mingw: always chmod(, 0666) before unlink()
[git/mingw/4msysgit.git] / sha1_file.c
blob14ec6047978352fe32b66d93c2d05c7304c312c9
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 #ifndef __MINGW32__
576 /* We leave these file descriptors open with sliding mmap;
577 * there is no point keeping them open across exec(), though.
579 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
580 if (fd_flag < 0)
581 die("cannot determine file descriptor flags");
582 fd_flag |= FD_CLOEXEC;
583 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
584 die("cannot set FD_CLOEXEC");
585 #endif
587 /* Verify we recognize this pack file format. */
588 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
589 die("file %s is far too short to be a packfile", p->pack_name);
590 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
591 die("file %s is not a GIT packfile", p->pack_name);
592 if (!pack_version_ok(hdr.hdr_version))
593 die("packfile %s is version %u and not supported"
594 " (try upgrading GIT to a newer version)",
595 p->pack_name, ntohl(hdr.hdr_version));
597 /* Verify the pack matches its index. */
598 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
599 die("packfile %s claims to have %u objects"
600 " while index size indicates %u objects",
601 p->pack_name, ntohl(hdr.hdr_entries),
602 num_packed_objects(p));
603 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
604 die("end of packfile %s is unavailable", p->pack_name);
605 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
606 die("packfile %s signature is unavailable", p->pack_name);
607 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
608 if (hashcmp(sha1, idx_sha1))
609 die("packfile %s does not match index", p->pack_name);
612 static int in_window(struct pack_window *win, unsigned long offset)
614 /* We must promise at least 20 bytes (one hash) after the
615 * offset is available from this window, otherwise the offset
616 * is not actually in this window and a different window (which
617 * has that one hash excess) must be used. This is to support
618 * the object header and delta base parsing routines below.
620 off_t win_off = win->offset;
621 return win_off <= offset
622 && (offset + 20) <= (win_off + win->len);
625 unsigned char* use_pack(struct packed_git *p,
626 struct pack_window **w_cursor,
627 unsigned long offset,
628 unsigned int *left)
630 struct pack_window *win = *w_cursor;
632 if (p->pack_fd == -1)
633 open_packed_git(p);
635 /* Since packfiles end in a hash of their content and its
636 * pointless to ask for an offset into the middle of that
637 * hash, and the in_window function above wouldn't match
638 * don't allow an offset too close to the end of the file.
640 if (offset > (p->pack_size - 20))
641 die("offset beyond end of packfile (truncated pack?)");
643 if (!win || !in_window(win, offset)) {
644 if (win)
645 win->inuse_cnt--;
646 for (win = p->windows; win; win = win->next) {
647 if (in_window(win, offset))
648 break;
650 if (!win) {
651 if (!page_size)
652 page_size = getpagesize();
653 win = xcalloc(1, sizeof(*win));
654 win->offset = (offset / page_size) * page_size;
655 win->len = p->pack_size - win->offset;
656 if (win->len > packed_git_window_size)
657 win->len = packed_git_window_size;
658 pack_mapped += win->len;
659 while (packed_git_limit < pack_mapped
660 && unuse_one_window(p))
661 ; /* nothing */
662 win->base = xmmap(NULL, win->len,
663 PROT_READ, MAP_PRIVATE,
664 p->pack_fd, win->offset);
665 if (win->base == MAP_FAILED)
666 die("packfile %s cannot be mapped: %s",
667 p->pack_name,
668 strerror(errno));
669 pack_mmap_calls++;
670 pack_open_windows++;
671 if (pack_mapped > peak_pack_mapped)
672 peak_pack_mapped = pack_mapped;
673 if (pack_open_windows > peak_pack_open_windows)
674 peak_pack_open_windows = pack_open_windows;
675 win->next = p->windows;
676 p->windows = win;
679 if (win != *w_cursor) {
680 win->last_used = pack_used_ctr++;
681 win->inuse_cnt++;
682 *w_cursor = win;
684 offset -= win->offset;
685 if (left)
686 *left = win->len - offset;
687 return win->base + offset;
690 struct packed_git *add_packed_git(char *path, int path_len, int local)
692 struct stat st;
693 struct packed_git *p;
694 unsigned long idx_size;
695 void *idx_map;
696 unsigned char sha1[20];
698 if (check_packed_git_idx(path, &idx_size, &idx_map))
699 return NULL;
701 /* do we have a corresponding .pack file? */
702 strcpy(path + path_len - 4, ".pack");
703 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
704 munmap(idx_map, idx_size);
705 return NULL;
707 /* ok, it looks sane as far as we can check without
708 * actually mapping the pack file.
