129496250c3673c1f62a5da5db728703501ddf0f
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
39static const char *no_log_pack_access = "no_log_pack_access";
40static const char *log_pack_access;
41
42/*
43 * This is meant to hold a *small* number of objects that you would
44 * want read_sha1_file() to be able to return, but yet you do not want
45 * to write them into the object store (e.g. a browse-only
46 * application).
47 */
48static struct cached_object {
49 unsigned char sha1[20];
50 enum object_type type;
51 void *buf;
52 unsigned long size;
53} *cached_objects;
54static int cached_object_nr, cached_object_alloc;
55
56static struct cached_object empty_tree = {
57 EMPTY_TREE_SHA1_BIN_LITERAL,
58 OBJ_TREE,
59 "",
60 0
61};
62
63static struct packed_git *last_found_pack;
64
65static struct cached_object *find_cached_object(const unsigned char *sha1)
66{
67 int i;
68 struct cached_object *co = cached_objects;
69
70 for (i = 0; i < cached_object_nr; i++, co++) {
71 if (!hashcmp(co->sha1, sha1))
72 return co;
73 }
74 if (!hashcmp(sha1, empty_tree.sha1))
75 return &empty_tree;
76 return NULL;
77}
78
79int mkdir_in_gitdir(const char *path)
80{
81 if (mkdir(path, 0777)) {
82 int saved_errno = errno;
83 struct stat st;
84 struct strbuf sb = STRBUF_INIT;
85
86 if (errno != EEXIST)
87 return -1;
88 /*
89 * Are we looking at a path in a symlinked worktree
90 * whose original repository does not yet have it?
91 * e.g. .git/rr-cache pointing at its original
92 * repository in which the user hasn't performed any
93 * conflict resolution yet?
94 */
95 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
96 strbuf_readlink(&sb, path, st.st_size) ||
97 !is_absolute_path(sb.buf) ||
98 mkdir(sb.buf, 0777)) {
99 strbuf_release(&sb);
100 errno = saved_errno;
101 return -1;
102 }
103 strbuf_release(&sb);
104 }
105 return adjust_shared_perm(path);
106}
107
108int safe_create_leading_directories(char *path)
109{
110 char *pos = path + offset_1st_component(path);
111 struct stat st;
112
113 while (pos) {
114 pos = strchr(pos, '/');
115 if (!pos)
116 break;
117 while (*++pos == '/')
118 ;
119 if (!*pos)
120 break;
121 *--pos = '\0';
122 if (!stat(path, &st)) {
123 /* path exists */
124 if (!S_ISDIR(st.st_mode)) {
125 *pos = '/';
126 return -3;
127 }
128 }
129 else if (mkdir(path, 0777)) {
130 if (errno == EEXIST &&
131 !stat(path, &st) && S_ISDIR(st.st_mode)) {
132 ; /* somebody created it since we checked */
133 } else {
134 *pos = '/';
135 return -1;
136 }
137 }
138 else if (adjust_shared_perm(path)) {
139 *pos = '/';
140 return -2;
141 }
142 *pos++ = '/';
143 }
144 return 0;
145}
146
147int safe_create_leading_directories_const(const char *path)
148{
149 /* path points to cache entries, so xstrdup before messing with it */
150 char *buf = xstrdup(path);
151 int result = safe_create_leading_directories(buf);
152 free(buf);
153 return result;
154}
155
156static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
157{
158 int i;
159 for (i = 0; i < 20; i++) {
160 static char hex[] = "0123456789abcdef";
161 unsigned int val = sha1[i];
162 char *pos = pathbuf + i*2 + (i > 0);
163 *pos++ = hex[val >> 4];
164 *pos = hex[val & 0xf];
165 }
166}
167
168/*
169 * NOTE! This returns a statically allocated buffer, so you have to be
170 * careful about using it. Do an "xstrdup()" if you need to save the
171 * filename.
172 *
173 * Also note that this returns the location for creating. Reading
174 * SHA1 file can happen from any alternate directory listed in the
175 * DB_ENVIRONMENT environment variable if it is not found in
176 * the primary object database.
177 */
178char *sha1_file_name(const unsigned char *sha1)
179{
180 static char buf[PATH_MAX];
181 const char *objdir;
182 int len;
183
184 objdir = get_object_directory();
185 len = strlen(objdir);
186
187 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
188 if (len + 43 > PATH_MAX)
189 die("insanely long object directory %s", objdir);
190 memcpy(buf, objdir, len);
191 buf[len] = '/';
192 buf[len+3] = '/';
193 buf[len+42] = '\0';
194 fill_sha1_path(buf + len + 1, sha1);
195 return buf;
196}
197
198static char *sha1_get_pack_name(const unsigned char *sha1,
199 char **name, char **base, const char *which)
200{
201 static const char hex[] = "0123456789abcdef";
202 char *buf;
203 int i;
204
205 if (!*base) {
206 const char *sha1_file_directory = get_object_directory();
207 int len = strlen(sha1_file_directory);
208 *base = xmalloc(len + 60);
209 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
210 sha1_file_directory, which);
211 *name = *base + len + 11;
212 }
213
214 buf = *name;
215
216 for (i = 0; i < 20; i++) {
217 unsigned int val = *sha1++;
218 *buf++ = hex[val >> 4];
219 *buf++ = hex[val & 0xf];
220 }
221
222 return *base;
223}
224
225char *sha1_pack_name(const unsigned char *sha1)
226{
227 static char *name, *base;
228
229 return sha1_get_pack_name(sha1, &name, &base, "pack");
230}
231
232char *sha1_pack_index_name(const unsigned char *sha1)
233{
234 static char *name, *base;
235
236 return sha1_get_pack_name(sha1, &name, &base, "idx");
237}
238
239struct alternate_object_database *alt_odb_list;
240static struct alternate_object_database **alt_odb_tail;
241
242/*
243 * Prepare alternate object database registry.
244 *
245 * The variable alt_odb_list points at the list of struct
246 * alternate_object_database. The elements on this list come from
247 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
248 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
249 * whose contents is similar to that environment variable but can be
250 * LF separated. Its base points at a statically allocated buffer that
251 * contains "/the/directory/corresponding/to/.git/objects/...", while
252 * its name points just after the slash at the end of ".git/objects/"
253 * in the example above, and has enough space to hold 40-byte hex
254 * SHA1, an extra slash for the first level indirection, and the
255 * terminating NUL.
256 */
257static int link_alt_odb_entry(const char *entry, const char *relative_base, int depth)
258{
259 const char *objdir = get_object_directory();
260 struct alternate_object_database *ent;
261 struct alternate_object_database *alt;
262 int pfxlen, entlen;
263 struct strbuf pathbuf = STRBUF_INIT;
264
265 if (!is_absolute_path(entry) && relative_base) {
266 strbuf_addstr(&pathbuf, real_path(relative_base));
267 strbuf_addch(&pathbuf, '/');
268 }
269 strbuf_addstr(&pathbuf, entry);
270
271 normalize_path_copy(pathbuf.buf, pathbuf.buf);
272
273 pfxlen = strlen(pathbuf.buf);
274
275 /*
276 * The trailing slash after the directory name is given by
277 * this function at the end. Remove duplicates.
278 */
279 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
280 pfxlen -= 1;
281
282 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
283 ent = xmalloc(sizeof(*ent) + entlen);
284 memcpy(ent->base, pathbuf.buf, pfxlen);
285 strbuf_release(&pathbuf);
286
287 ent->name = ent->base + pfxlen + 1;
288 ent->base[pfxlen + 3] = '/';
289 ent->base[pfxlen] = ent->base[entlen-1] = 0;
290
291 /* Detect cases where alternate disappeared */
292 if (!is_directory(ent->base)) {
293 error("object directory %s does not exist; "
294 "check .git/objects/info/alternates.",
295 ent->base);
296 free(ent);
297 return -1;
298 }
299
300 /* Prevent the common mistake of listing the same
301 * thing twice, or object directory itself.
302 */
303 for (alt = alt_odb_list; alt; alt = alt->next) {
304 if (!memcmp(ent->base, alt->base, pfxlen)) {
305 free(ent);
306 return -1;
307 }
308 }
309 if (!strcmp(ent->base, objdir)) {
310 free(ent);
311 return -1;
312 }
313
314 /* add the alternate entry */
315 *alt_odb_tail = ent;
316 alt_odb_tail = &(ent->next);
317 ent->next = NULL;
318
319 /* recursively add alternates */
320 read_info_alternates(ent->base, depth + 1);
321
322 ent->base[pfxlen] = '/';
323
324 return 0;
325}
326
327static void link_alt_odb_entries(const char *alt, int len, int sep,
328 const char *relative_base, int depth)
329{
330 struct string_list entries = STRING_LIST_INIT_NODUP;
331 char *alt_copy;
332 int i;
333
334 if (depth > 5) {
335 error("%s: ignoring alternate object stores, nesting too deep.",
336 relative_base);
337 return;
338 }
339
340 alt_copy = xmemdupz(alt, len);
341 string_list_split_in_place(&entries, alt_copy, sep, -1);
342 for (i = 0; i < entries.nr; i++) {
343 const char *entry = entries.items[i].string;
344 if (entry[0] == '\0' || entry[0] == '#')
345 continue;
346 if (!is_absolute_path(entry) && depth) {
347 error("%s: ignoring relative alternate object store %s",
348 relative_base, entry);
349 } else {
350 link_alt_odb_entry(entry, relative_base, depth);
351 }
352 }
353 string_list_clear(&entries, 0);
354 free(alt_copy);
355}
356
357void read_info_alternates(const char * relative_base, int depth)
358{
359 char *map;
360 size_t mapsz;
361 struct stat st;
362 const char alt_file_name[] = "info/alternates";
363 /* Given that relative_base is no longer than PATH_MAX,
364 ensure that "path" has enough space to append "/", the
365 file name, "info/alternates", and a trailing NUL. */
366 char path[PATH_MAX + 1 + sizeof alt_file_name];
367 int fd;
368
369 sprintf(path, "%s/%s", relative_base, alt_file_name);
370 fd = git_open_noatime(path);
371 if (fd < 0)
372 return;
373 if (fstat(fd, &st) || (st.st_size == 0)) {
374 close(fd);
375 return;
376 }
377 mapsz = xsize_t(st.st_size);
378 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
379 close(fd);
380
381 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
382
383 munmap(map, mapsz);
384}
385
386void add_to_alternates_file(const char *reference)
387{
388 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
389 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
390 char *alt = mkpath("%s\n", reference);
391 write_or_die(fd, alt, strlen(alt));
392 if (commit_lock_file(lock))
393 die("could not close alternates file");
394 if (alt_odb_tail)
395 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
396}
397
398void foreach_alt_odb(alt_odb_fn fn, void *cb)
399{
400 struct alternate_object_database *ent;
401
402 prepare_alt_odb();
403 for (ent = alt_odb_list; ent; ent = ent->next)
404 if (fn(ent, cb))
405 return;
406}
407
408void prepare_alt_odb(void)
409{
410 const char *alt;
411
412 if (alt_odb_tail)
413 return;
414
415 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
416 if (!alt) alt = "";
417
418 alt_odb_tail = &alt_odb_list;
419 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
420
421 read_info_alternates(get_object_directory(), 0);
422}
423
424static int has_loose_object_local(const unsigned char *sha1)
425{
426 char *name = sha1_file_name(sha1);
427 return !access(name, F_OK);
428}
429
430int has_loose_object_nonlocal(const unsigned char *sha1)
431{
432 struct alternate_object_database *alt;
433 prepare_alt_odb();
434 for (alt = alt_odb_list; alt; alt = alt->next) {
435 fill_sha1_path(alt->name, sha1);
436 if (!access(alt->base, F_OK))
437 return 1;
438 }
439 return 0;
440}
441
442static int has_loose_object(const unsigned char *sha1)
443{
444 return has_loose_object_local(sha1) ||
445 has_loose_object_nonlocal(sha1);
446}
447
448static unsigned int pack_used_ctr;
449static unsigned int pack_mmap_calls;
450static unsigned int peak_pack_open_windows;
451static unsigned int pack_open_windows;
452static unsigned int pack_open_fds;
453static unsigned int pack_max_fds;
454static size_t peak_pack_mapped;
455static size_t pack_mapped;
456struct packed_git *packed_git;
457
458void pack_report(void)
459{
460 fprintf(stderr,
461 "pack_report: getpagesize() = %10" SZ_FMT "\n"
462 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
463 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
464 sz_fmt(getpagesize()),
465 sz_fmt(packed_git_window_size),
466 sz_fmt(packed_git_limit));
467 fprintf(stderr,
468 "pack_report: pack_used_ctr = %10u\n"
469 "pack_report: pack_mmap_calls = %10u\n"
470 "pack_report: pack_open_windows = %10u / %10u\n"
471 "pack_report: pack_mapped = "
472 "%10" SZ_FMT " / %10" SZ_FMT "\n",
473 pack_used_ctr,
474 pack_mmap_calls,
475 pack_open_windows, peak_pack_open_windows,
476 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
477}
478
479static int check_packed_git_idx(const char *path, struct packed_git *p)
480{
481 void *idx_map;
482 struct pack_idx_header *hdr;
483 size_t idx_size;
484 uint32_t version, nr, i, *index;
485 int fd = git_open_noatime(path);
486 struct stat st;
487
488 if (fd < 0)
489 return -1;
490 if (fstat(fd, &st)) {
491 close(fd);
492 return -1;
493 }
494 idx_size = xsize_t(st.st_size);
495 if (idx_size < 4 * 256 + 20 + 20) {
496 close(fd);
497 return error("index file %s is too small", path);
498 }
499 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
500 close(fd);
501
502 hdr = idx_map;
503 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
504 version = ntohl(hdr->idx_version);
505 if (version < 2 || version > 2) {
506 munmap(idx_map, idx_size);
507 return error("index file %s is version %"PRIu32
508 " and is not supported by this binary"
509 " (try upgrading GIT to a newer version)",
510 path, version);
511 }
512 } else
513 version = 1;
514
515 nr = 0;
516 index = idx_map;
517 if (version > 1)
518 index += 2; /* skip index header */
519 for (i = 0; i < 256; i++) {
520 uint32_t n = ntohl(index[i]);
521 if (n < nr) {
522 munmap(idx_map, idx_size);
523 return error("non-monotonic index %s", path);
524 }
525 nr = n;
526 }
527
528 if (version == 1) {
529 /*
530 * Total size:
531 * - 256 index entries 4 bytes each
532 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
533 * - 20-byte SHA1 of the packfile
534 * - 20-byte SHA1 file checksum
535 */
536 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
537 munmap(idx_map, idx_size);
538 return error("wrong index v1 file size in %s", path);
539 }
540 } else if (version == 2) {
541 /*
542 * Minimum size:
543 * - 8 bytes of header
544 * - 256 index entries 4 bytes each
545 * - 20-byte sha1 entry * nr
546 * - 4-byte crc entry * nr
547 * - 4-byte offset entry * nr
548 * - 20-byte SHA1 of the packfile
549 * - 20-byte SHA1 file checksum
550 * And after the 4-byte offset table might be a
551 * variable sized table containing 8-byte entries
552 * for offsets larger than 2^31.
