Merge branch 'js/diff-ni'
[git/dscho.git] / sha1_file.c
blob7faa8bcd507f54a55cb9e2ac685654921475f995
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 size_t mapsz;
353 struct stat st;
354 char path[PATH_MAX];
355 int fd;
357 sprintf(path, "%s/info/alternates", relative_base);
358 fd = open(path, O_RDONLY);
359 if (fd < 0)
360 return;
361 if (fstat(fd, &st) || (st.st_size == 0)) {
362 close(fd);
363 return;
365 mapsz = xsize_t(st.st_size);
366 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
367 close(fd);
369 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
371 munmap(map, mapsz);
374 void prepare_alt_odb(void)
376 const char *alt;
378 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
379 if (!alt) alt = "";
381 if (alt_odb_tail)
382 return;
383 alt_odb_tail = &alt_odb_list;
384 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
386 read_info_alternates(get_object_directory(), 0);
389 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
391 char *name = sha1_file_name(sha1);
392 struct alternate_object_database *alt;
394 if (!stat(name, st))
395 return name;
396 prepare_alt_odb();
397 for (alt = alt_odb_list; alt; alt = alt->next) {
398 name = alt->name;
399 fill_sha1_path(name, sha1);
400 if (!stat(alt->base, st))
401 return alt->base;
403 return NULL;
406 static unsigned int pack_used_ctr;
407 static unsigned int pack_mmap_calls;
408 static unsigned int peak_pack_open_windows;
409 static unsigned int pack_open_windows;
410 static size_t peak_pack_mapped;
411 static size_t pack_mapped;
412 struct packed_git *packed_git;
414 void pack_report()
416 fprintf(stderr,
417 "pack_report: getpagesize() = %10" SZ_FMT "\n"
418 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
419 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
420 (size_t) getpagesize(),
421 packed_git_window_size,
422 packed_git_limit);
423 fprintf(stderr,
424 "pack_report: pack_used_ctr = %10u\n"
425 "pack_report: pack_mmap_calls = %10u\n"
426 "pack_report: pack_open_windows = %10u / %10u\n"
427 "pack_report: pack_mapped = "
428 "%10" SZ_FMT " / %10" SZ_FMT "\n",
429 pack_used_ctr,
430 pack_mmap_calls,
431 pack_open_windows, peak_pack_open_windows,
432 pack_mapped, peak_pack_mapped);
435 static int check_packed_git_idx(const char *path,
436 unsigned long *idx_size_,
437 void **idx_map_)
439 void *idx_map;
440 uint32_t *index;
441 size_t idx_size;
442 uint32_t nr, i;
443 int fd = open(path, O_RDONLY);
444 struct stat st;
445 if (fd < 0)
446 return -1;
447 if (fstat(fd, &st)) {
448 close(fd);
449 return -1;
451 idx_size = xsize_t(st.st_size);
452 if (idx_size < 4 * 256 + 20 + 20) {
453 close(fd);
454 return error("index file %s is too small", path);
456 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
457 close(fd);
459 index = idx_map;
460 *idx_map_ = idx_map;
461 *idx_size_ = idx_size;
463 /* a future index format would start with this, as older git
464 * binaries would fail the non-monotonic index check below.
465 * give a nicer warning to the user if we can.
467 if (index[0] == htonl(PACK_IDX_SIGNATURE)) {
468 munmap(idx_map, idx_size);
469 return error("index file %s is a newer version"
470 " and is not supported by this binary"
471 " (try upgrading GIT to a newer version)",
472 path);
475 nr = 0;
476 for (i = 0; i < 256; i++) {
477 uint32_t n = ntohl(index[i]);
478 if (n < nr) {
479 munmap(idx_map, idx_size);
480 return error("non-monotonic index %s", path);
482 nr = n;
486 * Total size:
487 * - 256 index entries 4 bytes each
488 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
489 * - 20-byte SHA1 of the packfile
490 * - 20-byte SHA1 file checksum
492 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
493 munmap(idx_map, idx_size);
494 return error("wrong index file size in %s", path);
497 return 0;
500 static void scan_windows(struct packed_git *p,
501 struct packed_git **lru_p,
502 struct pack_window **lru_w,
503 struct pack_window **lru_l)
505 struct pack_window *w, *w_l;
507 for (w_l = NULL, w = p->windows; w; w = w->next) {
508 if (!w->inuse_cnt) {
509 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
510 *lru_p = p;
511 *lru_w = w;
512 *lru_l = w_l;
515 w_l = w;
519 static int unuse_one_window(struct packed_git *current)
521 struct packed_git *p, *lru_p = NULL;
522 struct pack_window *lru_w = NULL, *lru_l = NULL;
524 if (current)
525 scan_windows(current, &lru_p, &lru_w, &lru_l);
526 for (p = packed_git; p; p = p->next)
527 scan_windows(p, &lru_p, &lru_w, &lru_l);
528 if (lru_p) {
529 munmap(lru_w->base, lru_w->len);
530 pack_mapped -= lru_w->len;
531 if (lru_l)
532 lru_l->next = lru_w->next;
533 else {
534 lru_p->windows = lru_w->next;
535 if (!lru_p->windows && lru_p != current) {
536 close(lru_p->pack_fd);
537 lru_p->pack_fd = -1;
540 free(lru_w);
541 pack_open_windows--;
542 return 1;
544 return 0;
547 void release_pack_memory(size_t need)
549 size_t cur = pack_mapped;
550 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
551 ; /* nothing */
554 void unuse_pack(struct pack_window **w_cursor)
556 struct pack_window *w = *w_cursor;
557 if (w) {
558 w->inuse_cnt--;
559 *w_cursor = NULL;
564 * Do not call this directly as this leaks p->pack_fd on error return;
565 * call open_packed_git() instead.
567 static int open_packed_git_1(struct packed_git *p)
569 struct stat st;
570 struct pack_header hdr;
571 unsigned char sha1[20];
572 unsigned char *idx_sha1;
573 long fd_flag;
575 p->pack_fd = open(p->pack_name, O_RDONLY);
576 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
577 return -1;
579 /* If we created the struct before we had the pack we lack size. */
580 if (!p->pack_size) {
581 if (!S_ISREG(st.st_mode))
582 return error("packfile %s not a regular file", p->pack_name);
583 p->pack_size = st.st_size;
584 } else if (p->pack_size != st.st_size)
585 return error("packfile %s size changed", p->pack_name);
587 /* We leave these file descriptors open with sliding mmap;
588 * there is no point keeping them open across exec(), though.
