Merge branch 'master' of git://repo.or.cz/alt-git
[git/mingw/4msysgit/wingit-dll.git] / sha1_file.c
blob48bbf6676360439ba117e5b948438a96aa37e7d5
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 static unsigned int sha1_file_open_flag = O_NOATIME;
40 const signed char hexval_table[256] = {
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
45 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
47 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
48 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
49 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
53 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
69 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
71 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
72 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
75 int get_sha1_hex(const char *hex, unsigned char *sha1)
77 int i;
78 for (i = 0; i < 20; i++) {
79 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
80 if (val & ~0xff)
81 return -1;
82 *sha1++ = val;
83 hex += 2;
85 return 0;
88 static inline int offset_1st_component(const char *path)
90 if (has_dos_drive_prefix(path))
91 return 2 + (path[2] == '/');
92 return *path == '/';
95 int safe_create_leading_directories(char *path)
97 char *pos = path + offset_1st_component(path);
98 struct stat st;
100 while (pos) {
101 pos = strchr(pos, '/');
102 if (!pos)
103 break;
104 *pos = 0;
105 if (!stat(path, &st)) {
106 /* path exists */
107 if (!S_ISDIR(st.st_mode)) {
108 *pos = '/';
109 return -3;
112 else if (mkdir(path, 0777)) {
113 *pos = '/';
114 return -1;
116 else if (adjust_shared_perm(path)) {
117 *pos = '/';
118 return -2;
120 *pos++ = '/';
122 return 0;
125 char * sha1_to_hex(const unsigned char *sha1)
127 static int bufno;
128 static char hexbuffer[4][50];
129 static const char hex[] = "0123456789abcdef";
130 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
131 int i;
133 for (i = 0; i < 20; i++) {
134 unsigned int val = *sha1++;
135 *buf++ = hex[val >> 4];
136 *buf++ = hex[val & 0xf];
138 *buf = '\0';
140 return buffer;
143 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
145 int i;
146 for (i = 0; i < 20; i++) {
147 static char hex[] = "0123456789abcdef";
148 unsigned int val = sha1[i];
149 char *pos = pathbuf + i*2 + (i > 0);
150 *pos++ = hex[val >> 4];
151 *pos = hex[val & 0xf];
156 * NOTE! This returns a statically allocated buffer, so you have to be
157 * careful about using it. Do an "xstrdup()" if you need to save the
158 * filename.
160 * Also note that this returns the location for creating. Reading
161 * SHA1 file can happen from any alternate directory listed in the
162 * DB_ENVIRONMENT environment variable if it is not found in
163 * the primary object database.
165 char *sha1_file_name(const unsigned char *sha1)
167 static char *name, *base;
169 if (!base) {
170 const char *sha1_file_directory = get_object_directory();
171 int len = strlen(sha1_file_directory);
172 base = xmalloc(len + 60);
173 memcpy(base, sha1_file_directory, len);
174 memset(base+len, 0, 60);
175 base[len] = '/';
176 base[len+3] = '/';
177 name = base + len + 1;
179 fill_sha1_path(name, sha1);
180 return base;
183 char *sha1_pack_name(const unsigned char *sha1)
185 static const char hex[] = "0123456789abcdef";
186 static char *name, *base, *buf;
187 int i;
189 if (!base) {
190 const char *sha1_file_directory = get_object_directory();
191 int len = strlen(sha1_file_directory);
192 base = xmalloc(len + 60);
193 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
194 name = base + len + 11;
197 buf = name;
199 for (i = 0; i < 20; i++) {
200 unsigned int val = *sha1++;
201 *buf++ = hex[val >> 4];
202 *buf++ = hex[val & 0xf];
205 return base;
208 char *sha1_pack_index_name(const unsigned char *sha1)
210 static const char hex[] = "0123456789abcdef";
211 static char *name, *base, *buf;
212 int i;
214 if (!base) {
215 const char *sha1_file_directory = get_object_directory();
216 int len = strlen(sha1_file_directory);
217 base = xmalloc(len + 60);
218 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
219 name = base + len + 11;
222 buf = name;
224 for (i = 0; i < 20; i++) {
225 unsigned int val = *sha1++;
226 *buf++ = hex[val >> 4];
227 *buf++ = hex[val & 0xf];
230 return base;
233 struct alternate_object_database *alt_odb_list;
234 static struct alternate_object_database **alt_odb_tail;
236 static void read_info_alternates(const char * alternates, int depth);
239 * Prepare alternate object database registry.
241 * The variable alt_odb_list points at the list of struct
242 * alternate_object_database. The elements on this list come from
243 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
244 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
245 * whose contents is similar to that environment variable but can be
246 * LF separated. Its base points at a statically allocated buffer that
247 * contains "/the/directory/corresponding/to/.git/objects/...", while
248 * its name points just after the slash at the end of ".git/objects/"
249 * in the example above, and has enough space to hold 40-byte hex
250 * SHA1, an extra slash for the first level indirection, and the
251 * terminating NUL.
253 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
255 struct stat st;
256 const char *objdir = get_object_directory();
257 struct alternate_object_database *ent;
258 struct alternate_object_database *alt;
259 /* 43 = 40-byte + 2 '/' + terminating NUL */
260 int pfxlen = len;
261 int entlen = pfxlen + 43;
262 int base_len = -1;
264 if (!is_absolute_path(entry) && relative_base) {
265 /* Relative alt-odb */
266 if (base_len < 0)
267 base_len = strlen(relative_base) + 1;
268 entlen += base_len;
269 pfxlen += base_len;
271 ent = xmalloc(sizeof(*ent) + entlen);
273 if (!is_absolute_path(entry) && relative_base) {
274 memcpy(ent->base, relative_base, base_len - 1);
275 ent->base[base_len - 1] = '/';
276 memcpy(ent->base + base_len, entry, len);
278 else
279 memcpy(ent->base, entry, pfxlen);
281 ent->name = ent->base + pfxlen + 1;
282 ent->base[pfxlen + 3] = '/';
283 ent->base[pfxlen] = ent->base[entlen-1] = 0;
285 /* Detect cases where alternate disappeared */
286 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
287 error("object directory %s does not exist; "
288 "check .git/objects/info/alternates.",
289 ent->base);
290 free(ent);
291 return -1;
294 /* Prevent the common mistake of listing the same
295 * thing twice, or object directory itself.
