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