710 p = xmalloc(sizeof(*p) + path_len + 2);
711 strcpy(p->pack_name, path);
712 p->index_size = idx_size;
713 p->pack_size = st.st_size;
714 p->index_base = idx_map;
715 p->next = NULL;
716 p->windows = NULL;
717 p->pack_fd = -1;
718 p->pack_local = local;
719 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
720 hashcpy(p->sha1, sha1);
721 return p;
724 struct packed_git *parse_pack_index(unsigned char *sha1)
726 char *path = sha1_pack_index_name(sha1);
727 return parse_pack_index_file(sha1, path);
730 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
732 struct packed_git *p;
733 unsigned long idx_size;
734 void *idx_map;
735 char *path;
737 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
738 return NULL;
740 path = sha1_pack_name(sha1);
742 p = xmalloc(sizeof(*p) + strlen(path) + 2);
743 strcpy(p->pack_name, path);
744 p->index_size = idx_size;
745 p->pack_size = 0;
746 p->index_base = idx_map;
747 p->next = NULL;
748 p->windows = NULL;
749 p->pack_fd = -1;
750 hashcpy(p->sha1, sha1);
751 return p;
754 void install_packed_git(struct packed_git *pack)
756 pack->next = packed_git;
757 packed_git = pack;
760 static void prepare_packed_git_one(char *objdir, int local)
762 char path[PATH_MAX];
763 int len;
764 DIR *dir;
765 struct dirent *de;
767 sprintf(path, "%s/pack", objdir);
768 len = strlen(path);
769 dir = opendir(path);
770 if (!dir) {
771 if (errno != ENOENT)
772 error("unable to open object pack directory: %s: %s",
773 path, strerror(errno));
774 return;
776 path[len++] = '/';
777 while ((de = readdir(dir)) != NULL) {
778 int namelen = strlen(de->d_name);
779 struct packed_git *p;
781 if (!has_extension(de->d_name, ".idx"))
782 continue;
784 /* we have .idx. Is it a file we can map? */
785 strcpy(path + len, de->d_name);
786 for (p = packed_git; p; p = p->next) {
787 if (!memcmp(path, p->pack_name, len + namelen - 4))
788 break;
790 if (p)
791 continue;
792 p = add_packed_git(path, len + namelen, local);
793 if (!p)
794 continue;
795 p->next = packed_git;
796 packed_git = p;
798 closedir(dir);
801 static int prepare_packed_git_run_once = 0;
802 void prepare_packed_git(void)
804 struct alternate_object_database *alt;
806 if (prepare_packed_git_run_once)
807 return;
808 prepare_packed_git_one(get_object_directory(), 1);
809 prepare_alt_odb();
810 for (alt = alt_odb_list; alt; alt = alt->next) {
811 alt->name[-1] = 0;
812 prepare_packed_git_one(alt->base, 0);
813 alt->name[-1] = '/';
815 prepare_packed_git_run_once = 1;
818 void reprepare_packed_git(void)
820 prepare_packed_git_run_once = 0;
821 prepare_packed_git();
824 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
826 unsigned char real_sha1[20];
827 hash_sha1_file(map, size, type, real_sha1);
828 return hashcmp(sha1, real_sha1) ? -1 : 0;
831 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
833 struct stat st;
834 void *map;
835 int fd;
836 char *filename = find_sha1_file(sha1, &st);
838 if (!filename) {
839 return NULL;
842 fd = open(filename, O_RDONLY | sha1_file_open_flag);
843 if (fd < 0) {
844 /* See if it works without O_NOATIME */
845 switch (sha1_file_open_flag) {
846 default:
847 fd = open(filename, O_RDONLY);
848 if (fd >= 0)
849 break;
850 /* Fallthrough */
851 case 0:
852 return NULL;
855 /* If it failed once, it will probably fail again.
856 * Stop using O_NOATIME
858 sha1_file_open_flag = 0;
860 map = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
861 close(fd);
862 *size = st.st_size;
863 return map;
866 int legacy_loose_object(unsigned char *map)
868 unsigned int word;
871 * Is it a zlib-compressed buffer? If so, the first byte
872 * must be 0x78 (15-bit window size, deflated), and the
873 * first 16-bit word is evenly divisible by 31
875 word = (map[0] << 8) + map[1];
876 if (map[0] == 0x78 && !(word % 31))
877 return 1;
878 else
879 return 0;
882 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
884 unsigned shift;
885 unsigned char c;
886 unsigned long size;
887 unsigned long used = 0;
889 c = buf[used++];
890 *type = (c >> 4) & 7;
891 size = c & 15;
892 shift = 4;
893 while (c & 0x80) {
894 if (len <= used)
895 return 0;
896 if (sizeof(long) * 8 <= shift)
897 return 0;
898 c = buf[used++];
899 size += (c & 0x7f) << shift;
900 shift += 7;
902 *sizep = size;
903 return used;
906 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
908 unsigned long size, used;
909 static const char valid_loose_object_type[8] = {
910 0, /* OBJ_EXT */
911 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
912 0, /* "delta" and others are invalid in a loose object */
914 enum object_type type;
916 /* Get the data stream */
917 memset(stream, 0, sizeof(*stream));
918 stream->next_in = map;
919 stream->avail_in = mapsize;
920 stream->next_out = buffer;
921 stream->avail_out = bufsiz;
923 if (legacy_loose_object(map)) {
924 inflateInit(stream);
925 return inflate(stream, 0);
928 used = unpack_object_header_gently(map, mapsize, &type, &size);
929 if (!used || !valid_loose_object_type[type])
930 return -1;
931 map += used;
932 mapsize -= used;
934 /* Set up the stream for the rest.. */
935 stream->next_in = map;
936 stream->avail_in = mapsize;
937 inflateInit(stream);
939 /* And generate the fake traditional header */
940 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
941 type_names[type], size);
942 return 0;
945 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
947 int bytes = strlen(buffer) + 1;
948 unsigned char *buf = xmalloc(1+size);
949 unsigned long n;
951 n = stream->total_out - bytes;
952 if (n > size)
953 n = size;
954 memcpy(buf, (char *) buffer + bytes, n);
955 bytes = n;
956 if (bytes < size) {
957 stream->next_out = buf + bytes;
958 stream->avail_out = size - bytes;
959 while (inflate(stream, Z_FINISH) == Z_OK)
960 /* nothing */;
962 buf[size] = 0;
963 inflateEnd(stream);
964 return buf;
968 * We used to just use "sscanf()", but that's actually way
969 * too permissive for what we want to check. So do an anal
970 * object header parse by hand.
