sha1_file.c: cleanup "offset" usage
[git/gitweb.git] / sha1_file.c
blob8147a4f5054694a1d715128b7e9bd761ade930ca
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 int safe_create_leading_directories(char *path)
85 char *pos = path;
86 struct stat st;
88 if (*pos == '/')
89 pos++;
91 while (pos) {
92 pos = strchr(pos, '/');
93 if (!pos)
94 break;
95 *pos = 0;
96 if (!stat(path, &st)) {
97 /* path exists */
98 if (!S_ISDIR(st.st_mode)) {
99 *pos = '/';
100 return -3;
103 else if (mkdir(path, 0777)) {
104 *pos = '/';
105 return -1;
107 else if (adjust_shared_perm(path)) {
108 *pos = '/';
109 return -2;
111 *pos++ = '/';
113 return 0;
116 char * sha1_to_hex(const unsigned char *sha1)
118 static int bufno;
119 static char hexbuffer[4][50];
120 static const char hex[] = "0123456789abcdef";
121 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
122 int i;
124 for (i = 0; i < 20; i++) {
125 unsigned int val = *sha1++;
126 *buf++ = hex[val >> 4];
127 *buf++ = hex[val & 0xf];
129 *buf = '\0';
131 return buffer;
134 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
136 int i;
137 for (i = 0; i < 20; i++) {
138 static char hex[] = "0123456789abcdef";
139 unsigned int val = sha1[i];
140 char *pos = pathbuf + i*2 + (i > 0);
141 *pos++ = hex[val >> 4];
142 *pos = hex[val & 0xf];
147 * NOTE! This returns a statically allocated buffer, so you have to be
148 * careful about using it. Do a "xstrdup()" if you need to save the
149 * filename.
151 * Also note that this returns the location for creating. Reading
152 * SHA1 file can happen from any alternate directory listed in the
153 * DB_ENVIRONMENT environment variable if it is not found in
154 * the primary object database.
156 char *sha1_file_name(const unsigned char *sha1)
158 static char *name, *base;
160 if (!base) {
161 const char *sha1_file_directory = get_object_directory();
162 int len = strlen(sha1_file_directory);
163 base = xmalloc(len + 60);
164 memcpy(base, sha1_file_directory, len);
165 memset(base+len, 0, 60);
166 base[len] = '/';
167 base[len+3] = '/';
168 name = base + len + 1;
170 fill_sha1_path(name, sha1);
171 return base;
174 char *sha1_pack_name(const unsigned char *sha1)
176 static const char hex[] = "0123456789abcdef";
177 static char *name, *base, *buf;
178 int i;
180 if (!base) {
181 const char *sha1_file_directory = get_object_directory();
182 int len = strlen(sha1_file_directory);
183 base = xmalloc(len + 60);
184 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
185 name = base + len + 11;
188 buf = name;
190 for (i = 0; i < 20; i++) {
191 unsigned int val = *sha1++;
192 *buf++ = hex[val >> 4];
193 *buf++ = hex[val & 0xf];
196 return base;
199 char *sha1_pack_index_name(const unsigned char *sha1)
201 static const char hex[] = "0123456789abcdef";
202 static char *name, *base, *buf;
203 int i;
205 if (!base) {
206 const char *sha1_file_directory = get_object_directory();
207 int len = strlen(sha1_file_directory);
208 base = xmalloc(len + 60);
209 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
210 name = base + len + 11;
213 buf = name;
215 for (i = 0; i < 20; i++) {
216 unsigned int val = *sha1++;
217 *buf++ = hex[val >> 4];
218 *buf++ = hex[val & 0xf];
221 return base;
224 struct alternate_object_database *alt_odb_list;
225 static struct alternate_object_database **alt_odb_tail;
227 static void read_info_alternates(const char * alternates, int depth);
230 * Prepare alternate object database registry.
232 * The variable alt_odb_list points at the list of struct
233 * alternate_object_database. The elements on this list come from
234 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
235 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
236 * whose contents is similar to that environment variable but can be
237 * LF separated. Its base points at a statically allocated buffer that
238 * contains "/the/directory/corresponding/to/.git/objects/...", while
239 * its name points just after the slash at the end of ".git/objects/"
240 * in the example above, and has enough space to hold 40-byte hex
241 * SHA1, an extra slash for the first level indirection, and the
242 * terminating NUL.
244 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
246 struct stat st;
247 const char *objdir = get_object_directory();
248 struct alternate_object_database *ent;
249 struct alternate_object_database *alt;
250 /* 43 = 40-byte + 2 '/' + terminating NUL */
251 int pfxlen = len;
252 int entlen = pfxlen + 43;
253 int base_len = -1;
255 if (*entry != '/' && relative_base) {
256 /* Relative alt-odb */
257 if (base_len < 0)
258 base_len = strlen(relative_base) + 1;
259 entlen += base_len;
260 pfxlen += base_len;
262 ent = xmalloc(sizeof(*ent) + entlen);
264 if (*entry != '/' && relative_base) {
265 memcpy(ent->base, relative_base, base_len - 1);
266 ent->base[base_len - 1] = '/';
267 memcpy(ent->base + base_len, entry, len);
269 else
270 memcpy(ent->base, entry, pfxlen);
272 ent->name = ent->base + pfxlen + 1;
273 ent->base[pfxlen + 3] = '/';
274 ent->base[pfxlen] = ent->base[entlen-1] = 0;
276 /* Detect cases where alternate disappeared */
277 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
278 error("object directory %s does not exist; "
279 "check .git/objects/info/alternates.",
280 ent->base);
281 free(ent);
282 return -1;
285 /* Prevent the common mistake of listing the same
286 * thing twice, or object directory itself.
