fast-import: Add tip about importing renames.
[git/jrn.git] / sha1_file.c
blob0d4bf80e74b16aa089acd06c24ad1b1a039e65b5
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
9 #include "cache.h"
10 #include "delta.h"
11 #include "pack.h"
12 #include "blob.h"
13 #include "commit.h"
14 #include "tag.h"
15 #include "tree.h"
17 #ifndef O_NOATIME
18 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
19 #define O_NOATIME 01000000
20 #else
21 #define O_NOATIME 0
22 #endif
23 #endif
25 #ifdef NO_C99_FORMAT
26 #define SZ_FMT "lu"
27 #else
28 #define SZ_FMT "zu"
29 #endif
31 const unsigned char null_sha1[20];
33 static unsigned int sha1_file_open_flag = O_NOATIME;
35 signed char hexval_table[256] = {
36 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
37 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
38 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
42 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
43 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
44 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
45 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
47 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
48 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
70 int get_sha1_hex(const char *hex, unsigned char *sha1)
72 int i;
73 for (i = 0; i < 20; i++) {
74 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
75 if (val & ~0xff)
76 return -1;
77 *sha1++ = val;
78 hex += 2;
80 return 0;
83 int safe_create_leading_directories(char *path)
85 char *pos = path;
86 struct stat st;
88 if (*pos == '/')
89 pos++;
91 while (pos) {
92 pos = strchr(pos, '/');
93 if (!pos)
94 break;
95 *pos = 0;
96 if (!stat(path, &st)) {
97 /* path exists */
98 if (!S_ISDIR(st.st_mode)) {
99 *pos = '/';
100 return -3;
103 else if (mkdir(path, 0777)) {
104 *pos = '/';
105 return -1;
107 else if (adjust_shared_perm(path)) {
108 *pos = '/';
109 return -2;
111 *pos++ = '/';
113 return 0;
116 char * sha1_to_hex(const unsigned char *sha1)
118 static int bufno;
119 static char hexbuffer[4][50];
120 static const char hex[] = "0123456789abcdef";
121 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
122 int i;
124 for (i = 0; i < 20; i++) {
125 unsigned int val = *sha1++;
126 *buf++ = hex[val >> 4];
127 *buf++ = hex[val & 0xf];
129 *buf = '\0';
131 return buffer;
134 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
136 int i;
137 for (i = 0; i < 20; i++) {
138 static char hex[] = "0123456789abcdef";
139 unsigned int val = sha1[i];
140 char *pos = pathbuf + i*2 + (i > 0);
141 *pos++ = hex[val >> 4];
142 *pos = hex[val & 0xf];
147 * NOTE! This returns a statically allocated buffer, so you have to be
148 * careful about using it. Do a "xstrdup()" if you need to save the
149 * filename.
151 * Also note that this returns the location for creating. Reading
152 * SHA1 file can happen from any alternate directory listed in the
153 * DB_ENVIRONMENT environment variable if it is not found in
154 * the primary object database.
156 char *sha1_file_name(const unsigned char *sha1)
158 static char *name, *base;
160 if (!base) {
161 const char *sha1_file_directory = get_object_directory();
162 int len = strlen(sha1_file_directory);
163 base = xmalloc(len + 60);
164 memcpy(base, sha1_file_directory, len);
165 memset(base+len, 0, 60);
166 base[len] = '/';
167 base[len+3] = '/';
168 name = base + len + 1;
170 fill_sha1_path(name, sha1);
171 return base;
174 char *sha1_pack_name(const unsigned char *sha1)
176 static const char hex[] = "0123456789abcdef";
177 static char *name, *base, *buf;
178 int i;
180 if (!base) {
181 const char *sha1_file_directory = get_object_directory();
182 int len = strlen(sha1_file_directory);
183 base = xmalloc(len + 60);
184 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
185 name = base + len + 11;
188 buf = name;
190 for (i = 0; i < 20; i++) {
191 unsigned int val = *sha1++;
192 *buf++ = hex[val >> 4];
193 *buf++ = hex[val & 0xf];
196 return base;
199 char *sha1_pack_index_name(const unsigned char *sha1)
201 static const char hex[] = "0123456789abcdef";
202 static char *name, *base, *buf;
203 int i;
205 if (!base) {
206 const char *sha1_file_directory = get_object_directory();
207 int len = strlen(sha1_file_directory);
208 base = xmalloc(len + 60);
209 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
210 name = base + len + 11;
213 buf = name;
215 for (i = 0; i < 20; i++) {
216 unsigned int val = *sha1++;
217 *buf++ = hex[val >> 4];
218 *buf++ = hex[val & 0xf];
221 return base;
224 struct alternate_object_database *alt_odb_list;
225 static struct alternate_object_database **alt_odb_tail;
227 static void read_info_alternates(const char * alternates, int depth);
230 * Prepare alternate object database registry.
232 * The variable alt_odb_list points at the list of struct
233 * alternate_object_database. The elements on this list come from
234 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
235 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
236 * whose contents is similar to that environment variable but can be
237 * LF separated. Its base points at a statically allocated buffer that
238 * contains "/the/directory/corresponding/to/.git/objects/...", while
239 * its name points just after the slash at the end of ".git/objects/"
240 * in the example above, and has enough space to hold 40-byte hex
241 * SHA1, an extra slash for the first level indirection, and the
242 * terminating NUL.
244 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
246 struct stat st;
247 const char *objdir = get_object_directory();
248 struct alternate_object_database *ent;
249 struct alternate_object_database *alt;
250 /* 43 = 40-byte + 2 '/' + terminating NUL */
251 int pfxlen = len;
252 int entlen = pfxlen + 43;
253 int base_len = -1;
255 if (*entry != '/' && relative_base) {
256 /* Relative alt-odb */
257 if (base_len < 0)
258 base_len = strlen(relative_base) + 1;
259 entlen += base_len;
260 pfxlen += base_len;
262 ent = xmalloc(sizeof(*ent) + entlen);
264 if (*entry != '/' && relative_base) {
265 memcpy(ent->base, relative_base, base_len - 1);
266 ent->base[base_len - 1] = '/';
267 memcpy(ent->base + base_len, entry, len);
269 else
270 memcpy(ent->base, entry, pfxlen);
272 ent->name = ent->base + pfxlen + 1;
273 ent->base[pfxlen + 3] = '/';
274 ent->base[pfxlen] = ent->base[entlen-1] = 0;
276 /* Detect cases where alternate disappeared */
277 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
278 error("object directory %s does not exist; "
279 "check .git/objects/info/alternates.",
280 ent->base);
281 free(ent);
282 return -1;
285 /* Prevent the common mistake of listing the same
286 * thing twice, or object directory itself.
