Add dangling objects tips.
[git/dscho.git] / sha1_file.c
blob498665e50c1a5053e29a80c37ad95b800bb0f86d
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 static size_t page_size;
411 struct packed_git *packed_git;
413 void pack_report()
415 fprintf(stderr,
416 "pack_report: getpagesize() = %10" SZ_FMT "\n"
417 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
418 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
419 page_size,
420 packed_git_window_size,
421 packed_git_limit);
422 fprintf(stderr,
423 "pack_report: pack_used_ctr = %10u\n"
424 "pack_report: pack_mmap_calls = %10u\n"
425 "pack_report: pack_open_windows = %10u / %10u\n"
426 "pack_report: pack_mapped = "
427 "%10" SZ_FMT " / %10" SZ_FMT "\n",
428 pack_used_ctr,
429 pack_mmap_calls,
430 pack_open_windows, peak_pack_open_windows,
431 pack_mapped, peak_pack_mapped);
434 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
435 void **idx_map_)
437 void *idx_map;
438 uint32_t *index;
439 unsigned long idx_size;
440 int nr, i;
441 int fd = open(path, O_RDONLY);
442 struct stat st;
443 if (fd < 0)
444 return -1;
445 if (fstat(fd, &st)) {
446 close(fd);
447 return -1;
449 idx_size = st.st_size;
450 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
451 close(fd);
453 index = idx_map;
454 *idx_map_ = idx_map;
455 *idx_size_ = idx_size;
457 /* check index map */
458 if (idx_size < 4*256 + 20 + 20)
459 return error("index file %s is too small", path);
461 /* a future index format would start with this, as older git
462 * binaries would fail the non-monotonic index check below.
463 * give a nicer warning to the user if we can.
465 if (index[0] == htonl(PACK_IDX_SIGNATURE))
466 return error("index file %s is a newer version"
467 " and is not supported by this binary"
468 " (try upgrading GIT to a newer version)",
469 path);
471 nr = 0;
472 for (i = 0; i < 256; i++) {
473 unsigned int n = ntohl(index[i]);
474 if (n < nr)
475 return error("non-monotonic index %s", path);
476 nr = n;
480 * Total size:
481 * - 256 index entries 4 bytes each
482 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
483 * - 20-byte SHA1 of the packfile
484 * - 20-byte SHA1 file checksum
486 if (idx_size != 4*256 + nr * 24 + 20 + 20)
487 return error("wrong index file size in %s", path);
489 return 0;
492 static void scan_windows(struct packed_git *p,
493 struct packed_git **lru_p,
494 struct pack_window **lru_w,
495 struct pack_window **lru_l)
497 struct pack_window *w, *w_l;
499 for (w_l = NULL, w = p->windows; w; w = w->next) {
500 if (!w->inuse_cnt) {
501 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
502 *lru_p = p;
503 *lru_w = w;
504 *lru_l = w_l;
507 w_l = w;
511 static int unuse_one_window(struct packed_git *current)
513 struct packed_git *p, *lru_p = NULL;
514 struct pack_window *lru_w = NULL, *lru_l = NULL;
516 if (current)
517 scan_windows(current, &lru_p, &lru_w, &lru_l);
518 for (p = packed_git; p; p = p->next)
519 scan_windows(p, &lru_p, &lru_w, &lru_l);
520 if (lru_p) {
521 munmap(lru_w->base, lru_w->len);
522 pack_mapped -= lru_w->len;
523 if (lru_l)
524 lru_l->next = lru_w->next;
525 else {
526 lru_p->windows = lru_w->next;
527 if (!lru_p->windows && lru_p != current) {
528 close(lru_p->pack_fd);
529 lru_p->pack_fd = -1;
532 free(lru_w);
533 pack_open_windows--;
534 return 1;
536 return 0;
539 void release_pack_memory(size_t need)
541 size_t cur = pack_mapped;
542 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
543 ; /* nothing */
546 void unuse_pack(struct pack_window **w_cursor)
548 struct pack_window *w = *w_cursor;
549 if (w) {
550 w->inuse_cnt--;
551 *w_cursor = NULL;
555 static void open_packed_git(struct packed_git *p)
557 struct stat st;
558 struct pack_header hdr;
559 unsigned char sha1[20];
560 unsigned char *idx_sha1;
561 long fd_flag;
563 p->pack_fd = open(p->pack_name, O_RDONLY);
564 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
565 die("packfile %s cannot be opened", p->pack_name);
567 /* If we created the struct before we had the pack we lack size. */
568 if (!p->pack_size) {
569 if (!S_ISREG(st.st_mode))
570 die("packfile %s not a regular file", p->pack_name);
571 p->pack_size = st.st_size;
572 } else if (p->pack_size != st.st_size)
573 die("packfile %s size changed", p->pack_name);
575 /* We leave these file descriptors open with sliding mmap;
576 * there is no point keeping them open across exec(), though.
578 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
579 if (fd_flag < 0)
580 die("cannot determine file descriptor flags");
581 fd_flag |= FD_CLOEXEC;
582 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
583 die("cannot set FD_CLOEXEC");
585 /* Verify we recognize this pack file format. */
586 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
587 die("file %s is far too short to be a packfile", p->pack_name);
588 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
589 die("file %s is not a GIT packfile", p->pack_name);
590 if (!pack_version_ok(hdr.hdr_version))
591 die("packfile %s is version %u and not supported"
592 " (try upgrading GIT to a newer version)",
593 p->pack_name, ntohl(hdr.hdr_version));
595 /* Verify the pack matches its index. */
596 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
597 die("packfile %s claims to have %u objects"
598 " while index size indicates %u objects",
599 p->pack_name, ntohl(hdr.hdr_entries),
600 num_packed_objects(p));
601 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
602 die("end of packfile %s is unavailable", p->pack_name);
603 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
604 die("packfile %s signature is unavailable", p->pack_name);
605 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
606 if (hashcmp(sha1, idx_sha1))
607 die("packfile %s does not match index", p->pack_name);
610 static int in_window(struct pack_window *win, unsigned long offset)
612 /* We must promise at least 20 bytes (one hash) after the
613 * offset is available from this window, otherwise the offset
614 * is not actually in this window and a different window (which
615 * has that one hash excess) must be used. This is to support
616 * the object header and delta base parsing routines below.