288 for (alt = alt_odb_list; alt; alt = alt->next) {
289 if (!memcmp(ent->base, alt->base, pfxlen)) {
290 free(ent);
291 return -1;
294 if (!memcmp(ent->base, objdir, pfxlen)) {
295 free(ent);
296 return -1;
299 /* add the alternate entry */
300 *alt_odb_tail = ent;
301 alt_odb_tail = &(ent->next);
302 ent->next = NULL;
304 /* recursively add alternates */
305 read_info_alternates(ent->base, depth + 1);
307 ent->base[pfxlen] = '/';
309 return 0;
312 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
313 const char *relative_base, int depth)
315 const char *cp, *last;
317 if (depth > 5) {
318 error("%s: ignoring alternate object stores, nesting too deep.",
319 relative_base);
320 return;
323 last = alt;
324 while (last < ep) {
325 cp = last;
326 if (cp < ep && *cp == '#') {
327 while (cp < ep && *cp != sep)
328 cp++;
329 last = cp + 1;
330 continue;
332 while (cp < ep && *cp != sep)
333 cp++;
334 if (last != cp) {
335 if ((*last != '/') && depth) {
336 error("%s: ignoring relative alternate object store %s",
337 relative_base, last);
338 } else {
339 link_alt_odb_entry(last, cp - last,
340 relative_base, depth);
343 while (cp < ep && *cp == sep)
344 cp++;
345 last = cp;
349 static void read_info_alternates(const char * relative_base, int depth)
351 char *map;
352 struct stat st;
353 char path[PATH_MAX];
354 int fd;
356 sprintf(path, "%s/info/alternates", relative_base);
357 fd = open(path, O_RDONLY);
358 if (fd < 0)
359 return;
360 if (fstat(fd, &st) || (st.st_size == 0)) {
361 close(fd);
362 return;
364 map = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
365 close(fd);
367 link_alt_odb_entries(map, map + st.st_size, '\n', relative_base, depth);
369 munmap(map, st.st_size);
372 void prepare_alt_odb(void)
374 const char *alt;
376 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
377 if (!alt) alt = "";
379 if (alt_odb_tail)
380 return;
381 alt_odb_tail = &alt_odb_list;
382 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
384 read_info_alternates(get_object_directory(), 0);
387 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
389 char *name = sha1_file_name(sha1);
390 struct alternate_object_database *alt;
392 if (!stat(name, st))
393 return name;
394 prepare_alt_odb();
395 for (alt = alt_odb_list; alt; alt = alt->next) {
396 name = alt->name;
397 fill_sha1_path(name, sha1);
398 if (!stat(alt->base, st))
399 return alt->base;
401 return NULL;
404 static unsigned int pack_used_ctr;
405 static unsigned int pack_mmap_calls;
406 static unsigned int peak_pack_open_windows;
407 static unsigned int pack_open_windows;
408 static size_t peak_pack_mapped;
409 static size_t pack_mapped;
410 struct packed_git *packed_git;
412 void pack_report()
414 fprintf(stderr,
415 "pack_report: getpagesize() = %10" SZ_FMT "\n"
416 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
417 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
418 (size_t) getpagesize(),
419 packed_git_window_size,
420 packed_git_limit);
421 fprintf(stderr,
422 "pack_report: pack_used_ctr = %10u\n"
423 "pack_report: pack_mmap_calls = %10u\n"
424 "pack_report: pack_open_windows = %10u / %10u\n"
425 "pack_report: pack_mapped = "
426 "%10" SZ_FMT " / %10" SZ_FMT "\n",
427 pack_used_ctr,
428 pack_mmap_calls,
429 pack_open_windows, peak_pack_open_windows,
430 pack_mapped, peak_pack_mapped);
433 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
434 void **idx_map_)
436 void *idx_map;
437 uint32_t *index;
438 unsigned long idx_size;
439 int nr, i;
440 int fd = open(path, O_RDONLY);
441 struct stat st;
442 if (fd < 0)
443 return -1;
444 if (fstat(fd, &st)) {
445 close(fd);
446 return -1;
448 idx_size = st.st_size;
449 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
450 close(fd);
452 index = idx_map;
453 *idx_map_ = idx_map;
454 *idx_size_ = idx_size;
456 /* check index map */
457 if (idx_size < 4*256 + 20 + 20)
458 return error("index file %s is too small", path);
460 /* a future index format would start with this, as older git
461 * binaries would fail the non-monotonic index check below.
462 * give a nicer warning to the user if we can.
464 if (index[0] == htonl(PACK_IDX_SIGNATURE))
465 return error("index file %s is a newer version"
466 " and is not supported by this binary"
467 " (try upgrading GIT to a newer version)",
468 path);
470 nr = 0;
471 for (i = 0; i < 256; i++) {
472 unsigned int n = ntohl(index[i]);
473 if (n < nr)
474 return error("non-monotonic index %s", path);
475 nr = n;
479 * Total size:
480 * - 256 index entries 4 bytes each
481 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
482 * - 20-byte SHA1 of the packfile
483 * - 20-byte SHA1 file checksum
485 if (idx_size != 4*256 + nr * 24 + 20 + 20)
486 return error("wrong index file size in %s", path);
488 return 0;
491 static void scan_windows(struct packed_git *p,
492 struct packed_git **lru_p,
493 struct pack_window **lru_w,
494 struct pack_window **lru_l)
496 struct pack_window *w, *w_l;
498 for (w_l = NULL, w = p->windows; w; w = w->next) {
499 if (!w->inuse_cnt) {
500 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
501 *lru_p = p;
502 *lru_w = w;
503 *lru_l = w_l;
506 w_l = w;
510 static int unuse_one_window(struct packed_git *current)
512 struct packed_git *p, *lru_p = NULL;
513 struct pack_window *lru_w = NULL, *lru_l = NULL;
515 if (current)
516 scan_windows(current, &lru_p, &lru_w, &lru_l);
517 for (p = packed_git; p; p = p->next)
518 scan_windows(p, &lru_p, &lru_w, &lru_l);
519 if (lru_p) {
520 munmap(lru_w->base, lru_w->len);
521 pack_mapped -= lru_w->len;
522 if (lru_l)
523 lru_l->next = lru_w->next;
524 else {
525 lru_p->windows = lru_w->next;
526 if (!lru_p->windows && lru_p != current) {
527 close(lru_p->pack_fd);
528 lru_p->pack_fd = -1;
531 free(lru_w);
532 pack_open_windows--;
533 return 1;
535 return 0;
538 void release_pack_memory(size_t need)
540 size_t cur = pack_mapped;
541 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
542 ; /* nothing */
545 void unuse_pack(struct pack_window **w_cursor)
547 struct pack_window *w = *w_cursor;
548 if (w) {
549 w->inuse_cnt--;
550 *w_cursor = NULL;
555 * Do not call this directly as this leaks p->pack_fd on error return;
556 * call open_packed_git() instead.
558 static int open_packed_git_1(struct packed_git *p)
560 struct stat st;
561 struct pack_header hdr;
562 unsigned char sha1[20];
563 unsigned char *idx_sha1;
564 long fd_flag;
566 p->pack_fd = open(p->pack_name, O_RDONLY);
567 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
568 return -1;
570 /* If we created the struct before we had the pack we lack size. */
571 if (!p->pack_size) {
572 if (!S_ISREG(st.st_mode))
573 return error("packfile %s not a regular file", p->pack_name);
574 p->pack_size = st.st_size;
575 } else if (p->pack_size != st.st_size)
576 return error("packfile %s size changed", p->pack_name);
578 /* We leave these file descriptors open with sliding mmap;
579 * there is no point keeping them open across exec(), though.
581 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
582 if (fd_flag < 0)
583 return error("cannot determine file descriptor flags");
584 fd_flag |= FD_CLOEXEC;
585 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
586 return error("cannot set FD_CLOEXEC");
588 /* Verify we recognize this pack file format. */
589 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
590 return error("file %s is far too short to be a packfile", p->pack_name);
591 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
592 return error("file %s is not a GIT packfile", p->pack_name);
593 if (!pack_version_ok(hdr.hdr_version))
594 return error("packfile %s is version %u and not supported"
595 " (try upgrading GIT to a newer version)",
596 p->pack_name, ntohl(hdr.hdr_version));
598 /* Verify the pack matches its index. */
599 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
600 return error("packfile %s claims to have %u objects"
601 " while index size indicates %u objects",
602 p->pack_name, ntohl(hdr.hdr_entries),
603 num_packed_objects(p));
604 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
605 return error("end of packfile %s is unavailable", p->pack_name);
606 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
607 return error("packfile %s signature is unavailable", p->pack_name);
608 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
609 if (hashcmp(sha1, idx_sha1))
610 return error("packfile %s does not match index", p->pack_name);
611 return 0;
614 static int open_packed_git(struct packed_git *p)
616 if (!open_packed_git_1(p))
617 return 0;
618 if (p->pack_fd != -1) {
619 close(p->pack_fd);
620 p->pack_fd = -1;
622 return -1;
625 static int in_window(struct pack_window *win, unsigned long offset)
627 /* We must promise at least 20 bytes (one hash) after the
628 * offset is available from this window, otherwise the offset
629 * is not actually in this window and a different window (which
630 * has that one hash excess) must be used. This is to support
631 * the object header and delta base parsing routines below.
