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
35static inline uintmax_t sz_fmt(size_t s) { return s; }
36
37const 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 */
45static struct cached_object {
46 unsigned char sha1[20];
47 enum object_type type;
48 void *buf;
49 unsigned long size;
50} *cached_objects;
51static int cached_object_nr, cached_object_alloc;
52
53static struct cached_object empty_tree = {
54 EMPTY_TREE_SHA1_BIN_LITERAL,
55 OBJ_TREE,
56 "",
57 0
58};
59
60static struct packed_git *last_found_pack;
61
62static 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
76int 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
105int 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
144int 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
153static 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 */
175char *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
195static 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
222char *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
229char *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
236struct alternate_object_database *alt_odb_list;
237static struct alternate_object_database **alt_odb_tail;
238
239static 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 */
256static 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
326static 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
356void 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
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, 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, 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
734static 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 */
756static 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
829static 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
841static 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
854unsigned 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
929static 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
937static void try_to_free_pack_memory(size_t size)
938{
939 release_pack_memory(size, -1);
940}
941
942struct 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
985struct 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
1000void 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
1009void (*report_garbage)(const char *desc, const char *path);
1010
1011static 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
1032static 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
1066static 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
1136static 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
1164static 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
1190static int prepare_packed_git_run_once = 0;
1191void 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
1208void reprepare_packed_git(void)
1209{
1210 discard_revindex();
1211 prepare_packed_git_run_once = 0;
1212 prepare_packed_git();
1213}
1214
1215static 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
1227static 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 */
1245int 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
1287static 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
1306static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1307{
1308 char *name = sha1_file_name(sha1);
1309 struct alternate_object_database *alt;
1310
1311 if (!lstat(name, st))
1312 return 0;
1313
1314 prepare_alt_odb();
1315 errno = ENOENT;
1316 for (alt = alt_odb_list; alt; alt = alt->next) {
1317 name = alt->name;
1318 fill_sha1_path(name, sha1);
1319 if (!lstat(alt->base, st))
1320 return 0;
1321 }
1322
1323 return -1;
1324}
1325
1326static int open_sha1_file(const unsigned char *sha1)
1327{
1328 int fd;
1329 char *name = sha1_file_name(sha1);
1330 struct alternate_object_database *alt;
1331
1332 fd = git_open_noatime(name);
1333 if (fd >= 0)
1334 return fd;
1335
1336 prepare_alt_odb();
1337 errno = ENOENT;
1338 for (alt = alt_odb_list; alt; alt = alt->next) {
1339 name = alt->name;
1340 fill_sha1_path(name, sha1);
1341 fd = git_open_noatime(alt->base);
1342 if (fd >= 0)
1343 return fd;
1344 }
1345 return -1;
1346}
1347
1348void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1349{
1350 void *map;
1351 int fd;
1352
1353 fd = open_sha1_file(sha1);
1354 map = NULL;
1355 if (fd >= 0) {
1356 struct stat st;
1357
1358 if (!fstat(fd, &st)) {
1359 *size = xsize_t(st.st_size);
1360 if (!*size) {
1361 /* mmap() is forbidden on empty files */
1362 error("object file %s is empty", sha1_file_name(sha1));
1363 return NULL;
1364 }
1365 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1366 }
1367 close(fd);
1368 }
1369 return map;
1370}
1371
1372/*
1373 * There used to be a second loose object header format which
1374 * was meant to mimic the in-pack format, allowing for direct
1375 * copy of the object data. This format turned up not to be
1376 * really worth it and we no longer write loose objects in that
1377 * format.
1378 */
1379static int experimental_loose_object(unsigned char *map)
1380{
1381 unsigned int word;
1382
1383 /*
1384 * We must determine if the buffer contains the standard
1385 * zlib-deflated stream or the experimental format based
1386 * on the in-pack object format. Compare the header byte
1387 * for each format:
1388 *
1389 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1390 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1391 *
1392 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1393 * in standard loose-object format, UNLESS it is a Git-pack
1394 * format object *exactly* 8 bytes in size when inflated.
1395 *
1396 * However, RFC1950 also specifies that the 1st 16-bit word
1397 * must be divisible by 31 - this checksum tells us our buffer
1398 * is in the standard format, giving a false positive only if
1399 * the 1st word of the Git-pack format object happens to be
1400 * divisible by 31, ie:
1401 * ((byte0 * 256) + byte1) % 31 = 0
1402 * => 0ttt10000www1000 % 31 = 0
1403 *
1404 * As it happens, this case can only arise for www=3 & ttt=1
1405 * - ie, a Commit object, which would have to be 8 bytes in
1406 * size. As no Commit can be that small, we find that the
1407 * combination of these two criteria (bitmask & checksum)
1408 * can always correctly determine the buffer format.
1409 */
1410 word = (map[0] << 8) + map[1];
1411 if ((map[0] & 0x8F) == 0x08 && !(word % 31))
1412 return 0;
1413 else
1414 return 1;
1415}
1416
1417unsigned long unpack_object_header_buffer(const unsigned char *buf,
1418 unsigned long len, enum object_type *type, unsigned long *sizep)
1419{
1420 unsigned shift;
1421 unsigned long size, c;
1422 unsigned long used = 0;
1423
1424 c = buf[used++];
1425 *type = (c >> 4) & 7;
1426 size = c & 15;
1427 shift = 4;
1428 while (c & 0x80) {
1429 if (len <= used || bitsizeof(long) <= shift) {
1430 error("bad object header");
1431 size = used = 0;
1432 break;
1433 }
1434 c = buf[used++];
1435 size += (c & 0x7f) << shift;
1436 shift += 7;
1437 }
1438 *sizep = size;
1439 return used;
1440}
1441
1442int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1443{
1444 unsigned long size, used;
1445 static const char valid_loose_object_type[8] = {
1446 0, /* OBJ_EXT */
1447 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1448 0, /* "delta" and others are invalid in a loose object */
1449 };
1450 enum object_type type;
1451
1452 /* Get the data stream */
1453 memset(stream, 0, sizeof(*stream));
1454 stream->next_in = map;
1455 stream->avail_in = mapsize;
1456 stream->next_out = buffer;
1457 stream->avail_out = bufsiz;
1458
1459 if (experimental_loose_object(map)) {
1460 /*
1461 * The old experimental format we no longer produce;
1462 * we can still read it.
1463 */
1464 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1465 if (!used || !valid_loose_object_type[type])
1466 return -1;
1467 map += used;
1468 mapsize -= used;
1469
1470 /* Set up the stream for the rest.. */
1471 stream->next_in = map;
1472 stream->avail_in = mapsize;
1473 git_inflate_init(stream);
1474
1475 /* And generate the fake traditional header */
1476 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1477 typename(type), size);
1478 return 0;
1479 }
1480 git_inflate_init(stream);
1481 return git_inflate(stream, 0);
1482}
1483
1484static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1485{
1486 int bytes = strlen(buffer) + 1;
1487 unsigned char *buf = xmallocz(size);
1488 unsigned long n;
1489 int status = Z_OK;
1490
1491 n = stream->total_out - bytes;
1492 if (n > size)
1493 n = size;
1494 memcpy(buf, (char *) buffer + bytes, n);
1495 bytes = n;
1496 if (bytes <= size) {
1497 /*
1498 * The above condition must be (bytes <= size), not
1499 * (bytes < size). In other words, even though we
1500 * expect no more output and set avail_out to zero,
1501 * the input zlib stream may have bytes that express
1502 * "this concludes the stream", and we *do* want to
1503 * eat that input.
