Merge branch 'master' of git://git.kernel.org/pub/scm/git/git
[git/jnareb-git.git] / sha1_file.c
blob6c0e251e02e20aed1eaa3bc1737e3e606cb4e23e
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
9 #include "cache.h"
10 #include "delta.h"
11 #include "pack.h"
12 #include "blob.h"
13 #include "commit.h"
14 #include "tag.h"
15 #include "tree.h"
16 #include "refs.h"
17 #include "pack-revindex.h"
18 #include "sha1-lookup.h"
20 #ifndef O_NOATIME
21 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
22 #define O_NOATIME 01000000
23 #else
24 #define O_NOATIME 0
25 #endif
26 #endif
28 #ifdef NO_C99_FORMAT
29 #define SZ_FMT "lu"
30 static unsigned long sz_fmt(size_t s) { return (unsigned long)s; }
31 #else
32 #define SZ_FMT "zu"
33 static size_t sz_fmt(size_t s) { return s; }
34 #endif
36 const unsigned char null_sha1[20];
38 const signed char hexval_table[256] = {
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
45 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
46 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
47 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
48 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
51 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
69 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
73 int get_sha1_hex(const char *hex, unsigned char *sha1)
75 int i;
76 for (i = 0; i < 20; i++) {
77 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
78 if (val & ~0xff)
79 return -1;
80 *sha1++ = val;
81 hex += 2;
83 return 0;
86 static inline int offset_1st_component(const char *path)
88 if (has_dos_drive_prefix(path))
89 return 2 + (path[2] == '/');
90 return *path == '/';
93 int safe_create_leading_directories(char *path)
95 char *pos = path + offset_1st_component(path);
96 struct stat st;
98 while (pos) {
99 pos = strchr(pos, '/');
100 if (!pos)
101 break;
102 while (*++pos == '/')
104 if (!*pos)
105 break;
106 *--pos = '\0';
107 if (!stat(path, &st)) {
108 /* path exists */
109 if (!S_ISDIR(st.st_mode)) {
110 *pos = '/';
111 return -3;
114 else if (mkdir(path, 0777)) {
115 *pos = '/';
116 return -1;
118 else if (adjust_shared_perm(path)) {
119 *pos = '/';
120 return -2;
122 *pos++ = '/';
124 return 0;
127 int safe_create_leading_directories_const(const char *path)
129 /* path points to cache entries, so xstrdup before messing with it */
130 char *buf = xstrdup(path);
131 int result = safe_create_leading_directories(buf);
132 free(buf);
133 return result;
136 char *sha1_to_hex(const unsigned char *sha1)
138 static int bufno;
139 static char hexbuffer[4][50];
140 static const char hex[] = "0123456789abcdef";
141 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
142 int i;
144 for (i = 0; i < 20; i++) {
145 unsigned int val = *sha1++;
146 *buf++ = hex[val >> 4];
147 *buf++ = hex[val & 0xf];
149 *buf = '\0';
151 return buffer;
154 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
156 int i;
157 for (i = 0; i < 20; i++) {
158 static char hex[] = "0123456789abcdef";
159 unsigned int val = sha1[i];
160 char *pos = pathbuf + i*2 + (i > 0);
161 *pos++ = hex[val >> 4];
162 *pos = hex[val & 0xf];
167 * NOTE! This returns a statically allocated buffer, so you have to be
168 * careful about using it. Do an "xstrdup()" if you need to save the
169 * filename.
171 * Also note that this returns the location for creating. Reading
172 * SHA1 file can happen from any alternate directory listed in the
173 * DB_ENVIRONMENT environment variable if it is not found in
174 * the primary object database.
176 char *sha1_file_name(const unsigned char *sha1)
178 static char *name, *base;
180 if (!base) {
181 const char *sha1_file_directory = get_object_directory();
182 int len = strlen(sha1_file_directory);
183 base = xmalloc(len + 60);
184 memcpy(base, sha1_file_directory, len);
185 memset(base+len, 0, 60);
186 base[len] = '/';
187 base[len+3] = '/';
188 name = base + len + 1;
190 fill_sha1_path(name, sha1);
191 return base;
194 static char *sha1_get_pack_name(const unsigned char *sha1,
195 char **name, char **base, const char *which)
197 static const char hex[] = "0123456789abcdef";
198 char *buf;
199 int i;
201 if (!*base) {
202 const char *sha1_file_directory = get_object_directory();
203 int len = strlen(sha1_file_directory);
204 *base = xmalloc(len + 60);
205 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
206 sha1_file_directory, which);
207 *name = *base + len + 11;
210 buf = *name;
212 for (i = 0; i < 20; i++) {
213 unsigned int val = *sha1++;
214 *buf++ = hex[val >> 4];
215 *buf++ = hex[val & 0xf];
218 return *base;
221 char *sha1_pack_name(const unsigned char *sha1)
223 static char *name, *base;
225 return sha1_get_pack_name(sha1, &name, &base, "pack");
228 char *sha1_pack_index_name(const unsigned char *sha1)
230 static char *name, *base;
232 return sha1_get_pack_name(sha1, &name, &base, "idx");
235 struct alternate_object_database *alt_odb_list;
236 static struct alternate_object_database **alt_odb_tail;
238 static void read_info_alternates(const char * alternates, int depth);
241 * Prepare alternate object database registry.
243 * The variable alt_odb_list points at the list of struct
244 * alternate_object_database. The elements on this list come from
245 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
246 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
247 * whose contents is similar to that environment variable but can be
248 * LF separated. Its base points at a statically allocated buffer that
249 * contains "/the/directory/corresponding/to/.git/objects/...", while
250 * its name points just after the slash at the end of ".git/objects/"
251 * in the example above, and has enough space to hold 40-byte hex
252 * SHA1, an extra slash for the first level indirection, and the
253 * terminating NUL.
255 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
257 const char *objdir = get_object_directory();
258 struct alternate_object_database *ent;
259 struct alternate_object_database *alt;
260 /* 43 = 40-byte + 2 '/' + terminating NUL */
261 int pfxlen = len;
262 int entlen = pfxlen + 43;
263 int base_len = -1;
265 if (!is_absolute_path(entry) && relative_base) {
266 /* Relative alt-odb */
267 if (base_len < 0)
268 base_len = strlen(relative_base) + 1;
269 entlen += base_len;
270 pfxlen += base_len;
272 ent = xmalloc(sizeof(*ent) + entlen);
274 if (!is_absolute_path(entry) && relative_base) {
275 memcpy(ent->base, relative_base, base_len - 1);
276 ent->base[base_len - 1] = '/';
277 memcpy(ent->base + base_len, entry, len);
279 else
280 memcpy(ent->base, entry, pfxlen);
282 ent->name = ent->base + pfxlen + 1;
283 ent->base[pfxlen + 3] = '/';
284 ent->base[pfxlen] = ent->base[entlen-1] = 0;
286 /* Detect cases where alternate disappeared */
287 if (!is_directory(ent->base)) {
288 error("object directory %s does not exist; "
289 "check .git/objects/info/alternates.",
290 ent->base);
291 free(ent);
292 return -1;
295 /* Prevent the common mistake of listing the same
296 * thing twice, or object directory itself.