633 off_t win_off = win->offset;
634 return win_off <= offset
635 && (offset + 20) <= (win_off + win->len);
638 unsigned char* use_pack(struct packed_git *p,
639 struct pack_window **w_cursor,
640 unsigned long offset,
641 unsigned int *left)
643 struct pack_window *win = *w_cursor;
645 if (p->pack_fd == -1 && open_packed_git(p))
646 die("packfile %s cannot be accessed", p->pack_name);
648 /* Since packfiles end in a hash of their content and its
649 * pointless to ask for an offset into the middle of that
650 * hash, and the in_window function above wouldn't match
651 * don't allow an offset too close to the end of the file.
653 if (offset > (p->pack_size - 20))
654 die("offset beyond end of packfile (truncated pack?)");
656 if (!win || !in_window(win, offset)) {
657 if (win)
658 win->inuse_cnt--;
659 for (win = p->windows; win; win = win->next) {
660 if (in_window(win, offset))
661 break;
663 if (!win) {
664 size_t window_align = packed_git_window_size / 2;
665 win = xcalloc(1, sizeof(*win));
666 win->offset = (offset / window_align) * window_align;
667 win->len = p->pack_size - win->offset;
668 if (win->len > packed_git_window_size)
669 win->len = packed_git_window_size;
670 pack_mapped += win->len;
671 while (packed_git_limit < pack_mapped
672 && unuse_one_window(p))
673 ; /* nothing */
674 win->base = xmmap(NULL, win->len,
675 PROT_READ, MAP_PRIVATE,
676 p->pack_fd, win->offset);
677 if (win->base == MAP_FAILED)
678 die("packfile %s cannot be mapped: %s",
679 p->pack_name,
680 strerror(errno));
681 pack_mmap_calls++;
682 pack_open_windows++;
683 if (pack_mapped > peak_pack_mapped)
684 peak_pack_mapped = pack_mapped;
685 if (pack_open_windows > peak_pack_open_windows)
686 peak_pack_open_windows = pack_open_windows;
687 win->next = p->windows;
688 p->windows = win;
691 if (win != *w_cursor) {
692 win->last_used = pack_used_ctr++;
693 win->inuse_cnt++;
694 *w_cursor = win;
696 offset -= win->offset;
697 if (left)
698 *left = win->len - offset;
699 return win->base + offset;
702 struct packed_git *add_packed_git(char *path, int path_len, int local)
704 struct stat st;
705 struct packed_git *p;
706 unsigned long idx_size;
707 void *idx_map;
708 unsigned char sha1[20];
710 if (check_packed_git_idx(path, &idx_size, &idx_map))
711 return NULL;
713 /* do we have a corresponding .pack file? */
714 strcpy(path + path_len - 4, ".pack");
715 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
716 munmap(idx_map, idx_size);
717 return NULL;
719 /* ok, it looks sane as far as we can check without
720 * actually mapping the pack file.
722 p = xmalloc(sizeof(*p) + path_len + 2);
723 strcpy(p->pack_name, path);
724 p->index_size = idx_size;
725 p->pack_size = st.st_size;
726 p->index_base = idx_map;
727 p->next = NULL;
728 p->windows = NULL;
729 p->pack_fd = -1;
730 p->pack_local = local;
731 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
732 hashcpy(p->sha1, sha1);
733 return p;
736 struct packed_git *parse_pack_index(unsigned char *sha1)
738 char *path = sha1_pack_index_name(sha1);
739 return parse_pack_index_file(sha1, path);
742 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
744 struct packed_git *p;
745 unsigned long idx_size;
746 void *idx_map;
747 char *path;
749 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
750 return NULL;
752 path = sha1_pack_name(sha1);
754 p = xmalloc(sizeof(*p) + strlen(path) + 2);
755 strcpy(p->pack_name, path);
756 p->index_size = idx_size;
757 p->pack_size = 0;
758 p->index_base = idx_map;
759 p->next = NULL;
760 p->windows = NULL;
761 p->pack_fd = -1;
762 hashcpy(p->sha1, sha1);
763 return p;
766 void install_packed_git(struct packed_git *pack)
768 pack->next = packed_git;
769 packed_git = pack;
772 static void prepare_packed_git_one(char *objdir, int local)
774 char path[PATH_MAX];
775 int len;
776 DIR *dir;
777 struct dirent *de;
779 sprintf(path, "%s/pack", objdir);
780 len = strlen(path);
781 dir = opendir(path);
782 if (!dir) {
783 if (errno != ENOENT)
784 error("unable to open object pack directory: %s: %s",
785 path, strerror(errno));
786 return;
788 path[len++] = '/';
789 while ((de = readdir(dir)) != NULL) {
790 int namelen = strlen(de->d_name);
791 struct packed_git *p;
793 if (!has_extension(de->d_name, ".idx"))
794 continue;
796 /* Don't reopen a pack we already have. */
797 strcpy(path + len, de->d_name);
798 for (p = packed_git; p; p = p->next) {
799 if (!memcmp(path, p->pack_name, len + namelen - 4))
800 break;
802 if (p)
803 continue;
804 /* See if it really is a valid .idx file with corresponding
805 * .pack file that we can map.
807 p = add_packed_git(path, len + namelen, local);
808 if (!p)
809 continue;
810 install_packed_git(p);
812 closedir(dir);
815 static int prepare_packed_git_run_once = 0;
816 void prepare_packed_git(void)
818 struct alternate_object_database *alt;
820 if (prepare_packed_git_run_once)
821 return;
822 prepare_packed_git_one(get_object_directory(), 1);
823 prepare_alt_odb();
824 for (alt = alt_odb_list; alt; alt = alt->next) {
825 alt->name[-1] = 0;
826 prepare_packed_git_one(alt->base, 0);
827 alt->name[-1] = '/';
829 prepare_packed_git_run_once = 1;
832 void reprepare_packed_git(void)
834 prepare_packed_git_run_once = 0;
835 prepare_packed_git();
838 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
840 unsigned char real_sha1[20];
841 hash_sha1_file(map, size, type, real_sha1);
842 return hashcmp(sha1, real_sha1) ? -1 : 0;
845 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
847 struct stat st;
848 void *map;
849 int fd;
850 char *filename = find_sha1_file(sha1, &st);
852 if (!filename) {
853 return NULL;
856 fd = open(filename, O_RDONLY | sha1_file_open_flag);
857 if (fd < 0) {
858 /* See if it works without O_NOATIME */
859 switch (sha1_file_open_flag) {
860 default:
861 fd = open(filename, O_RDONLY);
862 if (fd >= 0)
863 break;
864 /* Fallthrough */
865 case 0:
866 return NULL;
869 /* If it failed once, it will probably fail again.
