan_windows(current, &lru_p, &lru_w, &lru_l);
621 for (p = packed_git; p; p = p->next)
622 scan_windows(p, &lru_p, &lru_w, &lru_l);
623 if (lru_p) {
624 munmap(lru_w->base, lru_w->len);
625 pack_mapped -= lru_w->len;
626 if (lru_l)
627 lru_l->next = lru_w->next;
628 else {
629 lru_p->windows = lru_w->next;
630 if (!lru_p->windows && lru_p->pack_fd != -1
631 && lru_p->pack_fd != keep_fd) {
632 close(lru_p->pack_fd);
633 pack_open_fds--;
634 lru_p->pack_fd = -1;
635 }
636 }
637 free(lru_w);
638 pack_open_windows--;
639 return 1;
640 }
641 return 0;
642 }
643
644 void release_pack_memory(size_t need, int fd)
645 {
646 size_t cur = pack_mapped;
647 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
648 ; /* nothing */
649 }
650
651 void *xmmap(void *start, size_t length,
652 int prot, int flags, int fd, off_t offset)
653 {
654 void *ret = mmap(start, length, prot, flags, fd, offset);
655 if (ret == MAP_FAILED) {
656 if (!length)
657 return NULL;
658 release_pack_memory(length, fd);
659 ret = mmap(start, length, prot, flags, fd, offset);
660 if (ret == MAP_FAILED)
661 die_errno("Out of memory? mmap failed");
662 }
663 return ret;
664 }
665
666 void close_pack_windows(struct packed_git *p)
667 {
668 while (p->windows) {
669 struct pack_window *w = p->windows;
670
671 if (w->inuse_cnt)
672 die("pack '%s' still has open windows to it",
673 p->pack_name);
674 munmap(w->base, w->len);
675 pack_mapped -= w->len;
676 pack_open_windows--;
677 p->windows = w->next;
678 free(w);
679 }
680 }
681
682 void unuse_pack(struct pack_window **w_cursor)
683 {
684 struct pack_window *w = *w_cursor;
685 if (w) {
686 w->inuse_cnt--;
687 *w_cursor = NULL;
688 }
689 }
690
691 void close_pack_index(struct packed_git *p)
692 {
693 if (p->index_data) {
694 munmap((void *)p->index_data, p->index_size);
695 p->index_data = NULL;
696 }
697 }
698
699 /*
700 * This is used by git-repack in case a newly created pack happens to
701 * contain the same set of objects as an existing one. In that case
702 * the resulting file might be different even if its name would be the
703 * same. It is best to close any reference to the old pack before it is
704 * replaced on disk. Of course no index pointers nor windows for given pack
705 * must subsist at this point. If ever objects from this pack are requested
706 * again, the new version of the pack will be reinitialized through
707 * reprepare_packed_git().
708 */
709 void free_pack_by_name(const char *pack_name)
710 {
711 struct packed_git *p, **pp = &packed_git;
712
713 while (*pp) {
714 p = *pp;
715 if (strcmp(pack_name, p->pack_name) == 0) {
716 clear_delta_base_cache();
717 close_pack_windows(p);
718 if (p->pack_fd != -1) {
719 close(p->pack_fd);
720 pack_open_fds--;
721 }
722 close_pack_index(p);
723 free(p->bad_object_sha1);
724 *pp = p->next;
725 if (last_found_pack == p)
726 last_found_pack = NULL;
727 free(p);
728 return;
729 }
730 pp = &p->next;
731 }
732 }
733
734 static unsigned int get_max_fd_limit(void)
735 {
736 #ifdef RLIMIT_NOFILE
737 struct rlimit lim;
738
739 if (getrlimit(RLIMIT_NOFILE, &lim))
740 die_errno("cannot get RLIMIT_NOFILE");
741
742 return lim.rlim_cur;
743 #elif defined(_SC_OPEN_MAX)
744 return sysconf(_SC_OPEN_MAX);
745 #elif defined(OPEN_MAX)
746 return OPEN_MAX;
747 #else
748 return 1; /* see the caller ;-) */
749 #endif
750 }
751
752 /*
753 * Do not call this directly as this leaks p->pack_fd on error return;
754 * call open_packed_git() instead.
755 */
756 static int open_packed_git_1(struct packed_git *p)
757 {
758 struct stat st;
759 struct pack_header hdr;
760 unsigned char sha1[20];
761 unsigned char *idx_sha1;
762 long fd_flag;
763
764 if (!p->index_data && open_pack_index(p))
765 return error("packfile %s index unavailable", p->pack_name);
766
767 if (!pack_max_fds) {
768 unsigned int max_fds = get_max_fd_limit();
769
770 /* Save 3 for stdin/stdout/stderr, 22 for work */
771 if (25 < max_fds)
772 pack_max_fds = max_fds - 25;
773 else
774 pack_max_fds = 1;
775 }
776
777 while (pack_max_fds <= pack_open_fds && unuse_one_window(NULL, -1))
778 ; /* nothing */
779
780 p->pack_fd = git_open_noatime(p->pack_name);
781 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
782 return -1;
783 pack_open_fds++;
784
785 /* If we created the struct before we had the pack we lack size. */
786 if (!p->pack_size) {
787 if (!S_ISREG(st.st_mode))
788 return error("packfile %s not a regular file", p->pack_name);
789 p->pack_size = st.st_size;
790 } else if (p->pack_size != st.st_size)
791 return error("packfile %s size changed", p->pack_name);
792
793 /* We leave these file descriptors open with sliding mmap;
794 * there is no point keeping them open across exec(), though.
795 */
796 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
797 if (fd_flag < 0)
798 return error("cannot determine file descriptor flags");
799 fd_flag |= FD_CLOEXEC;
800 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
801 return error("cannot set FD_CLOEXEC");
802
803 /* Verify we recognize this pack file format. */
804 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
805 return error("file %s is far too short to be a packfile", p->pack_name);
806 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
807 return error("file %s is not a GIT packfile", p->pack_name);
808 if (!pack_version_ok(hdr.hdr_version))
809 return error("packfile %s is version %"PRIu32" and not"
810 " supported (try upgrading GIT to a newer version)",
811 p->pack_name, ntohl(hdr.hdr_version));
812
813 /* Verify the pack matches its index. */
814 if (p->num_objects != ntohl(hdr.hdr_entries))
815 return error("packfile %s claims to have %"PRIu32" objects"
816 " while index indicates %"PRIu32" objects",
817 p->pack_name, ntohl(hdr.hdr_entries),
818 p->num_objects);
819 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
820 return error("end of packfile %s is unavailable", p->pack_name);
821 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
822 return error("packfile %s signature is unavailable", p->pack_name);
823 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
824 if (hashcmp(sha1, idx_sha1))
825 return error("packfile %s does not match index", p->pack_name);
826 return 0;
827 }
828
829 static int open_packed_git(struct packed_git *p)
830 {
831 if (!open_packed_git_1(p))
832 return 0;
833 if (p->pack_fd != -1) {
834 close(p->pack_fd);
835 pack_open_fds--;
836 p->pack_fd = -1;
837 }
838 return -1;
839 }
840
841 static int in_window(struct pack_window *win, off_t offset)
842 {
843 /* We must promise at least 20 bytes (one hash) after the
844 * offset is available from this window, otherwise the offset
845 * is not actually in this window and a different window (which
846 * has that one hash excess) must be used. This is to support
847 * the object header and delta base parsing routines below.
848 */
849 off_t win_off = win->offset;
850 return win_off <= offset
851 && (offset + 20) <= (win_off + win->len);
852 }
853
854 unsigned char *use_pack(struct packed_git *p,
855 struct pack_window **w_cursor,
856 off_t offset,
857 unsigned long *left)
858 {
859 struct pack_window *win = *w_cursor;
860
861 /* Since packfiles end in a hash of their content and it's
862 * pointless to ask for an offset into the middle of that
863 * hash, and the in_window function above wouldn't match
864 * don't allow an offset too close to the end of the file.