553 */
554 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
555 unsigned long max_size = min_size;
556 if (nr)
557 max_size += (nr - 1)*8;
558 if (idx_size < min_size || idx_size > max_size) {
559 munmap(idx_map, idx_size);
560 return error("wrong index v2 file size in %s", path);
561 }
562 if (idx_size != min_size &&
563 /*
564 * make sure we can deal with large pack offsets.
565 * 31-bit signed offset won't be enough, neither
566 * 32-bit unsigned one will be.
567 */
568 (sizeof(off_t) <= 4)) {
569 munmap(idx_map, idx_size);
570 return error("pack too large for current definition of off_t in %s", path);
571 }
572 }
573
574 p->index_version = version;
575 p->index_data = idx_map;
576 p->index_size = idx_size;
577 p->num_objects = nr;
578 return 0;
579}
580
581int open_pack_index(struct packed_git *p)
582{
583 char *idx_name;
584 int ret;
585
586 if (p->index_data)
587 return 0;
588
589 idx_name = xstrdup(p->pack_name);
590 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
591 ret = check_packed_git_idx(idx_name, p);
592 free(idx_name);
593 return ret;
594}
595
596static void scan_windows(struct packed_git *p,
597 struct packed_git **lru_p,
598 struct pack_window **lru_w,
599 struct pack_window **lru_l)
600{
601 struct pack_window *w, *w_l;
602
603 for (w_l = NULL, w = p->windows; w; w = w->next) {
604 if (!w->inuse_cnt) {
605 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
606 *lru_p = p;
607 *lru_w = w;
608 *lru_l = w_l;
609 }
610 }
611 w_l = w;
612 }
613}
614
615static int unuse_one_window(struct packed_git *current)
616{
617 struct packed_git *p, *lru_p = NULL;
618 struct pack_window *lru_w = NULL, *lru_l = NULL;
619
620 if (current)
621 scan_windows(current, &lru_p, &lru_w, &lru_l);
622 for (p = packed_git; p; p = p->next)
623 scan_windows(p, &lru_p, &lru_w, &lru_l);
624 if (lru_p) {
625 munmap(lru_w->base, lru_w->len);
626 pack_mapped -= lru_w->len;
627 if (lru_l)
628 lru_l->next = lru_w->next;
629 else
630 lru_p->windows = lru_w->next;
631 free(lru_w);
632 pack_open_windows--;
633 return 1;
634 }
635 return 0;
636}
637
638void release_pack_memory(size_t need)
639{
640 size_t cur = pack_mapped;
641 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
642 ; /* nothing */
643}
644
645void *xmmap(void *start, size_t length,
646 int prot, int flags, int fd, off_t offset)
647{
648 void *ret = mmap(start, length, prot, flags, fd, offset);
649 if (ret == MAP_FAILED) {
650 if (!length)
651 return NULL;
652 release_pack_memory(length);
653 ret = mmap(start, length, prot, flags, fd, offset);
654 if (ret == MAP_FAILED)
655 die_errno("Out of memory? mmap failed");
656 }
657 return ret;
658}
659
660void close_pack_windows(struct packed_git *p)
661{
662 while (p->windows) {
663 struct pack_window *w = p->windows;
664
665 if (w->inuse_cnt)
666 die("pack '%s' still has open windows to it",
667 p->pack_name);
668 munmap(w->base, w->len);
669 pack_mapped -= w->len;
670 pack_open_windows--;
671 p->windows = w->next;
672 free(w);
673 }
674}
675
676/*
677 * The LRU pack is the one with the oldest MRU window, preferring packs
678 * with no used windows, or the oldest mtime if it has no windows allocated.
679 */
680static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
681{
682 struct pack_window *w, *this_mru_w;
683 int has_windows_inuse = 0;
684
685 /*
686 * Reject this pack if it has windows and the previously selected
687 * one does not. If this pack does not have windows, reject
688 * it if the pack file is newer than the previously selected one.
689 */
690 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
691 return;
692
693 for (w = this_mru_w = p->windows; w; w = w->next) {
694 /*
695 * Reject this pack if any of its windows are in use,
696 * but the previously selected pack did not have any
697 * inuse windows. Otherwise, record that this pack
698 * has windows in use.
699 */
700 if (w->inuse_cnt) {
701 if (*accept_windows_inuse)
702 has_windows_inuse = 1;
703 else
704 return;
705 }
706
707 if (w->last_used > this_mru_w->last_used)
708 this_mru_w = w;
709
710 /*
711 * Reject this pack if it has windows that have been
712 * used more recently than the previously selected pack.
713 * If the previously selected pack had windows inuse and
714 * we have not encountered a window in this pack that is
715 * inuse, skip this check since we prefer a pack with no
716 * inuse windows to one that has inuse windows.
717 */
718 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
719 this_mru_w->last_used > (*mru_w)->last_used)
720 return;
721 }
722
723 /*
724 * Select this pack.
725 */
726 *mru_w = this_mru_w;
727 *lru_p = p;
728 *accept_windows_inuse = has_windows_inuse;
729}
730
731static int close_one_pack(void)
732{
733 struct packed_git *p, *lru_p = NULL;
734 struct pack_window *mru_w = NULL;
735 int accept_windows_inuse = 1;
736
737 for (p = packed_git; p; p = p->next) {
738 if (p->pack_fd == -1)
739 continue;
740 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
741 }
742
743 if (lru_p) {
744 close(lru_p->pack_fd);
745 pack_open_fds--;
746 lru_p->pack_fd = -1;
747 return 1;
748 }
749
750 return 0;
751}
752
753void unuse_pack(struct pack_window **w_cursor)
754{
755 struct pack_window *w = *w_cursor;
756 if (w) {
757 w->inuse_cnt--;
758 *w_cursor = NULL;
759 }
760}
761
762void close_pack_index(struct packed_git *p)
763{
764 if (p->index_data) {
765 munmap((void *)p->index_data, p->index_size);
766 p->index_data = NULL;
767 }
768}
769
770/*
771 * This is used by git-repack in case a newly created pack happens to
772 * contain the same set of objects as an existing one. In that case
773 * the resulting file might be different even if its name would be the
774 * same. It is best to close any reference to the old pack before it is
775 * replaced on disk. Of course no index pointers nor windows for given pack
776 * must subsist at this point. If ever objects from this pack are requested
777 * again, the new version of the pack will be reinitialized through
778 * reprepare_packed_git().
779 */
780void free_pack_by_name(const char *pack_name)
781{
782 struct packed_git *p, **pp = &packed_git;
783
784 while (*pp) {
785 p = *pp;
786 if (strcmp(pack_name, p->pack_name) == 0) {
787 clear_delta_base_cache();
788 close_pack_windows(p);
789 if (p->pack_fd != -1) {
790 close(p->pack_fd);
791 pack_open_fds--;
792 }
793 close_pack_index(p);
794 free(p->bad_object_sha1);
795 *pp = p->next;
796 if (last_found_pack == p)
797 last_found_pack = NULL;
798 free(p);
799 return;
800 }
801 pp = &p->next;
802 }
803}
804
805static unsigned int get_max_fd_limit(void)
806{
807#ifdef RLIMIT_NOFILE
808 struct rlimit lim;
809
810 if (getrlimit(RLIMIT_NOFILE, &lim))
811 die_errno("cannot get RLIMIT_NOFILE");
812
813 return lim.rlim_cur;
814#elif defined(_SC_OPEN_MAX)
815 return sysconf(_SC_OPEN_MAX);
816#elif defined(OPEN_MAX)
817 return OPEN_MAX;
818#else
819 return 1; /* see the caller ;-) */
820#endif
821}
822
823/*
824 * Do not call this directly as this leaks p->pack_fd on error return;
825 * call open_packed_git() instead.
826 */
827static int open_packed_git_1(struct packed_git *p)
828{
829 struct stat st;
830 struct pack_header hdr;
831 unsigned char sha1[20];
832 unsigned char *idx_sha1;
833 long fd_flag;
834
835 if (!p->index_data && open_pack_index(p))
836 return error("packfile %s index unavailable", p->pack_name);
837
838 if (!pack_max_fds) {
839 unsigned int max_fds = get_max_fd_limit();
840
841 /* Save 3 for stdin/stdout/stderr, 22 for work */
842 if (25 < max_fds)
843 pack_max_fds = max_fds - 25;
844 else
845 pack_max_fds = 1;
846 }
847
848 while (pack_max_fds <= pack_open_fds && close_one_pack())
849 ; /* nothing */
850
851 p->pack_fd = git_open_noatime(p->pack_name);
852 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
853 return -1;
854 pack_open_fds++;
855
856 /* If we created the struct before we had the pack we lack size. */
857 if (!p->pack_size) {
858 if (!S_ISREG(st.st_mode))
859 return error("packfile %s not a regular file", p->pack_name);
860 p->pack_size = st.st_size;
861 } else if (p->pack_size != st.st_size)
862 return error("packfile %s size changed", p->pack_name);
863
864 /* We leave these file descriptors open with sliding mmap;
865 * there is no point keeping them open across exec(), though.
