Merge with git://repo.or.cz/git.git#next.
[git/mingw.git] / sha1_file.c
blob1abe323eca9a958cc377c1285e7fb2b6cc4a59a0
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;
563 static void open_packed_git(struct packed_git *p)
565 struct stat st;
566 struct pack_header hdr;
567 unsigned char sha1[20];
568 unsigned char *idx_sha1;
569 long fd_flag;
571 p->pack_fd = open(p->pack_name, O_RDONLY);
572 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
573 die("packfile %s cannot be opened", p->pack_name);
575 /* If we created the struct before we had the pack we lack size. */
576 if (!p->pack_size) {
577 if (!S_ISREG(st.st_mode))
578 die("packfile %s not a regular file", p->pack_name);
579 p->pack_size = st.st_size;
580 } else if (p->pack_size != st.st_size)
581 die("packfile %s size changed", p->pack_name);
583 #ifndef __MINGW32__
584 /* We leave these file descriptors open with sliding mmap;
585 * there is no point keeping them open across exec(), though.
587 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
588 if (fd_flag < 0)
589 die("cannot determine file descriptor flags");
590 fd_flag |= FD_CLOEXEC;
591 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
592 die("cannot set FD_CLOEXEC");
593 #endif
595 /* Verify we recognize this pack file format. */
596 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
597 die("file %s is far too short to be a packfile", p->pack_name);
598 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
599 die("file %s is not a GIT packfile", p->pack_name);
600 if (!pack_version_ok(hdr.hdr_version))
601 die("packfile %s is version %u and not supported"
602 " (try upgrading GIT to a newer version)",
603 p->pack_name, ntohl(hdr.hdr_version));
605 /* Verify the pack matches its index. */
606 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
607 die("packfile %s claims to have %u objects"
608 " while index size indicates %u objects",
609 p->pack_name, ntohl(hdr.hdr_entries),
610 num_packed_objects(p));
611 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
612 die("end of packfile %s is unavailable", p->pack_name);
613 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
614 die("packfile %s signature is unavailable", p->pack_name);
615 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
616 if (hashcmp(sha1, idx_sha1))
617 die("packfile %s does not match index", p->pack_name);
620 static int in_window(struct pack_window *win, unsigned long offset)
622 /* We must promise at least 20 bytes (one hash) after the
623 * offset is available from this window, otherwise the offset
624 * is not actually in this window and a different window (which
625 * has that one hash excess) must be used. This is to support
626 * the object header and delta base parsing routines below.
628 off_t win_off = win->offset;
629 return win_off <= offset
630 && (offset + 20) <= (win_off + win->len);
633 unsigned char* use_pack(struct packed_git *p,
634 struct pack_window **w_cursor,
635 unsigned long offset,
636 unsigned int *left)
638 struct pack_window *win = *w_cursor;
640 if (p->pack_fd == -1)
641 open_packed_git(p);
643 /* Since packfiles end in a hash of their content and its
644 * pointless to ask for an offset into the middle of that
645 * hash, and the in_window function above wouldn't match
646 * don't allow an offset too close to the end of the file.
648 if (offset > (p->pack_size - 20))
649 die("offset beyond end of packfile (truncated pack?)");
651 if (!win || !in_window(win, offset)) {
652 if (win)
653 win->inuse_cnt--;
654 for (win = p->windows; win; win = win->next) {
655 if (in_window(win, offset))
656 break;
658 if (!win) {
659 if (!page_size)
660 page_size = getpagesize();
661 win = xcalloc(1, sizeof(*win));
662 win->offset = (offset / page_size) * page_size;
663 win->len = p->pack_size - win->offset;
664 if (win->len > packed_git_window_size)
665 win->len = packed_git_window_size;
666 pack_mapped += win->len;
667 while (packed_git_limit < pack_mapped
668 && unuse_one_window(p))
669 ; /* nothing */
670 win->base = xmmap(NULL, win->len,
671 PROT_READ, MAP_PRIVATE,
672 p->pack_fd, win->offset);
673 if (win->base == MAP_FAILED)
674 die("packfile %s cannot be mapped: %s",
675 p->pack_name,
676 strerror(errno));
677 pack_mmap_calls++;
678 pack_open_windows++;
679 if (pack_mapped > peak_pack_mapped)
680 peak_pack_mapped = pack_mapped;
681 if (pack_open_windows > peak_pack_open_windows)
682 peak_pack_open_windows = pack_open_windows;
683 win->next = p->windows;
684 p->windows = win;
687 if (win != *w_cursor) {
688 win->last_used = pack_used_ctr++;
689 win->inuse_cnt++;
690 *w_cursor = win;
692 offset -= win->offset;
693 if (left)
694 *left = win->len - offset;
695 return win->base + offset;
698 struct packed_git *add_packed_git(char *path, int path_len, int local)
700 struct stat st;
701 struct packed_git *p;
702 unsigned long idx_size;
703 void *idx_map;
704 unsigned char sha1[20];
706 if (check_packed_git_idx(path, &idx_size, &idx_map))
707 return NULL;
709 /* do we have a corresponding .pack file? */
710 strcpy(path + path_len - 4, ".pack");
711 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
712 munmap(idx_map, idx_size);
713 return NULL;
715 /* ok, it looks sane as far as we can check without
716 * actually mapping the pack file.