1504 *
1505 * Otherwise we would not be able to test that we
1506 * consumed all the input to reach the expected size;
1507 * we also want to check that zlib tells us that all
1508 * went well with status == Z_STREAM_END at the end.
1509 */
1510 stream->next_out = buf + bytes;
1511 stream->avail_out = size - bytes;
1512 while (status == Z_OK)
1513 status = git_inflate(stream, Z_FINISH);
1514 }
1515 if (status == Z_STREAM_END && !stream->avail_in) {
1516 git_inflate_end(stream);
1517 return buf;
1518 }
1519
1520 if (status < 0)
1521 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1522 else if (stream->avail_in)
1523 error("garbage at end of loose object '%s'",
1524 sha1_to_hex(sha1));
1525 free(buf);
1526 return NULL;
1527}
1528
1529/*
1530 * We used to just use "sscanf()", but that's actually way
1531 * too permissive for what we want to check. So do an anal
1532 * object header parse by hand.
1533 */
1534int parse_sha1_header(const char *hdr, unsigned long *sizep)
1535{
1536 char type[10];
1537 int i;
1538 unsigned long size;
1539
1540 /*
1541 * The type can be at most ten bytes (including the
1542 * terminating '\0' that we add), and is followed by
1543 * a space.
1544 */
1545 i = 0;
1546 for (;;) {
1547 char c = *hdr++;
1548 if (c == ' ')
1549 break;
1550 type[i++] = c;
1551 if (i >= sizeof(type))
1552 return -1;
1553 }
1554 type[i] = 0;
1555
1556 /*
1557 * The length must follow immediately, and be in canonical
1558 * decimal format (ie "010" is not valid).
1559 */
1560 size = *hdr++ - '0';
1561 if (size > 9)
1562 return -1;
1563 if (size) {
1564 for (;;) {
1565 unsigned long c = *hdr - '0';
1566 if (c > 9)
1567 break;
1568 hdr++;
1569 size = size * 10 + c;
1570 }
1571 }
1572 *sizep = size;
1573
1574 /*
1575 * The length must be followed by a zero byte
1576 */
1577 return *hdr ? -1 : type_from_string(type);
1578}
1579
1580static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1581{
1582 int ret;
1583 git_zstream stream;
1584 char hdr[8192];
1585
1586 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1587 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1588 return NULL;
1589
1590 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1591}
1592
1593unsigned long get_size_from_delta(struct packed_git *p,
1594 struct pack_window **w_curs,
1595 off_t curpos)
1596{
1597 const unsigned char *data;
1598 unsigned char delta_head[20], *in;
1599 git_zstream stream;
1600 int st;
1601
1602 memset(&stream, 0, sizeof(stream));
1603 stream.next_out = delta_head;
1604 stream.avail_out = sizeof(delta_head);
1605
1606 git_inflate_init(&stream);
1607 do {
1608 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1609 stream.next_in = in;
1610 st = git_inflate(&stream, Z_FINISH);
1611 curpos += stream.next_in - in;
1612 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1613 stream.total_out < sizeof(delta_head));
1614 git_inflate_end(&stream);
1615 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1616 error("delta data unpack-initial failed");
1617 return 0;
1618 }
1619
1620 /* Examine the initial part of the delta to figure out
1621 * the result size.
1622 */
1623 data = delta_head;
1624
1625 /* ignore base size */
1626 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1627
1628 /* Read the result size */
1629 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1630}
1631
1632static off_t get_delta_base(struct packed_git *p,
1633 struct pack_window **w_curs,
1634 off_t *curpos,
1635 enum object_type type,
1636 off_t delta_obj_offset)
1637{
1638 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1639 off_t base_offset;
1640
1641 /* use_pack() assured us we have [base_info, base_info + 20)
1642 * as a range that we can look at without walking off the
1643 * end of the mapped window. Its actually the hash size
1644 * that is assured. An OFS_DELTA longer than the hash size
1645 * is stupid, as then a REF_DELTA would be smaller to store.
1646 */
1647 if (type == OBJ_OFS_DELTA) {
1648 unsigned used = 0;
1649 unsigned char c = base_info[used++];
1650 base_offset = c & 127;
1651 while (c & 128) {
1652 base_offset += 1;
1653 if (!base_offset || MSB(base_offset, 7))
1654 return 0; /* overflow */
1655 c = base_info[used++];
1656 base_offset = (base_offset << 7) + (c & 127);
1657 }
1658 base_offset = delta_obj_offset - base_offset;
1659 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1660 return 0; /* out of bound */
1661 *curpos += used;
1662 } else if (type == OBJ_REF_DELTA) {
1663 /* The base entry _must_ be in the same pack */
1664 base_offset = find_pack_entry_one(base_info, p);
1665 *curpos += 20;
1666 } else
1667 die("I am totally screwed");
1668 return base_offset;
1669}
1670
1671int unpack_object_header(struct packed_git *p,
1672 struct pack_window **w_curs,
1673 off_t *curpos,
1674 unsigned long *sizep)
1675{
1676 unsigned char *base;
1677 unsigned long left;
1678 unsigned long used;
1679 enum object_type type;
1680
1681 /* use_pack() assures us we have [base, base + 20) available
1682 * as a range that we can look at. (Its actually the hash
1683 * size that is assured.) With our object header encoding
1684 * the maximum deflated object size is 2^137, which is just
1685 * insane, so we know won't exceed what we have been given.