870 * Stop using O_NOATIME
872 sha1_file_open_flag = 0;
874 map = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
875 close(fd);
876 *size = st.st_size;
877 return map;
880 int legacy_loose_object(unsigned char *map)
882 unsigned int word;
885 * Is it a zlib-compressed buffer? If so, the first byte
886 * must be 0x78 (15-bit window size, deflated), and the
887 * first 16-bit word is evenly divisible by 31
889 word = (map[0] << 8) + map[1];
890 if (map[0] == 0x78 && !(word % 31))
891 return 1;
892 else
893 return 0;
896 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
898 unsigned shift;
899 unsigned char c;
900 unsigned long size;
901 unsigned long used = 0;
903 c = buf[used++];
904 *type = (c >> 4) & 7;
905 size = c & 15;
906 shift = 4;
907 while (c & 0x80) {
908 if (len <= used)
909 return 0;
910 if (sizeof(long) * 8 <= shift)
911 return 0;
912 c = buf[used++];
913 size += (c & 0x7f) << shift;
914 shift += 7;
916 *sizep = size;
917 return used;
920 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
922 unsigned long size, used;
923 static const char valid_loose_object_type[8] = {
924 0, /* OBJ_EXT */
925 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
926 0, /* "delta" and others are invalid in a loose object */
928 enum object_type type;
930 /* Get the data stream */
931 memset(stream, 0, sizeof(*stream));
932 stream->next_in = map;
933 stream->avail_in = mapsize;
934 stream->next_out = buffer;
935 stream->avail_out = bufsiz;
937 if (legacy_loose_object(map)) {
938 inflateInit(stream);
939 return inflate(stream, 0);
942 used = unpack_object_header_gently(map, mapsize, &type, &size);
943 if (!used || !valid_loose_object_type[type])
944 return -1;
945 map += used;
946 mapsize -= used;
948 /* Set up the stream for the rest.. */
949 stream->next_in = map;
950 stream->avail_in = mapsize;
951 inflateInit(stream);
953 /* And generate the fake traditional header */
954 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
955 type_names[type], size);
956 return 0;
959 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
961 int bytes = strlen(buffer) + 1;
962 unsigned char *buf = xmalloc(1+size);
963 unsigned long n;
965 n = stream->total_out - bytes;
966 if (n > size)
967 n = size;
968 memcpy(buf, (char *) buffer + bytes, n);
969 bytes = n;
970 if (bytes < size) {
971 stream->next_out = buf + bytes;
972 stream->avail_out = size - bytes;
973 while (inflate(stream, Z_FINISH) == Z_OK)
974 /* nothing */;
976 buf[size] = 0;
977 inflateEnd(stream);
978 return buf;
982 * We used to just use "sscanf()", but that's actually way
983 * too permissive for what we want to check. So do an anal
984 * object header parse by hand.
986 static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
988 int i;
989 unsigned long size;
992 * The type can be at most ten bytes (including the
993 * terminating '\0' that we add), and is followed by
994 * a space.
996 i = 10;
997 for (;;) {
998 char c = *hdr++;
999 if (c == ' ')
1000 break;
1001 if (!--i)
1002 return -1;
1003 *type++ = c;
1005 *type = 0;
1008 * The length must follow immediately, and be in canonical
1009 * decimal format (ie "010" is not valid).
1011 size = *hdr++ - '0';
1012 if (size > 9)
1013 return -1;
1014 if (size) {
1015 for (;;) {
1016 unsigned long c = *hdr - '0';
1017 if (c > 9)
1018 break;
1019 hdr++;
1020 size = size * 10 + c;
1023 *sizep = size;
1026 * The length must be followed by a zero byte
1028 return *hdr ? -1 : 0;
1031 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
1033 int ret;
1034 z_stream stream;
1035 char hdr[8192];
1037 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1038 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
1039 return NULL;
1041 return unpack_sha1_rest(&stream, hdr, *size);
1044 static unsigned long get_delta_base(struct packed_git *p,
1045 struct pack_window **w_curs,
1046 unsigned long *curpos,
1047 enum object_type kind,
1048 unsigned long delta_obj_offset)
1050 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1051 unsigned long base_offset;
1053 /* use_pack() assured us we have [base_info, base_info + 20)
1054 * as a range that we can look at without walking off the
1055 * end of the mapped window. Its actually the hash size
1056 * that is assured. An OFS_DELTA longer than the hash size
1057 * is stupid, as then a REF_DELTA would be smaller to store.
1059 if (kind == OBJ_OFS_DELTA) {
1060 unsigned used = 0;
1061 unsigned char c = base_info[used++];
1062 base_offset = c & 127;
1063 while (c & 128) {
1064 base_offset += 1;
1065 if (!base_offset || base_offset & ~(~0UL >> 7))
1066 die("offset value overflow for delta base object");
1067 c = base_info[used++];
1068 base_offset = (base_offset << 7) + (c & 127);
1070 base_offset = delta_obj_offset - base_offset;
1071 if (base_offset >= delta_obj_offset)
1072 die("delta base offset out of bound");
1073 *curpos += used;
1074 } else if (kind == OBJ_REF_DELTA) {
1075 /* The base entry _must_ be in the same pack */
1076 base_offset = find_pack_entry_one(base_info, p);
1077 if (!base_offset)
1078 die("failed to find delta-pack base object %s",
1079 sha1_to_hex(base_info));
1080 *curpos += 20;
1081 } else
1082 die("I am totally screwed");
1083 return base_offset;
1086 /* forward declaration for a mutually recursive function */
1087 static int packed_object_info(struct packed_git *p, unsigned long offset,
1088 char *type, unsigned long *sizep);
1090 static int packed_delta_info(struct packed_git *p,
1091 struct pack_window **w_curs,
1092 unsigned long curpos,
1093 enum object_type kind,
1094 unsigned long obj_offset,
1095 char *type,
1096 unsigned long *sizep)
1098 unsigned long base_offset;
1100 base_offset = get_delta_base(p, w_curs, &curpos, kind, obj_offset);
1102 /* We choose to only get the type of the base object and
1103 * ignore potentially corrupt pack file that expects the delta
1104 * based on a base with a wrong size. This saves tons of
1105 * inflate() calls.
1107 if (packed_object_info(p, base_offset, type, NULL))
1108 die("cannot get info for delta-pack base");
1110 if (sizep) {
1111 const unsigned char *data;
1112 unsigned char delta_head[20], *in;
1113 z_stream stream;
1114 int st;
1116 memset(&stream, 0, sizeof(stream));
1117 stream.next_out = delta_head;
1118 stream.avail_out = sizeof(delta_head);
1120 inflateInit(&stream);
1121 do {
1122 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1123 stream.next_in = in;
1124 st = inflate(&stream, Z_FINISH);
1125 curpos += stream.next_in - in;
1126 } while ((st == Z_OK || st == Z_BUF_ERROR)
1127 && stream.total_out < sizeof(delta_head));
1128 inflateEnd(&stream);
1129 if ((st != Z_STREAM_END) &&
1130 stream.total_out != sizeof(delta_head))
1131 die("delta data unpack-initial failed");
1133 /* Examine the initial part of the delta to figure out
1134 * the result size.