972 static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
974 int i;
975 unsigned long size;
978 * The type can be at most ten bytes (including the
979 * terminating '\0' that we add), and is followed by
980 * a space.
982 i = 10;
983 for (;;) {
984 char c = *hdr++;
985 if (c == ' ')
986 break;
987 if (!--i)
988 return -1;
989 *type++ = c;
991 *type = 0;
994 * The length must follow immediately, and be in canonical
995 * decimal format (ie "010" is not valid).
997 size = *hdr++ - '0';
998 if (size > 9)
999 return -1;
1000 if (size) {
1001 for (;;) {
1002 unsigned long c = *hdr - '0';
1003 if (c > 9)
1004 break;
1005 hdr++;
1006 size = size * 10 + c;
1009 *sizep = size;
1012 * The length must be followed by a zero byte
1014 return *hdr ? -1 : 0;
1017 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
1019 int ret;
1020 z_stream stream;
1021 char hdr[8192];
1023 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1024 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
1025 return NULL;
1027 return unpack_sha1_rest(&stream, hdr, *size);
1030 static unsigned long get_delta_base(struct packed_git *p,
1031 struct pack_window **w_curs,
1032 unsigned long offset,
1033 enum object_type kind,
1034 unsigned long delta_obj_offset,
1035 unsigned long *base_obj_offset)
1037 unsigned char *base_info = use_pack(p, w_curs, offset, NULL);
1038 unsigned long base_offset;
1040 /* use_pack() assured us we have [base_info, base_info + 20)
1041 * as a range that we can look at without walking off the
1042 * end of the mapped window. Its actually the hash size
1043 * that is assured. An OFS_DELTA longer than the hash size
1044 * is stupid, as then a REF_DELTA would be smaller to store.
1046 if (kind == OBJ_OFS_DELTA) {
1047 unsigned used = 0;
1048 unsigned char c = base_info[used++];
1049 base_offset = c & 127;
1050 while (c & 128) {
1051 base_offset += 1;
1052 if (!base_offset || base_offset & ~(~0UL >> 7))
1053 die("offset value overflow for delta base object");
1054 c = base_info[used++];
1055 base_offset = (base_offset << 7) + (c & 127);
1057 base_offset = delta_obj_offset - base_offset;
1058 if (base_offset >= delta_obj_offset)
1059 die("delta base offset out of bound");
1060 offset += used;
1061 } else if (kind == OBJ_REF_DELTA) {
1062 /* The base entry _must_ be in the same pack */
1063 base_offset = find_pack_entry_one(base_info, p);
1064 if (!base_offset)
1065 die("failed to find delta-pack base object %s",
1066 sha1_to_hex(base_info));
1067 offset += 20;
1068 } else
1069 die("I am totally screwed");
1070 *base_obj_offset = base_offset;
1071 return offset;
1074 /* forward declaration for a mutually recursive function */
1075 static int packed_object_info(struct packed_git *p, unsigned long offset,
1076 char *type, unsigned long *sizep);
1078 static int packed_delta_info(struct packed_git *p,
1079 struct pack_window **w_curs,
1080 unsigned long offset,
1081 enum object_type kind,
1082 unsigned long obj_offset,
1083 char *type,
1084 unsigned long *sizep)
1086 unsigned long base_offset;
1088 offset = get_delta_base(p, w_curs, offset, kind,
1089 obj_offset, &base_offset);
1091 /* We choose to only get the type of the base object and
1092 * ignore potentially corrupt pack file that expects the delta
1093 * based on a base with a wrong size. This saves tons of
1094 * inflate() calls.
1096 if (packed_object_info(p, base_offset, type, NULL))
1097 die("cannot get info for delta-pack base");
1099 if (sizep) {
1100 const unsigned char *data;
1101 unsigned char delta_head[20], *in;
1102 unsigned long result_size;
1103 z_stream stream;
1104 int st;
1106 memset(&stream, 0, sizeof(stream));
1107 stream.next_out = delta_head;
1108 stream.avail_out = sizeof(delta_head);
1110 inflateInit(&stream);
1111 do {
1112 in = use_pack(p, w_curs, offset, &stream.avail_in);
1113 stream.next_in = in;
1114 st = inflate(&stream, Z_FINISH);
1115 offset += stream.next_in - in;
1116 } while ((st == Z_OK || st == Z_BUF_ERROR)
1117 && stream.total_out < sizeof(delta_head));
1118 inflateEnd(&stream);
1119 if ((st != Z_STREAM_END) &&
1120 stream.total_out != sizeof(delta_head))
1121 die("delta data unpack-initial failed");
1123 /* Examine the initial part of the delta to figure out
1124 * the result size.