1686 */
1687 base = use_pack(p, w_curs, *curpos, &left);
1688 used = unpack_object_header_buffer(base, left, &type, sizep);
1689 if (!used) {
1690 type = OBJ_BAD;
1691 } else
1692 *curpos += used;
1693
1694 return type;
1695}
1696
1697static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1698{
1699 int type;
1700 struct revindex_entry *revidx;
1701 const unsigned char *sha1;
1702 revidx = find_pack_revindex(p, obj_offset);
1703 if (!revidx)
1704 return OBJ_BAD;
1705 sha1 = nth_packed_object_sha1(p, revidx->nr);
1706 mark_bad_packed_object(p, sha1);
1707 type = sha1_object_info(sha1, NULL);
1708 if (type <= OBJ_NONE)
1709 return OBJ_BAD;
1710 return type;
1711}
1712
1713#define POI_STACK_PREALLOC 64
1714
1715static enum object_type packed_to_object_type(struct packed_git *p,
1716 off_t obj_offset,
1717 enum object_type type,
1718 struct pack_window **w_curs,
1719 off_t curpos)
1720{
1721 off_t small_poi_stack[POI_STACK_PREALLOC];
1722 off_t *poi_stack = small_poi_stack;
1723 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1724
1725 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1726 off_t base_offset;
1727 unsigned long size;
1728 /* Push the object we're going to leave behind */
1729 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1730 poi_stack_alloc = alloc_nr(poi_stack_nr);
1731 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1732 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1733 } else {
1734 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1735 }
1736 poi_stack[poi_stack_nr++] = obj_offset;
1737 /* If parsing the base offset fails, just unwind */
1738 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1739 if (!base_offset)
1740 goto unwind;
1741 curpos = obj_offset = base_offset;
1742 type = unpack_object_header(p, w_curs, &curpos, &size);
1743 if (type <= OBJ_NONE) {
1744 /* If getting the base itself fails, we first
1745 * retry the base, otherwise unwind */
1746 type = retry_bad_packed_offset(p, base_offset);
1747 if (type > OBJ_NONE)
1748 goto out;
1749 goto unwind;
1750 }
1751 }
1752
1753 switch (type) {
1754 case OBJ_BAD:
1755 case OBJ_COMMIT:
1756 case OBJ_TREE:
1757 case OBJ_BLOB:
1758 case OBJ_TAG:
1759 break;
1760 default:
1761 error("unknown object type %i at offset %"PRIuMAX" in %s",
1762 type, (uintmax_t)obj_offset, p->pack_name);
1763 type = OBJ_BAD;
1764 }
1765
1766out:
1767 if (poi_stack != small_poi_stack)
1768 free(poi_stack);
1769 return type;
1770
1771unwind:
1772 while (poi_stack_nr) {
1773 obj_offset = poi_stack[--poi_stack_nr];
1774 type = retry_bad_packed_offset(p, obj_offset);
1775 if (type > OBJ_NONE)
1776 goto out;
1777 }
1778 type = OBJ_BAD;
1779 goto out;
1780}
1781
1782static int packed_object_info(struct packed_git *p, off_t obj_offset,
1783 struct object_info *oi)
1784{
1785 struct pack_window *w_curs = NULL;
1786 unsigned long size;
1787 off_t curpos = obj_offset;
1788 enum object_type type;
1789
1790 /*
1791 * We always get the representation type, but only convert it to
1792 * a "real" type later if the caller is interested.
1793 */
1794 type = unpack_object_header(p, &w_curs, &curpos, &size);
1795
1796 if (oi->sizep) {
1797 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1798 off_t tmp_pos = curpos;
1799 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1800 type, obj_offset);
1801 if (!base_offset) {
1802 type = OBJ_BAD;
1803 goto out;
1804 }
1805 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1806 if (*oi->sizep == 0) {
1807 type = OBJ_BAD;
1808 goto out;
1809 }
1810 } else {
1811 *oi->sizep = size;
1812 }
1813 }
1814
1815 if (oi->disk_sizep) {
1816 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1817 *oi->disk_sizep = revidx[1].offset - obj_offset;
1818 }
1819
1820 if (oi->typep) {
1821 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
1822 if (*oi->typep < 0) {
1823 type = OBJ_BAD;
1824 goto out;
1825 }
1826 }
1827
1828out:
1829 unuse_pack(&w_curs);
1830 return type;
1831}
1832
1833static void *unpack_compressed_entry(struct packed_git *p,
1834 struct pack_window **w_curs,
1835 off_t curpos,
1836 unsigned long size)
1837{
1838 int st;
1839 git_zstream stream;
1840 unsigned char *buffer, *in;
1841
1842 buffer = xmallocz(size);
1843 memset(&stream, 0, sizeof(stream));
1844 stream.next_out = buffer;
1845 stream.avail_out = size + 1;
1846
1847 git_inflate_init(&stream);
1848 do {
1849 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1850 stream.next_in = in;
1851 st = git_inflate(&stream, Z_FINISH);
1852 if (!stream.avail_out)
1853 break; /* the payload is larger than it should be */
1854 curpos += stream.next_in - in;
1855 } while (st == Z_OK || st == Z_BUF_ERROR);
1856 git_inflate_end(&stream);
1857 if ((st != Z_STREAM_END) || stream.total_out != size) {
1858 free(buffer);
1859 return NULL;
1860 }
1861
1862 return buffer;
1863}
1864
1865#define MAX_DELTA_CACHE (256)
1866
1867static size_t delta_base_cached;
1868
1869static struct delta_base_cache_lru_list {
1870 struct delta_base_cache_lru_list *prev;
1871 struct delta_base_cache_lru_list *next;
1872} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1873
1874static struct delta_base_cache_entry {
1875 struct delta_base_cache_lru_list lru;
1876 void *data;
1877 struct packed_git *p;
1878 off_t base_offset;
1879 unsigned long size;
1880 enum object_type type;
1881} delta_base_cache[MAX_DELTA_CACHE];
1882
1883static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1884{
1885 unsigned long hash;
1886
1887 hash = (unsigned long)p + (unsigned long)base_offset;
1888 hash += (hash >> 8) + (hash >> 16);
1889 return hash % MAX_DELTA_CACHE;
1890}
1891
1892static struct delta_base_cache_entry *
1893get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1894{
1895 unsigned long hash = pack_entry_hash(p, base_offset);
1896 return delta_base_cache + hash;
1897}
1898
1899static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
1900 struct packed_git *p, off_t base_offset)
1901{
1902 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1903}
1904
1905static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1906{
1907 struct delta_base_cache_entry *ent;
1908 ent = get_delta_base_cache_entry(p, base_offset);
1909 return eq_delta_base_cache_entry(ent, p, base_offset);
1910}
1911
1912static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1913{
1914 ent->data = NULL;
1915 ent->lru.next->prev = ent->lru.prev;
1916 ent->lru.prev->next = ent->lru.next;
1917 delta_base_cached -= ent->size;
1918}
1919
1920static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1921 unsigned long *base_size, enum object_type *type, int keep_cache)
1922{