618 off_t win_off = win->offset;
619 return win_off <= offset
620 && (offset + 20) <= (win_off + win->len);
623 unsigned char* use_pack(struct packed_git *p,
624 struct pack_window **w_cursor,
625 unsigned long offset,
626 unsigned int *left)
628 struct pack_window *win = *w_cursor;
630 if (p->pack_fd == -1)
631 open_packed_git(p);
633 /* Since packfiles end in a hash of their content and its
634 * pointless to ask for an offset into the middle of that
635 * hash, and the in_window function above wouldn't match
636 * don't allow an offset too close to the end of the file.
638 if (offset > (p->pack_size - 20))
639 die("offset beyond end of packfile (truncated pack?)");
641 if (!win || !in_window(win, offset)) {
642 if (win)
643 win->inuse_cnt--;
644 for (win = p->windows; win; win = win->next) {
645 if (in_window(win, offset))
646 break;
648 if (!win) {
649 if (!page_size)
650 page_size = getpagesize();
651 win = xcalloc(1, sizeof(*win));
652 win->offset = (offset / page_size) * page_size;
653 win->len = p->pack_size - win->offset;
654 if (win->len > packed_git_window_size)
655 win->len = packed_git_window_size;
656 pack_mapped += win->len;
657 while (packed_git_limit < pack_mapped
658 && unuse_one_window(p))
659 ; /* nothing */
660 win->base = xmmap(NULL, win->len,
661 PROT_READ, MAP_PRIVATE,
662 p->pack_fd, win->offset);
663 if (win->base == MAP_FAILED)
664 die("packfile %s cannot be mapped: %s",
665 p->pack_name,
666 strerror(errno));
667 pack_mmap_calls++;
668 pack_open_windows++;
669 if (pack_mapped > peak_pack_mapped)
670 peak_pack_mapped = pack_mapped;
671 if (pack_open_windows > peak_pack_open_windows)
672 peak_pack_open_windows = pack_open_windows;
673 win->next = p->windows;
674 p->windows = win;
677 if (win != *w_cursor) {
678 win->last_used = pack_used_ctr++;
679 win->inuse_cnt++;
680 *w_cursor = win;
682 offset -= win->offset;
683 if (left)
684 *left = win->len - offset;
685 return win->base + offset;
688 struct packed_git *add_packed_git(char *path, int path_len, int local)
690 struct stat st;
691 struct packed_git *p;
692 unsigned long idx_size;
693 void *idx_map;
694 unsigned char sha1[20];
696 if (check_packed_git_idx(path, &idx_size, &idx_map))
697 return NULL;
699 /* do we have a corresponding .pack file? */
700 strcpy(path + path_len - 4, ".pack");
701 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
702 munmap(idx_map, idx_size);
703 return NULL;
705 /* ok, it looks sane as far as we can check without
706 * actually mapping the pack file.
708 p = xmalloc(sizeof(*p) + path_len + 2);
709 strcpy(p->pack_name, path);
710 p->index_size = idx_size;
711 p->pack_size = st.st_size;
712 p->index_base = idx_map;
713 p->next = NULL;
714 p->windows = NULL;
715 p->pack_fd = -1;
716 p->pack_local = local;
717 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
718 hashcpy(p->sha1, sha1);
719 return p;
722 struct packed_git *parse_pack_index(unsigned char *sha1)
724 char *path = sha1_pack_index_name(sha1);
725 return parse_pack_index_file(sha1, path);
728 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
730 struct packed_git *p;
731 unsigned long idx_size;
732 void *idx_map;
733 char *path;
735 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
736 return NULL;
738 path = sha1_pack_name(sha1);
740 p = xmalloc(sizeof(*p) + strlen(path) + 2);
741 strcpy(p->pack_name, path);
742 p->index_size = idx_size;
743 p->pack_size = 0;
744 p->index_base = idx_map;
745 p->next = NULL;
746 p->windows = NULL;
747 p->pack_fd = -1;
748 hashcpy(p->sha1, sha1);
749 return p;
752 void install_packed_git(struct packed_git *pack)
754 pack->next = packed_git;
755 packed_git = pack;
758 static void prepare_packed_git_one(char *objdir, int local)
760 char path[PATH_MAX];
761 int len;
762 DIR *dir;
763 struct dirent *de;
765 sprintf(path, "%s/pack", objdir);
766 len = strlen(path);
767 dir = opendir(path);
768 if (!dir) {
769 if (errno != ENOENT)
770 error("unable to open object pack directory: %s: %s",
771 path, strerror(errno));
772 return;
774 path[len++] = '/';
775 while ((de = readdir(dir)) != NULL) {
776 int namelen = strlen(de->d_name);
777 struct packed_git *p;
779 if (!has_extension(de->d_name, ".idx"))
780 continue;
782 /* we have .idx. Is it a file we can map? */
783 strcpy(path + len, de->d_name);
784 for (p = packed_git; p; p = p->next) {
785 if (!memcmp(path, p->pack_name, len + namelen - 4))
786 break;
788 if (p)
789 continue;
790 p = add_packed_git(path, len + namelen, local);
791 if (!p)
792 continue;
793 p->next = packed_git;
794 packed_git = p;
796 closedir(dir);
799 static int prepare_packed_git_run_once = 0;
800 void prepare_packed_git(void)
802 struct alternate_object_database *alt;
804 if (prepare_packed_git_run_once)
805 return;
806 prepare_packed_git_one(get_object_directory(), 1);
807 prepare_alt_odb();
808 for (alt = alt_odb_list; alt; alt = alt->next) {
809 alt->name[-1] = 0;
810 prepare_packed_git_one(alt->base, 0);
811 alt->name[-1] = '/';
813 prepare_packed_git_run_once = 1;
816 void reprepare_packed_git(void)
818 prepare_packed_git_run_once = 0;
819 prepare_packed_git();
822 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
824 unsigned char real_sha1[20];
825 hash_sha1_file(map, size, type, real_sha1);
826 return hashcmp(sha1, real_sha1) ? -1 : 0;
829 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
831 struct stat st;
832 void *map;
833 int fd;
834 char *filename = find_sha1_file(sha1, &st);
836 if (!filename) {
837 return NULL;
840 fd = open(filename, O_RDONLY | sha1_file_open_flag);
841 if (fd < 0) {
842 /* See if it works without O_NOATIME */
843 switch (sha1_file_open_flag) {
844 default:
845 fd = open(filename, O_RDONLY);
846 if (fd >= 0)
847 break;
848 /* Fallthrough */
849 case 0:
850 return NULL;
853 /* If it failed once, it will probably fail again.