865 */
866 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
867 die("packfile %s cannot be accessed", p->pack_name);
868 if (offset > (p->pack_size - 20))
869 die("offset beyond end of packfile (truncated pack?)");
870
871 if (!win || !in_window(win, offset)) {
872 if (win)
873 win->inuse_cnt--;
874 for (win = p->windows; win; win = win->next) {
875 if (in_window(win, offset))
876 break;
877 }
878 if (!win) {
879 size_t window_align = packed_git_window_size / 2;
880 off_t len;
881
882 if (p->pack_fd == -1 && open_packed_git(p))
883 die("packfile %s cannot be accessed", p->pack_name);
884
885 win = xcalloc(1, sizeof(*win));
886 win->offset = (offset / window_align) * window_align;
887 len = p->pack_size - win->offset;
888 if (len > packed_git_window_size)
889 len = packed_git_window_size;
890 win->len = (size_t)len;
891 pack_mapped += win->len;
892 while (packed_git_limit < pack_mapped
893 && unuse_one_window(p, p->pack_fd))
894 ; /* nothing */
895 win->base = xmmap(NULL, win->len,
896 PROT_READ, MAP_PRIVATE,
897 p->pack_fd, win->offset);
898 if (win->base == MAP_FAILED)
899 die("packfile %s cannot be mapped: %s",
900 p->pack_name,
901 strerror(errno));
902 if (!win->offset && win->len == p->pack_size
903 && !p->do_not_close) {
904 close(p->pack_fd);
905 pack_open_fds--;
906 p->pack_fd = -1;
907 }
908 pack_mmap_calls++;
909 pack_open_windows++;
910 if (pack_mapped > peak_pack_mapped)
911 peak_pack_mapped = pack_mapped;
912 if (pack_open_windows > peak_pack_open_windows)
913 peak_pack_open_windows = pack_open_windows;
914 win->next = p->windows;
915 p->windows = win;
916 }
917 }
918 if (win != *w_cursor) {
919 win->last_used = pack_used_ctr++;
920 win->inuse_cnt++;
921 *w_cursor = win;
922 }
923 offset -= win->offset;
924 if (left)
925 *left = win->len - xsize_t(offset);
926 return win->base + offset;
927 }
928
929 static struct packed_git *alloc_packed_git(int extra)
930 {
931 struct packed_git *p = xmalloc(sizeof(*p) + extra);
932 memset(p, 0, sizeof(*p));
933 p->pack_fd = -1;
934 return p;
935 }
936
937 static void try_to_free_pack_memory(size_t size)
938 {
939 release_pack_memory(size, -1);
940 }
941
942 struct packed_git *add_packed_git(const char *path, int path_len, int local)
943 {
944 static int have_set_try_to_free_routine;
945 struct stat st;
946 struct packed_git *p = alloc_packed_git(path_len + 2);
947
948 if (!have_set_try_to_free_routine) {
949 have_set_try_to_free_routine = 1;
950 set_try_to_free_routine(try_to_free_pack_memory);
951 }
952
953 /*
954 * Make sure a corresponding .pack file exists and that
955 * the index looks sane.
956 */
957 path_len -= strlen(".idx");
958 if (path_len < 1) {
959 free(p);
960 return NULL;
961 }
962 memcpy(p->pack_name, path, path_len);
963
964 strcpy(p->pack_name + path_len, ".keep");
965 if (!access(p->pack_name, F_OK))
966 p->pack_keep = 1;
967
968 strcpy(p->pack_name + path_len, ".pack");
969 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
970 free(p);
971 return NULL;
972 }
973
974 /* ok, it looks sane as far as we can check without
975 * actually mapping the pack file.
976 */
977 p->pack_size = st.st_size;
978 p->pack_local = local;
979 p->mtime = st.st_mtime;
980 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
981 hashclr(p->sha1);
982 return p;
983 }
984
985 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
986 {
987 const char *path = sha1_pack_name(sha1);
988 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
989
990 strcpy(p->pack_name, path);
991 hashcpy(p->sha1, sha1);
992 if (check_packed_git_idx(idx_path, p)) {
993 free(p);
994 return NULL;
995 }
996
997 return p;
998 }
999
1000 void install_packed_git(struct packed_git *pack)
1001 {
1002 if (pack->pack_fd != -1)
1003 pack_open_fds++;
1004
1005 pack->next = packed_git;
1006 packed_git = pack;
1007 }
1008
1009 void (*report_garbage)(const char *desc, const char *path);
1010
1011 static void report_helper(const struct string_list *list,
1012 int seen_bits, int first, int last)
1013 {
1014 const char *msg;
1015 switch (seen_bits) {
1016 case 0:
1017 msg = "no corresponding .idx nor .pack";
1018 break;
1019 case 1:
1020 msg = "no corresponding .idx";
1021 break;
1022 case 2:
1023 msg = "no corresponding .pack";
1024 break;
1025 default:
1026 return;
1027 }
1028 for (; first < last; first++)
1029 report_garbage(msg, list->items[first].string);
1030 }
1031
1032 static void report_pack_garbage(struct string_list *list)
1033 {
1034 int i, baselen = -1, first = 0, seen_bits = 0;
1035
1036 if (!report_garbage)
1037 return;
1038
1039 sort_string_list(list);
1040
1041 for (i = 0; i < list->nr; i++) {
1042 const char *path = list->items[i].string;
1043 if (baselen != -1 &&
1044 strncmp(path, list->items[first].string, baselen)) {
1045 report_helper(list, seen_bits, first, i);
1046 baselen = -1;
1047 seen_bits = 0;
1048 }
1049 if (baselen == -1) {
1050 const char *dot = strrchr(path, '.');
1051 if (!dot) {
1052 report_garbage("garbage found", path);
1053 continue;
1054 }
1055 baselen = dot - path + 1;
1056 first = i;
1057 }
1058 if (!strcmp(path + baselen, "pack"))
1059 seen_bits |= 1;
1060 else if (!strcmp(path + baselen, "idx"))
1061 seen_bits |= 2;
1062 }
1063 report_helper(list, seen_bits, first, list->nr);
1064 }
1065
1066 static void prepare_packed_git_one(char *objdir, int local)
1067 {
1068 /* Ensure that this buffer is large enough so that we can
1069 append "/pack/" without clobbering the stack even if
1070 strlen(objdir) were PATH_MAX. */
1071 char path[PATH_MAX + 1 + 4 + 1 + 1];
1072 int len;
1073 DIR *dir;
1074 struct dirent *de;
1075 struct string_list garbage = STRING_LIST_INIT_DUP;
1076
1077 sprintf(path, "%s/pack", objdir);
1078 len = strlen(path);
1079 dir = opendir(path);
1080 if (!dir) {
1081 if (errno != ENOENT)
1082 error("unable to open object pack directory: %s: %s",
1083 path, strerror(errno));
1084 return;
1085 }
1086 path[len++] = '/';
1087 while ((de = readdir(dir)) != NULL) {
1088 int namelen = strlen(de->d_name);
1089 struct packed_git *p;
1090
1091 if (len + namelen + 1 > sizeof(path)) {
1092 if (report_garbage) {
1093 struct strbuf sb = STRBUF_INIT;
1094 strbuf_addf(&sb, "%.*s/%s", len - 1, path, de->d_name);
1095 report_garbage("path too long", sb.buf);
1096 strbuf_release(&sb);
1097 }
1098 continue;
1099 }
1100
1101 if (is_dot_or_dotdot(de->d_name))
1102 continue;
1103
1104 strcpy(path + len, de->d_name);
1105
1106 if (has_extension(de->d_name, ".idx")) {
1107 /* Don't reopen a pack we already have. */
1108 for (p = packed_git; p; p = p->next) {
1109 if (!memcmp(path, p->pack_name, len + namelen - 4))
1110 break;
1111 }
1112 if (p == NULL &&
1113 /*
1114 * See if it really is a valid .idx file with
1115 * corresponding .pack file that we can map.
1116 */
1117 (p = add_packed_git(path, len + namelen, local)) != NULL)
1118 install_packed_git(p);
1119 }
1120
1121 if (!report_garbage)
1122 continue;
1123
1124 if (has_extension(de->d_name, ".idx") ||
1125 has_extension(de->d_name, ".pack") ||
1126 has_extension(de->d_name, ".keep"))
1127 string_list_append(&garbage, path);
1128 else
1129 report_garbage("garbage found", path);
1130 }
1131 closedir(dir);
1132 report_pack_garbage(&garbage);
1133 string_list_clear(&garbage, 0);
1134 }
1135
1136 static int sort_pack(const void *a_, const void *b_)
1137 {
1138 struct packed_git *a = *((struct packed_git **)a_);
1139 struct packed_git *b = *((struct packed_git **)b_);
1140 int st;
1141
1142 /*
1143 * Local packs tend to contain objects specific to our
1144 * variant of the project than remote ones. In addition,
1145 * remote ones could be on a network mounted filesystem.
1146 * Favor local ones for these reasons.
1147 */
1148 st = a->pack_local - b->pack_local;
1149 if (st)
1150 return -st;
1151
1152 /*
1153 * Younger packs tend to contain more recent objects,
1154 * and more recent objects tend to get accessed more
1155 * often.
1156 */
1157 if (a->mtime < b->mtime)
1158 return 1;
1159 else if (a->mtime == b->mtime)
1160 return 0;
1161 return -1;
1162 }
1163
1164 static void rearrange_packed_git(void)
1165 {
1166 struct packed_git **ary, *p;
1167 int i, n;
1168
1169 for (n = 0, p = packed_git; p; p = p->next)
1170 n++;
1171 if (n < 2)
1172 return;
1173
1174 /* prepare an array of packed_git for easier sorting */
1175 ary = xcalloc(n, sizeof(struct packed_git *));
1176 for (n = 0, p = packed_git; p; p = p->next)
1177 ary[n++] = p;
1178
1179 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1180
1181 /* link them back again */
1182 for (i = 0; i < n - 1; i++)
1183 ary[i]->next = ary[i + 1];
1184 ary[n - 1]->next = NULL;
1185 packed_git = ary[0];
1186
1187 free(ary);
1188 }
1189
1190 static int prepare_packed_git_run_once = 0;
1191 void prepare_packed_git(void)
1192 {
1193 struct alternate_object_database *alt;
1194
1195 if (prepare_packed_git_run_once)
1196 return;
1197 prepare_packed_git_one(get_object_directory(), 1);
1198 prepare_alt_odb();
1199 for (alt = alt_odb_list; alt; alt = alt->next) {
1200 alt->name[-1] = 0;
1201 prepare_packed_git_one(alt->base, 0);
1202 alt->name[-1] = '/';
1203 }
1204 rearrange_packed_git();
1205 prepare_packed_git_run_once = 1;
1206 }
1207
1208 void reprepare_packed_git(void)
1209 {
1210 discard_revindex();
1211 prepare_packed_git_run_once = 0;
1212 prepare_packed_git();
1213 }
1214
1215 static void mark_bad_packed_object(struct packed_git *p,
1216 const unsigned char *sha1)
1217 {
1218 unsigned i;
1219 for (i = 0; i < p->num_bad_objects; i++)
1220 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1221 return;
1222 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1223 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1224 p->num_bad_objects++;
1225 }
1226
1227 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1228 {
1229 struct packed_git *p;
1230 unsigned i;
1231
1232 for (p = packed_git; p; p = p->next)
1233 for (i = 0; i < p->num_bad_objects; i++)
1234 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1235 return p;
1236 return NULL;
1237 }
1238
1239 /*
1240 * With an in-core object data in "map", rehash it to make sure the
1241 * object name actually matches "sha1" to detect object corruption.