590 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
591 if (fd_flag < 0)
592 return error("cannot determine file descriptor flags");
593 fd_flag |= FD_CLOEXEC;
594 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
595 return error("cannot set FD_CLOEXEC");
597 /* Verify we recognize this pack file format. */
598 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
599 return error("file %s is far too short to be a packfile", p->pack_name);
600 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
601 return error("file %s is not a GIT packfile", p->pack_name);
602 if (!pack_version_ok(hdr.hdr_version))
603 return error("packfile %s is version %u and not supported"
604 " (try upgrading GIT to a newer version)",
605 p->pack_name, ntohl(hdr.hdr_version));
607 /* Verify the pack matches its index. */
608 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
609 return error("packfile %s claims to have %u objects"
610 " while index size indicates %u objects",
611 p->pack_name, ntohl(hdr.hdr_entries),
612 num_packed_objects(p));
613 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
614 return error("end of packfile %s is unavailable", p->pack_name);
615 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
616 return error("packfile %s signature is unavailable", p->pack_name);
617 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
618 if (hashcmp(sha1, idx_sha1))
619 return error("packfile %s does not match index", p->pack_name);
620 return 0;
623 static int open_packed_git(struct packed_git *p)
625 if (!open_packed_git_1(p))
626 return 0;
627 if (p->pack_fd != -1) {
628 close(p->pack_fd);
629 p->pack_fd = -1;
631 return -1;
634 static int in_window(struct pack_window *win, off_t offset)
636 /* We must promise at least 20 bytes (one hash) after the
637 * offset is available from this window, otherwise the offset
638 * is not actually in this window and a different window (which
639 * has that one hash excess) must be used. This is to support
640 * the object header and delta base parsing routines below.
642 off_t win_off = win->offset;
643 return win_off <= offset
644 && (offset + 20) <= (win_off + win->len);
647 unsigned char* use_pack(struct packed_git *p,
648 struct pack_window **w_cursor,
649 off_t offset,
650 unsigned int *left)
652 struct pack_window *win = *w_cursor;
654 if (p->pack_fd == -1 && open_packed_git(p))
655 die("packfile %s cannot be accessed", p->pack_name);
657 /* Since packfiles end in a hash of their content and its
658 * pointless to ask for an offset into the middle of that
659 * hash, and the in_window function above wouldn't match
660 * don't allow an offset too close to the end of the file.
662 if (offset > (p->pack_size - 20))
663 die("offset beyond end of packfile (truncated pack?)");
665 if (!win || !in_window(win, offset)) {
666 if (win)
667 win->inuse_cnt--;
668 for (win = p->windows; win; win = win->next) {
669 if (in_window(win, offset))
670 break;
672 if (!win) {
673 size_t window_align = packed_git_window_size / 2;
674 off_t len;
675 win = xcalloc(1, sizeof(*win));
676 win->offset = (offset / window_align) * window_align;
677 len = p->pack_size - win->offset;
678 if (len > packed_git_window_size)
679 len = packed_git_window_size;
680 win->len = (size_t)len;
681 pack_mapped += win->len;
682 while (packed_git_limit < pack_mapped
683 && unuse_one_window(p))
684 ; /* nothing */
685 win->base = xmmap(NULL, win->len,
686 PROT_READ, MAP_PRIVATE,
687 p->pack_fd, win->offset);
688 if (win->base == MAP_FAILED)
689 die("packfile %s cannot be mapped: %s",
690 p->pack_name,
691 strerror(errno));
692 pack_mmap_calls++;
693 pack_open_windows++;
694 if (pack_mapped > peak_pack_mapped)
695 peak_pack_mapped = pack_mapped;
696 if (pack_open_windows > peak_pack_open_windows)
697 peak_pack_open_windows = pack_open_windows;
698 win->next = p->windows;
699 p->windows = win;
702 if (win != *w_cursor) {
703 win->last_used = pack_used_ctr++;
704 win->inuse_cnt++;
705 *w_cursor = win;
707 offset -= win->offset;
708 if (left)
709 *left = win->len - xsize_t(offset);
710 return win->base + offset;
713 struct packed_git *add_packed_git(char *path, int path_len, int local)
715 struct stat st;
716 struct packed_git *p;
717 unsigned long idx_size;
718 void *idx_map;
719 unsigned char sha1[20];
721 if (check_packed_git_idx(path, &idx_size, &idx_map))
722 return NULL;
724 /* do we have a corresponding .pack file? */
725 strcpy(path + path_len - 4, ".pack");
726 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
727 munmap(idx_map, idx_size);
728 return NULL;
730 /* ok, it looks sane as far as we can check without
731 * actually mapping the pack file.
733 p = xmalloc(sizeof(*p) + path_len + 2);
734 strcpy(p->pack_name, path);
735 p->index_size = idx_size;
736 p->pack_size = st.st_size;
737 p->index_base = idx_map;
738 p->next = NULL;
739 p->windows = NULL;
740 p->pack_fd = -1;
741 p->pack_local = local;
742 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
743 hashcpy(p->sha1, sha1);
744 return p;
747 struct packed_git *parse_pack_index(unsigned char *sha1)
749 char *path = sha1_pack_index_name(sha1);
750 return parse_pack_index_file(sha1, path);
753 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
755 struct packed_git *p;
756 unsigned long idx_size;
757 void *idx_map;
758 char *path;
760 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
761 return NULL;
763 path = sha1_pack_name(sha1);
765 p = xmalloc(sizeof(*p) + strlen(path) + 2);
766 strcpy(p->pack_name, path);
767 p->index_size = idx_size;
768 p->pack_size = 0;
769 p->index_base = idx_map;
770 p->next = NULL;
771 p->windows = NULL;
772 p->pack_fd = -1;
773 hashcpy(p->sha1, sha1);
774 return p;
777 void install_packed_git(struct packed_git *pack)
779 pack->next = packed_git;
780 packed_git = pack;
783 static void prepare_packed_git_one(char *objdir, int local)
785 char path[PATH_MAX];
786 int len;
787 DIR *dir;
788 struct dirent *de;
790 sprintf(path, "%s/pack", objdir);
791 len = strlen(path);
792 dir = opendir(path);
793 if (!dir) {
794 if (errno != ENOENT)
795 error("unable to open object pack directory: %s: %s",
796 path, strerror(errno));
797 return;
799 path[len++] = '/';
800 while ((de = readdir(dir)) != NULL) {
801 int namelen = strlen(de->d_name);
802 struct packed_git *p;
804 if (!has_extension(de->d_name, ".idx"))
805 continue;
807 /* Don't reopen a pack we already have. */
808 strcpy(path + len, de->d_name);
809 for (p = packed_git; p; p = p->next) {
810 if (!memcmp(path, p->pack_name, len + namelen - 4))
811 break;
813 if (p)
814 continue;
815 /* See if it really is a valid .idx file with corresponding
816 * .pack file that we can map.
