Cast 64 bit off_t to 32 bit size_t
[git/jnareb-git/bp-gitweb.git] / sha1_file.c
blob219a10f403e0c3aa2f8d60ecde5807cf51d1a659
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)
972 int bytes = strlen(buffer) + 1;
973 unsigned char *buf = xmalloc(1+size);
974 unsigned long n;
976 n = stream->total_out - bytes;
977 if (n > size)
978 n = size;
979 memcpy(buf, (char *) buffer + bytes, n);
980 bytes = n;
981 if (bytes < size) {
982 stream->next_out = buf + bytes;
983 stream->avail_out = size - bytes;
984 while (inflate(stream, Z_FINISH) == Z_OK)
985 /* nothing */;
987 buf[size] = 0;
988 inflateEnd(stream);
989 return buf;
993 * We used to just use "sscanf()", but that's actually way
994 * too permissive for what we want to check. So do an anal
995 * object header parse by hand.
997 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
999 char type[10];
1000 int i;
1001 unsigned long size;
1004 * The type can be at most ten bytes (including the
1005 * terminating '\0' that we add), and is followed by
1006 * a space.
1008 i = 0;
1009 for (;;) {
1010 char c = *hdr++;
1011 if (c == ' ')
1012 break;
1013 type[i++] = c;
1014 if (i >= sizeof(type))
1015 return -1;
1017 type[i] = 0;
1020 * The length must follow immediately, and be in canonical
1021 * decimal format (ie "010" is not valid).
1023 size = *hdr++ - '0';
1024 if (size > 9)
1025 return -1;
1026 if (size) {
1027 for (;;) {
1028 unsigned long c = *hdr - '0';
1029 if (c > 9)
1030 break;
1031 hdr++;
1032 size = size * 10 + c;
1035 *sizep = size;
1038 * The length must be followed by a zero byte
1040 return *hdr ? -1 : type_from_string(type);
1043 void * unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size)
1045 int ret;
1046 z_stream stream;
1047 char hdr[8192];
1049 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1050 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1051 return NULL;
1053 return unpack_sha1_rest(&stream, hdr, *size);
1056 static off_t get_delta_base(struct packed_git *p,
1057 struct pack_window **w_curs,
1058 off_t *curpos,
1059 enum object_type type,
1060 off_t delta_obj_offset)
1062 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1063 off_t base_offset;
1065 /* use_pack() assured us we have [base_info, base_info + 20)
1066 * as a range that we can look at without walking off the
1067 * end of the mapped window. Its actually the hash size
1068 * that is assured. An OFS_DELTA longer than the hash size
1069 * is stupid, as then a REF_DELTA would be smaller to store.
1071 if (type == OBJ_OFS_DELTA) {
1072 unsigned used = 0;
1073 unsigned char c = base_info[used++];
1074 base_offset = c & 127;
1075 while (c & 128) {
1076 base_offset += 1;
1077 if (!base_offset || base_offset & ~(~0UL >> 7))
1078 die("offset value overflow for delta base object");
1079 c = base_info[used++];
1080 base_offset = (base_offset << 7) + (c & 127);
1082 base_offset = delta_obj_offset - base_offset;
1083 if (base_offset >= delta_obj_offset)
1084 die("delta base offset out of bound");
1085 *curpos += used;
1086 } else if (type == OBJ_REF_DELTA) {
1087 /* The base entry _must_ be in the same pack */
1088 base_offset = find_pack_entry_one(base_info, p);
1089 if (!base_offset)
1090 die("failed to find delta-pack base object %s",
1091 sha1_to_hex(base_info));
1092 *curpos += 20;
1093 } else
1094 die("I am totally screwed");
1095 return base_offset;
1098 /* forward declaration for a mutually recursive function */
1099 static int packed_object_info(struct packed_git *p, off_t offset,
1100 unsigned long *sizep);
1102 static int packed_delta_info(struct packed_git *p,
1103 struct pack_window **w_curs,
1104 off_t curpos,
1105 enum object_type type,
1106 off_t obj_offset,
1107 unsigned long *sizep)
1109 off_t base_offset;
1111 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1112 type = packed_object_info(p, base_offset, NULL);
1114 /* We choose to only get the type of the base object and
1115 * ignore potentially corrupt pack file that expects the delta
1116 * based on a base with a wrong size. This saves tons of
1117 * inflate() calls.