1242 * With "map" == NULL, try reading the object named with "sha1" using
1243 * the streaming interface and rehash it to do the same.
1244 */
1245 int check_sha1_signature(const unsigned char *sha1, void *map,
1246 unsigned long size, const char *type)
1247 {
1248 unsigned char real_sha1[20];
1249 enum object_type obj_type;
1250 struct git_istream *st;
1251 git_SHA_CTX c;
1252 char hdr[32];
1253 int hdrlen;
1254
1255 if (map) {
1256 hash_sha1_file(map, size, type, real_sha1);
1257 return hashcmp(sha1, real_sha1) ? -1 : 0;
1258 }
1259
1260 st = open_istream(sha1, &obj_type, &size, NULL);
1261 if (!st)
1262 return -1;
1263
1264 /* Generate the header */
1265 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1266
1267 /* Sha1.. */
1268 git_SHA1_Init(&c);
1269 git_SHA1_Update(&c, hdr, hdrlen);
1270 for (;;) {
1271 char buf[1024 * 16];
1272 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1273
1274 if (readlen < 0) {
1275 close_istream(st);
1276 return -1;
1277 }
1278 if (!readlen)
1279 break;
1280 git_SHA1_Update(&c, buf, readlen);
1281 }
1282 git_SHA1_Final(real_sha1, &c);
1283 close_istream(st);
1284 return hashcmp(sha1, real_sha1) ? -1 : 0;
1285 }
1286
1287 static int git_open_noatime(const char *name)
1288 {
1289 static int sha1_file_open_flag = O_NOATIME;
1290
1291 for (;;) {
1292 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1293 if (fd >= 0)
1294 return fd;
1295
1296 /* Might the failure be due to O_NOATIME? */
1297 if (errno != ENOENT && sha1_file_open_flag) {
1298 sha1_file_open_flag = 0;
1299 continue;
1300 }
1301
1302 return -1;
1303 }
1304 }
1305
1306 static int open_sha1_file(const unsigned char *sha1)
1307 {
1308 int fd;
1309 char *name = sha1_file_name(sha1);
1310 struct alternate_object_database *alt;
1311
1312 fd = git_open_noatime(name);
1313 if (fd >= 0)
1314 return fd;
1315
1316 prepare_alt_odb();
1317 errno = ENOENT;
1318 for (alt = alt_odb_list; alt; alt = alt->next) {
1319 name = alt->name;
1320 fill_sha1_path(name, sha1);
1321 fd = git_open_noatime(alt->base);
1322 if (fd >= 0)
1323 return fd;
1324 }
1325 return -1;
1326 }
1327
1328 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1329 {
1330 void *map;
1331 int fd;
1332
1333 fd = open_sha1_file(sha1);
1334 map = NULL;
1335 if (fd >= 0) {
1336 struct stat st;
1337
1338 if (!fstat(fd, &st)) {
1339 *size = xsize_t(st.st_size);
1340 if (!*size) {
1341 /* mmap() is forbidden on empty files */
1342 error("object file %s is empty", sha1_file_name(sha1));
1343 return NULL;
1344 }
1345 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1346 }
1347 close(fd);
1348 }
1349 return map;
1350 }
1351
1352 /*
1353 * There used to be a second loose object header format which
1354 * was meant to mimic the in-pack format, allowing for direct
1355 * copy of the object data. This format turned up not to be
1356 * really worth it and we no longer write loose objects in that
1357 * format.
1358 */
1359 static int experimental_loose_object(unsigned char *map)
1360 {
1361 unsigned int word;
1362
1363 /*
1364 * We must determine if the buffer contains the standard
1365 * zlib-deflated stream or the experimental format based
1366 * on the in-pack object format. Compare the header byte
1367 * for each format:
1368 *
1369 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1370 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1371 *
1372 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1373 * in standard loose-object format, UNLESS it is a Git-pack
1374 * format object *exactly* 8 bytes in size when inflated.
1375 *
1376 * However, RFC1950 also specifies that the 1st 16-bit word
1377 * must be divisible by 31 - this checksum tells us our buffer
1378 * is in the standard format, giving a false positive only if
1379 * the 1st word of the Git-pack format object happens to be
1380 * divisible by 31, ie:
1381 * ((byte0 * 256) + byte1) % 31 = 0
1382 * => 0ttt10000www1000 % 31 = 0
1383 *
1384 * As it happens, this case can only arise for www=3 & ttt=1
1385 * - ie, a Commit object, which would have to be 8 bytes in
1386 * size. As no Commit can be that small, we find that the
1387 * combination of these two criteria (bitmask & checksum)
1388 * can always correctly determine the buffer format.
1389 */
1390 word = (map[0] << 8) + map[1];
1391 if ((map[0] & 0x8F) == 0x08 && !(word % 31))
1392 return 0;
1393 else
1394 return 1;
1395 }
1396
1397 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1398 unsigned long len, enum object_type *type, unsigned long *sizep)
1399 {
1400 unsigned shift;
1401 unsigned long size, c;
1402 unsigned long used = 0;
1403
1404 c = buf[used++];
1405 *type = (c >> 4) & 7;
1406 size = c & 15;
1407 shift = 4;
1408 while (c & 0x80) {
1409 if (len <= used || bitsizeof(long) <= shift) {
1410 error("bad object header");
1411 size = used = 0;
1412 break;
1413 }
1414 c = buf[used++];
1415 size += (c & 0x7f) << shift;
1416 shift += 7;
1417 }
1418 *sizep = size;
1419 return used;
1420 }
1421
1422 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1423 {
1424 unsigned long size, used;
1425 static const char valid_loose_object_type[8] = {
1426 0, /* OBJ_EXT */
1427 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1428 0, /* "delta" and others are invalid in a loose object */
1429 };
1430 enum object_type type;
1431
1432 /* Get the data stream */
1433 memset(stream, 0, sizeof(*stream));
1434 stream->next_in = map;
1435 stream->avail_in = mapsize;
1436 stream->next_out = buffer;
1437 stream->avail_out = bufsiz;
1438
1439 if (experimental_loose_object(map)) {
1440 /*
1441 * The old experimental format we no longer produce;
1442 * we can still read it.
1443 */
1444 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1445 if (!used || !valid_loose_object_type[type])
1446 return -1;
1447 map += used;
1448 mapsize -= used;
1449
1450 /* Set up the stream for the rest.. */
1451 stream->next_in = map;
1452 stream->avail_in = mapsize;
1453 git_inflate_init(stream);
1454
1455 /* And generate the fake traditional header */
1456 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1457 typename(type), size);
1458 return 0;
1459 }
1460 git_inflate_init(stream);
1461 return git_inflate(stream, 0);
1462 }
1463
1464 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1465 {
1466 int bytes = strlen(buffer) + 1;
1467 unsigned char *buf = xmallocz(size);
1468 unsigned long n;
1469 int status = Z_OK;
1470
1471 n = stream->total_out - bytes;
1472 if (n > size)
1473 n = size;
1474 memcpy(buf, (char *) buffer + bytes, n);
1475 bytes = n;
1476 if (bytes <= size) {
1477 /*
1478 * The above condition must be (bytes <= size), not
1479 * (bytes < size). In other words, even though we
1480 * expect no more output and set avail_out to zero,
1481 * the input zlib stream may have bytes that express
1482 * "this concludes the stream", and we *do* want to
1483 * eat that input.
1484 *
1485 * Otherwise we would not be able to test that we
1486 * consumed all the input to reach the expected size;
1487 * we also want to check that zlib tells us that all
1488 * went well with status == Z_STREAM_END at the end.
1489 */
1490 stream->next_out = buf + bytes;
1491 stream->avail_out = size - bytes;
1492 while (status == Z_OK)
1493 status = git_inflate(stream, Z_FINISH);
1494 }
1495 if (status == Z_STREAM_END && !stream->avail_in) {
1496 git_inflate_end(stream);
1497 return buf;
1498 }
1499
1500 if (status < 0)
1501 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1502 else if (stream->avail_in)
1503 error("garbage at end of loose object '%s'",
1504 sha1_to_hex(sha1));
1505 free(buf);
1506 return NULL;
1507 }
1508
1509 /*
1510 * We used to just use "sscanf()", but that's actually way
1511 * too permissive for what we want to check. So do an anal
1512 * object header parse by hand.
