unpack_sha1_file(): detect corrupt loose object files.
[git/trast.git] / sha1_file.c
blobac6b5e00b6dec913a39cc54e84829dc3fc6c782f
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 typename(type), size);
956 return 0;
959 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
961 int bytes = strlen(buffer) + 1;
962 unsigned char *buf = xmalloc(1+size);
963 unsigned long n;
964 int status = Z_OK;
966 n = stream->total_out - bytes;
967 if (n > size)
968 n = size;
969 memcpy(buf, (char *) buffer + bytes, n);
970 bytes = n;
971 if (bytes < size) {
972 stream->next_out = buf + bytes;
973 stream->avail_out = size - bytes;
974 while (status == Z_OK)
975 status = inflate(stream, Z_FINISH);
977 buf[size] = 0;
978 if ((status == Z_OK || status == Z_STREAM_END) && !stream->avail_in) {
979 inflateEnd(stream);
980 return buf;
983 if (status < 0)
984 error("corrupt loose object '%s'", sha1_to_hex(sha1));
985 else if (stream->avail_in)
986 error("garbage at end of loose object '%s'",
987 sha1_to_hex(sha1));
988 free(buf);
989 return NULL;
993 * We used to just use "sscanf()", but that's actually way
994 * too permissive for what we want to check. So do an anal
995 * object header parse by hand.
997 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
999 char type[10];
1000 int i;
1001 unsigned long size;
1004 * The type can be at most ten bytes (including the
1005 * terminating '\0' that we add), and is followed by
1006 * a space.
1008 i = 0;
1009 for (;;) {
1010 char c = *hdr++;
1011 if (c == ' ')
1012 break;
1013 type[i++] = c;
1014 if (i >= sizeof(type))
1015 return -1;
1017 type[i] = 0;
1020 * The length must follow immediately, and be in canonical
1021 * decimal format (ie "010" is not valid).
1023 size = *hdr++ - '0';
1024 if (size > 9)
1025 return -1;
1026 if (size) {
1027 for (;;) {
1028 unsigned long c = *hdr - '0';
1029 if (c > 9)
1030 break;
1031 hdr++;
1032 size = size * 10 + c;
1035 *sizep = size;
1038 * The length must be followed by a zero byte
1040 return *hdr ? -1 : type_from_string(type);
1043 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1045 int ret;
1046 z_stream stream;
1047 char hdr[8192];
1049 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1050 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1051 return NULL;
1053 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1056 static unsigned long get_delta_base(struct packed_git *p,
1057 struct pack_window **w_curs,
1058 unsigned long *curpos,
1059 enum object_type type,
1060 unsigned long delta_obj_offset)
1062 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1063 unsigned long base_offset;
1065 /* use_pack() assured us we have [base_info, base_info + 20)
1066 * as a range that we can look at without walking off the
1067 * end of the mapped window. Its actually the hash size
1068 * that is assured. An OFS_DELTA longer than the hash size
1069 * is stupid, as then a REF_DELTA would be smaller to store.
1071 if (type == OBJ_OFS_DELTA) {
1072 unsigned used = 0;
1073 unsigned char c = base_info[used++];
1074 base_offset = c & 127;
1075 while (c & 128) {
1076 base_offset += 1;
1077 if (!base_offset || base_offset & ~(~0UL >> 7))
1078 die("offset value overflow for delta base object");
1079 c = base_info[used++];
1080 base_offset = (base_offset << 7) + (c & 127);
1082 base_offset = delta_obj_offset - base_offset;
1083 if (base_offset >= delta_obj_offset)
1084 die("delta base offset out of bound");
1085 *curpos += used;
1086 } else if (type == OBJ_REF_DELTA) {
1087 /* The base entry _must_ be in the same pack */
1088 base_offset = find_pack_entry_one(base_info, p);
1089 if (!base_offset)
1090 die("failed to find delta-pack base object %s",
1091 sha1_to_hex(base_info));
1092 *curpos += 20;
1093 } else
1094 die("I am totally screwed");
1095 return base_offset;
1098 /* forward declaration for a mutually recursive function */
1099 static int packed_object_info(struct packed_git *p, unsigned long offset,
1100 unsigned long *sizep);
1102 static int packed_delta_info(struct packed_git *p,
1103 struct pack_window **w_curs,
1104 unsigned long curpos,
1105 enum object_type type,
1106 unsigned long obj_offset,
1107 unsigned long *sizep)
1109 unsigned long base_offset;
1111 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1112 type = packed_object_info(p, base_offset, NULL);
1114 /* We choose to only get the type of the base object and
1115 * ignore potentially corrupt pack file that expects the delta
1116 * based on a base with a wrong size. This saves tons of
1117 * inflate() calls.