1923 struct delta_base_cache_entry *ent;
1924 void *ret;
1925
1926 ent = get_delta_base_cache_entry(p, base_offset);
1927
1928 if (!eq_delta_base_cache_entry(ent, p, base_offset))
1929 return unpack_entry(p, base_offset, type, base_size);
1930
1931 ret = ent->data;
1932
1933 if (!keep_cache)
1934 clear_delta_base_cache_entry(ent);
1935 else
1936 ret = xmemdupz(ent->data, ent->size);
1937 *type = ent->type;
1938 *base_size = ent->size;
1939 return ret;
1940}
1941
1942static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1943{
1944 if (ent->data) {
1945 free(ent->data);
1946 ent->data = NULL;
1947 ent->lru.next->prev = ent->lru.prev;
1948 ent->lru.prev->next = ent->lru.next;
1949 delta_base_cached -= ent->size;
1950 }
1951}
1952
1953void clear_delta_base_cache(void)
1954{
1955 unsigned long p;
1956 for (p = 0; p < MAX_DELTA_CACHE; p++)
1957 release_delta_base_cache(&delta_base_cache[p]);
1958}
1959
1960static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1961 void *base, unsigned long base_size, enum object_type type)
1962{
1963 unsigned long hash = pack_entry_hash(p, base_offset);
1964 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1965 struct delta_base_cache_lru_list *lru;
1966
1967 release_delta_base_cache(ent);
1968 delta_base_cached += base_size;
1969
1970 for (lru = delta_base_cache_lru.next;
1971 delta_base_cached > delta_base_cache_limit
1972 && lru != &delta_base_cache_lru;
1973 lru = lru->next) {
1974 struct delta_base_cache_entry *f = (void *)lru;
1975 if (f->type == OBJ_BLOB)
1976 release_delta_base_cache(f);
1977 }
1978 for (lru = delta_base_cache_lru.next;
1979 delta_base_cached > delta_base_cache_limit
1980 && lru != &delta_base_cache_lru;
1981 lru = lru->next) {
1982 struct delta_base_cache_entry *f = (void *)lru;
1983 release_delta_base_cache(f);
1984 }
1985
1986 ent->p = p;
1987 ent->base_offset = base_offset;
1988 ent->type = type;
1989 ent->data = base;
1990 ent->size = base_size;
1991 ent->lru.next = &delta_base_cache_lru;
1992 ent->lru.prev = delta_base_cache_lru.prev;
1993 delta_base_cache_lru.prev->next = &ent->lru;
1994 delta_base_cache_lru.prev = &ent->lru;
1995}
1996
1997static void *read_object(const unsigned char *sha1, enum object_type *type,
1998 unsigned long *size);
1999
2000static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2001{
2002 static FILE *log_file;
2003
2004 if (!log_file) {
2005 log_file = fopen(log_pack_access, "w");
2006 if (!log_file) {
2007 error("cannot open pack access log '%s' for writing: %s",
2008 log_pack_access, strerror(errno));
2009 log_pack_access = NULL;
2010 return;
2011 }
2012 }
2013 fprintf(log_file, "%s %"PRIuMAX"\n",
2014 p->pack_name, (uintmax_t)obj_offset);
2015 fflush(log_file);
2016}
2017
2018int do_check_packed_object_crc;
2019
2020#define UNPACK_ENTRY_STACK_PREALLOC 64
2021struct unpack_entry_stack_ent {
2022 off_t obj_offset;
2023 off_t curpos;
2024 unsigned long size;
2025};
2026
2027void *unpack_entry(struct packed_git *p, off_t obj_offset,
2028 enum object_type *final_type, unsigned long *final_size)
2029{
2030 struct pack_window *w_curs = NULL;
2031 off_t curpos = obj_offset;
2032 void *data = NULL;
2033 unsigned long size;
2034 enum object_type type;
2035 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2036 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2037 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2038 int base_from_cache = 0;
2039
2040 if (log_pack_access)
2041 write_pack_access_log(p, obj_offset);
2042
2043 /* PHASE 1: drill down to the innermost base object */
2044 for (;;) {
2045 off_t base_offset;
2046 int i;
2047 struct delta_base_cache_entry *ent;
2048
2049 if (do_check_packed_object_crc && p->index_version > 1) {
2050 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2051 unsigned long len = revidx[1].offset - obj_offset;
2052 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2053 const unsigned char *sha1 =
2054 nth_packed_object_sha1(p, revidx->nr);
2055 error("bad packed object CRC for %s",
2056 sha1_to_hex(sha1));
2057 mark_bad_packed_object(p, sha1);
2058 unuse_pack(&w_curs);
2059 return NULL;
2060 }
2061 }
2062
2063 ent = get_delta_base_cache_entry(p, curpos);
2064 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2065 type = ent->type;
2066 data = ent->data;
2067 size = ent->size;
2068 clear_delta_base_cache_entry(ent);
2069 base_from_cache = 1;
2070 break;
2071 }
2072
2073 type = unpack_object_header(p, &w_curs, &curpos, &size);
2074 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2075 break;
2076
2077 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2078 if (!base_offset) {
2079 error("failed to validate delta base reference "
2080 "at offset %"PRIuMAX" from %s",
2081 (uintmax_t)curpos, p->pack_name);
2082 /* bail to phase 2, in hopes of recovery */
2083 data = NULL;
2084 break;
2085 }
2086
2087 /* push object, proceed to base */
2088 if (delta_stack_nr >= delta_stack_alloc
2089 && delta_stack == small_delta_stack) {
2090 delta_stack_alloc = alloc_nr(delta_stack_nr);
2091 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2092 memcpy(delta_stack, small_delta_stack,
2093 sizeof(*delta_stack)*delta_stack_nr);
2094 } else {
2095 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2096 }
2097 i = delta_stack_nr++;
2098 delta_stack[i].obj_offset = obj_offset;
2099 delta_stack[i].curpos = curpos;
2100 delta_stack[i].size = size;
2101
2102 curpos = obj_offset = base_offset;
2103 }
2104
2105 /* PHASE 2: handle the base */
2106 switch (type) {
2107 case OBJ_OFS_DELTA:
2108 case OBJ_REF_DELTA:
2109 if (data)
2110 die("BUG in unpack_entry: left loop at a valid delta");
2111 break;
2112 case OBJ_COMMIT:
2113 case OBJ_TREE:
2114 case OBJ_BLOB:
2115 case OBJ_TAG:
2116 if (!base_from_cache)
2117 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2118 break;
2119 default:
2120 data = NULL;
2121 error("unknown object type %i at offset %"PRIuMAX" in %s",
2122 type, (uintmax_t)obj_offset, p->pack_name);
2123 }
2124
2125 /* PHASE 3: apply deltas in order */
2126
2127 /* invariants:
2128 * 'data' holds the base data, or NULL if there was corruption
2129 */
2130 while (delta_stack_nr) {
2131 void *delta_data;
2132 void *base = data;
2133 unsigned long delta_size, base_size = size;
2134 int i;
2135
2136 data = NULL;
2137
2138 if (base)
2139 add_delta_base_cache(p, obj_offset, base, base_size, type);
2140
2141 if (!base) {
2142 /*
2143 * We're probably in deep shit, but let's try to fetch
2144 * the required base anyway from another pack or loose.
2145 * This is costly but should happen only in the presence
2146 * of a corrupted pack, and is better than failing outright.