1136 data = delta_head;
1138 /* ignore base size */
1139 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1141 /* Read the result size */
1142 *sizep = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1144 return 0;
1147 static int unpack_object_header(struct packed_git *p,
1148 struct pack_window **w_curs,
1149 unsigned long *curpos,
1150 unsigned long *sizep)
1152 unsigned char *base;
1153 unsigned int left;
1154 unsigned long used;
1155 enum object_type type;
1157 /* use_pack() assures us we have [base, base + 20) available
1158 * as a range that we can look at at. (Its actually the hash
1159 * size that is assured.) With our object header encoding
1160 * the maximum deflated object size is 2^137, which is just
1161 * insane, so we know won't exceed what we have been given.
1163 base = use_pack(p, w_curs, *curpos, &left);
1164 used = unpack_object_header_gently(base, left, &type, sizep);
1165 if (!used)
1166 die("object offset outside of pack file");
1167 *curpos += used;
1169 return type;
1172 void packed_object_info_detail(struct packed_git *p,
1173 unsigned long obj_offset,
1174 char *type,
1175 unsigned long *size,
1176 unsigned long *store_size,
1177 unsigned int *delta_chain_length,
1178 unsigned char *base_sha1)
1180 struct pack_window *w_curs = NULL;
1181 unsigned long curpos, dummy;
1182 unsigned char *next_sha1;
1183 enum object_type kind;
1185 *delta_chain_length = 0;
1186 curpos = obj_offset;
1187 kind = unpack_object_header(p, &w_curs, &curpos, size);
1189 for (;;) {
1190 switch (kind) {
1191 default:
1192 die("pack %s contains unknown object type %d",
1193 p->pack_name, kind);
1194 case OBJ_COMMIT:
1195 case OBJ_TREE:
1196 case OBJ_BLOB:
1197 case OBJ_TAG:
1198 strcpy(type, type_names[kind]);
1199 *store_size = 0; /* notyet */
1200 unuse_pack(&w_curs);
1201 return;
1202 case OBJ_OFS_DELTA:
1203 obj_offset = get_delta_base(p, &w_curs, &curpos, kind, obj_offset);
1204 if (*delta_chain_length == 0) {
1205 /* TODO: find base_sha1 as pointed by curpos */
1207 break;
1208 case OBJ_REF_DELTA:
1209 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1210 if (*delta_chain_length == 0)
1211 hashcpy(base_sha1, next_sha1);
1212 obj_offset = find_pack_entry_one(next_sha1, p);
1213 break;
1215 (*delta_chain_length)++;
1216 curpos = obj_offset;
1217 kind = unpack_object_header(p, &w_curs, &curpos, &dummy);
1221 static int packed_object_info(struct packed_git *p, unsigned long obj_offset,
1222 char *type, unsigned long *sizep)
1224 struct pack_window *w_curs = NULL;
1225 unsigned long size, curpos = obj_offset;
1226 enum object_type kind;
1228 kind = unpack_object_header(p, &w_curs, &curpos, &size);
1230 switch (kind) {
1231 case OBJ_OFS_DELTA:
1232 case OBJ_REF_DELTA:
1233 packed_delta_info(p, &w_curs, curpos, kind,
1234 obj_offset, type, sizep);
1235 break;
1236 case OBJ_COMMIT:
1237 case OBJ_TREE:
1238 case OBJ_BLOB:
1239 case OBJ_TAG:
1240 strcpy(type, type_names[kind]);
1241 if (sizep)
1242 *sizep = size;
1243 break;
1244 default:
1245 die("pack %s contains unknown object type %d",
1246 p->pack_name, kind);
1248 unuse_pack(&w_curs);
1249 return 0;
1252 static void *unpack_compressed_entry(struct packed_git *p,
1253 struct pack_window **w_curs,
1254 unsigned long curpos,
1255 unsigned long size)
1257 int st;
1258 z_stream stream;
1259 unsigned char *buffer, *in;
1261 buffer = xmalloc(size + 1);
1262 buffer[size] = 0;
1263 memset(&stream, 0, sizeof(stream));
1264 stream.next_out = buffer;
1265 stream.avail_out = size;
1267 inflateInit(&stream);
1268 do {
1269 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1270 stream.next_in = in;
1271 st = inflate(&stream, Z_FINISH);
1272 curpos += stream.next_in - in;
1273 } while (st == Z_OK || st == Z_BUF_ERROR);
1274 inflateEnd(&stream);
1275 if ((st != Z_STREAM_END) || stream.total_out != size) {
1276 free(buffer);
1277 return NULL;
1280 return buffer;
1283 static void *unpack_delta_entry(struct packed_git *p,
1284 struct pack_window **w_curs,
1285 unsigned long curpos,
1286 unsigned long delta_size,
1287 enum object_type kind,
1288 unsigned long obj_offset,
1289 char *type,
1290 unsigned long *sizep)
1292 void *delta_data, *result, *base;
1293 unsigned long result_size, base_size, base_offset;
1295 base_offset = get_delta_base(p, w_curs, &curpos, kind, obj_offset);
1296 base = unpack_entry(p, base_offset, type, &base_size);
1297 if (!base)
1298 die("failed to read delta base object at %lu from %s",
1299 base_offset, p->pack_name);
1301 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1302 result = patch_delta(base, base_size,
1303 delta_data, delta_size,
1304 &result_size);
1305 if (!result)
1306 die("failed to apply delta");
1307 free(delta_data);
1308 free(base);
1309 *sizep = result_size;
1310 return result;
1313 void *unpack_entry(struct packed_git *p, unsigned long obj_offset,
1314 char *type, unsigned long *sizep)
1316 struct pack_window *w_curs = NULL;
1317 unsigned long size, curpos = obj_offset;
1318 enum object_type kind;
1319 void *retval;
1321 kind = unpack_object_header(p, &w_curs, &curpos, &size);
1322 switch (kind) {
1323 case OBJ_OFS_DELTA:
1324 case OBJ_REF_DELTA:
1325 retval = unpack_delta_entry(p, &w_curs, curpos, size,
1326 kind, obj_offset, type, sizep);
1327 break;
1328 case OBJ_COMMIT:
1329 case OBJ_TREE:
1330 case OBJ_BLOB:
1331 case OBJ_TAG:
1332 strcpy(type, type_names[kind]);
1333 *sizep = size;
1334 retval = unpack_compressed_entry(p, &w_curs, curpos, size);
1335 break;
1336 default:
1337 die("unknown object type %i in %s", kind, p->pack_name);
1339 unuse_pack(&w_curs);
1340 return retval;
1343 int num_packed_objects(const struct packed_git *p)
1345 /* See check_packed_git_idx() */
1346 return (p->index_size - 20 - 20 - 4*256) / 24;
1349 int nth_packed_object_sha1(const struct packed_git *p, int n,
1350 unsigned char* sha1)
1352 void *index = p->index_base + 256;
1353 if (n < 0 || num_packed_objects(p) <= n)
1354 return -1;
1355 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1356 return 0;
1359 unsigned long find_pack_entry_one(const unsigned char *sha1,
1360 struct packed_git *p)
1362 uint32_t *level1_ofs = p->index_base;
1363 int hi = ntohl(level1_ofs[*sha1]);
1364 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1365 void *index = p->index_base + 256;
1367 do {
1368 int mi = (lo + hi) / 2;
1369 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1370 if (!cmp)
1371 return ntohl(*((uint32_t *)((char *)index + (24 * mi))));
1372 if (cmp > 0)
1373 hi = mi;
1374 else
1375 lo = mi+1;
1376 } while (lo < hi);
1377 return 0;
1380 static int matches_pack_name(struct packed_git *p, const char *ig)
1382 const char *last_c, *c;
1384 if (!strcmp(p->pack_name, ig))
1385 return 0;
1387 for (c = p->pack_name, last_c = c; *c;)
1388 if (*c == '/')
1389 last_c = ++c;
1390 else
1391 ++c;
1392 if (!strcmp(last_c, ig))
1393 return 0;
1395 return 1;
1398 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1400 struct packed_git *p;
1401 unsigned long offset;
1403 prepare_packed_git();
1405 for (p = packed_git; p; p = p->next) {
1406 if (ignore_packed) {
1407 const char **ig;
1408 for (ig = ignore_packed; *ig; ig++)
1409 if (!matches_pack_name(p, *ig))
1410 break;
1411 if (*ig)
1412 continue;
1414 offset = find_pack_entry_one(sha1, p);
1415 if (offset) {
1417 * We are about to tell the caller where they can
1418 * locate the requested object. We better make
1419 * sure the packfile is still here and can be
1420 * accessed before supplying that answer, as
1421 * it may have been deleted since the index
1422 * was loaded!