1126 data = delta_head;
1128 /* ignore base size */
1129 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1131 /* Read the result size */
1132 result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1133 *sizep = result_size;
1135 return 0;
1138 static unsigned long unpack_object_header(struct packed_git *p,
1139 struct pack_window **w_curs,
1140 unsigned long offset,
1141 enum object_type *type,
1142 unsigned long *sizep)
1144 unsigned char *base;
1145 unsigned int left;
1146 unsigned long used;
1148 /* use_pack() assures us we have [base, base + 20) available
1149 * as a range that we can look at at. (Its actually the hash
1150 * size that is assurred.) With our object header encoding
1151 * the maximum deflated object size is 2^137, which is just
1152 * insane, so we know won't exceed what we have been given.
1154 base = use_pack(p, w_curs, offset, &left);
1155 used = unpack_object_header_gently(base, left, type, sizep);
1156 if (!used)
1157 die("object offset outside of pack file");
1159 return offset + used;
1162 void packed_object_info_detail(struct packed_git *p,
1163 unsigned long offset,
1164 char *type,
1165 unsigned long *size,
1166 unsigned long *store_size,
1167 unsigned int *delta_chain_length,
1168 unsigned char *base_sha1)
1170 struct pack_window *w_curs = NULL;
1171 unsigned long obj_offset, val;
1172 unsigned char *next_sha1;
1173 enum object_type kind;
1175 *delta_chain_length = 0;
1176 obj_offset = offset;
1177 offset = unpack_object_header(p, &w_curs, offset, &kind, size);
1179 for (;;) {
1180 switch (kind) {
1181 default:
1182 die("pack %s contains unknown object type %d",
1183 p->pack_name, kind);
1184 case OBJ_COMMIT:
1185 case OBJ_TREE:
1186 case OBJ_BLOB:
1187 case OBJ_TAG:
1188 strcpy(type, type_names[kind]);
1189 *store_size = 0; /* notyet */
1190 unuse_pack(&w_curs);
1191 return;
1192 case OBJ_OFS_DELTA:
1193 get_delta_base(p, &w_curs, offset, kind,
1194 obj_offset, &offset);
1195 if (*delta_chain_length == 0) {
1196 /* TODO: find base_sha1 as pointed by offset */
1198 break;
1199 case OBJ_REF_DELTA:
1200 next_sha1 = use_pack(p, &w_curs, offset, NULL);
1201 if (*delta_chain_length == 0)
1202 hashcpy(base_sha1, next_sha1);
1203 offset = find_pack_entry_one(next_sha1, p);
1204 break;
1206 obj_offset = offset;
1207 offset = unpack_object_header(p, &w_curs, offset, &kind, &val);
1208 (*delta_chain_length)++;
1212 static int packed_object_info(struct packed_git *p, unsigned long offset,
1213 char *type, unsigned long *sizep)
1215 struct pack_window *w_curs = NULL;
1216 unsigned long size, obj_offset = offset;
1217 enum object_type kind;
1218 int r;
1220 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1222 switch (kind) {
1223 case OBJ_OFS_DELTA:
1224 case OBJ_REF_DELTA:
1225 r = packed_delta_info(p, &w_curs, offset, kind,
1226 obj_offset, type, sizep);
1227 unuse_pack(&w_curs);
1228 return r;
1229 case OBJ_COMMIT:
1230 case OBJ_TREE:
1231 case OBJ_BLOB:
1232 case OBJ_TAG:
1233 strcpy(type, type_names[kind]);
1234 unuse_pack(&w_curs);
1235 break;
1236 default:
1237 die("pack %s contains unknown object type %d",
1238 p->pack_name, kind);
1240 if (sizep)
1241 *sizep = size;
1242 return 0;
1245 static void *unpack_compressed_entry(struct packed_git *p,
1246 struct pack_window **w_curs,
1247 unsigned long offset,
1248 unsigned long size)
1250 int st;
1251 z_stream stream;
1252 unsigned char *buffer, *in;
1254 buffer = xmalloc(size + 1);
1255 buffer[size] = 0;
1256 memset(&stream, 0, sizeof(stream));
1257 stream.next_out = buffer;
1258 stream.avail_out = size;
1260 inflateInit(&stream);
1261 do {
1262 in = use_pack(p, w_curs, offset, &stream.avail_in);
1263 stream.next_in = in;
1264 st = inflate(&stream, Z_FINISH);
1265 offset += stream.next_in - in;
1266 } while (st == Z_OK || st == Z_BUF_ERROR);
1267 inflateEnd(&stream);
1268 if ((st != Z_STREAM_END) || stream.total_out != size) {
1269 free(buffer);
1270 return NULL;
1273 return buffer;
1276 static void *unpack_delta_entry(struct packed_git *p,
1277 struct pack_window **w_curs,
1278 unsigned long offset,
1279 unsigned long delta_size,