1513 */
1514 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1515 {
1516 char type[10];
1517 int i;
1518 unsigned long size;
1519
1520 /*
1521 * The type can be at most ten bytes (including the
1522 * terminating '\0' that we add), and is followed by
1523 * a space.
1524 */
1525 i = 0;
1526 for (;;) {
1527 char c = *hdr++;
1528 if (c == ' ')
1529 break;
1530 type[i++] = c;
1531 if (i >= sizeof(type))
1532 return -1;
1533 }
1534 type[i] = 0;
1535
1536 /*
1537 * The length must follow immediately, and be in canonical
1538 * decimal format (ie "010" is not valid).
1539 */
1540 size = *hdr++ - '0';
1541 if (size > 9)
1542 return -1;
1543 if (size) {
1544 for (;;) {
1545 unsigned long c = *hdr - '0';
1546 if (c > 9)
1547 break;
1548 hdr++;
1549 size = size * 10 + c;
1550 }
1551 }
1552 *sizep = size;
1553
1554 /*
1555 * The length must be followed by a zero byte
1556 */
1557 return *hdr ? -1 : type_from_string(type);
1558 }
1559
1560 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1561 {
1562 int ret;
1563 git_zstream stream;
1564 char hdr[8192];
1565
1566 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1567 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1568 return NULL;
1569
1570 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1571 }
1572
1573 unsigned long get_size_from_delta(struct packed_git *p,
1574 struct pack_window **w_curs,
1575 off_t curpos)
1576 {
1577 const unsigned char *data;
1578 unsigned char delta_head[20], *in;
1579 git_zstream stream;
1580 int st;
1581
1582 memset(&stream, 0, sizeof(stream));
1583 stream.next_out = delta_head;
1584 stream.avail_out = sizeof(delta_head);
1585
1586 git_inflate_init(&stream);
1587 do {
1588 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1589 stream.next_in = in;
1590 st = git_inflate(&stream, Z_FINISH);
1591 curpos += stream.next_in - in;
1592 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1593 stream.total_out < sizeof(delta_head));
1594 git_inflate_end(&stream);
1595 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1596 error("delta data unpack-initial failed");
1597 return 0;
1598 }
1599
1600 /* Examine the initial part of the delta to figure out
1601 * the result size.
1602 */
1603 data = delta_head;
1604
1605 /* ignore base size */
1606 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1607
1608 /* Read the result size */
1609 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1610 }
1611
1612 static off_t get_delta_base(struct packed_git *p,
1613 struct pack_window **w_curs,
1614 off_t *curpos,
1615 enum object_type type,
1616 off_t delta_obj_offset)
1617 {
1618 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1619 off_t base_offset;
1620
1621 /* use_pack() assured us we have [base_info, base_info + 20)
1622 * as a range that we can look at without walking off the
1623 * end of the mapped window. Its actually the hash size
1624 * that is assured. An OFS_DELTA longer than the hash size
1625 * is stupid, as then a REF_DELTA would be smaller to store.
1626 */
1627 if (type == OBJ_OFS_DELTA) {
1628 unsigned used = 0;
1629 unsigned char c = base_info[used++];
1630 base_offset = c & 127;
1631 while (c & 128) {
1632 base_offset += 1;
1633 if (!base_offset || MSB(base_offset, 7))
1634 return 0; /* overflow */
1635 c = base_info[used++];
1636 base_offset = (base_offset << 7) + (c & 127);
1637 }
1638 base_offset = delta_obj_offset - base_offset;
1639 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1640 return 0; /* out of bound */
1641 *curpos += used;
1642 } else if (type == OBJ_REF_DELTA) {
1643 /* The base entry _must_ be in the same pack */
1644 base_offset = find_pack_entry_one(base_info, p);
1645 *curpos += 20;
1646 } else
1647 die("I am totally screwed");
1648 return base_offset;
1649 }
1650
1651 int unpack_object_header(struct packed_git *p,
1652 struct pack_window **w_curs,
1653 off_t *curpos,
1654 unsigned long *sizep)
1655 {
1656 unsigned char *base;
1657 unsigned long left;
1658 unsigned long used;
1659 enum object_type type;
1660
1661 /* use_pack() assures us we have [base, base + 20) available
1662 * as a range that we can look at. (Its actually the hash
1663 * size that is assured.) With our object header encoding
1664 * the maximum deflated object size is 2^137, which is just
1665 * insane, so we know won't exceed what we have been given.
1666 */
1667 base = use_pack(p, w_curs, *curpos, &left);
1668 used = unpack_object_header_buffer(base, left, &type, sizep);
1669 if (!used) {
1670 type = OBJ_BAD;
1671 } else
1672 *curpos += used;
1673
1674 return type;
1675 }
1676
1677 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1678 {
1679 int type;
1680 struct revindex_entry *revidx;
1681 const unsigned char *sha1;
1682 revidx = find_pack_revindex(p, obj_offset);
1683 if (!revidx)
1684 return OBJ_BAD;
1685 sha1 = nth_packed_object_sha1(p, revidx->nr);
1686 mark_bad_packed_object(p, sha1);
1687 type = sha1_object_info(sha1, NULL);
1688 if (type <= OBJ_NONE)
1689 return OBJ_BAD;
1690 return type;
1691 }
1692
1693
1694 #define POI_STACK_PREALLOC 64
1695
1696 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1697 unsigned long *sizep, int *rtype)
1698 {
1699 struct pack_window *w_curs = NULL;
1700 unsigned long size;
1701 off_t curpos = obj_offset;
1702 enum object_type type;
1703 off_t small_poi_stack[POI_STACK_PREALLOC];
1704 off_t *poi_stack = small_poi_stack;
1705 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1706
1707 type = unpack_object_header(p, &w_curs, &curpos, &size);
1708
1709 if (rtype)
1710 *rtype = type; /* representation type */
1711
1712 if (sizep) {
1713 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1714 off_t tmp_pos = curpos;
1715 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1716 type, obj_offset);
1717 if (!base_offset) {
1718 type = OBJ_BAD;
1719 goto out;
1720 }
1721 *sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1722 if (*sizep == 0) {
1723 type = OBJ_BAD;
1724 goto out;
1725 }
1726 } else {
1727 *sizep = size;
1728 }
1729 }
1730
1731 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1732 off_t base_offset;
1733 /* Push the object we're going to leave behind */
1734 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1735 poi_stack_alloc = alloc_nr(poi_stack_nr);
1736 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1737 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1738 } else {
1739 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1740 }
1741 poi_stack[poi_stack_nr++] = obj_offset;
1742 /* If parsing the base offset fails, just unwind */
1743 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1744 if (!base_offset)
1745 goto unwind;
1746 curpos = obj_offset = base_offset;
1747 type = unpack_object_header(p, &w_curs, &curpos, &size);
1748 if (type <= OBJ_NONE) {
1749 /* If getting the base itself fails, we first
1750 * retry the base, otherwise unwind */
1751 type = retry_bad_packed_offset(p, base_offset);
1752 if (type > OBJ_NONE)
1753 goto out;
1754 goto unwind;
1755 }
1756 }
1757
1758 switch (type) {
1759 case OBJ_BAD:
1760 case OBJ_COMMIT:
1761 case OBJ_TREE:
1762 case OBJ_BLOB:
1763 case OBJ_TAG:
1764 break;
1765 default:
1766 error("unknown object type %i at offset %"PRIuMAX" in %s",
1767 type, (uintmax_t)obj_offset, p->pack_name);
1768 type = OBJ_BAD;
1769 }
1770
1771 out:
1772 if (poi_stack != small_poi_stack)
1773 free(poi_stack);
1774 unuse_pack(&w_curs);
1775 return type;
1776
1777 unwind:
1778 while (poi_stack_nr) {
1779 obj_offset = poi_stack[--poi_stack_nr];
1780 type = retry_bad_packed_offset(p, obj_offset);
1781 if (type > OBJ_NONE)
1782 goto out;
1783 }
1784 type = OBJ_BAD;
1785 goto out;
1786 }
1787
1788 static void *unpack_compressed_entry(struct packed_git *p,
1789 struct pack_window **w_curs,
1790 off_t curpos,
1791 unsigned long size)
1792 {
1793 int st;
1794 git_zstream stream;
1795 unsigned char *buffer, *in;
1796
1797 buffer = xmallocz(size);
1798 memset(&stream, 0, sizeof(stream));
1799 stream.next_out = buffer;
1800 stream.avail_out = size + 1;
1801
1802 git_inflate_init(&stream);
1803 do {
1804 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1805 stream.next_in = in;
1806 st = git_inflate(&stream, Z_FINISH);
1807 if (!stream.avail_out)
1808 break; /* the payload is larger than it should be */
1809 curpos += stream.next_in - in;
1810 } while (st == Z_OK || st == Z_BUF_ERROR);
1811 git_inflate_end(&stream);
1812 if ((st != Z_STREAM_END) || stream.total_out != size) {
1813 free(buffer);