718 p = xmalloc(sizeof(*p) + path_len + 2);
719 strcpy(p->pack_name, path);
720 p->index_size = idx_size;
721 p->pack_size = st.st_size;
722 p->index_base = idx_map;
723 p->next = NULL;
724 p->windows = NULL;
725 p->pack_fd = -1;
726 p->pack_local = local;
727 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
728 hashcpy(p->sha1, sha1);
729 return p;
732 struct packed_git *parse_pack_index(unsigned char *sha1)
734 char *path = sha1_pack_index_name(sha1);
735 return parse_pack_index_file(sha1, path);
738 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
740 struct packed_git *p;
741 unsigned long idx_size;
742 void *idx_map;
743 char *path;
745 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
746 return NULL;
748 path = sha1_pack_name(sha1);
750 p = xmalloc(sizeof(*p) + strlen(path) + 2);
751 strcpy(p->pack_name, path);
752 p->index_size = idx_size;
753 p->pack_size = 0;
754 p->index_base = idx_map;
755 p->next = NULL;
756 p->windows = NULL;
757 p->pack_fd = -1;
758 hashcpy(p->sha1, sha1);
759 return p;
762 void install_packed_git(struct packed_git *pack)
764 pack->next = packed_git;
765 packed_git = pack;
768 static void prepare_packed_git_one(char *objdir, int local)
770 char path[PATH_MAX];
771 int len;
772 DIR *dir;
773 struct dirent *de;
775 sprintf(path, "%s/pack", objdir);
776 len = strlen(path);
777 dir = opendir(path);
778 if (!dir) {
779 if (errno != ENOENT)
780 error("unable to open object pack directory: %s: %s",
781 path, strerror(errno));
782 return;
784 path[len++] = '/';
785 while ((de = readdir(dir)) != NULL) {
786 int namelen = strlen(de->d_name);
787 struct packed_git *p;
789 if (!has_extension(de->d_name, ".idx"))
790 continue;
792 /* we have .idx. Is it a file we can map? */
793 strcpy(path + len, de->d_name);
794 for (p = packed_git; p; p = p->next) {
795 if (!memcmp(path, p->pack_name, len + namelen - 4))
796 break;
798 if (p)
799 continue;
800 p = add_packed_git(path, len + namelen, local);
801 if (!p)
802 continue;
803 p->next = packed_git;
804 packed_git = p;
806 closedir(dir);
809 static int prepare_packed_git_run_once = 0;
810 void prepare_packed_git(void)
812 struct alternate_object_database *alt;
814 if (prepare_packed_git_run_once)
815 return;
816 prepare_packed_git_one(get_object_directory(), 1);
817 prepare_alt_odb();
818 for (alt = alt_odb_list; alt; alt = alt->next) {
819 alt->name[-1] = 0;
820 prepare_packed_git_one(alt->base, 0);
821 alt->name[-1] = '/';
823 prepare_packed_git_run_once = 1;
826 void reprepare_packed_git(void)
828 prepare_packed_git_run_once = 0;
829 prepare_packed_git();
832 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
834 unsigned char real_sha1[20];
835 hash_sha1_file(map, size, type, real_sha1);
836 return hashcmp(sha1, real_sha1) ? -1 : 0;
839 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
841 struct stat st;
842 void *map;
843 int fd;
844 char *filename = find_sha1_file(sha1, &st);
846 if (!filename) {
847 return NULL;
850 fd = open(filename, O_RDONLY | sha1_file_open_flag);
851 if (fd < 0) {
852 /* See if it works without O_NOATIME */
853 switch (sha1_file_open_flag) {
854 default:
855 fd = open(filename, O_RDONLY);
856 if (fd >= 0)
857 break;
858 /* Fallthrough */
859 case 0:
860 return NULL;
863 /* If it failed once, it will probably fail again.
864 * Stop using O_NOATIME
866 sha1_file_open_flag = 0;
868 map = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
869 close(fd);
870 *size = st.st_size;
871 return map;
874 int legacy_loose_object(unsigned char *map)
876 unsigned int word;
879 * Is it a zlib-compressed buffer? If so, the first byte
880 * must be 0x78 (15-bit window size, deflated), and the
881 * first 16-bit word is evenly divisible by 31
883 word = (map[0] << 8) + map[1];
884 if (map[0] == 0x78 && !(word % 31))
885 return 1;
886 else
887 return 0;
890 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
892 unsigned shift;
893 unsigned char c;
894 unsigned long size;
895 unsigned long used = 0;
897 c = buf[used++];
898 *type = (c >> 4) & 7;
899 size = c & 15;
900 shift = 4;
901 while (c & 0x80) {
902 if (len <= used)
903 return 0;
904 if (sizeof(long) * 8 <= shift)
905 return 0;
906 c = buf[used++];
907 size += (c & 0x7f) << shift;
908 shift += 7;
910 *sizep = size;
911 return used;
914 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
916 unsigned long size, used;
917 static const char valid_loose_object_type[8] = {
918 0, /* OBJ_EXT */
919 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
920 0, /* "delta" and others are invalid in a loose object */
922 enum object_type type;
924 /* Get the data stream */
925 memset(stream, 0, sizeof(*stream));
926 stream->next_in = map;
927 stream->avail_in = mapsize;
928 stream->next_out = buffer;
929 stream->avail_out = bufsiz;
931 if (legacy_loose_object(map)) {
932 inflateInit(stream);
933 return inflate(stream, 0);
936 used = unpack_object_header_gently(map, mapsize, &type, &size);
937 if (!used || !valid_loose_object_type[type])
938 return -1;
939 map += used;
940 mapsize -= used;
942 /* Set up the stream for the rest.. */
943 stream->next_in = map;
944 stream->avail_in = mapsize;
945 inflateInit(stream);
947 /* And generate the fake traditional header */
948 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
949 type_names[type], size);
950 return 0;
953 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
955 int bytes = strlen(buffer) + 1;
956 unsigned char *buf = xmalloc(1+size);
957 unsigned long n;
959 n = stream->total_out - bytes;
960 if (n > size)
961 n = size;
962 memcpy(buf, (char *) buffer + bytes, n);
963 bytes = n;
964 if (bytes < size) {
965 stream->next_out = buf + bytes;
966 stream->avail_out = size - bytes;
967 while (inflate(stream, Z_FINISH) == Z_OK)
968 /* nothing */;
970 buf[size] = 0;
971 inflateEnd(stream);
972 return buf;
976 * We used to just use "sscanf()", but that's actually way
977 * too permissive for what we want to check. So do an anal
978 * object header parse by hand.
