![[tokkee]](http://tokkee.org/images/avatar.png){kind=avatar}
1 /*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
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 */
68 };
70 int get_sha1_hex(const char *hex, unsigned char *sha1)
71 {
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;
79 }
80 return 0;
81 }
83 int safe_create_leading_directories(char *path)
84 {
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;
101 }
102 }
103 else if (mkdir(path, 0777)) {
104 *pos = '/';
105 return -1;
106 }
107 else if (adjust_shared_perm(path)) {
108 *pos = '/';
109 return -2;
110 }
111 *pos++ = '/';
112 }
113 return 0;
114 }
116 char * sha1_to_hex(const unsigned char *sha1)
117 {
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];
128 }
129 *buf = '\0';
131 return buffer;
132 }
134 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
135 {
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];
143 }
144 }
146 /*
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.
150 *
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.
155 */
156 char *sha1_file_name(const unsigned char *sha1)
157 {
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;
169 }
170 fill_sha1_path(name, sha1);
171 return base;
172 }
174 char *sha1_pack_name(const unsigned char *sha1)
175 {
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;
186 }
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];
194 }
196 return base;
197 }
199 char *sha1_pack_index_name(const unsigned char *sha1)
200 {
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;
211 }
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];
219 }
221 return base;
222 }
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);
229 /*
230 * Prepare alternate object database registry.
231 *
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.
243 */
244 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
245 {
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;
261 }
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);
268 }
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;
283 }
285 /* Prevent the common mistake of listing the same
286 * thing twice, or object directory itself.
287 */
288 for (alt = alt_odb_list; alt; alt = alt->next) {
289 if (!memcmp(ent->base, alt->base, pfxlen)) {
290 free(ent);
291 return -1;
292 }
293 }
294 if (!memcmp(ent->base, objdir, pfxlen)) {
295 free(ent);
296 return -1;
297 }
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;
310 }
312 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
313 const char *relative_base, int depth)
314 {
315 const char *cp, *last;
317 if (depth > 5) {
318 error("%s: ignoring alternate object stores, nesting too deep.",
319 relative_base);
320 return;
321 }
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;
331 }
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);
341 }
342 }
343 while (cp < ep && *cp == sep)
344 cp++;
345 last = cp;
346 }
347 }
349 static void read_info_alternates(const char * relative_base, int depth)
350 {
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;
363 }
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);
370 }
372 void prepare_alt_odb(void)
373 {
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);
385 }
387 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
388 {
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;
400 }
401 return NULL;
402 }
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()
414 {
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);
432 }
434 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
435 void **idx_map_)
436 {
437 void *idx_map;
438 unsigned int *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;
448 }
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 too small");
460 nr = 0;
461 for (i = 0; i < 256; i++) {
462 unsigned int n = ntohl(index[i]);
463 if (n < nr)
464 return error("non-monotonic index");
465 nr = n;
466 }
468 /*
469 * Total size:
470 * - 256 index entries 4 bytes each
471 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
472 * - 20-byte SHA1 of the packfile
473 * - 20-byte SHA1 file checksum
474 */
475 if (idx_size != 4*256 + nr * 24 + 20 + 20)
476 return error("wrong index file size");
478 return 0;
479 }
481 static void scan_windows(struct packed_git *p,
482 struct packed_git **lru_p,
483 struct pack_window **lru_w,
484 struct pack_window **lru_l)
485 {
486 struct pack_window *w, *w_l;
488 for (w_l = NULL, w = p->windows; w; w = w->next) {
489 if (!w->inuse_cnt) {
490 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
491 *lru_p = p;
492 *lru_w = w;
493 *lru_l = w_l;
494 }
495 }
496 w_l = w;
497 }
498 }
500 static int unuse_one_window(struct packed_git *current)
501 {
502 struct packed_git *p, *lru_p = NULL;
503 struct pack_window *lru_w = NULL, *lru_l = NULL;
505 if (current)
506 scan_windows(current, &lru_p, &lru_w, &lru_l);
507 for (p = packed_git; p; p = p->next)
508 scan_windows(p, &lru_p, &lru_w, &lru_l);
509 if (lru_p) {
510 munmap(lru_w->base, lru_w->len);
511 pack_mapped -= lru_w->len;
512 if (lru_l)
513 lru_l->next = lru_w->next;
514 else {
515 lru_p->windows = lru_w->next;
516 if (!lru_p->windows && lru_p != current) {
517 close(lru_p->pack_fd);
518 lru_p->pack_fd = -1;
519 }
520 }
521 free(lru_w);
522 pack_open_windows--;
523 return 1;
524 }
525 return 0;
526 }
528 void release_pack_memory(size_t need)
529 {
530 size_t cur = pack_mapped;
531 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
532 ; /* nothing */
533 }
535 void unuse_pack(struct pack_window **w_cursor)
536 {
537 struct pack_window *w = *w_cursor;
538 if (w) {
539 w->inuse_cnt--;
540 *w_cursor = NULL;
541 }
542 }
544 static void open_packed_git(struct packed_git *p)
545 {
546 struct stat st;
547 struct pack_header hdr;
548 unsigned char sha1[20];
549 unsigned char *idx_sha1;
550 long fd_flag;
552 p->pack_fd = open(p->pack_name, O_RDONLY);
553 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
554 die("packfile %s cannot be opened", p->pack_name);
556 /* If we created the struct before we had the pack we lack size. */
557 if (!p->pack_size) {
558 if (!S_ISREG(st.st_mode))
559 die("packfile %s not a regular file", p->pack_name);
560 p->pack_size = st.st_size;
561 } else if (p->pack_size != st.st_size)
562 die("packfile %s size changed", p->pack_name);
564 /* We leave these file descriptors open with sliding mmap;
565 * there is no point keeping them open across exec(), though.