1119 if (sizep) {
1120 const unsigned char *data;
1121 unsigned char delta_head[20], *in;
1122 z_stream stream;
1123 int st;
1125 memset(&stream, 0, sizeof(stream));
1126 stream.next_out = delta_head;
1127 stream.avail_out = sizeof(delta_head);
1129 inflateInit(&stream);
1130 do {
1131 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1132 stream.next_in = in;
1133 st = inflate(&stream, Z_FINISH);
1134 curpos += stream.next_in - in;
1135 } while ((st == Z_OK || st == Z_BUF_ERROR)
1136 && stream.total_out < sizeof(delta_head));
1137 inflateEnd(&stream);
1138 if ((st != Z_STREAM_END) &&
1139 stream.total_out != sizeof(delta_head))
1140 die("delta data unpack-initial failed");
1142 /* Examine the initial part of the delta to figure out
1143 * the result size.
1145 data = delta_head;
1147 /* ignore base size */
1148 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1150 /* Read the result size */
1151 *sizep = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1154 return type;
1157 static int unpack_object_header(struct packed_git *p,
1158 struct pack_window **w_curs,
1159 unsigned long *curpos,
1160 unsigned long *sizep)
1162 unsigned char *base;
1163 unsigned int left;
1164 unsigned long used;
1165 enum object_type type;
1167 /* use_pack() assures us we have [base, base + 20) available
1168 * as a range that we can look at at. (Its actually the hash
1169 * size that is assured.) With our object header encoding
1170 * the maximum deflated object size is 2^137, which is just
1171 * insane, so we know won't exceed what we have been given.
1173 base = use_pack(p, w_curs, *curpos, &left);
1174 used = unpack_object_header_gently(base, left, &type, sizep);
1175 if (!used)
1176 die("object offset outside of pack file");
1177 *curpos += used;
1179 return type;
1182 const char *packed_object_info_detail(struct packed_git *p,
1183 unsigned long obj_offset,
1184 unsigned long *size,
1185 unsigned long *store_size,
1186 unsigned int *delta_chain_length,
1187 unsigned char *base_sha1)
1189 struct pack_window *w_curs = NULL;
1190 unsigned long curpos, dummy;
1191 unsigned char *next_sha1;
1192 enum object_type type;
1194 *delta_chain_length = 0;
1195 curpos = obj_offset;
1196 type = unpack_object_header(p, &w_curs, &curpos, size);
1198 for (;;) {
1199 switch (type) {
1200 default:
1201 die("pack %s contains unknown object type %d",
1202 p->pack_name, type);
1203 case OBJ_COMMIT:
1204 case OBJ_TREE:
1205 case OBJ_BLOB:
1206 case OBJ_TAG:
1207 *store_size = 0; /* notyet */
1208 unuse_pack(&w_curs);
1209 return typename(type);
1210 case OBJ_OFS_DELTA:
1211 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1212 if (*delta_chain_length == 0) {
1213 /* TODO: find base_sha1 as pointed by curpos */
1215 break;
1216 case OBJ_REF_DELTA:
1217 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1218 if (*delta_chain_length == 0)
1219 hashcpy(base_sha1, next_sha1);
1220 obj_offset = find_pack_entry_one(next_sha1, p);
1221 break;
1223 (*delta_chain_length)++;
1224 curpos = obj_offset;
1225 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1229 static int packed_object_info(struct packed_git *p, unsigned long obj_offset,
1230 unsigned long *sizep)
1232 struct pack_window *w_curs = NULL;
1233 unsigned long size, curpos = obj_offset;
1234 enum object_type type;
1236 type = unpack_object_header(p, &w_curs, &curpos, &size);
1238 switch (type) {
1239 case OBJ_OFS_DELTA:
1240 case OBJ_REF_DELTA:
1241 type = packed_delta_info(p, &w_curs, curpos,
1242 type, obj_offset, sizep);
1243 break;
1244 case OBJ_COMMIT:
1245 case OBJ_TREE:
1246 case OBJ_BLOB:
1247 case OBJ_TAG:
1248 if (sizep)
1249 *sizep = size;
1250 break;
1251 default:
1252 die("pack %s contains unknown object type %d",
1253 p->pack_name, type);
1255 unuse_pack(&w_curs);
1256 return type;
