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