854 * Stop using O_NOATIME
856 sha1_file_open_flag = 0;
858 map = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
859 close(fd);
860 *size = st.st_size;
861 return map;
864 int legacy_loose_object(unsigned char *map)
866 unsigned int word;
869 * Is it a zlib-compressed buffer? If so, the first byte
870 * must be 0x78 (15-bit window size, deflated), and the
871 * first 16-bit word is evenly divisible by 31
873 word = (map[0] << 8) + map[1];
874 if (map[0] == 0x78 && !(word % 31))
875 return 1;
876 else
877 return 0;
880 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
882 unsigned shift;
883 unsigned char c;
884 unsigned long size;
885 unsigned long used = 0;
887 c = buf[used++];
888 *type = (c >> 4) & 7;
889 size = c & 15;
890 shift = 4;
891 while (c & 0x80) {
892 if (len <= used)
893 return 0;
894 if (sizeof(long) * 8 <= shift)
895 return 0;
896 c = buf[used++];
897 size += (c & 0x7f) << shift;
898 shift += 7;
900 *sizep = size;
901 return used;
904 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
906 unsigned long size, used;
907 static const char valid_loose_object_type[8] = {
908 0, /* OBJ_EXT */
909 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
910 0, /* "delta" and others are invalid in a loose object */
912 enum object_type type;
914 /* Get the data stream */
915 memset(stream, 0, sizeof(*stream));
916 stream->next_in = map;
917 stream->avail_in = mapsize;
918 stream->next_out = buffer;
919 stream->avail_out = bufsiz;
921 if (legacy_loose_object(map)) {
922 inflateInit(stream);
923 return inflate(stream, 0);
926 used = unpack_object_header_gently(map, mapsize, &type, &size);
927 if (!used || !valid_loose_object_type[type])
928 return -1;
929 map += used;
930 mapsize -= used;
932 /* Set up the stream for the rest.. */
933 stream->next_in = map;
934 stream->avail_in = mapsize;
935 inflateInit(stream);
937 /* And generate the fake traditional header */
938 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
939 type_names[type], size);
940 return 0;
943 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
945 int bytes = strlen(buffer) + 1;
946 unsigned char *buf = xmalloc(1+size);
947 unsigned long n;
949 n = stream->total_out - bytes;
950 if (n > size)
951 n = size;
952 memcpy(buf, (char *) buffer + bytes, n);
953 bytes = n;
954 if (bytes < size) {
955 stream->next_out = buf + bytes;
956 stream->avail_out = size - bytes;
957 while (inflate(stream, Z_FINISH) == Z_OK)
958 /* nothing */;
960 buf[size] = 0;
961 inflateEnd(stream);
962 return buf;
966 * We used to just use "sscanf()", but that's actually way
967 * too permissive for what we want to check. So do an anal
968 * object header parse by hand.
970 static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
972 int i;
973 unsigned long size;
976 * The type can be at most ten bytes (including the
977 * terminating '\0' that we add), and is followed by
978 * a space.
980 i = 10;
981 for (;;) {
982 char c = *hdr++;
983 if (c == ' ')
984 break;
985 if (!--i)
986 return -1;
987 *type++ = c;
989 *type = 0;
992 * The length must follow immediately, and be in canonical
993 * decimal format (ie "010" is not valid).
995 size = *hdr++ - '0';
996 if (size > 9)
997 return -1;
998 if (size) {
999 for (;;) {
1000 unsigned long c = *hdr - '0';
1001 if (c > 9)
1002 break;
1003 hdr++;
1004 size = size * 10 + c;
1007 *sizep = size;
1010 * The length must be followed by a zero byte
1012 return *hdr ? -1 : 0;
1015 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
1017 int ret;
1018 z_stream stream;
1019 char hdr[8192];
1021 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1022 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
1023 return NULL;
1025 return unpack_sha1_rest(&stream, hdr, *size);
1028 static unsigned long get_delta_base(struct packed_git *p,
1029 struct pack_window **w_curs,
1030 unsigned long offset,
1031 enum object_type kind,
1032 unsigned long delta_obj_offset,
1033 unsigned long *base_obj_offset)
1035 unsigned char *base_info = use_pack(p, w_curs, offset, NULL);
1036 unsigned long base_offset;
1038 /* use_pack() assured us we have [base_info, base_info + 20)
1039 * as a range that we can look at without walking off the
1040 * end of the mapped window. Its actually the hash size
1041 * that is assured. An OFS_DELTA longer than the hash size
1042 * is stupid, as then a REF_DELTA would be smaller to store.