866 */
867 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
868 if (fd_flag < 0)
869 return error("cannot determine file descriptor flags");
870 fd_flag |= FD_CLOEXEC;
871 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
872 return error("cannot set FD_CLOEXEC");
873
874 /* Verify we recognize this pack file format. */
875 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
876 return error("file %s is far too short to be a packfile", p->pack_name);
877 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
878 return error("file %s is not a GIT packfile", p->pack_name);
879 if (!pack_version_ok(hdr.hdr_version))
880 return error("packfile %s is version %"PRIu32" and not"
881 " supported (try upgrading GIT to a newer version)",
882 p->pack_name, ntohl(hdr.hdr_version));
883
884 /* Verify the pack matches its index. */
885 if (p->num_objects != ntohl(hdr.hdr_entries))
886 return error("packfile %s claims to have %"PRIu32" objects"
887 " while index indicates %"PRIu32" objects",
888 p->pack_name, ntohl(hdr.hdr_entries),
889 p->num_objects);
890 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
891 return error("end of packfile %s is unavailable", p->pack_name);
892 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
893 return error("packfile %s signature is unavailable", p->pack_name);
894 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
895 if (hashcmp(sha1, idx_sha1))
896 return error("packfile %s does not match index", p->pack_name);
897 return 0;
898}
899
900static int open_packed_git(struct packed_git *p)
901{
902 if (!open_packed_git_1(p))
903 return 0;
904 if (p->pack_fd != -1) {
905 close(p->pack_fd);
906 pack_open_fds--;
907 p->pack_fd = -1;
908 }
909 return -1;
910}
911
912static int in_window(struct pack_window *win, off_t offset)
913{
914 /* We must promise at least 20 bytes (one hash) after the
915 * offset is available from this window, otherwise the offset
916 * is not actually in this window and a different window (which
917 * has that one hash excess) must be used. This is to support
918 * the object header and delta base parsing routines below.
919 */
920 off_t win_off = win->offset;
921 return win_off <= offset
922 && (offset + 20) <= (win_off + win->len);
923}
924
925unsigned char *use_pack(struct packed_git *p,
926 struct pack_window **w_cursor,
927 off_t offset,
928 unsigned long *left)
929{
930 struct pack_window *win = *w_cursor;
931
932 /* Since packfiles end in a hash of their content and it's
933 * pointless to ask for an offset into the middle of that
934 * hash, and the in_window function above wouldn't match
935 * don't allow an offset too close to the end of the file.
936 */
937 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
938 die("packfile %s cannot be accessed", p->pack_name);
939 if (offset > (p->pack_size - 20))
940 die("offset beyond end of packfile (truncated pack?)");
941
942 if (!win || !in_window(win, offset)) {
943 if (win)
944 win->inuse_cnt--;
945 for (win = p->windows; win; win = win->next) {
946 if (in_window(win, offset))
947 break;
948 }
949 if (!win) {
950 size_t window_align = packed_git_window_size / 2;
951 off_t len;
952
953 if (p->pack_fd == -1 && open_packed_git(p))
954 die("packfile %s cannot be accessed", p->pack_name);
955
956 win = xcalloc(1, sizeof(*win));
957 win->offset = (offset / window_align) * window_align;
958 len = p->pack_size - win->offset;
959 if (len > packed_git_window_size)
960 len = packed_git_window_size;
961 win->len = (size_t)len;
962 pack_mapped += win->len;
963 while (packed_git_limit < pack_mapped
964 && unuse_one_window(p))
965 ; /* nothing */
966 win->base = xmmap(NULL, win->len,
967 PROT_READ, MAP_PRIVATE,
968 p->pack_fd, win->offset);
969 if (win->base == MAP_FAILED)
970 die("packfile %s cannot be mapped: %s",
971 p->pack_name,
972 strerror(errno));
973 if (!win->offset && win->len == p->pack_size
974 && !p->do_not_close) {
975 close(p->pack_fd);
976 pack_open_fds--;
977 p->pack_fd = -1;
978 }
979 pack_mmap_calls++;
980 pack_open_windows++;
981 if (pack_mapped > peak_pack_mapped)
982 peak_pack_mapped = pack_mapped;
983 if (pack_open_windows > peak_pack_open_windows)
984 peak_pack_open_windows = pack_open_windows;
985 win->next = p->windows;
986 p->windows = win;
987 }
988 }
989 if (win != *w_cursor) {
990 win->last_used = pack_used_ctr++;
991 win->inuse_cnt++;
992 *w_cursor = win;
993 }
994 offset -= win->offset;
995 if (left)
996 *left = win->len - xsize_t(offset);
997 return win->base + offset;
998}
999
1000static struct packed_git *alloc_packed_git(int extra)
1001{
1002 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1003 memset(p, 0, sizeof(*p));
1004 p->pack_fd = -1;
1005 return p;
1006}
1007
1008static void try_to_free_pack_memory(size_t size)
1009{
1010 release_pack_memory(size);
1011}
1012
1013struct packed_git *add_packed_git(const char *path, int path_len, int local)
1014{
1015 static int have_set_try_to_free_routine;
1016 struct stat st;
1017 struct packed_git *p = alloc_packed_git(path_len + 2);
1018
1019 if (!have_set_try_to_free_routine) {
1020 have_set_try_to_free_routine = 1;
1021 set_try_to_free_routine(try_to_free_pack_memory);
1022 }
1023
1024 /*
1025 * Make sure a corresponding .pack file exists and that
1026 * the index looks sane.
1027 */
1028 path_len -= strlen(".idx");
1029 if (path_len < 1) {
1030 free(p);
1031 return NULL;
1032 }
1033 memcpy(p->pack_name, path, path_len);
1034
1035 strcpy(p->pack_name + path_len, ".keep");
1036 if (!access(p->pack_name, F_OK))
1037 p->pack_keep = 1;
1038
1039 strcpy(p->pack_name + path_len, ".pack");
1040 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1041 free(p);
1042 return NULL;
1043 }
1044
1045 /* ok, it looks sane as far as we can check without
1046 * actually mapping the pack file.
1047 */
1048 p->pack_size = st.st_size;
1049 p->pack_local = local;
1050 p->mtime = st.st_mtime;
1051 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1052 hashclr(p->sha1);
1053 return p;
1054}
1055
1056struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1057{
1058 const char *path = sha1_pack_name(sha1);
1059 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
1060
1061 strcpy(p->pack_name, path);
1062 hashcpy(p->sha1, sha1);
1063 if (check_packed_git_idx(idx_path, p)) {
1064 free(p);
1065 return NULL;
1066 }
1067
1068 return p;
1069}
1070
1071void install_packed_git(struct packed_git *pack)
1072{
1073 if (pack->pack_fd != -1)
1074 pack_open_fds++;
1075
1076 pack->next = packed_git;
1077 packed_git = pack;
1078}
1079
1080void (*report_garbage)(const char *desc, const char *path);
1081
1082static void report_helper(const struct string_list *list,
1083 int seen_bits, int first, int last)
1084{
1085 const char *msg;
1086 switch (seen_bits) {
1087 case 0:
1088 msg = "no corresponding .idx nor .pack";
1089 break;
1090 case 1:
1091 msg = "no corresponding .idx";
1092 break;
1093 case 2:
1094 msg = "no corresponding .pack";
1095 break;
1096 default:
1097 return;
1098 }
1099 for (; first < last; first++)
1100 report_garbage(msg, list->items[first].string);
1101}
1102
1103static void report_pack_garbage(struct string_list *list)
1104{
1105 int i, baselen = -1, first = 0, seen_bits = 0;
1106
1107 if (!report_garbage)
1108 return;
1109
1110 sort_string_list(list);
1111
1112 for (i = 0; i < list->nr; i++) {
1113 const char *path = list->items[i].string;
1114 if (baselen != -1 &&
1115 strncmp(path, list->items[first].string, baselen)) {
1116 report_helper(list, seen_bits, first, i);
1117 baselen = -1;
1118 seen_bits = 0;
1119 }
1120 if (baselen == -1) {
1121 const char *dot = strrchr(path, '.');
1122 if (!dot) {
1123 report_garbage("garbage found", path);
1124 continue;
1125 }
1126 baselen = dot - path + 1;
1127 first = i;
1128 }
1129 if (!strcmp(path + baselen, "pack"))
1130 seen_bits |= 1;
1131 else if (!strcmp(path + baselen, "idx"))
1132 seen_bits |= 2;
1133 }
1134 report_helper(list, seen_bits, first, list->nr);
1135}
1136
1137static void prepare_packed_git_one(char *objdir, int local)
1138{
1139 /* Ensure that this buffer is large enough so that we can
1140 append "/pack/" without clobbering the stack even if
1141 strlen(objdir) were PATH_MAX. */
1142 char path[PATH_MAX + 1 + 4 + 1 + 1];
1143 int len;
1144 DIR *dir;
1145 struct dirent *de;
1146 struct string_list garbage = STRING_LIST_INIT_DUP;
1147
1148 sprintf(path, "%s/pack", objdir);
1149 len = strlen(path);
1150 dir = opendir(path);
1151 if (!dir) {
1152 if (errno != ENOENT)
1153 error("unable to open object pack directory: %s: %s",
1154 path, strerror(errno));
1155 return;
1156 }
1157 path[len++] = '/';
1158 while ((de = readdir(dir)) != NULL) {
1159 int namelen = strlen(de->d_name);
1160 struct packed_git *p;
1161
1162 if (len + namelen + 1 > sizeof(path)) {
1163 if (report_garbage) {
1164 struct strbuf sb = STRBUF_INIT;
1165 strbuf_addf(&sb, "%.*s/%s", len - 1, path, de->d_name);
1166 report_garbage("path too long", sb.buf);
1167 strbuf_release(&sb);
1168 }
1169 continue;
1170 }
1171
1172 if (is_dot_or_dotdot(de->d_name))
1173 continue;
1174
1175 strcpy(path + len, de->d_name);
1176
1177 if (has_extension(de->d_name, ".idx")) {
1178 /* Don't reopen a pack we already have. */
1179 for (p = packed_git; p; p = p->next) {
1180 if (!memcmp(path, p->pack_name, len + namelen - 4))
1181 break;
1182 }
1183 if (p == NULL &&
1184 /*
1185 * See if it really is a valid .idx file with
1186 * corresponding .pack file that we can map.
1187 */
1188 (p = add_packed_git(path, len + namelen, local)) != NULL)
1189 install_packed_git(p);
1190 }
1191
1192 if (!report_garbage)
1193 continue;
1194
1195 if (has_extension(de->d_name, ".idx") ||
1196 has_extension(de->d_name, ".pack") ||
1197 has_extension(de->d_name, ".bitmap") ||
1198 has_extension(de->d_name, ".keep"))
1199 string_list_append(&garbage, path);
1200 else
1201 report_garbage("garbage found", path);
1202 }
1203 closedir(dir);
1204 report_pack_garbage(&garbage);
1205 string_list_clear(&garbage, 0);
1206}
1207
1208static int sort_pack(const void *a_, const void *b_)
1209{
1210 struct packed_git *a = *((struct packed_git **)a_);
1211 struct packed_git *b = *((struct packed_git **)b_);
1212 int st;
1213
1214 /*
1215 * Local packs tend to contain objects specific to our
1216 * variant of the project than remote ones. In addition,
1217 * remote ones could be on a network mounted filesystem.
1218 * Favor local ones for these reasons.
1219 */
1220 st = a->pack_local - b->pack_local;
1221 if (st)
1222 return -st;
1223
1224 /*
1225 * Younger packs tend to contain more recent objects,
1226 * and more recent objects tend to get accessed more
1227 * often.