2147 */
2148 struct revindex_entry *revidx;
2149 const unsigned char *base_sha1;
2150 revidx = find_pack_revindex(p, obj_offset);
2151 if (revidx) {
2152 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2153 error("failed to read delta base object %s"
2154 " at offset %"PRIuMAX" from %s",
2155 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2156 p->pack_name);
2157 mark_bad_packed_object(p, base_sha1);
2158 base = read_object(base_sha1, &type, &base_size);
2159 }
2160 }
2161
2162 i = --delta_stack_nr;
2163 obj_offset = delta_stack[i].obj_offset;
2164 curpos = delta_stack[i].curpos;
2165 delta_size = delta_stack[i].size;
2166
2167 if (!base)
2168 continue;
2169
2170 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2171
2172 if (!delta_data) {
2173 error("failed to unpack compressed delta "
2174 "at offset %"PRIuMAX" from %s",
2175 (uintmax_t)curpos, p->pack_name);
2176 data = NULL;
2177 continue;
2178 }
2179
2180 data = patch_delta(base, base_size,
2181 delta_data, delta_size,
2182 &size);
2183 if (!data)
2184 die("failed to apply delta");
2185
2186 free (delta_data);
2187 }
2188
2189 *final_type = type;
2190 *final_size = size;
2191
2192 unuse_pack(&w_curs);
2193 return data;
2194}
2195
2196const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2197 uint32_t n)
2198{
2199 const unsigned char *index = p->index_data;
2200 if (!index) {
2201 if (open_pack_index(p))
2202 return NULL;
2203 index = p->index_data;
2204 }
2205 if (n >= p->num_objects)
2206 return NULL;
2207 index += 4 * 256;
2208 if (p->index_version == 1) {
2209 return index + 24 * n + 4;
2210 } else {
2211 index += 8;
2212 return index + 20 * n;
2213 }
2214}
2215
2216off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2217{
2218 const unsigned char *index = p->index_data;
2219 index += 4 * 256;
2220 if (p->index_version == 1) {
2221 return ntohl(*((uint32_t *)(index + 24 * n)));
2222 } else {
2223 uint32_t off;
2224 index += 8 + p->num_objects * (20 + 4);
2225 off = ntohl(*((uint32_t *)(index + 4 * n)));
2226 if (!(off & 0x80000000))
2227 return off;
2228 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2229 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2230 ntohl(*((uint32_t *)(index + 4)));
2231 }
2232}
2233
2234off_t find_pack_entry_one(const unsigned char *sha1,
2235 struct packed_git *p)
2236{
2237 const uint32_t *level1_ofs = p->index_data;
2238 const unsigned char *index = p->index_data;
2239 unsigned hi, lo, stride;
2240 static int use_lookup = -1;
2241 static int debug_lookup = -1;
2242
2243 if (debug_lookup < 0)
2244 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2245
2246 if (!index) {
2247 if (open_pack_index(p))
2248 return 0;
2249 level1_ofs = p->index_data;
2250 index = p->index_data;
2251 }
2252 if (p->index_version > 1) {
2253 level1_ofs += 2;
2254 index += 8;
2255 }
2256 index += 4 * 256;
2257 hi = ntohl(level1_ofs[*sha1]);
2258 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2259 if (p->index_version > 1) {
2260 stride = 20;
2261 } else {
2262 stride = 24;
2263 index += 4;
2264 }
2265
2266 if (debug_lookup)
2267 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2268 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2269
2270 if (use_lookup < 0)
2271 use_lookup = !!getenv("GIT_USE_LOOKUP");
2272 if (use_lookup) {
2273 int pos = sha1_entry_pos(index, stride, 0,
2274 lo, hi, p->num_objects, sha1);
2275 if (pos < 0)
2276 return 0;
2277 return nth_packed_object_offset(p, pos);
2278 }
2279
2280 do {
2281 unsigned mi = (lo + hi) / 2;
2282 int cmp = hashcmp(index + mi * stride, sha1);
2283
2284 if (debug_lookup)
2285 printf("lo %u hi %u rg %u mi %u\n",
2286 lo, hi, hi - lo, mi);
2287 if (!cmp)
2288 return nth_packed_object_offset(p, mi);
2289 if (cmp > 0)
2290 hi = mi;
2291 else
2292 lo = mi+1;
2293 } while (lo < hi);
2294 return 0;
2295}
2296
2297int is_pack_valid(struct packed_git *p)
2298{
2299 /* An already open pack is known to be valid. */
2300 if (p->pack_fd != -1)
2301 return 1;
2302
2303 /* If the pack has one window completely covering the
2304 * file size, the pack is known to be valid even if
2305 * the descriptor is not currently open.
2306 */
2307 if (p->windows) {
2308 struct pack_window *w = p->windows;
2309
2310 if (!w->offset && w->len == p->pack_size)
2311 return 1;
2312 }
2313
2314 /* Force the pack to open to prove its valid. */
2315 return !open_packed_git(p);
2316}
2317
2318static int fill_pack_entry(const unsigned char *sha1,
2319 struct pack_entry *e,
2320 struct packed_git *p)
2321{
2322 off_t offset;
2323
2324 if (p->num_bad_objects) {
2325 unsigned i;
2326 for (i = 0; i < p->num_bad_objects; i++)
2327 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2328 return 0;
2329 }
2330
2331 offset = find_pack_entry_one(sha1, p);
2332 if (!offset)
2333 return 0;
2334
2335 /*
2336 * We are about to tell the caller where they can locate the
2337 * requested object. We better make sure the packfile is
2338 * still here and can be accessed before supplying that
2339 * answer, as it may have been deleted since the index was
2340 * loaded!
2341 */
2342 if (!is_pack_valid(p)) {
2343 warning("packfile %s cannot be accessed", p->pack_name);
2344 return 0;
2345 }
2346 e->offset = offset;
2347 e->p = p;
2348 hashcpy(e->sha1, sha1);
2349 return 1;
2350}
2351
2352static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2353{
2354 struct packed_git *p;
2355
2356 prepare_packed_git();
2357 if (!packed_git)
2358 return 0;
2359
2360 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2361 return 1;
2362
2363 for (p = packed_git; p; p = p->next) {
2364 if (p == last_found_pack || !fill_pack_entry(sha1, e, p))
2365 continue;
2366
2367 last_found_pack = p;
2368 return 1;
2369 }
2370 return 0;
2371}
2372
2373struct packed_git *find_sha1_pack(const unsigned char *sha1,
2374 struct packed_git *packs)
2375{
2376 struct packed_git *p;
2377
2378 for (p = packs; p; p = p->next) {
2379 if (find_pack_entry_one(sha1, p))
2380 return p;
2381 }
2382 return NULL;
2383
2384}
2385
2386static int sha1_loose_object_info(const unsigned char *sha1,
2387 struct object_info *oi)
2388{
2389 int status;
2390 unsigned long mapsize, size;
2391 void *map;
2392 git_zstream stream;
2393 char hdr[32];
2394
2395 /*
2396 * If we don't care about type or size, then we don't
2397 * need to look inside the object at all.