566 */
567 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
568 if (fd_flag < 0)
569 die("cannot determine file descriptor flags");
570 fd_flag |= FD_CLOEXEC;
571 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
572 die("cannot set FD_CLOEXEC");
574 /* Verify we recognize this pack file format. */
575 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
576 die("file %s is far too short to be a packfile", p->pack_name);
577 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
578 die("file %s is not a GIT packfile", p->pack_name);
579 if (!pack_version_ok(hdr.hdr_version))
580 die("packfile %s is version %u and not supported"
581 " (try upgrading GIT to a newer version)",
582 p->pack_name, ntohl(hdr.hdr_version));
584 /* Verify the pack matches its index. */
585 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
586 die("packfile %s claims to have %u objects"
587 " while index size indicates %u objects",
588 p->pack_name, ntohl(hdr.hdr_entries),
589 num_packed_objects(p));
590 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
591 die("end of packfile %s is unavailable", p->pack_name);
592 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
593 die("packfile %s signature is unavailable", p->pack_name);
594 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
595 if (hashcmp(sha1, idx_sha1))
596 die("packfile %s does not match index", p->pack_name);
597 }
599 static int in_window(struct pack_window *win, unsigned long offset)
600 {
601 /* We must promise at least 20 bytes (one hash) after the
602 * offset is available from this window, otherwise the offset
603 * is not actually in this window and a different window (which
604 * has that one hash excess) must be used. This is to support
605 * the object header and delta base parsing routines below.
606 */
607 off_t win_off = win->offset;
608 return win_off <= offset
609 && (offset + 20) <= (win_off + win->len);
610 }
612 unsigned char* use_pack(struct packed_git *p,
613 struct pack_window **w_cursor,
614 unsigned long offset,
615 unsigned int *left)
616 {
617 struct pack_window *win = *w_cursor;
619 if (p->pack_fd == -1)
620 open_packed_git(p);
622 /* Since packfiles end in a hash of their content and its
623 * pointless to ask for an offset into the middle of that
624 * hash, and the in_window function above wouldn't match
625 * don't allow an offset too close to the end of the file.
626 */
627 if (offset > (p->pack_size - 20))
628 die("offset beyond end of packfile (truncated pack?)");
630 if (!win || !in_window(win, offset)) {
631 if (win)
632 win->inuse_cnt--;
633 for (win = p->windows; win; win = win->next) {
634 if (in_window(win, offset))
635 break;
636 }
637 if (!win) {
638 if (!page_size)
639 page_size = getpagesize();
640 win = xcalloc(1, sizeof(*win));
641 win->offset = (offset / page_size) * page_size;
642 win->len = p->pack_size - win->offset;
643 if (win->len > packed_git_window_size)
644 win->len = packed_git_window_size;
645 pack_mapped += win->len;
646 while (packed_git_limit < pack_mapped
647 && unuse_one_window(p))
648 ; /* nothing */
649 win->base = xmmap(NULL, win->len,
650 PROT_READ, MAP_PRIVATE,
651 p->pack_fd, win->offset);
652 if (win->base == MAP_FAILED)
653 die("packfile %s cannot be mapped: %s",
654 p->pack_name,
655 strerror(errno));
656 pack_mmap_calls++;
657 pack_open_windows++;
658 if (pack_mapped > peak_pack_mapped)
659 peak_pack_mapped = pack_mapped;
660 if (pack_open_windows > peak_pack_open_windows)
661 peak_pack_open_windows = pack_open_windows;
662 win->next = p->windows;
663 p->windows = win;
664 }
665 }
666 if (win != *w_cursor) {
667 win->last_used = pack_used_ctr++;
668 win->inuse_cnt++;
669 *w_cursor = win;
670 }
671 offset -= win->offset;
672 if (left)
673 *left = win->len - offset;
674 return win->base + offset;
675 }
677 struct packed_git *add_packed_git(char *path, int path_len, int local)
678 {
679 struct stat st;
680 struct packed_git *p;
681 unsigned long idx_size;
682 void *idx_map;
683 unsigned char sha1[20];
685 if (check_packed_git_idx(path, &idx_size, &idx_map))
686 return NULL;
688 /* do we have a corresponding .pack file? */
689 strcpy(path + path_len - 4, ".pack");
690 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
691 munmap(idx_map, idx_size);
692 return NULL;
693 }
694 /* ok, it looks sane as far as we can check without
695 * actually mapping the pack file.
696 */
697 p = xmalloc(sizeof(*p) + path_len + 2);
698 strcpy(p->pack_name, path);
699 p->index_size = idx_size;
700 p->pack_size = st.st_size;
701 p->index_base = idx_map;
702 p->next = NULL;
703 p->windows = NULL;
704 p->pack_fd = -1;
705 p->pack_local = local;
706 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
707 hashcpy(p->sha1, sha1);
708 return p;
709 }
711 struct packed_git *parse_pack_index(unsigned char *sha1)
712 {
713 char *path = sha1_pack_index_name(sha1);
714 return parse_pack_index_file(sha1, path);
715 }
717 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
718 {
719 struct packed_git *p;
720 unsigned long idx_size;
721 void *idx_map;
722 char *path;
724 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
725 return NULL;
727 path = sha1_pack_name(sha1);
729 p = xmalloc(sizeof(*p) + strlen(path) + 2);
730 strcpy(p->pack_name, path);
731 p->index_size = idx_size;
732 p->pack_size = 0;
733 p->index_base = idx_map;
734 p->next = NULL;
735 p->windows = NULL;
736 p->pack_fd = -1;
737 hashcpy(p->sha1, sha1);
738 return p;
739 }
741 void install_packed_git(struct packed_git *pack)
742 {
743 pack->next = packed_git;
744 packed_git = pack;
