Update README.MinGW.
[git/mingw.git] / sha1_file.c
blob93ce30eae9f143de3e4a95a50a25851f4fc91dc8
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"
17 #ifndef O_NOATIME
18 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
19 #define O_NOATIME 01000000
20 #else
21 #define O_NOATIME 0
22 #endif
23 #endif
25 #ifdef NO_C99_FORMAT
26 #define SZ_FMT "lu"
27 #else
28 #define SZ_FMT "zu"
29 #endif
31 const unsigned char null_sha1[20];
33 static unsigned int sha1_file_open_flag = O_NOATIME;
35 signed char hexval_table[256] = {
36 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
37 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
38 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
42 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
43 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
44 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
45 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
47 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
48 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
70 int get_sha1_hex(const char *hex, unsigned char *sha1)
72 int i;
73 for (i = 0; i < 20; i++) {
74 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
75 if (val & ~0xff)
76 return -1;
77 *sha1++ = val;
78 hex += 2;
80 return 0;
83 /* returns the number of chars to skip to first component */
84 static inline int is_path_absolute(const char *path)
86 #ifdef __MINGW32__
87 if (isalpha(path[0]) && path[1] == ':')
88 return 2 + (path[2] == '/');
89 /* TODO: C:dir/file 'relative' paths are not catered for */
90 #endif
91 return *path == '/';
94 int safe_create_leading_directories(char *path)
96 char *pos = path + is_path_absolute(path);
97 struct stat st;
99 while (pos) {
100 pos = strchr(pos, '/');
101 if (!pos)
102 break;
103 *pos = 0;
104 if (!stat(path, &st)) {
105 /* path exists */
106 if (!S_ISDIR(st.st_mode)) {
107 *pos = '/';
108 return -3;
111 else if (mkdir(path, 0777)) {
112 *pos = '/';
113 return -1;
115 else if (adjust_shared_perm(path)) {
116 *pos = '/';
117 return -2;
119 *pos++ = '/';
121 return 0;
124 char * sha1_to_hex(const unsigned char *sha1)
126 static int bufno;
127 static char hexbuffer[4][50];
128 static const char hex[] = "0123456789abcdef";
129 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
130 int i;
132 for (i = 0; i < 20; i++) {
133 unsigned int val = *sha1++;
134 *buf++ = hex[val >> 4];
135 *buf++ = hex[val & 0xf];
137 *buf = '\0';
139 return buffer;
142 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
144 int i;
145 for (i = 0; i < 20; i++) {
146 static char hex[] = "0123456789abcdef";
147 unsigned int val = sha1[i];
148 char *pos = pathbuf + i*2 + (i > 0);
149 *pos++ = hex[val >> 4];
150 *pos = hex[val & 0xf];
155 * NOTE! This returns a statically allocated buffer, so you have to be
156 * careful about using it. Do a "xstrdup()" if you need to save the
157 * filename.
159 * Also note that this returns the location for creating. Reading
160 * SHA1 file can happen from any alternate directory listed in the
161 * DB_ENVIRONMENT environment variable if it is not found in
162 * the primary object database.
164 char *sha1_file_name(const unsigned char *sha1)
166 static char *name, *base;
168 if (!base) {
169 const char *sha1_file_directory = get_object_directory();
170 int len = strlen(sha1_file_directory);
171 base = xmalloc(len + 60);
172 memcpy(base, sha1_file_directory, len);
173 memset(base+len, 0, 60);
174 base[len] = '/';
175 base[len+3] = '/';
176 name = base + len + 1;
178 fill_sha1_path(name, sha1);
179 return base;
182 char *sha1_pack_name(const unsigned char *sha1)
184 static const char hex[] = "0123456789abcdef";
185 static char *name, *base, *buf;
186 int i;
188 if (!base) {
189 const char *sha1_file_directory = get_object_directory();
190 int len = strlen(sha1_file_directory);
191 base = xmalloc(len + 60);
192 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
193 name = base + len + 11;
196 buf = name;
198 for (i = 0; i < 20; i++) {
199 unsigned int val = *sha1++;
200 *buf++ = hex[val >> 4];
201 *buf++ = hex[val & 0xf];
204 return base;
207 char *sha1_pack_index_name(const unsigned char *sha1)
209 static const char hex[] = "0123456789abcdef";
210 static char *name, *base, *buf;
211 int i;
213 if (!base) {
214 const char *sha1_file_directory = get_object_directory();
215 int len = strlen(sha1_file_directory);
216 base = xmalloc(len + 60);
217 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
218 name = base + len + 11;
221 buf = name;
223 for (i = 0; i < 20; i++) {
224 unsigned int val = *sha1++;
225 *buf++ = hex[val >> 4];
226 *buf++ = hex[val & 0xf];
229 return base;
232 struct alternate_object_database *alt_odb_list;
233 static struct alternate_object_database **alt_odb_tail;
235 static void read_info_alternates(const char * alternates, int depth);
238 * Prepare alternate object database registry.
240 * The variable alt_odb_list points at the list of struct
241 * alternate_object_database. The elements on this list come from
242 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
243 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
244 * whose contents is similar to that environment variable but can be
245 * LF separated. Its base points at a statically allocated buffer that
246 * contains "/the/directory/corresponding/to/.git/objects/...", while
247 * its name points just after the slash at the end of ".git/objects/"
248 * in the example above, and has enough space to hold 40-byte hex
249 * SHA1, an extra slash for the first level indirection, and the
250 * terminating NUL.
252 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
254 struct stat st;
255 const char *objdir = get_object_directory();
256 struct alternate_object_database *ent;
257 struct alternate_object_database *alt;
258 /* 43 = 40-byte + 2 '/' + terminating NUL */
259 int pfxlen = len;
260 int entlen = pfxlen + 43;
261 int base_len = -1;
263 if (!is_path_absolute(entry) && relative_base) {
264 /* Relative alt-odb */
265 if (base_len < 0)
266 base_len = strlen(relative_base) + 1;
267 entlen += base_len;
268 pfxlen += base_len;
270 ent = xmalloc(sizeof(*ent) + entlen);
272 if (!is_path_absolute(entry) && relative_base) {
273 memcpy(ent->base, relative_base, base_len - 1);
274 ent->base[base_len - 1] = '/';
275 memcpy(ent->base + base_len, entry, len);
277 else
278 memcpy(ent->base, entry, pfxlen);
280 ent->name = ent->base + pfxlen + 1;
281 ent->base[pfxlen + 3] = '/';
282 ent->base[pfxlen] = ent->base[entlen-1] = 0;
284 /* Detect cases where alternate disappeared */
285 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
286 error("object directory %s does not exist; "
287 "check .git/objects/info/alternates.",
288 ent->base);
289 free(ent);
290 return -1;
293 /* Prevent the common mistake of listing the same
294 * thing twice, or object directory itself.