1044 if (kind == OBJ_OFS_DELTA) {
1045 unsigned used = 0;
1046 unsigned char c = base_info[used++];
1047 base_offset = c & 127;
1048 while (c & 128) {
1049 base_offset += 1;
1050 if (!base_offset || base_offset & ~(~0UL >> 7))
1051 die("offset value overflow for delta base object");
1052 c = base_info[used++];
1053 base_offset = (base_offset << 7) + (c & 127);
1055 base_offset = delta_obj_offset - base_offset;
1056 if (base_offset >= delta_obj_offset)
1057 die("delta base offset out of bound");
1058 offset += used;
1059 } else if (kind == OBJ_REF_DELTA) {
1060 /* The base entry _must_ be in the same pack */
1061 base_offset = find_pack_entry_one(base_info, p);
1062 if (!base_offset)
1063 die("failed to find delta-pack base object %s",
1064 sha1_to_hex(base_info));
1065 offset += 20;
1066 } else
1067 die("I am totally screwed");
1068 *base_obj_offset = base_offset;
1069 return offset;
1072 /* forward declaration for a mutually recursive function */
1073 static int packed_object_info(struct packed_git *p, unsigned long offset,
1074 char *type, unsigned long *sizep);
1076 static int packed_delta_info(struct packed_git *p,
1077 struct pack_window **w_curs,
1078 unsigned long offset,
1079 enum object_type kind,
1080 unsigned long obj_offset,
1081 char *type,
1082 unsigned long *sizep)
1084 unsigned long base_offset;
1086 offset = get_delta_base(p, w_curs, offset, kind,
1087 obj_offset, &base_offset);
1089 /* We choose to only get the type of the base object and
1090 * ignore potentially corrupt pack file that expects the delta
1091 * based on a base with a wrong size. This saves tons of
1092 * inflate() calls.
1094 if (packed_object_info(p, base_offset, type, NULL))
1095 die("cannot get info for delta-pack base");
1097 if (sizep) {
1098 const unsigned char *data;
1099 unsigned char delta_head[20], *in;
1100 unsigned long result_size;
1101 z_stream stream;
1102 int st;
1104 memset(&stream, 0, sizeof(stream));
1105 stream.next_out = delta_head;
1106 stream.avail_out = sizeof(delta_head);
1108 inflateInit(&stream);
1109 do {
1110 in = use_pack(p, w_curs, offset, &stream.avail_in);
1111 stream.next_in = in;
1112 st = inflate(&stream, Z_FINISH);
1113 offset += stream.next_in - in;
1114 } while ((st == Z_OK || st == Z_BUF_ERROR)
1115 && stream.total_out < sizeof(delta_head));
1116 inflateEnd(&stream);
1117 if ((st != Z_STREAM_END) &&
1118 stream.total_out != sizeof(delta_head))
1119 die("delta data unpack-initial failed");
1121 /* Examine the initial part of the delta to figure out
1122 * the result size.
1124 data = delta_head;
1126 /* ignore base size */
1127 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1129 /* Read the result size */
1130 result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1131 *sizep = result_size;
1133 return 0;
1136 static unsigned long unpack_object_header(struct packed_git *p,
1137 struct pack_window **w_curs,
1138 unsigned long offset,
1139 enum object_type *type,
1140 unsigned long *sizep)
1142 unsigned char *base;
1143 unsigned int left;
1144 unsigned long used;
1146 /* use_pack() assures us we have [base, base + 20) available
1147 * as a range that we can look at at. (Its actually the hash
1148 * size that is assurred.) With our object header encoding
1149 * the maximum deflated object size is 2^137, which is just
1150 * insane, so we know won't exceed what we have been given.
1152 base = use_pack(p, w_curs, offset, &left);
1153 used = unpack_object_header_gently(base, left, type, sizep);
1154 if (!used)
1155 die("object offset outside of pack file");
1157 return offset + used;
1160 void packed_object_info_detail(struct packed_git *p,
1161 unsigned long offset,
1162 char *type,
1163 unsigned long *size,
1164 unsigned long *store_size,
1165 unsigned int *delta_chain_length,
1166 unsigned char *base_sha1)
1168 struct pack_window *w_curs = NULL;
1169 unsigned long obj_offset, val;
1170 unsigned char *next_sha1;
1171 enum object_type kind;
1173 *delta_chain_length = 0;
1174 obj_offset = offset;
1175 offset = unpack_object_header(p, &w_curs, offset, &kind, size);
1177 for (;;) {
1178 switch (kind) {
1179 default:
1180 die("pack %s contains unknown object type %d",
1181 p->pack_name, kind);
1182 case OBJ_COMMIT:
1183 case OBJ_TREE:
1184 case OBJ_BLOB:
1185 case OBJ_TAG:
1186 strcpy(type, type_names[kind]);
1187 *store_size = 0; /* notyet */
1188 unuse_pack(&w_curs);
1189 return;
1190 case OBJ_OFS_DELTA:
1191 get_delta_base(p, &w_curs, offset, kind,
1192 obj_offset, &offset);
1193 if (*delta_chain_length == 0) {
1194 /* TODO: find base_sha1 as pointed by offset */
1196 break;
1197 case OBJ_REF_DELTA:
1198 next_sha1 = use_pack(p, &w_curs, offset, NULL);
1199 if (*delta_chain_length == 0)
1200 hashcpy(base_sha1, next_sha1);
1201 offset = find_pack_entry_one(next_sha1, p);
1202 break;
1204 obj_offset = offset;
1205 offset = unpack_object_header(p, &w_curs, offset, &kind, &val);