297 for (alt = alt_odb_list; alt; alt = alt->next) {
298 if (!memcmp(ent->base, alt->base, pfxlen)) {
299 free(ent);
300 return -1;
303 if (!memcmp(ent->base, objdir, pfxlen)) {
304 free(ent);
305 return -1;
308 /* add the alternate entry */
309 *alt_odb_tail = ent;
310 alt_odb_tail = &(ent->next);
311 ent->next = NULL;
313 /* recursively add alternates */
314 read_info_alternates(ent->base, depth + 1);
316 ent->base[pfxlen] = '/';
318 return 0;
321 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
322 const char *relative_base, int depth)
324 const char *cp, *last;
326 if (depth > 5) {
327 error("%s: ignoring alternate object stores, nesting too deep.",
328 relative_base);
329 return;
332 last = alt;
333 while (last < ep) {
334 cp = last;
335 if (cp < ep && *cp == '#') {
336 while (cp < ep && *cp != sep)
337 cp++;
338 last = cp + 1;
339 continue;
341 while (cp < ep && *cp != sep)
342 cp++;
343 if (last != cp) {
344 if (!is_absolute_path(last) && depth) {
345 error("%s: ignoring relative alternate object store %s",
346 relative_base, last);
347 } else {
348 link_alt_odb_entry(last, cp - last,
349 relative_base, depth);
352 while (cp < ep && *cp == sep)
353 cp++;
354 last = cp;
358 static void read_info_alternates(const char * relative_base, int depth)
360 char *map;
361 size_t mapsz;
362 struct stat st;
363 const char alt_file_name[] = "info/alternates";
364 /* Given that relative_base is no longer than PATH_MAX,
365 ensure that "path" has enough space to append "/", the
366 file name, "info/alternates", and a trailing NUL. */
367 char path[PATH_MAX + 1 + sizeof alt_file_name];
368 int fd;
370 sprintf(path, "%s/%s", relative_base, alt_file_name);
371 fd = open(path, O_RDONLY);
372 if (fd < 0)
373 return;
374 if (fstat(fd, &st) || (st.st_size == 0)) {
375 close(fd);
376 return;
378 mapsz = xsize_t(st.st_size);
379 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
380 close(fd);
382 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
384 munmap(map, mapsz);
387 void prepare_alt_odb(void)
389 const char *alt;
391 if (alt_odb_tail)
392 return;
394 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
395 if (!alt) alt = "";
397 alt_odb_tail = &alt_odb_list;
398 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
400 read_info_alternates(get_object_directory(), 0);
403 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
405 char *name = sha1_file_name(sha1);
406 struct alternate_object_database *alt;
408 if (!stat(name, st))
409 return name;
410 prepare_alt_odb();
411 for (alt = alt_odb_list; alt; alt = alt->next) {
412 name = alt->name;
413 fill_sha1_path(name, sha1);
414 if (!stat(alt->base, st))
415 return alt->base;
417 return NULL;
420 static unsigned int pack_used_ctr;
421 static unsigned int pack_mmap_calls;
422 static unsigned int peak_pack_open_windows;
423 static unsigned int pack_open_windows;
424 static size_t peak_pack_mapped;
425 static size_t pack_mapped;
426 struct packed_git *packed_git;
428 void pack_report(void)
430 fprintf(stderr,
431 "pack_report: getpagesize() = %10" SZ_FMT "\n"
432 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
433 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
434 sz_fmt(getpagesize()),
435 sz_fmt(packed_git_window_size),
436 sz_fmt(packed_git_limit));
437 fprintf(stderr,
438 "pack_report: pack_used_ctr = %10u\n"
439 "pack_report: pack_mmap_calls = %10u\n"
440 "pack_report: pack_open_windows = %10u / %10u\n"
441 "pack_report: pack_mapped = "
442 "%10" SZ_FMT " / %10" SZ_FMT "\n",
443 pack_used_ctr,
444 pack_mmap_calls,
445 pack_open_windows, peak_pack_open_windows,
446 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
449 static int check_packed_git_idx(const char *path, struct packed_git *p)
451 void *idx_map;
452 struct pack_idx_header *hdr;
453 size_t idx_size;
454 uint32_t version, nr, i, *index;
455 int fd = open(path, O_RDONLY);
456 struct stat st;
458 if (fd < 0)
459 return -1;
460 if (fstat(fd, &st)) {
461 close(fd);
462 return -1;
464 idx_size = xsize_t(st.st_size);
465 if (idx_size < 4 * 256 + 20 + 20) {
466 close(fd);
467 return error("index file %s is too small", path);
469 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
470 close(fd);
472 hdr = idx_map;
473 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
474 version = ntohl(hdr->idx_version);
475 if (version < 2 || version > 2) {
476 munmap(idx_map, idx_size);
477 return error("index file %s is version %d"
478 " and is not supported by this binary"
479 " (try upgrading GIT to a newer version)",
480 path, version);
482 } else
483 version = 1;
485 nr = 0;
486 index = idx_map;
487 if (version > 1)
488 index += 2; /* skip index header */
489 for (i = 0; i < 256; i++) {
490 uint32_t n = ntohl(index[i]);
491 if (n < nr) {
492 munmap(idx_map, idx_size);
493 return error("non-monotonic index %s", path);
495 nr = n;
498 if (version == 1) {
500 * Total size:
501 * - 256 index entries 4 bytes each
502 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
503 * - 20-byte SHA1 of the packfile
504 * - 20-byte SHA1 file checksum
506 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
507 munmap(idx_map, idx_size);
508 return error("wrong index v1 file size in %s", path);
510 } else if (version == 2) {
512 * Minimum size:
513 * - 8 bytes of header
514 * - 256 index entries 4 bytes each
515 * - 20-byte sha1 entry * nr
516 * - 4-byte crc entry * nr
517 * - 4-byte offset entry * nr
518 * - 20-byte SHA1 of the packfile
519 * - 20-byte SHA1 file checksum
520 * And after the 4-byte offset table might be a
521 * variable sized table containing 8-byte entries
522 * for offsets larger than 2^31.
524 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
525 unsigned long max_size = min_size;
526 if (nr)
527 max_size += (nr - 1)*8;
528 if (idx_size < min_size || idx_size > max_size) {
529 munmap(idx_map, idx_size);
530 return error("wrong index v2 file size in %s", path);
532 if (idx_size != min_size &&
534 * make sure we can deal with large pack offsets.
535 * 31-bit signed offset won't be enough, neither
536 * 32-bit unsigned one will be.
538 (sizeof(off_t) <= 4)) {
539 munmap(idx_map, idx_size);
540 return error("pack too large for current definition of off_t in %s", path);
544 p->index_version = version;
545 p->index_data = idx_map;
546 p->index_size = idx_size;
547 p->num_objects = nr;
548 return 0;
551 int open_pack_index(struct packed_git *p)
553 char *idx_name;
554 int ret;
556 if (p->index_data)
557 return 0;
559 idx_name = xstrdup(p->pack_name);
560 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
561 ret = check_packed_git_idx(idx_name, p);
562 free(idx_name);
563 return ret;
566 static void scan_windows(struct packed_git *p,
567 struct packed_git **lru_p,
568 struct pack_window **lru_w,
569 struct pack_window **lru_l)
571 struct pack_window *w, *w_l;
573 for (w_l = NULL, w = p->windows; w; w = w->next) {
574 if (!w->inuse_cnt) {
575 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
576 *lru_p = p;
577 *lru_w = w;
578 *lru_l = w_l;
581 w_l = w;
585 static int unuse_one_window(struct packed_git *current, int keep_fd)
587 struct packed_git *p, *lru_p = NULL;
588 struct pack_window *lru_w = NULL, *lru_l = NULL;
590 if (current)
591 scan_windows(current, &lru_p, &lru_w, &lru_l);
592 for (p = packed_git; p; p = p->next)
593 scan_windows(p, &lru_p, &lru_w, &lru_l);
594 if (lru_p) {
595 munmap(lru_w->base, lru_w->len);
596 pack_mapped -= lru_w->len;
597 if (lru_l)
598 lru_l->next = lru_w->next;
599 else {
600 lru_p->windows = lru_w->next;
601 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
602 close(lru_p->pack_fd);
603 lru_p->pack_fd = -1;
606 free(lru_w);
607 pack_open_windows--;
608 return 1;
610 return 0;
613 void release_pack_memory(size_t need, int fd)
615 size_t cur = pack_mapped;
616 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
617 ; /* nothing */
620 void close_pack_windows(struct packed_git *p)
622 while (p->windows) {
623 struct pack_window *w = p->windows;
625 if (w->inuse_cnt)
626 die("pack '%s' still has open windows to it",
627 p->pack_name);
628 munmap(w->base, w->len);
629 pack_mapped -= w->len;
630 pack_open_windows--;
631 p->windows = w->next;
632 free(w);
636 void unuse_pack(struct pack_window **w_cursor)
638 struct pack_window *w = *w_cursor;
639 if (w) {
640 w->inuse_cnt--;
641 *w_cursor = NULL;
646 * Do not call this directly as this leaks p->pack_fd on error return;
647 * call open_packed_git() instead.
649 static int open_packed_git_1(struct packed_git *p)
651 struct stat st;
652 struct pack_header hdr;
653 unsigned char sha1[20];
654 unsigned char *idx_sha1;
655 long fd_flag;
657 if (!p->index_data && open_pack_index(p))
658 return error("packfile %s index unavailable", p->pack_name);
660 p->pack_fd = open(p->pack_name, O_RDONLY);
661 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
662 return -1;
664 /* If we created the struct before we had the pack we lack size. */
665 if (!p->pack_size) {
666 if (!S_ISREG(st.st_mode))
667 return error("packfile %s not a regular file", p->pack_name);
668 p->pack_size = st.st_size;
669 } else if (p->pack_size != st.st_size)
670 return error("packfile %s size changed", p->pack_name);
672 /* We leave these file descriptors open with sliding mmap;
673 * there is no point keeping them open across exec(), though.