1280 enum object_type kind,
1281 unsigned long obj_offset,
1282 char *type,
1283 unsigned long *sizep)
1285 void *delta_data, *result, *base;
1286 unsigned long result_size, base_size, base_offset;
1288 offset = get_delta_base(p, w_curs, offset, kind,
1289 obj_offset, &base_offset);
1290 base = unpack_entry(p, base_offset, type, &base_size);
1291 if (!base)
1292 die("failed to read delta base object at %lu from %s",
1293 base_offset, p->pack_name);
1295 delta_data = unpack_compressed_entry(p, w_curs, offset, delta_size);
1296 result = patch_delta(base, base_size,
1297 delta_data, delta_size,
1298 &result_size);
1299 if (!result)
1300 die("failed to apply delta");
1301 free(delta_data);
1302 free(base);
1303 *sizep = result_size;
1304 return result;
1307 void *unpack_entry(struct packed_git *p, unsigned long offset,
1308 char *type, unsigned long *sizep)
1310 struct pack_window *w_curs = NULL;
1311 unsigned long size, obj_offset = offset;
1312 enum object_type kind;
1313 void *retval;
1315 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1316 switch (kind) {
1317 case OBJ_OFS_DELTA:
1318 case OBJ_REF_DELTA:
1319 retval = unpack_delta_entry(p, &w_curs, offset, size,
1320 kind, obj_offset, type, sizep);
1321 break;
1322 case OBJ_COMMIT:
1323 case OBJ_TREE:
1324 case OBJ_BLOB:
1325 case OBJ_TAG:
1326 strcpy(type, type_names[kind]);
1327 *sizep = size;
1328 retval = unpack_compressed_entry(p, &w_curs, offset, size);
1329 break;
1330 default:
1331 die("unknown object type %i in %s", kind, p->pack_name);
1333 unuse_pack(&w_curs);
1334 return retval;
1337 int num_packed_objects(const struct packed_git *p)
1339 /* See check_packed_git_idx() */
1340 return (p->index_size - 20 - 20 - 4*256) / 24;
1343 int nth_packed_object_sha1(const struct packed_git *p, int n,
1344 unsigned char* sha1)
1346 void *index = p->index_base + 256;
1347 if (n < 0 || num_packed_objects(p) <= n)
1348 return -1;
1349 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1350 return 0;
1353 unsigned long find_pack_entry_one(const unsigned char *sha1,
1354 struct packed_git *p)
1356 uint32_t *level1_ofs = p->index_base;
1357 int hi = ntohl(level1_ofs[*sha1]);
1358 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1359 void *index = p->index_base + 256;
1361 do {
1362 int mi = (lo + hi) / 2;
1363 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1364 if (!cmp)
1365 return ntohl(*((uint32_t *)((char *)index + (24 * mi))));
1366 if (cmp > 0)
1367 hi = mi;
1368 else
1369 lo = mi+1;
1370 } while (lo < hi);
1371 return 0;
1374 static int matches_pack_name(struct packed_git *p, const char *ig)
1376 const char *last_c, *c;
1378 if (!strcmp(p->pack_name, ig))
1379 return 0;
1381 for (c = p->pack_name, last_c = c; *c;)
1382 if (*c == '/')
1383 last_c = ++c;
1384 else
1385 ++c;
1386 if (!strcmp(last_c, ig))
1387 return 0;
1389 return 1;
1392 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1394 struct packed_git *p;
1395 unsigned long offset;
1397 prepare_packed_git();
1399 for (p = packed_git; p; p = p->next) {
1400 if (ignore_packed) {
1401 const char **ig;
1402 for (ig = ignore_packed; *ig; ig++)
1403 if (!matches_pack_name(p, *ig))
1404 break;
1405 if (*ig)
1406 continue;
1408 offset = find_pack_entry_one(sha1, p);
1409 if (offset) {
1410 e->offset = offset;
1411 e->p = p;
1412 hashcpy(e->sha1, sha1);
1413 return 1;
1416 return 0;
1419 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1420 struct packed_git *packs)
1422 struct packed_git *p;
1424 for (p = packs; p; p = p->next) {
1425 if (find_pack_entry_one(sha1, p))
1426 return p;
1428 return NULL;
1432 static int sha1_loose_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1434 int status;
1435 unsigned long mapsize, size;
1436 void *map;
1437 z_stream stream;
1438 char hdr[128];
1440 map = map_sha1_file(sha1, &mapsize);
1441 if (!map)
1442 return error("unable to find %s", sha1_to_hex(sha1));
1443 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1444 status = error("unable to unpack %s header",
1445 sha1_to_hex(sha1));
1446 if (parse_sha1_header(hdr, type, &size) < 0)