1228 */
1229 if (a->mtime < b->mtime)
1230 return 1;
1231 else if (a->mtime == b->mtime)
1232 return 0;
1233 return -1;
1234}
1235
1236static void rearrange_packed_git(void)
1237{
1238 struct packed_git **ary, *p;
1239 int i, n;
1240
1241 for (n = 0, p = packed_git; p; p = p->next)
1242 n++;
1243 if (n < 2)
1244 return;
1245
1246 /* prepare an array of packed_git for easier sorting */
1247 ary = xcalloc(n, sizeof(struct packed_git *));
1248 for (n = 0, p = packed_git; p; p = p->next)
1249 ary[n++] = p;
1250
1251 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1252
1253 /* link them back again */
1254 for (i = 0; i < n - 1; i++)
1255 ary[i]->next = ary[i + 1];
1256 ary[n - 1]->next = NULL;
1257 packed_git = ary[0];
1258
1259 free(ary);
1260}
1261
1262static int prepare_packed_git_run_once = 0;
1263void prepare_packed_git(void)
1264{
1265 struct alternate_object_database *alt;
1266
1267 if (prepare_packed_git_run_once)
1268 return;
1269 prepare_packed_git_one(get_object_directory(), 1);
1270 prepare_alt_odb();
1271 for (alt = alt_odb_list; alt; alt = alt->next) {
1272 alt->name[-1] = 0;
1273 prepare_packed_git_one(alt->base, 0);
1274 alt->name[-1] = '/';
1275 }
1276 rearrange_packed_git();
1277 prepare_packed_git_run_once = 1;
1278}
1279
1280void reprepare_packed_git(void)
1281{
1282 discard_revindex();
1283 prepare_packed_git_run_once = 0;
1284 prepare_packed_git();
1285}
1286
1287static void mark_bad_packed_object(struct packed_git *p,
1288 const unsigned char *sha1)
1289{
1290 unsigned i;
1291 for (i = 0; i < p->num_bad_objects; i++)
1292 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1293 return;
1294 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1295 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1296 p->num_bad_objects++;
1297}
1298
1299static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1300{
1301 struct packed_git *p;
1302 unsigned i;
1303
1304 for (p = packed_git; p; p = p->next)
1305 for (i = 0; i < p->num_bad_objects; i++)
1306 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1307 return p;
1308 return NULL;
1309}
1310
1311/*
1312 * With an in-core object data in "map", rehash it to make sure the
1313 * object name actually matches "sha1" to detect object corruption.
1314 * With "map" == NULL, try reading the object named with "sha1" using
1315 * the streaming interface and rehash it to do the same.
1316 */
1317int check_sha1_signature(const unsigned char *sha1, void *map,
1318 unsigned long size, const char *type)
1319{
1320 unsigned char real_sha1[20];
1321 enum object_type obj_type;
1322 struct git_istream *st;
1323 git_SHA_CTX c;
1324 char hdr[32];
1325 int hdrlen;
1326
1327 if (map) {
1328 hash_sha1_file(map, size, type, real_sha1);
1329 return hashcmp(sha1, real_sha1) ? -1 : 0;
1330 }
1331
1332 st = open_istream(sha1, &obj_type, &size, NULL);
1333 if (!st)
1334 return -1;
1335
1336 /* Generate the header */
1337 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1338
1339 /* Sha1.. */
1340 git_SHA1_Init(&c);
1341 git_SHA1_Update(&c, hdr, hdrlen);
1342 for (;;) {
1343 char buf[1024 * 16];
1344 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1345
1346 if (readlen < 0) {
1347 close_istream(st);
1348 return -1;
1349 }
1350 if (!readlen)
1351 break;
1352 git_SHA1_Update(&c, buf, readlen);
1353 }
1354 git_SHA1_Final(real_sha1, &c);
1355 close_istream(st);
1356 return hashcmp(sha1, real_sha1) ? -1 : 0;
1357}
1358
1359int git_open_noatime(const char *name)
1360{
1361 static int sha1_file_open_flag = O_NOATIME;
1362
1363 for (;;) {
1364 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1365 if (fd >= 0)
1366 return fd;
1367
1368 /* Might the failure be due to O_NOATIME? */
1369 if (errno != ENOENT && sha1_file_open_flag) {
1370 sha1_file_open_flag = 0;
1371 continue;
1372 }
1373
1374 return -1;
1375 }
1376}
1377
1378static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1379{
1380 char *name = sha1_file_name(sha1);
1381 struct alternate_object_database *alt;
1382
1383 if (!lstat(name, st))
1384 return 0;
1385
1386 prepare_alt_odb();
1387 errno = ENOENT;
1388 for (alt = alt_odb_list; alt; alt = alt->next) {
1389 name = alt->name;
1390 fill_sha1_path(name, sha1);
1391 if (!lstat(alt->base, st))
1392 return 0;
1393 }
1394
1395 return -1;
1396}
1397
1398static int open_sha1_file(const unsigned char *sha1)
1399{
1400 int fd;
1401 char *name = sha1_file_name(sha1);
1402 struct alternate_object_database *alt;
1403
1404 fd = git_open_noatime(name);
1405 if (fd >= 0)
1406 return fd;
1407
1408 prepare_alt_odb();
1409 errno = ENOENT;
1410 for (alt = alt_odb_list; alt; alt = alt->next) {
1411 name = alt->name;
1412 fill_sha1_path(name, sha1);
1413 fd = git_open_noatime(alt->base);
1414 if (fd >= 0)
1415 return fd;
1416 }
1417 return -1;
1418}
1419
1420void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1421{
1422 void *map;
1423 int fd;
1424
1425 fd = open_sha1_file(sha1);
1426 map = NULL;
1427 if (fd >= 0) {
1428 struct stat st;
1429
1430 if (!fstat(fd, &st)) {
1431 *size = xsize_t(st.st_size);
1432 if (!*size) {
1433 /* mmap() is forbidden on empty files */
1434 error("object file %s is empty", sha1_file_name(sha1));
1435 return NULL;
1436 }
1437 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1438 }
1439 close(fd);
1440 }
1441 return map;
1442}
1443
1444/*
1445 * There used to be a second loose object header format which
1446 * was meant to mimic the in-pack format, allowing for direct
1447 * copy of the object data. This format turned up not to be
1448 * really worth it and we no longer write loose objects in that
1449 * format.
1450 */
1451static int experimental_loose_object(unsigned char *map)
1452{
1453 unsigned int word;
1454
1455 /*
1456 * We must determine if the buffer contains the standard
1457 * zlib-deflated stream or the experimental format based
1458 * on the in-pack object format. Compare the header byte
1459 * for each format:
1460 *
1461 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1462 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1463 *
1464 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1465 * in standard loose-object format, UNLESS it is a Git-pack
1466 * format object *exactly* 8 bytes in size when inflated.
1467 *
1468 * However, RFC1950 also specifies that the 1st 16-bit word
1469 * must be divisible by 31 - this checksum tells us our buffer
1470 * is in the standard format, giving a false positive only if
1471 * the 1st word of the Git-pack format object happens to be
1472 * divisible by 31, ie:
1473 * ((byte0 * 256) + byte1) % 31 = 0
1474 * => 0ttt10000www1000 % 31 = 0
1475 *
1476 * As it happens, this case can only arise for www=3 & ttt=1
1477 * - ie, a Commit object, which would have to be 8 bytes in
1478 * size. As no Commit can be that small, we find that the
1479 * combination of these two criteria (bitmask & checksum)
1480 * can always correctly determine the buffer format.
1481 */
1482 word = (map[0] << 8) + map[1];
1483 if ((map[0] & 0x8F) == 0x08 && !(word % 31))
1484 return 0;
1485 else
1486 return 1;
1487}
1488
1489unsigned long unpack_object_header_buffer(const unsigned char *buf,
1490 unsigned long len, enum object_type *type, unsigned long *sizep)
1491{
1492 unsigned shift;
1493 unsigned long size, c;
1494 unsigned long used = 0;
1495
1496 c = buf[used++];
1497 *type = (c >> 4) & 7;
1498 size = c & 15;
1499 shift = 4;
1500 while (c & 0x80) {
1501 if (len <= used || bitsizeof(long) <= shift) {
1502 error("bad object header");
1503 size = used = 0;
1504 break;
1505 }
1506 c = buf[used++];
1507 size += (c & 0x7f) << shift;
1508 shift += 7;
1509 }
1510 *sizep = size;
1511 return used;
1512}
1513
1514int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1515{
1516 unsigned long size, used;
1517 static const char valid_loose_object_type[8] = {
1518 0, /* OBJ_EXT */
1519 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1520 0, /* "delta" and others are invalid in a loose object */
1521 };
1522 enum object_type type;
1523
1524 /* Get the data stream */
1525 memset(stream, 0, sizeof(*stream));
1526 stream->next_in = map;
1527 stream->avail_in = mapsize;
1528 stream->next_out = buffer;
1529 stream->avail_out = bufsiz;
1530
1531 if (experimental_loose_object(map)) {
1532 /*
1533 * The old experimental format we no longer produce;
1534 * we can still read it.
1535 */
1536 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1537 if (!used || !valid_loose_object_type[type])
1538 return -1;
1539 map += used;
1540 mapsize -= used;
1541
1542 /* Set up the stream for the rest.. */
1543 stream->next_in = map;
1544 stream->avail_in = mapsize;
1545 git_inflate_init(stream);
1546
1547 /* And generate the fake traditional header */
1548 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1549 typename(type), size);
1550 return 0;
1551 }
1552 git_inflate_init(stream);
1553 return git_inflate(stream, 0);
1554}
1555
1556static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1557{
1558 int bytes = strlen(buffer) + 1;
1559 unsigned char *buf = xmallocz(size);
1560 unsigned long n;
1561 int status = Z_OK;
1562
1563 n = stream->total_out - bytes;
1564 if (n > size)
1565 n = size;
1566 memcpy(buf, (char *) buffer + bytes, n);
1567 bytes = n;
1568 if (bytes <= size) {
1569 /*
1570 * The above condition must be (bytes <= size), not
1571 * (bytes < size). In other words, even though we
1572 * expect no more output and set avail_out to zero,
1573 * the input zlib stream may have bytes that express
1574 * "this concludes the stream", and we *do* want to
1575 * eat that input.
1576 *
1577 * Otherwise we would not be able to test that we
1578 * consumed all the input to reach the expected size;
1579 * we also want to check that zlib tells us that all
1580 * went well with status == Z_STREAM_END at the end.
1581 */
1582 stream->next_out = buf + bytes;
1583 stream->avail_out = size - bytes;
1584 while (status == Z_OK)
1585 status = git_inflate(stream, Z_FINISH);
1586 }
1587 if (status == Z_STREAM_END && !stream->avail_in) {
1588 git_inflate_end(stream);
1589 return buf;
1590 }
1591
1592 if (status < 0)
1593 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1594 else if (stream->avail_in)
1595 error("garbage at end of loose object '%s'",
1596 sha1_to_hex(sha1));
1597 free(buf);
1598 return NULL;
1599}
1600
1601/*
1602 * We used to just use "sscanf()", but that's actually way
1603 * too permissive for what we want to check. So do an anal
1604 * object header parse by hand.
1605 */
1606int parse_sha1_header(const char *hdr, unsigned long *sizep)
1607{
1608 char type[10];
1609 int i;
1610 unsigned long size;
1611
1612 /*
1613 * The type can be at most ten bytes (including the
1614 * terminating '\0' that we add), and is followed by
1615 * a space.
1616 */
1617 i = 0;
1618 for (;;) {
1619 char c = *hdr++;
1620 if (c == ' ')
1621 break;
1622 type[i++] = c;
1623 if (i >= sizeof(type))
1624 return -1;
1625 }
1626 type[i] = 0;
1627
1628 /*
1629 * The length must follow immediately, and be in canonical
1630 * decimal format (ie "010" is not valid).