745 }
747 static void prepare_packed_git_one(char *objdir, int local)
748 {
749 char path[PATH_MAX];
750 int len;
751 DIR *dir;
752 struct dirent *de;
754 sprintf(path, "%s/pack", objdir);
755 len = strlen(path);
756 dir = opendir(path);
757 if (!dir) {
758 if (errno != ENOENT)
759 error("unable to open object pack directory: %s: %s",
760 path, strerror(errno));
761 return;
762 }
763 path[len++] = '/';
764 while ((de = readdir(dir)) != NULL) {
765 int namelen = strlen(de->d_name);
766 struct packed_git *p;
768 if (!has_extension(de->d_name, ".idx"))
769 continue;
771 /* we have .idx. Is it a file we can map? */
772 strcpy(path + len, de->d_name);
773 for (p = packed_git; p; p = p->next) {
774 if (!memcmp(path, p->pack_name, len + namelen - 4))
775 break;
776 }
777 if (p)
778 continue;
779 p = add_packed_git(path, len + namelen, local);
780 if (!p)
781 continue;
782 p->next = packed_git;
783 packed_git = p;
784 }
785 closedir(dir);
786 }
788 static int prepare_packed_git_run_once = 0;
789 void prepare_packed_git(void)
790 {
791 struct alternate_object_database *alt;
793 if (prepare_packed_git_run_once)
794 return;
795 prepare_packed_git_one(get_object_directory(), 1);
796 prepare_alt_odb();
797 for (alt = alt_odb_list; alt; alt = alt->next) {
798 alt->name[-1] = 0;
799 prepare_packed_git_one(alt->base, 0);
800 alt->name[-1] = '/';
801 }
802 prepare_packed_git_run_once = 1;
803 }
805 void reprepare_packed_git(void)
806 {
807 prepare_packed_git_run_once = 0;
808 prepare_packed_git();
809 }
811 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
812 {
813 unsigned char real_sha1[20];
814 hash_sha1_file(map, size, type, real_sha1);
815 return hashcmp(sha1, real_sha1) ? -1 : 0;
816 }
818 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
819 {
820 struct stat st;
821 void *map;
822 int fd;
823 char *filename = find_sha1_file(sha1, &st);
825 if (!filename) {
826 return NULL;
827 }
829 fd = open(filename, O_RDONLY | sha1_file_open_flag);
830 if (fd < 0) {
831 /* See if it works without O_NOATIME */
832 switch (sha1_file_open_flag) {
833 default:
834 fd = open(filename, O_RDONLY);
835 if (fd >= 0)
836 break;
837 /* Fallthrough */
838 case 0:
839 return NULL;
840 }
842 /* If it failed once, it will probably fail again.
843 * Stop using O_NOATIME
844 */
845 sha1_file_open_flag = 0;
846 }
847 map = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
848 close(fd);
849 *size = st.st_size;
850 return map;
851 }
853 int legacy_loose_object(unsigned char *map)
854 {
855 unsigned int word;
857 /*
858 * Is it a zlib-compressed buffer? If so, the first byte
859 * must be 0x78 (15-bit window size, deflated), and the
860 * first 16-bit word is evenly divisible by 31
861 */
862 word = (map[0] << 8) + map[1];
863 if (map[0] == 0x78 && !(word % 31))
864 return 1;
865 else
866 return 0;
867 }
869 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
870 {
871 unsigned shift;
872 unsigned char c;
873 unsigned long size;
874 unsigned long used = 0;
876 c = buf[used++];
877 *type = (c >> 4) & 7;
878 size = c & 15;
879 shift = 4;
880 while (c & 0x80) {
881 if (len <= used)
882 return 0;
883 if (sizeof(long) * 8 <= shift)
884 return 0;
885 c = buf[used++];
886 size += (c & 0x7f) << shift;
887 shift += 7;
888 }
889 *sizep = size;
890 return used;
891 }
893 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
894 {
895 unsigned long size, used;
896 static const char valid_loose_object_type[8] = {
897 0, /* OBJ_EXT */
898 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
899 0, /* "delta" and others are invalid in a loose object */
900 };
901 enum object_type type;
903 /* Get the data stream */
904 memset(stream, 0, sizeof(*stream));
905 stream->next_in = map;
906 stream->avail_in = mapsize;
907 stream->next_out = buffer;
908 stream->avail_out = bufsiz;
910 if (legacy_loose_object(map)) {
911 inflateInit(stream);
912 return inflate(stream, 0);
913 }
915 used = unpack_object_header_gently(map, mapsize, &type, &size);
916 if (!used || !valid_loose_object_type[type])
917 return -1;
918 map += used;
919 mapsize -= used;
921 /* Set up the stream for the rest.. */
922 stream->next_in = map;
923 stream->avail_in = mapsize;
924 inflateInit(stream);
926 /* And generate the fake traditional header */
927 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
928 type_names[type], size);
929 return 0;
930 }
932 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
933 {
934 int bytes = strlen(buffer) + 1;
935 unsigned char *buf = xmalloc(1+size);
936 unsigned long n;
938 n = stream->total_out - bytes;
939 if (n > size)
940 n = size;
941 memcpy(buf, (char *) buffer + bytes, n);
942 bytes = n;
943 if (bytes < size) {
944 stream->next_out = buf + bytes;
945 stream->avail_out = size - bytes;
946 while (inflate(stream, Z_FINISH) == Z_OK)
947 /* nothing */;
948 }
949 buf[size] = 0;
950 inflateEnd(stream);
951 return buf;
952 }
954 /*
955 * We used to just use "sscanf()", but that's actually way
956 * too permissive for what we want to check. So do an anal
957 * object header parse by hand.
958 */
959 static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
960 {
961 int i;
962 unsigned long size;
964 /*
965 * The type can be at most ten bytes (including the
966 * terminating '\0' that we add), and is followed by
967 * a space.
968 */
969 i = 10;
970 for (;;) {
971 char c = *hdr++;
972 if (c == ' ')
973 break;
974 if (!--i)
975 return -1;
976 *type++ = c;
977 }
978 *type = 0;
980 /*
981 * The length must follow immediately, and be in canonical
982 * decimal format (ie "010" is not valid).