818 p = add_packed_git(path, len + namelen, local);
819 if (!p)
820 continue;
821 install_packed_git(p);
823 closedir(dir);
826 static int prepare_packed_git_run_once = 0;
827 void prepare_packed_git(void)
829 struct alternate_object_database *alt;
831 if (prepare_packed_git_run_once)
832 return;
833 prepare_packed_git_one(get_object_directory(), 1);
834 prepare_alt_odb();
835 for (alt = alt_odb_list; alt; alt = alt->next) {
836 alt->name[-1] = 0;
837 prepare_packed_git_one(alt->base, 0);
838 alt->name[-1] = '/';
840 prepare_packed_git_run_once = 1;
843 void reprepare_packed_git(void)
845 prepare_packed_git_run_once = 0;
846 prepare_packed_git();
849 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
851 unsigned char real_sha1[20];
852 hash_sha1_file(map, size, type, real_sha1);
853 return hashcmp(sha1, real_sha1) ? -1 : 0;
856 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
858 struct stat st;
859 void *map;
860 int fd;
861 char *filename = find_sha1_file(sha1, &st);
863 if (!filename) {
864 return NULL;
867 fd = open(filename, O_RDONLY | sha1_file_open_flag);
868 if (fd < 0) {
869 /* See if it works without O_NOATIME */
870 switch (sha1_file_open_flag) {
871 default:
872 fd = open(filename, O_RDONLY);
873 if (fd >= 0)
874 break;
875 /* Fallthrough */
876 case 0:
877 return NULL;
880 /* If it failed once, it will probably fail again.
881 * Stop using O_NOATIME
883 sha1_file_open_flag = 0;
885 *size = xsize_t(st.st_size);
886 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
887 close(fd);
888 return map;
891 int legacy_loose_object(unsigned char *map)
893 unsigned int word;
896 * Is it a zlib-compressed buffer? If so, the first byte
897 * must be 0x78 (15-bit window size, deflated), and the
898 * first 16-bit word is evenly divisible by 31
900 word = (map[0] << 8) + map[1];
901 if (map[0] == 0x78 && !(word % 31))
902 return 1;
903 else
904 return 0;
907 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
909 unsigned shift;
910 unsigned char c;
911 unsigned long size;
912 unsigned long used = 0;
914 c = buf[used++];
915 *type = (c >> 4) & 7;
916 size = c & 15;
917 shift = 4;
918 while (c & 0x80) {
919 if (len <= used)
920 return 0;
921 if (sizeof(long) * 8 <= shift)
922 return 0;
923 c = buf[used++];
924 size += (c & 0x7f) << shift;
925 shift += 7;
927 *sizep = size;
928 return used;
931 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
933 unsigned long size, used;
934 static const char valid_loose_object_type[8] = {
935 0, /* OBJ_EXT */
936 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
937 0, /* "delta" and others are invalid in a loose object */
939 enum object_type type;
941 /* Get the data stream */
942 memset(stream, 0, sizeof(*stream));
943 stream->next_in = map;
944 stream->avail_in = mapsize;
945 stream->next_out = buffer;
946 stream->avail_out = bufsiz;
948 if (legacy_loose_object(map)) {
949 inflateInit(stream);
950 return inflate(stream, 0);
953 used = unpack_object_header_gently(map, mapsize, &type, &size);
954 if (!used || !valid_loose_object_type[type])
955 return -1;
956 map += used;
957 mapsize -= used;
959 /* Set up the stream for the rest.. */
960 stream->next_in = map;
961 stream->avail_in = mapsize;
962 inflateInit(stream);
964 /* And generate the fake traditional header */
965 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
966 typename(type), size);
967 return 0;
970 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
972 int bytes = strlen(buffer) + 1;
973 unsigned char *buf = xmalloc(1+size);
974 unsigned long n;
975 int status = Z_OK;
977 n = stream->total_out - bytes;
978 if (n > size)
979 n = size;
980 memcpy(buf, (char *) buffer + bytes, n);
981 bytes = n;
982 if (bytes < size) {
983 stream->next_out = buf + bytes;
984 stream->avail_out = size - bytes;
985 while (status == Z_OK)
986 status = inflate(stream, Z_FINISH);
988 buf[size] = 0;
989 if ((status == Z_OK || status == Z_STREAM_END) && !stream->avail_in) {
990 inflateEnd(stream);
991 return buf;
994 if (status < 0)
995 error("corrupt loose object '%s'", sha1_to_hex(sha1));
996 else if (stream->avail_in)
997 error("garbage at end of loose object '%s'",
998 sha1_to_hex(sha1));
999 free(buf);
1000 return NULL;
1004 * We used to just use "sscanf()", but that's actually way
1005 * too permissive for what we want to check. So do an anal
1006 * object header parse by hand.
1008 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1010 char type[10];
1011 int i;
1012 unsigned long size;
1015 * The type can be at most ten bytes (including the
1016 * terminating '\0' that we add), and is followed by
1017 * a space.
1019 i = 0;
1020 for (;;) {
1021 char c = *hdr++;
1022 if (c == ' ')
1023 break;
1024 type[i++] = c;
1025 if (i >= sizeof(type))
1026 return -1;
1028 type[i] = 0;
1031 * The length must follow immediately, and be in canonical
1032 * decimal format (ie "010" is not valid).