1119 if (sizep) {
1120 const unsigned char *data;
1121 unsigned char delta_head[20], *in;
1122 z_stream stream;
1123 int st;
1125 memset(&stream, 0, sizeof(stream));
1126 stream.next_out = delta_head;
1127 stream.avail_out = sizeof(delta_head);
1129 inflateInit(&stream);
1130 do {
1131 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1132 stream.next_in = in;
1133 st = inflate(&stream, Z_FINISH);
1134 curpos += stream.next_in - in;
1135 } while ((st == Z_OK || st == Z_BUF_ERROR)
1136 && stream.total_out < sizeof(delta_head));
1137 inflateEnd(&stream);
1138 if ((st != Z_STREAM_END) &&
1139 stream.total_out != sizeof(delta_head))
1140 die("delta data unpack-initial failed");
1142 /* Examine the initial part of the delta to figure out
1143 * the result size.
1145 data = delta_head;
1147 /* ignore base size */
1148 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1150 /* Read the result size */
1151 *sizep = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1154 return type;
1157 static int unpack_object_header(struct packed_git *p,
1158 struct pack_window **w_curs,
1159 off_t *curpos,
1160 unsigned long *sizep)
1162 unsigned char *base;
1163 unsigned int left;
1164 unsigned long used;
1165 enum object_type type;
1167 /* use_pack() assures us we have [base, base + 20) available
1168 * as a range that we can look at at. (Its actually the hash
1169 * size that is assured.) With our object header encoding
1170 * the maximum deflated object size is 2^137, which is just
1171 * insane, so we know won't exceed what we have been given.
1173 base = use_pack(p, w_curs, *curpos, &left);
1174 used = unpack_object_header_gently(base, left, &type, sizep);
1175 if (!used)
1176 die("object offset outside of pack file");
1177 *curpos += used;
1179 return type;
1182 const char *packed_object_info_detail(struct packed_git *p,
1183 off_t obj_offset,
1184 unsigned long *size,
1185 unsigned long *store_size,
1186 unsigned int *delta_chain_length,
1187 unsigned char *base_sha1)
1189 struct pack_window *w_curs = NULL;
1190 off_t curpos;
1191 unsigned long dummy;
1192 unsigned char *next_sha1;
1193 enum object_type type;
1195 *delta_chain_length = 0;
1196 curpos = obj_offset;
1197 type = unpack_object_header(p, &w_curs, &curpos, size);
1199 for (;;) {
1200 switch (type) {
1201 default:
1202 die("pack %s contains unknown object type %d",
1203 p->pack_name, type);
1204 case OBJ_COMMIT:
1205 case OBJ_TREE:
1206 case OBJ_BLOB:
1207 case OBJ_TAG:
1208 *store_size = 0; /* notyet */
1209 unuse_pack(&w_curs);
1210 return typename(type);
1211 case OBJ_OFS_DELTA:
1212 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1213 if (*delta_chain_length == 0) {
1214 /* TODO: find base_sha1 as pointed by curpos */
1215 hashclr(base_sha1);
1217 break;
1218 case OBJ_REF_DELTA:
1219 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1220 if (*delta_chain_length == 0)
1221 hashcpy(base_sha1, next_sha1);
1222 obj_offset = find_pack_entry_one(next_sha1, p);
1223 break;
1225 (*delta_chain_length)++;
1226 curpos = obj_offset;
1227 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1231 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1232 unsigned long *sizep)
1234 struct pack_window *w_curs = NULL;
1235 unsigned long size;
1236 off_t curpos = obj_offset;
1237 enum object_type type;
1239 type = unpack_object_header(p, &w_curs, &curpos, &size);
1241 switch (type) {
1242 case OBJ_OFS_DELTA:
1243 case OBJ_REF_DELTA:
1244 type = packed_delta_info(p, &w_curs, curpos,
1245 type, obj_offset, sizep);
1246 break;
1247 case OBJ_COMMIT:
1248 case OBJ_TREE:
1249 case OBJ_BLOB:
1250 case OBJ_TAG:
1251 if (sizep)
1252 *sizep = size;
1253 break;
1254 default:
1255 die("pack %s contains unknown object type %d",
1256 p->pack_name, type);
1258 unuse_pack(&w_curs);
1259 return type;
1262 static void *unpack_compressed_entry(struct packed_git *p,
1263 struct pack_window **w_curs,
1264 off_t curpos,
1265 unsigned long size)
1267 int st;
1268 z_stream stream;
1269 unsigned char *buffer, *in;