1447 status = error("unable to parse %s header", sha1_to_hex(sha1));
1448 else {
1449 status = 0;
1450 if (sizep)
1451 *sizep = size;
1453 inflateEnd(&stream);
1454 munmap(map, mapsize);
1455 return status;
1458 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1460 struct pack_entry e;
1462 if (!find_pack_entry(sha1, &e, NULL)) {
1463 reprepare_packed_git();
1464 if (!find_pack_entry(sha1, &e, NULL))
1465 return sha1_loose_object_info(sha1, type, sizep);
1467 return packed_object_info(e.p, e.offset, type, sizep);
1470 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1472 struct pack_entry e;
1474 if (!find_pack_entry(sha1, &e, NULL)) {
1475 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1476 return NULL;
1478 return unpack_entry(e.p, e.offset, type, size);
1481 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1483 unsigned long mapsize;
1484 void *map, *buf;
1485 struct pack_entry e;
1487 if (find_pack_entry(sha1, &e, NULL))
1488 return read_packed_sha1(sha1, type, size);
1489 map = map_sha1_file(sha1, &mapsize);
1490 if (map) {
1491 buf = unpack_sha1_file(map, mapsize, type, size);
1492 munmap(map, mapsize);
1493 return buf;
1495 reprepare_packed_git();
1496 if (find_pack_entry(sha1, &e, NULL))
1497 return read_packed_sha1(sha1, type, size);
1498 return NULL;
1501 void *read_object_with_reference(const unsigned char *sha1,
1502 const char *required_type,
1503 unsigned long *size,
1504 unsigned char *actual_sha1_return)
1506 char type[20];
1507 void *buffer;
1508 unsigned long isize;
1509 unsigned char actual_sha1[20];
1511 hashcpy(actual_sha1, sha1);
1512 while (1) {
1513 int ref_length = -1;
1514 const char *ref_type = NULL;
1516 buffer = read_sha1_file(actual_sha1, type, &isize);
1517 if (!buffer)
1518 return NULL;
1519 if (!strcmp(type, required_type)) {
1520 *size = isize;
1521 if (actual_sha1_return)
1522 hashcpy(actual_sha1_return, actual_sha1);
1523 return buffer;
1525 /* Handle references */
1526 else if (!strcmp(type, commit_type))
1527 ref_type = "tree ";
1528 else if (!strcmp(type, tag_type))
1529 ref_type = "object ";
1530 else {
1531 free(buffer);
1532 return NULL;
1534 ref_length = strlen(ref_type);
1536 if (memcmp(buffer, ref_type, ref_length) ||
1537 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1538 free(buffer);
1539 return NULL;
1541 free(buffer);
1542 /* Now we have the ID of the referred-to object in
1543 * actual_sha1. Check again. */
1547 static void write_sha1_file_prepare(void *buf, unsigned long len,
1548 const char *type, unsigned char *sha1,
1549 unsigned char *hdr, int *hdrlen)
1551 SHA_CTX c;
1553 /* Generate the header */
1554 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1556 /* Sha1.. */
1557 SHA1_Init(&c);
1558 SHA1_Update(&c, hdr, *hdrlen);
1559 SHA1_Update(&c, buf, len);
1560 SHA1_Final(sha1, &c);
1564 * Link the tempfile to the final place, possibly creating the
1565 * last directory level as you do so.
1567 * Returns the errno on failure, 0 on success.
1569 static int link_temp_to_file(const char *tmpfile, const char *filename)
1571 int ret;
1572 char *dir;
1574 if (!link(tmpfile, filename))
1575 return 0;
1578 * Try to mkdir the last path component if that failed.
1580 * Re-try the "link()" regardless of whether the mkdir
1581 * succeeds, since a race might mean that somebody
1582 * else succeeded.
1584 ret = errno;
1585 dir = strrchr(filename, '/');
1586 if (dir) {
1587 *dir = 0;
1588 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1589 *dir = '/';
1590 return -2;
1592 *dir = '/';
1593 if (!link(tmpfile, filename))
1594 return 0;
1595 ret = errno;
1597 return ret;
1601 * Move the just written object into its final resting place
1603 int move_temp_to_file(const char *tmpfile, const char *filename)
1605 int ret = link_temp_to_file(tmpfile, filename);
1608 * Coda hack - coda doesn't like cross-directory links,
1609 * so we fall back to a rename, which will mean that it
1610 * won't be able to check collisions, but that's not a
1611 * big deal.
1613 * The same holds for FAT formatted media.
1615 * When this succeeds, we just return 0. We have nothing
1616 * left to unlink.