298 for (alt = alt_odb_list; alt; alt = alt->next) {
299 if (!memcmp(ent->base, alt->base, pfxlen)) {
300 free(ent);
301 return -1;
304 if (!memcmp(ent->base, objdir, pfxlen)) {
305 free(ent);
306 return -1;
309 /* add the alternate entry */
310 *alt_odb_tail = ent;
311 alt_odb_tail = &(ent->next);
312 ent->next = NULL;
314 /* recursively add alternates */
315 read_info_alternates(ent->base, depth + 1);
317 ent->base[pfxlen] = '/';
319 return 0;
322 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
323 const char *relative_base, int depth)
325 const char *cp, *last;
327 if (depth > 5) {
328 error("%s: ignoring alternate object stores, nesting too deep.",
329 relative_base);
330 return;
333 last = alt;
334 while (last < ep) {
335 cp = last;
336 if (cp < ep && *cp == '#') {
337 while (cp < ep && *cp != sep)
338 cp++;
339 last = cp + 1;
340 continue;
342 while (cp < ep && *cp != sep)
343 cp++;
344 if (last != cp) {
345 if (!is_absolute_path(last) && depth) {
346 error("%s: ignoring relative alternate object store %s",
347 relative_base, last);
348 } else {
349 link_alt_odb_entry(last, cp - last,
350 relative_base, depth);
353 while (cp < ep && *cp == sep)
354 cp++;
355 last = cp;
359 static void read_info_alternates(const char * relative_base, int depth)
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;
371 sprintf(path, "%s/%s", relative_base, alt_file_name);
372 fd = open(path, O_RDONLY);
373 if (fd < 0)
374 return;
375 if (fstat(fd, &st) || (st.st_size == 0)) {
376 close(fd);
377 return;
379 mapsz = xsize_t(st.st_size);
380 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
381 close(fd);
383 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
385 munmap(map, mapsz);
388 void add_to_alternates_file(const char *reference)
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/objects\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, alt + strlen(alt), '\n', NULL, 0);
400 void foreach_alt_odb(alt_odb_fn fn, void *cb)
402 struct alternate_object_database *ent;
404 prepare_alt_odb();
405 for (ent = alt_odb_list; ent; ent = ent->next)
406 if (fn(ent, cb))
407 return;
410 void prepare_alt_odb(void)
412 const char *alt;
414 if (alt_odb_tail)
415 return;
417 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
418 if (!alt) alt = "";
420 alt_odb_tail = &alt_odb_list;
421 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
423 read_info_alternates(get_object_directory(), 0);
426 static int has_loose_object_local(const unsigned char *sha1)
428 char *name = sha1_file_name(sha1);
429 return !access(name, F_OK);
432 int has_loose_object_nonlocal(const unsigned char *sha1)
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;
441 return 0;
444 static int has_loose_object(const unsigned char *sha1)
446 return has_loose_object_local(sha1) ||
447 has_loose_object_nonlocal(sha1);
450 static unsigned int pack_used_ctr;
451 static unsigned int pack_mmap_calls;
452 static unsigned int peak_pack_open_windows;
453 static unsigned int pack_open_windows;
454 static size_t peak_pack_mapped;
455 static size_t pack_mapped;
456 struct packed_git *packed_git;
458 void pack_report(void)
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));
479 static int check_packed_git_idx(const char *path, struct packed_git *p)
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 = open(path, O_RDONLY);
486 struct stat st;
488 if (fd < 0)
489 return -1;
490 if (fstat(fd, &st)) {
491 close(fd);
492 return -1;
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);
499 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
500 close(fd);
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);
512 } else
513 version = 1;
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);
525 nr = n;
528 if (version == 1) {
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
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);
540 } else if (version == 2) {
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.
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);
562 if (idx_size != min_size &&
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.
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);
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;
581 int open_pack_index(struct packed_git *p)
583 char *idx_name;
584 int ret;
586 if (p->index_data)
587 return 0;
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;
596 static 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)
601 struct pack_window *w, *w_l;
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;
611 w_l = w;
615 static int unuse_one_window(struct packed_git *current, int keep_fd)
617 struct packed_git *p, *lru_p = NULL;
618 struct pack_window *lru_w = NULL, *lru_l = NULL;
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 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
632 close(lru_p->pack_fd);
633 lru_p->pack_fd = -1;
636 free(lru_w);
637 pack_open_windows--;
638 return 1;
640 return 0;
643 void release_pack_memory(size_t need, int fd)
645 size_t cur = pack_mapped;
646 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
647 ; /* nothing */
650 void close_pack_windows(struct packed_git *p)
652 while (p->windows) {
653 struct pack_window *w = p->windows;
655 if (w->inuse_cnt)
656 die("pack '%s' still has open windows to it",
657 p->pack_name);
658 munmap(w->base, w->len);
659 pack_mapped -= w->len;
660 pack_open_windows--;
661 p->windows = w->next;
662 free(w);
666 void unuse_pack(struct pack_window **w_cursor)
668 struct pack_window *w = *w_cursor;
669 if (w) {
670 w->inuse_cnt--;
671 *w_cursor = NULL;
676 * Do not call this directly as this leaks p->pack_fd on error return;
677 * call open_packed_git() instead.
679 static int open_packed_git_1(struct packed_git *p)
681 struct stat st;
682 struct pack_header hdr;
683 unsigned char sha1[20];
684 unsigned char *idx_sha1;
685 long fd_flag;
687 if (!p->index_data && open_pack_index(p))
688 return error("packfile %s index unavailable", p->pack_name);
690 p->pack_fd = open(p->pack_name, O_RDONLY);
691 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
692 return -1;
694 /* If we created the struct before we had the pack we lack size. */
695 if (!p->pack_size) {
696 if (!S_ISREG(st.st_mode))
697 return error("packfile %s not a regular file", p->pack_name);
698 p->pack_size = st.st_size;
699 } else if (p->pack_size != st.st_size)
700 return error("packfile %s size changed", p->pack_name);
702 /* We leave these file descriptors open with sliding mmap;
703 * there is no point keeping them open across exec(), though.
705 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
706 if (fd_flag < 0)
707 return error("cannot determine file descriptor flags");
708 fd_flag |= FD_CLOEXEC;
709 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
710 return error("cannot set FD_CLOEXEC");
712 /* Verify we recognize this pack file format. */
713 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
714 return error("file %s is far too short to be a packfile", p->pack_name);
715 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
716 return error("file %s is not a GIT packfile", p->pack_name);
717 if (!pack_version_ok(hdr.hdr_version))
718 return error("packfile %s is version %"PRIu32" and not"
719 " supported (try upgrading GIT to a newer version)",
720 p->pack_name, ntohl(hdr.hdr_version));
722 /* Verify the pack matches its index. */
723 if (p->num_objects != ntohl(hdr.hdr_entries))
724 return error("packfile %s claims to have %"PRIu32" objects"
725 " while index indicates %"PRIu32" objects",
726 p->pack_name, ntohl(hdr.hdr_entries),
727 p->num_objects);
728 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
729 return error("end of packfile %s is unavailable", p->pack_name);
730 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
731 return error("packfile %s signature is unavailable", p->pack_name);
732 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
733 if (hashcmp(sha1, idx_sha1))
734 return error("packfile %s does not match index", p->pack_name);
735 return 0;
738 static int open_packed_git(struct packed_git *p)
740 if (!open_packed_git_1(p))
741 return 0;
742 if (p->pack_fd != -1) {
743 close(p->pack_fd);
744 p->pack_fd = -1;
746 return -1;
749 static int in_window(struct pack_window *win, off_t offset)
751 /* We must promise at least 20 bytes (one hash) after the
752 * offset is available from this window, otherwise the offset
753 * is not actually in this window and a different window (which
754 * has that one hash excess) must be used. This is to support
755 * the object header and delta base parsing routines below.
757 off_t win_off = win->offset;
758 return win_off <= offset
759 && (offset + 20) <= (win_off + win->len);
762 unsigned char* use_pack(struct packed_git *p,
763 struct pack_window **w_cursor,
764 off_t offset,
765 unsigned int *left)
767 struct pack_window *win = *w_cursor;
769 if (p->pack_fd == -1 && open_packed_git(p))
770 die("packfile %s cannot be accessed", p->pack_name);
772 /* Since packfiles end in a hash of their content and its
773 * pointless to ask for an offset into the middle of that
774 * hash, and the in_window function above wouldn't match
775 * don't allow an offset too close to the end of the file.