1424 if (p->pack_fd == -1 && open_packed_git(p)) {
1425 error("packfile %s cannot be accessed", p->pack_name);
1426 continue;
1428 e->offset = offset;
1429 e->p = p;
1430 hashcpy(e->sha1, sha1);
1431 return 1;
1434 return 0;
1437 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1438 struct packed_git *packs)
1440 struct packed_git *p;
1442 for (p = packs; p; p = p->next) {
1443 if (find_pack_entry_one(sha1, p))
1444 return p;
1446 return NULL;
1450 static int sha1_loose_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1452 int status;
1453 unsigned long mapsize, size;
1454 void *map;
1455 z_stream stream;
1456 char hdr[32];
1458 map = map_sha1_file(sha1, &mapsize);
1459 if (!map)
1460 return error("unable to find %s", sha1_to_hex(sha1));
1461 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1462 status = error("unable to unpack %s header",
1463 sha1_to_hex(sha1));
1464 if (parse_sha1_header(hdr, type, &size) < 0)
1465 status = error("unable to parse %s header", sha1_to_hex(sha1));
1466 else {
1467 status = 0;
1468 if (sizep)
1469 *sizep = size;
1471 inflateEnd(&stream);
1472 munmap(map, mapsize);
1473 return status;
1476 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1478 struct pack_entry e;
1480 if (!find_pack_entry(sha1, &e, NULL)) {
1481 reprepare_packed_git();
1482 if (!find_pack_entry(sha1, &e, NULL))
1483 return sha1_loose_object_info(sha1, type, sizep);
1485 return packed_object_info(e.p, e.offset, type, sizep);
1488 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1490 struct pack_entry e;
1492 if (!find_pack_entry(sha1, &e, NULL))
1493 return NULL;
1494 else
1495 return unpack_entry(e.p, e.offset, type, size);
1499 * This is meant to hold a *small* number of objects that you would
1500 * want read_sha1_file() to be able to return, but yet you do not want
1501 * to write them into the object store (e.g. a browse-only
1502 * application).
1504 static struct cached_object {
1505 unsigned char sha1[20];
1506 const char *type;
1507 void *buf;
1508 unsigned long size;
1509 } *cached_objects;
1510 static int cached_object_nr, cached_object_alloc;
1512 static struct cached_object *find_cached_object(const unsigned char *sha1)
1514 int i;
1515 struct cached_object *co = cached_objects;
1517 for (i = 0; i < cached_object_nr; i++, co++) {
1518 if (!hashcmp(co->sha1, sha1))
1519 return co;
1521 return NULL;
1524 int pretend_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *sha1)
1526 struct cached_object *co;
1528 hash_sha1_file(buf, len, type, sha1);
1529 if (has_sha1_file(sha1) || find_cached_object(sha1))
1530 return 0;
1531 if (cached_object_alloc <= cached_object_nr) {
1532 cached_object_alloc = alloc_nr(cached_object_alloc);
1533 cached_objects = xrealloc(cached_objects,
1534 sizeof(*cached_objects) *
1535 cached_object_alloc);
1537 co = &cached_objects[cached_object_nr++];
1538 co->size = len;
1539 co->type = strdup(type);
1540 co->buf = xmalloc(len);
1541 memcpy(co->buf, buf, len);
1542 hashcpy(co->sha1, sha1);
1543 return 0;
1546 void *read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1548 unsigned long mapsize;
1549 void *map, *buf;
1550 struct cached_object *co;
1552 co = find_cached_object(sha1);
1553 if (co) {
1554 buf = xmalloc(co->size + 1);
1555 memcpy(buf, co->buf, co->size);
1556 ((char*)buf)[co->size] = 0;
1557 strcpy(type, co->type);
1558 *size = co->size;
1559 return buf;
1562 buf = read_packed_sha1(sha1, type, size);
1563 if (buf)
1564 return buf;
1565 map = map_sha1_file(sha1, &mapsize);
1566 if (map) {
1567 buf = unpack_sha1_file(map, mapsize, type, size);
1568 munmap(map, mapsize);
1569 return buf;
1571 reprepare_packed_git();
1572 return read_packed_sha1(sha1, type, size);
1575 void *read_object_with_reference(const unsigned char *sha1,
1576 const char *required_type,
1577 unsigned long *size,
1578 unsigned char *actual_sha1_return)
1580 char type[20];
1581 void *buffer;
1582 unsigned long isize;
1583 unsigned char actual_sha1[20];
1585 hashcpy(actual_sha1, sha1);
1586 while (1) {
1587 int ref_length = -1;
1588 const char *ref_type = NULL;
1590 buffer = read_sha1_file(actual_sha1, type, &isize);
1591 if (!buffer)
1592 return NULL;
1593 if (!strcmp(type, required_type)) {
1594 *size = isize;
1595 if (actual_sha1_return)
1596 hashcpy(actual_sha1_return, actual_sha1);
1597 return buffer;
1599 /* Handle references */
1600 else if (!strcmp(type, commit_type))
1601 ref_type = "tree ";
1602 else if (!strcmp(type, tag_type))
1603 ref_type = "object ";
1604 else {
1605 free(buffer);
1606 return NULL;
1608 ref_length = strlen(ref_type);
1610 if (memcmp(buffer, ref_type, ref_length) ||
1611 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1612 free(buffer);
1613 return NULL;
1615 free(buffer);
1616 /* Now we have the ID of the referred-to object in
1617 * actual_sha1. Check again. */
1621 static void write_sha1_file_prepare(void *buf, unsigned long len,
1622 const char *type, unsigned char *sha1,
1623 char *hdr, int *hdrlen)
1625 SHA_CTX c;
1627 /* Generate the header */
1628 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
1630 /* Sha1.. */
1631 SHA1_Init(&c);
1632 SHA1_Update(&c, hdr, *hdrlen);
1633 SHA1_Update(&c, buf, len);
1634 SHA1_Final(sha1, &c);
1638 * Link the tempfile to the final place, possibly creating the
1639 * last directory level as you do so.