1631 */
1632 size = *hdr++ - '0';
1633 if (size > 9)
1634 return -1;
1635 if (size) {
1636 for (;;) {
1637 unsigned long c = *hdr - '0';
1638 if (c > 9)
1639 break;
1640 hdr++;
1641 size = size * 10 + c;
1642 }
1643 }
1644 *sizep = size;
1645
1646 /*
1647 * The length must be followed by a zero byte
1648 */
1649 return *hdr ? -1 : type_from_string(type);
1650}
1651
1652static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1653{
1654 int ret;
1655 git_zstream stream;
1656 char hdr[8192];
1657
1658 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1659 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1660 return NULL;
1661
1662 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1663}
1664
1665unsigned long get_size_from_delta(struct packed_git *p,
1666 struct pack_window **w_curs,
1667 off_t curpos)
1668{
1669 const unsigned char *data;
1670 unsigned char delta_head[20], *in;
1671 git_zstream stream;
1672 int st;
1673
1674 memset(&stream, 0, sizeof(stream));
1675 stream.next_out = delta_head;
1676 stream.avail_out = sizeof(delta_head);
1677
1678 git_inflate_init(&stream);
1679 do {
1680 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1681 stream.next_in = in;
1682 st = git_inflate(&stream, Z_FINISH);
1683 curpos += stream.next_in - in;
1684 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1685 stream.total_out < sizeof(delta_head));
1686 git_inflate_end(&stream);
1687 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1688 error("delta data unpack-initial failed");
1689 return 0;
1690 }
1691
1692 /* Examine the initial part of the delta to figure out
1693 * the result size.
1694 */
1695 data = delta_head;
1696
1697 /* ignore base size */
1698 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1699
1700 /* Read the result size */
1701 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1702}
1703
1704static off_t get_delta_base(struct packed_git *p,
1705 struct pack_window **w_curs,
1706 off_t *curpos,
1707 enum object_type type,
1708 off_t delta_obj_offset)
1709{
1710 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1711 off_t base_offset;
1712
1713 /* use_pack() assured us we have [base_info, base_info + 20)
1714 * as a range that we can look at without walking off the
1715 * end of the mapped window. Its actually the hash size
1716 * that is assured. An OFS_DELTA longer than the hash size
1717 * is stupid, as then a REF_DELTA would be smaller to store.
1718 */
1719 if (type == OBJ_OFS_DELTA) {
1720 unsigned used = 0;
1721 unsigned char c = base_info[used++];
1722 base_offset = c & 127;
1723 while (c & 128) {
1724 base_offset += 1;
1725 if (!base_offset || MSB(base_offset, 7))
1726 return 0; /* overflow */
1727 c = base_info[used++];
1728 base_offset = (base_offset << 7) + (c & 127);
1729 }
1730 base_offset = delta_obj_offset - base_offset;
1731 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1732 return 0; /* out of bound */
1733 *curpos += used;
1734 } else if (type == OBJ_REF_DELTA) {
1735 /* The base entry _must_ be in the same pack */
1736 base_offset = find_pack_entry_one(base_info, p);
1737 *curpos += 20;
1738 } else
1739 die("I am totally screwed");
1740 return base_offset;
1741}
1742
1743int unpack_object_header(struct packed_git *p,
1744 struct pack_window **w_curs,
1745 off_t *curpos,
1746 unsigned long *sizep)
1747{
1748 unsigned char *base;
1749 unsigned long left;
1750 unsigned long used;
1751 enum object_type type;
1752
1753 /* use_pack() assures us we have [base, base + 20) available
1754 * as a range that we can look at. (Its actually the hash
1755 * size that is assured.) With our object header encoding
1756 * the maximum deflated object size is 2^137, which is just
1757 * insane, so we know won't exceed what we have been given.
1758 */
1759 base = use_pack(p, w_curs, *curpos, &left);
1760 used = unpack_object_header_buffer(base, left, &type, sizep);
1761 if (!used) {
1762 type = OBJ_BAD;
1763 } else
1764 *curpos += used;
1765
1766 return type;
1767}
1768
1769static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1770{
1771 int type;
1772 struct revindex_entry *revidx;
1773 const unsigned char *sha1;
1774 revidx = find_pack_revindex(p, obj_offset);
1775 if (!revidx)
1776 return OBJ_BAD;
1777 sha1 = nth_packed_object_sha1(p, revidx->nr);
1778 mark_bad_packed_object(p, sha1);
1779 type = sha1_object_info(sha1, NULL);
1780 if (type <= OBJ_NONE)
1781 return OBJ_BAD;
1782 return type;
1783}
1784
1785#define POI_STACK_PREALLOC 64
1786
1787static enum object_type packed_to_object_type(struct packed_git *p,
1788 off_t obj_offset,
1789 enum object_type type,
1790 struct pack_window **w_curs,
1791 off_t curpos)
1792{
1793 off_t small_poi_stack[POI_STACK_PREALLOC];
1794 off_t *poi_stack = small_poi_stack;
1795 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1796
1797 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1798 off_t base_offset;
1799 unsigned long size;
1800 /* Push the object we're going to leave behind */
1801 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1802 poi_stack_alloc = alloc_nr(poi_stack_nr);
1803 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1804 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1805 } else {
1806 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1807 }
1808 poi_stack[poi_stack_nr++] = obj_offset;
1809 /* If parsing the base offset fails, just unwind */
1810 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1811 if (!base_offset)
1812 goto unwind;
1813 curpos = obj_offset = base_offset;
1814 type = unpack_object_header(p, w_curs, &curpos, &size);
1815 if (type <= OBJ_NONE) {
1816 /* If getting the base itself fails, we first
1817 * retry the base, otherwise unwind */
1818 type = retry_bad_packed_offset(p, base_offset);
1819 if (type > OBJ_NONE)
1820 goto out;
1821 goto unwind;
1822 }
1823 }
1824
1825 switch (type) {
1826 case OBJ_BAD:
1827 case OBJ_COMMIT:
1828 case OBJ_TREE:
1829 case OBJ_BLOB:
1830 case OBJ_TAG:
1831 break;
1832 default:
1833 error("unknown object type %i at offset %"PRIuMAX" in %s",
1834 type, (uintmax_t)obj_offset, p->pack_name);
1835 type = OBJ_BAD;
1836 }
1837
1838out:
1839 if (poi_stack != small_poi_stack)
1840 free(poi_stack);
1841 return type;
1842
1843unwind:
1844 while (poi_stack_nr) {
1845 obj_offset = poi_stack[--poi_stack_nr];
1846 type = retry_bad_packed_offset(p, obj_offset);
1847 if (type > OBJ_NONE)
1848 goto out;
1849 }
1850 type = OBJ_BAD;
1851 goto out;
1852}
1853
1854static int packed_object_info(struct packed_git *p, off_t obj_offset,
1855 struct object_info *oi)
1856{
1857 struct pack_window *w_curs = NULL;
1858 unsigned long size;
1859 off_t curpos = obj_offset;
1860 enum object_type type;
1861
1862 /*
1863 * We always get the representation type, but only convert it to
1864 * a "real" type later if the caller is interested.
1865 */
1866 type = unpack_object_header(p, &w_curs, &curpos, &size);
1867
1868 if (oi->sizep) {
1869 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1870 off_t tmp_pos = curpos;
1871 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1872 type, obj_offset);
1873 if (!base_offset) {
1874 type = OBJ_BAD;
1875 goto out;
1876 }
1877 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1878 if (*oi->sizep == 0) {
1879 type = OBJ_BAD;
1880 goto out;
1881 }
1882 } else {
1883 *oi->sizep = size;
1884 }
1885 }
1886
1887 if (oi->disk_sizep) {
1888 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1889 *oi->disk_sizep = revidx[1].offset - obj_offset;
1890 }
1891
1892 if (oi->typep) {
1893 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
1894 if (*oi->typep < 0) {
1895 type = OBJ_BAD;
1896 goto out;
1897 }
1898 }
1899
1900out:
1901 unuse_pack(&w_curs);
1902 return type;
1903}
1904
1905static void *unpack_compressed_entry(struct packed_git *p,
1906 struct pack_window **w_curs,
1907 off_t curpos,
1908 unsigned long size)
1909{
1910 int st;
1911 git_zstream stream;
1912 unsigned char *buffer, *in;
1913
1914 buffer = xmallocz(size);
1915 memset(&stream, 0, sizeof(stream));
1916 stream.next_out = buffer;
1917 stream.avail_out = size + 1;
1918
1919 git_inflate_init(&stream);
1920 do {
1921 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1922 stream.next_in = in;
1923 st = git_inflate(&stream, Z_FINISH);
1924 if (!stream.avail_out)
1925 break; /* the payload is larger than it should be */
1926 curpos += stream.next_in - in;
1927 } while (st == Z_OK || st == Z_BUF_ERROR);
1928 git_inflate_end(&stream);
1929 if ((st != Z_STREAM_END) || stream.total_out != size) {
1930 free(buffer);
1931 return NULL;
1932 }
1933
1934 return buffer;
1935}
1936
1937#define MAX_DELTA_CACHE (256)
1938
1939static size_t delta_base_cached;
1940
1941static struct delta_base_cache_lru_list {
1942 struct delta_base_cache_lru_list *prev;
1943 struct delta_base_cache_lru_list *next;
1944} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1945
1946static struct delta_base_cache_entry {
1947 struct delta_base_cache_lru_list lru;
1948 void *data;
1949 struct packed_git *p;
1950 off_t base_offset;
1951 unsigned long size;
1952 enum object_type type;
1953} delta_base_cache[MAX_DELTA_CACHE];
1954
1955static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1956{
1957 unsigned long hash;
1958
1959 hash = (unsigned long)p + (unsigned long)base_offset;
1960 hash += (hash >> 8) + (hash >> 16);
1961 return hash % MAX_DELTA_CACHE;
1962}
1963
1964static struct delta_base_cache_entry *
1965get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1966{
1967 unsigned long hash = pack_entry_hash(p, base_offset);
1968 return delta_base_cache + hash;
1969}
1970
1971static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
1972 struct packed_git *p, off_t base_offset)
1973{
1974 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1975}
1976
1977static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1978{
1979 struct delta_base_cache_entry *ent;
1980 ent = get_delta_base_cache_entry(p, base_offset);
1981 return eq_delta_base_cache_entry(ent, p, base_offset);
1982}
1983
1984static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1985{
1986 ent->data = NULL;
1987 ent->lru.next->prev = ent->lru.prev;
1988 ent->lru.prev->next = ent->lru.next;
1989 delta_base_cached -= ent->size;
1990}
1991
1992static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1993 unsigned long *base_size, enum object_type *type, int keep_cache)
1994{
1995 struct delta_base_cache_entry *ent;
1996 void *ret;
1997
1998 ent = get_delta_base_cache_entry(p, base_offset);
1999
2000 if (!eq_delta_base_cache_entry(ent, p, base_offset))
2001 return unpack_entry(p, base_offset, type, base_size);
2002
2003 ret = ent->data;
2004
2005 if (!keep_cache)
2006 clear_delta_base_cache_entry(ent);
2007 else
2008 ret = xmemdupz(ent->data, ent->size);
2009 *type = ent->type;
2010 *base_size = ent->size;
2011 return ret;
2012}
2013
2014static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2015{
2016 if (ent->data) {
2017 free(ent->data);
2018 ent->data = NULL;
2019 ent->lru.next->prev = ent->lru.prev;
2020 ent->lru.prev->next = ent->lru.next;
2021 delta_base_cached -= ent->size;
2022 }
2023}
2024
2025void clear_delta_base_cache(void)
2026{
2027 unsigned long p;
2028 for (p = 0; p < MAX_DELTA_CACHE; p++)
2029 release_delta_base_cache(&delta_base_cache[p]);
2030}
2031
2032static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2033 void *base, unsigned long base_size, enum object_type type)