980 static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
982 int i;
983 unsigned long size;
986 * The type can be at most ten bytes (including the
987 * terminating '\0' that we add), and is followed by
988 * a space.
990 i = 10;
991 for (;;) {
992 char c = *hdr++;
993 if (c == ' ')
994 break;
995 if (!--i)
996 return -1;
997 *type++ = c;
999 *type = 0;
1002 * The length must follow immediately, and be in canonical
1003 * decimal format (ie "010" is not valid).
1005 size = *hdr++ - '0';
1006 if (size > 9)
1007 return -1;
1008 if (size) {
1009 for (;;) {
1010 unsigned long c = *hdr - '0';
1011 if (c > 9)
1012 break;
1013 hdr++;
1014 size = size * 10 + c;
1017 *sizep = size;
1020 * The length must be followed by a zero byte
1022 return *hdr ? -1 : 0;
1025 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
1027 int ret;
1028 z_stream stream;
1029 char hdr[8192];
1031 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1032 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
1033 return NULL;
1035 return unpack_sha1_rest(&stream, hdr, *size);
1038 static unsigned long get_delta_base(struct packed_git *p,
1039 struct pack_window **w_curs,
1040 unsigned long offset,
1041 enum object_type kind,
1042 unsigned long delta_obj_offset,
1043 unsigned long *base_obj_offset)
1045 unsigned char *base_info = use_pack(p, w_curs, offset, NULL);
1046 unsigned long base_offset;
1048 /* use_pack() assured us we have [base_info, base_info + 20)
1049 * as a range that we can look at without walking off the
1050 * end of the mapped window. Its actually the hash size
1051 * that is assured. An OFS_DELTA longer than the hash size
1052 * is stupid, as then a REF_DELTA would be smaller to store.
1054 if (kind == OBJ_OFS_DELTA) {
1055 unsigned used = 0;
1056 unsigned char c = base_info[used++];
1057 base_offset = c & 127;
1058 while (c & 128) {
1059 base_offset += 1;
1060 if (!base_offset || base_offset & ~(~0UL >> 7))
1061 die("offset value overflow for delta base object");
1062 c = base_info[used++];
1063 base_offset = (base_offset << 7) + (c & 127);
1065 base_offset = delta_obj_offset - base_offset;
1066 if (base_offset >= delta_obj_offset)
1067 die("delta base offset out of bound");
1068 offset += used;
1069 } else if (kind == OBJ_REF_DELTA) {
1070 /* The base entry _must_ be in the same pack */
1071 base_offset = find_pack_entry_one(base_info, p);
1072 if (!base_offset)
1073 die("failed to find delta-pack base object %s",
1074 sha1_to_hex(base_info));
1075 offset += 20;
1076 } else
1077 die("I am totally screwed");
1078 *base_obj_offset = base_offset;
1079 return offset;
1082 /* forward declaration for a mutually recursive function */
1083 static int packed_object_info(struct packed_git *p, unsigned long offset,
1084 char *type, unsigned long *sizep);
1086 static int packed_delta_info(struct packed_git *p,
1087 struct pack_window **w_curs,
1088 unsigned long offset,
1089 enum object_type kind,
1090 unsigned long obj_offset,
1091 char *type,
1092 unsigned long *sizep)
1094 unsigned long base_offset;
1096 offset = get_delta_base(p, w_curs, offset, kind,
1097 obj_offset, &base_offset);
1099 /* We choose to only get the type of the base object and
1100 * ignore potentially corrupt pack file that expects the delta
1101 * based on a base with a wrong size. This saves tons of
1102 * inflate() calls.