983 */
984 size = *hdr++ - '0';
985 if (size > 9)
986 return -1;
987 if (size) {
988 for (;;) {
989 unsigned long c = *hdr - '0';
990 if (c > 9)
991 break;
992 hdr++;
993 size = size * 10 + c;
994 }
995 }
996 *sizep = size;
998 /*
999 * The length must be followed by a zero byte
1000 */
1001 return *hdr ? -1 : 0;
1002 }
1004 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
1005 {
1006 int ret;
1007 z_stream stream;
1008 char hdr[8192];
1010 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1011 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
1012 return NULL;
1014 return unpack_sha1_rest(&stream, hdr, *size);
1015 }
1017 static unsigned long get_delta_base(struct packed_git *p,
1018 struct pack_window **w_curs,
1019 unsigned long offset,
1020 enum object_type kind,
1021 unsigned long delta_obj_offset,
1022 unsigned long *base_obj_offset)
1023 {
1024 unsigned char *base_info = use_pack(p, w_curs, offset, NULL);
1025 unsigned long base_offset;
1027 /* use_pack() assured us we have [base_info, base_info + 20)
1028 * as a range that we can look at without walking off the
1029 * end of the mapped window. Its actually the hash size
1030 * that is assured. An OFS_DELTA longer than the hash size
1031 * is stupid, as then a REF_DELTA would be smaller to store.
1032 */
1033 if (kind == OBJ_OFS_DELTA) {
1034 unsigned used = 0;
1035 unsigned char c = base_info[used++];
1036 base_offset = c & 127;
1037 while (c & 128) {
1038 base_offset += 1;
1039 if (!base_offset || base_offset & ~(~0UL >> 7))
1040 die("offset value overflow for delta base object");
1041 c = base_info[used++];
1042 base_offset = (base_offset << 7) + (c & 127);
1043 }
1044 base_offset = delta_obj_offset - base_offset;
1045 if (base_offset >= delta_obj_offset)
1046 die("delta base offset out of bound");
1047 offset += used;
1048 } else if (kind == OBJ_REF_DELTA) {
1049 /* The base entry _must_ be in the same pack */
1050 base_offset = find_pack_entry_one(base_info, p);
1051 if (!base_offset)
1052 die("failed to find delta-pack base object %s",
1053 sha1_to_hex(base_info));
1054 offset += 20;
1055 } else
1056 die("I am totally screwed");
1057 *base_obj_offset = base_offset;
1058 return offset;
1059 }
1061 /* forward declaration for a mutually recursive function */
1062 static int packed_object_info(struct packed_git *p, unsigned long offset,
1063 char *type, unsigned long *sizep);
1065 static int packed_delta_info(struct packed_git *p,
1066 struct pack_window **w_curs,
1067 unsigned long offset,
1068 enum object_type kind,
1069 unsigned long obj_offset,
1070 char *type,
1071 unsigned long *sizep)
1072 {
1073 unsigned long base_offset;
1075 offset = get_delta_base(p, w_curs, offset, kind,
1076 obj_offset, &base_offset);
1078 /* We choose to only get the type of the base object and
1079 * ignore potentially corrupt pack file that expects the delta
1080 * based on a base with a wrong size. This saves tons of
1081 * inflate() calls.
1082 */
1083 if (packed_object_info(p, base_offset, type, NULL))
1084 die("cannot get info for delta-pack base");
1086 if (sizep) {
1087 const unsigned char *data;
1088 unsigned char delta_head[20], *in;
1089 unsigned long result_size;
1090 z_stream stream;
1091 int st;
1093 memset(&stream, 0, sizeof(stream));
1094 stream.next_out = delta_head;
1095 stream.avail_out = sizeof(delta_head);
1097 inflateInit(&stream);
1098 do {
1099 in = use_pack(p, w_curs, offset, &stream.avail_in);
1100 stream.next_in = in;
1101 st = inflate(&stream, Z_FINISH);
1102 offset += stream.next_in - in;
1103 } while ((st == Z_OK || st == Z_BUF_ERROR)
1104 && stream.total_out < sizeof(delta_head));
1105 inflateEnd(&stream);
1106 if ((st != Z_STREAM_END) &&
1107 stream.total_out != sizeof(delta_head))
1108 die("delta data unpack-initial failed");
1110 /* Examine the initial part of the delta to figure out
1111 * the result size.
1112 */
1113 data = delta_head;
1115 /* ignore base size */
1116 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1118 /* Read the result size */
1119 result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1120 *sizep = result_size;
1121 }
1122 return 0;
1123 }
1125 static unsigned long unpack_object_header(struct packed_git *p,
1126 struct pack_window **w_curs,
1127 unsigned long offset,
1128 enum object_type *type,
1129 unsigned long *sizep)
1130 {
1131 unsigned char *base;
1132 unsigned int left;
1133 unsigned long used;
1135 /* use_pack() assures us we have [base, base + 20) available
1136 * as a range that we can look at at. (Its actually the hash
1137 * size that is assurred.) With our object header encoding
1138 * the maximum deflated object size is 2^137, which is just
1139 * insane, so we know won't exceed what we have been given.