675 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
676 if (fd_flag < 0)
677 return error("cannot determine file descriptor flags");
678 fd_flag |= FD_CLOEXEC;
679 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
680 return error("cannot set FD_CLOEXEC");
682 /* Verify we recognize this pack file format. */
683 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
684 return error("file %s is far too short to be a packfile", p->pack_name);
685 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
686 return error("file %s is not a GIT packfile", p->pack_name);
687 if (!pack_version_ok(hdr.hdr_version))
688 return error("packfile %s is version %u and not supported"
689 " (try upgrading GIT to a newer version)",
690 p->pack_name, ntohl(hdr.hdr_version));
692 /* Verify the pack matches its index. */
693 if (p->num_objects != ntohl(hdr.hdr_entries))
694 return error("packfile %s claims to have %u objects"
695 " while index indicates %u objects",
696 p->pack_name, ntohl(hdr.hdr_entries),
697 p->num_objects);
698 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
699 return error("end of packfile %s is unavailable", p->pack_name);
700 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
701 return error("packfile %s signature is unavailable", p->pack_name);
702 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
703 if (hashcmp(sha1, idx_sha1))
704 return error("packfile %s does not match index", p->pack_name);
705 return 0;
708 static int open_packed_git(struct packed_git *p)
710 if (!open_packed_git_1(p))
711 return 0;
712 if (p->pack_fd != -1) {
713 close(p->pack_fd);
714 p->pack_fd = -1;
716 return -1;
719 static int in_window(struct pack_window *win, off_t offset)
721 /* We must promise at least 20 bytes (one hash) after the
722 * offset is available from this window, otherwise the offset
723 * is not actually in this window and a different window (which
724 * has that one hash excess) must be used. This is to support
725 * the object header and delta base parsing routines below.
727 off_t win_off = win->offset;
728 return win_off <= offset
729 && (offset + 20) <= (win_off + win->len);
732 unsigned char* use_pack(struct packed_git *p,
733 struct pack_window **w_cursor,
734 off_t offset,
735 unsigned int *left)
737 struct pack_window *win = *w_cursor;
739 if (p->pack_fd == -1 && open_packed_git(p))
740 die("packfile %s cannot be accessed", p->pack_name);
742 /* Since packfiles end in a hash of their content and its
743 * pointless to ask for an offset into the middle of that
744 * hash, and the in_window function above wouldn't match
745 * don't allow an offset too close to the end of the file.
747 if (offset > (p->pack_size - 20))
748 die("offset beyond end of packfile (truncated pack?)");
750 if (!win || !in_window(win, offset)) {
751 if (win)
752 win->inuse_cnt--;
753 for (win = p->windows; win; win = win->next) {
754 if (in_window(win, offset))
755 break;
757 if (!win) {
758 size_t window_align = packed_git_window_size / 2;
759 off_t len;
760 win = xcalloc(1, sizeof(*win));
761 win->offset = (offset / window_align) * window_align;
762 len = p->pack_size - win->offset;
763 if (len > packed_git_window_size)
764 len = packed_git_window_size;
765 win->len = (size_t)len;
766 pack_mapped += win->len;
767 while (packed_git_limit < pack_mapped
768 && unuse_one_window(p, p->pack_fd))
769 ; /* nothing */
770 win->base = xmmap(NULL, win->len,
771 PROT_READ, MAP_PRIVATE,
772 p->pack_fd, win->offset);
773 if (win->base == MAP_FAILED)
774 die("packfile %s cannot be mapped: %s",
775 p->pack_name,
776 strerror(errno));
777 pack_mmap_calls++;
778 pack_open_windows++;
779 if (pack_mapped > peak_pack_mapped)
780 peak_pack_mapped = pack_mapped;
781 if (pack_open_windows > peak_pack_open_windows)
782 peak_pack_open_windows = pack_open_windows;
783 win->next = p->windows;
784 p->windows = win;
787 if (win != *w_cursor) {
788 win->last_used = pack_used_ctr++;
789 win->inuse_cnt++;
790 *w_cursor = win;
792 offset -= win->offset;
793 if (left)
794 *left = win->len - xsize_t(offset);
795 return win->base + offset;
798 struct packed_git *add_packed_git(const char *path, int path_len, int local)
800 struct stat st;
801 struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
804 * Make sure a corresponding .pack file exists and that
805 * the index looks sane.
807 path_len -= strlen(".idx");
808 if (path_len < 1)
809 return NULL;
810 memcpy(p->pack_name, path, path_len);
811 strcpy(p->pack_name + path_len, ".pack");
812 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
813 free(p);
814 return NULL;
817 /* ok, it looks sane as far as we can check without
818 * actually mapping the pack file.
820 p->index_version = 0;
821 p->index_data = NULL;
822 p->index_size = 0;
823 p->num_objects = 0;
824 p->pack_size = st.st_size;
825 p->next = NULL;
826 p->windows = NULL;
827 p->pack_fd = -1;
828 p->pack_local = local;
829 p->mtime = st.st_mtime;
830 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
831 hashclr(p->sha1);
832 return p;
835 struct packed_git *parse_pack_index(unsigned char *sha1)
837 char *path = sha1_pack_index_name(sha1);
838 return parse_pack_index_file(sha1, path);
841 struct packed_git *parse_pack_index_file(const unsigned char *sha1,
842 const char *idx_path)
844 const char *path = sha1_pack_name(sha1);
845 struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
847 if (check_packed_git_idx(idx_path, p)) {
848 free(p);
849 return NULL;
852 strcpy(p->pack_name, path);
853 p->pack_size = 0;
854 p->next = NULL;
855 p->windows = NULL;
856 p->pack_fd = -1;
857 hashcpy(p->sha1, sha1);
858 return p;
861 void install_packed_git(struct packed_git *pack)
863 pack->next = packed_git;
864 packed_git = pack;
867 static void prepare_packed_git_one(char *objdir, int local)
869 /* Ensure that this buffer is large enough so that we can
870 append "/pack/" without clobbering the stack even if
871 strlen(objdir) were PATH_MAX. */
872 char path[PATH_MAX + 1 + 4 + 1 + 1];
873 int len;
874 DIR *dir;
875 struct dirent *de;
877 sprintf(path, "%s/pack", objdir);
878 len = strlen(path);
879 dir = opendir(path);
880 if (!dir) {
881 if (errno != ENOENT)
882 error("unable to open object pack directory: %s: %s",
883 path, strerror(errno));
884 return;
886 path[len++] = '/';
887 while ((de = readdir(dir)) != NULL) {
888 int namelen = strlen(de->d_name);
889 struct packed_git *p;
891 if (!has_extension(de->d_name, ".idx"))
892 continue;
894 if (len + namelen + 1 > sizeof(path))
895 continue;
897 /* Don't reopen a pack we already have. */
898 strcpy(path + len, de->d_name);
899 for (p = packed_git; p; p = p->next) {
900 if (!memcmp(path, p->pack_name, len + namelen - 4))
901 break;
903 if (p)
904 continue;
905 /* See if it really is a valid .idx file with corresponding
906 * .pack file that we can map.
908 p = add_packed_git(path, len + namelen, local);
909 if (!p)
910 continue;
911 install_packed_git(p);
913 closedir(dir);
916 static int sort_pack(const void *a_, const void *b_)
918 struct packed_git *a = *((struct packed_git **)a_);
919 struct packed_git *b = *((struct packed_git **)b_);
920 int st;
923 * Local packs tend to contain objects specific to our
924 * variant of the project than remote ones. In addition,
925 * remote ones could be on a network mounted filesystem.
926 * Favor local ones for these reasons.
928 st = a->pack_local - b->pack_local;
929 if (st)
930 return -st;
933 * Younger packs tend to contain more recent objects,
934 * and more recent objects tend to get accessed more
935 * often.