2398 */
2399 if (!oi->typep && !oi->sizep) {
2400 if (oi->disk_sizep) {
2401 struct stat st;
2402 if (stat_sha1_file(sha1, &st) < 0)
2403 return -1;
2404 *oi->disk_sizep = st.st_size;
2405 }
2406 return 0;
2407 }
2408
2409 map = map_sha1_file(sha1, &mapsize);
2410 if (!map)
2411 return error("unable to find %s", sha1_to_hex(sha1));
2412 if (oi->disk_sizep)
2413 *oi->disk_sizep = mapsize;
2414 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2415 status = error("unable to unpack %s header",
2416 sha1_to_hex(sha1));
2417 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2418 status = error("unable to parse %s header", sha1_to_hex(sha1));
2419 else if (oi->sizep)
2420 *oi->sizep = size;
2421 git_inflate_end(&stream);
2422 munmap(map, mapsize);
2423 if (oi->typep)
2424 *oi->typep = status;
2425 return 0;
2426}
2427
2428/* returns enum object_type or negative */
2429int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
2430{
2431 struct cached_object *co;
2432 struct pack_entry e;
2433 int rtype;
2434
2435 co = find_cached_object(sha1);
2436 if (co) {
2437 if (oi->typep)
2438 *(oi->typep) = co->type;
2439 if (oi->sizep)
2440 *(oi->sizep) = co->size;
2441 if (oi->disk_sizep)
2442 *(oi->disk_sizep) = 0;
2443 oi->whence = OI_CACHED;
2444 return 0;
2445 }
2446
2447 if (!find_pack_entry(sha1, &e)) {
2448 /* Most likely it's a loose object. */
2449 if (!sha1_loose_object_info(sha1, oi)) {
2450 oi->whence = OI_LOOSE;
2451 return 0;
2452 }
2453
2454 /* Not a loose object; someone else may have just packed it. */
2455 reprepare_packed_git();
2456 if (!find_pack_entry(sha1, &e))
2457 return -1;
2458 }
2459
2460 rtype = packed_object_info(e.p, e.offset, oi);
2461 if (rtype < 0) {
2462 mark_bad_packed_object(e.p, sha1);
2463 return sha1_object_info_extended(sha1, oi);
2464 } else if (in_delta_base_cache(e.p, e.offset)) {
2465 oi->whence = OI_DBCACHED;
2466 } else {
2467 oi->whence = OI_PACKED;
2468 oi->u.packed.offset = e.offset;
2469 oi->u.packed.pack = e.p;
2470 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2471 rtype == OBJ_OFS_DELTA);
2472 }
2473
2474 return 0;
2475}
2476
2477int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2478{
2479 enum object_type type;
2480 struct object_info oi = {0};
2481
2482 oi.typep = &type;
2483 oi.sizep = sizep;
2484 if (sha1_object_info_extended(sha1, &oi) < 0)
2485 return -1;
2486 return type;
2487}
2488
2489static void *read_packed_sha1(const unsigned char *sha1,
2490 enum object_type *type, unsigned long *size)
2491{
2492 struct pack_entry e;
2493 void *data;
2494
2495 if (!find_pack_entry(sha1, &e))
2496 return NULL;
2497 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2498 if (!data) {
2499 /*
2500 * We're probably in deep shit, but let's try to fetch
2501 * the required object anyway from another pack or loose.
2502 * This should happen only in the presence of a corrupted
2503 * pack, and is better than failing outright.
2504 */
2505 error("failed to read object %s at offset %"PRIuMAX" from %s",
2506 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2507 mark_bad_packed_object(e.p, sha1);
2508 data = read_object(sha1, type, size);
2509 }
2510 return data;
2511}
2512
2513int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2514 unsigned char *sha1)
2515{
2516 struct cached_object *co;
2517
2518 hash_sha1_file(buf, len, typename(type), sha1);
2519 if (has_sha1_file(sha1) || find_cached_object(sha1))
2520 return 0;
2521 if (cached_object_alloc <= cached_object_nr) {
2522 cached_object_alloc = alloc_nr(cached_object_alloc);
2523 cached_objects = xrealloc(cached_objects,
2524 sizeof(*cached_objects) *
2525 cached_object_alloc);
2526 }
2527 co = &cached_objects[cached_object_nr++];
2528 co->size = len;
2529 co->type = type;
2530 co->buf = xmalloc(len);
2531 memcpy(co->buf, buf, len);
2532 hashcpy(co->sha1, sha1);
2533 return 0;
2534}
2535
2536static void *read_object(const unsigned char *sha1, enum object_type *type,
2537 unsigned long *size)
2538{
2539 unsigned long mapsize;
2540 void *map, *buf;
2541 struct cached_object *co;
2542
2543 co = find_cached_object(sha1);
2544 if (co) {
2545 *type = co->type;
2546 *size = co->size;
2547 return xmemdupz(co->buf, co->size);
2548 }
2549
2550 buf = read_packed_sha1(sha1, type, size);
2551 if (buf)
2552 return buf;
2553 map = map_sha1_file(sha1, &mapsize);
2554 if (map) {
2555 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2556 munmap(map, mapsize);
2557 return buf;
2558 }
2559 reprepare_packed_git();
2560 return read_packed_sha1(sha1, type, size);
2561}
2562
2563/*
2564 * This function dies on corrupt objects; the callers who want to
2565 * deal with them should arrange to call read_object() and give error
2566 * messages themselves.
2567 */
2568void *read_sha1_file_extended(const unsigned char *sha1,
2569 enum object_type *type,
2570 unsigned long *size,
2571 unsigned flag)
2572{
2573 void *data;
2574 char *path;
2575 const struct packed_git *p;
2576 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2577 ? lookup_replace_object(sha1) : sha1;
2578
2579 errno = 0;
2580 data = read_object(repl, type, size);
2581 if (data)
2582 return data;
2583
2584 if (errno && errno != ENOENT)
2585 die_errno("failed to read object %s", sha1_to_hex(sha1));
2586
2587 /* die if we replaced an object with one that does not exist */
2588 if (repl != sha1)
2589 die("replacement %s not found for %s",
2590 sha1_to_hex(repl), sha1_to_hex(sha1));
2591
2592 if (has_loose_object(repl)) {
2593 path = sha1_file_name(sha1);
2594 die("loose object %s (stored in %s) is corrupt",
2595 sha1_to_hex(repl), path);
2596 }
2597
2598 if ((p = has_packed_and_bad(repl)) != NULL)
2599 die("packed object %s (stored in %s) is corrupt",
2600 sha1_to_hex(repl), p->pack_name);
2601
2602 return NULL;
2603}
2604
2605void *read_object_with_reference(const unsigned char *sha1,
2606 const char *required_type_name,
2607 unsigned long *size,
2608 unsigned char *actual_sha1_return)
2609{
2610 enum object_type type, required_type;
2611 void *buffer;
2612 unsigned long isize;
2613 unsigned char actual_sha1[20];
2614
2615 required_type = type_from_string(required_type_name);
2616 hashcpy(actual_sha1, sha1);
2617 while (1) {
2618 int ref_length = -1;
2619 const char *ref_type = NULL;
2620
2621 buffer = read_sha1_file(actual_sha1, &type, &isize);
2622 if (!buffer)
2623 return NULL;
2624 if (type == required_type) {
2625 *size = isize;
2626 if (actual_sha1_return)
2627 hashcpy(actual_sha1_return, actual_sha1);
2628 return buffer;
2629 }
2630 /* Handle references */
2631 else if (type == OBJ_COMMIT)
2632 ref_type = "tree ";
2633 else if (type == OBJ_TAG)
2634 ref_type = "object ";
2635 else {
2636 free(buffer);
2637 return NULL;
2638 }
2639 ref_length = strlen(ref_type);
2640
2641 if (ref_length + 40 > isize ||
2642 memcmp(buffer, ref_type, ref_length) ||
2643 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2644 free(buffer);
2645 return NULL;
2646 }
2647 free(buffer);
2648 /* Now we have the ID of the referred-to object in
2649 * actual_sha1. Check again. */
2650 }
2651}
2652
2653static void write_sha1_file_prepare(const void *buf, unsigned long len,
2654 const char *type, unsigned char *sha1,
2655 char *hdr, int *hdrlen)
2656{
2657 git_SHA_CTX c;
2658
2659 /* Generate the header */
2660 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2661
2662 /* Sha1.. */
2663 git_SHA1_Init(&c);
2664 git_SHA1_Update(&c, hdr, *hdrlen);
2665 git_SHA1_Update(&c, buf, len);
2666 git_SHA1_Final(sha1, &c);
2667}
2668
2669/*
2670 * Move the just written object into its final resting place.