1206 (*delta_chain_length)++;
1210 static int packed_object_info(struct packed_git *p, unsigned long offset,
1211 char *type, unsigned long *sizep)
1213 struct pack_window *w_curs = NULL;
1214 unsigned long size, obj_offset = offset;
1215 enum object_type kind;
1216 int r;
1218 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1220 switch (kind) {
1221 case OBJ_OFS_DELTA:
1222 case OBJ_REF_DELTA:
1223 r = packed_delta_info(p, &w_curs, offset, kind,
1224 obj_offset, type, sizep);
1225 unuse_pack(&w_curs);
1226 return r;
1227 case OBJ_COMMIT:
1228 case OBJ_TREE:
1229 case OBJ_BLOB:
1230 case OBJ_TAG:
1231 strcpy(type, type_names[kind]);
1232 unuse_pack(&w_curs);
1233 break;
1234 default:
1235 die("pack %s contains unknown object type %d",
1236 p->pack_name, kind);
1238 if (sizep)
1239 *sizep = size;
1240 return 0;
1243 static void *unpack_compressed_entry(struct packed_git *p,
1244 struct pack_window **w_curs,
1245 unsigned long offset,
1246 unsigned long size)
1248 int st;
1249 z_stream stream;
1250 unsigned char *buffer, *in;
1252 buffer = xmalloc(size + 1);
1253 buffer[size] = 0;
1254 memset(&stream, 0, sizeof(stream));
1255 stream.next_out = buffer;
1256 stream.avail_out = size;
1258 inflateInit(&stream);
1259 do {
1260 in = use_pack(p, w_curs, offset, &stream.avail_in);
1261 stream.next_in = in;
1262 st = inflate(&stream, Z_FINISH);
1263 offset += stream.next_in - in;
1264 } while (st == Z_OK || st == Z_BUF_ERROR);
1265 inflateEnd(&stream);
1266 if ((st != Z_STREAM_END) || stream.total_out != size) {
1267 free(buffer);
1268 return NULL;
1271 return buffer;
1274 static void *unpack_delta_entry(struct packed_git *p,
1275 struct pack_window **w_curs,
1276 unsigned long offset,
1277 unsigned long delta_size,
1278 enum object_type kind,
1279 unsigned long obj_offset,
1280 char *type,
1281 unsigned long *sizep)
1283 void *delta_data, *result, *base;
1284 unsigned long result_size, base_size, base_offset;
1286 offset = get_delta_base(p, w_curs, offset, kind,
1287 obj_offset, &base_offset);
1288 base = unpack_entry(p, base_offset, type, &base_size);
1289 if (!base)
1290 die("failed to read delta base object at %lu from %s",
1291 base_offset, p->pack_name);
1293 delta_data = unpack_compressed_entry(p, w_curs, offset, delta_size);
1294 result = patch_delta(base, base_size,
1295 delta_data, delta_size,
1296 &result_size);
1297 if (!result)
1298 die("failed to apply delta");
1299 free(delta_data);
1300 free(base);
1301 *sizep = result_size;
1302 return result;
1305 void *unpack_entry(struct packed_git *p, unsigned long offset,
1306 char *type, unsigned long *sizep)
1308 struct pack_window *w_curs = NULL;
1309 unsigned long size, obj_offset = offset;
1310 enum object_type kind;
1311 void *retval;
1313 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1314 switch (kind) {
1315 case OBJ_OFS_DELTA:
1316 case OBJ_REF_DELTA:
1317 retval = unpack_delta_entry(p, &w_curs, offset, size,
1318 kind, obj_offset, type, sizep);
1319 break;
1320 case OBJ_COMMIT:
1321 case OBJ_TREE:
1322 case OBJ_BLOB:
1323 case OBJ_TAG:
1324 strcpy(type, type_names[kind]);
1325 *sizep = size;
1326 retval = unpack_compressed_entry(p, &w_curs, offset, size);
1327 break;
1328 default:
1329 die("unknown object type %i in %s", kind, p->pack_name);
1331 unuse_pack(&w_curs);
1332 return retval;
1335 int num_packed_objects(const struct packed_git *p)
1337 /* See check_packed_git_idx() */
1338 return (p->index_size - 20 - 20 - 4*256) / 24;
1341 int nth_packed_object_sha1(const struct packed_git *p, int n,
1342 unsigned char* sha1)
1344 void *index = p->index_base + 256;
1345 if (n < 0 || num_packed_objects(p) <= n)
1346 return -1;
1347 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1348 return 0;
1351 unsigned long find_pack_entry_one(const unsigned char *sha1,
1352 struct packed_git *p)
1354 uint32_t *level1_ofs = p->index_base;
1355 int hi = ntohl(level1_ofs[*sha1]);
1356 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1357 void *index = p->index_base + 256;
1359 do {
1360 int mi = (lo + hi) / 2;
1361 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1362 if (!cmp)
1363 return ntohl(*((uint32_t *)((char *)index + (24 * mi))));
1364 if (cmp > 0)
1365 hi = mi;
1366 else
1367 lo = mi+1;
1368 } while (lo < hi);
1369 return 0;
1372 static int matches_pack_name(struct packed_git *p, const char *ig)
1374 const char *last_c, *c;
1376 if (!strcmp(p->pack_name, ig))
1377 return 0;
1379 for (c = p->pack_name, last_c = c; *c;)
1380 if (*c == '/')
1381 last_c = ++c;
1382 else
1383 ++c;
1384 if (!strcmp(last_c, ig))
1385 return 0;
1387 return 1;
1390 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1392 struct packed_git *p;
1393 unsigned long offset;
1395 prepare_packed_git();
1397 for (p = packed_git; p; p = p->next) {
1398 if (ignore_packed) {
1399 const char **ig;
1400 for (ig = ignore_packed; *ig; ig++)
1401 if (!matches_pack_name(p, *ig))