1271 buffer = xmalloc(size + 1);
1272 buffer[size] = 0;
1273 memset(&stream, 0, sizeof(stream));
1274 stream.next_out = buffer;
1275 stream.avail_out = size;
1277 inflateInit(&stream);
1278 do {
1279 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1280 stream.next_in = in;
1281 st = inflate(&stream, Z_FINISH);
1282 curpos += stream.next_in - in;
1283 } while (st == Z_OK || st == Z_BUF_ERROR);
1284 inflateEnd(&stream);
1285 if ((st != Z_STREAM_END) || stream.total_out != size) {
1286 free(buffer);
1287 return NULL;
1290 return buffer;
1293 static void *unpack_delta_entry(struct packed_git *p,
1294 struct pack_window **w_curs,
1295 off_t curpos,
1296 unsigned long delta_size,
1297 off_t obj_offset,
1298 enum object_type *type,
1299 unsigned long *sizep)
1301 void *delta_data, *result, *base;
1302 unsigned long base_size;
1303 off_t base_offset;
1305 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1306 base = unpack_entry(p, base_offset, type, &base_size);
1307 if (!base)
1308 die("failed to read delta base object"
1309 " at %"PRIuMAX" from %s",
1310 (uintmax_t)base_offset, p->pack_name);
1312 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1313 result = patch_delta(base, base_size,
1314 delta_data, delta_size,
1315 sizep);
1316 if (!result)
1317 die("failed to apply delta");
1318 free(delta_data);
1319 free(base);
1320 return result;
1323 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1324 enum object_type *type, unsigned long *sizep)
1326 struct pack_window *w_curs = NULL;
1327 off_t curpos = obj_offset;
1328 void *data;
1330 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1331 switch (*type) {
1332 case OBJ_OFS_DELTA:
1333 case OBJ_REF_DELTA:
1334 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1335 obj_offset, type, sizep);
1336 break;
1337 case OBJ_COMMIT:
1338 case OBJ_TREE:
1339 case OBJ_BLOB:
1340 case OBJ_TAG:
1341 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1342 break;
1343 default:
1344 die("unknown object type %i in %s", *type, p->pack_name);
1346 unuse_pack(&w_curs);
1347 return data;
1350 uint32_t num_packed_objects(const struct packed_git *p)
1352 /* See check_packed_git_idx() */
1353 return (uint32_t)((p->index_size - 20 - 20 - 4*256) / 24);
1356 int nth_packed_object_sha1(const struct packed_git *p, uint32_t n,
1357 unsigned char* sha1)
1359 void *index = p->index_base + 256;
1360 if (num_packed_objects(p) <= n)
1361 return -1;
1362 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1363 return 0;
1366 off_t find_pack_entry_one(const unsigned char *sha1,
1367 struct packed_git *p)
1369 uint32_t *level1_ofs = p->index_base;
1370 int hi = ntohl(level1_ofs[*sha1]);
1371 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1372 void *index = p->index_base + 256;
1374 do {
1375 int mi = (lo + hi) / 2;
1376 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1377 if (!cmp)
1378 return ntohl(*((uint32_t *)((char *)index + (24 * mi))));
1379 if (cmp > 0)
1380 hi = mi;
1381 else
1382 lo = mi+1;
1383 } while (lo < hi);
1384 return 0;
1387 static int matches_pack_name(struct packed_git *p, const char *ig)
1389 const char *last_c, *c;
1391 if (!strcmp(p->pack_name, ig))
1392 return 0;
1394 for (c = p->pack_name, last_c = c; *c;)
1395 if (*c == '/')
1396 last_c = ++c;
1397 else
1398 ++c;
1399 if (!strcmp(last_c, ig))
1400 return 0;
1402 return 1;
1405 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1407 struct packed_git *p;
1408 off_t offset;
1410 prepare_packed_git();
1412 for (p = packed_git; p; p = p->next) {
1413 if (ignore_packed) {
1414 const char **ig;
1415 for (ig = ignore_packed; *ig; ig++)
1416 if (!matches_pack_name(p, *ig))
1417 break;
1418 if (*ig)
1419 continue;
1421 offset = find_pack_entry_one(sha1, p);
1422 if (offset) {
1424 * We are about to tell the caller where they can
1425 * locate the requested object. We better make
1426 * sure the packfile is still here and can be
1427 * accessed before supplying that answer, as
1428 * it may have been deleted since the index
1429 * was loaded!