1618 if (ret && ret != EEXIST) {
1619 if (!rename(tmpfile, filename))
1620 return 0;
1621 ret = errno;
1623 unlink(tmpfile);
1624 if (ret) {
1625 if (ret != EEXIST) {
1626 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1628 /* FIXME!!! Collision check here ? */
1631 return 0;
1634 static int write_buffer(int fd, const void *buf, size_t len)
1636 if (write_in_full(fd, buf, len) < 0)
1637 return error("file write error (%s)", strerror(errno));
1638 return 0;
1641 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1643 int hdr_len;
1644 unsigned char c;
1646 c = (type << 4) | (len & 15);
1647 len >>= 4;
1648 hdr_len = 1;
1649 while (len) {
1650 *hdr++ = c | 0x80;
1651 hdr_len++;
1652 c = (len & 0x7f);
1653 len >>= 7;
1655 *hdr = c;
1656 return hdr_len;
1659 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1661 int obj_type, hdr;
1663 if (use_legacy_headers) {
1664 while (deflate(stream, 0) == Z_OK)
1665 /* nothing */;
1666 return;
1668 if (!strcmp(type, blob_type))
1669 obj_type = OBJ_BLOB;
1670 else if (!strcmp(type, tree_type))
1671 obj_type = OBJ_TREE;
1672 else if (!strcmp(type, commit_type))
1673 obj_type = OBJ_COMMIT;
1674 else if (!strcmp(type, tag_type))
1675 obj_type = OBJ_TAG;
1676 else
1677 die("trying to generate bogus object of type '%s'", type);
1678 hdr = write_binary_header(stream->next_out, obj_type, len);
1679 stream->total_out = hdr;
1680 stream->next_out += hdr;
1681 stream->avail_out -= hdr;
1684 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1685 unsigned char *sha1)
1687 unsigned char hdr[50];
1688 int hdrlen;
1689 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1690 return 0;
1693 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1695 int size;
1696 unsigned char *compressed;
1697 z_stream stream;
1698 unsigned char sha1[20];
1699 char *filename;
1700 static char tmpfile[PATH_MAX];
1701 unsigned char hdr[50];
1702 int fd, hdrlen;
1704 /* Normally if we have it in the pack then we do not bother writing
1705 * it out into .git/objects/??/?{38} file.
1707 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1708 filename = sha1_file_name(sha1);
1709 if (returnsha1)
1710 hashcpy(returnsha1, sha1);
1711 if (has_sha1_file(sha1))
1712 return 0;
1713 fd = open(filename, O_RDONLY);
1714 if (fd >= 0) {
1716 * FIXME!!! We might do collision checking here, but we'd
1717 * need to uncompress the old file and check it. Later.
1719 close(fd);
1720 return 0;
1723 if (errno != ENOENT) {
1724 return error("sha1 file %s: %s\n", filename, strerror(errno));
1727 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1729 fd = mkstemp(tmpfile);
1730 if (fd < 0) {
1731 if (errno == EPERM)
1732 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1733 else
1734 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1737 /* Set it up */
1738 memset(&stream, 0, sizeof(stream));
1739 deflateInit(&stream, zlib_compression_level);
1740 size = 8 + deflateBound(&stream, len+hdrlen);
1741 compressed = xmalloc(size);
1743 /* Compress it */
1744 stream.next_out = compressed;
1745 stream.avail_out = size;
1747 /* First header.. */
1748 stream.next_in = hdr;
1749 stream.avail_in = hdrlen;
1750 setup_object_header(&stream, type, len);
1752 /* Then the data itself.. */
1753 stream.next_in = buf;
1754 stream.avail_in = len;
1755 while (deflate(&stream, Z_FINISH) == Z_OK)
1756 /* nothing */;
1757 deflateEnd(&stream);
1758 size = stream.total_out;
1760 if (write_buffer(fd, compressed, size) < 0)
1761 die("unable to write sha1 file");
1762 fchmod(fd, 0444);
1763 close(fd);
1764 free(compressed);
1766 return move_temp_to_file(tmpfile, filename);
1770 * We need to unpack and recompress the object for writing
1771 * it out to a different file.
1773 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1775 size_t size;
1776 z_stream stream;
1777 unsigned char *unpacked;
1778 unsigned long len;
1779 char type[20];
1780 char hdr[50];
1781 int hdrlen;
1782 void *buf;
1784 /* need to unpack and recompress it by itself */
1785 unpacked = read_packed_sha1(sha1, type, &len);
1787 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1789 /* Set it up */
1790 memset(&stream, 0, sizeof(stream));
1791 deflateInit(&stream, zlib_compression_level);
1792 size = deflateBound(&stream, len + hdrlen);
1793 buf = xmalloc(size);
1795 /* Compress it */
1796 stream.next_out = buf;
1797 stream.avail_out = size;
1799 /* First header.. */
1800 stream.next_in = (void *)hdr;
1801 stream.avail_in = hdrlen;
1802 while (deflate(&stream, 0) == Z_OK)
1803 /* nothing */;
1805 /* Then the data itself.. */
1806 stream.next_in = unpacked;
1807 stream.avail_in = len;
1808 while (deflate(&stream, Z_FINISH) == Z_OK)
1809 /* nothing */;
1810 deflateEnd(&stream);
1811 free(unpacked);
1813 *objsize = stream.total_out;
1814 return buf;
1817 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1819 int retval;
1820 unsigned long objsize;
1821 void *buf = map_sha1_file(sha1, &objsize);
1823 if (buf) {
1824 retval = write_buffer(fd, buf, objsize);
1825 munmap(buf, objsize);
1826 return retval;
1829 buf = repack_object(sha1, &objsize);
1830 retval = write_buffer(fd, buf, objsize);
1831 free(buf);