296 for (alt = alt_odb_list; alt; alt = alt->next) {
297 if (!memcmp(ent->base, alt->base, pfxlen)) {
298 free(ent);
299 return -1;
302 if (!memcmp(ent->base, objdir, pfxlen)) {
303 free(ent);
304 return -1;
307 /* add the alternate entry */
308 *alt_odb_tail = ent;
309 alt_odb_tail = &(ent->next);
310 ent->next = NULL;
312 /* recursively add alternates */
313 read_info_alternates(ent->base, depth + 1);
315 ent->base[pfxlen] = '/';
317 return 0;
320 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
321 const char *relative_base, int depth)
323 const char *cp, *last;
325 if (depth > 5) {
326 error("%s: ignoring alternate object stores, nesting too deep.",
327 relative_base);
328 return;
331 last = alt;
332 while (last < ep) {
333 cp = last;
334 if (cp < ep && *cp == '#') {
335 while (cp < ep && *cp != sep)
336 cp++;
337 last = cp + 1;
338 continue;
340 while (cp < ep && *cp != sep)
341 cp++;
342 if (last != cp) {
343 if (!is_path_absolute(last) && depth) {
344 error("%s: ignoring relative alternate object store %s",
345 relative_base, last);
346 } else {
347 link_alt_odb_entry(last, cp - last,
348 relative_base, depth);
351 while (cp < ep && *cp == sep)
352 cp++;
353 last = cp;
357 static void read_info_alternates(const char * relative_base, int depth)
359 char *map;
360 struct stat st;
361 char path[PATH_MAX];
362 int fd;
364 sprintf(path, "%s/info/alternates", relative_base);
365 fd = open(path, O_RDONLY);
366 if (fd < 0)
367 return;
368 if (fstat(fd, &st) || (st.st_size == 0)) {
369 close(fd);
370 return;
372 map = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
373 close(fd);
375 link_alt_odb_entries(map, map + st.st_size, '\n', relative_base, depth);
377 munmap(map, st.st_size);
380 void prepare_alt_odb(void)
382 const char *alt;
384 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
385 if (!alt) alt = "";
387 if (alt_odb_tail)
388 return;
389 alt_odb_tail = &alt_odb_list;
390 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
392 read_info_alternates(get_object_directory(), 0);
395 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
397 char *name = sha1_file_name(sha1);
398 struct alternate_object_database *alt;
400 if (!stat(name, st))
401 return name;
402 prepare_alt_odb();
403 for (alt = alt_odb_list; alt; alt = alt->next) {
404 name = alt->name;
405 fill_sha1_path(name, sha1);
406 if (!stat(alt->base, st))
407 return alt->base;
409 return NULL;
412 static unsigned int pack_used_ctr;
413 static unsigned int pack_mmap_calls;
414 static unsigned int peak_pack_open_windows;
415 static unsigned int pack_open_windows;
416 static size_t peak_pack_mapped;
417 static size_t pack_mapped;
418 static size_t page_size;
419 struct packed_git *packed_git;
421 void pack_report()
423 fprintf(stderr,
424 "pack_report: getpagesize() = %10" SZ_FMT "\n"
425 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
426 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
427 page_size,
428 packed_git_window_size,
429 packed_git_limit);
430 fprintf(stderr,
431 "pack_report: pack_used_ctr = %10u\n"
432 "pack_report: pack_mmap_calls = %10u\n"
433 "pack_report: pack_open_windows = %10u / %10u\n"
434 "pack_report: pack_mapped = "
435 "%10" SZ_FMT " / %10" SZ_FMT "\n",
436 pack_used_ctr,
437 pack_mmap_calls,
438 pack_open_windows, peak_pack_open_windows,
439 pack_mapped, peak_pack_mapped);
442 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
443 void **idx_map_)
445 void *idx_map;
446 uint32_t *index;
447 unsigned long idx_size;
448 int nr, i;
449 int fd = open(path, O_RDONLY);
450 struct stat st;
451 if (fd < 0)
452 return -1;
453 if (fstat(fd, &st)) {
454 close(fd);
455 return -1;
457 idx_size = st.st_size;
458 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
459 close(fd);
461 index = idx_map;
462 *idx_map_ = idx_map;
463 *idx_size_ = idx_size;
465 /* check index map */
466 if (idx_size < 4*256 + 20 + 20)
467 return error("index file %s is too small", path);
469 /* a future index format would start with this, as older git
470 * binaries would fail the non-monotonic index check below.
471 * give a nicer warning to the user if we can.
473 if (index[0] == htonl(PACK_IDX_SIGNATURE))
474 return error("index file %s is a newer version"
475 " and is not supported by this binary"
476 " (try upgrading GIT to a newer version)",
477 path);
479 nr = 0;
480 for (i = 0; i < 256; i++) {
481 unsigned int n = ntohl(index[i]);
482 if (n < nr)
483 return error("non-monotonic index %s", path);
484 nr = n;
488 * Total size:
489 * - 256 index entries 4 bytes each
490 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
491 * - 20-byte SHA1 of the packfile
492 * - 20-byte SHA1 file checksum
494 if (idx_size != 4*256 + nr * 24 + 20 + 20)
495 return error("wrong index file size in %s", path);
497 return 0;
500 static void scan_windows(struct packed_git *p,
501 struct packed_git **lru_p,
502 struct pack_window **lru_w,
503 struct pack_window **lru_l)
505 struct pack_window *w, *w_l;
507 for (w_l = NULL, w = p->windows; w; w = w->next) {
508 if (!w->inuse_cnt) {
509 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
510 *lru_p = p;
511 *lru_w = w;
512 *lru_l = w_l;
515 w_l = w;
519 static int unuse_one_window(struct packed_git *current)
521 struct packed_git *p, *lru_p = NULL;
522 struct pack_window *lru_w = NULL, *lru_l = NULL;
524 if (current)
525 scan_windows(current, &lru_p, &lru_w, &lru_l);
526 for (p = packed_git; p; p = p->next)
527 scan_windows(p, &lru_p, &lru_w, &lru_l);
528 if (lru_p) {
529 munmap(lru_w->base, lru_w->len);
530 pack_mapped -= lru_w->len;
531 if (lru_l)
532 lru_l->next = lru_w->next;
533 else {
534 lru_p->windows = lru_w->next;
535 if (!lru_p->windows && lru_p != current) {
536 close(lru_p->pack_fd);
537 lru_p->pack_fd = -1;
540 free(lru_w);
541 pack_open_windows--;
542 return 1;
544 return 0;
547 void release_pack_memory(size_t need)
549 size_t cur = pack_mapped;
550 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
551 ; /* nothing */
554 void unuse_pack(struct pack_window **w_cursor)
556 struct pack_window *w = *w_cursor;
557 if (w) {
558 w->inuse_cnt--;
559 *w_cursor = NULL;
564 * Do not call this directly as this leaks p->pack_fd on error return;
565 * call open_packed_git() instead.
567 static int open_packed_git_1(struct packed_git *p)
569 struct stat st;
570 struct pack_header hdr;
571 unsigned char sha1[20];
572 unsigned char *idx_sha1;
573 long fd_flag;
575 p->pack_fd = open(p->pack_name, O_RDONLY);
576 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
577 return -1;
579 /* If we created the struct before we had the pack we lack size. */
580 if (!p->pack_size) {
581 if (!S_ISREG(st.st_mode))
582 return error("packfile %s not a regular file", p->pack_name);
583 p->pack_size = st.st_size;
584 } else if (p->pack_size != st.st_size)
585 return error("packfile %s size changed", p->pack_name);
587 #ifndef __MINGW32__
588 /* We leave these file descriptors open with sliding mmap;
589 * there is no point keeping them open across exec(), though.