1431 if (p->pack_fd == -1 && open_packed_git(p)) {
1432 error("packfile %s cannot be accessed", p->pack_name);
1433 continue;
1435 e->offset = offset;
1436 e->p = p;
1437 hashcpy(e->sha1, sha1);
1438 return 1;
1441 return 0;
1444 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1445 struct packed_git *packs)
1447 struct packed_git *p;
1449 for (p = packs; p; p = p->next) {
1450 if (find_pack_entry_one(sha1, p))
1451 return p;
1453 return NULL;
1457 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1459 int status;
1460 unsigned long mapsize, size;
1461 void *map;
1462 z_stream stream;
1463 char hdr[32];
1465 map = map_sha1_file(sha1, &mapsize);
1466 if (!map)
1467 return error("unable to find %s", sha1_to_hex(sha1));
1468 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1469 status = error("unable to unpack %s header",
1470 sha1_to_hex(sha1));
1471 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1472 status = error("unable to parse %s header", sha1_to_hex(sha1));
1473 else if (sizep)
1474 *sizep = size;
1475 inflateEnd(&stream);
1476 munmap(map, mapsize);
1477 return status;
1480 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1482 struct pack_entry e;
1484 if (!find_pack_entry(sha1, &e, NULL)) {
1485 reprepare_packed_git();
1486 if (!find_pack_entry(sha1, &e, NULL))
1487 return sha1_loose_object_info(sha1, sizep);
1489 return packed_object_info(e.p, e.offset, sizep);
1492 static void *read_packed_sha1(const unsigned char *sha1,
1493 enum object_type *type, unsigned long *size)
1495 struct pack_entry e;
1497 if (!find_pack_entry(sha1, &e, NULL))
1498 return NULL;
1499 else
1500 return unpack_entry(e.p, e.offset, type, size);
1504 * This is meant to hold a *small* number of objects that you would
1505 * want read_sha1_file() to be able to return, but yet you do not want
1506 * to write them into the object store (e.g. a browse-only
1507 * application).