2671 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2672 * "moving" is only a part of what it does, when no patch between
2673 * master to pu changes the call sites of this function.
2674 */
2675int move_temp_to_file(const char *tmpfile, const char *filename)
2676{
2677 int ret = 0;
2678
2679 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2680 goto try_rename;
2681 else if (link(tmpfile, filename))
2682 ret = errno;
2683
2684 /*
2685 * Coda hack - coda doesn't like cross-directory links,
2686 * so we fall back to a rename, which will mean that it
2687 * won't be able to check collisions, but that's not a
2688 * big deal.
2689 *
2690 * The same holds for FAT formatted media.
2691 *
2692 * When this succeeds, we just return. We have nothing
2693 * left to unlink.
2694 */
2695 if (ret && ret != EEXIST) {
2696 try_rename:
2697 if (!rename(tmpfile, filename))
2698 goto out;
2699 ret = errno;
2700 }
2701 unlink_or_warn(tmpfile);
2702 if (ret) {
2703 if (ret != EEXIST) {
2704 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2705 }
2706 /* FIXME!!! Collision check here ? */
2707 }
2708
2709out:
2710 if (adjust_shared_perm(filename))
2711 return error("unable to set permission to '%s'", filename);
2712 return 0;
2713}
2714
2715static int write_buffer(int fd, const void *buf, size_t len)
2716{
2717 if (write_in_full(fd, buf, len) < 0)
2718 return error("file write error (%s)", strerror(errno));
2719 return 0;
2720}
2721
2722int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2723 unsigned char *sha1)
2724{
2725 char hdr[32];
2726 int hdrlen;
2727 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2728 return 0;
2729}
2730
2731/* Finalize a file on disk, and close it. */
2732static void close_sha1_file(int fd)
2733{
2734 if (fsync_object_files)
2735 fsync_or_die(fd, "sha1 file");
2736 if (close(fd) != 0)
2737 die_errno("error when closing sha1 file");
2738}
2739
2740/* Size of directory component, including the ending '/' */
2741static inline int directory_size(const char *filename)
2742{
2743 const char *s = strrchr(filename, '/');
2744 if (!s)
2745 return 0;
2746 return s - filename + 1;
2747}
2748
2749/*
2750 * This creates a temporary file in the same directory as the final
2751 * 'filename'
2752 *
2753 * We want to avoid cross-directory filename renames, because those
2754 * can have problems on various filesystems (FAT, NFS, Coda).
2755 */
2756static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2757{
2758 int fd, dirlen = directory_size(filename);
2759
2760 if (dirlen + 20 > bufsiz) {
2761 errno = ENAMETOOLONG;
2762 return -1;
2763 }
2764 memcpy(buffer, filename, dirlen);
2765 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2766 fd = git_mkstemp_mode(buffer, 0444);
2767 if (fd < 0 && dirlen && errno == ENOENT) {
2768 /* Make sure the directory exists */
2769 memcpy(buffer, filename, dirlen);
2770 buffer[dirlen-1] = 0;
2771 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2772 return -1;
2773
2774 /* Try again */
2775 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2776 fd = git_mkstemp_mode(buffer, 0444);
2777 }
2778 return fd;
2779}
2780
2781static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2782 const void *buf, unsigned long len, time_t mtime)
2783{
2784 int fd, ret;
2785 unsigned char compressed[4096];
2786 git_zstream stream;
2787 git_SHA_CTX c;
2788 unsigned char parano_sha1[20];
2789 char *filename;
2790 static char tmp_file[PATH_MAX];
2791
2792 filename = sha1_file_name(sha1);
2793 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
2794 if (fd < 0) {
2795 if (errno == EACCES)
2796 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2797 else
2798 return error("unable to create temporary file: %s", strerror(errno));
2799 }
2800
2801 /* Set it up */
2802 memset(&stream, 0, sizeof(stream));
2803 git_deflate_init(&stream, zlib_compression_level);
2804 stream.next_out = compressed;
2805 stream.avail_out = sizeof(compressed);
2806 git_SHA1_Init(&c);
2807
2808 /* First header.. */
2809 stream.next_in = (unsigned char *)hdr;
2810 stream.avail_in = hdrlen;
2811 while (git_deflate(&stream, 0) == Z_OK)
2812 ; /* nothing */
2813 git_SHA1_Update(&c, hdr, hdrlen);
2814
2815 /* Then the data itself.. */
2816 stream.next_in = (void *)buf;
2817 stream.avail_in = len;
2818 do {
2819 unsigned char *in0 = stream.next_in;
2820 ret = git_deflate(&stream, Z_FINISH);
2821 git_SHA1_Update(&c, in0, stream.next_in - in0);
2822 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2823 die("unable to write sha1 file");
2824 stream.next_out = compressed;
2825 stream.avail_out = sizeof(compressed);
2826 } while (ret == Z_OK);
2827
2828 if (ret != Z_STREAM_END)
2829 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2830 ret = git_deflate_end_gently(&stream);
2831 if (ret != Z_OK)
2832 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2833 git_SHA1_Final(parano_sha1, &c);
2834 if (hashcmp(sha1, parano_sha1) != 0)
2835 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2836
2837 close_sha1_file(fd);
2838
2839 if (mtime) {
2840 struct utimbuf utb;
2841 utb.actime = mtime;
2842 utb.modtime = mtime;
2843 if (utime(tmp_file, &utb) < 0)
2844 warning("failed utime() on %s: %s",
2845 tmp_file, strerror(errno));
2846 }
2847
2848 return move_temp_to_file(tmp_file, filename);
2849}
2850
2851int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2852{
2853 unsigned char sha1[20];
2854 char hdr[32];
2855 int hdrlen;
2856
2857 /* Normally if we have it in the pack then we do not bother writing
2858 * it out into .git/objects/??/?{38} file.