288 for (alt = alt_odb_list; alt; alt = alt->next) {
289 if (!memcmp(ent->base, alt->base, pfxlen)) {
290 free(ent);
291 return -1;
294 if (!memcmp(ent->base, objdir, pfxlen)) {
295 free(ent);
296 return -1;
299 /* add the alternate entry */
300 *alt_odb_tail = ent;
301 alt_odb_tail = &(ent->next);
302 ent->next = NULL;
304 /* recursively add alternates */
305 read_info_alternates(ent->base, depth + 1);
307 ent->base[pfxlen] = '/';
309 return 0;
312 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
313 const char *relative_base, int depth)
315 const char *cp, *last;
317 if (depth > 5) {
318 error("%s: ignoring alternate object stores, nesting too deep.",
319 relative_base);
320 return;
323 last = alt;
324 while (last < ep) {
325 cp = last;
326 if (cp < ep && *cp == '#') {
327 while (cp < ep && *cp != sep)
328 cp++;
329 last = cp + 1;
330 continue;
332 while (cp < ep && *cp != sep)
333 cp++;
334 if (last != cp) {
335 if ((*last != '/') && depth) {
336 error("%s: ignoring relative alternate object store %s",
337 relative_base, last);
338 } else {
339 link_alt_odb_entry(last, cp - last,
340 relative_base, depth);
343 while (cp < ep && *cp == sep)
344 cp++;
345 last = cp;
349 static void read_info_alternates(const char * relative_base, int depth)
351 char *map;
352 struct stat st;
353 char path[PATH_MAX];
354 int fd;
356 sprintf(path, "%s/info/alternates", relative_base);
357 fd = open(path, O_RDONLY);
358 if (fd < 0)
359 return;
360 if (fstat(fd, &st) || (st.st_size == 0)) {
361 close(fd);
362 return;
364 map = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
365 close(fd);
367 link_alt_odb_entries(map, map + st.st_size, '\n', relative_base, depth);
369 munmap(map, st.st_size);
372 void prepare_alt_odb(void)
374 const char *alt;
376 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
377 if (!alt) alt = "";
379 if (alt_odb_tail)
380 return;
381 alt_odb_tail = &alt_odb_list;
382 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
384 read_info_alternates(get_object_directory(), 0);
387 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
389 char *name = sha1_file_name(sha1);
390 struct alternate_object_database *alt;
392 if (!stat(name, st))
393 return name;
394 prepare_alt_odb();
395 for (alt = alt_odb_list; alt; alt = alt->next) {
396 name = alt->name;
397 fill_sha1_path(name, sha1);
398 if (!stat(alt->base, st))
399 return alt->base;
401 return NULL;
404 static unsigned int pack_used_ctr;
405 static unsigned int pack_mmap_calls;
406 static unsigned int peak_pack_open_windows;
407 static unsigned int pack_open_windows;
408 static size_t peak_pack_mapped;
409 static size_t pack_mapped;
410 static size_t page_size;
411 struct packed_git *packed_git;
413 void pack_report()
415 fprintf(stderr,
416 "pack_report: getpagesize() = %10" SZ_FMT "\n"
417 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
418 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
419 page_size,
420 packed_git_window_size,
421 packed_git_limit);
422 fprintf(stderr,
423 "pack_report: pack_used_ctr = %10u\n"
424 "pack_report: pack_mmap_calls = %10u\n"
425 "pack_report: pack_open_windows = %10u / %10u\n"
426 "pack_report: pack_mapped = "
427 "%10" SZ_FMT " / %10" SZ_FMT "\n",
428 pack_used_ctr,
429 pack_mmap_calls,
430 pack_open_windows, peak_pack_open_windows,
431 pack_mapped, peak_pack_mapped);
434 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
435 void **idx_map_)
437 void *idx_map;
438 uint32_t *index;
439 unsigned long idx_size;
440 int nr, i;
441 int fd = open(path, O_RDONLY);
442 struct stat st;
443 if (fd < 0)
444 return -1;
445 if (fstat(fd, &st)) {
446 close(fd);
447 return -1;
449 idx_size = st.st_size;
450 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
451 close(fd);
453 index = idx_map;
454 *idx_map_ = idx_map;
455 *idx_size_ = idx_size;
457 /* check index map */
458 if (idx_size < 4*256 + 20 + 20)
459 return error("index file %s is too small", path);
461 /* a future index format would start with this, as older git
462 * binaries would fail the non-monotonic index check below.
463 * give a nicer warning to the user if we can.
465 if (index[0] == htonl(PACK_IDX_SIGNATURE))
466 return error("index file %s is a newer version"
467 " and is not supported by this binary"
468 " (try upgrading GIT to a newer version)",
469 path);
471 nr = 0;
472 for (i = 0; i < 256; i++) {
473 unsigned int n = ntohl(index[i]);
474 if (n < nr)
475 return error("non-monotonic index %s", path);
476 nr = n;
480 * Total size:
481 * - 256 index entries 4 bytes each
482 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
483 * - 20-byte SHA1 of the packfile
484 * - 20-byte SHA1 file checksum
486 if (idx_size != 4*256 + nr * 24 + 20 + 20)
487 return error("wrong index file size in %s", path);
489 return 0;
492 static void scan_windows(struct packed_git *p,
493 struct packed_git **lru_p,
494 struct pack_window **lru_w,
495 struct pack_window **lru_l)
497 struct pack_window *w, *w_l;
499 for (w_l = NULL, w = p->windows; w; w = w->next) {
500 if (!w->inuse_cnt) {
501 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
502 *lru_p = p;
503 *lru_w = w;
504 *lru_l = w_l;
507 w_l = w;
511 static int unuse_one_window(struct packed_git *current)
513 struct packed_git *p, *lru_p = NULL;
514 struct pack_window *lru_w = NULL, *lru_l = NULL;
516 if (current)
517 scan_windows(current, &lru_p, &lru_w, &lru_l);
518 for (p = packed_git; p; p = p->next)
519 scan_windows(p, &lru_p, &lru_w, &lru_l);
520 if (lru_p) {
521 munmap(lru_w->base, lru_w->len);
522 pack_mapped -= lru_w->len;
523 if (lru_l)
524 lru_l->next = lru_w->next;
525 else {
526 lru_p->windows = lru_w->next;
527 if (!lru_p->windows && lru_p != current) {
528 close(lru_p->pack_fd);
529 lru_p->pack_fd = -1;
532 free(lru_w);
533 pack_open_windows--;
534 return 1;
536 return 0;
539 void release_pack_memory(size_t need)
541 size_t cur = pack_mapped;
542 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
543 ; /* nothing */
546 void unuse_pack(struct pack_window **w_cursor)
548 struct pack_window *w = *w_cursor;
549 if (w) {
550 w->inuse_cnt--;
551 *w_cursor = NULL;
556 * Do not call this directly as this leaks p->pack_fd on error return;
557 * call open_packed_git() instead.