1641 * Returns the errno on failure, 0 on success.
1643 static int link_temp_to_file(const char *tmpfile, const char *filename)
1645 int ret;
1646 char *dir;
1648 if (!link(tmpfile, filename))
1649 return 0;
1652 * Try to mkdir the last path component if that failed.
1654 * Re-try the "link()" regardless of whether the mkdir
1655 * succeeds, since a race might mean that somebody
1656 * else succeeded.
1658 ret = errno;
1659 dir = strrchr(filename, '/');
1660 if (dir) {
1661 *dir = 0;
1662 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1663 *dir = '/';
1664 return -2;
1666 *dir = '/';
1667 if (!link(tmpfile, filename))
1668 return 0;
1669 ret = errno;
1671 return ret;
1675 * Move the just written object into its final resting place
1677 int move_temp_to_file(const char *tmpfile, const char *filename)
1679 int ret = link_temp_to_file(tmpfile, filename);
1682 * Coda hack - coda doesn't like cross-directory links,
1683 * so we fall back to a rename, which will mean that it
1684 * won't be able to check collisions, but that's not a
1685 * big deal.
1687 * The same holds for FAT formatted media.
1689 * When this succeeds, we just return 0. We have nothing
1690 * left to unlink.
1692 if (ret && ret != EEXIST) {
1693 if (!rename(tmpfile, filename))
1694 return 0;
1695 ret = errno;
1697 unlink(tmpfile);
1698 if (ret) {
1699 if (ret != EEXIST) {
1700 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1702 /* FIXME!!! Collision check here ? */
1705 return 0;
1708 static int write_buffer(int fd, const void *buf, size_t len)
1710 if (write_in_full(fd, buf, len) < 0)
1711 return error("file write error (%s)", strerror(errno));
1712 return 0;
1715 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1717 int hdr_len;
1718 unsigned char c;
1720 c = (type << 4) | (len & 15);
1721 len >>= 4;
1722 hdr_len = 1;
1723 while (len) {
1724 *hdr++ = c | 0x80;
1725 hdr_len++;
1726 c = (len & 0x7f);
1727 len >>= 7;
1729 *hdr = c;
1730 return hdr_len;
1733 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1735 int obj_type, hdrlen;
1737 if (use_legacy_headers) {
1738 while (deflate(stream, 0) == Z_OK)
1739 /* nothing */;
1740 return;
1742 if (!strcmp(type, blob_type))
1743 obj_type = OBJ_BLOB;
1744 else if (!strcmp(type, tree_type))
1745 obj_type = OBJ_TREE;
1746 else if (!strcmp(type, commit_type))
1747 obj_type = OBJ_COMMIT;
1748 else if (!strcmp(type, tag_type))
1749 obj_type = OBJ_TAG;
1750 else
1751 die("trying to generate bogus object of type '%s'", type);
1752 hdrlen = write_binary_header(stream->next_out, obj_type, len);
1753 stream->total_out = hdrlen;
1754 stream->next_out += hdrlen;
1755 stream->avail_out -= hdrlen;
1758 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1759 unsigned char *sha1)
1761 char hdr[32];
1762 int hdrlen;
1763 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1764 return 0;
1767 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1769 int size;
1770 unsigned char *compressed;
1771 z_stream stream;
1772 unsigned char sha1[20];
1773 char *filename;
1774 static char tmpfile[PATH_MAX];
1775 char hdr[32];
1776 int fd, hdrlen;
1778 /* Normally if we have it in the pack then we do not bother writing
1779 * it out into .git/objects/??/?{38} file.
1781 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1782 filename = sha1_file_name(sha1);
1783 if (returnsha1)
1784 hashcpy(returnsha1, sha1);
1785 if (has_sha1_file(sha1))
1786 return 0;
1787 fd = open(filename, O_RDONLY);
1788 if (fd >= 0) {
1790 * FIXME!!! We might do collision checking here, but we'd
1791 * need to uncompress the old file and check it. Later.
1793 close(fd);
1794 return 0;
1797 if (errno != ENOENT) {
1798 return error("sha1 file %s: %s\n", filename, strerror(errno));
1801 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1803 fd = mkstemp(tmpfile);
1804 if (fd < 0) {
1805 if (errno == EPERM)
1806 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1807 else
1808 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1811 /* Set it up */
1812 memset(&stream, 0, sizeof(stream));
1813 deflateInit(&stream, zlib_compression_level);
1814 size = 8 + deflateBound(&stream, len+hdrlen);
1815 compressed = xmalloc(size);
1817 /* Compress it */
1818 stream.next_out = compressed;
1819 stream.avail_out = size;
1821 /* First header.. */
1822 stream.next_in = (unsigned char *)hdr;
1823 stream.avail_in = hdrlen;
1824 setup_object_header(&stream, type, len);
1826 /* Then the data itself.. */
1827 stream.next_in = buf;
1828 stream.avail_in = len;
1829 while (deflate(&stream, Z_FINISH) == Z_OK)
1830 /* nothing */;
1831 deflateEnd(&stream);
1832 size = stream.total_out;
1834 if (write_buffer(fd, compressed, size) < 0)
1835 die("unable to write sha1 file");
1836 fchmod(fd, 0444);
1837 close(fd);
1838 free(compressed);
1840 return move_temp_to_file(tmpfile, filename);
1844 * We need to unpack and recompress the object for writing
1845 * it out to a different file.