591 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
592 if (fd_flag < 0)
593 return error("cannot determine file descriptor flags");
594 fd_flag |= FD_CLOEXEC;
595 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
596 return error("cannot set FD_CLOEXEC");
597 #endif
599 /* Verify we recognize this pack file format. */
600 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
601 return error("file %s is far too short to be a packfile", p->pack_name);
602 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
603 return error("file %s is not a GIT packfile", p->pack_name);
604 if (!pack_version_ok(hdr.hdr_version))
605 return error("packfile %s is version %u and not supported"
606 " (try upgrading GIT to a newer version)",
607 p->pack_name, ntohl(hdr.hdr_version));
609 /* Verify the pack matches its index. */
610 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
611 return error("packfile %s claims to have %u objects"
612 " while index size indicates %u objects",
613 p->pack_name, ntohl(hdr.hdr_entries),
614 num_packed_objects(p));
615 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
616 return error("end of packfile %s is unavailable", p->pack_name);
617 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
618 return error("packfile %s signature is unavailable", p->pack_name);
619 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
620 if (hashcmp(sha1, idx_sha1))
621 return error("packfile %s does not match index", p->pack_name);
622 return 0;
625 static int open_packed_git(struct packed_git *p)
627 if (!open_packed_git_1(p))
628 return 0;
629 if (p->pack_fd != -1) {
630 close(p->pack_fd);
631 p->pack_fd = -1;
633 return -1;
636 static int in_window(struct pack_window *win, unsigned long offset)
638 /* We must promise at least 20 bytes (one hash) after the
639 * offset is available from this window, otherwise the offset
640 * is not actually in this window and a different window (which
641 * has that one hash excess) must be used. This is to support
642 * the object header and delta base parsing routines below.
644 off_t win_off = win->offset;
645 return win_off <= offset
646 && (offset + 20) <= (win_off + win->len);
649 unsigned char* use_pack(struct packed_git *p,
650 struct pack_window **w_cursor,
651 unsigned long offset,
652 unsigned int *left)
654 struct pack_window *win = *w_cursor;
656 if (p->pack_fd == -1 && open_packed_git(p))
657 die("packfile %s cannot be accessed", p->pack_name);
659 /* Since packfiles end in a hash of their content and its
660 * pointless to ask for an offset into the middle of that
661 * hash, and the in_window function above wouldn't match
662 * don't allow an offset too close to the end of the file.
664 if (offset > (p->pack_size - 20))
665 die("offset beyond end of packfile (truncated pack?)");
667 if (!win || !in_window(win, offset)) {
668 if (win)
669 win->inuse_cnt--;
670 for (win = p->windows; win; win = win->next) {
671 if (in_window(win, offset))
672 break;
674 if (!win) {
675 if (!page_size)
676 page_size = getpagesize();
677 win = xcalloc(1, sizeof(*win));
678 win->offset = (offset / page_size) * page_size;
679 win->len = p->pack_size - win->offset;
680 if (win->len > packed_git_window_size)
681 win->len = packed_git_window_size;
682 pack_mapped += win->len;
683 while (packed_git_limit < pack_mapped
684 && unuse_one_window(p))
685 ; /* nothing */
686 win->base = xmmap(NULL, win->len,
687 PROT_READ, MAP_PRIVATE,
688 p->pack_fd, win->offset);
689 if (win->base == MAP_FAILED)
690 die("packfile %s cannot be mapped: %s",
691 p->pack_name,
692 strerror(errno));
693 pack_mmap_calls++;
694 pack_open_windows++;
695 if (pack_mapped > peak_pack_mapped)
696 peak_pack_mapped = pack_mapped;
697 if (pack_open_windows > peak_pack_open_windows)
698 peak_pack_open_windows = pack_open_windows;
699 win->next = p->windows;
700 p->windows = win;
703 if (win != *w_cursor) {
704 win->last_used = pack_used_ctr++;
705 win->inuse_cnt++;
706 *w_cursor = win;
708 offset -= win->offset;
709 if (left)
710 *left = win->len - offset;
711 return win->base + offset;
714 struct packed_git *add_packed_git(char *path, int path_len, int local)
716 struct stat st;
717 struct packed_git *p;
718 unsigned long idx_size;
719 void *idx_map;
720 unsigned char sha1[20];
722 if (check_packed_git_idx(path, &idx_size, &idx_map))
723 return NULL;
725 /* do we have a corresponding .pack file? */
726 strcpy(path + path_len - 4, ".pack");
727 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
728 munmap(idx_map, idx_size);
729 return NULL;
731 /* ok, it looks sane as far as we can check without
732 * actually mapping the pack file.
734 p = xmalloc(sizeof(*p) + path_len + 2);
735 strcpy(p->pack_name, path);
736 p->index_size = idx_size;
737 p->pack_size = st.st_size;
738 p->index_base = idx_map;
739 p->next = NULL;
740 p->windows = NULL;
741 p->pack_fd = -1;
742 p->pack_local = local;
743 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
744 hashcpy(p->sha1, sha1);
745 return p;
748 struct packed_git *parse_pack_index(unsigned char *sha1)
750 char *path = sha1_pack_index_name(sha1);
751 return parse_pack_index_file(sha1, path);
754 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
756 struct packed_git *p;
757 unsigned long idx_size;
758 void *idx_map;
759 char *path;
761 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
762 return NULL;
764 path = sha1_pack_name(sha1);
766 p = xmalloc(sizeof(*p) + strlen(path) + 2);
767 strcpy(p->pack_name, path);
768 p->index_size = idx_size;
769 p->pack_size = 0;
770 p->index_base = idx_map;
771 p->next = NULL;
772 p->windows = NULL;
773 p->pack_fd = -1;
774 hashcpy(p->sha1, sha1);
775 return p;
778 void install_packed_git(struct packed_git *pack)
780 pack->next = packed_git;
781 packed_git = pack;
784 static void prepare_packed_git_one(char *objdir, int local)
786 char path[PATH_MAX];
787 int len;
788 DIR *dir;
789 struct dirent *de;
791 sprintf(path, "%s/pack", objdir);
792 len = strlen(path);
793 dir = opendir(path);
794 if (!dir) {
795 if (errno != ENOENT)
796 error("unable to open object pack directory: %s: %s",
797 path, strerror(errno));
798 return;
800 path[len++] = '/';
801 while ((de = readdir(dir)) != NULL) {
802 int namelen = strlen(de->d_name);
803 struct packed_git *p;
805 if (!has_extension(de->d_name, ".idx"))
806 continue;
808 /* Don't reopen a pack we already have. */
809 strcpy(path + len, de->d_name);
810 for (p = packed_git; p; p = p->next) {
811 if (!memcmp(path, p->pack_name, len + namelen - 4))
812 break;
814 if (p)
815 continue;
816 /* See if it really is a valid .idx file with corresponding
817 * .pack file that we can map.