1140 */
1141 base = use_pack(p, w_curs, offset, &left);
1142 used = unpack_object_header_gently(base, left, type, sizep);
1143 if (!used)
1144 die("object offset outside of pack file");
1146 return offset + used;
1147 }
1149 void packed_object_info_detail(struct packed_git *p,
1150 unsigned long offset,
1151 char *type,
1152 unsigned long *size,
1153 unsigned long *store_size,
1154 unsigned int *delta_chain_length,
1155 unsigned char *base_sha1)
1156 {
1157 struct pack_window *w_curs = NULL;
1158 unsigned long obj_offset, val;
1159 unsigned char *next_sha1;
1160 enum object_type kind;
1162 *delta_chain_length = 0;
1163 obj_offset = offset;
1164 offset = unpack_object_header(p, &w_curs, offset, &kind, size);
1166 for (;;) {
1167 switch (kind) {
1168 default:
1169 die("pack %s contains unknown object type %d",
1170 p->pack_name, kind);
1171 case OBJ_COMMIT:
1172 case OBJ_TREE:
1173 case OBJ_BLOB:
1174 case OBJ_TAG:
1175 strcpy(type, type_names[kind]);
1176 *store_size = 0; /* notyet */
1177 unuse_pack(&w_curs);
1178 return;
1179 case OBJ_OFS_DELTA:
1180 get_delta_base(p, &w_curs, offset, kind,
1181 obj_offset, &offset);
1182 if (*delta_chain_length == 0) {
1183 /* TODO: find base_sha1 as pointed by offset */
1184 }
1185 break;
1186 case OBJ_REF_DELTA:
1187 next_sha1 = use_pack(p, &w_curs, offset, NULL);
1188 if (*delta_chain_length == 0)
1189 hashcpy(base_sha1, next_sha1);
1190 offset = find_pack_entry_one(next_sha1, p);
1191 break;
1192 }
1193 obj_offset = offset;
1194 offset = unpack_object_header(p, &w_curs, offset, &kind, &val);
1195 (*delta_chain_length)++;
1196 }
1197 }
1199 static int packed_object_info(struct packed_git *p, unsigned long offset,
1200 char *type, unsigned long *sizep)
1201 {
1202 struct pack_window *w_curs = NULL;
1203 unsigned long size, obj_offset = offset;
1204 enum object_type kind;
1205 int r;
1207 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1209 switch (kind) {
1210 case OBJ_OFS_DELTA:
1211 case OBJ_REF_DELTA:
1212 r = packed_delta_info(p, &w_curs, offset, kind,
1213 obj_offset, type, sizep);
1214 unuse_pack(&w_curs);
1215 return r;
1216 case OBJ_COMMIT:
1217 case OBJ_TREE:
1218 case OBJ_BLOB:
1219 case OBJ_TAG:
1220 strcpy(type, type_names[kind]);
1221 unuse_pack(&w_curs);
1222 break;
1223 default:
1224 die("pack %s contains unknown object type %d",
1225 p->pack_name, kind);
1226 }
1227 if (sizep)
1228 *sizep = size;
1229 return 0;
1230 }
1232 static void *unpack_compressed_entry(struct packed_git *p,
1233 struct pack_window **w_curs,
1234 unsigned long offset,
1235 unsigned long size)
1236 {
1237 int st;
1238 z_stream stream;
1239 unsigned char *buffer, *in;
1241 buffer = xmalloc(size + 1);
1242 buffer[size] = 0;
1243 memset(&stream, 0, sizeof(stream));
1244 stream.next_out = buffer;
1245 stream.avail_out = size;
1247 inflateInit(&stream);
1248 do {
1249 in = use_pack(p, w_curs, offset, &stream.avail_in);
1250 stream.next_in = in;
1251 st = inflate(&stream, Z_FINISH);
1252 offset += stream.next_in - in;
1253 } while (st == Z_OK || st == Z_BUF_ERROR);
1254 inflateEnd(&stream);
1255 if ((st != Z_STREAM_END) || stream.total_out != size) {
1256 free(buffer);
1257 return NULL;
1258 }
1260 return buffer;
1261 }
1263 static void *unpack_delta_entry(struct packed_git *p,
1264 struct pack_window **w_curs,
1265 unsigned long offset,
1266 unsigned long delta_size,
1267 enum object_type kind,
1268 unsigned long obj_offset,
1269 char *type,
1270 unsigned long *sizep)
1271 {
1272 void *delta_data, *result, *base;
1273 unsigned long result_size, base_size, base_offset;
1275 offset = get_delta_base(p, w_curs, offset, kind,
1276 obj_offset, &base_offset);
1277 base = unpack_entry(p, base_offset, type, &base_size);
1278 if (!base)
1279 die("failed to read delta base object at %lu from %s",
1280 base_offset, p->pack_name);
1282 delta_data = unpack_compressed_entry(p, w_curs, offset, delta_size);
1283 result = patch_delta(base, base_size,
1284 delta_data, delta_size,
1285 &result_size);
1286 if (!result)
1287 die("failed to apply delta");
1288 free(delta_data);
1289 free(base);
1290 *sizep = result_size;
1291 return result;
1292 }
1294 void *unpack_entry(struct packed_git *p, unsigned long offset,
1295 char *type, unsigned long *sizep)
1296 {
1297 struct pack_window *w_curs = NULL;
1298 unsigned long size, obj_offset = offset;
1299 enum object_type kind;
1300 void *retval;
1302 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1303 switch (kind) {
1304 case OBJ_OFS_DELTA:
1305 case OBJ_REF_DELTA:
1306 retval = unpack_delta_entry(p, &w_curs, offset, size,
1307 kind, obj_offset, type, sizep);
1308 break;
1309 case OBJ_COMMIT:
1310 case OBJ_TREE:
1311 case OBJ_BLOB:
1312 case OBJ_TAG:
1313 strcpy(type, type_names[kind]);
1314 *sizep = size;
1315 retval = unpack_compressed_entry(p, &w_curs, offset, size);
1316 break;
1317 default:
1318 die("unknown object type %i in %s", kind, p->pack_name);
1319 }
1320 unuse_pack(&w_curs);
1321 return retval;
1322 }
1324 int num_packed_objects(const struct packed_git *p)
1325 {
1326 /* See check_packed_git_idx() */
1327 return (p->index_size - 20 - 20 - 4*256) / 24;
1328 }
1330 int nth_packed_object_sha1(const struct packed_git *p, int n,
1331 unsigned char* sha1)
1332 {
1333 void *index = p->index_base + 256;
1334 if (n < 0 || num_packed_objects(p) <= n)
1335 return -1;
1336 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1337 return 0;
1338 }
1340 unsigned long find_pack_entry_one(const unsigned char *sha1,
1341 struct packed_git *p)
1342 {
1343 unsigned int *level1_ofs = p->index_base;