1847 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1849 size_t size;
1850 z_stream stream;
1851 unsigned char *unpacked;
1852 unsigned long len;
1853 char type[20];
1854 char hdr[32];
1855 int hdrlen;
1856 void *buf;
1858 /* need to unpack and recompress it by itself */
1859 unpacked = read_packed_sha1(sha1, type, &len);
1860 if (!unpacked)
1861 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1863 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1865 /* Set it up */
1866 memset(&stream, 0, sizeof(stream));
1867 deflateInit(&stream, zlib_compression_level);
1868 size = deflateBound(&stream, len + hdrlen);
1869 buf = xmalloc(size);
1871 /* Compress it */
1872 stream.next_out = buf;
1873 stream.avail_out = size;
1875 /* First header.. */
1876 stream.next_in = (void *)hdr;
1877 stream.avail_in = hdrlen;
1878 while (deflate(&stream, 0) == Z_OK)
1879 /* nothing */;
1881 /* Then the data itself.. */
1882 stream.next_in = unpacked;
1883 stream.avail_in = len;
1884 while (deflate(&stream, Z_FINISH) == Z_OK)
1885 /* nothing */;
1886 deflateEnd(&stream);
1887 free(unpacked);
1889 *objsize = stream.total_out;
1890 return buf;
1893 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1895 int retval;
1896 unsigned long objsize;
1897 void *buf = map_sha1_file(sha1, &objsize);
1899 if (buf) {
1900 retval = write_buffer(fd, buf, objsize);
1901 munmap(buf, objsize);
1902 return retval;
1905 buf = repack_object(sha1, &objsize);
1906 retval = write_buffer(fd, buf, objsize);
1907 free(buf);
1908 return retval;
1911 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1912 size_t bufsize, size_t *bufposn)
1914 char tmpfile[PATH_MAX];
1915 int local;
1916 z_stream stream;
1917 unsigned char real_sha1[20];
1918 unsigned char discard[4096];
1919 int ret;
1920 SHA_CTX c;
1922 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1924 local = mkstemp(tmpfile);
1925 if (local < 0) {
1926 if (errno == EPERM)
1927 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1928 else
1929 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1932 memset(&stream, 0, sizeof(stream));
1934 inflateInit(&stream);
1936 SHA1_Init(&c);
1938 do {
1939 ssize_t size;
1940 if (*bufposn) {
1941 stream.avail_in = *bufposn;
1942 stream.next_in = (unsigned char *) buffer;
1943 do {
1944 stream.next_out = discard;
1945 stream.avail_out = sizeof(discard);
1946 ret = inflate(&stream, Z_SYNC_FLUSH);
1947 SHA1_Update(&c, discard, sizeof(discard) -
1948 stream.avail_out);
1949 } while (stream.avail_in && ret == Z_OK);
1950 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1951 die("unable to write sha1 file");
1952 memmove(buffer, buffer + *bufposn - stream.avail_in,
1953 stream.avail_in);
1954 *bufposn = stream.avail_in;
1955 if (ret != Z_OK)
1956 break;
1958 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
1959 if (size <= 0) {
1960 close(local);
1961 unlink(tmpfile);
1962 if (!size)
1963 return error("Connection closed?");
1964 perror("Reading from connection");
1965 return -1;
1967 *bufposn += size;
1968 } while (1);
1969 inflateEnd(&stream);
1971 close(local);
1972 SHA1_Final(real_sha1, &c);
1973 if (ret != Z_STREAM_END) {
1974 unlink(tmpfile);
1975 return error("File %s corrupted", sha1_to_hex(sha1));
1977 if (hashcmp(sha1, real_sha1)) {
1978 unlink(tmpfile);
1979 return error("File %s has bad hash", sha1_to_hex(sha1));
1982 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1985 int has_pack_index(const unsigned char *sha1)
1987 struct stat st;
1988 if (stat(sha1_pack_index_name(sha1), &st))
1989 return 0;
1990 return 1;
1993 int has_pack_file(const unsigned char *sha1)
1995 struct stat st;
1996 if (stat(sha1_pack_name(sha1), &st))
1997 return 0;
1998 return 1;
2001 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2003 struct pack_entry e;
2004 return find_pack_entry(sha1, &e, ignore_packed);
2007 int has_sha1_file(const unsigned char *sha1)
2009 struct stat st;
2010 struct pack_entry e;
2012 if (find_pack_entry(sha1, &e, NULL))
2013 return 1;
2014 return find_sha1_file(sha1, &st) ? 1 : 0;
2018 * reads from fd as long as possible into a supplied buffer of size bytes.
2019 * If necessary the buffer's size is increased using realloc()
2021 * returns 0 if anything went fine and -1 otherwise
2023 * NOTE: both buf and size may change, but even when -1 is returned
2024 * you still have to free() it yourself.
2026 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
2028 char* buf = *return_buf;
2029 unsigned long size = *return_size;
2030 int iret;
2031 unsigned long off = 0;
2033 do {
2034 iret = xread(fd, buf + off, size - off);
2035 if (iret > 0) {
2036 off += iret;
2037 if (off == size) {
2038 size *= 2;
2039 buf = xrealloc(buf, size);
2042 } while (iret > 0);
2044 *return_buf = buf;
2045 *return_size = off;
2047 if (iret < 0)
2048 return -1;
2049 return 0;
2052 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2054 unsigned long size = 4096;
2055 char *buf = xmalloc(size);
2056 int ret;
2058 if (read_pipe(fd, &buf, &size)) {
2059 free(buf);
2060 return -1;
2063 if (!type)
2064 type = blob_type;
2065 if (write_object)
2066 ret = write_sha1_file(buf, size, type, sha1);
2067 else
2068 ret = hash_sha1_file(buf, size, type, sha1);
2069 free(buf);
2070 return ret;
2073 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
2075 unsigned long size = st->st_size;
2076 void *buf;
2077 int ret, re_allocated = 0;
2079 buf = "";
2080 if (size)
2081 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2082 close(fd);
2084 if (!type)
2085 type = blob_type;
2088 * Convert blobs to git internal format
2090 if (!strcmp(type, blob_type)) {
2091 unsigned long nsize = size;
2092 char *nbuf = buf;
2093 if (convert_to_git(NULL, &nbuf, &nsize)) {
2094 if (size)
2095 munmap(buf, size);
2096 size = nsize;
2097 buf = nbuf;
2098 re_allocated = 1;
2102 if (write_object)
2103 ret = write_sha1_file(buf, size, type, sha1);
2104 else
2105 ret = hash_sha1_file(buf, size, type, sha1);
2106 if (re_allocated) {
2107 free(buf);
2108 return ret;
2110 if (size)
2111 munmap(buf, size);
2112 return ret;
2115 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2117 int fd;
2118 char *target;
2120 switch (st->st_mode & S_IFMT) {
2121 case S_IFREG:
2122 fd = open(path, O_RDONLY);
2123 if (fd < 0)
2124 return error("open(\"%s\"): %s", path,
2125 strerror(errno));
2126 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
2127 return error("%s: failed to insert into database",
2128 path);
2129 break;
2130 case S_IFLNK:
2131 target = xmalloc(st->st_size+1);
2132 if (readlink(path, target, st->st_size+1) != st->st_size) {
2133 char *errstr = strerror(errno);
2134 free(target);
2135 return error("readlink(\"%s\"): %s", path,
2136 errstr);
2138 if (!write_object)
2139 hash_sha1_file(target, st->st_size, blob_type, sha1);
2140 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
2141 return error("%s: failed to insert into database",
2142 path);
2143 free(target);
2144 break;
2145 default:
2146 return error("%s: unsupported file type", path);
2148 return 0;
2151 int read_pack_header(int fd, struct pack_header *header)
2153 char *c = (char*)header;
2154 ssize_t remaining = sizeof(struct pack_header);
2155 do {
2156 ssize_t r = xread(fd, c, remaining);
2157 if (r <= 0)
2158 /* "eof before pack header was fully read" */
2159 return PH_ERROR_EOF;
2160 remaining -= r;
2161 c += r;
2162 } while (remaining > 0);
2163 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2164 /* "protocol error (pack signature mismatch detected)" */
2165 return PH_ERROR_PACK_SIGNATURE;
2166 if (!pack_version_ok(header->hdr_version))
2167 /* "protocol error (pack version unsupported)" */
2168 return PH_ERROR_PROTOCOL;
2169 return 0;