559 static int open_packed_git_1(struct packed_git *p)
561 struct stat st;
562 struct pack_header hdr;
563 unsigned char sha1[20];
564 unsigned char *idx_sha1;
565 long fd_flag;
567 p->pack_fd = open(p->pack_name, O_RDONLY);
568 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
569 return -1;
571 /* If we created the struct before we had the pack we lack size. */
572 if (!p->pack_size) {
573 if (!S_ISREG(st.st_mode))
574 return error("packfile %s not a regular file", p->pack_name);
575 p->pack_size = st.st_size;
576 } else if (p->pack_size != st.st_size)
577 return error("packfile %s size changed", p->pack_name);
579 /* We leave these file descriptors open with sliding mmap;
580 * there is no point keeping them open across exec(), though.
582 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
583 if (fd_flag < 0)
584 return error("cannot determine file descriptor flags");
585 fd_flag |= FD_CLOEXEC;
586 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
587 return error("cannot set FD_CLOEXEC");
589 /* Verify we recognize this pack file format. */
590 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
591 return error("file %s is far too short to be a packfile", p->pack_name);
592 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
593 return error("file %s is not a GIT packfile", p->pack_name);
594 if (!pack_version_ok(hdr.hdr_version))
595 return error("packfile %s is version %u and not supported"
596 " (try upgrading GIT to a newer version)",
597 p->pack_name, ntohl(hdr.hdr_version));
599 /* Verify the pack matches its index. */
600 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
601 return error("packfile %s claims to have %u objects"
602 " while index size indicates %u objects",
603 p->pack_name, ntohl(hdr.hdr_entries),
604 num_packed_objects(p));
605 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
606 return error("end of packfile %s is unavailable", p->pack_name);
607 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
608 return error("packfile %s signature is unavailable", p->pack_name);
609 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
610 if (hashcmp(sha1, idx_sha1))
611 return error("packfile %s does not match index", p->pack_name);
612 return 0;
615 static int open_packed_git(struct packed_git *p)
617 if (!open_packed_git_1(p))
618 return 0;
619 if (p->pack_fd != -1) {
620 close(p->pack_fd);
621 p->pack_fd = -1;
623 return -1;
626 static int in_window(struct pack_window *win, unsigned long offset)
628 /* We must promise at least 20 bytes (one hash) after the
629 * offset is available from this window, otherwise the offset
630 * is not actually in this window and a different window (which
631 * has that one hash excess) must be used. This is to support
632 * the object header and delta base parsing routines below.
634 off_t win_off = win->offset;
635 return win_off <= offset
636 && (offset + 20) <= (win_off + win->len);
639 unsigned char* use_pack(struct packed_git *p,
640 struct pack_window **w_cursor,
641 unsigned long offset,
642 unsigned int *left)
644 struct pack_window *win = *w_cursor;
646 if (p->pack_fd == -1 && open_packed_git(p))
647 die("packfile %s cannot be accessed", p->pack_name);
649 /* Since packfiles end in a hash of their content and its
650 * pointless to ask for an offset into the middle of that
651 * hash, and the in_window function above wouldn't match
652 * don't allow an offset too close to the end of the file.
654 if (offset > (p->pack_size - 20))
655 die("offset beyond end of packfile (truncated pack?)");
657 if (!win || !in_window(win, offset)) {
658 if (win)
659 win->inuse_cnt--;
660 for (win = p->windows; win; win = win->next) {
661 if (in_window(win, offset))
662 break;
664 if (!win) {
665 if (!page_size)
666 page_size = getpagesize();
667 win = xcalloc(1, sizeof(*win));
668 win->offset = (offset / page_size) * page_size;
669 win->len = p->pack_size - win->offset;
670 if (win->len > packed_git_window_size)
671 win->len = packed_git_window_size;
672 pack_mapped += win->len;
673 while (packed_git_limit < pack_mapped
674 && unuse_one_window(p))
675 ; /* nothing */
676 win->base = xmmap(NULL, win->len,
677 PROT_READ, MAP_PRIVATE,
678 p->pack_fd, win->offset);
679 if (win->base == MAP_FAILED)
680 die("packfile %s cannot be mapped: %s",
681 p->pack_name,
682 strerror(errno));
683 pack_mmap_calls++;
684 pack_open_windows++;
685 if (pack_mapped > peak_pack_mapped)
686 peak_pack_mapped = pack_mapped;
687 if (pack_open_windows > peak_pack_open_windows)
688 peak_pack_open_windows = pack_open_windows;
689 win->next = p->windows;
690 p->windows = win;
693 if (win != *w_cursor) {
694 win->last_used = pack_used_ctr++;
695 win->inuse_cnt++;
696 *w_cursor = win;
698 offset -= win->offset;
699 if (left)
700 *left = win->len - offset;
701 return win->base + offset;
704 struct packed_git *add_packed_git(char *path, int path_len, int local)
706 struct stat st;
707 struct packed_git *p;
708 unsigned long idx_size;
709 void *idx_map;
710 unsigned char sha1[20];
712 if (check_packed_git_idx(path, &idx_size, &idx_map))
713 return NULL;
715 /* do we have a corresponding .pack file? */
716 strcpy(path + path_len - 4, ".pack");
717 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
718 munmap(idx_map, idx_size);
719 return NULL;
721 /* ok, it looks sane as far as we can check without
722 * actually mapping the pack file.
724 p = xmalloc(sizeof(*p) + path_len + 2);
725 strcpy(p->pack_name, path);
726 p->index_size = idx_size;
727 p->pack_size = st.st_size;
728 p->index_base = idx_map;
729 p->next = NULL;
730 p->windows = NULL;
731 p->pack_fd = -1;
732 p->pack_local = local;
733 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
734 hashcpy(p->sha1, sha1);
735 return p;
738 struct packed_git *parse_pack_index(unsigned char *sha1)
740 char *path = sha1_pack_index_name(sha1);
741 return parse_pack_index_file(sha1, path);
744 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
746 struct packed_git *p;
747 unsigned long idx_size;
748 void *idx_map;
749 char *path;
751 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
752 return NULL;
754 path = sha1_pack_name(sha1);
756 p = xmalloc(sizeof(*p) + strlen(path) + 2);
757 strcpy(p->pack_name, path);
758 p->index_size = idx_size;
759 p->pack_size = 0;
760 p->index_base = idx_map;
761 p->next = NULL;
762 p->windows = NULL;
763 p->pack_fd = -1;
764 hashcpy(p->sha1, sha1);
765 return p;
768 void install_packed_git(struct packed_git *pack)
770 pack->next = packed_git;
771 packed_git = pack;
774 static void prepare_packed_git_one(char *objdir, int local)
776 char path[PATH_MAX];
777 int len;
778 DIR *dir;
779 struct dirent *de;
781 sprintf(path, "%s/pack", objdir);
782 len = strlen(path);
783 dir = opendir(path);
784 if (!dir) {
785 if (errno != ENOENT)
786 error("unable to open object pack directory: %s: %s",
787 path, strerror(errno));
788 return;
790 path[len++] = '/';
791 while ((de = readdir(dir)) != NULL) {
792 int namelen = strlen(de->d_name);
793 struct packed_git *p;
795 if (!has_extension(de->d_name, ".idx"))
796 continue;
798 /* Don't reopen a pack we already have. */
799 strcpy(path + len, de->d_name);
800 for (p = packed_git; p; p = p->next) {
801 if (!memcmp(path, p->pack_name, len + namelen - 4))
802 break;
804 if (p)
805 continue;
806 /* See if it really is a valid .idx file with corresponding
807 * .pack file that we can map.