1509 static struct cached_object {
1510 unsigned char sha1[20];
1511 enum object_type type;
1512 void *buf;
1513 unsigned long size;
1514 } *cached_objects;
1515 static int cached_object_nr, cached_object_alloc;
1517 static struct cached_object *find_cached_object(const unsigned char *sha1)
1519 int i;
1520 struct cached_object *co = cached_objects;
1522 for (i = 0; i < cached_object_nr; i++, co++) {
1523 if (!hashcmp(co->sha1, sha1))
1524 return co;
1526 return NULL;
1529 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1530 unsigned char *sha1)
1532 struct cached_object *co;
1534 hash_sha1_file(buf, len, typename(type), sha1);
1535 if (has_sha1_file(sha1) || find_cached_object(sha1))
1536 return 0;
1537 if (cached_object_alloc <= cached_object_nr) {
1538 cached_object_alloc = alloc_nr(cached_object_alloc);
1539 cached_objects = xrealloc(cached_objects,
1540 sizeof(*cached_objects) *
1541 cached_object_alloc);
1543 co = &cached_objects[cached_object_nr++];
1544 co->size = len;
1545 co->type = type;
1546 co->buf = xmalloc(len);
1547 memcpy(co->buf, buf, len);
1548 hashcpy(co->sha1, sha1);
1549 return 0;
1552 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1553 unsigned long *size)
1555 unsigned long mapsize;
1556 void *map, *buf;
1557 struct cached_object *co;
1559 co = find_cached_object(sha1);
1560 if (co) {
1561 buf = xmalloc(co->size + 1);
1562 memcpy(buf, co->buf, co->size);
1563 ((char*)buf)[co->size] = 0;
1564 *type = co->type;
1565 *size = co->size;
1566 return buf;
1569 buf = read_packed_sha1(sha1, type, size);
1570 if (buf)
1571 return buf;
1572 map = map_sha1_file(sha1, &mapsize);
1573 if (map) {
1574 buf = unpack_sha1_file(map, mapsize, type, size);
1575 munmap(map, mapsize);
1576 return buf;
1578 reprepare_packed_git();
1579 return read_packed_sha1(sha1, type, size);
1582 void *read_object_with_reference(const unsigned char *sha1,
1583 const char *required_type_name,
1584 unsigned long *size,
1585 unsigned char *actual_sha1_return)
1587 enum object_type type, required_type;
1588 void *buffer;
1589 unsigned long isize;
1590 unsigned char actual_sha1[20];
1592 required_type = type_from_string(required_type_name);
1593 hashcpy(actual_sha1, sha1);
1594 while (1) {
1595 int ref_length = -1;
1596 const char *ref_type = NULL;
1598 buffer = read_sha1_file(actual_sha1, &type, &isize);
1599 if (!buffer)
1600 return NULL;
1601 if (type == required_type) {
1602 *size = isize;
1603 if (actual_sha1_return)
1604 hashcpy(actual_sha1_return, actual_sha1);
1605 return buffer;
1607 /* Handle references */
1608 else if (type == OBJ_COMMIT)
1609 ref_type = "tree ";
1610 else if (type == OBJ_TAG)
1611 ref_type = "object ";
1612 else {
1613 free(buffer);
1614 return NULL;
1616 ref_length = strlen(ref_type);
1618 if (memcmp(buffer, ref_type, ref_length) ||
1619 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1620 free(buffer);
1621 return NULL;
1623 free(buffer);
1624 /* Now we have the ID of the referred-to object in
1625 * actual_sha1. Check again. */
1629 static void write_sha1_file_prepare(void *buf, unsigned long len,
1630 const char *type, unsigned char *sha1,
1631 char *hdr, int *hdrlen)
1633 SHA_CTX c;
1635 /* Generate the header */
1636 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
1638 /* Sha1.. */
1639 SHA1_Init(&c);
1640 SHA1_Update(&c, hdr, *hdrlen);
1641 SHA1_Update(&c, buf, len);
1642 SHA1_Final(sha1, &c);
1646 * Link the tempfile to the final place, possibly creating the
1647 * last directory level as you do so.
1649 * Returns the errno on failure, 0 on success.
1651 static int link_temp_to_file(const char *tmpfile, const char *filename)
1653 int ret;
1654 char *dir;
1656 if (!link(tmpfile, filename))
1657 return 0;
1660 * Try to mkdir the last path component if that failed.
1662 * Re-try the "link()" regardless of whether the mkdir
1663 * succeeds, since a race might mean that somebody
1664 * else succeeded.
1666 ret = errno;
1667 dir = strrchr(filename, '/');
1668 if (dir) {
1669 *dir = 0;
1670 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1671 *dir = '/';
1672 return -2;
1674 *dir = '/';
1675 if (!link(tmpfile, filename))
1676 return 0;
1677 ret = errno;
1679 return ret;
1683 * Move the just written object into its final resting place
1685 int move_temp_to_file(const char *tmpfile, const char *filename)
1687 int ret = link_temp_to_file(tmpfile, filename);
1690 * Coda hack - coda doesn't like cross-directory links,
1691 * so we fall back to a rename, which will mean that it
1692 * won't be able to check collisions, but that's not a
1693 * big deal.
1695 * The same holds for FAT formatted media.
1697 * When this succeeds, we just return 0. We have nothing
1698 * left to unlink.