777 if (offset > (p->pack_size - 20))
778 die("offset beyond end of packfile (truncated pack?)");
780 if (!win || !in_window(win, offset)) {
781 if (win)
782 win->inuse_cnt--;
783 for (win = p->windows; win; win = win->next) {
784 if (in_window(win, offset))
785 break;
787 if (!win) {
788 size_t window_align = packed_git_window_size / 2;
789 off_t len;
790 win = xcalloc(1, sizeof(*win));
791 win->offset = (offset / window_align) * window_align;
792 len = p->pack_size - win->offset;
793 if (len > packed_git_window_size)
794 len = packed_git_window_size;
795 win->len = (size_t)len;
796 pack_mapped += win->len;
797 while (packed_git_limit < pack_mapped
798 && unuse_one_window(p, p->pack_fd))
799 ; /* nothing */
800 win->base = xmmap(NULL, win->len,
801 PROT_READ, MAP_PRIVATE,
802 p->pack_fd, win->offset);
803 if (win->base == MAP_FAILED)
804 die("packfile %s cannot be mapped: %s",
805 p->pack_name,
806 strerror(errno));
807 pack_mmap_calls++;
808 pack_open_windows++;
809 if (pack_mapped > peak_pack_mapped)
810 peak_pack_mapped = pack_mapped;
811 if (pack_open_windows > peak_pack_open_windows)
812 peak_pack_open_windows = pack_open_windows;
813 win->next = p->windows;
814 p->windows = win;
817 if (win != *w_cursor) {
818 win->last_used = pack_used_ctr++;
819 win->inuse_cnt++;
820 *w_cursor = win;
822 offset -= win->offset;
823 if (left)
824 *left = win->len - xsize_t(offset);
825 return win->base + offset;
828 static struct packed_git *alloc_packed_git(int extra)
830 struct packed_git *p = xmalloc(sizeof(*p) + extra);
831 memset(p, 0, sizeof(*p));
832 p->pack_fd = -1;
833 return p;
836 struct packed_git *add_packed_git(const char *path, int path_len, int local)
838 struct stat st;
839 struct packed_git *p = alloc_packed_git(path_len + 2);
842 * Make sure a corresponding .pack file exists and that
843 * the index looks sane.
845 path_len -= strlen(".idx");
846 if (path_len < 1) {
847 free(p);
848 return NULL;
850 memcpy(p->pack_name, path, path_len);
852 strcpy(p->pack_name + path_len, ".keep");
853 if (!access(p->pack_name, F_OK))
854 p->pack_keep = 1;
856 strcpy(p->pack_name + path_len, ".pack");
857 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
858 free(p);
859 return NULL;
862 /* ok, it looks sane as far as we can check without
863 * actually mapping the pack file.
865 p->pack_size = st.st_size;
866 p->pack_local = local;
867 p->mtime = st.st_mtime;
868 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
869 hashclr(p->sha1);
870 return p;
873 struct packed_git *parse_pack_index(unsigned char *sha1)
875 const char *idx_path = sha1_pack_index_name(sha1);
876 const char *path = sha1_pack_name(sha1);
877 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
879 strcpy(p->pack_name, path);
880 hashcpy(p->sha1, sha1);
881 if (check_packed_git_idx(idx_path, p)) {
882 free(p);
883 return NULL;
886 return p;
889 void install_packed_git(struct packed_git *pack)
891 pack->next = packed_git;
892 packed_git = pack;
895 static void prepare_packed_git_one(char *objdir, int local)
897 /* Ensure that this buffer is large enough so that we can
898 append "/pack/" without clobbering the stack even if
899 strlen(objdir) were PATH_MAX. */
900 char path[PATH_MAX + 1 + 4 + 1 + 1];
901 int len;
902 DIR *dir;
903 struct dirent *de;
905 sprintf(path, "%s/pack", objdir);
906 len = strlen(path);
907 dir = opendir(path);
908 if (!dir) {
909 if (errno != ENOENT)
910 error("unable to open object pack directory: %s: %s",
911 path, strerror(errno));
912 return;
914 path[len++] = '/';
915 while ((de = readdir(dir)) != NULL) {
916 int namelen = strlen(de->d_name);
917 struct packed_git *p;
919 if (!has_extension(de->d_name, ".idx"))
920 continue;
922 if (len + namelen + 1 > sizeof(path))
923 continue;
925 /* Don't reopen a pack we already have. */
926 strcpy(path + len, de->d_name);
927 for (p = packed_git; p; p = p->next) {
928 if (!memcmp(path, p->pack_name, len + namelen - 4))
929 break;
931 if (p)
932 continue;
933 /* See if it really is a valid .idx file with corresponding
934 * .pack file that we can map.
936 p = add_packed_git(path, len + namelen, local);
937 if (!p)
938 continue;
939 install_packed_git(p);
941 closedir(dir);
944 static int sort_pack(const void *a_, const void *b_)
946 struct packed_git *a = *((struct packed_git **)a_);
947 struct packed_git *b = *((struct packed_git **)b_);
948 int st;
951 * Local packs tend to contain objects specific to our
952 * variant of the project than remote ones. In addition,
953 * remote ones could be on a network mounted filesystem.
954 * Favor local ones for these reasons.
956 st = a->pack_local - b->pack_local;
957 if (st)
958 return -st;
961 * Younger packs tend to contain more recent objects,
962 * and more recent objects tend to get accessed more
963 * often.
965 if (a->mtime < b->mtime)
966 return 1;
967 else if (a->mtime == b->mtime)
968 return 0;
969 return -1;
972 static void rearrange_packed_git(void)
974 struct packed_git **ary, *p;
975 int i, n;
977 for (n = 0, p = packed_git; p; p = p->next)
978 n++;
979 if (n < 2)
980 return;
982 /* prepare an array of packed_git for easier sorting */
983 ary = xcalloc(n, sizeof(struct packed_git *));
984 for (n = 0, p = packed_git; p; p = p->next)
985 ary[n++] = p;
987 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
989 /* link them back again */
990 for (i = 0; i < n - 1; i++)
991 ary[i]->next = ary[i + 1];
992 ary[n - 1]->next = NULL;
993 packed_git = ary[0];
995 free(ary);
998 static int prepare_packed_git_run_once = 0;
999 void prepare_packed_git(void)
1001 struct alternate_object_database *alt;
1003 if (prepare_packed_git_run_once)
1004 return;
1005 prepare_packed_git_one(get_object_directory(), 1);
1006 prepare_alt_odb();
1007 for (alt = alt_odb_list; alt; alt = alt->next) {
1008 alt->name[-1] = 0;
1009 prepare_packed_git_one(alt->base, 0);
1010 alt->name[-1] = '/';
1012 rearrange_packed_git();
1013 prepare_packed_git_run_once = 1;
1016 void reprepare_packed_git(void)
1018 discard_revindex();
1019 prepare_packed_git_run_once = 0;
1020 prepare_packed_git();
1023 static void mark_bad_packed_object(struct packed_git *p,
1024 const unsigned char *sha1)
1026 unsigned i;
1027 for (i = 0; i < p->num_bad_objects; i++)
1028 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1029 return;
1030 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1031 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1032 p->num_bad_objects++;
1035 static int has_packed_and_bad(const unsigned char *sha1)
1037 struct packed_git *p;
1038 unsigned i;
1040 for (p = packed_git; p; p = p->next)
1041 for (i = 0; i < p->num_bad_objects; i++)
1042 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1043 return 1;
1044 return 0;
1047 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1049 unsigned char real_sha1[20];
1050 hash_sha1_file(map, size, type, real_sha1);
1051 return hashcmp(sha1, real_sha1) ? -1 : 0;
1054 static int git_open_noatime(const char *name)
1056 static int sha1_file_open_flag = O_NOATIME;
1057 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1059 /* Might the failure be due to O_NOATIME? */
1060 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1061 fd = open(name, O_RDONLY);
1062 if (fd >= 0)
1063 sha1_file_open_flag = 0;
1065 return fd;
1068 static int open_sha1_file(const unsigned char *sha1)
1070 int fd;
1071 char *name = sha1_file_name(sha1);
1072 struct alternate_object_database *alt;
1074 fd = git_open_noatime(name);
1075 if (fd >= 0)
1076 return fd;
1078 prepare_alt_odb();
1079 errno = ENOENT;
1080 for (alt = alt_odb_list; alt; alt = alt->next) {
1081 name = alt->name;
1082 fill_sha1_path(name, sha1);
1083 fd = git_open_noatime(alt->base);
1084 if (fd >= 0)
1085 return fd;
1087 return -1;
1090 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1092 void *map;
1093 int fd;
1095 fd = open_sha1_file(sha1);
1096 map = NULL;
1097 if (fd >= 0) {
1098 struct stat st;
1100 if (!fstat(fd, &st)) {
1101 *size = xsize_t(st.st_size);
1102 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1104 close(fd);
1106 return map;
1109 static int legacy_loose_object(unsigned char *map)
1111 unsigned int word;
1114 * Is it a zlib-compressed buffer? If so, the first byte
1115 * must be 0x78 (15-bit window size, deflated), and the
1116 * first 16-bit word is evenly divisible by 31
1118 word = (map[0] << 8) + map[1];
1119 if (map[0] == 0x78 && !(word % 31))
1120 return 1;
1121 else
1122 return 0;
1125 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1126 unsigned long len, enum object_type *type, unsigned long *sizep)
1128 unsigned shift;
1129 unsigned char c;
1130 unsigned long size;
1131 unsigned long used = 0;
1133 c = buf[used++];
1134 *type = (c >> 4) & 7;
1135 size = c & 15;
1136 shift = 4;
1137 while (c & 0x80) {
1138 if (len <= used || sizeof(long) * 8 <= shift) {
1139 error("bad object header");
1140 return 0;
1142 c = buf[used++];
1143 size += (c & 0x7f) << shift;
1144 shift += 7;
1146 *sizep = size;
1147 return used;
1150 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1152 unsigned long size, used;
1153 static const char valid_loose_object_type[8] = {
1154 0, /* OBJ_EXT */
1155 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1156 0, /* "delta" and others are invalid in a loose object */
1158 enum object_type type;
1160 /* Get the data stream */
1161 memset(stream, 0, sizeof(*stream));
1162 stream->next_in = map;
1163 stream->avail_in = mapsize;
1164 stream->next_out = buffer;
1165 stream->avail_out = bufsiz;
1167 if (legacy_loose_object(map)) {
1168 inflateInit(stream);
1169 return inflate(stream, 0);
1174 * There used to be a second loose object header format which
1175 * was meant to mimic the in-pack format, allowing for direct
1176 * copy of the object data. This format turned up not to be
1177 * really worth it and we don't write it any longer. But we
1178 * can still read it.