809 p = add_packed_git(path, len + namelen, local);
810 if (!p)
811 continue;
812 install_packed_git(p);
814 closedir(dir);
817 static int prepare_packed_git_run_once = 0;
818 void prepare_packed_git(void)
820 struct alternate_object_database *alt;
822 if (prepare_packed_git_run_once)
823 return;
824 prepare_packed_git_one(get_object_directory(), 1);
825 prepare_alt_odb();
826 for (alt = alt_odb_list; alt; alt = alt->next) {
827 alt->name[-1] = 0;
828 prepare_packed_git_one(alt->base, 0);
829 alt->name[-1] = '/';
831 prepare_packed_git_run_once = 1;
834 void reprepare_packed_git(void)
836 prepare_packed_git_run_once = 0;
837 prepare_packed_git();
840 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
842 unsigned char real_sha1[20];
843 hash_sha1_file(map, size, type, real_sha1);
844 return hashcmp(sha1, real_sha1) ? -1 : 0;
847 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
849 struct stat st;
850 void *map;
851 int fd;
852 char *filename = find_sha1_file(sha1, &st);
854 if (!filename) {
855 return NULL;
858 fd = open(filename, O_RDONLY | sha1_file_open_flag);
859 if (fd < 0) {
860 /* See if it works without O_NOATIME */
861 switch (sha1_file_open_flag) {
862 default:
863 fd = open(filename, O_RDONLY);
864 if (fd >= 0)
865 break;
866 /* Fallthrough */
867 case 0:
868 return NULL;
871 /* If it failed once, it will probably fail again.
872 * Stop using O_NOATIME
874 sha1_file_open_flag = 0;
876 map = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
877 close(fd);
878 *size = st.st_size;
879 return map;
882 int legacy_loose_object(unsigned char *map)
884 unsigned int word;
887 * Is it a zlib-compressed buffer? If so, the first byte
888 * must be 0x78 (15-bit window size, deflated), and the
889 * first 16-bit word is evenly divisible by 31
891 word = (map[0] << 8) + map[1];
892 if (map[0] == 0x78 && !(word % 31))
893 return 1;
894 else
895 return 0;
898 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
900 unsigned shift;
901 unsigned char c;
902 unsigned long size;
903 unsigned long used = 0;
905 c = buf[used++];
906 *type = (c >> 4) & 7;
907 size = c & 15;
908 shift = 4;
909 while (c & 0x80) {
910 if (len <= used)
911 return 0;
912 if (sizeof(long) * 8 <= shift)
913 return 0;
914 c = buf[used++];
915 size += (c & 0x7f) << shift;
916 shift += 7;
918 *sizep = size;
919 return used;
922 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
924 unsigned long size, used;
925 static const char valid_loose_object_type[8] = {
926 0, /* OBJ_EXT */
927 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
928 0, /* "delta" and others are invalid in a loose object */
930 enum object_type type;
932 /* Get the data stream */
933 memset(stream, 0, sizeof(*stream));
934 stream->next_in = map;
935 stream->avail_in = mapsize;
936 stream->next_out = buffer;
937 stream->avail_out = bufsiz;
939 if (legacy_loose_object(map)) {
940 inflateInit(stream);
941 return inflate(stream, 0);
944 used = unpack_object_header_gently(map, mapsize, &type, &size);
945 if (!used || !valid_loose_object_type[type])
946 return -1;
947 map += used;
948 mapsize -= used;
950 /* Set up the stream for the rest.. */
951 stream->next_in = map;
952 stream->avail_in = mapsize;
953 inflateInit(stream);
955 /* And generate the fake traditional header */
956 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
957 type_names[type], size);
958 return 0;
961 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
963 int bytes = strlen(buffer) + 1;
964 unsigned char *buf = xmalloc(1+size);
965 unsigned long n;
967 n = stream->total_out - bytes;
968 if (n > size)
969 n = size;
970 memcpy(buf, (char *) buffer + bytes, n);
971 bytes = n;
972 if (bytes < size) {
973 stream->next_out = buf + bytes;
974 stream->avail_out = size - bytes;
975 while (inflate(stream, Z_FINISH) == Z_OK)
976 /* nothing */;
978 buf[size] = 0;
979 inflateEnd(stream);
980 return buf;
984 * We used to just use "sscanf()", but that's actually way
985 * too permissive for what we want to check. So do an anal
986 * object header parse by hand.
988 static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
990 int i;
991 unsigned long size;
994 * The type can be at most ten bytes (including the
995 * terminating '\0' that we add), and is followed by
996 * a space.
998 i = 10;
999 for (;;) {
1000 char c = *hdr++;
1001 if (c == ' ')
1002 break;
1003 if (!--i)
1004 return -1;
1005 *type++ = c;
1007 *type = 0;
1010 * The length must follow immediately, and be in canonical
1011 * decimal format (ie "010" is not valid).
1013 size = *hdr++ - '0';
1014 if (size > 9)
1015 return -1;
1016 if (size) {
1017 for (;;) {
1018 unsigned long c = *hdr - '0';
1019 if (c > 9)
1020 break;
1021 hdr++;
1022 size = size * 10 + c;
1025 *sizep = size;
1028 * The length must be followed by a zero byte
1030 return *hdr ? -1 : 0;
1033 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
1035 int ret;
1036 z_stream stream;
1037 char hdr[8192];
1039 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1040 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
1041 return NULL;
1043 return unpack_sha1_rest(&stream, hdr, *size);
1046 static unsigned long get_delta_base(struct packed_git *p,
1047 struct pack_window **w_curs,
1048 unsigned long offset,
1049 enum object_type kind,
1050 unsigned long delta_obj_offset,
1051 unsigned long *base_obj_offset)
1053 unsigned char *base_info = use_pack(p, w_curs, offset, NULL);
1054 unsigned long base_offset;
1056 /* use_pack() assured us we have [base_info, base_info + 20)
1057 * as a range that we can look at without walking off the
1058 * end of the mapped window. Its actually the hash size
1059 * that is assured. An OFS_DELTA longer than the hash size
1060 * is stupid, as then a REF_DELTA would be smaller to store.