1104 if (packed_object_info(p, base_offset, type, NULL))
1105 die("cannot get info for delta-pack base");
1107 if (sizep) {
1108 const unsigned char *data;
1109 unsigned char delta_head[20], *in;
1110 unsigned long result_size;
1111 z_stream stream;
1112 int st;
1114 memset(&stream, 0, sizeof(stream));
1115 stream.next_out = delta_head;
1116 stream.avail_out = sizeof(delta_head);
1118 inflateInit(&stream);
1119 do {
1120 in = use_pack(p, w_curs, offset, &stream.avail_in);
1121 stream.next_in = in;
1122 st = inflate(&stream, Z_FINISH);
1123 offset += stream.next_in - in;
1124 } while ((st == Z_OK || st == Z_BUF_ERROR)
1125 && stream.total_out < sizeof(delta_head));
1126 inflateEnd(&stream);
1127 if ((st != Z_STREAM_END) &&
1128 stream.total_out != sizeof(delta_head))
1129 die("delta data unpack-initial failed");
1131 /* Examine the initial part of the delta to figure out
1132 * the result size.
1134 data = delta_head;
1136 /* ignore base size */
1137 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1139 /* Read the result size */
1140 result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1141 *sizep = result_size;
1143 return 0;
1146 static unsigned long unpack_object_header(struct packed_git *p,
1147 struct pack_window **w_curs,
1148 unsigned long offset,
1149 enum object_type *type,
1150 unsigned long *sizep)
1152 unsigned char *base;
1153 unsigned int left;
1154 unsigned long used;
1156 /* use_pack() assures us we have [base, base + 20) available
1157 * as a range that we can look at at. (Its actually the hash
1158 * size that is assurred.) With our object header encoding
1159 * the maximum deflated object size is 2^137, which is just
1160 * insane, so we know won't exceed what we have been given.
1162 base = use_pack(p, w_curs, offset, &left);
1163 used = unpack_object_header_gently(base, left, type, sizep);
1164 if (!used)
1165 die("object offset outside of pack file");
1167 return offset + used;
1170 void packed_object_info_detail(struct packed_git *p,
1171 unsigned long offset,
1172 char *type,
1173 unsigned long *size,
1174 unsigned long *store_size,
1175 unsigned int *delta_chain_length,
1176 unsigned char *base_sha1)
1178 struct pack_window *w_curs = NULL;
1179 unsigned long obj_offset, val;
1180 unsigned char *next_sha1;
1181 enum object_type kind;
1183 *delta_chain_length = 0;
1184 obj_offset = offset;
1185 offset = unpack_object_header(p, &w_curs, offset, &kind, size);
1187 for (;;) {
1188 switch (kind) {
1189 default:
1190 die("pack %s contains unknown object type %d",
1191 p->pack_name, kind);
1192 case OBJ_COMMIT:
1193 case OBJ_TREE:
1194 case OBJ_BLOB:
1195 case OBJ_TAG:
1196 strcpy(type, type_names[kind]);
1197 *store_size = 0; /* notyet */
1198 unuse_pack(&w_curs);
1199 return;
1200 case OBJ_OFS_DELTA:
1201 get_delta_base(p, &w_curs, offset, kind,
1202 obj_offset, &offset);
1203 if (*delta_chain_length == 0) {
1204 /* TODO: find base_sha1 as pointed by offset */
1206 break;
1207 case OBJ_REF_DELTA:
1208 next_sha1 = use_pack(p, &w_curs, offset, NULL);
1209 if (*delta_chain_length == 0)
1210 hashcpy(base_sha1, next_sha1);
1211 offset = find_pack_entry_one(next_sha1, p);
1212 break;
1214 obj_offset = offset;
1215 offset = unpack_object_header(p, &w_curs, offset, &kind, &val);
1216 (*delta_chain_length)++;
1220 static int packed_object_info(struct packed_git *p, unsigned long offset,
1221 char *type, unsigned long *sizep)
1223 struct pack_window *w_curs = NULL;
1224 unsigned long size, obj_offset = offset;
1225 enum object_type kind;
1226 int r;
1228 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1230 switch (kind) {
1231 case OBJ_OFS_DELTA:
1232 case OBJ_REF_DELTA:
1233 r = packed_delta_info(p, &w_curs, offset, kind,
1234 obj_offset, type, sizep);
1235 unuse_pack(&w_curs);
1236 return r;
1237 case OBJ_COMMIT:
1238 case OBJ_TREE:
1239 case OBJ_BLOB:
1240 case OBJ_TAG:
1241 strcpy(type, type_names[kind]);
1242 unuse_pack(&w_curs);
1243 break;
1244 default:
1245 die("pack %s contains unknown object type %d",
1246 p->pack_name, kind);
1248 if (sizep)
1249 *sizep = size;
1250 return 0;
1253 static void *unpack_compressed_entry(struct packed_git *p,
1254 struct pack_window **w_curs,
1255 unsigned long offset,
1256 unsigned long size)
1258 int st;
1259 z_stream stream;
1260 unsigned char *buffer, *in;
1262 buffer = xmalloc(size + 1);
1263 buffer[size] = 0;
1264 memset(&stream, 0, sizeof(stream));
1265 stream.next_out = buffer;
1266 stream.avail_out = size;
1268 inflateInit(&stream);
1269 do {
1270 in = use_pack(p, w_curs, offset, &stream.avail_in);
1271 stream.next_in = in;
1272 st = inflate(&stream, Z_FINISH);
1273 offset += stream.next_in - in;
1274 } while (st == Z_OK || st == Z_BUF_ERROR);
1275 inflateEnd(&stream);
1276 if ((st != Z_STREAM_END) || stream.total_out != size) {
1277 free(buffer);
1278 return NULL;
1281 return buffer;
1284 static void *unpack_delta_entry(struct packed_git *p,
1285 struct pack_window **w_curs,
1286 unsigned long offset,
1287 unsigned long delta_size,
1288 enum object_type kind,