1344 int hi = ntohl(level1_ofs[*sha1]);
1345 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1346 void *index = p->index_base + 256;
1348 do {
1349 int mi = (lo + hi) / 2;
1350 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1351 if (!cmp)
1352 return ntohl(*((unsigned int *) ((char *) index + (24 * mi))));
1353 if (cmp > 0)
1354 hi = mi;
1355 else
1356 lo = mi+1;
1357 } while (lo < hi);
1358 return 0;
1359 }
1361 static int matches_pack_name(struct packed_git *p, const char *ig)
1362 {
1363 const char *last_c, *c;
1365 if (!strcmp(p->pack_name, ig))
1366 return 0;
1368 for (c = p->pack_name, last_c = c; *c;)
1369 if (*c == '/')
1370 last_c = ++c;
1371 else
1372 ++c;
1373 if (!strcmp(last_c, ig))
1374 return 0;
1376 return 1;
1377 }
1379 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1380 {
1381 struct packed_git *p;
1382 unsigned long offset;
1384 prepare_packed_git();
1386 for (p = packed_git; p; p = p->next) {
1387 if (ignore_packed) {
1388 const char **ig;
1389 for (ig = ignore_packed; *ig; ig++)
1390 if (!matches_pack_name(p, *ig))
1391 break;
1392 if (*ig)
1393 continue;
1394 }
1395 offset = find_pack_entry_one(sha1, p);
1396 if (offset) {
1397 e->offset = offset;
1398 e->p = p;
1399 hashcpy(e->sha1, sha1);
1400 return 1;
1401 }
1402 }
1403 return 0;
1404 }
1406 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1407 struct packed_git *packs)
1408 {
1409 struct packed_git *p;
1411 for (p = packs; p; p = p->next) {
1412 if (find_pack_entry_one(sha1, p))
1413 return p;
1414 }
1415 return NULL;
1417 }
1419 static int sha1_loose_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1420 {
1421 int status;
1422 unsigned long mapsize, size;
1423 void *map;
1424 z_stream stream;
1425 char hdr[128];
1427 map = map_sha1_file(sha1, &mapsize);
1428 if (!map)
1429 return error("unable to find %s", sha1_to_hex(sha1));
1430 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1431 status = error("unable to unpack %s header",
1432 sha1_to_hex(sha1));
1433 if (parse_sha1_header(hdr, type, &size) < 0)
1434 status = error("unable to parse %s header", sha1_to_hex(sha1));
1435 else {
1436 status = 0;
1437 if (sizep)
1438 *sizep = size;
1439 }
1440 inflateEnd(&stream);
1441 munmap(map, mapsize);
1442 return status;
1443 }
1445 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1446 {
1447 struct pack_entry e;
1449 if (!find_pack_entry(sha1, &e, NULL)) {
1450 reprepare_packed_git();
1451 if (!find_pack_entry(sha1, &e, NULL))
1452 return sha1_loose_object_info(sha1, type, sizep);
1453 }
1454 return packed_object_info(e.p, e.offset, type, sizep);
1455 }
1457 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1458 {
1459 struct pack_entry e;
1461 if (!find_pack_entry(sha1, &e, NULL)) {
1462 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1463 return NULL;
1464 }
1465 return unpack_entry(e.p, e.offset, type, size);
1466 }
1468 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1469 {
1470 unsigned long mapsize;
1471 void *map, *buf;
1472 struct pack_entry e;
1474 if (find_pack_entry(sha1, &e, NULL))
1475 return read_packed_sha1(sha1, type, size);
1476 map = map_sha1_file(sha1, &mapsize);
1477 if (map) {
1478 buf = unpack_sha1_file(map, mapsize, type, size);
1479 munmap(map, mapsize);
1480 return buf;
1481 }
1482 reprepare_packed_git();
1483 if (find_pack_entry(sha1, &e, NULL))
1484 return read_packed_sha1(sha1, type, size);
1485 return NULL;
1486 }
1488 void *read_object_with_reference(const unsigned char *sha1,
1489 const char *required_type,
1490 unsigned long *size,
1491 unsigned char *actual_sha1_return)
1492 {
1493 char type[20];
1494 void *buffer;
1495 unsigned long isize;
1496 unsigned char actual_sha1[20];
1498 hashcpy(actual_sha1, sha1);
1499 while (1) {
1500 int ref_length = -1;
1501 const char *ref_type = NULL;
1503 buffer = read_sha1_file(actual_sha1, type, &isize);
1504 if (!buffer)
1505 return NULL;
1506 if (!strcmp(type, required_type)) {
1507 *size = isize;
1508 if (actual_sha1_return)
1509 hashcpy(actual_sha1_return, actual_sha1);
1510 return buffer;
1511 }
1512 /* Handle references */
1513 else if (!strcmp(type, commit_type))
1514 ref_type = "tree ";
1515 else if (!strcmp(type, tag_type))
1516 ref_type = "object ";
1517 else {
1518 free(buffer);
1519 return NULL;
1520 }
1521 ref_length = strlen(ref_type);
1523 if (memcmp(buffer, ref_type, ref_length) ||
1524 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1525 free(buffer);
1526 return NULL;
1527 }
1528 free(buffer);
1529 /* Now we have the ID of the referred-to object in
1530 * actual_sha1. Check again. */
1531 }
1532 }
1534 static void write_sha1_file_prepare(void *buf, unsigned long len,
1535 const char *type, unsigned char *sha1,
1536 unsigned char *hdr, int *hdrlen)
1537 {
1538 SHA_CTX c;
1540 /* Generate the header */
1541 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1543 /* Sha1.. */
1544 SHA1_Init(&c);
1545 SHA1_Update(&c, hdr, *hdrlen);
1546 SHA1_Update(&c, buf, len);
1547 SHA1_Final(sha1, &c);
1548 }
1550 /*
1551 * Link the tempfile to the final place, possibly creating the
1552 * last directory level as you do so.
1553 *
1554 * Returns the errno on failure, 0 on success.
1555 */
1556 static int link_temp_to_file(const char *tmpfile, const char *filename)
1557 {
1558 int ret;
1559 char *dir;
1561 if (!link(tmpfile, filename))
1562 return 0;
1564 /*
1565 * Try to mkdir the last path component if that failed.
1566 *
1567 * Re-try the "link()" regardless of whether the mkdir
1568 * succeeds, since a race might mean that somebody
1569 * else succeeded.