1402 break;
1403 if (*ig)
1404 continue;
1406 offset = find_pack_entry_one(sha1, p);
1407 if (offset) {
1408 e->offset = offset;
1409 e->p = p;
1410 hashcpy(e->sha1, sha1);
1411 return 1;
1414 return 0;
1417 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1418 struct packed_git *packs)
1420 struct packed_git *p;
1422 for (p = packs; p; p = p->next) {
1423 if (find_pack_entry_one(sha1, p))
1424 return p;
1426 return NULL;
1430 static int sha1_loose_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1432 int status;
1433 unsigned long mapsize, size;
1434 void *map;
1435 z_stream stream;
1436 char hdr[128];
1438 map = map_sha1_file(sha1, &mapsize);
1439 if (!map)
1440 return error("unable to find %s", sha1_to_hex(sha1));
1441 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1442 status = error("unable to unpack %s header",
1443 sha1_to_hex(sha1));
1444 if (parse_sha1_header(hdr, type, &size) < 0)
1445 status = error("unable to parse %s header", sha1_to_hex(sha1));
1446 else {
1447 status = 0;
1448 if (sizep)
1449 *sizep = size;
1451 inflateEnd(&stream);
1452 munmap(map, mapsize);
1453 return status;
1456 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1458 struct pack_entry e;
1460 if (!find_pack_entry(sha1, &e, NULL)) {
1461 reprepare_packed_git();
1462 if (!find_pack_entry(sha1, &e, NULL))
1463 return sha1_loose_object_info(sha1, type, sizep);
1465 return packed_object_info(e.p, e.offset, type, sizep);
1468 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1470 struct pack_entry e;
1472 if (!find_pack_entry(sha1, &e, NULL))
1473 return NULL;
1474 else
1475 return unpack_entry(e.p, e.offset, type, size);
1478 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1480 unsigned long mapsize;
1481 void *map, *buf;
1483 buf = read_packed_sha1(sha1, type, size);
1484 if (buf)
1485 return buf;
1486 map = map_sha1_file(sha1, &mapsize);
1487 if (map) {
1488 buf = unpack_sha1_file(map, mapsize, type, size);
1489 munmap(map, mapsize);
1490 return buf;
1492 reprepare_packed_git();
1493 return read_packed_sha1(sha1, type, size);
1496 void *read_object_with_reference(const unsigned char *sha1,
1497 const char *required_type,
1498 unsigned long *size,
1499 unsigned char *actual_sha1_return)
1501 char type[20];
1502 void *buffer;
1503 unsigned long isize;
1504 unsigned char actual_sha1[20];
1506 hashcpy(actual_sha1, sha1);
1507 while (1) {
1508 int ref_length = -1;
1509 const char *ref_type = NULL;
1511 buffer = read_sha1_file(actual_sha1, type, &isize);
1512 if (!buffer)
1513 return NULL;
1514 if (!strcmp(type, required_type)) {
1515 *size = isize;
1516 if (actual_sha1_return)
1517 hashcpy(actual_sha1_return, actual_sha1);
1518 return buffer;
1520 /* Handle references */
1521 else if (!strcmp(type, commit_type))
1522 ref_type = "tree ";
1523 else if (!strcmp(type, tag_type))
1524 ref_type = "object ";
1525 else {
1526 free(buffer);
1527 return NULL;
1529 ref_length = strlen(ref_type);
1531 if (memcmp(buffer, ref_type, ref_length) ||
1532 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1533 free(buffer);
1534 return NULL;
1536 free(buffer);
1537 /* Now we have the ID of the referred-to object in
1538 * actual_sha1. Check again. */
1542 static void write_sha1_file_prepare(void *buf, unsigned long len,
1543 const char *type, unsigned char *sha1,
1544 unsigned char *hdr, int *hdrlen)
1546 SHA_CTX c;
1548 /* Generate the header */
1549 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1551 /* Sha1.. */
1552 SHA1_Init(&c);
1553 SHA1_Update(&c, hdr, *hdrlen);
1554 SHA1_Update(&c, buf, len);
1555 SHA1_Final(sha1, &c);
1559 * Link the tempfile to the final place, possibly creating the
1560 * last directory level as you do so.
1562 * Returns the errno on failure, 0 on success.
1564 static int link_temp_to_file(const char *tmpfile, const char *filename)
1566 int ret;
1567 char *dir;
1569 if (!link(tmpfile, filename))
1570 return 0;
1573 * Try to mkdir the last path component if that failed.
1575 * Re-try the "link()" regardless of whether the mkdir
1576 * succeeds, since a race might mean that somebody
1577 * else succeeded.
1579 ret = errno;
1580 dir = strrchr(filename, '/');
1581 if (dir) {
1582 *dir = 0;
1583 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1584 *dir = '/';
1585 return -2;
1587 *dir = '/';
1588 if (!link(tmpfile, filename))
1589 return 0;
1590 ret = errno;
1592 return ret;
1596 * Move the just written object into its final resting place
1598 int move_temp_to_file(const char *tmpfile, const char *filename)
1600 int ret = link_temp_to_file(tmpfile, filename);
1603 * Coda hack - coda doesn't like cross-directory links,
1604 * so we fall back to a rename, which will mean that it
1605 * won't be able to check collisions, but that's not a
1606 * big deal.
1608 * The same holds for FAT formatted media.
1610 * When this succeeds, we just return 0. We have nothing
1611 * left to unlink.