937 if (a->mtime < b->mtime)
938 return 1;
939 else if (a->mtime == b->mtime)
940 return 0;
941 return -1;
944 static void rearrange_packed_git(void)
946 struct packed_git **ary, *p;
947 int i, n;
949 for (n = 0, p = packed_git; p; p = p->next)
950 n++;
951 if (n < 2)
952 return;
954 /* prepare an array of packed_git for easier sorting */
955 ary = xcalloc(n, sizeof(struct packed_git *));
956 for (n = 0, p = packed_git; p; p = p->next)
957 ary[n++] = p;
959 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
961 /* link them back again */
962 for (i = 0; i < n - 1; i++)
963 ary[i]->next = ary[i + 1];
964 ary[n - 1]->next = NULL;
965 packed_git = ary[0];
967 free(ary);
970 static int prepare_packed_git_run_once = 0;
971 void prepare_packed_git(void)
973 struct alternate_object_database *alt;
975 if (prepare_packed_git_run_once)
976 return;
977 prepare_packed_git_one(get_object_directory(), 1);
978 prepare_alt_odb();
979 for (alt = alt_odb_list; alt; alt = alt->next) {
980 alt->name[-1] = 0;
981 prepare_packed_git_one(alt->base, 0);
982 alt->name[-1] = '/';
984 rearrange_packed_git();
985 prepare_packed_git_run_once = 1;
988 void reprepare_packed_git(void)
990 prepare_packed_git_run_once = 0;
991 prepare_packed_git();
994 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
996 unsigned char real_sha1[20];
997 hash_sha1_file(map, size, type, real_sha1);
998 return hashcmp(sha1, real_sha1) ? -1 : 0;
1001 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1003 struct stat st;
1004 void *map;
1005 int fd;
1006 char *filename = find_sha1_file(sha1, &st);
1008 if (!filename) {
1009 return NULL;
1012 fd = open(filename, O_RDONLY | sha1_file_open_flag);
1013 if (fd < 0) {
1014 /* See if it works without O_NOATIME */
1015 switch (sha1_file_open_flag) {
1016 default:
1017 fd = open(filename, O_RDONLY);
1018 if (fd >= 0)
1019 break;
1020 /* Fallthrough */
1021 case 0:
1022 return NULL;
1025 /* If it failed once, it will probably fail again.
1026 * Stop using O_NOATIME
1028 sha1_file_open_flag = 0;
1030 *size = xsize_t(st.st_size);
1031 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1032 close(fd);
1033 return map;
1036 static int legacy_loose_object(unsigned char *map)
1038 unsigned int word;
1041 * Is it a zlib-compressed buffer? If so, the first byte
1042 * must be 0x78 (15-bit window size, deflated), and the
1043 * first 16-bit word is evenly divisible by 31
1045 word = (map[0] << 8) + map[1];
1046 if (map[0] == 0x78 && !(word % 31))
1047 return 1;
1048 else
1049 return 0;
1052 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1054 unsigned shift;
1055 unsigned char c;
1056 unsigned long size;
1057 unsigned long used = 0;
1059 c = buf[used++];
1060 *type = (c >> 4) & 7;
1061 size = c & 15;
1062 shift = 4;
1063 while (c & 0x80) {
1064 if (len <= used)
1065 return 0;
1066 if (sizeof(long) * 8 <= shift)
1067 return 0;
1068 c = buf[used++];
1069 size += (c & 0x7f) << shift;
1070 shift += 7;
1072 *sizep = size;
1073 return used;
1076 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1078 unsigned long size, used;
1079 static const char valid_loose_object_type[8] = {
1080 0, /* OBJ_EXT */
1081 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1082 0, /* "delta" and others are invalid in a loose object */
1084 enum object_type type;
1086 /* Get the data stream */
1087 memset(stream, 0, sizeof(*stream));
1088 stream->next_in = map;
1089 stream->avail_in = mapsize;
1090 stream->next_out = buffer;
1091 stream->avail_out = bufsiz;
1093 if (legacy_loose_object(map)) {
1094 inflateInit(stream);
1095 return inflate(stream, 0);
1100 * There used to be a second loose object header format which
1101 * was meant to mimic the in-pack format, allowing for direct
1102 * copy of the object data. This format turned up not to be
1103 * really worth it and we don't write it any longer. But we
1104 * can still read it.
1106 used = unpack_object_header_gently(map, mapsize, &type, &size);
1107 if (!used || !valid_loose_object_type[type])
1108 return -1;
1109 map += used;
1110 mapsize -= used;
1112 /* Set up the stream for the rest.. */
1113 stream->next_in = map;
1114 stream->avail_in = mapsize;
1115 inflateInit(stream);
1117 /* And generate the fake traditional header */
1118 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1119 typename(type), size);
1120 return 0;
1123 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1125 int bytes = strlen(buffer) + 1;
1126 unsigned char *buf = xmalloc(1+size);
1127 unsigned long n;
1128 int status = Z_OK;
1130 n = stream->total_out - bytes;
1131 if (n > size)
1132 n = size;
1133 memcpy(buf, (char *) buffer + bytes, n);
1134 bytes = n;
1135 if (bytes <= size) {
1137 * The above condition must be (bytes <= size), not
1138 * (bytes < size). In other words, even though we
1139 * expect no more output and set avail_out to zer0,
1140 * the input zlib stream may have bytes that express
1141 * "this concludes the stream", and we *do* want to
1142 * eat that input.
1144 * Otherwise we would not be able to test that we
1145 * consumed all the input to reach the expected size;
1146 * we also want to check that zlib tells us that all
1147 * went well with status == Z_STREAM_END at the end.
1149 stream->next_out = buf + bytes;
1150 stream->avail_out = size - bytes;
1151 while (status == Z_OK)
1152 status = inflate(stream, Z_FINISH);
1154 buf[size] = 0;
1155 if (status == Z_STREAM_END && !stream->avail_in) {
1156 inflateEnd(stream);
1157 return buf;
1160 if (status < 0)
1161 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1162 else if (stream->avail_in)
1163 error("garbage at end of loose object '%s'",
1164 sha1_to_hex(sha1));
1165 free(buf);
1166 return NULL;
1170 * We used to just use "sscanf()", but that's actually way
1171 * too permissive for what we want to check. So do an anal
1172 * object header parse by hand.
1174 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1176 char type[10];
1177 int i;
1178 unsigned long size;
1181 * The type can be at most ten bytes (including the
1182 * terminating '\0' that we add), and is followed by
1183 * a space.
1185 i = 0;
1186 for (;;) {
1187 char c = *hdr++;
1188 if (c == ' ')
1189 break;
1190 type[i++] = c;
1191 if (i >= sizeof(type))
1192 return -1;
1194 type[i] = 0;
1197 * The length must follow immediately, and be in canonical
1198 * decimal format (ie "010" is not valid).
1200 size = *hdr++ - '0';
1201 if (size > 9)
1202 return -1;
1203 if (size) {
1204 for (;;) {
1205 unsigned long c = *hdr - '0';
1206 if (c > 9)
1207 break;
1208 hdr++;
1209 size = size * 10 + c;
1212 *sizep = size;
1215 * The length must be followed by a zero byte
1217 return *hdr ? -1 : type_from_string(type);
1220 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1222 int ret;
1223 z_stream stream;
1224 char hdr[8192];
1226 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1227 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1228 return NULL;
1230 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1233 unsigned long get_size_from_delta(struct packed_git *p,
1234 struct pack_window **w_curs,
1235 off_t curpos)
1237 const unsigned char *data;
1238 unsigned char delta_head[20], *in;
1239 z_stream stream;
1240 int st;
1242 memset(&stream, 0, sizeof(stream));
1243 stream.next_out = delta_head;
1244 stream.avail_out = sizeof(delta_head);
1246 inflateInit(&stream);
1247 do {
1248 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1249 stream.next_in = in;
1250 st = inflate(&stream, Z_FINISH);
1251 curpos += stream.next_in - in;
1252 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1253 stream.total_out < sizeof(delta_head));
1254 inflateEnd(&stream);
1255 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1256 die("delta data unpack-initial failed");
1258 /* Examine the initial part of the delta to figure out
1259 * the result size.
1261 data = delta_head;
1263 /* ignore base size */
1264 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1266 /* Read the result size */
1267 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1270 static off_t get_delta_base(struct packed_git *p,
1271 struct pack_window **w_curs,
1272 off_t *curpos,
1273 enum object_type type,
1274 off_t delta_obj_offset)
1276 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1277 off_t base_offset;
1279 /* use_pack() assured us we have [base_info, base_info + 20)
1280 * as a range that we can look at without walking off the
1281 * end of the mapped window. Its actually the hash size
1282 * that is assured. An OFS_DELTA longer than the hash size
1283 * is stupid, as then a REF_DELTA would be smaller to store.