819 p = add_packed_git(path, len + namelen, local);
820 if (!p)
821 continue;
822 install_packed_git(p);
824 closedir(dir);
827 static int prepare_packed_git_run_once = 0;
828 void prepare_packed_git(void)
830 struct alternate_object_database *alt;
832 if (prepare_packed_git_run_once)
833 return;
834 prepare_packed_git_one(get_object_directory(), 1);
835 prepare_alt_odb();
836 for (alt = alt_odb_list; alt; alt = alt->next) {
837 alt->name[-1] = 0;
838 prepare_packed_git_one(alt->base, 0);
839 alt->name[-1] = '/';
841 prepare_packed_git_run_once = 1;
844 void reprepare_packed_git(void)
846 prepare_packed_git_run_once = 0;
847 prepare_packed_git();
850 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
852 unsigned char real_sha1[20];
853 hash_sha1_file(map, size, type, real_sha1);
854 return hashcmp(sha1, real_sha1) ? -1 : 0;
857 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
859 struct stat st;
860 void *map;
861 int fd;
862 char *filename = find_sha1_file(sha1, &st);
864 if (!filename) {
865 return NULL;
868 fd = open(filename, O_RDONLY | sha1_file_open_flag);
869 if (fd < 0) {
870 /* See if it works without O_NOATIME */
871 switch (sha1_file_open_flag) {
872 default:
873 fd = open(filename, O_RDONLY);
874 if (fd >= 0)
875 break;
876 /* Fallthrough */
877 case 0:
878 return NULL;
881 /* If it failed once, it will probably fail again.
882 * Stop using O_NOATIME
884 sha1_file_open_flag = 0;
886 map = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
887 close(fd);
888 *size = st.st_size;
889 return map;
892 int legacy_loose_object(unsigned char *map)
894 unsigned int word;
897 * Is it a zlib-compressed buffer? If so, the first byte
898 * must be 0x78 (15-bit window size, deflated), and the
899 * first 16-bit word is evenly divisible by 31
901 word = (map[0] << 8) + map[1];
902 if (map[0] == 0x78 && !(word % 31))
903 return 1;
904 else
905 return 0;
908 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
910 unsigned shift;
911 unsigned char c;
912 unsigned long size;
913 unsigned long used = 0;
915 c = buf[used++];
916 *type = (c >> 4) & 7;
917 size = c & 15;
918 shift = 4;
919 while (c & 0x80) {
920 if (len <= used)
921 return 0;
922 if (sizeof(long) * 8 <= shift)
923 return 0;
924 c = buf[used++];
925 size += (c & 0x7f) << shift;
926 shift += 7;
928 *sizep = size;
929 return used;
932 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
934 unsigned long size, used;
935 static const char valid_loose_object_type[8] = {
936 0, /* OBJ_EXT */
937 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
938 0, /* "delta" and others are invalid in a loose object */
940 enum object_type type;
942 /* Get the data stream */
943 memset(stream, 0, sizeof(*stream));
944 stream->next_in = map;
945 stream->avail_in = mapsize;
946 stream->next_out = buffer;
947 stream->avail_out = bufsiz;
949 if (legacy_loose_object(map)) {
950 inflateInit(stream);
951 return inflate(stream, 0);
954 used = unpack_object_header_gently(map, mapsize, &type, &size);
955 if (!used || !valid_loose_object_type[type])
956 return -1;
957 map += used;
958 mapsize -= used;
960 /* Set up the stream for the rest.. */
961 stream->next_in = map;
962 stream->avail_in = mapsize;
963 inflateInit(stream);
965 /* And generate the fake traditional header */
966 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
967 type_names[type], size);
968 return 0;
971 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
973 int bytes = strlen(buffer) + 1;
974 unsigned char *buf = xmalloc(1+size);
975 unsigned long n;
977 n = stream->total_out - bytes;
978 if (n > size)
979 n = size;
980 memcpy(buf, (char *) buffer + bytes, n);
981 bytes = n;
982 if (bytes < size) {
983 stream->next_out = buf + bytes;
984 stream->avail_out = size - bytes;
985 while (inflate(stream, Z_FINISH) == Z_OK)
986 /* nothing */;
988 buf[size] = 0;
989 inflateEnd(stream);
990 return buf;
994 * We used to just use "sscanf()", but that's actually way
995 * too permissive for what we want to check. So do an anal
996 * object header parse by hand.
998 static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
1000 int i;
1001 unsigned long size;
1004 * The type can be at most ten bytes (including the
1005 * terminating '\0' that we add), and is followed by
1006 * a space.
1008 i = 10;
1009 for (;;) {
1010 char c = *hdr++;
1011 if (c == ' ')
1012 break;
1013 if (!--i)
1014 return -1;
1015 *type++ = c;
1017 *type = 0;
1020 * The length must follow immediately, and be in canonical
1021 * decimal format (ie "010" is not valid).
1023 size = *hdr++ - '0';
1024 if (size > 9)
1025 return -1;
1026 if (size) {
1027 for (;;) {
1028 unsigned long c = *hdr - '0';
1029 if (c > 9)
1030 break;
1031 hdr++;
1032 size = size * 10 + c;
1035 *sizep = size;
1038 * The length must be followed by a zero byte
1040 return *hdr ? -1 : 0;
1043 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
1045 int ret;
1046 z_stream stream;
1047 char hdr[8192];
1049 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1050 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
1051 return NULL;
1053 return unpack_sha1_rest(&stream, hdr, *size);
1056 static unsigned long get_delta_base(struct packed_git *p,
1057 struct pack_window **w_curs,
1058 unsigned long offset,
1059 enum object_type kind,
1060 unsigned long delta_obj_offset,
1061 unsigned long *base_obj_offset)
1063 unsigned char *base_info = use_pack(p, w_curs, offset, NULL);
1064 unsigned long base_offset;
1066 /* use_pack() assured us we have [base_info, base_info + 20)
1067 * as a range that we can look at without walking off the
1068 * end of the mapped window. Its actually the hash size
1069 * that is assured. An OFS_DELTA longer than the hash size
1070 * is stupid, as then a REF_DELTA would be smaller to store.
1072 if (kind == OBJ_OFS_DELTA) {
1073 unsigned used = 0;
1074 unsigned char c = base_info[used++];
1075 base_offset = c & 127;
1076 while (c & 128) {
1077 base_offset += 1;
1078 if (!base_offset || base_offset & ~(~0UL >> 7))
1079 die("offset value overflow for delta base object");
1080 c = base_info[used++];
1081 base_offset = (base_offset << 7) + (c & 127);
1083 base_offset = delta_obj_offset - base_offset;
1084 if (base_offset >= delta_obj_offset)
1085 die("delta base offset out of bound");
1086 offset += used;
1087 } else if (kind == OBJ_REF_DELTA) {
1088 /* The base entry _must_ be in the same pack */
1089 base_offset = find_pack_entry_one(base_info, p);
1090 if (!base_offset)
1091 die("failed to find delta-pack base object %s",
1092 sha1_to_hex(base_info));
1093 offset += 20;
1094 } else
1095 die("I am totally screwed");
1096 *base_obj_offset = base_offset;
1097 return offset;
1100 /* forward declaration for a mutually recursive function */
1101 static int packed_object_info(struct packed_git *p, unsigned long offset,
1102 char *type, unsigned long *sizep);
1104 static int packed_delta_info(struct packed_git *p,
1105 struct pack_window **w_curs,
1106 unsigned long offset,
1107 enum object_type kind,
1108 unsigned long obj_offset,
1109 char *type,
1110 unsigned long *sizep)
1112 unsigned long base_offset;
1114 offset = get_delta_base(p, w_curs, offset, kind,
1115 obj_offset, &base_offset);
1117 /* We choose to only get the type of the base object and
1118 * ignore potentially corrupt pack file that expects the delta
1119 * based on a base with a wrong size. This saves tons of
1120 * inflate() calls.