2859 */
2860 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2861 if (returnsha1)
2862 hashcpy(returnsha1, sha1);
2863 if (has_sha1_file(sha1))
2864 return 0;
2865 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2866}
2867
2868int force_object_loose(const unsigned char *sha1, time_t mtime)
2869{
2870 void *buf;
2871 unsigned long len;
2872 enum object_type type;
2873 char hdr[32];
2874 int hdrlen;
2875 int ret;
2876
2877 if (has_loose_object(sha1))
2878 return 0;
2879 buf = read_packed_sha1(sha1, &type, &len);
2880 if (!buf)
2881 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2882 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2883 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2884 free(buf);
2885
2886 return ret;
2887}
2888
2889int has_pack_index(const unsigned char *sha1)
2890{
2891 struct stat st;
2892 if (stat(sha1_pack_index_name(sha1), &st))
2893 return 0;
2894 return 1;
2895}
2896
2897int has_sha1_pack(const unsigned char *sha1)
2898{
2899 struct pack_entry e;
2900 return find_pack_entry(sha1, &e);
2901}
2902
2903int has_sha1_file(const unsigned char *sha1)
2904{
2905 struct pack_entry e;
2906
2907 if (find_pack_entry(sha1, &e))
2908 return 1;
2909 return has_loose_object(sha1);
2910}
2911
2912static void check_tree(const void *buf, size_t size)
2913{
2914 struct tree_desc desc;
2915 struct name_entry entry;
2916
2917 init_tree_desc(&desc, buf, size);
2918 while (tree_entry(&desc, &entry))
2919 /* do nothing
2920 * tree_entry() will die() on malformed entries */
2921 ;
2922}
2923
2924static void check_commit(const void *buf, size_t size)
2925{
2926 struct commit c;
2927 memset(&c, 0, sizeof(c));
2928 if (parse_commit_buffer(&c, buf, size))
2929 die("corrupt commit");
2930}
2931
2932static void check_tag(const void *buf, size_t size)
2933{
2934 struct tag t;
2935 memset(&t, 0, sizeof(t));
2936 if (parse_tag_buffer(&t, buf, size))
2937 die("corrupt tag");
2938}
2939
2940static int index_mem(unsigned char *sha1, void *buf, size_t size,
2941 enum object_type type,
2942 const char *path, unsigned flags)
2943{
2944 int ret, re_allocated = 0;
2945 int write_object = flags & HASH_WRITE_OBJECT;
2946
2947 if (!type)
2948 type = OBJ_BLOB;
2949
2950 /*
2951 * Convert blobs to git internal format
2952 */
2953 if ((type == OBJ_BLOB) && path) {
2954 struct strbuf nbuf = STRBUF_INIT;
2955 if (convert_to_git(path, buf, size, &nbuf,
2956 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
2957 buf = strbuf_detach(&nbuf, &size);
2958 re_allocated = 1;
2959 }
2960 }
2961 if (flags & HASH_FORMAT_CHECK) {
2962 if (type == OBJ_TREE)
2963 check_tree(buf, size);
2964 if (type == OBJ_COMMIT)
2965 check_commit(buf, size);
2966 if (type == OBJ_TAG)
2967 check_tag(buf, size);
2968 }
2969
2970 if (write_object)
2971 ret = write_sha1_file(buf, size, typename(type), sha1);
2972 else
2973 ret = hash_sha1_file(buf, size, typename(type), sha1);
2974 if (re_allocated)
2975 free(buf);
2976 return ret;
2977}
2978
2979static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2980 const char *path, unsigned flags)
2981{
2982 struct strbuf sbuf = STRBUF_INIT;
2983 int ret;
2984
2985 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2986 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2987 else
2988 ret = -1;
2989 strbuf_release(&sbuf);
2990 return ret;
2991}
2992
2993#define SMALL_FILE_SIZE (32*1024)
2994
2995static int index_core(unsigned char *sha1, int fd, size_t size,
2996 enum object_type type, const char *path,
2997 unsigned flags)
2998{
2999 int ret;
3000
3001 if (!size) {
3002 ret = index_mem(sha1, NULL, size, type, path, flags);
3003 } else if (size <= SMALL_FILE_SIZE) {
3004 char *buf = xmalloc(size);
3005 if (size == read_in_full(fd, buf, size))
3006 ret = index_mem(sha1, buf, size, type, path, flags);
3007 else
3008 ret = error("short read %s", strerror(errno));
3009 free(buf);
3010 } else {
3011 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3012 ret = index_mem(sha1, buf, size, type, path, flags);
3013 munmap(buf, size);
3014 }
3015 return ret;
3016}
3017
3018/*
3019 * This creates one packfile per large blob unless bulk-checkin
3020 * machinery is "plugged".
3021 *
3022 * This also bypasses the usual "convert-to-git" dance, and that is on
3023 * purpose. We could write a streaming version of the converting
3024 * functions and insert that before feeding the data to fast-import
3025 * (or equivalent in-core API described above). However, that is
3026 * somewhat complicated, as we do not know the size of the filter
3027 * result, which we need to know beforehand when writing a git object.
3028 * Since the primary motivation for trying to stream from the working
3029 * tree file and to avoid mmaping it in core is to deal with large
3030 * binary blobs, they generally do not want to get any conversion, and
3031 * callers should avoid this code path when filters are requested.
3032 */
3033static int index_stream(unsigned char *sha1, int fd, size_t size,
3034 enum object_type type, const char *path,
3035 unsigned flags)
3036{
3037 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3038}
3039
3040int index_fd(unsigned char *sha1, int fd, struct stat *st,
3041 enum object_type type, const char *path, unsigned flags)
3042{
3043 int ret;
3044 size_t size = xsize_t(st->st_size);
3045
3046 if (!S_ISREG(st->st_mode))
3047 ret = index_pipe(sha1, fd, type, path, flags);
3048 else if (size <= big_file_threshold || type != OBJ_BLOB ||
3049 (path && would_convert_to_git(path, NULL, 0, 0)))
3050 ret = index_core(sha1, fd, size, type, path, flags);
3051 else
3052 ret = index_stream(sha1, fd, size, type, path, flags);
3053 close(fd);
3054 return ret;
3055}
3056
3057int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3058{
3059 int fd;
3060 struct strbuf sb = STRBUF_INIT;
3061
3062 switch (st->st_mode & S_IFMT) {
3063 case S_IFREG:
3064 fd = open(path, O_RDONLY);
3065 if (fd < 0)
3066 return error("open(\"%s\"): %s", path,
3067 strerror(errno));
3068 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3069 return error("%s: failed to insert into database",
3070 path);
3071 break;
3072 case S_IFLNK:
3073 if (strbuf_readlink(&sb, path, st->st_size)) {
3074 char *errstr = strerror(errno);
3075 return error("readlink(\"%s\"): %s", path,
3076 errstr);
3077 }
3078 if (!(flags & HASH_WRITE_OBJECT))
3079 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3080 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3081 return error("%s: failed to insert into database",
3082 path);
3083 strbuf_release(&sb);
3084 break;
3085 case S_IFDIR:
3086 return resolve_gitlink_ref(path, "HEAD", sha1);
3087 default:
3088 return error("%s: unsupported file type", path);
3089 }
3090 return 0;
3091}
3092
3093int read_pack_header(int fd, struct pack_header *header)
3094{
3095 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3096 /* "eof before pack header was fully read" */
3097 return PH_ERROR_EOF;
3098
3099 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3100 /* "protocol error (pack signature mismatch detected)" */
3101 return PH_ERROR_PACK_SIGNATURE;
3102 if (!pack_version_ok(header->hdr_version))
3103 /* "protocol error (pack version unsupported)" */
3104 return PH_ERROR_PROTOCOL;
3105 return 0;
3106}
3107
3108void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3109{
3110 enum object_type type = sha1_object_info(sha1, NULL);
3111 if (type < 0)
3112 die("%s is not a valid object", sha1_to_hex(sha1));
3113 if (type != expect)
3114 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3115 typename(expect));
3116}