1700 if (ret && ret != EEXIST) {
1701 if (!rename(tmpfile, filename))
1702 return 0;
1703 ret = errno;
1705 unlink(tmpfile);
1706 if (ret) {
1707 if (ret != EEXIST) {
1708 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1710 /* FIXME!!! Collision check here ? */
1713 return 0;
1716 static int write_buffer(int fd, const void *buf, size_t len)
1718 if (write_in_full(fd, buf, len) < 0)
1719 return error("file write error (%s)", strerror(errno));
1720 return 0;
1723 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1725 int hdr_len;
1726 unsigned char c;
1728 c = (type << 4) | (len & 15);
1729 len >>= 4;
1730 hdr_len = 1;
1731 while (len) {
1732 *hdr++ = c | 0x80;
1733 hdr_len++;
1734 c = (len & 0x7f);
1735 len >>= 7;
1737 *hdr = c;
1738 return hdr_len;
1741 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1743 int obj_type, hdrlen;
1745 if (use_legacy_headers) {
1746 while (deflate(stream, 0) == Z_OK)
1747 /* nothing */;
1748 return;
1750 obj_type = type_from_string(type);
1751 hdrlen = write_binary_header(stream->next_out, obj_type, len);
1752 stream->total_out = hdrlen;
1753 stream->next_out += hdrlen;
1754 stream->avail_out -= hdrlen;
1757 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1758 unsigned char *sha1)
1760 char hdr[32];
1761 int hdrlen;
1762 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1763 return 0;
1766 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1768 int size;
1769 unsigned char *compressed;
1770 z_stream stream;
1771 unsigned char sha1[20];
1772 char *filename;
1773 static char tmpfile[PATH_MAX];
1774 char hdr[32];
1775 int fd, hdrlen;
1777 /* Normally if we have it in the pack then we do not bother writing
1778 * it out into .git/objects/??/?{38} file.
1780 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1781 filename = sha1_file_name(sha1);
1782 if (returnsha1)
1783 hashcpy(returnsha1, sha1);
1784 if (has_sha1_file(sha1))
1785 return 0;
1786 fd = open(filename, O_RDONLY);
1787 if (fd >= 0) {
1789 * FIXME!!! We might do collision checking here, but we'd
1790 * need to uncompress the old file and check it. Later.
1792 close(fd);
1793 return 0;
1796 if (errno != ENOENT) {
1797 return error("sha1 file %s: %s\n", filename, strerror(errno));
1800 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1802 fd = mkstemp(tmpfile);
1803 if (fd < 0) {
1804 if (errno == EPERM)
1805 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1806 else
1807 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1810 /* Set it up */
1811 memset(&stream, 0, sizeof(stream));
1812 deflateInit(&stream, zlib_compression_level);
1813 size = 8 + deflateBound(&stream, len+hdrlen);
1814 compressed = xmalloc(size);
1816 /* Compress it */
1817 stream.next_out = compressed;
1818 stream.avail_out = size;
1820 /* First header.. */
1821 stream.next_in = (unsigned char *)hdr;
1822 stream.avail_in = hdrlen;
1823 setup_object_header(&stream, type, len);
1825 /* Then the data itself.. */
1826 stream.next_in = buf;
1827 stream.avail_in = len;
1828 while (deflate(&stream, Z_FINISH) == Z_OK)
1829 /* nothing */;
1830 deflateEnd(&stream);
1831 size = stream.total_out;
1833 if (write_buffer(fd, compressed, size) < 0)
1834 die("unable to write sha1 file");
1835 fchmod(fd, 0444);
1836 close(fd);
1837 free(compressed);
1839 return move_temp_to_file(tmpfile, filename);
1843 * We need to unpack and recompress the object for writing
1844 * it out to a different file.