1814 return NULL;
1815 }
1816
1817 return buffer;
1818 }
1819
1820 #define MAX_DELTA_CACHE (256)
1821
1822 static size_t delta_base_cached;
1823
1824 static struct delta_base_cache_lru_list {
1825 struct delta_base_cache_lru_list *prev;
1826 struct delta_base_cache_lru_list *next;
1827 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1828
1829 static struct delta_base_cache_entry {
1830 struct delta_base_cache_lru_list lru;
1831 void *data;
1832 struct packed_git *p;
1833 off_t base_offset;
1834 unsigned long size;
1835 enum object_type type;
1836 } delta_base_cache[MAX_DELTA_CACHE];
1837
1838 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1839 {
1840 unsigned long hash;
1841
1842 hash = (unsigned long)p + (unsigned long)base_offset;
1843 hash += (hash >> 8) + (hash >> 16);
1844 return hash % MAX_DELTA_CACHE;
1845 }
1846
1847 static struct delta_base_cache_entry *
1848 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1849 {
1850 unsigned long hash = pack_entry_hash(p, base_offset);
1851 return delta_base_cache + hash;
1852 }
1853
1854 static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
1855 struct packed_git *p, off_t base_offset)
1856 {
1857 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1858 }
1859
1860 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1861 {
1862 struct delta_base_cache_entry *ent;
1863 ent = get_delta_base_cache_entry(p, base_offset);
1864 return eq_delta_base_cache_entry(ent, p, base_offset);
1865 }
1866
1867 static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1868 {
1869 ent->data = NULL;
1870 ent->lru.next->prev = ent->lru.prev;
1871 ent->lru.prev->next = ent->lru.next;
1872 delta_base_cached -= ent->size;
1873 }
1874
1875 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1876 unsigned long *base_size, enum object_type *type, int keep_cache)
1877 {
1878 struct delta_base_cache_entry *ent;
1879 void *ret;
1880
1881 ent = get_delta_base_cache_entry(p, base_offset);
1882
1883 if (!eq_delta_base_cache_entry(ent, p, base_offset))
1884 return unpack_entry(p, base_offset, type, base_size);
1885
1886 ret = ent->data;
1887
1888 if (!keep_cache)
1889 clear_delta_base_cache_entry(ent);
1890 else
1891 ret = xmemdupz(ent->data, ent->size);
1892 *type = ent->type;
1893 *base_size = ent->size;
1894 return ret;
1895 }
1896
1897 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1898 {
1899 if (ent->data) {
1900 free(ent->data);
1901 ent->data = NULL;
1902 ent->lru.next->prev = ent->lru.prev;
1903 ent->lru.prev->next = ent->lru.next;
1904 delta_base_cached -= ent->size;
1905 }
1906 }
1907
1908 void clear_delta_base_cache(void)
1909 {
1910 unsigned long p;
1911 for (p = 0; p < MAX_DELTA_CACHE; p++)
1912 release_delta_base_cache(&delta_base_cache[p]);
1913 }
1914
1915 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1916 void *base, unsigned long base_size, enum object_type type)
1917 {
1918 unsigned long hash = pack_entry_hash(p, base_offset);
1919 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1920 struct delta_base_cache_lru_list *lru;
1921
1922 release_delta_base_cache(ent);
1923 delta_base_cached += base_size;
1924
1925 for (lru = delta_base_cache_lru.next;
1926 delta_base_cached > delta_base_cache_limit
1927 && lru != &delta_base_cache_lru;
1928 lru = lru->next) {
1929 struct delta_base_cache_entry *f = (void *)lru;
1930 if (f->type == OBJ_BLOB)
1931 release_delta_base_cache(f);
1932 }
1933 for (lru = delta_base_cache_lru.next;
1934 delta_base_cached > delta_base_cache_limit
1935 && lru != &delta_base_cache_lru;
1936 lru = lru->next) {
1937 struct delta_base_cache_entry *f = (void *)lru;
1938 release_delta_base_cache(f);
1939 }
1940
1941 ent->p = p;
1942 ent->base_offset = base_offset;
1943 ent->type = type;
1944 ent->data = base;
1945 ent->size = base_size;
1946 ent->lru.next = &delta_base_cache_lru;
1947 ent->lru.prev = delta_base_cache_lru.prev;
1948 delta_base_cache_lru.prev->next = &ent->lru;
1949 delta_base_cache_lru.prev = &ent->lru;
1950 }
1951
1952 static void *read_object(const unsigned char *sha1, enum object_type *type,
1953 unsigned long *size);
1954
1955 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1956 {
1957 static FILE *log_file;
1958
1959 if (!log_file) {
1960 log_file = fopen(log_pack_access, "w");
1961 if (!log_file) {
1962 error("cannot open pack access log '%s' for writing: %s",
1963 log_pack_access, strerror(errno));
1964 log_pack_access = NULL;
1965 return;
1966 }
1967 }
1968 fprintf(log_file, "%s %"PRIuMAX"\n",
1969 p->pack_name, (uintmax_t)obj_offset);
1970 fflush(log_file);
1971 }
1972
1973 int do_check_packed_object_crc;
1974
1975 #define UNPACK_ENTRY_STACK_PREALLOC 64
1976 struct unpack_entry_stack_ent {
1977 off_t obj_offset;
1978 off_t curpos;
1979 unsigned long size;
1980 };
1981
1982 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1983 enum object_type *final_type, unsigned long *final_size)
1984 {
1985 struct pack_window *w_curs = NULL;
1986 off_t curpos = obj_offset;
1987 void *data = NULL;
1988 unsigned long size;
1989 enum object_type type;
1990 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1991 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1992 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1993 int base_from_cache = 0;
1994
1995 if (log_pack_access)
1996 write_pack_access_log(p, obj_offset);
1997
1998 /* PHASE 1: drill down to the innermost base object */
1999 for (;;) {
2000 off_t base_offset;
2001 int i;
2002 struct delta_base_cache_entry *ent;
2003
2004 if (do_check_packed_object_crc && p->index_version > 1) {
2005 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2006 unsigned long len = revidx[1].offset - obj_offset;
2007 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2008 const unsigned char *sha1 =
2009 nth_packed_object_sha1(p, revidx->nr);
2010 error("bad packed object CRC for %s",
2011 sha1_to_hex(sha1));
2012 mark_bad_packed_object(p, sha1);
2013 unuse_pack(&w_curs);
2014 return NULL;
2015 }
2016 }
2017
2018 ent = get_delta_base_cache_entry(p, curpos);
2019 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2020 type = ent->type;
2021 data = ent->data;
2022 size = ent->size;
2023 clear_delta_base_cache_entry(ent);
2024 base_from_cache = 1;
2025 break;
2026 }
2027
2028 type = unpack_object_header(p, &w_curs, &curpos, &size);
2029 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2030 break;
2031
2032 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2033 if (!base_offset) {
2034 error("failed to validate delta base reference "
2035 "at offset %"PRIuMAX" from %s",
2036 (uintmax_t)curpos, p->pack_name);
2037 /* bail to phase 2, in hopes of recovery */
2038 data = NULL;
2039 break;
2040 }
2041
2042 /* push object, proceed to base */
2043 if (delta_stack_nr >= delta_stack_alloc
2044 && delta_stack == small_delta_stack) {
2045 delta_stack_alloc = alloc_nr(delta_stack_nr);
2046 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2047 memcpy(delta_stack, small_delta_stack,
2048 sizeof(*delta_stack)*delta_stack_nr);
2049 } else {
2050 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2051 }
2052 i = delta_stack_nr++;
2053 delta_stack[i].obj_offset = obj_offset;
2054 delta_stack[i].curpos = curpos;
2055 delta_stack[i].size = size;
2056
2057 curpos = obj_offset = base_offset;
2058 }
2059
2060 /* PHASE 2: handle the base */
2061 switch (type) {
2062 case OBJ_OFS_DELTA:
2063 case OBJ_REF_DELTA:
2064 if (data)
2065 die("BUG in unpack_entry: left loop at a valid delta");
2066 break;
2067 case OBJ_COMMIT:
2068 case OBJ_TREE:
2069 case OBJ_BLOB:
2070 case OBJ_TAG:
2071 if (!base_from_cache)
2072 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2073 break;
2074 default:
2075 data = NULL;
2076 error("unknown object type %i at offset %"PRIuMAX" in %s",
2077 type, (uintmax_t)obj_offset, p->pack_name);
2078 }
2079
2080 /* PHASE 3: apply deltas in order */
2081
2082 /* invariants:
2083 * 'data' holds the base data, or NULL if there was corruption
2084 */
2085 while (delta_stack_nr) {
2086 void *delta_data;
2087 void *base = data;
2088 unsigned long delta_size, base_size = size;
2089 int i;
2090
2091 data = NULL;
2092
2093 if (base)
2094 add_delta_base_cache(p, obj_offset, base, base_size, type);
2095
2096 if (!base) {
2097 /*
2098 * We're probably in deep shit, but let's try to fetch
2099 * the required base anyway from another pack or loose.
2100 * This is costly but should happen only in the presence
2101 * of a corrupted pack, and is better than failing outright.