1289 unsigned long obj_offset,
1290 char *type,
1291 unsigned long *sizep)
1293 void *delta_data, *result, *base;
1294 unsigned long result_size, base_size, base_offset;
1296 offset = get_delta_base(p, w_curs, offset, kind,
1297 obj_offset, &base_offset);
1298 base = unpack_entry(p, base_offset, type, &base_size);
1299 if (!base)
1300 die("failed to read delta base object at %lu from %s",
1301 base_offset, p->pack_name);
1303 delta_data = unpack_compressed_entry(p, w_curs, offset, delta_size);
1304 result = patch_delta(base, base_size,
1305 delta_data, delta_size,
1306 &result_size);
1307 if (!result)
1308 die("failed to apply delta");
1309 free(delta_data);
1310 free(base);
1311 *sizep = result_size;
1312 return result;
1315 void *unpack_entry(struct packed_git *p, unsigned long offset,
1316 char *type, unsigned long *sizep)
1318 struct pack_window *w_curs = NULL;
1319 unsigned long size, obj_offset = offset;
1320 enum object_type kind;
1321 void *retval;
1323 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1324 switch (kind) {
1325 case OBJ_OFS_DELTA:
1326 case OBJ_REF_DELTA:
1327 retval = unpack_delta_entry(p, &w_curs, offset, size,
1328 kind, obj_offset, type, sizep);
1329 break;
1330 case OBJ_COMMIT:
1331 case OBJ_TREE:
1332 case OBJ_BLOB:
1333 case OBJ_TAG:
1334 strcpy(type, type_names[kind]);
1335 *sizep = size;
1336 retval = unpack_compressed_entry(p, &w_curs, offset, size);
1337 break;
1338 default:
1339 die("unknown object type %i in %s", kind, p->pack_name);
1341 unuse_pack(&w_curs);
1342 return retval;
1345 int num_packed_objects(const struct packed_git *p)
1347 /* See check_packed_git_idx() */
1348 return (p->index_size - 20 - 20 - 4*256) / 24;
1351 int nth_packed_object_sha1(const struct packed_git *p, int n,
1352 unsigned char* sha1)
1354 void *index = p->index_base + 256;
1355 if (n < 0 || num_packed_objects(p) <= n)
1356 return -1;
1357 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1358 return 0;
1361 unsigned long find_pack_entry_one(const unsigned char *sha1,
1362 struct packed_git *p)
1364 uint32_t *level1_ofs = p->index_base;
1365 int hi = ntohl(level1_ofs[*sha1]);
1366 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1367 void *index = p->index_base + 256;
1369 do {
1370 int mi = (lo + hi) / 2;
1371 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1372 if (!cmp)
1373 return ntohl(*((uint32_t *)((char *)index + (24 * mi))));
1374 if (cmp > 0)
1375 hi = mi;
1376 else
1377 lo = mi+1;
1378 } while (lo < hi);
1379 return 0;
1382 static int matches_pack_name(struct packed_git *p, const char *ig)
1384 const char *last_c, *c;
1386 if (!strcmp(p->pack_name, ig))
1387 return 0;
1389 for (c = p->pack_name, last_c = c; *c;)
1390 if (*c == '/')
1391 last_c = ++c;
1392 else
1393 ++c;
1394 if (!strcmp(last_c, ig))
1395 return 0;
1397 return 1;
1400 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1402 struct packed_git *p;
1403 unsigned long offset;
1405 prepare_packed_git();
1407 for (p = packed_git; p; p = p->next) {
1408 if (ignore_packed) {
1409 const char **ig;
1410 for (ig = ignore_packed; *ig; ig++)
1411 if (!matches_pack_name(p, *ig))
1412 break;
1413 if (*ig)
1414 continue;
1416 offset = find_pack_entry_one(sha1, p);
1417 if (offset) {
1418 e->offset = offset;
1419 e->p = p;
1420 hashcpy(e->sha1, sha1);
1421 return 1;
1424 return 0;
1427 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1428 struct packed_git *packs)
1430 struct packed_git *p;
1432 for (p = packs; p; p = p->next) {
1433 if (find_pack_entry_one(sha1, p))
1434 return p;
1436 return NULL;
1440 static int sha1_loose_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1442 int status;
1443 unsigned long mapsize, size;
1444 void *map;
1445 z_stream stream;
1446 char hdr[128];
1448 map = map_sha1_file(sha1, &mapsize);
1449 if (!map)
1450 return error("unable to find %s", sha1_to_hex(sha1));
1451 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1452 status = error("unable to unpack %s header",
1453 sha1_to_hex(sha1));
1454 if (parse_sha1_header(hdr, type, &size) < 0)
1455 status = error("unable to parse %s header", sha1_to_hex(sha1));
1456 else {
1457 status = 0;
1458 if (sizep)
1459 *sizep = size;
1461 inflateEnd(&stream);
1462 munmap(map, mapsize);
1463 return status;
1466 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1468 struct pack_entry e;
1470 if (!find_pack_entry(sha1, &e, NULL)) {
1471 reprepare_packed_git();
1472 if (!find_pack_entry(sha1, &e, NULL))
1473 return sha1_loose_object_info(sha1, type, sizep);
1475 return packed_object_info(e.p, e.offset, type, sizep);
1478 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1480 struct pack_entry e;
1482 if (!find_pack_entry(sha1, &e, NULL))
1483 return NULL;
1484 else
1485 return unpack_entry(e.p, e.offset, type, size);
1488 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1490 unsigned long mapsize;
1491 void *map, *buf;
1493 buf = read_packed_sha1(sha1, type, size);
1494 if (buf)
1495 return buf;
1496 map = map_sha1_file(sha1, &mapsize);
1497 if (map) {