1613 if (ret && ret != EEXIST) {
1614 if (!rename(tmpfile, filename))
1615 return 0;
1616 ret = errno;
1618 unlink(tmpfile);
1619 if (ret) {
1620 if (ret != EEXIST) {
1621 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1623 /* FIXME!!! Collision check here ? */
1626 return 0;
1629 static int write_buffer(int fd, const void *buf, size_t len)
1631 if (write_in_full(fd, buf, len) < 0)
1632 return error("file write error (%s)", strerror(errno));
1633 return 0;
1636 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1638 int hdr_len;
1639 unsigned char c;
1641 c = (type << 4) | (len & 15);
1642 len >>= 4;
1643 hdr_len = 1;
1644 while (len) {
1645 *hdr++ = c | 0x80;
1646 hdr_len++;
1647 c = (len & 0x7f);
1648 len >>= 7;
1650 *hdr = c;
1651 return hdr_len;
1654 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1656 int obj_type, hdr;
1658 if (use_legacy_headers) {
1659 while (deflate(stream, 0) == Z_OK)
1660 /* nothing */;
1661 return;
1663 if (!strcmp(type, blob_type))
1664 obj_type = OBJ_BLOB;
1665 else if (!strcmp(type, tree_type))
1666 obj_type = OBJ_TREE;
1667 else if (!strcmp(type, commit_type))
1668 obj_type = OBJ_COMMIT;
1669 else if (!strcmp(type, tag_type))
1670 obj_type = OBJ_TAG;
1671 else
1672 die("trying to generate bogus object of type '%s'", type);
1673 hdr = write_binary_header(stream->next_out, obj_type, len);
1674 stream->total_out = hdr;
1675 stream->next_out += hdr;
1676 stream->avail_out -= hdr;
1679 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1680 unsigned char *sha1)
1682 unsigned char hdr[50];
1683 int hdrlen;
1684 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1685 return 0;
1688 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1690 int size;
1691 unsigned char *compressed;
1692 z_stream stream;
1693 unsigned char sha1[20];
1694 char *filename;
1695 static char tmpfile[PATH_MAX];
1696 unsigned char hdr[50];
1697 int fd, hdrlen;
1699 /* Normally if we have it in the pack then we do not bother writing
1700 * it out into .git/objects/??/?{38} file.
1702 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1703 filename = sha1_file_name(sha1);
1704 if (returnsha1)
1705 hashcpy(returnsha1, sha1);
1706 if (has_sha1_file(sha1))
1707 return 0;
1708 fd = open(filename, O_RDONLY);
1709 if (fd >= 0) {
1711 * FIXME!!! We might do collision checking here, but we'd
1712 * need to uncompress the old file and check it. Later.
1714 close(fd);
1715 return 0;
1718 if (errno != ENOENT) {
1719 return error("sha1 file %s: %s\n", filename, strerror(errno));
1722 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1724 fd = mkstemp(tmpfile);
1725 if (fd < 0) {
1726 if (errno == EPERM)
1727 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1728 else
1729 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1732 /* Set it up */
1733 memset(&stream, 0, sizeof(stream));
1734 deflateInit(&stream, zlib_compression_level);
1735 size = 8 + deflateBound(&stream, len+hdrlen);
1736 compressed = xmalloc(size);
1738 /* Compress it */
1739 stream.next_out = compressed;
1740 stream.avail_out = size;
1742 /* First header.. */
1743 stream.next_in = hdr;
1744 stream.avail_in = hdrlen;
1745 setup_object_header(&stream, type, len);
1747 /* Then the data itself.. */
1748 stream.next_in = buf;
1749 stream.avail_in = len;
1750 while (deflate(&stream, Z_FINISH) == Z_OK)
1751 /* nothing */;
1752 deflateEnd(&stream);
1753 size = stream.total_out;
1755 if (write_buffer(fd, compressed, size) < 0)
1756 die("unable to write sha1 file");
1757 fchmod(fd, 0444);
1758 close(fd);
1759 free(compressed);
1761 return move_temp_to_file(tmpfile, filename);
1765 * We need to unpack and recompress the object for writing
1766 * it out to a different file.
1768 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1770 size_t size;
1771 z_stream stream;
1772 unsigned char *unpacked;
1773 unsigned long len;
1774 char type[20];
1775 char hdr[50];
1776 int hdrlen;
1777 void *buf;
1779 /* need to unpack and recompress it by itself */
1780 unpacked = read_packed_sha1(sha1, type, &len);
1781 if (!unpacked)
1782 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1784 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1786 /* Set it up */
1787 memset(&stream, 0, sizeof(stream));
1788 deflateInit(&stream, zlib_compression_level);
1789 size = deflateBound(&stream, len + hdrlen);
1790 buf = xmalloc(size);
1792 /* Compress it */
1793 stream.next_out = buf;
1794 stream.avail_out = size;
1796 /* First header.. */
1797 stream.next_in = (void *)hdr;
1798 stream.avail_in = hdrlen;
1799 while (deflate(&stream, 0) == Z_OK)
1800 /* nothing */;
1802 /* Then the data itself.. */
1803 stream.next_in = unpacked;
1804 stream.avail_in = len;
1805 while (deflate(&stream, Z_FINISH) == Z_OK)
1806 /* nothing */;
1807 deflateEnd(&stream);
1808 free(unpacked);
1810 *objsize = stream.total_out;
1811 return buf;
1814 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1816 int retval;
1817 unsigned long objsize;
1818 void *buf = map_sha1_file(sha1, &objsize);
1820 if (buf) {
1821 retval = write_buffer(fd, buf, objsize);
1822 munmap(buf, objsize);
1823 return retval;
1826 buf = repack_object(sha1, &objsize);
1827 retval = write_buffer(fd, buf, objsize);
1828 free(buf);
1829 return retval;
1832 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1833 size_t bufsize, size_t *bufposn)
1835 char tmpfile[PATH_MAX];
1836 int local;
1837 z_stream stream;
1838 unsigned char real_sha1[20];
1839 unsigned char discard[4096];
1840 int ret;
1841 SHA_CTX c;