2102 */
2103 struct revindex_entry *revidx;
2104 const unsigned char *base_sha1;
2105 revidx = find_pack_revindex(p, obj_offset);
2106 if (revidx) {
2107 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2108 error("failed to read delta base object %s"
2109 " at offset %"PRIuMAX" from %s",
2110 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2111 p->pack_name);
2112 mark_bad_packed_object(p, base_sha1);
2113 base = read_object(base_sha1, &type, &base_size);
2114 }
2115 }
2116
2117 i = --delta_stack_nr;
2118 obj_offset = delta_stack[i].obj_offset;
2119 curpos = delta_stack[i].curpos;
2120 delta_size = delta_stack[i].size;
2121
2122 if (!base)
2123 continue;
2124
2125 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2126
2127 if (!delta_data) {
2128 error("failed to unpack compressed delta "
2129 "at offset %"PRIuMAX" from %s",
2130 (uintmax_t)curpos, p->pack_name);
2131 data = NULL;
2132 continue;
2133 }
2134
2135 data = patch_delta(base, base_size,
2136 delta_data, delta_size,
2137 &size);
2138 if (!data)
2139 die("failed to apply delta");
2140
2141 free (delta_data);
2142 }
2143
2144 *final_type = type;
2145 *final_size = size;
2146
2147 unuse_pack(&w_curs);
2148 return data;
2149 }
2150
2151 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2152 uint32_t n)
2153 {
2154 const unsigned char *index = p->index_data;
2155 if (!index) {
2156 if (open_pack_index(p))
2157 return NULL;
2158 index = p->index_data;
2159 }
2160 if (n >= p->num_objects)
2161 return NULL;
2162 index += 4 * 256;
2163 if (p->index_version == 1) {
2164 return index + 24 * n + 4;
2165 } else {
2166 index += 8;
2167 return index + 20 * n;
2168 }
2169 }
2170
2171 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2172 {
2173 const unsigned char *index = p->index_data;
2174 index += 4 * 256;
2175 if (p->index_version == 1) {
2176 return ntohl(*((uint32_t *)(index + 24 * n)));
2177 } else {
2178 uint32_t off;
2179 index += 8 + p->num_objects * (20 + 4);
2180 off = ntohl(*((uint32_t *)(index + 4 * n)));
2181 if (!(off & 0x80000000))
2182 return off;
2183 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2184 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2185 ntohl(*((uint32_t *)(index + 4)));
2186 }
2187 }
2188
2189 off_t find_pack_entry_one(const unsigned char *sha1,
2190 struct packed_git *p)
2191 {
2192 const uint32_t *level1_ofs = p->index_data;
2193 const unsigned char *index = p->index_data;
2194 unsigned hi, lo, stride;
2195 static int use_lookup = -1;
2196 static int debug_lookup = -1;
2197
2198 if (debug_lookup < 0)
2199 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2200
2201 if (!index) {
2202 if (open_pack_index(p))
2203 return 0;
2204 level1_ofs = p->index_data;
2205 index = p->index_data;
2206 }
2207 if (p->index_version > 1) {
2208 level1_ofs += 2;
2209 index += 8;
2210 }
2211 index += 4 * 256;
2212 hi = ntohl(level1_ofs[*sha1]);
2213 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2214 if (p->index_version > 1) {
2215 stride = 20;
2216 } else {
2217 stride = 24;
2218 index += 4;
2219 }
2220
2221 if (debug_lookup)
2222 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2223 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2224
2225 if (use_lookup < 0)
2226 use_lookup = !!getenv("GIT_USE_LOOKUP");
2227 if (use_lookup) {
2228 int pos = sha1_entry_pos(index, stride, 0,
2229 lo, hi, p->num_objects, sha1);
2230 if (pos < 0)
2231 return 0;
2232 return nth_packed_object_offset(p, pos);
2233 }
2234
2235 do {
2236 unsigned mi = (lo + hi) / 2;
2237 int cmp = hashcmp(index + mi * stride, sha1);
2238
2239 if (debug_lookup)
2240 printf("lo %u hi %u rg %u mi %u\n",
2241 lo, hi, hi - lo, mi);
2242 if (!cmp)
2243 return nth_packed_object_offset(p, mi);
2244 if (cmp > 0)
2245 hi = mi;
2246 else
2247 lo = mi+1;
2248 } while (lo < hi);
2249 return 0;
2250 }
2251
2252 int is_pack_valid(struct packed_git *p)
2253 {
2254 /* An already open pack is known to be valid. */
2255 if (p->pack_fd != -1)
2256 return 1;
2257
2258 /* If the pack has one window completely covering the
2259 * file size, the pack is known to be valid even if
2260 * the descriptor is not currently open.
2261 */
2262 if (p->windows) {
2263 struct pack_window *w = p->windows;
2264
2265 if (!w->offset && w->len == p->pack_size)
2266 return 1;
2267 }
2268
2269 /* Force the pack to open to prove its valid. */
2270 return !open_packed_git(p);
2271 }
2272
2273 static int fill_pack_entry(const unsigned char *sha1,
2274 struct pack_entry *e,
2275 struct packed_git *p)
2276 {
2277 off_t offset;
2278
2279 if (p->num_bad_objects) {
2280 unsigned i;
2281 for (i = 0; i < p->num_bad_objects; i++)
2282 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2283 return 0;
2284 }
2285
2286 offset = find_pack_entry_one(sha1, p);
2287 if (!offset)
2288 return 0;
2289
2290 /*
2291 * We are about to tell the caller where they can locate the
2292 * requested object. We better make sure the packfile is
2293 * still here and can be accessed before supplying that
2294 * answer, as it may have been deleted since the index was
2295 * loaded!
2296 */
2297 if (!is_pack_valid(p)) {
2298 warning("packfile %s cannot be accessed", p->pack_name);
2299 return 0;
2300 }
2301 e->offset = offset;
2302 e->p = p;
2303 hashcpy(e->sha1, sha1);
2304 return 1;
2305 }
2306
2307 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2308 {
2309 struct packed_git *p;
2310
2311 prepare_packed_git();
2312 if (!packed_git)
2313 return 0;
2314
2315 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2316 return 1;
2317
2318 for (p = packed_git; p; p = p->next) {
2319 if (p == last_found_pack || !fill_pack_entry(sha1, e, p))
2320 continue;
2321
2322 last_found_pack = p;
2323 return 1;
2324 }
2325 return 0;
2326 }
2327
2328 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2329 struct packed_git *packs)
2330 {
2331 struct packed_git *p;
2332
2333 for (p = packs; p; p = p->next) {
2334 if (find_pack_entry_one(sha1, p))
2335 return p;
2336 }
2337 return NULL;
2338
2339 }
2340
2341 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
2342 {
2343 int status;
2344 unsigned long mapsize, size;
2345 void *map;
2346 git_zstream stream;
2347 char hdr[32];
2348
2349 map = map_sha1_file(sha1, &mapsize);
2350 if (!map)
2351 return error("unable to find %s", sha1_to_hex(sha1));
2352 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2353 status = error("unable to unpack %s header",
2354 sha1_to_hex(sha1));
2355 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2356 status = error("unable to parse %s header", sha1_to_hex(sha1));
2357 else if (sizep)
2358 *sizep = size;
2359 git_inflate_end(&stream);
2360 munmap(map, mapsize);
2361 return status;
2362 }
2363
2364 /* returns enum object_type or negative */
2365 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
2366 {
2367 struct cached_object *co;
2368 struct pack_entry e;
2369 int status, rtype;
2370
2371 co = find_cached_object(sha1);
2372 if (co) {
2373 if (oi->sizep)
2374 *(oi->sizep) = co->size;
2375 oi->whence = OI_CACHED;
2376 return co->type;
2377 }
2378
2379 if (!find_pack_entry(sha1, &e)) {
2380 /* Most likely it's a loose object. */
2381 status = sha1_loose_object_info(sha1, oi->sizep);
2382 if (status >= 0) {
2383 oi->whence = OI_LOOSE;
2384 return status;
2385 }
2386
2387 /* Not a loose object; someone else may have just packed it. */
2388 reprepare_packed_git();
2389 if (!find_pack_entry(sha1, &e))
2390 return status;
2391 }
2392
2393 status = packed_object_info(e.p, e.offset, oi->sizep, &rtype);
2394 if (status < 0) {
2395 mark_bad_packed_object(e.p, sha1);
2396 status = sha1_object_info_extended(sha1, oi);
2397 } else if (in_delta_base_cache(e.p, e.offset)) {
2398 oi->whence = OI_DBCACHED;
2399 } else {
2400 oi->whence = OI_PACKED;
2401 oi->u.packed.offset = e.offset;
2402 oi->u.packed.pack = e.p;
2403 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2404 rtype == OBJ_OFS_DELTA);
2405 }
2406
2407 return status;
2408 }
2409
2410 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2411 {
2412 struct object_info oi;
2413
2414 oi.sizep = sizep;
2415 return sha1_object_info_extended(sha1, &oi);
2416 }
2417
2418 static void *read_packed_sha1(const unsigned char *sha1,
2419 enum object_type *type, unsigned long *size)
2420 {
2421 struct pack_entry e;
2422 void *data;
2423
2424 if (!find_pack_entry(sha1, &e))
2425 return NULL;
2426 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2427 if (!data) {
2428 /*
2429 * We're probably in deep shit, but let's try to fetch
2430 * the required object anyway from another pack or loose.
2431 * This should happen only in the presence of a corrupted
2432 * pack, and is better than failing outright.