1122 if (packed_object_info(p, base_offset, type, NULL))
1123 die("cannot get info for delta-pack base");
1125 if (sizep) {
1126 const unsigned char *data;
1127 unsigned char delta_head[20], *in;
1128 unsigned long result_size;
1129 z_stream stream;
1130 int st;
1132 memset(&stream, 0, sizeof(stream));
1133 stream.next_out = delta_head;
1134 stream.avail_out = sizeof(delta_head);
1136 inflateInit(&stream);
1137 do {
1138 in = use_pack(p, w_curs, offset, &stream.avail_in);
1139 stream.next_in = in;
1140 st = inflate(&stream, Z_FINISH);
1141 offset += stream.next_in - in;
1142 } while ((st == Z_OK || st == Z_BUF_ERROR)
1143 && stream.total_out < sizeof(delta_head));
1144 inflateEnd(&stream);
1145 if ((st != Z_STREAM_END) &&
1146 stream.total_out != sizeof(delta_head))
1147 die("delta data unpack-initial failed");
1149 /* Examine the initial part of the delta to figure out
1150 * the result size.
1152 data = delta_head;
1154 /* ignore base size */
1155 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1157 /* Read the result size */
1158 result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1159 *sizep = result_size;
1161 return 0;
1164 static unsigned long unpack_object_header(struct packed_git *p,
1165 struct pack_window **w_curs,
1166 unsigned long offset,
1167 enum object_type *type,
1168 unsigned long *sizep)
1170 unsigned char *base;
1171 unsigned int left;
1172 unsigned long used;
1174 /* use_pack() assures us we have [base, base + 20) available
1175 * as a range that we can look at at. (Its actually the hash
1176 * size that is assured.) With our object header encoding
1177 * the maximum deflated object size is 2^137, which is just
1178 * insane, so we know won't exceed what we have been given.
1180 base = use_pack(p, w_curs, offset, &left);
1181 used = unpack_object_header_gently(base, left, type, sizep);
1182 if (!used)
1183 die("object offset outside of pack file");
1185 return offset + used;
1188 void packed_object_info_detail(struct packed_git *p,
1189 unsigned long offset,
1190 char *type,
1191 unsigned long *size,
1192 unsigned long *store_size,
1193 unsigned int *delta_chain_length,
1194 unsigned char *base_sha1)
1196 struct pack_window *w_curs = NULL;
1197 unsigned long obj_offset, val;
1198 unsigned char *next_sha1;
1199 enum object_type kind;
1201 *delta_chain_length = 0;
1202 obj_offset = offset;
1203 offset = unpack_object_header(p, &w_curs, offset, &kind, size);
1205 for (;;) {
1206 switch (kind) {
1207 default:
1208 die("pack %s contains unknown object type %d",
1209 p->pack_name, kind);
1210 case OBJ_COMMIT:
1211 case OBJ_TREE:
1212 case OBJ_BLOB:
1213 case OBJ_TAG:
1214 strcpy(type, type_names[kind]);
1215 *store_size = 0; /* notyet */
1216 unuse_pack(&w_curs);
1217 return;
1218 case OBJ_OFS_DELTA:
1219 get_delta_base(p, &w_curs, offset, kind,
1220 obj_offset, &offset);
1221 if (*delta_chain_length == 0) {
1222 /* TODO: find base_sha1 as pointed by offset */
1224 break;
1225 case OBJ_REF_DELTA:
1226 next_sha1 = use_pack(p, &w_curs, offset, NULL);
1227 if (*delta_chain_length == 0)
1228 hashcpy(base_sha1, next_sha1);
1229 offset = find_pack_entry_one(next_sha1, p);
1230 break;
1232 obj_offset = offset;
1233 offset = unpack_object_header(p, &w_curs, offset, &kind, &val);
1234 (*delta_chain_length)++;
1238 static int packed_object_info(struct packed_git *p, unsigned long offset,
1239 char *type, unsigned long *sizep)
1241 struct pack_window *w_curs = NULL;
1242 unsigned long size, obj_offset = offset;
1243 enum object_type kind;
1244 int r;
1246 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1248 switch (kind) {
1249 case OBJ_OFS_DELTA:
1250 case OBJ_REF_DELTA:
1251 r = packed_delta_info(p, &w_curs, offset, kind,
1252 obj_offset, type, sizep);
1253 unuse_pack(&w_curs);
1254 return r;
1255 case OBJ_COMMIT:
1256 case OBJ_TREE:
1257 case OBJ_BLOB:
1258 case OBJ_TAG:
1259 strcpy(type, type_names[kind]);
1260 unuse_pack(&w_curs);
1261 break;
1262 default:
1263 die("pack %s contains unknown object type %d",
1264 p->pack_name, kind);
1266 if (sizep)
1267 *sizep = size;
1268 return 0;
1271 static void *unpack_compressed_entry(struct packed_git *p,
1272 struct pack_window **w_curs,
1273 unsigned long offset,
1274 unsigned long size)
1276 int st;
1277 z_stream stream;
1278 unsigned char *buffer, *in;
1280 buffer = xmalloc(size + 1);
1281 buffer[size] = 0;
1282 memset(&stream, 0, sizeof(stream));
1283 stream.next_out = buffer;
1284 stream.avail_out = size;
1286 inflateInit(&stream);
1287 do {
1288 in = use_pack(p, w_curs, offset, &stream.avail_in);
1289 stream.next_in = in;
1290 st = inflate(&stream, Z_FINISH);
1291 offset += stream.next_in - in;
1292 } while (st == Z_OK || st == Z_BUF_ERROR);
1293 inflateEnd(&stream);
1294 if ((st != Z_STREAM_END) || stream.total_out != size) {
1295 free(buffer);
1296 return NULL;
1299 return buffer;
1302 static void *unpack_delta_entry(struct packed_git *p,
1303 struct pack_window **w_curs,
1304 unsigned long offset,
1305 unsigned long delta_size,
1306 enum object_type kind,
1307 unsigned long obj_offset,
1308 char *type,
1309 unsigned long *sizep)
1311 void *delta_data, *result, *base;
1312 unsigned long result_size, base_size, base_offset;
1314 offset = get_delta_base(p, w_curs, offset, kind,
1315 obj_offset, &base_offset);
1316 base = unpack_entry(p, base_offset, type, &base_size);
1317 if (!base)
1318 die("failed to read delta base object at %lu from %s",
1319 base_offset, p->pack_name);
1321 delta_data = unpack_compressed_entry(p, w_curs, offset, delta_size);
1322 result = patch_delta(base, base_size,
1323 delta_data, delta_size,
1324 &result_size);
1325 if (!result)
1326 die("failed to apply delta");
1327 free(delta_data);
1328 free(base);
1329 *sizep = result_size;
1330 return result;
1333 void *unpack_entry(struct packed_git *p, unsigned long offset,
1334 char *type, unsigned long *sizep)
1336 struct pack_window *w_curs = NULL;
1337 unsigned long size, obj_offset = offset;
1338 enum object_type kind;
1339 void *retval;
1341 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1342 switch (kind) {
1343 case OBJ_OFS_DELTA:
1344 case OBJ_REF_DELTA:
1345 retval = unpack_delta_entry(p, &w_curs, offset, size,
1346 kind, obj_offset, type, sizep);
1347 break;
1348 case OBJ_COMMIT:
1349 case OBJ_TREE:
1350 case OBJ_BLOB:
1351 case OBJ_TAG:
1352 strcpy(type, type_names[kind]);
1353 *sizep = size;
1354 retval = unpack_compressed_entry(p, &w_curs, offset, size);
1355 break;
1356 default:
1357 die("unknown object type %i in %s", kind, p->pack_name);
1359 unuse_pack(&w_curs);
1360 return retval;
1363 int num_packed_objects(const struct packed_git *p)
1365 /* See check_packed_git_idx() */
1366 return (p->index_size - 20 - 20 - 4*256) / 24;
1369 int nth_packed_object_sha1(const struct packed_git *p, int n,
1370 unsigned char* sha1)
1372 void *index = p->index_base + 256;
1373 if (n < 0 || num_packed_objects(p) <= n)