1285 if (type == OBJ_OFS_DELTA) {
1286 unsigned used = 0;
1287 unsigned char c = base_info[used++];
1288 base_offset = c & 127;
1289 while (c & 128) {
1290 base_offset += 1;
1291 if (!base_offset || MSB(base_offset, 7))
1292 die("offset value overflow for delta base object");
1293 c = base_info[used++];
1294 base_offset = (base_offset << 7) + (c & 127);
1296 base_offset = delta_obj_offset - base_offset;
1297 if (base_offset >= delta_obj_offset)
1298 die("delta base offset out of bound");
1299 *curpos += used;
1300 } else if (type == OBJ_REF_DELTA) {
1301 /* The base entry _must_ be in the same pack */
1302 base_offset = find_pack_entry_one(base_info, p);
1303 if (!base_offset)
1304 die("failed to find delta-pack base object %s",
1305 sha1_to_hex(base_info));
1306 *curpos += 20;
1307 } else
1308 die("I am totally screwed");
1309 return base_offset;
1312 /* forward declaration for a mutually recursive function */
1313 static int packed_object_info(struct packed_git *p, off_t offset,
1314 unsigned long *sizep);
1316 static int packed_delta_info(struct packed_git *p,
1317 struct pack_window **w_curs,
1318 off_t curpos,
1319 enum object_type type,
1320 off_t obj_offset,
1321 unsigned long *sizep)
1323 off_t base_offset;
1325 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1326 type = packed_object_info(p, base_offset, NULL);
1328 /* We choose to only get the type of the base object and
1329 * ignore potentially corrupt pack file that expects the delta
1330 * based on a base with a wrong size. This saves tons of
1331 * inflate() calls.
1333 if (sizep)
1334 *sizep = get_size_from_delta(p, w_curs, curpos);
1336 return type;
1339 static int unpack_object_header(struct packed_git *p,
1340 struct pack_window **w_curs,
1341 off_t *curpos,
1342 unsigned long *sizep)
1344 unsigned char *base;
1345 unsigned int left;
1346 unsigned long used;
1347 enum object_type type;
1349 /* use_pack() assures us we have [base, base + 20) available
1350 * as a range that we can look at at. (Its actually the hash
1351 * size that is assured.) With our object header encoding
1352 * the maximum deflated object size is 2^137, which is just
1353 * insane, so we know won't exceed what we have been given.
1355 base = use_pack(p, w_curs, *curpos, &left);
1356 used = unpack_object_header_gently(base, left, &type, sizep);
1357 if (!used)
1358 die("object offset outside of pack file");
1359 *curpos += used;
1361 return type;
1364 const char *packed_object_info_detail(struct packed_git *p,
1365 off_t obj_offset,
1366 unsigned long *size,
1367 unsigned long *store_size,
1368 unsigned int *delta_chain_length,
1369 unsigned char *base_sha1)
1371 struct pack_window *w_curs = NULL;
1372 off_t curpos;
1373 unsigned long dummy;
1374 unsigned char *next_sha1;
1375 enum object_type type;
1376 struct revindex_entry *revidx;
1378 *delta_chain_length = 0;
1379 curpos = obj_offset;
1380 type = unpack_object_header(p, &w_curs, &curpos, size);
1382 revidx = find_pack_revindex(p, obj_offset);
1383 *store_size = revidx[1].offset - obj_offset;
1385 for (;;) {
1386 switch (type) {
1387 default:
1388 die("pack %s contains unknown object type %d",
1389 p->pack_name, type);
1390 case OBJ_COMMIT:
1391 case OBJ_TREE:
1392 case OBJ_BLOB:
1393 case OBJ_TAG:
1394 unuse_pack(&w_curs);
1395 return typename(type);
1396 case OBJ_OFS_DELTA:
1397 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1398 if (*delta_chain_length == 0) {
1399 revidx = find_pack_revindex(p, obj_offset);
1400 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1402 break;
1403 case OBJ_REF_DELTA:
1404 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1405 if (*delta_chain_length == 0)
1406 hashcpy(base_sha1, next_sha1);
1407 obj_offset = find_pack_entry_one(next_sha1, p);
1408 break;
1410 (*delta_chain_length)++;
1411 curpos = obj_offset;
1412 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1416 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1417 unsigned long *sizep)
1419 struct pack_window *w_curs = NULL;
1420 unsigned long size;
1421 off_t curpos = obj_offset;
1422 enum object_type type;
1424 type = unpack_object_header(p, &w_curs, &curpos, &size);
1426 switch (type) {
1427 case OBJ_OFS_DELTA:
1428 case OBJ_REF_DELTA:
1429 type = packed_delta_info(p, &w_curs, curpos,
1430 type, obj_offset, sizep);
1431 break;
1432 case OBJ_COMMIT:
1433 case OBJ_TREE:
1434 case OBJ_BLOB:
1435 case OBJ_TAG:
1436 if (sizep)
1437 *sizep = size;
1438 break;
1439 default:
1440 die("pack %s contains unknown object type %d",
1441 p->pack_name, type);
1443 unuse_pack(&w_curs);
1444 return type;
1447 static void *unpack_compressed_entry(struct packed_git *p,
1448 struct pack_window **w_curs,
1449 off_t curpos,
1450 unsigned long size)
1452 int st;
1453 z_stream stream;
1454 unsigned char *buffer, *in;
1456 buffer = xmalloc(size + 1);
1457 buffer[size] = 0;
1458 memset(&stream, 0, sizeof(stream));
1459 stream.next_out = buffer;
1460 stream.avail_out = size;
1462 inflateInit(&stream);
1463 do {
1464 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1465 stream.next_in = in;
1466 st = inflate(&stream, Z_FINISH);
1467 curpos += stream.next_in - in;
1468 } while (st == Z_OK || st == Z_BUF_ERROR);
1469 inflateEnd(&stream);
1470 if ((st != Z_STREAM_END) || stream.total_out != size) {
1471 free(buffer);
1472 return NULL;
1475 return buffer;
1478 #define MAX_DELTA_CACHE (256)
1480 static size_t delta_base_cached;
1482 static struct delta_base_cache_lru_list {
1483 struct delta_base_cache_lru_list *prev;
1484 struct delta_base_cache_lru_list *next;
1485 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1487 static struct delta_base_cache_entry {
1488 struct delta_base_cache_lru_list lru;
1489 void *data;
1490 struct packed_git *p;
1491 off_t base_offset;
1492 unsigned long size;
1493 enum object_type type;
1494 } delta_base_cache[MAX_DELTA_CACHE];
1496 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1498 unsigned long hash;
1500 hash = (unsigned long)p + (unsigned long)base_offset;
1501 hash += (hash >> 8) + (hash >> 16);
1502 return hash % MAX_DELTA_CACHE;
1505 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1506 unsigned long *base_size, enum object_type *type, int keep_cache)
1508 void *ret;
1509 unsigned long hash = pack_entry_hash(p, base_offset);
1510 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1512 ret = ent->data;
1513 if (ret && ent->p == p && ent->base_offset == base_offset)
1514 goto found_cache_entry;
1515 return unpack_entry(p, base_offset, type, base_size);
1517 found_cache_entry:
1518 if (!keep_cache) {
1519 ent->data = NULL;
1520 ent->lru.next->prev = ent->lru.prev;
1521 ent->lru.prev->next = ent->lru.next;
1522 delta_base_cached -= ent->size;
1523 } else {
1524 ret = xmemdupz(ent->data, ent->size);
1526 *type = ent->type;
1527 *base_size = ent->size;
1528 return ret;
1531 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1533 if (ent->data) {
1534 free(ent->data);
1535 ent->data = NULL;
1536 ent->lru.next->prev = ent->lru.prev;
1537 ent->lru.prev->next = ent->lru.next;
1538 delta_base_cached -= ent->size;
1542 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1543 void *base, unsigned long base_size, enum object_type type)
1545 unsigned long hash = pack_entry_hash(p, base_offset);
1546 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1547 struct delta_base_cache_lru_list *lru;
1549 release_delta_base_cache(ent);
1550 delta_base_cached += base_size;
1552 for (lru = delta_base_cache_lru.next;
1553 delta_base_cached > delta_base_cache_limit
1554 && lru != &delta_base_cache_lru;
1555 lru = lru->next) {
1556 struct delta_base_cache_entry *f = (void *)lru;
1557 if (f->type == OBJ_BLOB)
1558 release_delta_base_cache(f);
1560 for (lru = delta_base_cache_lru.next;
1561 delta_base_cached > delta_base_cache_limit
1562 && lru != &delta_base_cache_lru;