1034 size = *hdr++ - '0';
1035 if (size > 9)
1036 return -1;
1037 if (size) {
1038 for (;;) {
1039 unsigned long c = *hdr - '0';
1040 if (c > 9)
1041 break;
1042 hdr++;
1043 size = size * 10 + c;
1046 *sizep = size;
1049 * The length must be followed by a zero byte
1051 return *hdr ? -1 : type_from_string(type);
1054 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1056 int ret;
1057 z_stream stream;
1058 char hdr[8192];
1060 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1061 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1062 return NULL;
1064 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1067 static off_t get_delta_base(struct packed_git *p,
1068 struct pack_window **w_curs,
1069 off_t *curpos,
1070 enum object_type type,
1071 off_t delta_obj_offset)
1073 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1074 off_t base_offset;
1076 /* use_pack() assured us we have [base_info, base_info + 20)
1077 * as a range that we can look at without walking off the
1078 * end of the mapped window. Its actually the hash size
1079 * that is assured. An OFS_DELTA longer than the hash size
1080 * is stupid, as then a REF_DELTA would be smaller to store.
1082 if (type == OBJ_OFS_DELTA) {
1083 unsigned used = 0;
1084 unsigned char c = base_info[used++];
1085 base_offset = c & 127;
1086 while (c & 128) {
1087 base_offset += 1;
1088 if (!base_offset || base_offset & ~(~0UL >> 7))
1089 die("offset value overflow for delta base object");
1090 c = base_info[used++];
1091 base_offset = (base_offset << 7) + (c & 127);
1093 base_offset = delta_obj_offset - base_offset;
1094 if (base_offset >= delta_obj_offset)
1095 die("delta base offset out of bound");
1096 *curpos += used;
1097 } else if (type == OBJ_REF_DELTA) {
1098 /* The base entry _must_ be in the same pack */
1099 base_offset = find_pack_entry_one(base_info, p);
1100 if (!base_offset)
1101 die("failed to find delta-pack base object %s",
1102 sha1_to_hex(base_info));
1103 *curpos += 20;
1104 } else
1105 die("I am totally screwed");
1106 return base_offset;
1109 /* forward declaration for a mutually recursive function */
1110 static int packed_object_info(struct packed_git *p, off_t offset,
1111 unsigned long *sizep);
1113 static int packed_delta_info(struct packed_git *p,
1114 struct pack_window **w_curs,
1115 off_t curpos,
1116 enum object_type type,
1117 off_t obj_offset,
1118 unsigned long *sizep)
1120 off_t base_offset;
1122 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1123 type = packed_object_info(p, base_offset, NULL);
1125 /* We choose to only get the type of the base object and
1126 * ignore potentially corrupt pack file that expects the delta
1127 * based on a base with a wrong size. This saves tons of
1128 * inflate() calls.
1130 if (sizep) {
1131 const unsigned char *data;
1132 unsigned char delta_head[20], *in;
1133 z_stream stream;
1134 int st;
1136 memset(&stream, 0, sizeof(stream));
1137 stream.next_out = delta_head;
1138 stream.avail_out = sizeof(delta_head);
1140 inflateInit(&stream);
1141 do {
1142 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1143 stream.next_in = in;
1144 st = inflate(&stream, Z_FINISH);
1145 curpos += stream.next_in - in;
1146 } while ((st == Z_OK || st == Z_BUF_ERROR)
1147 && stream.total_out < sizeof(delta_head));
1148 inflateEnd(&stream);
1149 if ((st != Z_STREAM_END) &&
1150 stream.total_out != sizeof(delta_head))
1151 die("delta data unpack-initial failed");
1153 /* Examine the initial part of the delta to figure out
1154 * the result size.
1156 data = delta_head;
1158 /* ignore base size */
1159 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1161 /* Read the result size */
1162 *sizep = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1165 return type;
1168 static int unpack_object_header(struct packed_git *p,
1169 struct pack_window **w_curs,
1170 off_t *curpos,
1171 unsigned long *sizep)
1173 unsigned char *base;
1174 unsigned int left;
1175 unsigned long used;
1176 enum object_type type;
1178 /* use_pack() assures us we have [base, base + 20) available
1179 * as a range that we can look at at. (Its actually the hash
1180 * size that is assured.) With our object header encoding
1181 * the maximum deflated object size is 2^137, which is just
1182 * insane, so we know won't exceed what we have been given.