1374 return -1;
1375 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1376 return 0;
1379 unsigned long find_pack_entry_one(const unsigned char *sha1,
1380 struct packed_git *p)
1382 uint32_t *level1_ofs = p->index_base;
1383 int hi = ntohl(level1_ofs[*sha1]);
1384 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1385 void *index = p->index_base + 256;
1387 do {
1388 int mi = (lo + hi) / 2;
1389 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1390 if (!cmp)
1391 return ntohl(*((uint32_t *)((char *)index + (24 * mi))));
1392 if (cmp > 0)
1393 hi = mi;
1394 else
1395 lo = mi+1;
1396 } while (lo < hi);
1397 return 0;
1400 static int matches_pack_name(struct packed_git *p, const char *ig)
1402 const char *last_c, *c;
1404 if (!strcmp(p->pack_name, ig))
1405 return 0;
1407 for (c = p->pack_name, last_c = c; *c;)
1408 if (*c == '/')
1409 last_c = ++c;
1410 else
1411 ++c;
1412 if (!strcmp(last_c, ig))
1413 return 0;
1415 return 1;
1418 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1420 struct packed_git *p;
1421 unsigned long offset;
1423 prepare_packed_git();
1425 for (p = packed_git; p; p = p->next) {
1426 if (ignore_packed) {
1427 const char **ig;
1428 for (ig = ignore_packed; *ig; ig++)
1429 if (!matches_pack_name(p, *ig))
1430 break;
1431 if (*ig)
1432 continue;
1434 offset = find_pack_entry_one(sha1, p);
1435 if (offset) {
1437 * We are about to tell the caller where they can
1438 * locate the requested object. We better make
1439 * sure the packfile is still here and can be
1440 * accessed before supplying that answer, as
1441 * it may have been deleted since the index
1442 * was loaded!
1444 if (p->pack_fd == -1 && open_packed_git(p)) {
1445 error("packfile %s cannot be accessed", p->pack_name);
1446 continue;
1448 e->offset = offset;
1449 e->p = p;
1450 hashcpy(e->sha1, sha1);
1451 return 1;
1454 return 0;
1457 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1458 struct packed_git *packs)
1460 struct packed_git *p;
1462 for (p = packs; p; p = p->next) {
1463 if (find_pack_entry_one(sha1, p))
1464 return p;
1466 return NULL;
1470 static int sha1_loose_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1472 int status;
1473 unsigned long mapsize, size;
1474 void *map;
1475 z_stream stream;
1476 char hdr[128];
1478 map = map_sha1_file(sha1, &mapsize);
1479 if (!map)
1480 return error("unable to find %s", sha1_to_hex(sha1));
1481 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1482 status = error("unable to unpack %s header",
1483 sha1_to_hex(sha1));
1484 if (parse_sha1_header(hdr, type, &size) < 0)
1485 status = error("unable to parse %s header", sha1_to_hex(sha1));
1486 else {
1487 status = 0;
1488 if (sizep)
1489 *sizep = size;
1491 inflateEnd(&stream);
1492 munmap(map, mapsize);
1493 return status;
1496 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1498 struct pack_entry e;
1500 if (!find_pack_entry(sha1, &e, NULL)) {
1501 reprepare_packed_git();
1502 if (!find_pack_entry(sha1, &e, NULL))
1503 return sha1_loose_object_info(sha1, type, sizep);
1505 return packed_object_info(e.p, e.offset, type, sizep);
1508 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1510 struct pack_entry e;
1512 if (!find_pack_entry(sha1, &e, NULL))
1513 return NULL;
1514 else
1515 return unpack_entry(e.p, e.offset, type, size);
1519 * This is meant to hold a *small* number of objects that you would
1520 * want read_sha1_file() to be able to return, but yet you do not want
1521 * to write them into the object store (e.g. a browse-only
1522 * application).
1524 static struct cached_object {
1525 unsigned char sha1[20];
1526 const char *type;
1527 void *buf;
1528 unsigned long size;
1529 } *cached_objects;
1530 static int cached_object_nr, cached_object_alloc;
1532 static struct cached_object *find_cached_object(const unsigned char *sha1)
1534 int i;
1535 struct cached_object *co = cached_objects;
1537 for (i = 0; i < cached_object_nr; i++, co++) {
1538 if (!hashcmp(co->sha1, sha1))
1539 return co;
1541 return NULL;
1544 int pretend_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *sha1)
1546 struct cached_object *co;
1548 hash_sha1_file(buf, len, type, sha1);
1549 if (has_sha1_file(sha1) || find_cached_object(sha1))
1550 return 0;
1551 if (cached_object_alloc <= cached_object_nr) {
1552 cached_object_alloc = alloc_nr(cached_object_alloc);
1553 cached_objects = xrealloc(cached_objects,
1554 sizeof(*cached_objects) *
1555 cached_object_alloc);
1557 co = &cached_objects[cached_object_nr++];
1558 co->size = len;
1559 co->type = strdup(type);
1560 hashcpy(co->sha1, sha1);
1561 return 0;
1564 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1566 unsigned long mapsize;
1567 void *map, *buf;
1568 struct cached_object *co;
1570 co = find_cached_object(sha1);
1571 if (co) {
1572 buf = xmalloc(co->size + 1);
1573 memcpy(buf, co->buf, co->size);
1574 ((char*)buf)[co->size] = 0;
1575 strcpy(type, co->type);
1576 *size = co->size;
1577 return buf;
1580 buf = read_packed_sha1(sha1, type, size);
1581 if (buf)
1582 return buf;
1583 map = map_sha1_file(sha1, &mapsize);
1584 if (map) {
1585 buf = unpack_sha1_file(map, mapsize, type, size);
1586 munmap(map, mapsize);
1587 return buf;
1589 reprepare_packed_git();
1590 return read_packed_sha1(sha1, type, size);
1593 void *read_object_with_reference(const unsigned char *sha1,
1594 const char *required_type,
1595 unsigned long *size,
1596 unsigned char *actual_sha1_return)
1598 char type[20];
1599 void *buffer;
1600 unsigned long isize;
1601 unsigned char actual_sha1[20];
1603 hashcpy(actual_sha1, sha1);
1604 while (1) {
1605 int ref_length = -1;
1606 const char *ref_type = NULL;
1608 buffer = read_sha1_file(actual_sha1, type, &isize);
1609 if (!buffer)
1610 return NULL;
1611 if (!strcmp(type, required_type)) {
1612 *size = isize;
1613 if (actual_sha1_return)
1614 hashcpy(actual_sha1_return, actual_sha1);
1615 return buffer;
1617 /* Handle references */
1618 else if (!strcmp(type, commit_type))
1619 ref_type = "tree ";
1620 else if (!strcmp(type, tag_type))
1621 ref_type = "object ";
1622 else {
1623 free(buffer);
1624 return NULL;
1626 ref_length = strlen(ref_type);
1628 if (memcmp(buffer, ref_type, ref_length) ||
1629 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1630 free(buffer);
1631 return NULL;
1633 free(buffer);
1634 /* Now we have the ID of the referred-to object in
1635 * actual_sha1. Check again. */
1639 static void write_sha1_file_prepare(void *buf, unsigned long len,
1640 const char *type, unsigned char *sha1,
1641 unsigned char *hdr, int *hdrlen)
1643 SHA_CTX c;
1645 /* Generate the header */
1646 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1648 /* Sha1.. */
1649 SHA1_Init(&c);
1650 SHA1_Update(&c, hdr, *hdrlen);
1651 SHA1_Update(&c, buf, len);
1652 SHA1_Final(sha1, &c);
1656 * Link the tempfile to the final place, possibly creating the
1657 * last directory level as you do so.