1563 lru = lru->next) {
1564 struct delta_base_cache_entry *f = (void *)lru;
1565 release_delta_base_cache(f);
1568 ent->p = p;
1569 ent->base_offset = base_offset;
1570 ent->type = type;
1571 ent->data = base;
1572 ent->size = base_size;
1573 ent->lru.next = &delta_base_cache_lru;
1574 ent->lru.prev = delta_base_cache_lru.prev;
1575 delta_base_cache_lru.prev->next = &ent->lru;
1576 delta_base_cache_lru.prev = &ent->lru;
1579 static void *unpack_delta_entry(struct packed_git *p,
1580 struct pack_window **w_curs,
1581 off_t curpos,
1582 unsigned long delta_size,
1583 off_t obj_offset,
1584 enum object_type *type,
1585 unsigned long *sizep)
1587 void *delta_data, *result, *base;
1588 unsigned long base_size;
1589 off_t base_offset;
1591 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1592 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1593 if (!base)
1594 die("failed to read delta base object"
1595 " at %"PRIuMAX" from %s",
1596 (uintmax_t)base_offset, p->pack_name);
1598 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1599 if (!delta_data)
1600 die("failed to unpack compressed delta"
1601 " at %"PRIuMAX" from %s",
1602 (uintmax_t)curpos, p->pack_name);
1603 result = patch_delta(base, base_size,
1604 delta_data, delta_size,
1605 sizep);
1606 if (!result)
1607 die("failed to apply delta");
1608 free(delta_data);
1609 add_delta_base_cache(p, base_offset, base, base_size, *type);
1610 return result;
1613 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1614 enum object_type *type, unsigned long *sizep)
1616 struct pack_window *w_curs = NULL;
1617 off_t curpos = obj_offset;
1618 void *data;
1620 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1621 switch (*type) {
1622 case OBJ_OFS_DELTA:
1623 case OBJ_REF_DELTA:
1624 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1625 obj_offset, type, sizep);
1626 break;
1627 case OBJ_COMMIT:
1628 case OBJ_TREE:
1629 case OBJ_BLOB:
1630 case OBJ_TAG:
1631 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1632 break;
1633 default:
1634 die("unknown object type %i in %s", *type, p->pack_name);
1636 unuse_pack(&w_curs);
1637 return data;
1640 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1641 uint32_t n)
1643 const unsigned char *index = p->index_data;
1644 if (!index) {
1645 if (open_pack_index(p))
1646 return NULL;
1647 index = p->index_data;
1649 if (n >= p->num_objects)
1650 return NULL;
1651 index += 4 * 256;
1652 if (p->index_version == 1) {
1653 return index + 24 * n + 4;
1654 } else {
1655 index += 8;
1656 return index + 20 * n;
1660 static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1662 const unsigned char *index = p->index_data;
1663 index += 4 * 256;
1664 if (p->index_version == 1) {
1665 return ntohl(*((uint32_t *)(index + 24 * n)));
1666 } else {
1667 uint32_t off;
1668 index += 8 + p->num_objects * (20 + 4);
1669 off = ntohl(*((uint32_t *)(index + 4 * n)));
1670 if (!(off & 0x80000000))
1671 return off;
1672 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1673 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1674 ntohl(*((uint32_t *)(index + 4)));
1678 off_t find_pack_entry_one(const unsigned char *sha1,
1679 struct packed_git *p)
1681 const uint32_t *level1_ofs = p->index_data;
1682 const unsigned char *index = p->index_data;
1683 unsigned hi, lo, stride;
1684 static int use_lookup = -1;
1685 static int debug_lookup = -1;
1687 if (debug_lookup < 0)
1688 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1690 if (!index) {
1691 if (open_pack_index(p))
1692 return 0;
1693 level1_ofs = p->index_data;
1694 index = p->index_data;
1696 if (p->index_version > 1) {
1697 level1_ofs += 2;
1698 index += 8;
1700 index += 4 * 256;
1701 hi = ntohl(level1_ofs[*sha1]);
1702 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1703 if (p->index_version > 1) {
1704 stride = 20;
1705 } else {
1706 stride = 24;
1707 index += 4;
1710 if (debug_lookup)
1711 printf("%02x%02x%02x... lo %u hi %u nr %u\n",
1712 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1714 if (use_lookup < 0)
1715 use_lookup = !!getenv("GIT_USE_LOOKUP");
1716 if (use_lookup) {
1717 int pos = sha1_entry_pos(index, stride, 0,
1718 lo, hi, p->num_objects, sha1);
1719 if (pos < 0)
1720 return 0;
1721 return nth_packed_object_offset(p, pos);
1724 do {
1725 unsigned mi = (lo + hi) / 2;
1726 int cmp = hashcmp(index + mi * stride, sha1);
1728 if (debug_lookup)
1729 printf("lo %u hi %u rg %u mi %u\n",
1730 lo, hi, hi - lo, mi);
1731 if (!cmp)
1732 return nth_packed_object_offset(p, mi);
1733 if (cmp > 0)
1734 hi = mi;
1735 else
1736 lo = mi+1;
1737 } while (lo < hi);
1738 return 0;
1741 int matches_pack_name(struct packed_git *p, const char *name)
1743 const char *last_c, *c;
1745 if (!strcmp(p->pack_name, name))
1746 return 1;
1748 for (c = p->pack_name, last_c = c; *c;)
1749 if (*c == '/')
1750 last_c = ++c;
1751 else
1752 ++c;
1753 if (!strcmp(last_c, name))
1754 return 1;
1756 return 0;
1759 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1761 static struct packed_git *last_found = (void *)1;
1762 struct packed_git *p;
1763 off_t offset;
1765 prepare_packed_git();
1766 if (!packed_git)
1767 return 0;
1768 p = (last_found == (void *)1) ? packed_git : last_found;
1770 do {
1771 if (ignore_packed) {
1772 const char **ig;
1773 for (ig = ignore_packed; *ig; ig++)
1774 if (matches_pack_name(p, *ig))
1775 break;
1776 if (*ig)
1777 goto next;
1780 offset = find_pack_entry_one(sha1, p);
1781 if (offset) {
1783 * We are about to tell the caller where they can
1784 * locate the requested object. We better make
1785 * sure the packfile is still here and can be
1786 * accessed before supplying that answer, as
1787 * it may have been deleted since the index
1788 * was loaded!
1790 if (p->pack_fd == -1 && open_packed_git(p)) {
1791 error("packfile %s cannot be accessed", p->pack_name);
1792 goto next;
1794 e->offset = offset;
1795 e->p = p;
1796 hashcpy(e->sha1, sha1);
1797 last_found = p;
1798 return 1;
1801 next:
1802 if (p == last_found)
1803 p = packed_git;
1804 else
1805 p = p->next;
1806 if (p == last_found)
1807 p = p->next;
1808 } while (p);
1809 return 0;
1812 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1813 struct packed_git *packs)
1815 struct packed_git *p;
1817 for (p = packs; p; p = p->next) {
1818 if (find_pack_entry_one(sha1, p))
1819 return p;
1821 return NULL;
1825 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1827 int status;
1828 unsigned long mapsize, size;
1829 void *map;
1830 z_stream stream;
1831 char hdr[32];
1833 map = map_sha1_file(sha1, &mapsize);
1834 if (!map)
1835 return error("unable to find %s", sha1_to_hex(sha1));
1836 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1837 status = error("unable to unpack %s header",
1838 sha1_to_hex(sha1));
1839 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1840 status = error("unable to parse %s header", sha1_to_hex(sha1));
1841 else if (sizep)
1842 *sizep = size;
1843 inflateEnd(&stream);
1844 munmap(map, mapsize);
1845 return status;
1848 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1850 struct pack_entry e;
1852 if (!find_pack_entry(sha1, &e, NULL)) {
1853 reprepare_packed_git();
1854 if (!find_pack_entry(sha1, &e, NULL))
1855 return sha1_loose_object_info(sha1, sizep);
1857 return packed_object_info(e.p, e.offset, sizep);
1860 static void *read_packed_sha1(const unsigned char *sha1,
1861 enum object_type *type, unsigned long *size)
1863 struct pack_entry e;
1865 if (!find_pack_entry(sha1, &e, NULL))
1866 return NULL;
1867 else
1868 return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1872 * This is meant to hold a *small* number of objects that you would
1873 * want read_sha1_file() to be able to return, but yet you do not want
1874 * to write them into the object store (e.g. a browse-only
1875 * application).