1259 static void *unpack_compressed_entry(struct packed_git *p,
1260 struct pack_window **w_curs,
1261 unsigned long curpos,
1262 unsigned long size)
1264 int st;
1265 z_stream stream;
1266 unsigned char *buffer, *in;
1268 buffer = xmalloc(size + 1);
1269 buffer[size] = 0;
1270 memset(&stream, 0, sizeof(stream));
1271 stream.next_out = buffer;
1272 stream.avail_out = size;
1274 inflateInit(&stream);
1275 do {
1276 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1277 stream.next_in = in;
1278 st = inflate(&stream, Z_FINISH);
1279 curpos += stream.next_in - in;
1280 } while (st == Z_OK || st == Z_BUF_ERROR);
1281 inflateEnd(&stream);
1282 if ((st != Z_STREAM_END) || stream.total_out != size) {
1283 free(buffer);
1284 return NULL;
1287 return buffer;
1290 static void *unpack_delta_entry(struct packed_git *p,
1291 struct pack_window **w_curs,
1292 unsigned long curpos,
1293 unsigned long delta_size,
1294 unsigned long obj_offset,
1295 enum object_type *type,
1296 unsigned long *sizep)
1298 void *delta_data, *result, *base;
1299 unsigned long base_size, base_offset;
1301 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1302 base = unpack_entry(p, base_offset, type, &base_size);
1303 if (!base)
1304 die("failed to read delta base object at %lu from %s",
1305 base_offset, p->pack_name);
1307 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1308 result = patch_delta(base, base_size,
1309 delta_data, delta_size,
1310 sizep);
1311 if (!result)
1312 die("failed to apply delta");
1313 free(delta_data);
1314 free(base);
1315 return result;
1318 void *unpack_entry(struct packed_git *p, unsigned long obj_offset,
1319 enum object_type *type, unsigned long *sizep)
1321 struct pack_window *w_curs = NULL;
1322 unsigned long curpos = obj_offset;
1323 void *data;
1325 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1326 switch (*type) {
1327 case OBJ_OFS_DELTA:
1328 case OBJ_REF_DELTA:
1329 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1330 obj_offset, type, sizep);
1331 break;
1332 case OBJ_COMMIT:
1333 case OBJ_TREE:
1334 case OBJ_BLOB:
1335 case OBJ_TAG:
1336 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1337 break;
1338 default:
1339 die("unknown object type %i in %s", *type, p->pack_name);
1341 unuse_pack(&w_curs);
1342 return data;
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) {
1419 * We are about to tell the caller where they can
1420 * locate the requested object. We better make
1421 * sure the packfile is still here and can be
1422 * accessed before supplying that answer, as
1423 * it may have been deleted since the index
1424 * was loaded!
1426 if (p->pack_fd == -1 && open_packed_git(p)) {
1427 error("packfile %s cannot be accessed", p->pack_name);
1428 continue;
1430 e->offset = offset;
1431 e->p = p;
1432 hashcpy(e->sha1, sha1);
1433 return 1;
1436 return 0;
1439 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1440 struct packed_git *packs)
1442 struct packed_git *p;
1444 for (p = packs; p; p = p->next) {
1445 if (find_pack_entry_one(sha1, p))
1446 return p;
1448 return NULL;
1452 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1454 int status;
1455 unsigned long mapsize, size;
1456 void *map;
1457 z_stream stream;
1458 char hdr[32];
1460 map = map_sha1_file(sha1, &mapsize);
1461 if (!map)
1462 return error("unable to find %s", sha1_to_hex(sha1));
1463 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1464 status = error("unable to unpack %s header",
1465 sha1_to_hex(sha1));
1466 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1467 status = error("unable to parse %s header", sha1_to_hex(sha1));
1468 else if (sizep)
1469 *sizep = size;
1470 inflateEnd(&stream);
1471 munmap(map, mapsize);
1472 return status;
1475 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1477 struct pack_entry e;