1659 * Returns the errno on failure, 0 on success.
1661 static int link_temp_to_file(const char *tmpfile, const char *filename)
1663 int ret;
1664 char *dir;
1666 if (!link(tmpfile, filename))
1667 return 0;
1670 * Try to mkdir the last path component if that failed.
1672 * Re-try the "link()" regardless of whether the mkdir
1673 * succeeds, since a race might mean that somebody
1674 * else succeeded.
1676 ret = errno;
1677 dir = strrchr(filename, '/');
1678 if (dir) {
1679 *dir = 0;
1680 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1681 *dir = '/';
1682 return -2;
1684 *dir = '/';
1685 if (!link(tmpfile, filename))
1686 return 0;
1687 ret = errno;
1689 return ret;
1693 * Move the just written object into its final resting place
1695 int move_temp_to_file(const char *tmpfile, const char *filename)
1697 int ret = link_temp_to_file(tmpfile, filename);
1700 * Coda hack - coda doesn't like cross-directory links,
1701 * so we fall back to a rename, which will mean that it
1702 * won't be able to check collisions, but that's not a
1703 * big deal.
1705 * The same holds for FAT formatted media.
1707 * When this succeeds, we just return 0. We have nothing
1708 * left to unlink.
1710 if (ret && ret != EEXIST) {
1711 if (!rename(tmpfile, filename))
1712 return 0;
1713 ret = errno;
1715 unlink(tmpfile);
1716 if (ret) {
1717 if (ret != EEXIST) {
1718 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1720 /* FIXME!!! Collision check here ? */
1723 return 0;
1726 static int write_buffer(int fd, const void *buf, size_t len)
1728 if (write_in_full(fd, buf, len) < 0)
1729 return error("file write error (%s)", strerror(errno));
1730 return 0;
1733 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1735 int hdr_len;
1736 unsigned char c;
1738 c = (type << 4) | (len & 15);
1739 len >>= 4;
1740 hdr_len = 1;
1741 while (len) {
1742 *hdr++ = c | 0x80;
1743 hdr_len++;
1744 c = (len & 0x7f);
1745 len >>= 7;
1747 *hdr = c;
1748 return hdr_len;
1751 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1753 int obj_type, hdr;
1755 if (use_legacy_headers) {
1756 while (deflate(stream, 0) == Z_OK)
1757 /* nothing */;
1758 return;
1760 if (!strcmp(type, blob_type))
1761 obj_type = OBJ_BLOB;
1762 else if (!strcmp(type, tree_type))
1763 obj_type = OBJ_TREE;
1764 else if (!strcmp(type, commit_type))
1765 obj_type = OBJ_COMMIT;
1766 else if (!strcmp(type, tag_type))
1767 obj_type = OBJ_TAG;
1768 else
1769 die("trying to generate bogus object of type '%s'", type);
1770 hdr = write_binary_header(stream->next_out, obj_type, len);
1771 stream->total_out = hdr;
1772 stream->next_out += hdr;
1773 stream->avail_out -= hdr;
1776 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1777 unsigned char *sha1)
1779 unsigned char hdr[50];
1780 int hdrlen;
1781 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1782 return 0;
1785 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1787 int size;
1788 unsigned char *compressed;
1789 z_stream stream;
1790 unsigned char sha1[20];
1791 char *filename;
1792 static char tmpfile[PATH_MAX];
1793 unsigned char hdr[50];
1794 int fd, hdrlen;
1796 /* Normally if we have it in the pack then we do not bother writing
1797 * it out into .git/objects/??/?{38} file.
1799 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1800 filename = sha1_file_name(sha1);
1801 if (returnsha1)
1802 hashcpy(returnsha1, sha1);
1803 if (has_sha1_file(sha1))
1804 return 0;
1805 fd = open(filename, O_RDONLY);
1806 if (fd >= 0) {
1808 * FIXME!!! We might do collision checking here, but we'd
1809 * need to uncompress the old file and check it. Later.
1811 close(fd);
1812 return 0;
1815 if (errno != ENOENT) {
1816 return error("sha1 file %s: %s\n", filename, strerror(errno));
1819 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1821 fd = mkstemp(tmpfile);
1822 if (fd < 0) {
1823 if (errno == EPERM)
1824 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1825 else
1826 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1829 /* Set it up */
1830 memset(&stream, 0, sizeof(stream));
1831 deflateInit(&stream, zlib_compression_level);
1832 size = 8 + deflateBound(&stream, len+hdrlen);
1833 compressed = xmalloc(size);
1835 /* Compress it */
1836 stream.next_out = compressed;
1837 stream.avail_out = size;
1839 /* First header.. */
1840 stream.next_in = hdr;
1841 stream.avail_in = hdrlen;
1842 setup_object_header(&stream, type, len);
1844 /* Then the data itself.. */
1845 stream.next_in = buf;
1846 stream.avail_in = len;
1847 while (deflate(&stream, Z_FINISH) == Z_OK)
1848 /* nothing */;
1849 deflateEnd(&stream);
1850 size = stream.total_out;
1852 if (write_buffer(fd, compressed, size) < 0)
1853 die("unable to write sha1 file");
1854 fchmod(fd, 0444);
1855 close(fd);
1856 free(compressed);
1858 return move_temp_to_file(tmpfile, filename);
1862 * We need to unpack and recompress the object for writing
1863 * it out to a different file.