1846 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1848 size_t size;
1849 z_stream stream;
1850 unsigned char *unpacked;
1851 unsigned long len;
1852 enum object_type type;
1853 char hdr[32];
1854 int hdrlen;
1855 void *buf;
1857 /* need to unpack and recompress it by itself */
1858 unpacked = read_packed_sha1(sha1, &type, &len);
1859 if (!unpacked)
1860 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1862 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1864 /* Set it up */
1865 memset(&stream, 0, sizeof(stream));
1866 deflateInit(&stream, zlib_compression_level);
1867 size = deflateBound(&stream, len + hdrlen);
1868 buf = xmalloc(size);
1870 /* Compress it */
1871 stream.next_out = buf;
1872 stream.avail_out = size;
1874 /* First header.. */
1875 stream.next_in = (void *)hdr;
1876 stream.avail_in = hdrlen;
1877 while (deflate(&stream, 0) == Z_OK)
1878 /* nothing */;
1880 /* Then the data itself.. */
1881 stream.next_in = unpacked;
1882 stream.avail_in = len;
1883 while (deflate(&stream, Z_FINISH) == Z_OK)
1884 /* nothing */;
1885 deflateEnd(&stream);
1886 free(unpacked);
1888 *objsize = stream.total_out;
1889 return buf;
1892 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1894 int retval;
1895 unsigned long objsize;
1896 void *buf = map_sha1_file(sha1, &objsize);
1898 if (buf) {
1899 retval = write_buffer(fd, buf, objsize);
1900 munmap(buf, objsize);
1901 return retval;
1904 buf = repack_object(sha1, &objsize);
1905 retval = write_buffer(fd, buf, objsize);
1906 free(buf);
1907 return retval;
1910 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1911 size_t bufsize, size_t *bufposn)
1913 char tmpfile[PATH_MAX];
1914 int local;
1915 z_stream stream;
1916 unsigned char real_sha1[20];
1917 unsigned char discard[4096];
1918 int ret;
1919 SHA_CTX c;
1921 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1923 local = mkstemp(tmpfile);
1924 if (local < 0) {
1925 if (errno == EPERM)
1926 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1927 else
1928 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1931 memset(&stream, 0, sizeof(stream));
1933 inflateInit(&stream);
1935 SHA1_Init(&c);
1937 do {
1938 ssize_t size;
1939 if (*bufposn) {
1940 stream.avail_in = *bufposn;
1941 stream.next_in = (unsigned char *) buffer;
1942 do {
1943 stream.next_out = discard;
1944 stream.avail_out = sizeof(discard);
1945 ret = inflate(&stream, Z_SYNC_FLUSH);
1946 SHA1_Update(&c, discard, sizeof(discard) -
1947 stream.avail_out);
1948 } while (stream.avail_in && ret == Z_OK);
1949 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1950 die("unable to write sha1 file");
1951 memmove(buffer, buffer + *bufposn - stream.avail_in,
1952 stream.avail_in);
1953 *bufposn = stream.avail_in;
1954 if (ret != Z_OK)
1955 break;
1957 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
1958 if (size <= 0) {
1959 close(local);
1960 unlink(tmpfile);
1961 if (!size)
1962 return error("Connection closed?");
1963 perror("Reading from connection");
1964 return -1;
1966 *bufposn += size;
1967 } while (1);
1968 inflateEnd(&stream);
1970 close(local);
1971 SHA1_Final(real_sha1, &c);
1972 if (ret != Z_STREAM_END) {
1973 unlink(tmpfile);
1974 return error("File %s corrupted", sha1_to_hex(sha1));
1976 if (hashcmp(sha1, real_sha1)) {
1977 unlink(tmpfile);
1978 return error("File %s has bad hash", sha1_to_hex(sha1));
1981 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1984 int has_pack_index(const unsigned char *sha1)
1986 struct stat st;
1987 if (stat(sha1_pack_index_name(sha1), &st))
1988 return 0;
1989 return 1;
1992 int has_pack_file(const unsigned char *sha1)
1994 struct stat st;
1995 if (stat(sha1_pack_name(sha1), &st))
1996 return 0;
1997 return 1;
2000 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2002 struct pack_entry e;
2003 return find_pack_entry(sha1, &e, ignore_packed);
2006 int has_sha1_file(const unsigned char *sha1)
2008 struct stat st;
2009 struct pack_entry e;
2011 if (find_pack_entry(sha1, &e, NULL))
2012 return 1;
2013 return find_sha1_file(sha1, &st) ? 1 : 0;
2017 * reads from fd as long as possible into a supplied buffer of size bytes.