1570 */
1571 ret = errno;
1572 dir = strrchr(filename, '/');
1573 if (dir) {
1574 *dir = 0;
1575 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1576 *dir = '/';
1577 return -2;
1578 }
1579 *dir = '/';
1580 if (!link(tmpfile, filename))
1581 return 0;
1582 ret = errno;
1583 }
1584 return ret;
1585 }
1587 /*
1588 * Move the just written object into its final resting place
1589 */
1590 int move_temp_to_file(const char *tmpfile, const char *filename)
1591 {
1592 int ret = link_temp_to_file(tmpfile, filename);
1594 /*
1595 * Coda hack - coda doesn't like cross-directory links,
1596 * so we fall back to a rename, which will mean that it
1597 * won't be able to check collisions, but that's not a
1598 * big deal.
1599 *
1600 * The same holds for FAT formatted media.
1601 *
1602 * When this succeeds, we just return 0. We have nothing
1603 * left to unlink.
1604 */
1605 if (ret && ret != EEXIST) {
1606 if (!rename(tmpfile, filename))
1607 return 0;
1608 ret = errno;
1609 }
1610 unlink(tmpfile);
1611 if (ret) {
1612 if (ret != EEXIST) {
1613 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1614 }
1615 /* FIXME!!! Collision check here ? */
1616 }
1618 return 0;
1619 }
1621 static int write_buffer(int fd, const void *buf, size_t len)
1622 {
1623 if (write_in_full(fd, buf, len) < 0)
1624 return error("file write error (%s)", strerror(errno));
1625 return 0;
1626 }
1628 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1629 {
1630 int hdr_len;
1631 unsigned char c;
1633 c = (type << 4) | (len & 15);
1634 len >>= 4;
1635 hdr_len = 1;
1636 while (len) {
1637 *hdr++ = c | 0x80;
1638 hdr_len++;
1639 c = (len & 0x7f);
1640 len >>= 7;
1641 }
1642 *hdr = c;
1643 return hdr_len;
1644 }
1646 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1647 {
1648 int obj_type, hdr;
1650 if (use_legacy_headers) {
1651 while (deflate(stream, 0) == Z_OK)
1652 /* nothing */;
1653 return;
1654 }
1655 if (!strcmp(type, blob_type))
1656 obj_type = OBJ_BLOB;
1657 else if (!strcmp(type, tree_type))
1658 obj_type = OBJ_TREE;
1659 else if (!strcmp(type, commit_type))
1660 obj_type = OBJ_COMMIT;
1661 else if (!strcmp(type, tag_type))
1662 obj_type = OBJ_TAG;
1663 else
1664 die("trying to generate bogus object of type '%s'", type);
1665 hdr = write_binary_header(stream->next_out, obj_type, len);
1666 stream->total_out = hdr;
1667 stream->next_out += hdr;
1668 stream->avail_out -= hdr;
1669 }
1671 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1672 unsigned char *sha1)
1673 {
1674 unsigned char hdr[50];
1675 int hdrlen;
1676 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1677 return 0;
1678 }
1680 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1681 {
1682 int size;
1683 unsigned char *compressed;
1684 z_stream stream;
1685 unsigned char sha1[20];
1686 char *filename;
1687 static char tmpfile[PATH_MAX];
1688 unsigned char hdr[50];
1689 int fd, hdrlen;
1691 /* Normally if we have it in the pack then we do not bother writing
1692 * it out into .git/objects/??/?{38} file.
1693 */
1694 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1695 filename = sha1_file_name(sha1);
1696 if (returnsha1)
1697 hashcpy(returnsha1, sha1);
1698 if (has_sha1_file(sha1))
1699 return 0;
1700 fd = open(filename, O_RDONLY);
1701 if (fd >= 0) {
1702 /*
1703 * FIXME!!! We might do collision checking here, but we'd
1704 * need to uncompress the old file and check it. Later.
1705 */
1706 close(fd);
1707 return 0;
1708 }
1710 if (errno != ENOENT) {
1711 return error("sha1 file %s: %s\n", filename, strerror(errno));
1712 }
1714 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1716 fd = mkstemp(tmpfile);
1717 if (fd < 0) {
1718 if (errno == EPERM)
1719 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1720 else
1721 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1722 }
1724 /* Set it up */
1725 memset(&stream, 0, sizeof(stream));
1726 deflateInit(&stream, zlib_compression_level);
1727 size = 8 + deflateBound(&stream, len+hdrlen);
1728 compressed = xmalloc(size);
1730 /* Compress it */
1731 stream.next_out = compressed;
1732 stream.avail_out = size;
1734 /* First header.. */
1735 stream.next_in = hdr;
1736 stream.avail_in = hdrlen;
1737 setup_object_header(&stream, type, len);
1739 /* Then the data itself.. */
1740 stream.next_in = buf;
1741 stream.avail_in = len;
1742 while (deflate(&stream, Z_FINISH) == Z_OK)
1743 /* nothing */;
1744 deflateEnd(&stream);
1745 size = stream.total_out;
1747 if (write_buffer(fd, compressed, size) < 0)
1748 die("unable to write sha1 file");
1749 fchmod(fd, 0444);
1750 close(fd);
1751 free(compressed);
1753 return move_temp_to_file(tmpfile, filename);
1754 }
1756 /*
1757 * We need to unpack and recompress the object for writing
1758 * it out to a different file.