2433 */
2434 error("failed to read object %s at offset %"PRIuMAX" from %s",
2435 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2436 mark_bad_packed_object(e.p, sha1);
2437 data = read_object(sha1, type, size);
2438 }
2439 return data;
2440 }
2441
2442 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2443 unsigned char *sha1)
2444 {
2445 struct cached_object *co;
2446
2447 hash_sha1_file(buf, len, typename(type), sha1);
2448 if (has_sha1_file(sha1) || find_cached_object(sha1))
2449 return 0;
2450 if (cached_object_alloc <= cached_object_nr) {
2451 cached_object_alloc = alloc_nr(cached_object_alloc);
2452 cached_objects = xrealloc(cached_objects,
2453 sizeof(*cached_objects) *
2454 cached_object_alloc);
2455 }
2456 co = &cached_objects[cached_object_nr++];
2457 co->size = len;
2458 co->type = type;
2459 co->buf = xmalloc(len);
2460 memcpy(co->buf, buf, len);
2461 hashcpy(co->sha1, sha1);
2462 return 0;
2463 }
2464
2465 static void *read_object(const unsigned char *sha1, enum object_type *type,
2466 unsigned long *size)
2467 {
2468 unsigned long mapsize;
2469 void *map, *buf;
2470 struct cached_object *co;
2471
2472 co = find_cached_object(sha1);
2473 if (co) {
2474 *type = co->type;
2475 *size = co->size;
2476 return xmemdupz(co->buf, co->size);
2477 }
2478
2479 buf = read_packed_sha1(sha1, type, size);
2480 if (buf)
2481 return buf;
2482 map = map_sha1_file(sha1, &mapsize);
2483 if (map) {
2484 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2485 munmap(map, mapsize);
2486 return buf;
2487 }
2488 reprepare_packed_git();
2489 return read_packed_sha1(sha1, type, size);
2490 }
2491
2492 /*
2493 * This function dies on corrupt objects; the callers who want to
2494 * deal with them should arrange to call read_object() and give error
2495 * messages themselves.
2496 */
2497 void *read_sha1_file_extended(const unsigned char *sha1,
2498 enum object_type *type,
2499 unsigned long *size,
2500 unsigned flag)
2501 {
2502 void *data;
2503 char *path;
2504 const struct packed_git *p;
2505 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2506 ? lookup_replace_object(sha1) : sha1;
2507
2508 errno = 0;
2509 data = read_object(repl, type, size);
2510 if (data)
2511 return data;
2512
2513 if (errno && errno != ENOENT)
2514 die_errno("failed to read object %s", sha1_to_hex(sha1));
2515
2516 /* die if we replaced an object with one that does not exist */
2517 if (repl != sha1)
2518 die("replacement %s not found for %s",
2519 sha1_to_hex(repl), sha1_to_hex(sha1));
2520
2521 if (has_loose_object(repl)) {
2522 path = sha1_file_name(sha1);
2523 die("loose object %s (stored in %s) is corrupt",
2524 sha1_to_hex(repl), path);
2525 }
2526
2527 if ((p = has_packed_and_bad(repl)) != NULL)
2528 die("packed object %s (stored in %s) is corrupt",
2529 sha1_to_hex(repl), p->pack_name);
2530
2531 return NULL;
2532 }
2533
2534 void *read_object_with_reference(const unsigned char *sha1,
2535 const char *required_type_name,
2536 unsigned long *size,
2537 unsigned char *actual_sha1_return)
2538 {
2539 enum object_type type, required_type;
2540 void *buffer;
2541 unsigned long isize;
2542 unsigned char actual_sha1[20];
2543
2544 required_type = type_from_string(required_type_name);
2545 hashcpy(actual_sha1, sha1);
2546 while (1) {
2547 int ref_length = -1;
2548 const char *ref_type = NULL;
2549
2550 buffer = read_sha1_file(actual_sha1, &type, &isize);
2551 if (!buffer)
2552 return NULL;
2553 if (type == required_type) {
2554 *size = isize;
2555 if (actual_sha1_return)
2556 hashcpy(actual_sha1_return, actual_sha1);
2557 return buffer;
2558 }
2559 /* Handle references */
2560 else if (type == OBJ_COMMIT)
2561 ref_type = "tree ";
2562 else if (type == OBJ_TAG)
2563 ref_type = "object ";
2564 else {
2565 free(buffer);
2566 return NULL;
2567 }
2568 ref_length = strlen(ref_type);
2569
2570 if (ref_length + 40 > isize ||
2571 memcmp(buffer, ref_type, ref_length) ||
2572 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2573 free(buffer);
2574 return NULL;
2575 }
2576 free(buffer);
2577 /* Now we have the ID of the referred-to object in
2578 * actual_sha1. Check again. */
2579 }
2580 }
2581
2582 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2583 const char *type, unsigned char *sha1,
2584 char *hdr, int *hdrlen)
2585 {
2586 git_SHA_CTX c;
2587
2588 /* Generate the header */
2589 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2590
2591 /* Sha1.. */
2592 git_SHA1_Init(&c);
2593 git_SHA1_Update(&c, hdr, *hdrlen);
2594 git_SHA1_Update(&c, buf, len);
2595 git_SHA1_Final(sha1, &c);
2596 }
2597
2598 /*
2599 * Move the just written object into its final resting place.
2600 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2601 * "moving" is only a part of what it does, when no patch between
2602 * master to pu changes the call sites of this function.
2603 */
2604 int move_temp_to_file(const char *tmpfile, const char *filename)
2605 {
2606 int ret = 0;
2607
2608 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2609 goto try_rename;
2610 else if (link(tmpfile, filename))
2611 ret = errno;
2612
2613 /*
2614 * Coda hack - coda doesn't like cross-directory links,
2615 * so we fall back to a rename, which will mean that it
2616 * won't be able to check collisions, but that's not a
2617 * big deal.
2618 *
2619 * The same holds for FAT formatted media.
2620 *
2621 * When this succeeds, we just return. We have nothing
2622 * left to unlink.
2623 */
2624 if (ret && ret != EEXIST) {
2625 try_rename:
2626 if (!rename(tmpfile, filename))
2627 goto out;
2628 ret = errno;
2629 }
2630 unlink_or_warn(tmpfile);
2631 if (ret) {
2632 if (ret != EEXIST) {
2633 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2634 }
2635 /* FIXME!!! Collision check here ? */
2636 }
2637
2638 out:
2639 if (adjust_shared_perm(filename))
2640 return error("unable to set permission to '%s'", filename);
2641 return 0;
2642 }
2643
2644 static int write_buffer(int fd, const void *buf, size_t len)
2645 {
2646 if (write_in_full(fd, buf, len) < 0)
2647 return error("file write error (%s)", strerror(errno));
2648 return 0;
2649 }
2650
2651 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2652 unsigned char *sha1)
2653 {
2654 char hdr[32];
2655 int hdrlen;
2656 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2657 return 0;
2658 }
2659
2660 /* Finalize a file on disk, and close it. */
2661 static void close_sha1_file(int fd)
2662 {
2663 if (fsync_object_files)
2664 fsync_or_die(fd, "sha1 file");
2665 if (close(fd) != 0)
2666 die_errno("error when closing sha1 file");
2667 }
2668
2669 /* Size of directory component, including the ending '/' */
2670 static inline int directory_size(const char *filename)
2671 {
2672 const char *s = strrchr(filename, '/');
2673 if (!s)
2674 return 0;
2675 return s - filename + 1;
2676 }
2677
2678 /*
2679 * This creates a temporary file in the same directory as the final
2680 * 'filename'
2681 *
2682 * We want to avoid cross-directory filename renames, because those
2683 * can have problems on various filesystems (FAT, NFS, Coda).
2684 */
2685 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2686 {
2687 int fd, dirlen = directory_size(filename);
2688
2689 if (dirlen + 20 > bufsiz) {
2690 errno = ENAMETOOLONG;
2691 return -1;
2692 }
2693 memcpy(buffer, filename, dirlen);
2694 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2695 fd = git_mkstemp_mode(buffer, 0444);
2696 if (fd < 0 && dirlen && errno == ENOENT) {
2697 /* Make sure the directory exists */
2698 memcpy(buffer, filename, dirlen);
2699 buffer[dirlen-1] = 0;
2700 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2701 return -1;
2702
2703 /* Try again */
2704 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2705 fd = git_mkstemp_mode(buffer, 0444);
2706 }
2707 return fd;
2708 }
2709
2710 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2711 const void *buf, unsigned long len, time_t mtime)
2712 {
2713 int fd, ret;
2714 unsigned char compressed[4096];
2715 git_zstream stream;
2716 git_SHA_CTX c;
2717 unsigned char parano_sha1[20];
2718 char *filename;
2719 static char tmp_file[PATH_MAX];
2720
2721 filename = sha1_file_name(sha1);
2722 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
2723 if (fd < 0) {
2724 if (errno == EACCES)
2725 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2726 else
2727 return error("unable to create temporary file: %s", strerror(errno));
2728 }
2729
2730 /* Set it up */
2731 memset(&stream, 0, sizeof(stream));
2732 git_deflate_init(&stream, zlib_compression_level);
2733 stream.next_out = compressed;
2734 stream.avail_out = sizeof(compressed);
2735 git_SHA1_Init(&c);
2736
2737 /* First header.. */
2738 stream.next_in = (unsigned char *)hdr;
2739 stream.avail_in = hdrlen;
2740 while (git_deflate(&stream, 0) == Z_OK)
2741 ; /* nothing */
2742 git_SHA1_Update(&c, hdr, hdrlen);
2743
2744 /* Then the data itself.. */
2745 stream.next_in = (void *)buf;
2746 stream.avail_in = len;
2747 do {
2748 unsigned char *in0 = stream.next_in;
2749 ret = git_deflate(&stream, Z_FINISH);
2750 git_SHA1_Update(&c, in0, stream.next_in - in0);
2751 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2752 die("unable to write sha1 file");
2753 stream.next_out = compressed;
2754 stream.avail_out = sizeof(compressed);
2755 } while (ret == Z_OK);
2756
2757 if (ret != Z_STREAM_END)
2758 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2759 ret = git_deflate_end_gently(&stream);
2760 if (ret != Z_OK)
2761 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2762 git_SHA1_Final(parano_sha1, &c);
2763 if (hashcmp(sha1, parano_sha1) != 0)
2764 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2765
2766 close_sha1_file(fd);
2767
2768 if (mtime) {
2769 struct utimbuf utb;
2770 utb.actime = mtime;
2771 utb.modtime = mtime;
2772 if (utime(tmp_file, &utb) < 0)
2773 warning("failed utime() on %s: %s",
2774 tmp_file, strerror(errno));
2775 }
2776
2777 return move_temp_to_file(tmp_file, filename);
2778 }
2779
2780 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2781 {
2782 unsigned char sha1[20];
2783 char hdr[32];
2784 int hdrlen;
2785
2786 /* Normally if we have it in the pack then we do not bother writing
2787 * it out into .git/objects/??/?{38} file.