1498 buf = unpack_sha1_file(map, mapsize, type, size);
1499 munmap(map, mapsize);
1500 return buf;
1502 reprepare_packed_git();
1503 return read_packed_sha1(sha1, type, size);
1506 void *read_object_with_reference(const unsigned char *sha1,
1507 const char *required_type,
1508 unsigned long *size,
1509 unsigned char *actual_sha1_return)
1511 char type[20];
1512 void *buffer;
1513 unsigned long isize;
1514 unsigned char actual_sha1[20];
1516 hashcpy(actual_sha1, sha1);
1517 while (1) {
1518 int ref_length = -1;
1519 const char *ref_type = NULL;
1521 buffer = read_sha1_file(actual_sha1, type, &isize);
1522 if (!buffer)
1523 return NULL;
1524 if (!strcmp(type, required_type)) {
1525 *size = isize;
1526 if (actual_sha1_return)
1527 hashcpy(actual_sha1_return, actual_sha1);
1528 return buffer;
1530 /* Handle references */
1531 else if (!strcmp(type, commit_type))
1532 ref_type = "tree ";
1533 else if (!strcmp(type, tag_type))
1534 ref_type = "object ";
1535 else {
1536 free(buffer);
1537 return NULL;
1539 ref_length = strlen(ref_type);
1541 if (memcmp(buffer, ref_type, ref_length) ||
1542 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1543 free(buffer);
1544 return NULL;
1546 free(buffer);
1547 /* Now we have the ID of the referred-to object in
1548 * actual_sha1. Check again. */
1552 static void write_sha1_file_prepare(void *buf, unsigned long len,
1553 const char *type, unsigned char *sha1,
1554 unsigned char *hdr, int *hdrlen)
1556 SHA_CTX c;
1558 /* Generate the header */
1559 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1561 /* Sha1.. */
1562 SHA1_Init(&c);
1563 SHA1_Update(&c, hdr, *hdrlen);
1564 SHA1_Update(&c, buf, len);
1565 SHA1_Final(sha1, &c);
1569 * Link the tempfile to the final place, possibly creating the
1570 * last directory level as you do so.
1572 * Returns the errno on failure, 0 on success.
1574 static int link_temp_to_file(const char *tmpfile, const char *filename)
1576 int ret;
1577 char *dir;
1579 if (!link(tmpfile, filename))
1580 return 0;
1583 * Try to mkdir the last path component if that failed.
1585 * Re-try the "link()" regardless of whether the mkdir
1586 * succeeds, since a race might mean that somebody
1587 * else succeeded.
1589 ret = errno;
1590 dir = strrchr(filename, '/');
1591 if (dir) {
1592 *dir = 0;
1593 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1594 *dir = '/';
1595 return -2;
1597 *dir = '/';
1598 if (!link(tmpfile, filename))
1599 return 0;
1600 ret = errno;
1602 return ret;
1606 * Move the just written object into its final resting place
1608 int move_temp_to_file(const char *tmpfile, const char *filename)
1610 int ret = link_temp_to_file(tmpfile, filename);
1613 * Coda hack - coda doesn't like cross-directory links,
1614 * so we fall back to a rename, which will mean that it
1615 * won't be able to check collisions, but that's not a
1616 * big deal.
1618 * The same holds for FAT formatted media.
1620 * When this succeeds, we just return 0. We have nothing
1621 * left to unlink.
1623 if (ret && ret != EEXIST) {
1624 if (!rename(tmpfile, filename))
1625 return 0;
1626 ret = errno;
1628 unlink(tmpfile);
1629 if (ret) {
1630 if (ret != EEXIST) {
1631 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1633 /* FIXME!!! Collision check here ? */
1636 return 0;
1639 static int write_buffer(int fd, const void *buf, size_t len)
1641 if (write_in_full(fd, buf, len) < 0)
1642 return error("file write error (%s)", strerror(errno));
1643 return 0;
1646 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1648 int hdr_len;
1649 unsigned char c;
1651 c = (type << 4) | (len & 15);
1652 len >>= 4;
1653 hdr_len = 1;
1654 while (len) {
1655 *hdr++ = c | 0x80;
1656 hdr_len++;
1657 c = (len & 0x7f);
1658 len >>= 7;
1660 *hdr = c;
1661 return hdr_len;
1664 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1666 int obj_type, hdr;
1668 if (use_legacy_headers) {
1669 while (deflate(stream, 0) == Z_OK)
1670 /* nothing */;
1671 return;
1673 if (!strcmp(type, blob_type))
1674 obj_type = OBJ_BLOB;
1675 else if (!strcmp(type, tree_type))
1676 obj_type = OBJ_TREE;
1677 else if (!strcmp(type, commit_type))
1678 obj_type = OBJ_COMMIT;
1679 else if (!strcmp(type, tag_type))
1680 obj_type = OBJ_TAG;
1681 else
1682 die("trying to generate bogus object of type '%s'", type);
1683 hdr = write_binary_header(stream->next_out, obj_type, len);
1684 stream->total_out = hdr;
1685 stream->next_out += hdr;
1686 stream->avail_out -= hdr;
1689 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1690 unsigned char *sha1)
1692 unsigned char hdr[50];
1693 int hdrlen;
1694 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1695 return 0;
1698 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1700 int size;
1701 unsigned char *compressed;
1702 z_stream stream;
1703 unsigned char sha1[20];
1704 char *filename;
1705 static char tmpfile[PATH_MAX];
1706 unsigned char hdr[50];
1707 int fd, hdrlen;
1709 /* Normally if we have it in the pack then we do not bother writing
1710 * it out into .git/objects/??/?{38} file.