2018 * If necessary the buffer's size is increased using realloc()
2020 * returns 0 if anything went fine and -1 otherwise
2022 * NOTE: both buf and size may change, but even when -1 is returned
2023 * you still have to free() it yourself.
2025 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
2027 char* buf = *return_buf;
2028 unsigned long size = *return_size;
2029 int iret;
2030 unsigned long off = 0;
2032 do {
2033 iret = xread(fd, buf + off, size - off);
2034 if (iret > 0) {
2035 off += iret;
2036 if (off == size) {
2037 size *= 2;
2038 buf = xrealloc(buf, size);
2041 } while (iret > 0);
2043 *return_buf = buf;
2044 *return_size = off;
2046 if (iret < 0)
2047 return -1;
2048 return 0;
2051 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2053 unsigned long size = 4096;
2054 char *buf = xmalloc(size);
2055 int ret;
2057 if (read_pipe(fd, &buf, &size)) {
2058 free(buf);
2059 return -1;
2062 if (!type)
2063 type = blob_type;
2064 if (write_object)
2065 ret = write_sha1_file(buf, size, type, sha1);
2066 else
2067 ret = hash_sha1_file(buf, size, type, sha1);
2068 free(buf);
2069 return ret;
2072 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2073 enum object_type type, const char *path)
2075 size_t size = xsize_t(st->st_size);
2076 void *buf = NULL;
2077 int ret, re_allocated = 0;
2079 if (size)
2080 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2081 close(fd);
2083 if (!type)
2084 type = OBJ_BLOB;
2087 * Convert blobs to git internal format
2089 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2090 unsigned long nsize = size;
2091 char *nbuf = buf;
2092 if (convert_to_git(path, &nbuf, &nsize)) {
2093 if (size)
2094 munmap(buf, size);
2095 size = nsize;
2096 buf = nbuf;
2097 re_allocated = 1;
2101 if (write_object)
2102 ret = write_sha1_file(buf, size, typename(type), sha1);
2103 else
2104 ret = hash_sha1_file(buf, size, typename(type), sha1);
2105 if (re_allocated) {
2106 free(buf);
2107 return ret;
2109 if (size)
2110 munmap(buf, size);
2111 return ret;
2114 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2116 int fd;
2117 char *target;
2118 size_t len;
2120 switch (st->st_mode & S_IFMT) {
2121 case S_IFREG:
2122 fd = open(path, O_RDONLY);
2123 if (fd < 0)
2124 return error("open(\"%s\"): %s", path,
2125 strerror(errno));
2126 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2127 return error("%s: failed to insert into database",
2128 path);
2129 break;
2130 case S_IFLNK:
2131 len = xsize_t(st->st_size);
2132 target = xmalloc(len + 1);
2133 if (readlink(path, target, len + 1) != st->st_size) {
2134 char *errstr = strerror(errno);
2135 free(target);
2136 return error("readlink(\"%s\"): %s", path,
2137 errstr);
2139 if (!write_object)
2140 hash_sha1_file(target, len, blob_type, sha1);
2141 else if (write_sha1_file(target, len, blob_type, sha1))
2142 return error("%s: failed to insert into database",
2143 path);
2144 free(target);
2145 break;
2146 default:
2147 return error("%s: unsupported file type", path);
2149 return 0;
2152 int read_pack_header(int fd, struct pack_header *header)
2154 char *c = (char*)header;
2155 ssize_t remaining = sizeof(struct pack_header);
2156 do {
2157 ssize_t r = xread(fd, c, remaining);
2158 if (r <= 0)
2159 /* "eof before pack header was fully read" */
2160 return PH_ERROR_EOF;
2161 remaining -= r;
2162 c += r;
2163 } while (remaining > 0);
2164 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2165 /* "protocol error (pack signature mismatch detected)" */
2166 return PH_ERROR_PACK_SIGNATURE;
2167 if (!pack_version_ok(header->hdr_version))
2168 /* "protocol error (pack version unsupported)" */
2169 return PH_ERROR_PROTOCOL;
2170 return 0;