1832 return retval;
1835 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1836 size_t bufsize, size_t *bufposn)
1838 char tmpfile[PATH_MAX];
1839 int local;
1840 z_stream stream;
1841 unsigned char real_sha1[20];
1842 unsigned char discard[4096];
1843 int ret;
1844 SHA_CTX c;
1846 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1848 local = mkstemp(tmpfile);
1849 if (local < 0) {
1850 if (errno == EPERM)
1851 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1852 else
1853 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1856 memset(&stream, 0, sizeof(stream));
1858 inflateInit(&stream);
1860 SHA1_Init(&c);
1862 do {
1863 ssize_t size;
1864 if (*bufposn) {
1865 stream.avail_in = *bufposn;
1866 stream.next_in = (unsigned char *) buffer;
1867 do {
1868 stream.next_out = discard;
1869 stream.avail_out = sizeof(discard);
1870 ret = inflate(&stream, Z_SYNC_FLUSH);
1871 SHA1_Update(&c, discard, sizeof(discard) -
1872 stream.avail_out);
1873 } while (stream.avail_in && ret == Z_OK);
1874 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1875 die("unable to write sha1 file");
1876 memmove(buffer, buffer + *bufposn - stream.avail_in,
1877 stream.avail_in);
1878 *bufposn = stream.avail_in;
1879 if (ret != Z_OK)
1880 break;
1882 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
1883 if (size <= 0) {
1884 close(local);
1885 unlink(tmpfile);
1886 if (!size)
1887 return error("Connection closed?");
1888 perror("Reading from connection");
1889 return -1;
1891 *bufposn += size;
1892 } while (1);
1893 inflateEnd(&stream);
1895 close(local);
1896 SHA1_Final(real_sha1, &c);
1897 if (ret != Z_STREAM_END) {
1898 unlink(tmpfile);
1899 return error("File %s corrupted", sha1_to_hex(sha1));
1901 if (hashcmp(sha1, real_sha1)) {
1902 unlink(tmpfile);
1903 return error("File %s has bad hash", sha1_to_hex(sha1));
1906 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1909 int has_pack_index(const unsigned char *sha1)
1911 struct stat st;
1912 if (stat(sha1_pack_index_name(sha1), &st))
1913 return 0;
1914 return 1;
1917 int has_pack_file(const unsigned char *sha1)
1919 struct stat st;
1920 if (stat(sha1_pack_name(sha1), &st))
1921 return 0;
1922 return 1;
1925 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1927 struct pack_entry e;
1928 return find_pack_entry(sha1, &e, ignore_packed);
1931 int has_sha1_file(const unsigned char *sha1)
1933 struct stat st;
1934 struct pack_entry e;
1936 if (find_pack_entry(sha1, &e, NULL))
1937 return 1;
1938 return find_sha1_file(sha1, &st) ? 1 : 0;
1942 * reads from fd as long as possible into a supplied buffer of size bytes.
1943 * If necessary the buffer's size is increased using realloc()
1945 * returns 0 if anything went fine and -1 otherwise
1947 * NOTE: both buf and size may change, but even when -1 is returned
1948 * you still have to free() it yourself.
1950 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
1952 char* buf = *return_buf;
1953 unsigned long size = *return_size;
1954 int iret;
1955 unsigned long off = 0;
1957 do {
1958 iret = xread(fd, buf + off, size - off);
1959 if (iret > 0) {
1960 off += iret;
1961 if (off == size) {
1962 size *= 2;
1963 buf = xrealloc(buf, size);
1966 } while (iret > 0);
1968 *return_buf = buf;
1969 *return_size = off;
1971 if (iret < 0)
1972 return -1;
1973 return 0;
1976 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
1978 unsigned long size = 4096;
1979 char *buf = xmalloc(size);
1980 int ret;
1982 if (read_pipe(fd, &buf, &size)) {
1983 free(buf);
1984 return -1;
1987 if (!type)
1988 type = blob_type;
1989 if (write_object)
1990 ret = write_sha1_file(buf, size, type, sha1);
1991 else
1992 ret = hash_sha1_file(buf, size, type, sha1);
1993 free(buf);
1994 return ret;
1997 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1999 unsigned long size = st->st_size;
2000 void *buf;
2001 int ret;
2003 buf = "";
2004 if (size)
2005 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2006 close(fd);
2008 if (!type)
2009 type = blob_type;
2010 if (write_object)
2011 ret = write_sha1_file(buf, size, type, sha1);
2012 else
2013 ret = hash_sha1_file(buf, size, type, sha1);
2014 if (size)
2015 munmap(buf, size);
2016 return ret;
2019 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2021 int fd;
2022 char *target;
2024 switch (st->st_mode & S_IFMT) {
2025 case S_IFREG:
2026 fd = open(path, O_RDONLY);
2027 if (fd < 0)
2028 return error("open(\"%s\"): %s", path,
2029 strerror(errno));
2030 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
2031 return error("%s: failed to insert into database",
2032 path);
2033 break;
2034 case S_IFLNK:
2035 target = xmalloc(st->st_size+1);
2036 if (readlink(path, target, st->st_size+1) != st->st_size) {
2037 char *errstr = strerror(errno);
2038 free(target);
2039 return error("readlink(\"%s\"): %s", path,
2040 errstr);
2042 if (!write_object)
2043 hash_sha1_file(target, st->st_size, blob_type, sha1);
2044 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
2045 return error("%s: failed to insert into database",
2046 path);
2047 free(target);
2048 break;
2049 default:
2050 return error("%s: unsupported file type", path);
2052 return 0;