1479 if (!find_pack_entry(sha1, &e, NULL)) {
1480 reprepare_packed_git();
1481 if (!find_pack_entry(sha1, &e, NULL))
1482 return sha1_loose_object_info(sha1, sizep);
1484 return packed_object_info(e.p, e.offset, sizep);
1487 static void *read_packed_sha1(const unsigned char *sha1,
1488 enum object_type *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 enum object_type 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, enum object_type type,
1525 unsigned char *sha1)
1527 struct cached_object *co;
1529 hash_sha1_file(buf, len, typename(type), sha1);
1530 if (has_sha1_file(sha1) || find_cached_object(sha1))
1531 return 0;
1532 if (cached_object_alloc <= cached_object_nr) {
1533 cached_object_alloc = alloc_nr(cached_object_alloc);
1534 cached_objects = xrealloc(cached_objects,
1535 sizeof(*cached_objects) *
1536 cached_object_alloc);
1538 co = &cached_objects[cached_object_nr++];
1539 co->size = len;
1540 co->type = type;
1541 co->buf = xmalloc(len);
1542 memcpy(co->buf, buf, len);
1543 hashcpy(co->sha1, sha1);
1544 return 0;
1547 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1548 unsigned long *size)
1550 unsigned long mapsize;
1551 void *map, *buf;
1552 struct cached_object *co;
1554 co = find_cached_object(sha1);
1555 if (co) {
1556 buf = xmalloc(co->size + 1);
1557 memcpy(buf, co->buf, co->size);
1558 ((char*)buf)[co->size] = 0;
1559 *type = co->type;
1560 *size = co->size;
1561 return buf;
1564 buf = read_packed_sha1(sha1, type, size);
1565 if (buf)
1566 return buf;
1567 map = map_sha1_file(sha1, &mapsize);
1568 if (map) {
1569 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1570 munmap(map, mapsize);
1571 return buf;
1573 reprepare_packed_git();
1574 return read_packed_sha1(sha1, type, size);
1577 void *read_object_with_reference(const unsigned char *sha1,
1578 const char *required_type_name,
1579 unsigned long *size,
1580 unsigned char *actual_sha1_return)
1582 enum object_type type, required_type;
1583 void *buffer;
1584 unsigned long isize;
1585 unsigned char actual_sha1[20];
1587 required_type = type_from_string(required_type_name);
1588 hashcpy(actual_sha1, sha1);
1589 while (1) {
1590 int ref_length = -1;
1591 const char *ref_type = NULL;
1593 buffer = read_sha1_file(actual_sha1, &type, &isize);
1594 if (!buffer)
1595 return NULL;
1596 if (type == required_type) {
1597 *size = isize;
1598 if (actual_sha1_return)
1599 hashcpy(actual_sha1_return, actual_sha1);
1600 return buffer;
1602 /* Handle references */
1603 else if (type == OBJ_COMMIT)
1604 ref_type = "tree ";
1605 else if (type == OBJ_TAG)
1606 ref_type = "object ";
1607 else {
1608 free(buffer);
1609 return NULL;
1611 ref_length = strlen(ref_type);
1613 if (memcmp(buffer, ref_type, ref_length) ||
1614 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1615 free(buffer);
1616 return NULL;
1618 free(buffer);
1619 /* Now we have the ID of the referred-to object in
1620 * actual_sha1. Check again. */
1624 static void write_sha1_file_prepare(void *buf, unsigned long len,
1625 const char *type, unsigned char *sha1,
1626 char *hdr, int *hdrlen)
1628 SHA_CTX c;
1630 /* Generate the header */
1631 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
1633 /* Sha1.. */
1634 SHA1_Init(&c);
1635 SHA1_Update(&c, hdr, *hdrlen);
1636 SHA1_Update(&c, buf, len);
1637 SHA1_Final(sha1, &c);
1641 * Link the tempfile to the final place, possibly creating the
1642 * last directory level as you do so.
1644 * Returns the errno on failure, 0 on success.
1646 static int link_temp_to_file(const char *tmpfile, const char *filename)
1648 int ret;
1649 char *dir;
1651 if (!link(tmpfile, filename))
1652 return 0;
1655 * Try to mkdir the last path component if that failed.
1657 * Re-try the "link()" regardless of whether the mkdir
1658 * succeeds, since a race might mean that somebody
1659 * else succeeded.