2034{
2035 unsigned long hash = pack_entry_hash(p, base_offset);
2036 struct delta_base_cache_entry *ent = delta_base_cache + hash;
2037 struct delta_base_cache_lru_list *lru;
2038
2039 release_delta_base_cache(ent);
2040 delta_base_cached += base_size;
2041
2042 for (lru = delta_base_cache_lru.next;
2043 delta_base_cached > delta_base_cache_limit
2044 && lru != &delta_base_cache_lru;
2045 lru = lru->next) {
2046 struct delta_base_cache_entry *f = (void *)lru;
2047 if (f->type == OBJ_BLOB)
2048 release_delta_base_cache(f);
2049 }
2050 for (lru = delta_base_cache_lru.next;
2051 delta_base_cached > delta_base_cache_limit
2052 && lru != &delta_base_cache_lru;
2053 lru = lru->next) {
2054 struct delta_base_cache_entry *f = (void *)lru;
2055 release_delta_base_cache(f);
2056 }
2057
2058 ent->p = p;
2059 ent->base_offset = base_offset;
2060 ent->type = type;
2061 ent->data = base;
2062 ent->size = base_size;
2063 ent->lru.next = &delta_base_cache_lru;
2064 ent->lru.prev = delta_base_cache_lru.prev;
2065 delta_base_cache_lru.prev->next = &ent->lru;
2066 delta_base_cache_lru.prev = &ent->lru;
2067}
2068
2069static void *read_object(const unsigned char *sha1, enum object_type *type,
2070 unsigned long *size);
2071
2072static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2073{
2074 static FILE *log_file;
2075
2076 if (!log_pack_access)
2077 log_pack_access = getenv("GIT_TRACE_PACK_ACCESS");
2078 if (!log_pack_access)
2079 log_pack_access = no_log_pack_access;
2080 if (log_pack_access == no_log_pack_access)
2081 return;
2082
2083 if (!log_file) {
2084 log_file = fopen(log_pack_access, "w");
2085 if (!log_file) {
2086 error("cannot open pack access log '%s' for writing: %s",
2087 log_pack_access, strerror(errno));
2088 log_pack_access = no_log_pack_access;
2089 return;
2090 }
2091 }
2092 fprintf(log_file, "%s %"PRIuMAX"\n",
2093 p->pack_name, (uintmax_t)obj_offset);
2094 fflush(log_file);
2095}
2096
2097int do_check_packed_object_crc;
2098
2099#define UNPACK_ENTRY_STACK_PREALLOC 64
2100struct unpack_entry_stack_ent {
2101 off_t obj_offset;
2102 off_t curpos;
2103 unsigned long size;
2104};
2105
2106void *unpack_entry(struct packed_git *p, off_t obj_offset,
2107 enum object_type *final_type, unsigned long *final_size)
2108{
2109 struct pack_window *w_curs = NULL;
2110 off_t curpos = obj_offset;
2111 void *data = NULL;
2112 unsigned long size;
2113 enum object_type type;
2114 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2115 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2116 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2117 int base_from_cache = 0;
2118
2119 if (log_pack_access != no_log_pack_access)
2120 write_pack_access_log(p, obj_offset);
2121
2122 /* PHASE 1: drill down to the innermost base object */
2123 for (;;) {
2124 off_t base_offset;
2125 int i;
2126 struct delta_base_cache_entry *ent;
2127
2128 ent = get_delta_base_cache_entry(p, curpos);
2129 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2130 type = ent->type;
2131 data = ent->data;
2132 size = ent->size;
2133 clear_delta_base_cache_entry(ent);
2134 base_from_cache = 1;
2135 break;
2136 }
2137
2138 if (do_check_packed_object_crc && p->index_version > 1) {
2139 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2140 unsigned long len = revidx[1].offset - obj_offset;
2141 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2142 const unsigned char *sha1 =
2143 nth_packed_object_sha1(p, revidx->nr);
2144 error("bad packed object CRC for %s",
2145 sha1_to_hex(sha1));
2146 mark_bad_packed_object(p, sha1);
2147 unuse_pack(&w_curs);
2148 return NULL;
2149 }
2150 }
2151
2152 type = unpack_object_header(p, &w_curs, &curpos, &size);
2153 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2154 break;
2155
2156 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2157 if (!base_offset) {
2158 error("failed to validate delta base reference "
2159 "at offset %"PRIuMAX" from %s",
2160 (uintmax_t)curpos, p->pack_name);
2161 /* bail to phase 2, in hopes of recovery */
2162 data = NULL;
2163 break;
2164 }
2165
2166 /* push object, proceed to base */
2167 if (delta_stack_nr >= delta_stack_alloc
2168 && delta_stack == small_delta_stack) {
2169 delta_stack_alloc = alloc_nr(delta_stack_nr);
2170 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2171 memcpy(delta_stack, small_delta_stack,
2172 sizeof(*delta_stack)*delta_stack_nr);
2173 } else {
2174 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2175 }
2176 i = delta_stack_nr++;
2177 delta_stack[i].obj_offset = obj_offset;
2178 delta_stack[i].curpos = curpos;
2179 delta_stack[i].size = size;
2180
2181 curpos = obj_offset = base_offset;
2182 }
2183
2184 /* PHASE 2: handle the base */
2185 switch (type) {
2186 case OBJ_OFS_DELTA:
2187 case OBJ_REF_DELTA:
2188 if (data)
2189 die("BUG in unpack_entry: left loop at a valid delta");
2190 break;
2191 case OBJ_COMMIT:
2192 case OBJ_TREE:
2193 case OBJ_BLOB:
2194 case OBJ_TAG:
2195 if (!base_from_cache)
2196 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2197 break;
2198 default:
2199 data = NULL;
2200 error("unknown object type %i at offset %"PRIuMAX" in %s",
2201 type, (uintmax_t)obj_offset, p->pack_name);
2202 }
2203
2204 /* PHASE 3: apply deltas in order */
2205
2206 /* invariants:
2207 * 'data' holds the base data, or NULL if there was corruption
2208 */
2209 while (delta_stack_nr) {
2210 void *delta_data;
2211 void *base = data;
2212 unsigned long delta_size, base_size = size;
2213 int i;
2214
2215 data = NULL;
2216
2217 if (base)
2218 add_delta_base_cache(p, obj_offset, base, base_size, type);
2219
2220 if (!base) {
2221 /*
2222 * We're probably in deep shit, but let's try to fetch
2223 * the required base anyway from another pack or loose.
2224 * This is costly but should happen only in the presence
2225 * of a corrupted pack, and is better than failing outright.
2226 */
2227 struct revindex_entry *revidx;
2228 const unsigned char *base_sha1;
2229 revidx = find_pack_revindex(p, obj_offset);
2230 if (revidx) {
2231 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2232 error("failed to read delta base object %s"
2233 " at offset %"PRIuMAX" from %s",
2234 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2235 p->pack_name);
2236 mark_bad_packed_object(p, base_sha1);
2237 base = read_object(base_sha1, &type, &base_size);
2238 }
2239 }
2240
2241 i = --delta_stack_nr;
2242 obj_offset = delta_stack[i].obj_offset;
2243 curpos = delta_stack[i].curpos;
2244 delta_size = delta_stack[i].size;
2245
2246 if (!base)
2247 continue;
2248
2249 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2250
2251 if (!delta_data) {
2252 error("failed to unpack compressed delta "
2253 "at offset %"PRIuMAX" from %s",
2254 (uintmax_t)curpos, p->pack_name);
2255 data = NULL;
2256 continue;
2257 }
2258
2259 data = patch_delta(base, base_size,
2260 delta_data, delta_size,
2261 &size);
2262
2263 /*
2264 * We could not apply the delta; warn the user, but keep going.
2265 * Our failure will be noticed either in the next iteration of
2266 * the loop, or if this is the final delta, in the caller when
2267 * we return NULL. Those code paths will take care of making
2268 * a more explicit warning and retrying with another copy of
2269 * the object.
2270 */
2271 if (!data)
2272 error("failed to apply delta");
2273
2274 free(delta_data);
2275 }
2276
2277 *final_type = type;
2278 *final_size = size;
2279
2280 unuse_pack(&w_curs);
2281 return data;
2282}
2283
2284const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2285 uint32_t n)
2286{
2287 const unsigned char *index = p->index_data;
2288 if (!index) {
2289 if (open_pack_index(p))
2290 return NULL;
2291 index = p->index_data;
2292 }
2293 if (n >= p->num_objects)
2294 return NULL;
2295 index += 4 * 256;
2296 if (p->index_version == 1) {
2297 return index + 24 * n + 4;
2298 } else {
2299 index += 8;
2300 return index + 20 * n;
2301 }
2302}
2303
2304off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2305{
2306 const unsigned char *index = p->index_data;
2307 index += 4 * 256;
2308 if (p->index_version == 1) {
2309 return ntohl(*((uint32_t *)(index + 24 * n)));
2310 } else {
2311 uint32_t off;
2312 index += 8 + p->num_objects * (20 + 4);
2313 off = ntohl(*((uint32_t *)(index + 4 * n)));
2314 if (!(off & 0x80000000))
2315 return off;
2316 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2317 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2318 ntohl(*((uint32_t *)(index + 4)));
2319 }
2320}
2321
2322off_t find_pack_entry_one(const unsigned char *sha1,
2323 struct packed_git *p)
2324{
2325 const uint32_t *level1_ofs = p->index_data;
2326 const unsigned char *index = p->index_data;
2327 unsigned hi, lo, stride;
2328 static int use_lookup = -1;
2329 static int debug_lookup = -1;
2330
2331 if (debug_lookup < 0)
2332 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2333
2334 if (!index) {
2335 if (open_pack_index(p))
2336 return 0;
2337 level1_ofs = p->index_data;
2338 index = p->index_data;
2339 }
2340 if (p->index_version > 1) {
2341 level1_ofs += 2;
2342 index += 8;
2343 }
2344 index += 4 * 256;
2345 hi = ntohl(level1_ofs[*sha1]);
2346 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2347 if (p->index_version > 1) {
2348 stride = 20;
2349 } else {
2350 stride = 24;
2351 index += 4;
2352 }
2353
2354 if (debug_lookup)
2355 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2356 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2357
2358 if (use_lookup < 0)
2359 use_lookup = !!getenv("GIT_USE_LOOKUP");
2360 if (use_lookup) {
2361 int pos = sha1_entry_pos(index, stride, 0,
2362 lo, hi, p->num_objects, sha1);
2363 if (pos < 0)
2364 return 0;
2365 return nth_packed_object_offset(p, pos);
2366 }
2367
2368 do {
2369 unsigned mi = (lo + hi) / 2;
2370 int cmp = hashcmp(index + mi * stride, sha1);
2371
2372 if (debug_lookup)
2373 printf("lo %u hi %u rg %u mi %u\n",
2374 lo, hi, hi - lo, mi);
2375 if (!cmp)
2376 return nth_packed_object_offset(p, mi);
2377 if (cmp > 0)
2378 hi = mi;
2379 else
2380 lo = mi+1;
2381 } while (lo < hi);
2382 return 0;
2383}
2384
2385int is_pack_valid(struct packed_git *p)
2386{
2387 /* An already open pack is known to be valid. */
2388 if (p->pack_fd != -1)
2389 return 1;
2390
2391 /* If the pack has one window completely covering the
2392 * file size, the pack is known to be valid even if
2393 * the descriptor is not currently open.
2394 */
2395 if (p->windows) {
2396 struct pack_window *w = p->windows;
2397
2398 if (!w->offset && w->len == p->pack_size)
2399 return 1;
2400 }
2401
2402 /* Force the pack to open to prove its valid. */
2403 return !open_packed_git(p);
2404}
2405
2406static int fill_pack_entry(const unsigned char *sha1,
2407 struct pack_entry *e,
2408 struct packed_git *p)
2409{
2410 off_t offset;
2411
2412 if (p->num_bad_objects) {
2413 unsigned i;
2414 for (i = 0; i < p->num_bad_objects; i++)
2415 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2416 return 0;
2417 }
2418
2419 offset = find_pack_entry_one(sha1, p);
2420 if (!offset)
2421 return 0;
2422
2423 /*
2424 * We are about to tell the caller where they can locate the
2425 * requested object. We better make sure the packfile is
2426 * still here and can be accessed before supplying that
2427 * answer, as it may have been deleted since the index was
2428 * loaded!