1184 base = use_pack(p, w_curs, *curpos, &left);
1185 used = unpack_object_header_gently(base, left, &type, sizep);
1186 if (!used)
1187 die("object offset outside of pack file");
1188 *curpos += used;
1190 return type;
1193 const char *packed_object_info_detail(struct packed_git *p,
1194 off_t obj_offset,
1195 unsigned long *size,
1196 unsigned long *store_size,
1197 unsigned int *delta_chain_length,
1198 unsigned char *base_sha1)
1200 struct pack_window *w_curs = NULL;
1201 off_t curpos;
1202 unsigned long dummy;
1203 unsigned char *next_sha1;
1204 enum object_type type;
1206 *delta_chain_length = 0;
1207 curpos = obj_offset;
1208 type = unpack_object_header(p, &w_curs, &curpos, size);
1210 for (;;) {
1211 switch (type) {
1212 default:
1213 die("pack %s contains unknown object type %d",
1214 p->pack_name, type);
1215 case OBJ_COMMIT:
1216 case OBJ_TREE:
1217 case OBJ_BLOB:
1218 case OBJ_TAG:
1219 *store_size = 0; /* notyet */
1220 unuse_pack(&w_curs);
1221 return typename(type);
1222 case OBJ_OFS_DELTA:
1223 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1224 if (*delta_chain_length == 0) {
1225 /* TODO: find base_sha1 as pointed by curpos */
1226 hashclr(base_sha1);
1228 break;
1229 case OBJ_REF_DELTA:
1230 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1231 if (*delta_chain_length == 0)
1232 hashcpy(base_sha1, next_sha1);
1233 obj_offset = find_pack_entry_one(next_sha1, p);
1234 break;
1236 (*delta_chain_length)++;
1237 curpos = obj_offset;
1238 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1242 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1243 unsigned long *sizep)
1245 struct pack_window *w_curs = NULL;
1246 unsigned long size;
1247 off_t curpos = obj_offset;
1248 enum object_type type;
1250 type = unpack_object_header(p, &w_curs, &curpos, &size);
1252 switch (type) {
1253 case OBJ_OFS_DELTA:
1254 case OBJ_REF_DELTA:
1255 type = packed_delta_info(p, &w_curs, curpos,
1256 type, obj_offset, sizep);
1257 break;
1258 case OBJ_COMMIT:
1259 case OBJ_TREE:
1260 case OBJ_BLOB:
1261 case OBJ_TAG:
1262 if (sizep)
1263 *sizep = size;
1264 break;
1265 default:
1266 die("pack %s contains unknown object type %d",
1267 p->pack_name, type);
1269 unuse_pack(&w_curs);
1270 return type;
1273 static void *unpack_compressed_entry(struct packed_git *p,
1274 struct pack_window **w_curs,
1275 off_t curpos,
1276 unsigned long size)
1278 int st;
1279 z_stream stream;
1280 unsigned char *buffer, *in;
1282 buffer = xmalloc(size + 1);
1283 buffer[size] = 0;
1284 memset(&stream, 0, sizeof(stream));
1285 stream.next_out = buffer;
1286 stream.avail_out = size;
1288 inflateInit(&stream);
1289 do {
1290 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1291 stream.next_in = in;
1292 st = inflate(&stream, Z_FINISH);
1293 curpos += stream.next_in - in;
1294 } while (st == Z_OK || st == Z_BUF_ERROR);
1295 inflateEnd(&stream);
1296 if ((st != Z_STREAM_END) || stream.total_out != size) {
1297 free(buffer);
1298 return NULL;
1301 return buffer;
1304 static void *unpack_delta_entry(struct packed_git *p,
1305 struct pack_window **w_curs,
1306 off_t curpos,
1307 unsigned long delta_size,
1308 off_t obj_offset,
1309 enum object_type *type,
1310 unsigned long *sizep)
1312 void *delta_data, *result, *base;
1313 unsigned long base_size;
1314 off_t base_offset;
1316 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1317 base = unpack_entry(p, base_offset, type, &base_size);
1318 if (!base)
1319 die("failed to read delta base object"
1320 " at %"PRIuMAX" from %s",
1321 (uintmax_t)base_offset, p->pack_name);
1323 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1324 result = patch_delta(base, base_size,
1325 delta_data, delta_size,
1326 sizep);
1327 if (!result)
1328 die("failed to apply delta");
1329 free(delta_data);
1330 free(base);
1331 return result;
1334 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1335 enum object_type *type, unsigned long *sizep)
1337 struct pack_window *w_curs = NULL;
1338 off_t curpos = obj_offset;
1339 void *data;
1341 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1342 switch (*type) {
1343 case OBJ_OFS_DELTA:
1344 case OBJ_REF_DELTA:
1345 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1346 obj_offset, type, sizep);
1347 break;
1348 case OBJ_COMMIT:
1349 case OBJ_TREE:
1350 case OBJ_BLOB:
1351 case OBJ_TAG:
1352 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1353 break;
1354 default:
1355 die("unknown object type %i in %s", *type, p->pack_name);
1357 unuse_pack(&w_curs);
1358 return data;
1361 uint32_t num_packed_objects(const struct packed_git *p)
1363 /* See check_packed_git_idx() */
1364 return (uint32_t)((p->index_size - 20 - 20 - 4*256) / 24);
1367 int nth_packed_object_sha1(const struct packed_git *p, uint32_t n,
1368 unsigned char* sha1)
1370 void *index = p->index_base + 256;
1371 if (num_packed_objects(p) <= n)
1372 return -1;
1373 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1374 return 0;
1377 off_t find_pack_entry_one(const unsigned char *sha1,
1378 struct packed_git *p)
1380 uint32_t *level1_ofs = p->index_base;
1381 int hi = ntohl(level1_ofs[*sha1]);
1382 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1383 void *index = p->index_base + 256;
1385 do {
1386 int mi = (lo + hi) / 2;
1387 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1388 if (!cmp)
1389 return ntohl(*((uint32_t *)((char *)index + (24 * mi))));
1390 if (cmp > 0)
1391 hi = mi;
1392 else
1393 lo = mi+1;
1394 } while (lo < hi);
1395 return 0;
1398 static int matches_pack_name(struct packed_git *p, const char *ig)
1400 const char *last_c, *c;
1402 if (!strcmp(p->pack_name, ig))
1403 return 0;
1405 for (c = p->pack_name, last_c = c; *c;)
1406 if (*c == '/')
1407 last_c = ++c;
1408 else
1409 ++c;
1410 if (!strcmp(last_c, ig))
1411 return 0;
1413 return 1;
1416 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1418 struct packed_git *p;
1419 off_t offset;
1421 prepare_packed_git();
1423 for (p = packed_git; p; p = p->next) {
1424 if (ignore_packed) {
1425 const char **ig;
1426 for (ig = ignore_packed; *ig; ig++)
1427 if (!matches_pack_name(p, *ig))
1428 break;
1429 if (*ig)
1430 continue;
1432 offset = find_pack_entry_one(sha1, p);
1433 if (offset) {
1435 * We are about to tell the caller where they can
1436 * locate the requested object. We better make
1437 * sure the packfile is still here and can be
1438 * accessed before supplying that answer, as
1439 * it may have been deleted since the index
1440 * was loaded!