1180 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1181 if (!used || !valid_loose_object_type[type])
1182 return -1;
1183 map += used;
1184 mapsize -= used;
1186 /* Set up the stream for the rest.. */
1187 stream->next_in = map;
1188 stream->avail_in = mapsize;
1189 inflateInit(stream);
1191 /* And generate the fake traditional header */
1192 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1193 typename(type), size);
1194 return 0;
1197 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1199 int bytes = strlen(buffer) + 1;
1200 unsigned char *buf = xmalloc(1+size);
1201 unsigned long n;
1202 int status = Z_OK;
1204 n = stream->total_out - bytes;
1205 if (n > size)
1206 n = size;
1207 memcpy(buf, (char *) buffer + bytes, n);
1208 bytes = n;
1209 if (bytes <= size) {
1211 * The above condition must be (bytes <= size), not
1212 * (bytes < size). In other words, even though we
1213 * expect no more output and set avail_out to zer0,
1214 * the input zlib stream may have bytes that express
1215 * "this concludes the stream", and we *do* want to
1216 * eat that input.
1218 * Otherwise we would not be able to test that we
1219 * consumed all the input to reach the expected size;
1220 * we also want to check that zlib tells us that all
1221 * went well with status == Z_STREAM_END at the end.
1223 stream->next_out = buf + bytes;
1224 stream->avail_out = size - bytes;
1225 while (status == Z_OK)
1226 status = inflate(stream, Z_FINISH);
1228 buf[size] = 0;
1229 if (status == Z_STREAM_END && !stream->avail_in) {
1230 inflateEnd(stream);
1231 return buf;
1234 if (status < 0)
1235 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1236 else if (stream->avail_in)
1237 error("garbage at end of loose object '%s'",
1238 sha1_to_hex(sha1));
1239 free(buf);
1240 return NULL;
1244 * We used to just use "sscanf()", but that's actually way
1245 * too permissive for what we want to check. So do an anal
1246 * object header parse by hand.
1248 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1250 char type[10];
1251 int i;
1252 unsigned long size;
1255 * The type can be at most ten bytes (including the
1256 * terminating '\0' that we add), and is followed by
1257 * a space.
1259 i = 0;
1260 for (;;) {
1261 char c = *hdr++;
1262 if (c == ' ')
1263 break;
1264 type[i++] = c;
1265 if (i >= sizeof(type))
1266 return -1;
1268 type[i] = 0;
1271 * The length must follow immediately, and be in canonical
1272 * decimal format (ie "010" is not valid).
1274 size = *hdr++ - '0';
1275 if (size > 9)
1276 return -1;
1277 if (size) {
1278 for (;;) {
1279 unsigned long c = *hdr - '0';
1280 if (c > 9)
1281 break;
1282 hdr++;
1283 size = size * 10 + c;
1286 *sizep = size;
1289 * The length must be followed by a zero byte
1291 return *hdr ? -1 : type_from_string(type);
1294 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1296 int ret;
1297 z_stream stream;
1298 char hdr[8192];
1300 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1301 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1302 return NULL;
1304 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1307 unsigned long get_size_from_delta(struct packed_git *p,
1308 struct pack_window **w_curs,
1309 off_t curpos)
1311 const unsigned char *data;
1312 unsigned char delta_head[20], *in;
1313 z_stream stream;
1314 int st;
1316 memset(&stream, 0, sizeof(stream));
1317 stream.next_out = delta_head;
1318 stream.avail_out = sizeof(delta_head);
1320 inflateInit(&stream);
1321 do {
1322 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1323 stream.next_in = in;
1324 st = inflate(&stream, Z_FINISH);
1325 curpos += stream.next_in - in;
1326 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1327 stream.total_out < sizeof(delta_head));
1328 inflateEnd(&stream);
1329 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1330 error("delta data unpack-initial failed");
1331 return 0;
1334 /* Examine the initial part of the delta to figure out
1335 * the result size.
1337 data = delta_head;
1339 /* ignore base size */
1340 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1342 /* Read the result size */
1343 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1346 static off_t get_delta_base(struct packed_git *p,
1347 struct pack_window **w_curs,
1348 off_t *curpos,
1349 enum object_type type,
1350 off_t delta_obj_offset)
1352 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1353 off_t base_offset;
1355 /* use_pack() assured us we have [base_info, base_info + 20)
1356 * as a range that we can look at without walking off the
1357 * end of the mapped window. Its actually the hash size
1358 * that is assured. An OFS_DELTA longer than the hash size
1359 * is stupid, as then a REF_DELTA would be smaller to store.
1361 if (type == OBJ_OFS_DELTA) {
1362 unsigned used = 0;
1363 unsigned char c = base_info[used++];
1364 base_offset = c & 127;
1365 while (c & 128) {
1366 base_offset += 1;
1367 if (!base_offset || MSB(base_offset, 7))
1368 return 0; /* overflow */
1369 c = base_info[used++];
1370 base_offset = (base_offset << 7) + (c & 127);
1372 base_offset = delta_obj_offset - base_offset;
1373 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1374 return 0; /* out of bound */
1375 *curpos += used;
1376 } else if (type == OBJ_REF_DELTA) {
1377 /* The base entry _must_ be in the same pack */
1378 base_offset = find_pack_entry_one(base_info, p);
1379 *curpos += 20;
1380 } else
1381 die("I am totally screwed");
1382 return base_offset;
1385 /* forward declaration for a mutually recursive function */
1386 static int packed_object_info(struct packed_git *p, off_t offset,
1387 unsigned long *sizep);
1389 static int packed_delta_info(struct packed_git *p,
1390 struct pack_window **w_curs,
1391 off_t curpos,
1392 enum object_type type,
1393 off_t obj_offset,
1394 unsigned long *sizep)
1396 off_t base_offset;
1398 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1399 if (!base_offset)
1400 return OBJ_BAD;
1401 type = packed_object_info(p, base_offset, NULL);
1402 if (type <= OBJ_NONE) {
1403 struct revindex_entry *revidx;
1404 const unsigned char *base_sha1;
1405 revidx = find_pack_revindex(p, base_offset);
1406 if (!revidx)
1407 return OBJ_BAD;
1408 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1409 mark_bad_packed_object(p, base_sha1);
1410 type = sha1_object_info(base_sha1, NULL);
1411 if (type <= OBJ_NONE)
1412 return OBJ_BAD;
1415 /* We choose to only get the type of the base object and
1416 * ignore potentially corrupt pack file that expects the delta
1417 * based on a base with a wrong size. This saves tons of
1418 * inflate() calls.