1712 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1713 filename = sha1_file_name(sha1);
1714 if (returnsha1)
1715 hashcpy(returnsha1, sha1);
1716 if (has_sha1_file(sha1))
1717 return 0;
1718 fd = open(filename, O_RDONLY);
1719 if (fd >= 0) {
1721 * FIXME!!! We might do collision checking here, but we'd
1722 * need to uncompress the old file and check it. Later.
1724 close(fd);
1725 return 0;
1728 if (errno != ENOENT) {
1729 return error("sha1 file %s: %s\n", filename, strerror(errno));
1732 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1734 fd = mkstemp(tmpfile);
1735 if (fd < 0) {
1736 if (errno == EPERM)
1737 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1738 else
1739 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1742 /* Set it up */
1743 memset(&stream, 0, sizeof(stream));
1744 deflateInit(&stream, zlib_compression_level);
1745 size = 8 + deflateBound(&stream, len+hdrlen);
1746 compressed = xmalloc(size);
1748 /* Compress it */
1749 stream.next_out = compressed;
1750 stream.avail_out = size;
1752 /* First header.. */
1753 stream.next_in = hdr;
1754 stream.avail_in = hdrlen;
1755 setup_object_header(&stream, type, len);
1757 /* Then the data itself.. */
1758 stream.next_in = buf;
1759 stream.avail_in = len;
1760 while (deflate(&stream, Z_FINISH) == Z_OK)
1761 /* nothing */;
1762 deflateEnd(&stream);
1763 size = stream.total_out;
1765 if (write_buffer(fd, compressed, size) < 0)
1766 die("unable to write sha1 file");
1767 fchmod(fd, 0444);
1768 close(fd);
1769 free(compressed);
1771 return move_temp_to_file(tmpfile, filename);
1775 * We need to unpack and recompress the object for writing
1776 * it out to a different file.
1778 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1780 size_t size;
1781 z_stream stream;
1782 unsigned char *unpacked;
1783 unsigned long len;
1784 char type[20];
1785 char hdr[50];
1786 int hdrlen;
1787 void *buf;
1789 /* need to unpack and recompress it by itself */
1790 unpacked = read_packed_sha1(sha1, type, &len);
1791 if (!unpacked)
1792 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1794 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1796 /* Set it up */
1797 memset(&stream, 0, sizeof(stream));
1798 deflateInit(&stream, zlib_compression_level);
1799 size = deflateBound(&stream, len + hdrlen);
1800 buf = xmalloc(size);
1802 /* Compress it */
1803 stream.next_out = buf;
1804 stream.avail_out = size;
1806 /* First header.. */
1807 stream.next_in = (void *)hdr;
1808 stream.avail_in = hdrlen;
1809 while (deflate(&stream, 0) == Z_OK)
1810 /* nothing */;
1812 /* Then the data itself.. */
1813 stream.next_in = unpacked;
1814 stream.avail_in = len;
1815 while (deflate(&stream, Z_FINISH) == Z_OK)
1816 /* nothing */;
1817 deflateEnd(&stream);
1818 free(unpacked);
1820 *objsize = stream.total_out;
1821 return buf;
1824 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1826 int retval;
1827 unsigned long objsize;
1828 void *buf = map_sha1_file(sha1, &objsize);
1830 if (buf) {
1831 retval = write_buffer(fd, buf, objsize);
1832 munmap(buf, objsize);
1833 return retval;
1836 buf = repack_object(sha1, &objsize);
1837 retval = write_buffer(fd, buf, objsize);
1838 free(buf);
1839 return retval;
1842 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1843 size_t bufsize, size_t *bufposn)
1845 char tmpfile[PATH_MAX];
1846 int local;
1847 z_stream stream;
1848 unsigned char real_sha1[20];
1849 unsigned char discard[4096];
1850 int ret;
1851 SHA_CTX c;
1853 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1855 local = mkstemp(tmpfile);
1856 if (local < 0) {
1857 if (errno == EPERM)
1858 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1859 else
1860 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1863 memset(&stream, 0, sizeof(stream));
1865 inflateInit(&stream);
1867 SHA1_Init(&c);
1869 do {
1870 ssize_t size;
1871 if (*bufposn) {
1872 stream.avail_in = *bufposn;
1873 stream.next_in = (unsigned char *) buffer;
1874 do {
1875 stream.next_out = discard;
1876 stream.avail_out = sizeof(discard);
1877 ret = inflate(&stream, Z_SYNC_FLUSH);
1878 SHA1_Update(&c, discard, sizeof(discard) -
1879 stream.avail_out);
1880 } while (stream.avail_in && ret == Z_OK);
1881 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1882 die("unable to write sha1 file");
1883 memmove(buffer, buffer + *bufposn - stream.avail_in,
1884 stream.avail_in);
1885 *bufposn = stream.avail_in;
1886 if (ret != Z_OK)
1887 break;
1889 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