1442 if (p->pack_fd == -1 && open_packed_git(p)) {
1443 error("packfile %s cannot be accessed", p->pack_name);
1444 continue;
1446 e->offset = offset;
1447 e->p = p;
1448 hashcpy(e->sha1, sha1);
1449 return 1;
1452 return 0;
1455 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1456 struct packed_git *packs)
1458 struct packed_git *p;
1460 for (p = packs; p; p = p->next) {
1461 if (find_pack_entry_one(sha1, p))
1462 return p;
1464 return NULL;
1468 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1470 int status;
1471 unsigned long mapsize, size;
1472 void *map;
1473 z_stream stream;
1474 char hdr[32];
1476 map = map_sha1_file(sha1, &mapsize);
1477 if (!map)
1478 return error("unable to find %s", sha1_to_hex(sha1));
1479 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1480 status = error("unable to unpack %s header",
1481 sha1_to_hex(sha1));
1482 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1483 status = error("unable to parse %s header", sha1_to_hex(sha1));
1484 else if (sizep)
1485 *sizep = size;
1486 inflateEnd(&stream);
1487 munmap(map, mapsize);
1488 return status;
1491 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1493 struct pack_entry e;
1495 if (!find_pack_entry(sha1, &e, NULL)) {
1496 reprepare_packed_git();
1497 if (!find_pack_entry(sha1, &e, NULL))
1498 return sha1_loose_object_info(sha1, sizep);
1500 return packed_object_info(e.p, e.offset, sizep);
1503 static void *read_packed_sha1(const unsigned char *sha1,
1504 enum object_type *type, unsigned long *size)
1506 struct pack_entry e;
1508 if (!find_pack_entry(sha1, &e, NULL))
1509 return NULL;
1510 else
1511 return unpack_entry(e.p, e.offset, type, size);
1515 * This is meant to hold a *small* number of objects that you would
1516 * want read_sha1_file() to be able to return, but yet you do not want
1517 * to write them into the object store (e.g. a browse-only
1518 * application).
1520 static struct cached_object {
1521 unsigned char sha1[20];
1522 enum object_type type;
1523 void *buf;
1524 unsigned long size;
1525 } *cached_objects;
1526 static int cached_object_nr, cached_object_alloc;
1528 static struct cached_object *find_cached_object(const unsigned char *sha1)
1530 int i;
1531 struct cached_object *co = cached_objects;
1533 for (i = 0; i < cached_object_nr; i++, co++) {
1534 if (!hashcmp(co->sha1, sha1))
1535 return co;
1537 return NULL;
1540 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1541 unsigned char *sha1)
1543 struct cached_object *co;
1545 hash_sha1_file(buf, len, typename(type), sha1);
1546 if (has_sha1_file(sha1) || find_cached_object(sha1))
1547 return 0;
1548 if (cached_object_alloc <= cached_object_nr) {
1549 cached_object_alloc = alloc_nr(cached_object_alloc);
1550 cached_objects = xrealloc(cached_objects,
1551 sizeof(*cached_objects) *
1552 cached_object_alloc);
1554 co = &cached_objects[cached_object_nr++];
1555 co->size = len;
1556 co->type = type;
1557 co->buf = xmalloc(len);
1558 memcpy(co->buf, buf, len);
1559 hashcpy(co->sha1, sha1);
1560 return 0;
1563 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1564 unsigned long *size)
1566 unsigned long mapsize;
1567 void *map, *buf;
1568 struct cached_object *co;
1570 co = find_cached_object(sha1);
1571 if (co) {
1572 buf = xmalloc(co->size + 1);
1573 memcpy(buf, co->buf, co->size);
1574 ((char*)buf)[co->size] = 0;
1575 *type = co->type;
1576 *size = co->size;
1577 return buf;
1580 buf = read_packed_sha1(sha1, type, size);
1581 if (buf)
1582 return buf;
1583 map = map_sha1_file(sha1, &mapsize);
1584 if (map) {
1585 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1586 munmap(map, mapsize);
1587 return buf;
1589 reprepare_packed_git();
1590 return read_packed_sha1(sha1, type, size);
1593 void *read_object_with_reference(const unsigned char *sha1,
1594 const char *required_type_name,
1595 unsigned long *size,
1596 unsigned char *actual_sha1_return)
1598 enum object_type type, required_type;
1599 void *buffer;
1600 unsigned long isize;
1601 unsigned char actual_sha1[20];
1603 required_type = type_from_string(required_type_name);
1604 hashcpy(actual_sha1, sha1);
1605 while (1) {
1606 int ref_length = -1;
1607 const char *ref_type = NULL;
1609 buffer = read_sha1_file(actual_sha1, &type, &isize);
1610 if (!buffer)
1611 return NULL;
1612 if (type == required_type) {
1613 *size = isize;
1614 if (actual_sha1_return)
1615 hashcpy(actual_sha1_return, actual_sha1);
1616 return buffer;
1618 /* Handle references */
1619 else if (type == OBJ_COMMIT)
1620 ref_type = "tree ";
1621 else if (type == OBJ_TAG)
1622 ref_type = "object ";
1623 else {
1624 free(buffer);
1625 return NULL;
1627 ref_length = strlen(ref_type);
1629 if (memcmp(buffer, ref_type, ref_length) ||
1630 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1631 free(buffer);
1632 return NULL;
1634 free(buffer);
1635 /* Now we have the ID of the referred-to object in
1636 * actual_sha1. Check again. */
1640 static void write_sha1_file_prepare(void *buf, unsigned long len,
1641 const char *type, unsigned char *sha1,
1642 char *hdr, int *hdrlen)
1644 SHA_CTX c;
1646 /* Generate the header */
1647 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
1649 /* Sha1.. */
1650 SHA1_Init(&c);
1651 SHA1_Update(&c, hdr, *hdrlen);
1652 SHA1_Update(&c, buf, len);
1653 SHA1_Final(sha1, &c);
1657 * Link the tempfile to the final place, possibly creating the
1658 * last directory level as you do so.
1660 * Returns the errno on failure, 0 on success.
1662 static int link_temp_to_file(const char *tmpfile, const char *filename)
1664 int ret;
1665 char *dir;
1667 if (!link(tmpfile, filename))
1668 return 0;
1671 * Try to mkdir the last path component if that failed.
1673 * Re-try the "link()" regardless of whether the mkdir
1674 * succeeds, since a race might mean that somebody
1675 * else succeeded.
1677 ret = errno;
1678 dir = strrchr(filename, '/');
1679 if (dir) {
1680 *dir = 0;
1681 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1682 *dir = '/';
1683 return -2;
1685 *dir = '/';
1686 if (!link(tmpfile, filename))
1687 return 0;
1688 ret = errno;
1690 return ret;
1694 * Move the just written object into its final resting place
1696 int move_temp_to_file(const char *tmpfile, const char *filename)
1698 int ret = link_temp_to_file(tmpfile, filename);
1701 * Coda hack - coda doesn't like cross-directory links,
1702 * so we fall back to a rename, which will mean that it
1703 * won't be able to check collisions, but that's not a
1704 * big deal.