1877 static struct cached_object {
1878 unsigned char sha1[20];
1879 enum object_type type;
1880 void *buf;
1881 unsigned long size;
1882 } *cached_objects;
1883 static int cached_object_nr, cached_object_alloc;
1885 static struct cached_object empty_tree = {
1886 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
1887 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
1888 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
1889 OBJ_TREE,
1894 static struct cached_object *find_cached_object(const unsigned char *sha1)
1896 int i;
1897 struct cached_object *co = cached_objects;
1899 for (i = 0; i < cached_object_nr; i++, co++) {
1900 if (!hashcmp(co->sha1, sha1))
1901 return co;
1903 if (!hashcmp(sha1, empty_tree.sha1))
1904 return &empty_tree;
1905 return NULL;
1908 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1909 unsigned char *sha1)
1911 struct cached_object *co;
1913 hash_sha1_file(buf, len, typename(type), sha1);
1914 if (has_sha1_file(sha1) || find_cached_object(sha1))
1915 return 0;
1916 if (cached_object_alloc <= cached_object_nr) {
1917 cached_object_alloc = alloc_nr(cached_object_alloc);
1918 cached_objects = xrealloc(cached_objects,
1919 sizeof(*cached_objects) *
1920 cached_object_alloc);
1922 co = &cached_objects[cached_object_nr++];
1923 co->size = len;
1924 co->type = type;
1925 co->buf = xmalloc(len);
1926 memcpy(co->buf, buf, len);
1927 hashcpy(co->sha1, sha1);
1928 return 0;
1931 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1932 unsigned long *size)
1934 unsigned long mapsize;
1935 void *map, *buf;
1936 struct cached_object *co;
1938 co = find_cached_object(sha1);
1939 if (co) {
1940 *type = co->type;
1941 *size = co->size;
1942 return xmemdupz(co->buf, co->size);
1945 buf = read_packed_sha1(sha1, type, size);
1946 if (buf)
1947 return buf;
1948 map = map_sha1_file(sha1, &mapsize);
1949 if (map) {
1950 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1951 munmap(map, mapsize);
1952 return buf;
1954 reprepare_packed_git();
1955 return read_packed_sha1(sha1, type, size);
1958 void *read_object_with_reference(const unsigned char *sha1,
1959 const char *required_type_name,
1960 unsigned long *size,
1961 unsigned char *actual_sha1_return)
1963 enum object_type type, required_type;
1964 void *buffer;
1965 unsigned long isize;
1966 unsigned char actual_sha1[20];
1968 required_type = type_from_string(required_type_name);
1969 hashcpy(actual_sha1, sha1);
1970 while (1) {
1971 int ref_length = -1;
1972 const char *ref_type = NULL;
1974 buffer = read_sha1_file(actual_sha1, &type, &isize);
1975 if (!buffer)
1976 return NULL;
1977 if (type == required_type) {
1978 *size = isize;
1979 if (actual_sha1_return)
1980 hashcpy(actual_sha1_return, actual_sha1);
1981 return buffer;
1983 /* Handle references */
1984 else if (type == OBJ_COMMIT)
1985 ref_type = "tree ";
1986 else if (type == OBJ_TAG)
1987 ref_type = "object ";
1988 else {
1989 free(buffer);
1990 return NULL;
1992 ref_length = strlen(ref_type);
1994 if (ref_length + 40 > isize ||
1995 memcmp(buffer, ref_type, ref_length) ||
1996 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1997 free(buffer);
1998 return NULL;
2000 free(buffer);
2001 /* Now we have the ID of the referred-to object in
2002 * actual_sha1. Check again. */
2006 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2007 const char *type, unsigned char *sha1,
2008 char *hdr, int *hdrlen)
2010 SHA_CTX c;
2012 /* Generate the header */
2013 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2015 /* Sha1.. */
2016 SHA1_Init(&c);
2017 SHA1_Update(&c, hdr, *hdrlen);
2018 SHA1_Update(&c, buf, len);
2019 SHA1_Final(sha1, &c);
2023 * Link the tempfile to the final place, possibly creating the
2024 * last directory level as you do so.
2026 * Returns the errno on failure, 0 on success.
2028 static int link_temp_to_file(const char *tmpfile, const char *filename)
2030 int ret;
2031 char *dir;
2033 if (!link(tmpfile, filename))
2034 return 0;
2037 * Try to mkdir the last path component if that failed.
2039 * Re-try the "link()" regardless of whether the mkdir
2040 * succeeds, since a race might mean that somebody
2041 * else succeeded.
2043 ret = errno;
2044 dir = strrchr(filename, '/');
2045 if (dir) {
2046 *dir = 0;
2047 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
2048 *dir = '/';
2049 return -2;
2051 *dir = '/';
2052 if (!link(tmpfile, filename))
2053 return 0;
2054 ret = errno;
2056 return ret;
2060 * Move the just written object into its final resting place
2062 int move_temp_to_file(const char *tmpfile, const char *filename)
2064 int ret = link_temp_to_file(tmpfile, filename);
2067 * Coda hack - coda doesn't like cross-directory links,
2068 * so we fall back to a rename, which will mean that it
2069 * won't be able to check collisions, but that's not a
2070 * big deal.
2072 * The same holds for FAT formatted media.
2074 * When this succeeds, we just return 0. We have nothing
2075 * left to unlink.
2077 if (ret && ret != EEXIST) {
2078 if (!rename(tmpfile, filename))
2079 return 0;
2080 ret = errno;
2082 unlink(tmpfile);
2083 if (ret) {
2084 if (ret != EEXIST) {
2085 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2087 /* FIXME!!! Collision check here ? */
2090 return 0;
2093 static int write_buffer(int fd, const void *buf, size_t len)
2095 if (write_in_full(fd, buf, len) < 0)
2096 return error("file write error (%s)", strerror(errno));
2097 return 0;
2100 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2101 unsigned char *sha1)
2103 char hdr[32];
2104 int hdrlen;
2105 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2106 return 0;
2109 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2111 int size, ret;
2112 unsigned char *compressed;
2113 z_stream stream;
2114 unsigned char sha1[20];
2115 char *filename;
2116 static char tmpfile[PATH_MAX];
2117 char hdr[32];
2118 int fd, hdrlen;
2120 /* Normally if we have it in the pack then we do not bother writing
2121 * it out into .git/objects/??/?{38} file.
2123 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2124 filename = sha1_file_name(sha1);
2125 if (returnsha1)
2126 hashcpy(returnsha1, sha1);
2127 if (has_sha1_file(sha1))
2128 return 0;
2129 fd = open(filename, O_RDONLY);
2130 if (fd >= 0) {
2132 * FIXME!!! We might do collision checking here, but we'd
2133 * need to uncompress the old file and check it. Later.
2135 close(fd);
2136 return 0;
2139 if (errno != ENOENT) {
2140 return error("sha1 file %s: %s\n", filename, strerror(errno));
2143 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2145 fd = mkstemp(tmpfile);
2146 if (fd < 0) {
2147 if (errno == EPERM)
2148 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2149 else
2150 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2153 /* Set it up */
2154 memset(&stream, 0, sizeof(stream));
2155 deflateInit(&stream, zlib_compression_level);
2156 size = 8 + deflateBound(&stream, len+hdrlen);
2157 compressed = xmalloc(size);
2159 /* Compress it */
2160 stream.next_out = compressed;
2161 stream.avail_out = size;
2163 /* First header.. */
2164 stream.next_in = (unsigned char *)hdr;
2165 stream.avail_in = hdrlen;
2166 while (deflate(&stream, 0) == Z_OK)
2167 /* nothing */;
2169 /* Then the data itself.. */
2170 stream.next_in = buf;
2171 stream.avail_in = len;
2172 ret = deflate(&stream, Z_FINISH);
2173 if (ret != Z_STREAM_END)
2174 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2176 ret = deflateEnd(&stream);
2177 if (ret != Z_OK)
2178 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2180 size = stream.total_out;
2182 if (write_buffer(fd, compressed, size) < 0)
2183 die("unable to write sha1 file");
2184 fchmod(fd, 0444);
2185 if (close(fd))
2186 die("unable to write sha1 file");
2187 free(compressed);
2189 return move_temp_to_file(tmpfile, filename);
2193 * We need to unpack and recompress the object for writing
2194 * it out to a different file.