1420 if (sizep) {
1421 *sizep = get_size_from_delta(p, w_curs, curpos);
1422 if (*sizep == 0)
1423 type = OBJ_BAD;
1426 return type;
1429 static int unpack_object_header(struct packed_git *p,
1430 struct pack_window **w_curs,
1431 off_t *curpos,
1432 unsigned long *sizep)
1434 unsigned char *base;
1435 unsigned int left;
1436 unsigned long used;
1437 enum object_type type;
1439 /* use_pack() assures us we have [base, base + 20) available
1440 * as a range that we can look at at. (Its actually the hash
1441 * size that is assured.) With our object header encoding
1442 * the maximum deflated object size is 2^137, which is just
1443 * insane, so we know won't exceed what we have been given.
1445 base = use_pack(p, w_curs, *curpos, &left);
1446 used = unpack_object_header_buffer(base, left, &type, sizep);
1447 if (!used) {
1448 type = OBJ_BAD;
1449 } else
1450 *curpos += used;
1452 return type;
1455 const char *packed_object_info_detail(struct packed_git *p,
1456 off_t obj_offset,
1457 unsigned long *size,
1458 unsigned long *store_size,
1459 unsigned int *delta_chain_length,
1460 unsigned char *base_sha1)
1462 struct pack_window *w_curs = NULL;
1463 off_t curpos;
1464 unsigned long dummy;
1465 unsigned char *next_sha1;
1466 enum object_type type;
1467 struct revindex_entry *revidx;
1469 *delta_chain_length = 0;
1470 curpos = obj_offset;
1471 type = unpack_object_header(p, &w_curs, &curpos, size);
1473 revidx = find_pack_revindex(p, obj_offset);
1474 *store_size = revidx[1].offset - obj_offset;
1476 for (;;) {
1477 switch (type) {
1478 default:
1479 die("pack %s contains unknown object type %d",
1480 p->pack_name, type);
1481 case OBJ_COMMIT:
1482 case OBJ_TREE:
1483 case OBJ_BLOB:
1484 case OBJ_TAG:
1485 unuse_pack(&w_curs);
1486 return typename(type);
1487 case OBJ_OFS_DELTA:
1488 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1489 if (!obj_offset)
1490 die("pack %s contains bad delta base reference of type %s",
1491 p->pack_name, typename(type));
1492 if (*delta_chain_length == 0) {
1493 revidx = find_pack_revindex(p, obj_offset);
1494 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1496 break;
1497 case OBJ_REF_DELTA:
1498 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1499 if (*delta_chain_length == 0)
1500 hashcpy(base_sha1, next_sha1);
1501 obj_offset = find_pack_entry_one(next_sha1, p);
1502 break;
1504 (*delta_chain_length)++;
1505 curpos = obj_offset;
1506 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1510 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1511 unsigned long *sizep)
1513 struct pack_window *w_curs = NULL;
1514 unsigned long size;
1515 off_t curpos = obj_offset;
1516 enum object_type type;
1518 type = unpack_object_header(p, &w_curs, &curpos, &size);
1520 switch (type) {
1521 case OBJ_OFS_DELTA:
1522 case OBJ_REF_DELTA:
1523 type = packed_delta_info(p, &w_curs, curpos,
1524 type, obj_offset, sizep);
1525 break;
1526 case OBJ_COMMIT:
1527 case OBJ_TREE:
1528 case OBJ_BLOB:
1529 case OBJ_TAG:
1530 if (sizep)
1531 *sizep = size;
1532 break;
1533 default:
1534 error("unknown object type %i at offset %"PRIuMAX" in %s",
1535 type, (uintmax_t)obj_offset, p->pack_name);
1536 type = OBJ_BAD;
1538 unuse_pack(&w_curs);
1539 return type;
1542 static void *unpack_compressed_entry(struct packed_git *p,
1543 struct pack_window **w_curs,
1544 off_t curpos,
1545 unsigned long size)
1547 int st;
1548 z_stream stream;
1549 unsigned char *buffer, *in;
1551 buffer = xmalloc(size + 1);
1552 buffer[size] = 0;
1553 memset(&stream, 0, sizeof(stream));
1554 stream.next_out = buffer;
1555 stream.avail_out = size;
1557 inflateInit(&stream);
1558 do {
1559 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1560 stream.next_in = in;
1561 st = inflate(&stream, Z_FINISH);
1562 curpos += stream.next_in - in;
1563 } while (st == Z_OK || st == Z_BUF_ERROR);
1564 inflateEnd(&stream);
1565 if ((st != Z_STREAM_END) || stream.total_out != size) {
1566 free(buffer);
1567 return NULL;
1570 return buffer;
1573 #define MAX_DELTA_CACHE (256)
1575 static size_t delta_base_cached;
1577 static struct delta_base_cache_lru_list {
1578 struct delta_base_cache_lru_list *prev;
1579 struct delta_base_cache_lru_list *next;
1580 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1582 static struct delta_base_cache_entry {
1583 struct delta_base_cache_lru_list lru;
1584 void *data;
1585 struct packed_git *p;
1586 off_t base_offset;
1587 unsigned long size;
1588 enum object_type type;
1589 } delta_base_cache[MAX_DELTA_CACHE];
1591 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1593 unsigned long hash;
1595 hash = (unsigned long)p + (unsigned long)base_offset;
1596 hash += (hash >> 8) + (hash >> 16);
1597 return hash % MAX_DELTA_CACHE;
1600 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1601 unsigned long *base_size, enum object_type *type, int keep_cache)
1603 void *ret;
1604 unsigned long hash = pack_entry_hash(p, base_offset);
1605 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1607 ret = ent->data;
1608 if (!ret || ent->p != p || ent->base_offset != base_offset)
1609 return unpack_entry(p, base_offset, type, base_size);
1611 if (!keep_cache) {
1612 ent->data = NULL;
1613 ent->lru.next->prev = ent->lru.prev;
1614 ent->lru.prev->next = ent->lru.next;
1615 delta_base_cached -= ent->size;
1616 } else {
1617 ret = xmemdupz(ent->data, ent->size);
1619 *type = ent->type;
1620 *base_size = ent->size;
1621 return ret;
1624 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1626 if (ent->data) {
1627 free(ent->data);
1628 ent->data = NULL;
1629 ent->lru.next->prev = ent->lru.prev;
1630 ent->lru.prev->next = ent->lru.next;
1631 delta_base_cached -= ent->size;
1635 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1636 void *base, unsigned long base_size, enum object_type type)
1638 unsigned long hash = pack_entry_hash(p, base_offset);
1639 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1640 struct delta_base_cache_lru_list *lru;
1642 release_delta_base_cache(ent);
1643 delta_base_cached += base_size;
1645 for (lru = delta_base_cache_lru.next;
1646 delta_base_cached > delta_base_cache_limit
1647 && lru != &delta_base_cache_lru;
1648 lru = lru->next) {
1649 struct delta_base_cache_entry *f = (void *)lru;
1650 if (f->type == OBJ_BLOB)
1651 release_delta_base_cache(f);
1653 for (lru = delta_base_cache_lru.next;
1654 delta_base_cached > delta_base_cache_limit
1655 && lru != &delta_base_cache_lru;
1656 lru = lru->next) {
1657 struct delta_base_cache_entry *f = (void *)lru;
1658 release_delta_base_cache(f);
1661 ent->p = p;
1662 ent->base_offset = base_offset;
1663 ent->type = type;
1664 ent->data = base;
1665 ent->size = base_size;
1666 ent->lru.next = &delta_base_cache_lru;
1667 ent->lru.prev = delta_base_cache_lru.prev;
1668 delta_base_cache_lru.prev->next = &ent->lru;
1669 delta_base_cache_lru.prev = &ent->lru;
1672 static void *unpack_delta_entry(struct packed_git *p,
1673 struct pack_window **w_curs,
1674 off_t curpos,
1675 unsigned long delta_size,
1676 off_t obj_offset,
1677 enum object_type *type,
1678 unsigned long *sizep)
1680 void *delta_data, *result, *base;
1681 unsigned long base_size;
1682 off_t base_offset;
1684 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1685 if (!base_offset) {
1686 error("failed to validate delta base reference "
1687 "at offset %"PRIuMAX" from %s",
1688 (uintmax_t)curpos, p->pack_name);
1689 return NULL;
1691 unuse_pack(w_curs);
1692 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1693 if (!base) {
1695 * We're probably in deep shit, but let's try to fetch
1696 * the required base anyway from another pack or loose.