1661 ret = errno;
1662 dir = strrchr(filename, '/');
1663 if (dir) {
1664 *dir = 0;
1665 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1666 *dir = '/';
1667 return -2;
1669 *dir = '/';
1670 if (!link(tmpfile, filename))
1671 return 0;
1672 ret = errno;
1674 return ret;
1678 * Move the just written object into its final resting place
1680 int move_temp_to_file(const char *tmpfile, const char *filename)
1682 int ret = link_temp_to_file(tmpfile, filename);
1685 * Coda hack - coda doesn't like cross-directory links,
1686 * so we fall back to a rename, which will mean that it
1687 * won't be able to check collisions, but that's not a
1688 * big deal.
1690 * The same holds for FAT formatted media.
1692 * When this succeeds, we just return 0. We have nothing
1693 * left to unlink.
1695 if (ret && ret != EEXIST) {
1696 if (!rename(tmpfile, filename))
1697 return 0;
1698 ret = errno;
1700 unlink(tmpfile);
1701 if (ret) {
1702 if (ret != EEXIST) {
1703 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1705 /* FIXME!!! Collision check here ? */
1708 return 0;
1711 static int write_buffer(int fd, const void *buf, size_t len)
1713 if (write_in_full(fd, buf, len) < 0)
1714 return error("file write error (%s)", strerror(errno));
1715 return 0;
1718 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1720 int hdr_len;
1721 unsigned char c;
1723 c = (type << 4) | (len & 15);
1724 len >>= 4;
1725 hdr_len = 1;
1726 while (len) {
1727 *hdr++ = c | 0x80;
1728 hdr_len++;
1729 c = (len & 0x7f);
1730 len >>= 7;
1732 *hdr = c;
1733 return hdr_len;
1736 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1738 int obj_type, hdrlen;
1740 if (use_legacy_headers) {
1741 while (deflate(stream, 0) == Z_OK)
1742 /* nothing */;
1743 return;
1745 obj_type = type_from_string(type);
1746 hdrlen = write_binary_header(stream->next_out, obj_type, len);
1747 stream->total_out = hdrlen;
1748 stream->next_out += hdrlen;
1749 stream->avail_out -= hdrlen;
1752 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1753 unsigned char *sha1)
1755 char hdr[32];
1756 int hdrlen;
1757 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1758 return 0;
1761 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1763 int size;
1764 unsigned char *compressed;
1765 z_stream stream;
1766 unsigned char sha1[20];
1767 char *filename;
1768 static char tmpfile[PATH_MAX];
1769 char hdr[32];
1770 int fd, hdrlen;
1772 /* Normally if we have it in the pack then we do not bother writing
1773 * it out into .git/objects/??/?{38} file.
1775 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1776 filename = sha1_file_name(sha1);
1777 if (returnsha1)
1778 hashcpy(returnsha1, sha1);
1779 if (has_sha1_file(sha1))
1780 return 0;
1781 fd = open(filename, O_RDONLY);
1782 if (fd >= 0) {
1784 * FIXME!!! We might do collision checking here, but we'd
1785 * need to uncompress the old file and check it. Later.
1787 close(fd);
1788 return 0;
1791 if (errno != ENOENT) {
1792 return error("sha1 file %s: %s\n", filename, strerror(errno));
1795 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1797 fd = mkstemp(tmpfile);
1798 if (fd < 0) {
1799 if (errno == EPERM)
1800 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1801 else
1802 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1805 /* Set it up */
1806 memset(&stream, 0, sizeof(stream));
1807 deflateInit(&stream, zlib_compression_level);
1808 size = 8 + deflateBound(&stream, len+hdrlen);
1809 compressed = xmalloc(size);
1811 /* Compress it */
1812 stream.next_out = compressed;
1813 stream.avail_out = size;
1815 /* First header.. */
1816 stream.next_in = (unsigned char *)hdr;
1817 stream.avail_in = hdrlen;
1818 setup_object_header(&stream, type, len);
1820 /* Then the data itself.. */
1821 stream.next_in = buf;
1822 stream.avail_in = len;
1823 while (deflate(&stream, Z_FINISH) == Z_OK)
1824 /* nothing */;
1825 deflateEnd(&stream);
1826 size = stream.total_out;
1828 if (write_buffer(fd, compressed, size) < 0)
1829 die("unable to write sha1 file");
1830 fchmod(fd, 0444);
1831 close(fd);
1832 free(compressed);
1834 return move_temp_to_file(tmpfile, filename);
1838 * We need to unpack and recompress the object for writing
1839 * it out to a different file.