2196 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
2198 size_t size;
2199 z_stream stream;
2200 unsigned char *unpacked;
2201 unsigned long len;
2202 enum object_type type;
2203 char hdr[32];
2204 int hdrlen;
2205 void *buf;
2207 /* need to unpack and recompress it by itself */
2208 unpacked = read_packed_sha1(sha1, &type, &len);
2209 if (!unpacked)
2210 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2212 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2214 /* Set it up */
2215 memset(&stream, 0, sizeof(stream));
2216 deflateInit(&stream, zlib_compression_level);
2217 size = deflateBound(&stream, len + hdrlen);
2218 buf = xmalloc(size);
2220 /* Compress it */
2221 stream.next_out = buf;
2222 stream.avail_out = size;
2224 /* First header.. */
2225 stream.next_in = (void *)hdr;
2226 stream.avail_in = hdrlen;
2227 while (deflate(&stream, 0) == Z_OK)
2228 /* nothing */;
2230 /* Then the data itself.. */
2231 stream.next_in = unpacked;
2232 stream.avail_in = len;
2233 while (deflate(&stream, Z_FINISH) == Z_OK)
2234 /* nothing */;
2235 deflateEnd(&stream);
2236 free(unpacked);
2238 *objsize = stream.total_out;
2239 return buf;
2242 int write_sha1_to_fd(int fd, const unsigned char *sha1)
2244 int retval;
2245 unsigned long objsize;
2246 void *buf = map_sha1_file(sha1, &objsize);
2248 if (buf) {
2249 retval = write_buffer(fd, buf, objsize);
2250 munmap(buf, objsize);
2251 return retval;
2254 buf = repack_object(sha1, &objsize);
2255 retval = write_buffer(fd, buf, objsize);
2256 free(buf);
2257 return retval;
2260 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2261 size_t bufsize, size_t *bufposn)
2263 char tmpfile[PATH_MAX];
2264 int local;
2265 z_stream stream;
2266 unsigned char real_sha1[20];
2267 unsigned char discard[4096];
2268 int ret;
2269 SHA_CTX c;
2271 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2273 local = mkstemp(tmpfile);
2274 if (local < 0) {
2275 if (errno == EPERM)
2276 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2277 else
2278 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2281 memset(&stream, 0, sizeof(stream));
2283 inflateInit(&stream);
2285 SHA1_Init(&c);
2287 do {
2288 ssize_t size;
2289 if (*bufposn) {
2290 stream.avail_in = *bufposn;
2291 stream.next_in = (unsigned char *) buffer;
2292 do {
2293 stream.next_out = discard;
2294 stream.avail_out = sizeof(discard);
2295 ret = inflate(&stream, Z_SYNC_FLUSH);
2296 SHA1_Update(&c, discard, sizeof(discard) -
2297 stream.avail_out);
2298 } while (stream.avail_in && ret == Z_OK);
2299 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2300 die("unable to write sha1 file");
2301 memmove(buffer, buffer + *bufposn - stream.avail_in,
2302 stream.avail_in);
2303 *bufposn = stream.avail_in;
2304 if (ret != Z_OK)
2305 break;
2307 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
2308 if (size <= 0) {
2309 close(local);
2310 unlink(tmpfile);
2311 if (!size)
2312 return error("Connection closed?");
2313 perror("Reading from connection");
2314 return -1;
2316 *bufposn += size;
2317 } while (1);
2318 inflateEnd(&stream);
2320 fchmod(local, 0444);
2321 if (close(local) != 0)
2322 die("unable to write sha1 file");
2323 SHA1_Final(real_sha1, &c);
2324 if (ret != Z_STREAM_END) {
2325 unlink(tmpfile);
2326 return error("File %s corrupted", sha1_to_hex(sha1));
2328 if (hashcmp(sha1, real_sha1)) {
2329 unlink(tmpfile);
2330 return error("File %s has bad hash", sha1_to_hex(sha1));
2333 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
2336 int has_pack_index(const unsigned char *sha1)
2338 struct stat st;
2339 if (stat(sha1_pack_index_name(sha1), &st))
2340 return 0;
2341 return 1;
2344 int has_pack_file(const unsigned char *sha1)
2346 struct stat st;
2347 if (stat(sha1_pack_name(sha1), &st))
2348 return 0;
2349 return 1;
2352 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2354 struct pack_entry e;
2355 return find_pack_entry(sha1, &e, ignore_packed);
2358 int has_sha1_file(const unsigned char *sha1)
2360 struct stat st;
2361 struct pack_entry e;
2363 if (find_pack_entry(sha1, &e, NULL))
2364 return 1;
2365 return find_sha1_file(sha1, &st) ? 1 : 0;
2368 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2370 struct strbuf buf;
2371 int ret;
2373 strbuf_init(&buf, 0);
2374 if (strbuf_read(&buf, fd, 4096) < 0) {
2375 strbuf_release(&buf);
2376 return -1;
2379 if (!type)
2380 type = blob_type;
2381 if (write_object)
2382 ret = write_sha1_file(buf.buf, buf.len, type, sha1);
2383 else
2384 ret = hash_sha1_file(buf.buf, buf.len, type, sha1);
2385 strbuf_release(&buf);
2387 return ret;
2390 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2391 enum object_type type, const char *path)
2393 size_t size = xsize_t(st->st_size);
2394 void *buf = NULL;
2395 int ret, re_allocated = 0;
2397 if (size)
2398 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2399 close(fd);
2401 if (!type)
2402 type = OBJ_BLOB;
2405 * Convert blobs to git internal format
2407 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2408 struct strbuf nbuf;
2409 strbuf_init(&nbuf, 0);
2410 if (convert_to_git(path, buf, size, &nbuf,
2411 write_object ? safe_crlf : 0)) {
2412 munmap(buf, size);
2413 buf = strbuf_detach(&nbuf, &size);
2414 re_allocated = 1;
2418 if (write_object)
2419 ret = write_sha1_file(buf, size, typename(type), sha1);
2420 else
2421 ret = hash_sha1_file(buf, size, typename(type), sha1);
2422 if (re_allocated) {
2423 free(buf);
2424 return ret;
2426 if (size)
2427 munmap(buf, size);
2428 return ret;
2431 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2433 int fd;
2434 char *target;
2435 size_t len;
2437 switch (st->st_mode & S_IFMT) {
2438 case S_IFREG:
2439 fd = open(path, O_RDONLY);
2440 if (fd < 0)
2441 return error("open(\"%s\"): %s", path,
2442 strerror(errno));
2443 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2444 return error("%s: failed to insert into database",
2445 path);
2446 break;
2447 case S_IFLNK:
2448 len = xsize_t(st->st_size);
2449 target = xmalloc(len + 1);
2450 if (readlink(path, target, len + 1) != st->st_size) {
2451 char *errstr = strerror(errno);
2452 free(target);
2453 return error("readlink(\"%s\"): %s", path,
2454 errstr);
2456 if (!write_object)
2457 hash_sha1_file(target, len, blob_type, sha1);
2458 else if (write_sha1_file(target, len, blob_type, sha1))
2459 return error("%s: failed to insert into database",
2460 path);
2461 free(target);
2462 break;
2463 case S_IFDIR:
2464 return resolve_gitlink_ref(path, "HEAD", sha1);
2465 default:
2466 return error("%s: unsupported file type", path);
2468 return 0;
2471 int read_pack_header(int fd, struct pack_header *header)
2473 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2474 /* "eof before pack header was fully read" */
2475 return PH_ERROR_EOF;
2477 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2478 /* "protocol error (pack signature mismatch detected)" */
2479 return PH_ERROR_PACK_SIGNATURE;
2480 if (!pack_version_ok(header->hdr_version))
2481 /* "protocol error (pack version unsupported)" */
2482 return PH_ERROR_PROTOCOL;
2483 return 0;