1697 * This is costly but should happen only in the presence
1698 * of a corrupted pack, and is better than failing outright.
1700 struct revindex_entry *revidx;
1701 const unsigned char *base_sha1;
1702 revidx = find_pack_revindex(p, base_offset);
1703 if (!revidx)
1704 return NULL;
1705 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1706 error("failed to read delta base object %s"
1707 " at offset %"PRIuMAX" from %s",
1708 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1709 p->pack_name);
1710 mark_bad_packed_object(p, base_sha1);
1711 base = read_object(base_sha1, type, &base_size);
1712 if (!base)
1713 return NULL;
1716 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1717 if (!delta_data) {
1718 error("failed to unpack compressed delta "
1719 "at offset %"PRIuMAX" from %s",
1720 (uintmax_t)curpos, p->pack_name);
1721 free(base);
1722 return NULL;
1724 result = patch_delta(base, base_size,
1725 delta_data, delta_size,
1726 sizep);
1727 if (!result)
1728 die("failed to apply delta");
1729 free(delta_data);
1730 add_delta_base_cache(p, base_offset, base, base_size, *type);
1731 return result;
1734 int do_check_packed_object_crc;
1736 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1737 enum object_type *type, unsigned long *sizep)
1739 struct pack_window *w_curs = NULL;
1740 off_t curpos = obj_offset;
1741 void *data;
1743 if (do_check_packed_object_crc && p->index_version > 1) {
1744 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1745 unsigned long len = revidx[1].offset - obj_offset;
1746 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1747 const unsigned char *sha1 =
1748 nth_packed_object_sha1(p, revidx->nr);
1749 error("bad packed object CRC for %s",
1750 sha1_to_hex(sha1));
1751 mark_bad_packed_object(p, sha1);
1752 unuse_pack(&w_curs);
1753 return NULL;
1757 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1758 switch (*type) {
1759 case OBJ_OFS_DELTA:
1760 case OBJ_REF_DELTA:
1761 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1762 obj_offset, type, sizep);
1763 break;
1764 case OBJ_COMMIT:
1765 case OBJ_TREE:
1766 case OBJ_BLOB:
1767 case OBJ_TAG:
1768 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1769 break;
1770 default:
1771 data = NULL;
1772 error("unknown object type %i at offset %"PRIuMAX" in %s",
1773 *type, (uintmax_t)obj_offset, p->pack_name);
1775 unuse_pack(&w_curs);
1776 return data;
1779 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1780 uint32_t n)
1782 const unsigned char *index = p->index_data;
1783 if (!index) {
1784 if (open_pack_index(p))
1785 return NULL;
1786 index = p->index_data;
1788 if (n >= p->num_objects)
1789 return NULL;
1790 index += 4 * 256;
1791 if (p->index_version == 1) {
1792 return index + 24 * n + 4;
1793 } else {
1794 index += 8;
1795 return index + 20 * n;
1799 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1801 const unsigned char *index = p->index_data;
1802 index += 4 * 256;
1803 if (p->index_version == 1) {
1804 return ntohl(*((uint32_t *)(index + 24 * n)));
1805 } else {
1806 uint32_t off;
1807 index += 8 + p->num_objects * (20 + 4);
1808 off = ntohl(*((uint32_t *)(index + 4 * n)));
1809 if (!(off & 0x80000000))
1810 return off;
1811 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1812 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1813 ntohl(*((uint32_t *)(index + 4)));
1817 off_t find_pack_entry_one(const unsigned char *sha1,
1818 struct packed_git *p)
1820 const uint32_t *level1_ofs = p->index_data;
1821 const unsigned char *index = p->index_data;
1822 unsigned hi, lo, stride;
1823 static int use_lookup = -1;
1824 static int debug_lookup = -1;
1826 if (debug_lookup < 0)
1827 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1829 if (!index) {
1830 if (open_pack_index(p))
1831 return 0;
1832 level1_ofs = p->index_data;
1833 index = p->index_data;
1835 if (p->index_version > 1) {
1836 level1_ofs += 2;
1837 index += 8;
1839 index += 4 * 256;
1840 hi = ntohl(level1_ofs[*sha1]);
1841 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1842 if (p->index_version > 1) {
1843 stride = 20;
1844 } else {
1845 stride = 24;
1846 index += 4;
1849 if (debug_lookup)
1850 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1851 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1853 if (use_lookup < 0)
1854 use_lookup = !!getenv("GIT_USE_LOOKUP");
1855 if (use_lookup) {
1856 int pos = sha1_entry_pos(index, stride, 0,
1857 lo, hi, p->num_objects, sha1);
1858 if (pos < 0)
1859 return 0;
1860 return nth_packed_object_offset(p, pos);
1863 do {
1864 unsigned mi = (lo + hi) / 2;
1865 int cmp = hashcmp(index + mi * stride, sha1);
1867 if (debug_lookup)
1868 printf("lo %u hi %u rg %u mi %u\n",
1869 lo, hi, hi - lo, mi);
1870 if (!cmp)
1871 return nth_packed_object_offset(p, mi);
1872 if (cmp > 0)
1873 hi = mi;
1874 else
1875 lo = mi+1;
1876 } while (lo < hi);
1877 return 0;
1880 int matches_pack_name(struct packed_git *p, const char *name)
1882 const char *last_c, *c;
1884 if (!strcmp(p->pack_name, name))
1885 return 1;
1887 for (c = p->pack_name, last_c = c; *c;)
1888 if (*c == '/')
1889 last_c = ++c;
1890 else
1891 ++c;
1892 if (!strcmp(last_c, name))
1893 return 1;
1895 return 0;
1898 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1900 static struct packed_git *last_found = (void *)1;
1901 struct packed_git *p;
1902 off_t offset;
1904 prepare_packed_git();
1905 if (!packed_git)
1906 return 0;
1907 p = (last_found == (void *)1) ? packed_git : last_found;
1909 do {
1910 if (ignore_packed) {
1911 const char **ig;
1912 for (ig = ignore_packed; *ig; ig++)
1913 if (matches_pack_name(p, *ig))
1914 break;
1915 if (*ig)
1916 goto next;
1919 if (p->num_bad_objects) {
1920 unsigned i;
1921 for (i = 0; i < p->num_bad_objects; i++)
1922 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1923 goto next;
1926 offset = find_pack_entry_one(sha1, p);
1927 if (offset) {
1929 * We are about to tell the caller where they can
1930 * locate the requested object. We better make
1931 * sure the packfile is still here and can be
1932 * accessed before supplying that answer, as
1933 * it may have been deleted since the index
1934 * was loaded!