1890 if (size <= 0) {
1891 close(local);
1892 unlink(tmpfile);
1893 if (!size)
1894 return error("Connection closed?");
1895 perror("Reading from connection");
1896 return -1;
1898 *bufposn += size;
1899 } while (1);
1900 inflateEnd(&stream);
1902 close(local);
1903 SHA1_Final(real_sha1, &c);
1904 if (ret != Z_STREAM_END) {
1905 unlink(tmpfile);
1906 return error("File %s corrupted", sha1_to_hex(sha1));
1908 if (hashcmp(sha1, real_sha1)) {
1909 unlink(tmpfile);
1910 return error("File %s has bad hash", sha1_to_hex(sha1));
1913 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1916 int has_pack_index(const unsigned char *sha1)
1918 struct stat st;
1919 if (stat(sha1_pack_index_name(sha1), &st))
1920 return 0;
1921 return 1;
1924 int has_pack_file(const unsigned char *sha1)
1926 struct stat st;
1927 if (stat(sha1_pack_name(sha1), &st))
1928 return 0;
1929 return 1;
1932 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1934 struct pack_entry e;
1935 return find_pack_entry(sha1, &e, ignore_packed);
1938 int has_sha1_file(const unsigned char *sha1)
1940 struct stat st;
1941 struct pack_entry e;
1943 if (find_pack_entry(sha1, &e, NULL))
1944 return 1;
1945 return find_sha1_file(sha1, &st) ? 1 : 0;
1949 * reads from fd as long as possible into a supplied buffer of size bytes.
1950 * If necessary the buffer's size is increased using realloc()
1952 * returns 0 if anything went fine and -1 otherwise
1954 * NOTE: both buf and size may change, but even when -1 is returned
1955 * you still have to free() it yourself.
1957 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
1959 char* buf = *return_buf;
1960 unsigned long size = *return_size;
1961 int iret;
1962 unsigned long off = 0;
1964 do {
1965 iret = xread(fd, buf + off, size - off);
1966 if (iret > 0) {
1967 off += iret;
1968 if (off == size) {
1969 size *= 2;
1970 buf = xrealloc(buf, size);
1973 } while (iret > 0);
1975 *return_buf = buf;
1976 *return_size = off;
1978 if (iret < 0)
1979 return -1;
1980 return 0;
1983 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
1985 unsigned long size = 4096;
1986 char *buf = xmalloc(size);
1987 int ret;
1989 if (read_pipe(fd, &buf, &size)) {
1990 free(buf);
1991 return -1;
1994 if (!type)
1995 type = blob_type;
1996 if (write_object)
1997 ret = write_sha1_file(buf, size, type, sha1);
1998 else
1999 ret = hash_sha1_file(buf, size, type, sha1);
2000 free(buf);
2001 return ret;
2004 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
2006 unsigned long size = st->st_size;
2007 void *buf;
2008 int ret;
2010 buf = "";
2011 if (size)
2012 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2013 close(fd);
2015 if (!type)
2016 type = blob_type;
2017 if (write_object)
2018 ret = write_sha1_file(buf, size, type, sha1);
2019 else
2020 ret = hash_sha1_file(buf, size, type, sha1);
2021 if (size)
2022 munmap(buf, size);
2023 return ret;
2026 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2028 int fd;
2029 char *target;
2031 switch (st->st_mode & S_IFMT) {
2032 case S_IFREG:
2033 fd = open(path, O_RDONLY);
2034 if (fd < 0)
2035 return error("open(\"%s\"): %s", path,
2036 strerror(errno));
2037 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
2038 return error("%s: failed to insert into database",
2039 path);
2040 break;
2041 case S_IFLNK:
2042 target = xmalloc(st->st_size+1);
2043 if (readlink(path, target, st->st_size+1) != st->st_size) {
2044 char *errstr = strerror(errno);
2045 free(target);
2046 return error("readlink(\"%s\"): %s", path,
2047 errstr);
2049 if (!write_object)
2050 hash_sha1_file(target, st->st_size, blob_type, sha1);
2051 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
2052 return error("%s: failed to insert into database",
2053 path);
2054 free(target);
2055 break;
2056 default:
2057 return error("%s: unsupported file type", path);
2059 return 0;
2062 int read_pack_header(int fd, struct pack_header *header)
2064 char *c = (char*)header;
2065 ssize_t remaining = sizeof(struct pack_header);
2066 do {
2067 ssize_t r = xread(fd, c, remaining);
2068 if (r <= 0)
2069 /* "eof before pack header was fully read" */
2070 return PH_ERROR_EOF;
2071 remaining -= r;
2072 c += r;
2073 } while (remaining > 0);
2074 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2075 /* "protocol error (pack signature mismatch detected)" */
2076 return PH_ERROR_PACK_SIGNATURE;
2077 if (!pack_version_ok(header->hdr_version))
2078 /* "protocol error (pack version unsupported)" */
2079 return PH_ERROR_PROTOCOL;
2080 return 0;