1841 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1843 size_t size;
1844 z_stream stream;
1845 unsigned char *unpacked;
1846 unsigned long len;
1847 enum object_type type;
1848 char hdr[32];
1849 int hdrlen;
1850 void *buf;
1852 /* need to unpack and recompress it by itself */
1853 unpacked = read_packed_sha1(sha1, &type, &len);
1854 if (!unpacked)
1855 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1857 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1859 /* Set it up */
1860 memset(&stream, 0, sizeof(stream));
1861 deflateInit(&stream, zlib_compression_level);
1862 size = deflateBound(&stream, len + hdrlen);
1863 buf = xmalloc(size);
1865 /* Compress it */
1866 stream.next_out = buf;
1867 stream.avail_out = size;
1869 /* First header.. */
1870 stream.next_in = (void *)hdr;
1871 stream.avail_in = hdrlen;
1872 while (deflate(&stream, 0) == Z_OK)
1873 /* nothing */;
1875 /* Then the data itself.. */
1876 stream.next_in = unpacked;
1877 stream.avail_in = len;
1878 while (deflate(&stream, Z_FINISH) == Z_OK)
1879 /* nothing */;
1880 deflateEnd(&stream);
1881 free(unpacked);
1883 *objsize = stream.total_out;
1884 return buf;
1887 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1889 int retval;
1890 unsigned long objsize;
1891 void *buf = map_sha1_file(sha1, &objsize);
1893 if (buf) {
1894 retval = write_buffer(fd, buf, objsize);
1895 munmap(buf, objsize);
1896 return retval;
1899 buf = repack_object(sha1, &objsize);
1900 retval = write_buffer(fd, buf, objsize);
1901 free(buf);
1902 return retval;
1905 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1906 size_t bufsize, size_t *bufposn)
1908 char tmpfile[PATH_MAX];
1909 int local;
1910 z_stream stream;
1911 unsigned char real_sha1[20];
1912 unsigned char discard[4096];
1913 int ret;
1914 SHA_CTX c;
1916 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1918 local = mkstemp(tmpfile);
1919 if (local < 0) {
1920 if (errno == EPERM)
1921 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1922 else
1923 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1926 memset(&stream, 0, sizeof(stream));
1928 inflateInit(&stream);
1930 SHA1_Init(&c);
1932 do {
1933 ssize_t size;
1934 if (*bufposn) {
1935 stream.avail_in = *bufposn;
1936 stream.next_in = (unsigned char *) buffer;
1937 do {
1938 stream.next_out = discard;
1939 stream.avail_out = sizeof(discard);
1940 ret = inflate(&stream, Z_SYNC_FLUSH);
1941 SHA1_Update(&c, discard, sizeof(discard) -
1942 stream.avail_out);
1943 } while (stream.avail_in && ret == Z_OK);
1944 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1945 die("unable to write sha1 file");
1946 memmove(buffer, buffer + *bufposn - stream.avail_in,
1947 stream.avail_in);
1948 *bufposn = stream.avail_in;
1949 if (ret != Z_OK)
1950 break;
1952 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
1953 if (size <= 0) {
1954 close(local);
1955 unlink(tmpfile);
1956 if (!size)
1957 return error("Connection closed?");
1958 perror("Reading from connection");
1959 return -1;
1961 *bufposn += size;
1962 } while (1);
1963 inflateEnd(&stream);
1965 close(local);
1966 SHA1_Final(real_sha1, &c);
1967 if (ret != Z_STREAM_END) {
1968 unlink(tmpfile);
1969 return error("File %s corrupted", sha1_to_hex(sha1));
1971 if (hashcmp(sha1, real_sha1)) {
1972 unlink(tmpfile);
1973 return error("File %s has bad hash", sha1_to_hex(sha1));
1976 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1979 int has_pack_index(const unsigned char *sha1)
1981 struct stat st;
1982 if (stat(sha1_pack_index_name(sha1), &st))
1983 return 0;
1984 return 1;
1987 int has_pack_file(const unsigned char *sha1)
1989 struct stat st;
1990 if (stat(sha1_pack_name(sha1), &st))
1991 return 0;
1992 return 1;
1995 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1997 struct pack_entry e;
1998 return find_pack_entry(sha1, &e, ignore_packed);
2001 int has_sha1_file(const unsigned char *sha1)
2003 struct stat st;
2004 struct pack_entry e;
2006 if (find_pack_entry(sha1, &e, NULL))
2007 return 1;
2008 return find_sha1_file(sha1, &st) ? 1 : 0;
2012 * reads from fd as long as possible into a supplied buffer of size bytes.