1865 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1867 size_t size;
1868 z_stream stream;
1869 unsigned char *unpacked;
1870 unsigned long len;
1871 char type[20];
1872 char hdr[50];
1873 int hdrlen;
1874 void *buf;
1876 /* need to unpack and recompress it by itself */
1877 unpacked = read_packed_sha1(sha1, type, &len);
1878 if (!unpacked)
1879 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1881 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1883 /* Set it up */
1884 memset(&stream, 0, sizeof(stream));
1885 deflateInit(&stream, zlib_compression_level);
1886 size = deflateBound(&stream, len + hdrlen);
1887 buf = xmalloc(size);
1889 /* Compress it */
1890 stream.next_out = buf;
1891 stream.avail_out = size;
1893 /* First header.. */
1894 stream.next_in = (void *)hdr;
1895 stream.avail_in = hdrlen;
1896 while (deflate(&stream, 0) == Z_OK)
1897 /* nothing */;
1899 /* Then the data itself.. */
1900 stream.next_in = unpacked;
1901 stream.avail_in = len;
1902 while (deflate(&stream, Z_FINISH) == Z_OK)
1903 /* nothing */;
1904 deflateEnd(&stream);
1905 free(unpacked);
1907 *objsize = stream.total_out;
1908 return buf;
1911 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1913 int retval;
1914 unsigned long objsize;
1915 void *buf = map_sha1_file(sha1, &objsize);
1917 if (buf) {
1918 retval = write_buffer(fd, buf, objsize);
1919 munmap(buf, objsize);
1920 return retval;
1923 buf = repack_object(sha1, &objsize);
1924 retval = write_buffer(fd, buf, objsize);
1925 free(buf);
1926 return retval;
1929 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1930 size_t bufsize, size_t *bufposn)
1932 char tmpfile[PATH_MAX];
1933 int local;
1934 z_stream stream;
1935 unsigned char real_sha1[20];
1936 unsigned char discard[4096];
1937 int ret;
1938 SHA_CTX c;
1940 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1942 local = mkstemp(tmpfile);
1943 if (local < 0) {
1944 if (errno == EPERM)
1945 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1946 else
1947 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1950 memset(&stream, 0, sizeof(stream));
1952 inflateInit(&stream);
1954 SHA1_Init(&c);
1956 do {
1957 ssize_t size;
1958 if (*bufposn) {
1959 stream.avail_in = *bufposn;
1960 stream.next_in = (unsigned char *) buffer;
1961 do {
1962 stream.next_out = discard;
1963 stream.avail_out = sizeof(discard);
1964 ret = inflate(&stream, Z_SYNC_FLUSH);
1965 SHA1_Update(&c, discard, sizeof(discard) -
1966 stream.avail_out);
1967 } while (stream.avail_in && ret == Z_OK);
1968 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1969 die("unable to write sha1 file");
1970 memmove(buffer, buffer + *bufposn - stream.avail_in,
1971 stream.avail_in);
1972 *bufposn = stream.avail_in;
1973 if (ret != Z_OK)
1974 break;
1976 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
1977 if (size <= 0) {
1978 close(local);
1979 unlink(tmpfile);
1980 if (!size)
1981 return error("Connection closed?");
1982 perror("Reading from connection");
1983 return -1;
1985 *bufposn += size;
1986 } while (1);
1987 inflateEnd(&stream);
1989 close(local);
1990 SHA1_Final(real_sha1, &c);
1991 if (ret != Z_STREAM_END) {
1992 unlink(tmpfile);
1993 return error("File %s corrupted", sha1_to_hex(sha1));
1995 if (hashcmp(sha1, real_sha1)) {
1996 unlink(tmpfile);
1997 return error("File %s has bad hash", sha1_to_hex(sha1));
2000 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
2003 int has_pack_index(const unsigned char *sha1)
2005 struct stat st;
2006 if (stat(sha1_pack_index_name(sha1), &st))
2007 return 0;
2008 return 1;
2011 int has_pack_file(const unsigned char *sha1)
2013 struct stat st;
2014 if (stat(sha1_pack_name(sha1), &st))
2015 return 0;
2016 return 1;
2019 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2021 struct pack_entry e;
2022 return find_pack_entry(sha1, &e, ignore_packed);
2025 int has_sha1_file(const unsigned char *sha1)
2027 struct stat st;
2028 struct pack_entry e;
2030 if (find_pack_entry(sha1, &e, NULL))
2031 return 1;
2032 return find_sha1_file(sha1, &st) ? 1 : 0;
2036 * reads from fd as long as possible into a supplied buffer of size bytes.
2037 * If necessary the buffer's size is increased using realloc()
2039 * returns 0 if anything went fine and -1 otherwise
2041 * NOTE: both buf and size may change, but even when -1 is returned
2042 * you still have to free() it yourself.
2044 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
2046 char* buf = *return_buf;
2047 unsigned long size = *return_size;
2048 int iret;
2049 unsigned long off = 0;
2051 do {
2052 iret = xread(fd, buf + off, size - off);
2053 if (iret > 0) {
2054 off += iret;
2055 if (off == size) {
2056 size *= 2;
2057 buf = xrealloc(buf, size);
2060 } while (iret > 0);
2062 *return_buf = buf;
2063 *return_size = off;
2065 if (iret < 0)
2066 return -1;
2067 return 0;
2070 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2072 unsigned long size = 4096;
2073 char *buf = xmalloc(size);
2074 int ret;
2076 if (read_pipe(fd, &buf, &size)) {
2077 free(buf);
2078 return -1;
2081 if (!type)
2082 type = blob_type;
2083 if (write_object)
2084 ret = write_sha1_file(buf, size, type, sha1);
2085 else
2086 ret = hash_sha1_file(buf, size, type, sha1);
2087 free(buf);
2088 return ret;
2091 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
2093 unsigned long size = st->st_size;
2094 void *buf;
2095 int ret;
2097 buf = "";
2098 if (size)
2099 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2100 close(fd);
2102 if (!type)
2103 type = blob_type;
2104 /* FIXME: CRLF -> LF conversion here for blobs! We'll need the path! */
2105 if (write_object)
2106 ret = write_sha1_file(buf, size, type, sha1);
2107 else
2108 ret = hash_sha1_file(buf, size, type, sha1);
2109 if (size)
2110 munmap(buf, size);
2111 return ret;
2114 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2116 int fd;
2117 char *target;
2119 switch (st->st_mode & S_IFMT) {
2120 case S_IFREG:
2121 fd = open(path, O_RDONLY);
2122 if (fd < 0)
2123 return error("open(\"%s\"): %s", path,
2124 strerror(errno));
2125 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
2126 return error("%s: failed to insert into database",
2127 path);
2128 break;
2129 case S_IFLNK:
2130 target = xmalloc(st->st_size+1);
2131 if (readlink(path, target, st->st_size+1) != st->st_size) {
2132 char *errstr = strerror(errno);
2133 free(target);
2134 return error("readlink(\"%s\"): %s", path,
2135 errstr);
2137 if (!write_object)
2138 hash_sha1_file(target, st->st_size, blob_type, sha1);
2139 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
2140 return error("%s: failed to insert into database",
2141 path);
2142 free(target);
2143 break;
2144 default:
2145 return error("%s: unsupported file type", path);
2147 return 0;
2150 int read_pack_header(int fd, struct pack_header *header)
2152 char *c = (char*)header;
2153 ssize_t remaining = sizeof(struct pack_header);
2154 do {
2155 ssize_t r = xread(fd, c, remaining);
2156 if (r <= 0)
2157 /* "eof before pack header was fully read" */
2158 return PH_ERROR_EOF;
2159 remaining -= r;
2160 c += r;
2161 } while (remaining > 0);
2162 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2163 /* "protocol error (pack signature mismatch detected)" */
2164 return PH_ERROR_PACK_SIGNATURE;
2165 if (!pack_version_ok(header->hdr_version))
2166 /* "protocol error (pack version unsupported)" */
2167 return PH_ERROR_PROTOCOL;
2168 return 0;