1843 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1845 local = mkstemp(tmpfile);
1846 if (local < 0) {
1847 if (errno == EPERM)
1848 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1849 else
1850 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1853 memset(&stream, 0, sizeof(stream));
1855 inflateInit(&stream);
1857 SHA1_Init(&c);
1859 do {
1860 ssize_t size;
1861 if (*bufposn) {
1862 stream.avail_in = *bufposn;
1863 stream.next_in = (unsigned char *) buffer;
1864 do {
1865 stream.next_out = discard;
1866 stream.avail_out = sizeof(discard);
1867 ret = inflate(&stream, Z_SYNC_FLUSH);
1868 SHA1_Update(&c, discard, sizeof(discard) -
1869 stream.avail_out);
1870 } while (stream.avail_in && ret == Z_OK);
1871 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1872 die("unable to write sha1 file");
1873 memmove(buffer, buffer + *bufposn - stream.avail_in,
1874 stream.avail_in);
1875 *bufposn = stream.avail_in;
1876 if (ret != Z_OK)
1877 break;
1879 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
1880 if (size <= 0) {
1881 close(local);
1882 unlink(tmpfile);
1883 if (!size)
1884 return error("Connection closed?");
1885 perror("Reading from connection");
1886 return -1;
1888 *bufposn += size;
1889 } while (1);
1890 inflateEnd(&stream);
1892 close(local);
1893 SHA1_Final(real_sha1, &c);
1894 if (ret != Z_STREAM_END) {
1895 unlink(tmpfile);
1896 return error("File %s corrupted", sha1_to_hex(sha1));
1898 if (hashcmp(sha1, real_sha1)) {
1899 unlink(tmpfile);
1900 return error("File %s has bad hash", sha1_to_hex(sha1));
1903 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1906 int has_pack_index(const unsigned char *sha1)
1908 struct stat st;
1909 if (stat(sha1_pack_index_name(sha1), &st))
1910 return 0;
1911 return 1;
1914 int has_pack_file(const unsigned char *sha1)
1916 struct stat st;
1917 if (stat(sha1_pack_name(sha1), &st))
1918 return 0;
1919 return 1;
1922 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1924 struct pack_entry e;
1925 return find_pack_entry(sha1, &e, ignore_packed);
1928 int has_sha1_file(const unsigned char *sha1)
1930 struct stat st;
1931 struct pack_entry e;
1933 if (find_pack_entry(sha1, &e, NULL))
1934 return 1;
1935 return find_sha1_file(sha1, &st) ? 1 : 0;
1939 * reads from fd as long as possible into a supplied buffer of size bytes.
1940 * If necessary the buffer's size is increased using realloc()
1942 * returns 0 if anything went fine and -1 otherwise
1944 * NOTE: both buf and size may change, but even when -1 is returned
1945 * you still have to free() it yourself.
1947 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
1949 char* buf = *return_buf;
1950 unsigned long size = *return_size;
1951 int iret;
1952 unsigned long off = 0;
1954 do {
1955 iret = xread(fd, buf + off, size - off);
1956 if (iret > 0) {
1957 off += iret;
1958 if (off == size) {
1959 size *= 2;
1960 buf = xrealloc(buf, size);
1963 } while (iret > 0);
1965 *return_buf = buf;
1966 *return_size = off;
1968 if (iret < 0)
1969 return -1;
1970 return 0;
1973 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
1975 unsigned long size = 4096;
1976 char *buf = xmalloc(size);
1977 int ret;
1979 if (read_pipe(fd, &buf, &size)) {
1980 free(buf);
1981 return -1;
1984 if (!type)
1985 type = blob_type;
1986 if (write_object)
1987 ret = write_sha1_file(buf, size, type, sha1);
1988 else
1989 ret = hash_sha1_file(buf, size, type, sha1);
1990 free(buf);
1991 return ret;
1994 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1996 unsigned long size = st->st_size;
1997 void *buf;
1998 int ret;
2000 buf = "";
2001 if (size)
2002 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2003 close(fd);
2005 if (!type)
2006 type = blob_type;
2007 if (write_object)
2008 ret = write_sha1_file(buf, size, type, sha1);
2009 else
2010 ret = hash_sha1_file(buf, size, type, sha1);
2011 if (size)
2012 munmap(buf, size);
2013 return ret;
2016 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2018 int fd;
2019 char *target;
2021 switch (st->st_mode & S_IFMT) {
2022 case S_IFREG:
2023 fd = open(path, O_RDONLY);
2024 if (fd < 0)
2025 return error("open(\"%s\"): %s", path,
2026 strerror(errno));
2027 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
2028 return error("%s: failed to insert into database",
2029 path);
2030 break;
2031 case S_IFLNK:
2032 target = xmalloc(st->st_size+1);
2033 if (readlink(path, target, st->st_size+1) != st->st_size) {
2034 char *errstr = strerror(errno);
2035 free(target);
2036 return error("readlink(\"%s\"): %s", path,
2037 errstr);
2039 if (!write_object)
2040 hash_sha1_file(target, st->st_size, blob_type, sha1);
2041 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
2042 return error("%s: failed to insert into database",
2043 path);
2044 free(target);
2045 break;
2046 default:
2047 return error("%s: unsupported file type", path);
2049 return 0;
2052 int read_pack_header(int fd, struct pack_header *header)
2054 char *c = (char*)header;
2055 ssize_t remaining = sizeof(struct pack_header);
2056 do {
2057 ssize_t r = xread(fd, c, remaining);
2058 if (r <= 0)
2059 /* "eof before pack header was fully read" */
2060 return PH_ERROR_EOF;
2061 remaining -= r;
2062 c += r;
2063 } while (remaining > 0);
2064 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2065 /* "protocol error (pack signature mismatch detected)" */
2066 return PH_ERROR_PACK_SIGNATURE;
2067 if (!pack_version_ok(header->hdr_version))
2068 /* "protocol error (pack version unsupported)" */
2069 return PH_ERROR_PROTOCOL;
2070 return 0;