2429 */
2430 if (!is_pack_valid(p)) {
2431 warning("packfile %s cannot be accessed", p->pack_name);
2432 return 0;
2433 }
2434 e->offset = offset;
2435 e->p = p;
2436 hashcpy(e->sha1, sha1);
2437 return 1;
2438}
2439
2440static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2441{
2442 struct packed_git *p;
2443
2444 prepare_packed_git();
2445 if (!packed_git)
2446 return 0;
2447
2448 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2449 return 1;
2450
2451 for (p = packed_git; p; p = p->next) {
2452 if (p == last_found_pack || !fill_pack_entry(sha1, e, p))
2453 continue;
2454
2455 last_found_pack = p;
2456 return 1;
2457 }
2458 return 0;
2459}
2460
2461struct packed_git *find_sha1_pack(const unsigned char *sha1,
2462 struct packed_git *packs)
2463{
2464 struct packed_git *p;
2465
2466 for (p = packs; p; p = p->next) {
2467 if (find_pack_entry_one(sha1, p))
2468 return p;
2469 }
2470 return NULL;
2471
2472}
2473
2474static int sha1_loose_object_info(const unsigned char *sha1,
2475 struct object_info *oi)
2476{
2477 int status;
2478 unsigned long mapsize, size;
2479 void *map;
2480 git_zstream stream;
2481 char hdr[32];
2482
2483 /*
2484 * If we don't care about type or size, then we don't
2485 * need to look inside the object at all.
2486 */
2487 if (!oi->typep && !oi->sizep) {
2488 if (oi->disk_sizep) {
2489 struct stat st;
2490 if (stat_sha1_file(sha1, &st) < 0)
2491 return -1;
2492 *oi->disk_sizep = st.st_size;
2493 }
2494 return 0;
2495 }
2496
2497 map = map_sha1_file(sha1, &mapsize);
2498 if (!map)
2499 return -1;
2500 if (oi->disk_sizep)
2501 *oi->disk_sizep = mapsize;
2502 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2503 status = error("unable to unpack %s header",
2504 sha1_to_hex(sha1));
2505 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2506 status = error("unable to parse %s header", sha1_to_hex(sha1));
2507 else if (oi->sizep)
2508 *oi->sizep = size;
2509 git_inflate_end(&stream);
2510 munmap(map, mapsize);
2511 if (oi->typep)
2512 *oi->typep = status;
2513 return 0;
2514}
2515
2516/* returns enum object_type or negative */
2517int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
2518{
2519 struct cached_object *co;
2520 struct pack_entry e;
2521 int rtype;
2522
2523 co = find_cached_object(sha1);
2524 if (co) {
2525 if (oi->typep)
2526 *(oi->typep) = co->type;
2527 if (oi->sizep)
2528 *(oi->sizep) = co->size;
2529 if (oi->disk_sizep)
2530 *(oi->disk_sizep) = 0;
2531 oi->whence = OI_CACHED;
2532 return 0;
2533 }
2534
2535 if (!find_pack_entry(sha1, &e)) {
2536 /* Most likely it's a loose object. */
2537 if (!sha1_loose_object_info(sha1, oi)) {
2538 oi->whence = OI_LOOSE;
2539 return 0;
2540 }
2541
2542 /* Not a loose object; someone else may have just packed it. */
2543 reprepare_packed_git();
2544 if (!find_pack_entry(sha1, &e))
2545 return -1;
2546 }
2547
2548 rtype = packed_object_info(e.p, e.offset, oi);
2549 if (rtype < 0) {
2550 mark_bad_packed_object(e.p, sha1);
2551 return sha1_object_info_extended(sha1, oi);
2552 } else if (in_delta_base_cache(e.p, e.offset)) {
2553 oi->whence = OI_DBCACHED;
2554 } else {
2555 oi->whence = OI_PACKED;
2556 oi->u.packed.offset = e.offset;
2557 oi->u.packed.pack = e.p;
2558 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2559 rtype == OBJ_OFS_DELTA);
2560 }
2561
2562 return 0;
2563}
2564
2565int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2566{
2567 enum object_type type;
2568 struct object_info oi = {NULL};
2569
2570 oi.typep = &type;
2571 oi.sizep = sizep;
2572 if (sha1_object_info_extended(sha1, &oi) < 0)
2573 return -1;
2574 return type;
2575}
2576
2577static void *read_packed_sha1(const unsigned char *sha1,
2578 enum object_type *type, unsigned long *size)
2579{
2580 struct pack_entry e;
2581 void *data;
2582
2583 if (!find_pack_entry(sha1, &e))
2584 return NULL;
2585 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2586 if (!data) {
2587 /*
2588 * We're probably in deep shit, but let's try to fetch
2589 * the required object anyway from another pack or loose.
2590 * This should happen only in the presence of a corrupted
2591 * pack, and is better than failing outright.
2592 */
2593 error("failed to read object %s at offset %"PRIuMAX" from %s",
2594 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2595 mark_bad_packed_object(e.p, sha1);
2596 data = read_object(sha1, type, size);
2597 }
2598 return data;
2599}
2600
2601int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2602 unsigned char *sha1)
2603{
2604 struct cached_object *co;
2605
2606 hash_sha1_file(buf, len, typename(type), sha1);
2607 if (has_sha1_file(sha1) || find_cached_object(sha1))
2608 return 0;
2609 if (cached_object_alloc <= cached_object_nr) {
2610 cached_object_alloc = alloc_nr(cached_object_alloc);
2611 cached_objects = xrealloc(cached_objects,
2612 sizeof(*cached_objects) *
2613 cached_object_alloc);
2614 }
2615 co = &cached_objects[cached_object_nr++];
2616 co->size = len;
2617 co->type = type;
2618 co->buf = xmalloc(len);
2619 memcpy(co->buf, buf, len);
2620 hashcpy(co->sha1, sha1);
2621 return 0;
2622}
2623
2624static void *read_object(const unsigned char *sha1, enum object_type *type,
2625 unsigned long *size)
2626{
2627 unsigned long mapsize;
2628 void *map, *buf;
2629 struct cached_object *co;
2630
2631 co = find_cached_object(sha1);
2632 if (co) {
2633 *type = co->type;
2634 *size = co->size;
2635 return xmemdupz(co->buf, co->size);
2636 }
2637
2638 buf = read_packed_sha1(sha1, type, size);
2639 if (buf)
2640 return buf;
2641 map = map_sha1_file(sha1, &mapsize);
2642 if (map) {
2643 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2644 munmap(map, mapsize);
2645 return buf;
2646 }
2647 reprepare_packed_git();
2648 return read_packed_sha1(sha1, type, size);
2649}
2650
2651/*
2652 * This function dies on corrupt objects; the callers who want to
2653 * deal with them should arrange to call read_object() and give error
2654 * messages themselves.
2655 */
2656void *read_sha1_file_extended(const unsigned char *sha1,
2657 enum object_type *type,
2658 unsigned long *size,
2659 unsigned flag)
2660{
2661 void *data;
2662 char *path;
2663 const struct packed_git *p;
2664 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2665 ? lookup_replace_object(sha1) : sha1;
2666
2667 errno = 0;
2668 data = read_object(repl, type, size);
2669 if (data)
2670 return data;
2671
2672 if (errno && errno != ENOENT)
2673 die_errno("failed to read object %s", sha1_to_hex(sha1));
2674
2675 /* die if we replaced an object with one that does not exist */
2676 if (repl != sha1)
2677 die("replacement %s not found for %s",
2678 sha1_to_hex(repl), sha1_to_hex(sha1));
2679
2680 if (has_loose_object(repl)) {
2681 path = sha1_file_name(sha1);
2682 die("loose object %s (stored in %s) is corrupt",
2683 sha1_to_hex(repl), path);
2684 }
2685
2686 if ((p = has_packed_and_bad(repl)) != NULL)
2687 die("packed object %s (stored in %s) is corrupt",
2688 sha1_to_hex(repl), p->pack_name);
2689
2690 return NULL;
2691}
2692
2693void *read_object_with_reference(const unsigned char *sha1,
2694 const char *required_type_name,
2695 unsigned long *size,
2696 unsigned char *actual_sha1_return)
2697{
2698 enum object_type type, required_type;
2699 void *buffer;
2700 unsigned long isize;
2701 unsigned char actual_sha1[20];
2702
2703 required_type = type_from_string(required_type_name);
2704 hashcpy(actual_sha1, sha1);
2705 while (1) {
2706 int ref_length = -1;
2707 const char *ref_type = NULL;
2708
2709 buffer = read_sha1_file(actual_sha1, &type, &isize);
2710 if (!buffer)
2711 return NULL;
2712 if (type == required_type) {
2713 *size = isize;
2714 if (actual_sha1_return)
2715 hashcpy(actual_sha1_return, actual_sha1);
2716 return buffer;
2717 }
2718 /* Handle references */
2719 else if (type == OBJ_COMMIT)
2720 ref_type = "tree ";
2721 else if (type == OBJ_TAG)
2722 ref_type = "object ";
2723 else {
2724 free(buffer);
2725 return NULL;
2726 }
2727 ref_length = strlen(ref_type);
2728
2729 if (ref_length + 40 > isize ||
2730 memcmp(buffer, ref_type, ref_length) ||
2731 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2732 free(buffer);
2733 return NULL;
2734 }
2735 free(buffer);
2736 /* Now we have the ID of the referred-to object in
2737 * actual_sha1. Check again. */
2738 }
2739}
2740
2741static void write_sha1_file_prepare(const void *buf, unsigned long len,
2742 const char *type, unsigned char *sha1,
2743 char *hdr, int *hdrlen)
2744{
2745 git_SHA_CTX c;
2746
2747 /* Generate the header */
2748 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2749
2750 /* Sha1.. */
2751 git_SHA1_Init(&c);
2752 git_SHA1_Update(&c, hdr, *hdrlen);
2753 git_SHA1_Update(&c, buf, len);
2754 git_SHA1_Final(sha1, &c);
2755}
2756
2757/*
2758 * Move the just written object into its final resting place.
2759 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2760 * "moving" is only a part of what it does, when no patch between
2761 * master to pu changes the call sites of this function.
2762 */
2763int move_temp_to_file(const char *tmpfile, const char *filename)
2764{
2765 int ret = 0;
2766
2767 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2768 goto try_rename;
2769 else if (link(tmpfile, filename))
2770 ret = errno;
2771
2772 /*
2773 * Coda hack - coda doesn't like cross-directory links,
2774 * so we fall back to a rename, which will mean that it
2775 * won't be able to check collisions, but that's not a
2776 * big deal.
2777 *
2778 * The same holds for FAT formatted media.
2779 *
2780 * When this succeeds, we just return. We have nothing
2781 * left to unlink.
2782 */
2783 if (ret && ret != EEXIST) {
2784 try_rename:
2785 if (!rename(tmpfile, filename))
2786 goto out;
2787 ret = errno;
2788 }
2789 unlink_or_warn(tmpfile);
2790 if (ret) {
2791 if (ret != EEXIST) {
2792 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2793 }
2794 /* FIXME!!! Collision check here ? */
2795 }
2796
2797out:
2798 if (adjust_shared_perm(filename))
2799 return error("unable to set permission to '%s'", filename);
2800 return 0;
2801}
2802
2803static int write_buffer(int fd, const void *buf, size_t len)
2804{
2805 if (write_in_full(fd, buf, len) < 0)
2806 return error("file write error (%s)", strerror(errno));
2807 return 0;
2808}
2809
2810int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2811 unsigned char *sha1)
2812{
2813 char hdr[32];
2814 int hdrlen;
2815 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2816 return 0;
2817}
2818
2819/* Finalize a file on disk, and close it. */
2820static void close_sha1_file(int fd)
2821{
2822 if (fsync_object_files)
2823 fsync_or_die(fd, "sha1 file");
2824 if (close(fd) != 0)
2825 die_errno("error when closing sha1 file");
2826}
2827
2828/* Size of directory component, including the ending '/' */
2829static inline int directory_size(const char *filename)
2830{
2831 const char *s = strrchr(filename, '/');
2832 if (!s)
2833 return 0;
2834 return s - filename + 1;
2835}
2836
2837/*
2838 * This creates a temporary file in the same directory as the final
2839 * 'filename'
2840 *
2841 * We want to avoid cross-directory filename renames, because those
2842 * can have problems on various filesystems (FAT, NFS, Coda).