1936 if (p->pack_fd == -1 && open_packed_git(p)) {
1937 error("packfile %s cannot be accessed", p->pack_name);
1938 goto next;
1940 e->offset = offset;
1941 e->p = p;
1942 hashcpy(e->sha1, sha1);
1943 last_found = p;
1944 return 1;
1947 next:
1948 if (p == last_found)
1949 p = packed_git;
1950 else
1951 p = p->next;
1952 if (p == last_found)
1953 p = p->next;
1954 } while (p);
1955 return 0;
1958 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1959 struct packed_git *packs)
1961 struct packed_git *p;
1963 for (p = packs; p; p = p->next) {
1964 if (find_pack_entry_one(sha1, p))
1965 return p;
1967 return NULL;
1971 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1973 int status;
1974 unsigned long mapsize, size;
1975 void *map;
1976 z_stream stream;
1977 char hdr[32];
1979 map = map_sha1_file(sha1, &mapsize);
1980 if (!map)
1981 return error("unable to find %s", sha1_to_hex(sha1));
1982 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1983 status = error("unable to unpack %s header",
1984 sha1_to_hex(sha1));
1985 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1986 status = error("unable to parse %s header", sha1_to_hex(sha1));
1987 else if (sizep)
1988 *sizep = size;
1989 inflateEnd(&stream);
1990 munmap(map, mapsize);
1991 return status;
1994 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1996 struct pack_entry e;
1997 int status;
1999 if (!find_pack_entry(sha1, &e, NULL)) {
2000 /* Most likely it's a loose object. */
2001 status = sha1_loose_object_info(sha1, sizep);
2002 if (status >= 0)
2003 return status;
2005 /* Not a loose object; someone else may have just packed it. */
2006 reprepare_packed_git();
2007 if (!find_pack_entry(sha1, &e, NULL))
2008 return status;
2011 status = packed_object_info(e.p, e.offset, sizep);
2012 if (status < 0) {
2013 mark_bad_packed_object(e.p, sha1);
2014 status = sha1_object_info(sha1, sizep);
2017 return status;
2020 static void *read_packed_sha1(const unsigned char *sha1,
2021 enum object_type *type, unsigned long *size)
2023 struct pack_entry e;
2024 void *data;
2026 if (!find_pack_entry(sha1, &e, NULL))
2027 return NULL;
2028 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2029 if (!data) {
2031 * We're probably in deep shit, but let's try to fetch
2032 * the required object anyway from another pack or loose.
2033 * This should happen only in the presence of a corrupted
2034 * pack, and is better than failing outright.
2036 error("failed to read object %s at offset %"PRIuMAX" from %s",
2037 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2038 mark_bad_packed_object(e.p, sha1);
2039 data = read_object(sha1, type, size);
2041 return data;
2045 * This is meant to hold a *small* number of objects that you would
2046 * want read_sha1_file() to be able to return, but yet you do not want
2047 * to write them into the object store (e.g. a browse-only
2048 * application).
2050 static struct cached_object {
2051 unsigned char sha1[20];
2052 enum object_type type;
2053 void *buf;
2054 unsigned long size;
2055 } *cached_objects;
2056 static int cached_object_nr, cached_object_alloc;
2058 static struct cached_object empty_tree = {
2059 EMPTY_TREE_SHA1_BIN,
2060 OBJ_TREE,
2065 static struct cached_object *find_cached_object(const unsigned char *sha1)
2067 int i;
2068 struct cached_object *co = cached_objects;
2070 for (i = 0; i < cached_object_nr; i++, co++) {
2071 if (!hashcmp(co->sha1, sha1))
2072 return co;
2074 if (!hashcmp(sha1, empty_tree.sha1))
2075 return &empty_tree;
2076 return NULL;
2079 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2080 unsigned char *sha1)
2082 struct cached_object *co;
2084 hash_sha1_file(buf, len, typename(type), sha1);
2085 if (has_sha1_file(sha1) || find_cached_object(sha1))
2086 return 0;
2087 if (cached_object_alloc <= cached_object_nr) {
2088 cached_object_alloc = alloc_nr(cached_object_alloc);
2089 cached_objects = xrealloc(cached_objects,
2090 sizeof(*cached_objects) *
2091 cached_object_alloc);
2093 co = &cached_objects[cached_object_nr++];
2094 co->size = len;
2095 co->type = type;
2096 co->buf = xmalloc(len);
2097 memcpy(co->buf, buf, len);
2098 hashcpy(co->sha1, sha1);
2099 return 0;
2102 void *read_object(const unsigned char *sha1, enum object_type *type,
2103 unsigned long *size)
2105 unsigned long mapsize;
2106 void *map, *buf;
2107 struct cached_object *co;
2109 co = find_cached_object(sha1);
2110 if (co) {
2111 *type = co->type;
2112 *size = co->size;
2113 return xmemdupz(co->buf, co->size);
2116 buf = read_packed_sha1(sha1, type, size);
2117 if (buf)
2118 return buf;
2119 map = map_sha1_file(sha1, &mapsize);
2120 if (map) {
2121 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2122 munmap(map, mapsize);
2123 return buf;
2125 reprepare_packed_git();
2126 return read_packed_sha1(sha1, type, size);
2129 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
2130 unsigned long *size)
2132 void *data = read_object(sha1, type, size);
2133 /* legacy behavior is to die on corrupted objects */
2134 if (!data && (has_loose_object(sha1) || has_packed_and_bad(sha1)))
2135 die("object %s is corrupted", sha1_to_hex(sha1));
2136 return data;
2139 void *read_object_with_reference(const unsigned char *sha1,
2140 const char *required_type_name,
2141 unsigned long *size,
2142 unsigned char *actual_sha1_return)
2144 enum object_type type, required_type;
2145 void *buffer;
2146 unsigned long isize;
2147 unsigned char actual_sha1[20];
2149 required_type = type_from_string(required_type_name);
2150 hashcpy(actual_sha1, sha1);
2151 while (1) {
2152 int ref_length = -1;
2153 const char *ref_type = NULL;
2155 buffer = read_sha1_file(actual_sha1, &type, &isize);
2156 if (!buffer)
2157 return NULL;
2158 if (type == required_type) {
2159 *size = isize;
2160 if (actual_sha1_return)
2161 hashcpy(actual_sha1_return, actual_sha1);
2162 return buffer;
2164 /* Handle references */
2165 else if (type == OBJ_COMMIT)
2166 ref_type = "tree ";
2167 else if (type == OBJ_TAG)
2168 ref_type = "object ";
2169 else {
2170 free(buffer);
2171 return NULL;
2173 ref_length = strlen(ref_type);
2175 if (ref_length + 40 > isize ||
2176 memcmp(buffer, ref_type, ref_length) ||
2177 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2178 free(buffer);
2179 return NULL;
2181 free(buffer);
2182 /* Now we have the ID of the referred-to object in
2183 * actual_sha1. Check again. */
2187 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2188 const char *type, unsigned char *sha1,
2189 char *hdr, int *hdrlen)
2191 git_SHA_CTX c;
2193 /* Generate the header */
2194 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2196 /* Sha1.. */
2197 git_SHA1_Init(&c);
2198 git_SHA1_Update(&c, hdr, *hdrlen);
2199 git_SHA1_Update(&c, buf, len);
2200 git_SHA1_Final(sha1, &c);
2204 * Move the just written object into its final resting place
2206 int move_temp_to_file(const char *tmpfile, const char *filename)
2208 int ret = 0;
2209 if (link(tmpfile, filename))
2210 ret = errno;
2213 * Coda hack - coda doesn't like cross-directory links,
2214 * so we fall back to a rename, which will mean that it
2215 * won't be able to check collisions, but that's not a
2216 * big deal.
2218 * The same holds for FAT formatted media.
2220 * When this succeeds, we just return 0. We have nothing
2221 * left to unlink.