1062 if (kind == OBJ_OFS_DELTA) {
1063 unsigned used = 0;
1064 unsigned char c = base_info[used++];
1065 base_offset = c & 127;
1066 while (c & 128) {
1067 base_offset += 1;
1068 if (!base_offset || base_offset & ~(~0UL >> 7))
1069 die("offset value overflow for delta base object");
1070 c = base_info[used++];
1071 base_offset = (base_offset << 7) + (c & 127);
1073 base_offset = delta_obj_offset - base_offset;
1074 if (base_offset >= delta_obj_offset)
1075 die("delta base offset out of bound");
1076 offset += used;
1077 } else if (kind == OBJ_REF_DELTA) {
1078 /* The base entry _must_ be in the same pack */
1079 base_offset = find_pack_entry_one(base_info, p);
1080 if (!base_offset)
1081 die("failed to find delta-pack base object %s",
1082 sha1_to_hex(base_info));
1083 offset += 20;
1084 } else
1085 die("I am totally screwed");
1086 *base_obj_offset = base_offset;
1087 return offset;
1090 /* forward declaration for a mutually recursive function */
1091 static int packed_object_info(struct packed_git *p, unsigned long offset,
1092 char *type, unsigned long *sizep);
1094 static int packed_delta_info(struct packed_git *p,
1095 struct pack_window **w_curs,
1096 unsigned long offset,
1097 enum object_type kind,
1098 unsigned long obj_offset,
1099 char *type,
1100 unsigned long *sizep)
1102 unsigned long base_offset;
1104 offset = get_delta_base(p, w_curs, offset, kind,
1105 obj_offset, &base_offset);
1107 /* We choose to only get the type of the base object and
1108 * ignore potentially corrupt pack file that expects the delta
1109 * based on a base with a wrong size. This saves tons of
1110 * inflate() calls.
1112 if (packed_object_info(p, base_offset, type, NULL))
1113 die("cannot get info for delta-pack base");
1115 if (sizep) {
1116 const unsigned char *data;
1117 unsigned char delta_head[20], *in;
1118 unsigned long result_size;
1119 z_stream stream;
1120 int st;
1122 memset(&stream, 0, sizeof(stream));
1123 stream.next_out = delta_head;
1124 stream.avail_out = sizeof(delta_head);
1126 inflateInit(&stream);
1127 do {
1128 in = use_pack(p, w_curs, offset, &stream.avail_in);
1129 stream.next_in = in;
1130 st = inflate(&stream, Z_FINISH);
1131 offset += stream.next_in - in;
1132 } while ((st == Z_OK || st == Z_BUF_ERROR)
1133 && stream.total_out < sizeof(delta_head));
1134 inflateEnd(&stream);
1135 if ((st != Z_STREAM_END) &&
1136 stream.total_out != sizeof(delta_head))
1137 die("delta data unpack-initial failed");
1139 /* Examine the initial part of the delta to figure out
1140 * the result size.
1142 data = delta_head;
1144 /* ignore base size */
1145 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1147 /* Read the result size */
1148 result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1149 *sizep = result_size;
1151 return 0;
1154 static unsigned long unpack_object_header(struct packed_git *p,
1155 struct pack_window **w_curs,
1156 unsigned long offset,
1157 enum object_type *type,
1158 unsigned long *sizep)
1160 unsigned char *base;
1161 unsigned int left;
1162 unsigned long used;
1164 /* use_pack() assures us we have [base, base + 20) available
1165 * as a range that we can look at at. (Its actually the hash
1166 * size that is assured.) With our object header encoding
1167 * the maximum deflated object size is 2^137, which is just
1168 * insane, so we know won't exceed what we have been given.
1170 base = use_pack(p, w_curs, offset, &left);
1171 used = unpack_object_header_gently(base, left, type, sizep);
1172 if (!used)
1173 die("object offset outside of pack file");
1175 return offset + used;
1178 void packed_object_info_detail(struct packed_git *p,
1179 unsigned long offset,
1180 char *type,
1181 unsigned long *size,
1182 unsigned long *store_size,
1183 unsigned int *delta_chain_length,
1184 unsigned char *base_sha1)
1186 struct pack_window *w_curs = NULL;
1187 unsigned long obj_offset, val;
1188 unsigned char *next_sha1;
1189 enum object_type kind;
1191 *delta_chain_length = 0;
1192 obj_offset = offset;
1193 offset = unpack_object_header(p, &w_curs, offset, &kind, size);
1195 for (;;) {
1196 switch (kind) {
1197 default:
1198 die("pack %s contains unknown object type %d",
1199 p->pack_name, kind);
1200 case OBJ_COMMIT:
1201 case OBJ_TREE:
1202 case OBJ_BLOB:
1203 case OBJ_TAG:
1204 strcpy(type, type_names[kind]);
1205 *store_size = 0; /* notyet */
1206 unuse_pack(&w_curs);
1207 return;
1208 case OBJ_OFS_DELTA:
1209 get_delta_base(p, &w_curs, offset, kind,
1210 obj_offset, &offset);
1211 if (*delta_chain_length == 0) {
1212 /* TODO: find base_sha1 as pointed by offset */
1214 break;
1215 case OBJ_REF_DELTA:
1216 next_sha1 = use_pack(p, &w_curs, offset, NULL);
1217 if (*delta_chain_length == 0)
1218 hashcpy(base_sha1, next_sha1);
1219 offset = find_pack_entry_one(next_sha1, p);
1220 break;
1222 obj_offset = offset;
1223 offset = unpack_object_header(p, &w_curs, offset, &kind, &val);
1224 (*delta_chain_length)++;
1228 static int packed_object_info(struct packed_git *p, unsigned long offset,
1229 char *type, unsigned long *sizep)
1231 struct pack_window *w_curs = NULL;
1232 unsigned long size, obj_offset = offset;
1233 enum object_type kind;
1234 int r;
1236 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1238 switch (kind) {
1239 case OBJ_OFS_DELTA:
1240 case OBJ_REF_DELTA:
1241 r = packed_delta_info(p, &w_curs, offset, kind,
1242 obj_offset, type, sizep);
1243 unuse_pack(&w_curs);
1244 return r;
1245 case OBJ_COMMIT:
1246 case OBJ_TREE:
1247 case OBJ_BLOB:
1248 case OBJ_TAG:
1249 strcpy(type, type_names[kind]);
1250 unuse_pack(&w_curs);
1251 break;
1252 default:
1253 die("pack %s contains unknown object type %d",
1254 p->pack_name, kind);
1256 if (sizep)
1257 *sizep = size;
1258 return 0;
1261 static void *unpack_compressed_entry(struct packed_git *p,
1262 struct pack_window **w_curs,
1263 unsigned long offset,
1264 unsigned long size)
1266 int st;
1267 z_stream stream;
1268 unsigned char *buffer, *in;
1270 buffer = xmalloc(size + 1);
1271 buffer[size] = 0;
1272 memset(&stream, 0, sizeof(stream));