2788 */
2789 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2790 if (returnsha1)
2791 hashcpy(returnsha1, sha1);
2792 if (has_sha1_file(sha1))
2793 return 0;
2794 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2795 }
2796
2797 int force_object_loose(const unsigned char *sha1, time_t mtime)
2798 {
2799 void *buf;
2800 unsigned long len;
2801 enum object_type type;
2802 char hdr[32];
2803 int hdrlen;
2804 int ret;
2805
2806 if (has_loose_object(sha1))
2807 return 0;
2808 buf = read_packed_sha1(sha1, &type, &len);
2809 if (!buf)
2810 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2811 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2812 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2813 free(buf);
2814
2815 return ret;
2816 }
2817
2818 int has_pack_index(const unsigned char *sha1)
2819 {
2820 struct stat st;
2821 if (stat(sha1_pack_index_name(sha1), &st))
2822 return 0;
2823 return 1;
2824 }
2825
2826 int has_sha1_pack(const unsigned char *sha1)
2827 {
2828 struct pack_entry e;
2829 return find_pack_entry(sha1, &e);
2830 }
2831
2832 int has_sha1_file(const unsigned char *sha1)
2833 {
2834 struct pack_entry e;
2835
2836 if (find_pack_entry(sha1, &e))
2837 return 1;
2838 return has_loose_object(sha1);
2839 }
2840
2841 static void check_tree(const void *buf, size_t size)
2842 {
2843 struct tree_desc desc;
2844 struct name_entry entry;
2845
2846 init_tree_desc(&desc, buf, size);
2847 while (tree_entry(&desc, &entry))
2848 /* do nothing
2849 * tree_entry() will die() on malformed entries */
2850 ;
2851 }
2852
2853 static void check_commit(const void *buf, size_t size)
2854 {
2855 struct commit c;
2856 memset(&c, 0, sizeof(c));
2857 if (parse_commit_buffer(&c, buf, size))
2858 die("corrupt commit");
2859 }
2860
2861 static void check_tag(const void *buf, size_t size)
2862 {
2863 struct tag t;
2864 memset(&t, 0, sizeof(t));
2865 if (parse_tag_buffer(&t, buf, size))
2866 die("corrupt tag");
2867 }
2868
2869 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2870 enum object_type type,
2871 const char *path, unsigned flags)
2872 {
2873 int ret, re_allocated = 0;
2874 int write_object = flags & HASH_WRITE_OBJECT;
2875
2876 if (!type)
2877 type = OBJ_BLOB;
2878
2879 /*
2880 * Convert blobs to git internal format
2881 */
2882 if ((type == OBJ_BLOB) && path) {
2883 struct strbuf nbuf = STRBUF_INIT;
2884 if (convert_to_git(path, buf, size, &nbuf,
2885 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
2886 buf = strbuf_detach(&nbuf, &size);
2887 re_allocated = 1;
2888 }
2889 }
2890 if (flags & HASH_FORMAT_CHECK) {
2891 if (type == OBJ_TREE)
2892 check_tree(buf, size);
2893 if (type == OBJ_COMMIT)
2894 check_commit(buf, size);
2895 if (type == OBJ_TAG)
2896 check_tag(buf, size);
2897 }
2898
2899 if (write_object)
2900 ret = write_sha1_file(buf, size, typename(type), sha1);
2901 else
2902 ret = hash_sha1_file(buf, size, typename(type), sha1);
2903 if (re_allocated)
2904 free(buf);
2905 return ret;
2906 }
2907
2908 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2909 const char *path, unsigned flags)
2910 {
2911 struct strbuf sbuf = STRBUF_INIT;
2912 int ret;
2913
2914 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2915 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2916 else
2917 ret = -1;
2918 strbuf_release(&sbuf);
2919 return ret;
2920 }
2921
2922 #define SMALL_FILE_SIZE (32*1024)
2923
2924 static int index_core(unsigned char *sha1, int fd, size_t size,
2925 enum object_type type, const char *path,
2926 unsigned flags)
2927 {
2928 int ret;
2929
2930 if (!size) {
2931 ret = index_mem(sha1, NULL, size, type, path, flags);
2932 } else if (size <= SMALL_FILE_SIZE) {
2933 char *buf = xmalloc(size);
2934 if (size == read_in_full(fd, buf, size))
2935 ret = index_mem(sha1, buf, size, type, path, flags);
2936 else
2937 ret = error("short read %s", strerror(errno));
2938 free(buf);
2939 } else {
2940 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2941 ret = index_mem(sha1, buf, size, type, path, flags);
2942 munmap(buf, size);
2943 }
2944 return ret;
2945 }
2946
2947 /*
2948 * This creates one packfile per large blob unless bulk-checkin
2949 * machinery is "plugged".
2950 *
2951 * This also bypasses the usual "convert-to-git" dance, and that is on
2952 * purpose. We could write a streaming version of the converting
2953 * functions and insert that before feeding the data to fast-import
2954 * (or equivalent in-core API described above). However, that is
2955 * somewhat complicated, as we do not know the size of the filter
2956 * result, which we need to know beforehand when writing a git object.
2957 * Since the primary motivation for trying to stream from the working
2958 * tree file and to avoid mmaping it in core is to deal with large
2959 * binary blobs, they generally do not want to get any conversion, and
2960 * callers should avoid this code path when filters are requested.
2961 */
2962 static int index_stream(unsigned char *sha1, int fd, size_t size,
2963 enum object_type type, const char *path,
2964 unsigned flags)
2965 {
2966 return index_bulk_checkin(sha1, fd, size, type, path, flags);
2967 }
2968
2969 int index_fd(unsigned char *sha1, int fd, struct stat *st,
2970 enum object_type type, const char *path, unsigned flags)
2971 {
2972 int ret;
2973 size_t size = xsize_t(st->st_size);
2974
2975 if (!S_ISREG(st->st_mode))
2976 ret = index_pipe(sha1, fd, type, path, flags);
2977 else if (size <= big_file_threshold || type != OBJ_BLOB ||
2978 (path && would_convert_to_git(path, NULL, 0, 0)))
2979 ret = index_core(sha1, fd, size, type, path, flags);
2980 else
2981 ret = index_stream(sha1, fd, size, type, path, flags);
2982 close(fd);
2983 return ret;
2984 }
2985
2986 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
2987 {
2988 int fd;
2989 struct strbuf sb = STRBUF_INIT;
2990
2991 switch (st->st_mode & S_IFMT) {
2992 case S_IFREG:
2993 fd = open(path, O_RDONLY);
2994 if (fd < 0)
2995 return error("open(\"%s\"): %s", path,
2996 strerror(errno));
2997 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
2998 return error("%s: failed to insert into database",
2999 path);
3000 break;
3001 case S_IFLNK:
3002 if (strbuf_readlink(&sb, path, st->st_size)) {
3003 char *errstr = strerror(errno);
3004 return error("readlink(\"%s\"): %s", path,
3005 errstr);
3006 }
3007 if (!(flags & HASH_WRITE_OBJECT))
3008 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3009 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3010 return error("%s: failed to insert into database",
3011 path);
3012 strbuf_release(&sb);
3013 break;
3014 case S_IFDIR:
3015 return resolve_gitlink_ref(path, "HEAD", sha1);
3016 default:
3017 return error("%s: unsupported file type", path);
3018 }
3019 return 0;
3020 }
3021
3022 int read_pack_header(int fd, struct pack_header *header)
3023 {
3024 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3025 /* "eof before pack header was fully read" */
3026 return PH_ERROR_EOF;
3027
3028 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3029 /* "protocol error (pack signature mismatch detected)" */
3030 return PH_ERROR_PACK_SIGNATURE;
3031 if (!pack_version_ok(header->hdr_version))
3032 /* "protocol error (pack version unsupported)" */
3033 return PH_ERROR_PROTOCOL;
3034 return 0;
3035 }
3036
3037 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3038 {
3039 enum object_type type = sha1_object_info(sha1, NULL);
3040 if (type < 0)
3041 die("%s is not a valid object", sha1_to_hex(sha1));
3042 if (type != expect)
3043 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3044 typename(expect));
3045 }