2843 */
2844static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2845{
2846 int fd, dirlen = directory_size(filename);
2847
2848 if (dirlen + 20 > bufsiz) {
2849 errno = ENAMETOOLONG;
2850 return -1;
2851 }
2852 memcpy(buffer, filename, dirlen);
2853 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2854 fd = git_mkstemp_mode(buffer, 0444);
2855 if (fd < 0 && dirlen && errno == ENOENT) {
2856 /* Make sure the directory exists */
2857 memcpy(buffer, filename, dirlen);
2858 buffer[dirlen-1] = 0;
2859 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2860 return -1;
2861
2862 /* Try again */
2863 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2864 fd = git_mkstemp_mode(buffer, 0444);
2865 }
2866 return fd;
2867}
2868
2869static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2870 const void *buf, unsigned long len, time_t mtime)
2871{
2872 int fd, ret;
2873 unsigned char compressed[4096];
2874 git_zstream stream;
2875 git_SHA_CTX c;
2876 unsigned char parano_sha1[20];
2877 char *filename;
2878 static char tmp_file[PATH_MAX];
2879
2880 filename = sha1_file_name(sha1);
2881 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
2882 if (fd < 0) {
2883 if (errno == EACCES)
2884 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2885 else
2886 return error("unable to create temporary file: %s", strerror(errno));
2887 }
2888
2889 /* Set it up */
2890 memset(&stream, 0, sizeof(stream));
2891 git_deflate_init(&stream, zlib_compression_level);
2892 stream.next_out = compressed;
2893 stream.avail_out = sizeof(compressed);
2894 git_SHA1_Init(&c);
2895
2896 /* First header.. */
2897 stream.next_in = (unsigned char *)hdr;
2898 stream.avail_in = hdrlen;
2899 while (git_deflate(&stream, 0) == Z_OK)
2900 ; /* nothing */
2901 git_SHA1_Update(&c, hdr, hdrlen);
2902
2903 /* Then the data itself.. */
2904 stream.next_in = (void *)buf;
2905 stream.avail_in = len;
2906 do {
2907 unsigned char *in0 = stream.next_in;
2908 ret = git_deflate(&stream, Z_FINISH);
2909 git_SHA1_Update(&c, in0, stream.next_in - in0);
2910 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2911 die("unable to write sha1 file");
2912 stream.next_out = compressed;
2913 stream.avail_out = sizeof(compressed);
2914 } while (ret == Z_OK);
2915
2916 if (ret != Z_STREAM_END)
2917 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2918 ret = git_deflate_end_gently(&stream);
2919 if (ret != Z_OK)
2920 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2921 git_SHA1_Final(parano_sha1, &c);
2922 if (hashcmp(sha1, parano_sha1) != 0)
2923 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2924
2925 close_sha1_file(fd);
2926
2927 if (mtime) {
2928 struct utimbuf utb;
2929 utb.actime = mtime;
2930 utb.modtime = mtime;
2931 if (utime(tmp_file, &utb) < 0)
2932 warning("failed utime() on %s: %s",
2933 tmp_file, strerror(errno));
2934 }
2935
2936 return move_temp_to_file(tmp_file, filename);
2937}
2938
2939int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2940{
2941 unsigned char sha1[20];
2942 char hdr[32];
2943 int hdrlen;
2944
2945 /* Normally if we have it in the pack then we do not bother writing
2946 * it out into .git/objects/??/?{38} file.
2947 */
2948 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2949 if (returnsha1)
2950 hashcpy(returnsha1, sha1);
2951 if (has_sha1_file(sha1))
2952 return 0;
2953 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2954}
2955
2956int force_object_loose(const unsigned char *sha1, time_t mtime)
2957{
2958 void *buf;
2959 unsigned long len;
2960 enum object_type type;
2961 char hdr[32];
2962 int hdrlen;
2963 int ret;
2964
2965 if (has_loose_object(sha1))
2966 return 0;
2967 buf = read_packed_sha1(sha1, &type, &len);
2968 if (!buf)
2969 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2970 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2971 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2972 free(buf);
2973
2974 return ret;
2975}
2976
2977int has_pack_index(const unsigned char *sha1)
2978{
2979 struct stat st;
2980 if (stat(sha1_pack_index_name(sha1), &st))
2981 return 0;
2982 return 1;
2983}
2984
2985int has_sha1_pack(const unsigned char *sha1)
2986{
2987 struct pack_entry e;
2988 return find_pack_entry(sha1, &e);
2989}
2990
2991int has_sha1_file(const unsigned char *sha1)
2992{
2993 struct pack_entry e;
2994
2995 if (find_pack_entry(sha1, &e))
2996 return 1;
2997 if (has_loose_object(sha1))
2998 return 1;
2999 reprepare_packed_git();
3000 return find_pack_entry(sha1, &e);
3001}
3002
3003static void check_tree(const void *buf, size_t size)
3004{
3005 struct tree_desc desc;
3006 struct name_entry entry;
3007
3008 init_tree_desc(&desc, buf, size);
3009 while (tree_entry(&desc, &entry))
3010 /* do nothing
3011 * tree_entry() will die() on malformed entries */
3012 ;
3013}
3014
3015static void check_commit(const void *buf, size_t size)
3016{
3017 struct commit c;
3018 memset(&c, 0, sizeof(c));
3019 if (parse_commit_buffer(&c, buf, size))
3020 die("corrupt commit");
3021}
3022
3023static void check_tag(const void *buf, size_t size)
3024{
3025 struct tag t;
3026 memset(&t, 0, sizeof(t));
3027 if (parse_tag_buffer(&t, buf, size))
3028 die("corrupt tag");
3029}
3030
3031static int index_mem(unsigned char *sha1, void *buf, size_t size,
3032 enum object_type type,
3033 const char *path, unsigned flags)
3034{
3035 int ret, re_allocated = 0;
3036 int write_object = flags & HASH_WRITE_OBJECT;
3037
3038 if (!type)
3039 type = OBJ_BLOB;
3040
3041 /*
3042 * Convert blobs to git internal format
3043 */
3044 if ((type == OBJ_BLOB) && path) {
3045 struct strbuf nbuf = STRBUF_INIT;
3046 if (convert_to_git(path, buf, size, &nbuf,
3047 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3048 buf = strbuf_detach(&nbuf, &size);
3049 re_allocated = 1;
3050 }
3051 }
3052 if (flags & HASH_FORMAT_CHECK) {
3053 if (type == OBJ_TREE)
3054 check_tree(buf, size);
3055 if (type == OBJ_COMMIT)
3056 check_commit(buf, size);
3057 if (type == OBJ_TAG)
3058 check_tag(buf, size);
3059 }
3060
3061 if (write_object)
3062 ret = write_sha1_file(buf, size, typename(type), sha1);
3063 else
3064 ret = hash_sha1_file(buf, size, typename(type), sha1);
3065 if (re_allocated)
3066 free(buf);
3067 return ret;
3068}
3069
3070static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3071 const char *path, unsigned flags)
3072{
3073 struct strbuf sbuf = STRBUF_INIT;
3074 int ret;
3075
3076 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3077 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3078 else
3079 ret = -1;
3080 strbuf_release(&sbuf);
3081 return ret;
3082}
3083
3084#define SMALL_FILE_SIZE (32*1024)
3085
3086static int index_core(unsigned char *sha1, int fd, size_t size,
3087 enum object_type type, const char *path,
3088 unsigned flags)
3089{
3090 int ret;
3091
3092 if (!size) {
3093 ret = index_mem(sha1, NULL, size, type, path, flags);
3094 } else if (size <= SMALL_FILE_SIZE) {
3095 char *buf = xmalloc(size);
3096 if (size == read_in_full(fd, buf, size))
3097 ret = index_mem(sha1, buf, size, type, path, flags);
3098 else
3099 ret = error("short read %s", strerror(errno));
3100 free(buf);
3101 } else {
3102 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3103 ret = index_mem(sha1, buf, size, type, path, flags);
3104 munmap(buf, size);
3105 }
3106 return ret;
3107}
3108
3109/*
3110 * This creates one packfile per large blob unless bulk-checkin
3111 * machinery is "plugged".
3112 *
3113 * This also bypasses the usual "convert-to-git" dance, and that is on
3114 * purpose. We could write a streaming version of the converting
3115 * functions and insert that before feeding the data to fast-import
3116 * (or equivalent in-core API described above). However, that is
3117 * somewhat complicated, as we do not know the size of the filter
3118 * result, which we need to know beforehand when writing a git object.
3119 * Since the primary motivation for trying to stream from the working
3120 * tree file and to avoid mmaping it in core is to deal with large
3121 * binary blobs, they generally do not want to get any conversion, and
3122 * callers should avoid this code path when filters are requested.
3123 */
3124static int index_stream(unsigned char *sha1, int fd, size_t size,
3125 enum object_type type, const char *path,
3126 unsigned flags)
3127{
3128 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3129}
3130
3131int index_fd(unsigned char *sha1, int fd, struct stat *st,
3132 enum object_type type, const char *path, unsigned flags)
3133{
3134 int ret;
3135 size_t size = xsize_t(st->st_size);
3136
3137 if (!S_ISREG(st->st_mode))
3138 ret = index_pipe(sha1, fd, type, path, flags);
3139 else if (size <= big_file_threshold || type != OBJ_BLOB ||
3140 (path && would_convert_to_git(path, NULL, 0, 0)))
3141 ret = index_core(sha1, fd, size, type, path, flags);
3142 else
3143 ret = index_stream(sha1, fd, size, type, path, flags);
3144 close(fd);
3145 return ret;
3146}
3147
3148int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3149{
3150 int fd;
3151 struct strbuf sb = STRBUF_INIT;
3152
3153 switch (st->st_mode & S_IFMT) {
3154 case S_IFREG:
3155 fd = open(path, O_RDONLY);
3156 if (fd < 0)
3157 return error("open(\"%s\"): %s", path,
3158 strerror(errno));
3159 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3160 return error("%s: failed to insert into database",
3161 path);
3162 break;
3163 case S_IFLNK:
3164 if (strbuf_readlink(&sb, path, st->st_size)) {
3165 char *errstr = strerror(errno);
3166 return error("readlink(\"%s\"): %s", path,
3167 errstr);
3168 }
3169 if (!(flags & HASH_WRITE_OBJECT))
3170 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3171 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3172 return error("%s: failed to insert into database",
3173 path);
3174 strbuf_release(&sb);
3175 break;
3176 case S_IFDIR:
3177 return resolve_gitlink_ref(path, "HEAD", sha1);
3178 default:
3179 return error("%s: unsupported file type", path);
3180 }
3181 return 0;
3182}
3183
3184int read_pack_header(int fd, struct pack_header *header)
3185{
3186 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3187 /* "eof before pack header was fully read" */
3188 return PH_ERROR_EOF;
3189
3190 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3191 /* "protocol error (pack signature mismatch detected)" */
3192 return PH_ERROR_PACK_SIGNATURE;
3193 if (!pack_version_ok(header->hdr_version))
3194 /* "protocol error (pack version unsupported)" */
3195 return PH_ERROR_PROTOCOL;
3196 return 0;
3197}
3198
3199void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3200{
3201 enum object_type type = sha1_object_info(sha1, NULL);
3202 if (type < 0)
3203 die("%s is not a valid object", sha1_to_hex(sha1));
3204 if (type != expect)
3205 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3206 typename(expect));
3207}