1706 * The same holds for FAT formatted media.
1708 * When this succeeds, we just return 0. We have nothing
1709 * left to unlink.
1711 if (ret && ret != EEXIST) {
1712 if (!rename(tmpfile, filename))
1713 return 0;
1714 ret = errno;
1716 unlink(tmpfile);
1717 if (ret) {
1718 if (ret != EEXIST) {
1719 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1721 /* FIXME!!! Collision check here ? */
1724 return 0;
1727 static int write_buffer(int fd, const void *buf, size_t len)
1729 if (write_in_full(fd, buf, len) < 0)
1730 return error("file write error (%s)", strerror(errno));
1731 return 0;
1734 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1736 int hdr_len;
1737 unsigned char c;
1739 c = (type << 4) | (len & 15);
1740 len >>= 4;
1741 hdr_len = 1;
1742 while (len) {
1743 *hdr++ = c | 0x80;
1744 hdr_len++;
1745 c = (len & 0x7f);
1746 len >>= 7;
1748 *hdr = c;
1749 return hdr_len;
1752 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1754 int obj_type, hdrlen;
1756 if (use_legacy_headers) {
1757 while (deflate(stream, 0) == Z_OK)
1758 /* nothing */;
1759 return;
1761 obj_type = type_from_string(type);
1762 hdrlen = write_binary_header(stream->next_out, obj_type, len);
1763 stream->total_out = hdrlen;
1764 stream->next_out += hdrlen;
1765 stream->avail_out -= hdrlen;
1768 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1769 unsigned char *sha1)
1771 char hdr[32];
1772 int hdrlen;
1773 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1774 return 0;
1777 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1779 int size;
1780 unsigned char *compressed;
1781 z_stream stream;
1782 unsigned char sha1[20];
1783 char *filename;
1784 static char tmpfile[PATH_MAX];
1785 char hdr[32];
1786 int fd, hdrlen;
1788 /* Normally if we have it in the pack then we do not bother writing
1789 * it out into .git/objects/??/?{38} file.
1791 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1792 filename = sha1_file_name(sha1);
1793 if (returnsha1)
1794 hashcpy(returnsha1, sha1);
1795 if (has_sha1_file(sha1))
1796 return 0;
1797 fd = open(filename, O_RDONLY);
1798 if (fd >= 0) {
1800 * FIXME!!! We might do collision checking here, but we'd
1801 * need to uncompress the old file and check it. Later.
1803 close(fd);
1804 return 0;
1807 if (errno != ENOENT) {
1808 return error("sha1 file %s: %s\n", filename, strerror(errno));
1811 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1813 fd = mkstemp(tmpfile);
1814 if (fd < 0) {
1815 if (errno == EPERM)
1816 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1817 else
1818 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1821 /* Set it up */
1822 memset(&stream, 0, sizeof(stream));
1823 deflateInit(&stream, zlib_compression_level);
1824 size = 8 + deflateBound(&stream, len+hdrlen);
1825 compressed = xmalloc(size);
1827 /* Compress it */
1828 stream.next_out = compressed;
1829 stream.avail_out = size;
1831 /* First header.. */
1832 stream.next_in = (unsigned char *)hdr;
1833 stream.avail_in = hdrlen;
1834 setup_object_header(&stream, type, len);
1836 /* Then the data itself.. */
1837 stream.next_in = buf;
1838 stream.avail_in = len;
1839 while (deflate(&stream, Z_FINISH) == Z_OK)
1840 /* nothing */;
1841 deflateEnd(&stream);
1842 size = stream.total_out;
1844 if (write_buffer(fd, compressed, size) < 0)
1845 die("unable to write sha1 file");
1846 fchmod(fd, 0444);
1847 close(fd);
1848 free(compressed);
1850 return move_temp_to_file(tmpfile, filename);
1854 * We need to unpack and recompress the object for writing
1855 * it out to a different file.
1857 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1859 size_t size;
1860 z_stream stream;
1861 unsigned char *unpacked;
1862 unsigned long len;
1863 enum object_type type;
1864 char hdr[32];
1865 int hdrlen;
1866 void *buf;
1868 /* need to unpack and recompress it by itself */
1869 unpacked = read_packed_sha1(sha1, &type, &len);
1870 if (!unpacked)
1871 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1873 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1875 /* Set it up */
1876 memset(&stream, 0, sizeof(stream));
1877 deflateInit(&stream, zlib_compression_level);
1878 size = deflateBound(&stream, len + hdrlen);
1879 buf = xmalloc(size);
1881 /* Compress it */
1882 stream.next_out = buf;
1883 stream.avail_out = size;
1885 /* First header.. */
1886 stream.next_in = (void *)hdr;
1887 stream.avail_in = hdrlen;
1888 while (deflate(&stream, 0) == Z_OK)
1889 /* nothing */;
1891 /* Then the data itself.. */
1892 stream.next_in = unpacked;
1893 stream.avail_in = len;
1894 while (deflate(&stream, Z_FINISH) == Z_OK)
1895 /* nothing */;
1896 deflateEnd(&stream);
1897 free(unpacked);
1899 *objsize = stream.total_out;
1900 return buf;
1903 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1905 int retval;
1906 unsigned long objsize;
1907 void *buf = map_sha1_file(sha1, &objsize);
1909 if (buf) {
1910 retval = write_buffer(fd, buf, objsize);
1911 munmap(buf, objsize);
1912 return retval;
1915 buf = repack_object(sha1, &objsize);
1916 retval = write_buffer(fd, buf, objsize);
1917 free(buf);
1918 return retval;
1921 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1922 size_t bufsize, size_t *bufposn)
1924 char tmpfile[PATH_MAX];
1925 int local;
1926 z_stream stream;
1927 unsigned char real_sha1[20];
1928 unsigned char discard[4096];
1929 int ret;
1930 SHA_CTX c;
1932 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1934 local = mkstemp(tmpfile);
1935 if (local < 0) {
1936 if (errno == EPERM)
1937 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1938 else
1939 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1942 memset(&stream, 0, sizeof(stream));
1944 inflateInit(&stream);