1273 stream.next_out = buffer;
1274 stream.avail_out = size;
1276 inflateInit(&stream);
1277 do {
1278 in = use_pack(p, w_curs, offset, &stream.avail_in);
1279 stream.next_in = in;
1280 st = inflate(&stream, Z_FINISH);
1281 offset += stream.next_in - in;
1282 } while (st == Z_OK || st == Z_BUF_ERROR);
1283 inflateEnd(&stream);
1284 if ((st != Z_STREAM_END) || stream.total_out != size) {
1285 free(buffer);
1286 return NULL;
1289 return buffer;
1292 static void *unpack_delta_entry(struct packed_git *p,
1293 struct pack_window **w_curs,
1294 unsigned long offset,
1295 unsigned long delta_size,
1296 enum object_type kind,
1297 unsigned long obj_offset,
1298 char *type,
1299 unsigned long *sizep)
1301 void *delta_data, *result, *base;
1302 unsigned long result_size, base_size, base_offset;
1304 offset = get_delta_base(p, w_curs, offset, kind,
1305 obj_offset, &base_offset);
1306 base = unpack_entry(p, base_offset, type, &base_size);
1307 if (!base)
1308 die("failed to read delta base object at %lu from %s",
1309 base_offset, p->pack_name);
1311 delta_data = unpack_compressed_entry(p, w_curs, offset, delta_size);
1312 result = patch_delta(base, base_size,
1313 delta_data, delta_size,
1314 &result_size);
1315 if (!result)
1316 die("failed to apply delta");
1317 free(delta_data);
1318 free(base);
1319 *sizep = result_size;
1320 return result;
1323 void *unpack_entry(struct packed_git *p, unsigned long offset,
1324 char *type, unsigned long *sizep)
1326 struct pack_window *w_curs = NULL;
1327 unsigned long size, obj_offset = offset;
1328 enum object_type kind;
1329 void *retval;
1331 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1332 switch (kind) {
1333 case OBJ_OFS_DELTA:
1334 case OBJ_REF_DELTA:
1335 retval = unpack_delta_entry(p, &w_curs, offset, size,
1336 kind, obj_offset, type, sizep);
1337 break;
1338 case OBJ_COMMIT:
1339 case OBJ_TREE:
1340 case OBJ_BLOB:
1341 case OBJ_TAG:
1342 strcpy(type, type_names[kind]);
1343 *sizep = size;
1344 retval = unpack_compressed_entry(p, &w_curs, offset, size);
1345 break;
1346 default:
1347 die("unknown object type %i in %s", kind, p->pack_name);
1349 unuse_pack(&w_curs);
1350 return retval;
1353 int num_packed_objects(const struct packed_git *p)
1355 /* See check_packed_git_idx() */
1356 return (p->index_size - 20 - 20 - 4*256) / 24;
1359 int nth_packed_object_sha1(const struct packed_git *p, int n,
1360 unsigned char* sha1)
1362 void *index = p->index_base + 256;
1363 if (n < 0 || num_packed_objects(p) <= n)
1364 return -1;
1365 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1366 return 0;
1369 unsigned long find_pack_entry_one(const unsigned char *sha1,
1370 struct packed_git *p)
1372 uint32_t *level1_ofs = p->index_base;
1373 int hi = ntohl(level1_ofs[*sha1]);
1374 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1375 void *index = p->index_base + 256;
1377 do {
1378 int mi = (lo + hi) / 2;
1379 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1380 if (!cmp)
1381 return ntohl(*((uint32_t *)((char *)index + (24 * mi))));
1382 if (cmp > 0)
1383 hi = mi;
1384 else
1385 lo = mi+1;
1386 } while (lo < hi);
1387 return 0;
1390 static int matches_pack_name(struct packed_git *p, const char *ig)
1392 const char *last_c, *c;
1394 if (!strcmp(p->pack_name, ig))
1395 return 0;
1397 for (c = p->pack_name, last_c = c; *c;)
1398 if (*c == '/')
1399 last_c = ++c;
1400 else
1401 ++c;
1402 if (!strcmp(last_c, ig))
1403 return 0;
1405 return 1;
1408 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1410 struct packed_git *p;
1411 unsigned long offset;
1413 prepare_packed_git();
1415 for (p = packed_git; p; p = p->next) {
1416 if (ignore_packed) {
1417 const char **ig;
1418 for (ig = ignore_packed; *ig; ig++)
1419 if (!matches_pack_name(p, *ig))
1420 break;
1421 if (*ig)
1422 continue;
1424 offset = find_pack_entry_one(sha1, p);
1425 if (offset) {
1427 * We are about to tell the caller where they can
1428 * locate the requested object. We better make
1429 * sure the packfile is still here and can be
1430 * accessed before supplying that answer, as
1431 * it may have been deleted since the index
1432 * was loaded!
1434 if (p->pack_fd == -1 && open_packed_git(p)) {
1435 error("packfile %s cannot be accessed", p->pack_name);
1436 continue;
1438 e->offset = offset;
1439 e->p = p;
1440 hashcpy(e->sha1, sha1);
1441 return 1;
1444 return 0;
1447 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1448 struct packed_git *packs)
1450 struct packed_git *p;
1452 for (p = packs; p; p = p->next) {
1453 if (find_pack_entry_one(sha1, p))
1454 return p;
1456 return NULL;
1460 static int sha1_loose_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1462 int status;
1463 unsigned long mapsize, size;
1464 void *map;
1465 z_stream stream;
1466 char hdr[128];
1468 map = map_sha1_file(sha1, &mapsize);
1469 if (!map)
1470 return error("unable to find %s", sha1_to_hex(sha1));
1471 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1472 status = error("unable to unpack %s header",
1473 sha1_to_hex(sha1));
1474 if (parse_sha1_header(hdr, type, &size) < 0)
1475 status = error("unable to parse %s header", sha1_to_hex(sha1));
1476 else {
1477 status = 0;
1478 if (sizep)
1479 *sizep = size;
1481 inflateEnd(&stream);
1482 munmap(map, mapsize);
1483 return status;
1486 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1488 struct pack_entry e;
1490 if (!find_pack_entry(sha1, &e, NULL)) {
1491 reprepare_packed_git();
1492 if (!find_pack_entry(sha1, &e, NULL))
1493 return sha1_loose_object_info(sha1, type, sizep);
1495 return packed_object_info(e.p, e.offset, type, sizep);
1498 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1500 struct pack_entry e;
1502 if (!find_pack_entry(sha1, &e, NULL))
1503 return NULL;
1504 else
1505 return unpack_entry(e.p, e.offset, type, size);
1509 * This is meant to hold a *small* number of objects that you would
1510 * want read_sha1_file() to be able to return, but yet you do not want
1511 * to write them into the object store (e.g. a browse-only
1512 * application).