2223 if (ret && ret != EEXIST) {
2224 if (!rename(tmpfile, filename))
2225 return 0;
2226 ret = errno;
2228 unlink(tmpfile);
2229 if (ret) {
2230 if (ret != EEXIST) {
2231 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2233 /* FIXME!!! Collision check here ? */
2236 return 0;
2239 static int write_buffer(int fd, const void *buf, size_t len)
2241 if (write_in_full(fd, buf, len) < 0)
2242 return error("file write error (%s)", strerror(errno));
2243 return 0;
2246 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2247 unsigned char *sha1)
2249 char hdr[32];
2250 int hdrlen;
2251 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2252 return 0;
2255 /* Finalize a file on disk, and close it. */
2256 static void close_sha1_file(int fd)
2258 if (fsync_object_files)
2259 fsync_or_die(fd, "sha1 file");
2260 fchmod(fd, 0444);
2261 if (close(fd) != 0)
2262 die("unable to write sha1 file");
2265 /* Size of directory component, including the ending '/' */
2266 static inline int directory_size(const char *filename)
2268 const char *s = strrchr(filename, '/');
2269 if (!s)
2270 return 0;
2271 return s - filename + 1;
2275 * This creates a temporary file in the same directory as the final
2276 * 'filename'
2278 * We want to avoid cross-directory filename renames, because those
2279 * can have problems on various filesystems (FAT, NFS, Coda).
2281 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2283 int fd, dirlen = directory_size(filename);
2285 if (dirlen + 20 > bufsiz) {
2286 errno = ENAMETOOLONG;
2287 return -1;
2289 memcpy(buffer, filename, dirlen);
2290 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2291 fd = mkstemp(buffer);
2292 if (fd < 0 && dirlen && errno == ENOENT) {
2293 /* Make sure the directory exists */
2294 memcpy(buffer, filename, dirlen);
2295 buffer[dirlen-1] = 0;
2296 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2297 return -1;
2299 /* Try again */
2300 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2301 fd = mkstemp(buffer);
2303 return fd;
2306 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2307 void *buf, unsigned long len, time_t mtime)
2309 int fd, size, ret;
2310 unsigned char *compressed;
2311 z_stream stream;
2312 char *filename;
2313 static char tmpfile[PATH_MAX];
2315 filename = sha1_file_name(sha1);
2316 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2317 if (fd < 0) {
2318 if (errno == EACCES)
2319 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2320 else
2321 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2324 /* Set it up */
2325 memset(&stream, 0, sizeof(stream));
2326 deflateInit(&stream, zlib_compression_level);
2327 size = 8 + deflateBound(&stream, len+hdrlen);
2328 compressed = xmalloc(size);
2330 /* Compress it */
2331 stream.next_out = compressed;
2332 stream.avail_out = size;
2334 /* First header.. */
2335 stream.next_in = (unsigned char *)hdr;
2336 stream.avail_in = hdrlen;
2337 while (deflate(&stream, 0) == Z_OK)
2338 /* nothing */;
2340 /* Then the data itself.. */
2341 stream.next_in = buf;
2342 stream.avail_in = len;
2343 ret = deflate(&stream, Z_FINISH);
2344 if (ret != Z_STREAM_END)
2345 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2347 ret = deflateEnd(&stream);
2348 if (ret != Z_OK)
2349 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2351 size = stream.total_out;
2353 if (write_buffer(fd, compressed, size) < 0)
2354 die("unable to write sha1 file");
2355 close_sha1_file(fd);
2356 free(compressed);
2358 if (mtime) {
2359 struct utimbuf utb;
2360 utb.actime = mtime;
2361 utb.modtime = mtime;
2362 if (utime(tmpfile, &utb) < 0)
2363 warning("failed utime() on %s: %s",
2364 tmpfile, strerror(errno));
2367 return move_temp_to_file(tmpfile, filename);
2370 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2372 unsigned char sha1[20];
2373 char hdr[32];
2374 int hdrlen;
2376 /* Normally if we have it in the pack then we do not bother writing
2377 * it out into .git/objects/??/?{38} file.
2379 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2380 if (returnsha1)
2381 hashcpy(returnsha1, sha1);
2382 if (has_sha1_file(sha1))
2383 return 0;
2384 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2387 int force_object_loose(const unsigned char *sha1, time_t mtime)
2389 void *buf;
2390 unsigned long len;
2391 enum object_type type;
2392 char hdr[32];
2393 int hdrlen;
2394 int ret;
2396 if (has_loose_object(sha1))
2397 return 0;
2398 buf = read_packed_sha1(sha1, &type, &len);
2399 if (!buf)
2400 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2401 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2402 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2403 free(buf);
2405 return ret;
2408 int has_pack_index(const unsigned char *sha1)
2410 struct stat st;
2411 if (stat(sha1_pack_index_name(sha1), &st))
2412 return 0;
2413 return 1;
2416 int has_pack_file(const unsigned char *sha1)
2418 struct stat st;
2419 if (stat(sha1_pack_name(sha1), &st))
2420 return 0;
2421 return 1;
2424 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2426 struct pack_entry e;
2427 return find_pack_entry(sha1, &e, ignore_packed);
2430 int has_sha1_file(const unsigned char *sha1)
2432 struct pack_entry e;
2434 if (find_pack_entry(sha1, &e, NULL))
2435 return 1;
2436 return has_loose_object(sha1);
2439 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2440 int write_object, enum object_type type, const char *path)
2442 int ret, re_allocated = 0;
2444 if (!type)
2445 type = OBJ_BLOB;
2448 * Convert blobs to git internal format
2450 if ((type == OBJ_BLOB) && path) {
2451 struct strbuf nbuf = STRBUF_INIT;
2452 if (convert_to_git(path, buf, size, &nbuf,
2453 write_object ? safe_crlf : 0)) {
2454 buf = strbuf_detach(&nbuf, &size);
2455 re_allocated = 1;
2459 if (write_object)
2460 ret = write_sha1_file(buf, size, typename(type), sha1);
2461 else
2462 ret = hash_sha1_file(buf, size, typename(type), sha1);
2463 if (re_allocated)
2464 free(buf);
2465 return ret;
2468 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2469 enum object_type type, const char *path)
2471 int ret;
2472 size_t size = xsize_t(st->st_size);
2474 if (!S_ISREG(st->st_mode)) {
2475 struct strbuf sbuf = STRBUF_INIT;
2476 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2477 ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2478 type, path);
2479 else
2480 ret = -1;
2481 strbuf_release(&sbuf);
2482 } else if (size) {
2483 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2484 ret = index_mem(sha1, buf, size, write_object, type, path);
2485 munmap(buf, size);
2486 } else
2487 ret = index_mem(sha1, NULL, size, write_object, type, path);
2488 close(fd);
2489 return ret;
2492 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2494 int fd;
2495 char *target;
2496 size_t len;
2498 switch (st->st_mode & S_IFMT) {
2499 case S_IFREG:
2500 fd = open(path, O_RDONLY);
2501 if (fd < 0)
2502 return error("open(\"%s\"): %s", path,
2503 strerror(errno));
2504 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2505 return error("%s: failed to insert into database",
2506 path);
2507 break;
2508 case S_IFLNK:
2509 len = xsize_t(st->st_size);
2510 target = xmalloc(len + 1);
2511 if (readlink(path, target, len + 1) != st->st_size) {
2512 char *errstr = strerror(errno);
2513 free(target);
2514 return error("readlink(\"%s\"): %s", path,
2515 errstr);
2517 if (!write_object)
2518 hash_sha1_file(target, len, blob_type, sha1);
2519 else if (write_sha1_file(target, len, blob_type, sha1))
2520 return error("%s: failed to insert into database",
2521 path);
2522 free(target);
2523 break;
2524 case S_IFDIR:
2525 return resolve_gitlink_ref(path, "HEAD", sha1);
2526 default:
2527 return error("%s: unsupported file type", path);
2529 return 0;
2532 int read_pack_header(int fd, struct pack_header *header)
2534 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2535 /* "eof before pack header was fully read" */
2536 return PH_ERROR_EOF;
2538 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2539 /* "protocol error (pack signature mismatch detected)" */
2540 return PH_ERROR_PACK_SIGNATURE;
2541 if (!pack_version_ok(header->hdr_version))
2542 /* "protocol error (pack version unsupported)" */
2543 return PH_ERROR_PROTOCOL;
2544 return 0;