2013 * If necessary the buffer's size is increased using realloc()
2015 * returns 0 if anything went fine and -1 otherwise
2017 * NOTE: both buf and size may change, but even when -1 is returned
2018 * you still have to free() it yourself.
2020 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
2022 char* buf = *return_buf;
2023 unsigned long size = *return_size;
2024 int iret;
2025 unsigned long off = 0;
2027 do {
2028 iret = xread(fd, buf + off, size - off);
2029 if (iret > 0) {
2030 off += iret;
2031 if (off == size) {
2032 size *= 2;
2033 buf = xrealloc(buf, size);
2036 } while (iret > 0);
2038 *return_buf = buf;
2039 *return_size = off;
2041 if (iret < 0)
2042 return -1;
2043 return 0;
2046 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2048 unsigned long size = 4096;
2049 char *buf = xmalloc(size);
2050 int ret;
2052 if (read_pipe(fd, &buf, &size)) {
2053 free(buf);
2054 return -1;
2057 if (!type)
2058 type = blob_type;
2059 if (write_object)
2060 ret = write_sha1_file(buf, size, type, sha1);
2061 else
2062 ret = hash_sha1_file(buf, size, type, sha1);
2063 free(buf);
2064 return ret;
2067 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2068 enum object_type type, const char *path)
2070 unsigned long size = st->st_size;
2071 void *buf;
2072 int ret, re_allocated = 0;
2074 buf = "";
2075 if (size)
2076 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2077 close(fd);
2079 if (!type)
2080 type = OBJ_BLOB;
2083 * Convert blobs to git internal format
2085 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2086 unsigned long nsize = size;
2087 char *nbuf = buf;
2088 if (convert_to_git(path, &nbuf, &nsize)) {
2089 if (size)
2090 munmap(buf, size);
2091 size = nsize;
2092 buf = nbuf;
2093 re_allocated = 1;
2097 if (write_object)
2098 ret = write_sha1_file(buf, size, typename(type), sha1);
2099 else
2100 ret = hash_sha1_file(buf, size, typename(type), sha1);
2101 if (re_allocated) {
2102 free(buf);
2103 return ret;
2105 if (size)
2106 munmap(buf, size);
2107 return ret;
2110 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2112 int fd;
2113 char *target;
2115 switch (st->st_mode & S_IFMT) {
2116 case S_IFREG:
2117 fd = open(path, O_RDONLY);
2118 if (fd < 0)
2119 return error("open(\"%s\"): %s", path,
2120 strerror(errno));
2121 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2122 return error("%s: failed to insert into database",
2123 path);
2124 break;
2125 case S_IFLNK:
2126 target = xmalloc(st->st_size+1);
2127 if (readlink(path, target, st->st_size+1) != st->st_size) {
2128 char *errstr = strerror(errno);
2129 free(target);
2130 return error("readlink(\"%s\"): %s", path,
2131 errstr);
2133 if (!write_object)
2134 hash_sha1_file(target, st->st_size, blob_type, sha1);
2135 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
2136 return error("%s: failed to insert into database",
2137 path);
2138 free(target);
2139 break;
2140 default:
2141 return error("%s: unsupported file type", path);
2143 return 0;
2146 int read_pack_header(int fd, struct pack_header *header)
2148 char *c = (char*)header;
2149 ssize_t remaining = sizeof(struct pack_header);
2150 do {
2151 ssize_t r = xread(fd, c, remaining);
2152 if (r <= 0)
2153 /* "eof before pack header was fully read" */
2154 return PH_ERROR_EOF;
2155 remaining -= r;
2156 c += r;
2157 } while (remaining > 0);
2158 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2159 /* "protocol error (pack signature mismatch detected)" */
2160 return PH_ERROR_PACK_SIGNATURE;
2161 if (!pack_version_ok(header->hdr_version))
2162 /* "protocol error (pack version unsupported)" */
2163 return PH_ERROR_PROTOCOL;
2164 return 0;