1514 static struct cached_object {
1515 unsigned char sha1[20];
1516 const char *type;
1517 void *buf;
1518 unsigned long size;
1519 } *cached_objects;
1520 static int cached_object_nr, cached_object_alloc;
1522 static struct cached_object *find_cached_object(const unsigned char *sha1)
1524 int i;
1525 struct cached_object *co = cached_objects;
1527 for (i = 0; i < cached_object_nr; i++, co++) {
1528 if (!hashcmp(co->sha1, sha1))
1529 return co;
1531 return NULL;
1534 int pretend_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *sha1)
1536 struct cached_object *co;
1538 hash_sha1_file(buf, len, type, sha1);
1539 if (has_sha1_file(sha1) || find_cached_object(sha1))
1540 return 0;
1541 if (cached_object_alloc <= cached_object_nr) {
1542 cached_object_alloc = alloc_nr(cached_object_alloc);
1543 cached_objects = xrealloc(cached_objects,
1544 sizeof(*cached_objects) *
1545 cached_object_alloc);
1547 co = &cached_objects[cached_object_nr++];
1548 co->size = len;
1549 co->type = strdup(type);
1550 hashcpy(co->sha1, sha1);
1551 return 0;
1554 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1556 unsigned long mapsize;
1557 void *map, *buf;
1558 struct cached_object *co;
1560 co = find_cached_object(sha1);
1561 if (co) {
1562 buf = xmalloc(co->size + 1);
1563 memcpy(buf, co->buf, co->size);
1564 ((char*)buf)[co->size] = 0;
1565 strcpy(type, co->type);
1566 *size = co->size;
1567 return buf;
1570 buf = read_packed_sha1(sha1, type, size);
1571 if (buf)
1572 return buf;
1573 map = map_sha1_file(sha1, &mapsize);
1574 if (map) {
1575 buf = unpack_sha1_file(map, mapsize, type, size);
1576 munmap(map, mapsize);
1577 return buf;
1579 reprepare_packed_git();
1580 return read_packed_sha1(sha1, type, size);
1583 void *read_object_with_reference(const unsigned char *sha1,
1584 const char *required_type,
1585 unsigned long *size,
1586 unsigned char *actual_sha1_return)
1588 char type[20];
1589 void *buffer;
1590 unsigned long isize;
1591 unsigned char actual_sha1[20];
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 (!strcmp(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 (!strcmp(type, commit_type))
1609 ref_type = "tree ";
1610 else if (!strcmp(type, tag_type))
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 unsigned char *hdr, int *hdrlen)
1633 SHA_CTX c;
1635 /* Generate the header */
1636 *hdrlen = sprintf((char *)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, hdr;
1745 if (use_legacy_headers) {
1746 while (deflate(stream, 0) == Z_OK)
1747 /* nothing */;
1748 return;
1750 if (!strcmp(type, blob_type))
1751 obj_type = OBJ_BLOB;
1752 else if (!strcmp(type, tree_type))
1753 obj_type = OBJ_TREE;
1754 else if (!strcmp(type, commit_type))
1755 obj_type = OBJ_COMMIT;
1756 else if (!strcmp(type, tag_type))
1757 obj_type = OBJ_TAG;
1758 else
1759 die("trying to generate bogus object of type '%s'", type);
1760 hdr = write_binary_header(stream->next_out, obj_type, len);
1761 stream->total_out = hdr;
1762 stream->next_out += hdr;
1763 stream->avail_out -= hdr;
1766 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1767 unsigned char *sha1)
1769 unsigned char hdr[50];
1770 int hdrlen;
1771 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1772 return 0;
1775 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1777 int size;
1778 unsigned char *compressed;
1779 z_stream stream;
1780 unsigned char sha1[20];
1781 char *filename;
1782 static char tmpfile[PATH_MAX];
1783 unsigned char hdr[50];
1784 int fd, hdrlen;
1786 /* Normally if we have it in the pack then we do not bother writing
1787 * it out into .git/objects/??/?{38} file.
1789 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1790 filename = sha1_file_name(sha1);
1791 if (returnsha1)
1792 hashcpy(returnsha1, sha1);
1793 if (has_sha1_file(sha1))
1794 return 0;
1795 fd = open(filename, O_RDONLY);
1796 if (fd >= 0) {
1798 * FIXME!!! We might do collision checking here, but we'd
1799 * need to uncompress the old file and check it. Later.
1801 close(fd);
1802 return 0;
1805 if (errno != ENOENT) {
1806 return error("sha1 file %s: %s\n", filename, strerror(errno));
1809 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1811 fd = mkstemp(tmpfile);
1812 if (fd < 0) {
1813 if (errno == EPERM)
1814 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1815 else
1816 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1819 /* Set it up */
1820 memset(&stream, 0, sizeof(stream));
1821 deflateInit(&stream, zlib_compression_level);
1822 size = 8 + deflateBound(&stream, len+hdrlen);
1823 compressed = xmalloc(size);
1825 /* Compress it */
1826 stream.next_out = compressed;
1827 stream.avail_out = size;
1829 /* First header.. */
1830 stream.next_in = hdr;
1831 stream.avail_in = hdrlen;
1832 setup_object_header(&stream, type, len);
1834 /* Then the data itself.. */
1835 stream.next_in = buf;
1836 stream.avail_in = len;
1837 while (deflate(&stream, Z_FINISH) == Z_OK)
1838 /* nothing */;
1839 deflateEnd(&stream);
1840 size = stream.total_out;
1842 if (write_buffer(fd, compressed, size) < 0)
1843 die("unable to write sha1 file");
1844 fchmod(fd, 0444);
1845 close(fd);
1846 free(compressed);
1848 return move_temp_to_file(tmpfile, filename);
1852 * We need to unpack and recompress the object for writing
1853 * it out to a different file.