1946 SHA1_Init(&c);
1948 do {
1949 ssize_t size;
1950 if (*bufposn) {
1951 stream.avail_in = *bufposn;
1952 stream.next_in = (unsigned char *) buffer;
1953 do {
1954 stream.next_out = discard;
1955 stream.avail_out = sizeof(discard);
1956 ret = inflate(&stream, Z_SYNC_FLUSH);
1957 SHA1_Update(&c, discard, sizeof(discard) -
1958 stream.avail_out);
1959 } while (stream.avail_in && ret == Z_OK);
1960 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1961 die("unable to write sha1 file");
1962 memmove(buffer, buffer + *bufposn - stream.avail_in,
1963 stream.avail_in);
1964 *bufposn = stream.avail_in;
1965 if (ret != Z_OK)
1966 break;
1968 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
1969 if (size <= 0) {
1970 close(local);
1971 unlink(tmpfile);
1972 if (!size)
1973 return error("Connection closed?");
1974 perror("Reading from connection");
1975 return -1;
1977 *bufposn += size;
1978 } while (1);
1979 inflateEnd(&stream);
1981 close(local);
1982 SHA1_Final(real_sha1, &c);
1983 if (ret != Z_STREAM_END) {
1984 unlink(tmpfile);
1985 return error("File %s corrupted", sha1_to_hex(sha1));
1987 if (hashcmp(sha1, real_sha1)) {
1988 unlink(tmpfile);
1989 return error("File %s has bad hash", sha1_to_hex(sha1));
1992 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1995 int has_pack_index(const unsigned char *sha1)
1997 struct stat st;
1998 if (stat(sha1_pack_index_name(sha1), &st))
1999 return 0;
2000 return 1;
2003 int has_pack_file(const unsigned char *sha1)
2005 struct stat st;
2006 if (stat(sha1_pack_name(sha1), &st))
2007 return 0;
2008 return 1;
2011 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2013 struct pack_entry e;
2014 return find_pack_entry(sha1, &e, ignore_packed);
2017 int has_sha1_file(const unsigned char *sha1)
2019 struct stat st;
2020 struct pack_entry e;
2022 if (find_pack_entry(sha1, &e, NULL))
2023 return 1;
2024 return find_sha1_file(sha1, &st) ? 1 : 0;
2028 * reads from fd as long as possible into a supplied buffer of size bytes.
2029 * If necessary the buffer's size is increased using realloc()
2031 * returns 0 if anything went fine and -1 otherwise
2033 * NOTE: both buf and size may change, but even when -1 is returned
2034 * you still have to free() it yourself.
2036 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
2038 char* buf = *return_buf;
2039 unsigned long size = *return_size;
2040 int iret;
2041 unsigned long off = 0;
2043 do {
2044 iret = xread(fd, buf + off, size - off);
2045 if (iret > 0) {
2046 off += iret;
2047 if (off == size) {
2048 size *= 2;
2049 buf = xrealloc(buf, size);
2052 } while (iret > 0);
2054 *return_buf = buf;
2055 *return_size = off;
2057 if (iret < 0)
2058 return -1;
2059 return 0;
2062 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2064 unsigned long size = 4096;
2065 char *buf = xmalloc(size);
2066 int ret;
2068 if (read_pipe(fd, &buf, &size)) {
2069 free(buf);
2070 return -1;
2073 if (!type)
2074 type = blob_type;
2075 if (write_object)
2076 ret = write_sha1_file(buf, size, type, sha1);
2077 else
2078 ret = hash_sha1_file(buf, size, type, sha1);
2079 free(buf);
2080 return ret;
2083 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2084 enum object_type type, const char *path)
2086 size_t size = xsize_t(st->st_size);
2087 void *buf = NULL;
2088 int ret, re_allocated = 0;
2090 if (size)
2091 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2092 close(fd);
2094 if (!type)
2095 type = OBJ_BLOB;
2098 * Convert blobs to git internal format
2100 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2101 unsigned long nsize = size;
2102 char *nbuf = buf;
2103 if (convert_to_git(path, &nbuf, &nsize)) {
2104 if (size)
2105 munmap(buf, size);
2106 size = nsize;
2107 buf = nbuf;
2108 re_allocated = 1;
2112 if (write_object)
2113 ret = write_sha1_file(buf, size, typename(type), sha1);
2114 else
2115 ret = hash_sha1_file(buf, size, typename(type), sha1);
2116 if (re_allocated) {
2117 free(buf);
2118 return ret;
2120 if (size)
2121 munmap(buf, size);
2122 return ret;
2125 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2127 int fd;
2128 char *target;
2129 size_t len;
2131 switch (st->st_mode & S_IFMT) {
2132 case S_IFREG:
2133 fd = open(path, O_RDONLY);
2134 if (fd < 0)
2135 return error("open(\"%s\"): %s", path,
2136 strerror(errno));
2137 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2138 return error("%s: failed to insert into database",
2139 path);
2140 break;
2141 case S_IFLNK:
2142 len = xsize_t(st->st_size);
2143 target = xmalloc(len + 1);
2144 if (readlink(path, target, len + 1) != st->st_size) {
2145 char *errstr = strerror(errno);
2146 free(target);
2147 return error("readlink(\"%s\"): %s", path,
2148 errstr);
2150 if (!write_object)
2151 hash_sha1_file(target, len, blob_type, sha1);
2152 else if (write_sha1_file(target, len, blob_type, sha1))
2153 return error("%s: failed to insert into database",
2154 path);
2155 free(target);
2156 break;
2157 default:
2158 return error("%s: unsupported file type", path);
2160 return 0;
2163 int read_pack_header(int fd, struct pack_header *header)
2165 char *c = (char*)header;
2166 ssize_t remaining = sizeof(struct pack_header);
2167 do {
2168 ssize_t r = xread(fd, c, remaining);
2169 if (r <= 0)
2170 /* "eof before pack header was fully read" */
2171 return PH_ERROR_EOF;
2172 remaining -= r;
2173 c += r;
2174 } while (remaining > 0);
2175 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2176 /* "protocol error (pack signature mismatch detected)" */
2177 return PH_ERROR_PACK_SIGNATURE;
2178 if (!pack_version_ok(header->hdr_version))
2179 /* "protocol error (pack version unsupported)" */
2180 return PH_ERROR_PROTOCOL;
2181 return 0;