1855 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1857 size_t size;
1858 z_stream stream;
1859 unsigned char *unpacked;
1860 unsigned long len;
1861 char type[20];
1862 char hdr[50];
1863 int hdrlen;
1864 void *buf;
1866 /* need to unpack and recompress it by itself */
1867 unpacked = read_packed_sha1(sha1, type, &len);
1868 if (!unpacked)
1869 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1871 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1873 /* Set it up */
1874 memset(&stream, 0, sizeof(stream));
1875 deflateInit(&stream, zlib_compression_level);
1876 size = deflateBound(&stream, len + hdrlen);
1877 buf = xmalloc(size);
1879 /* Compress it */
1880 stream.next_out = buf;
1881 stream.avail_out = size;
1883 /* First header.. */
1884 stream.next_in = (void *)hdr;
1885 stream.avail_in = hdrlen;
1886 while (deflate(&stream, 0) == Z_OK)
1887 /* nothing */;
1889 /* Then the data itself.. */
1890 stream.next_in = unpacked;
1891 stream.avail_in = len;
1892 while (deflate(&stream, Z_FINISH) == Z_OK)
1893 /* nothing */;
1894 deflateEnd(&stream);
1895 free(unpacked);
1897 *objsize = stream.total_out;
1898 return buf;
1901 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1903 int retval;
1904 unsigned long objsize;
1905 void *buf = map_sha1_file(sha1, &objsize);
1907 if (buf) {
1908 retval = write_buffer(fd, buf, objsize);
1909 munmap(buf, objsize);
1910 return retval;
1913 buf = repack_object(sha1, &objsize);
1914 retval = write_buffer(fd, buf, objsize);
1915 free(buf);
1916 return retval;
1919 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1920 size_t bufsize, size_t *bufposn)
1922 char tmpfile[PATH_MAX];
1923 int local;
1924 z_stream stream;
1925 unsigned char real_sha1[20];
1926 unsigned char discard[4096];
1927 int ret;
1928 SHA_CTX c;
1930 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1932 local = mkstemp(tmpfile);
1933 if (local < 0) {
1934 if (errno == EPERM)
1935 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1936 else
1937 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1940 memset(&stream, 0, sizeof(stream));
1942 inflateInit(&stream);
1944 SHA1_Init(&c);
1946 do {
1947 ssize_t size;
1948 if (*bufposn) {
1949 stream.avail_in = *bufposn;
1950 stream.next_in = (unsigned char *) buffer;
1951 do {
1952 stream.next_out = discard;
1953 stream.avail_out = sizeof(discard);
1954 ret = inflate(&stream, Z_SYNC_FLUSH);
1955 SHA1_Update(&c, discard, sizeof(discard) -
1956 stream.avail_out);
1957 } while (stream.avail_in && ret == Z_OK);
1958 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1959 die("unable to write sha1 file");
1960 memmove(buffer, buffer + *bufposn - stream.avail_in,
1961 stream.avail_in);
1962 *bufposn = stream.avail_in;
1963 if (ret != Z_OK)
1964 break;
1966 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
1967 if (size <= 0) {
1968 close(local);
1969 unlink(tmpfile);
1970 if (!size)
1971 return error("Connection closed?");
1972 perror("Reading from connection");
1973 return -1;
1975 *bufposn += size;
1976 } while (1);
1977 inflateEnd(&stream);
1979 close(local);
1980 SHA1_Final(real_sha1, &c);
1981 if (ret != Z_STREAM_END) {
1982 unlink(tmpfile);
1983 return error("File %s corrupted", sha1_to_hex(sha1));
1985 if (hashcmp(sha1, real_sha1)) {
1986 unlink(tmpfile);
1987 return error("File %s has bad hash", sha1_to_hex(sha1));
1990 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1993 int has_pack_index(const unsigned char *sha1)
1995 struct stat st;
1996 if (stat(sha1_pack_index_name(sha1), &st))
1997 return 0;
1998 return 1;
2001 int has_pack_file(const unsigned char *sha1)
2003 struct stat st;
2004 if (stat(sha1_pack_name(sha1), &st))
2005 return 0;
2006 return 1;
2009 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2011 struct pack_entry e;
2012 return find_pack_entry(sha1, &e, ignore_packed);
2015 int has_sha1_file(const unsigned char *sha1)
2017 struct stat st;
2018 struct pack_entry e;
2020 if (find_pack_entry(sha1, &e, NULL))
2021 return 1;
2022 return find_sha1_file(sha1, &st) ? 1 : 0;
2026 * reads from fd as long as possible into a supplied buffer of size bytes.
2027 * If necessary the buffer's size is increased using realloc()
2029 * returns 0 if anything went fine and -1 otherwise
2031 * NOTE: both buf and size may change, but even when -1 is returned
2032 * you still have to free() it yourself.
2034 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
2036 char* buf = *return_buf;
2037 unsigned long size = *return_size;
2038 int iret;
2039 unsigned long off = 0;
2041 do {
2042 iret = xread(fd, buf + off, size - off);
2043 if (iret > 0) {
2044 off += iret;
2045 if (off == size) {
2046 size *= 2;
2047 buf = xrealloc(buf, size);
2050 } while (iret > 0);
2052 *return_buf = buf;
2053 *return_size = off;
2055 if (iret < 0)
2056 return -1;
2057 return 0;
2060 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2062 unsigned long size = 4096;
2063 char *buf = xmalloc(size);
2064 int ret;
2066 if (read_pipe(fd, &buf, &size)) {
2067 free(buf);
2068 return -1;
2071 if (!type)
2072 type = blob_type;
2073 if (write_object)
2074 ret = write_sha1_file(buf, size, type, sha1);
2075 else
2076 ret = hash_sha1_file(buf, size, type, sha1);
2077 free(buf);
2078 return ret;
2081 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
2083 unsigned long size = st->st_size;
2084 void *buf;
2085 int ret;
2087 buf = "";
2088 if (size)
2089 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2090 close(fd);
2092 if (!type)
2093 type = blob_type;
2094 if (write_object)
2095 ret = write_sha1_file(buf, size, type, sha1);
2096 else
2097 ret = hash_sha1_file(buf, size, type, sha1);
2098 if (size)
2099 munmap(buf, size);
2100 return ret;
2103 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2105 int fd;
2106 char *target;
2108 switch (st->st_mode & S_IFMT) {
2109 case S_IFREG:
2110 fd = open(path, O_RDONLY);
2111 if (fd < 0)
2112 return error("open(\"%s\"): %s", path,
2113 strerror(errno));
2114 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
2115 return error("%s: failed to insert into database",
2116 path);
2117 break;
2118 case S_IFLNK:
2119 target = xmalloc(st->st_size+1);
2120 if (readlink(path, target, st->st_size+1) != st->st_size) {
2121 char *errstr = strerror(errno);
2122 free(target);
2123 return error("readlink(\"%s\"): %s", path,
2124 errstr);
2126 if (!write_object)
2127 hash_sha1_file(target, st->st_size, blob_type, sha1);
2128 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
2129 return error("%s: failed to insert into database",
2130 path);
2131 free(target);
2132 break;
2133 default:
2134 return error("%s: unsupported file type", path);
2136 return 0;
2139 int read_pack_header(int fd, struct pack_header *header)
2141 char *c = (char*)header;
2142 ssize_t remaining = sizeof(struct pack_header);
2143 do {
2144 ssize_t r = xread(fd, c, remaining);
2145 if (r <= 0)
2146 /* "eof before pack header was fully read" */
2147 return PH_ERROR_EOF;
2148 remaining -= r;
2149 c += r;
2150 } while (remaining > 0);
2151 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2152 /* "protocol error (pack signature mismatch detected)" */
2153 return PH_ERROR_PACK_SIGNATURE;
2154 if (!pack_version_ok(header->hdr_version))
2155 /* "protocol error (pack version unsupported)" */
2156 return PH_ERROR_PROTOCOL;
2157 return 0;