1759 */
1760 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1761 {
1762 size_t size;
1763 z_stream stream;
1764 unsigned char *unpacked;
1765 unsigned long len;
1766 char type[20];
1767 char hdr[50];
1768 int hdrlen;
1769 void *buf;
1771 /* need to unpack and recompress it by itself */
1772 unpacked = read_packed_sha1(sha1, type, &len);
1774 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1776 /* Set it up */
1777 memset(&stream, 0, sizeof(stream));
1778 deflateInit(&stream, zlib_compression_level);
1779 size = deflateBound(&stream, len + hdrlen);
1780 buf = xmalloc(size);
1782 /* Compress it */
1783 stream.next_out = buf;
1784 stream.avail_out = size;
1786 /* First header.. */
1787 stream.next_in = (void *)hdr;
1788 stream.avail_in = hdrlen;
1789 while (deflate(&stream, 0) == Z_OK)
1790 /* nothing */;
1792 /* Then the data itself.. */
1793 stream.next_in = unpacked;
1794 stream.avail_in = len;
1795 while (deflate(&stream, Z_FINISH) == Z_OK)
1796 /* nothing */;
1797 deflateEnd(&stream);
1798 free(unpacked);
1800 *objsize = stream.total_out;
1801 return buf;
1802 }
1804 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1805 {
1806 int retval;
1807 unsigned long objsize;
1808 void *buf = map_sha1_file(sha1, &objsize);
1810 if (buf) {
1811 retval = write_buffer(fd, buf, objsize);
1812 munmap(buf, objsize);
1813 return retval;
1814 }
1816 buf = repack_object(sha1, &objsize);
1817 retval = write_buffer(fd, buf, objsize);
1818 free(buf);
1819 return retval;
1820 }
1822 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1823 size_t bufsize, size_t *bufposn)
1824 {
1825 char tmpfile[PATH_MAX];
1826 int local;
1827 z_stream stream;
1828 unsigned char real_sha1[20];
1829 unsigned char discard[4096];
1830 int ret;
1831 SHA_CTX c;
1833 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1835 local = mkstemp(tmpfile);
1836 if (local < 0) {
1837 if (errno == EPERM)
1838 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1839 else
1840 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1841 }
1843 memset(&stream, 0, sizeof(stream));
1845 inflateInit(&stream);
1847 SHA1_Init(&c);
1849 do {
1850 ssize_t size;
1851 if (*bufposn) {
1852 stream.avail_in = *bufposn;
1853 stream.next_in = (unsigned char *) buffer;
1854 do {
1855 stream.next_out = discard;
1856 stream.avail_out = sizeof(discard);
1857 ret = inflate(&stream, Z_SYNC_FLUSH);
1858 SHA1_Update(&c, discard, sizeof(discard) -
1859 stream.avail_out);
1860 } while (stream.avail_in && ret == Z_OK);
1861 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1862 die("unable to write sha1 file");
1863 memmove(buffer, buffer + *bufposn - stream.avail_in,
1864 stream.avail_in);
1865 *bufposn = stream.avail_in;
1866 if (ret != Z_OK)
1867 break;
1868 }
1869 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
1870 if (size <= 0) {
1871 close(local);
1872 unlink(tmpfile);
1873 if (!size)
1874 return error("Connection closed?");
1875 perror("Reading from connection");
1876 return -1;
1877 }
1878 *bufposn += size;
1879 } while (1);
1880 inflateEnd(&stream);
1882 close(local);
1883 SHA1_Final(real_sha1, &c);
1884 if (ret != Z_STREAM_END) {
1885 unlink(tmpfile);
1886 return error("File %s corrupted", sha1_to_hex(sha1));
1887 }
1888 if (hashcmp(sha1, real_sha1)) {
1889 unlink(tmpfile);
1890 return error("File %s has bad hash", sha1_to_hex(sha1));
1891 }
1893 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1894 }
1896 int has_pack_index(const unsigned char *sha1)
1897 {
1898 struct stat st;
1899 if (stat(sha1_pack_index_name(sha1), &st))
1900 return 0;
1901 return 1;
1902 }
1904 int has_pack_file(const unsigned char *sha1)
1905 {
1906 struct stat st;
1907 if (stat(sha1_pack_name(sha1), &st))
1908 return 0;
1909 return 1;
1910 }
1912 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1913 {
1914 struct pack_entry e;
1915 return find_pack_entry(sha1, &e, ignore_packed);
1916 }
1918 int has_sha1_file(const unsigned char *sha1)
1919 {
1920 struct stat st;
1921 struct pack_entry e;
1923 if (find_pack_entry(sha1, &e, NULL))
1924 return 1;
1925 return find_sha1_file(sha1, &st) ? 1 : 0;
1926 }
1928 /*
1929 * reads from fd as long as possible into a supplied buffer of size bytes.
1930 * If necessary the buffer's size is increased using realloc()
1931 *
1932 * returns 0 if anything went fine and -1 otherwise
1933 *
1934 * NOTE: both buf and size may change, but even when -1 is returned
1935 * you still have to free() it yourself.
1936 */
1937 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
1938 {
1939 char* buf = *return_buf;
1940 unsigned long size = *return_size;
1941 int iret;
1942 unsigned long off = 0;
1944 do {
1945 iret = xread(fd, buf + off, size - off);
1946 if (iret > 0) {
1947 off += iret;
1948 if (off == size) {
1949 size *= 2;
1950 buf = xrealloc(buf, size);
1951 }
1952 }
1953 } while (iret > 0);
1955 *return_buf = buf;
1956 *return_size = off;
1958 if (iret < 0)
1959 return -1;
1960 return 0;
1961 }
1963 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
1964 {
1965 unsigned long size = 4096;
1966 char *buf = xmalloc(size);
1967 int ret;
1969 if (read_pipe(fd, &buf, &size)) {
1970 free(buf);
1971 return -1;
1972 }
1974 if (!type)
1975 type = blob_type;
1976 if (write_object)
1977 ret = write_sha1_file(buf, size, type, sha1);
1978 else
1979 ret = hash_sha1_file(buf, size, type, sha1);
1980 free(buf);
1981 return ret;
1982 }
1984 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1985 {
1986 unsigned long size = st->st_size;
1987 void *buf;
1988 int ret;
1990 buf = "";
1991 if (size)
1992 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1993 close(fd);
1995 if (!type)
1996 type = blob_type;
1997 if (write_object)
1998 ret = write_sha1_file(buf, size, type, sha1);
1999 else
2000 ret = hash_sha1_file(buf, size, type, sha1);
2001 if (size)
2002 munmap(buf, size);
2003 return ret;
2004 }
2006 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2007 {
2008 int fd;
2009 char *target;
2011 switch (st->st_mode & S_IFMT) {
2012 case S_IFREG:
2013 fd = open(path, O_RDONLY);
2014 if (fd < 0)
2015 return error("open(\"%s\"): %s", path,
2016 strerror(errno));
2017 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
2018 return error("%s: failed to insert into database",
2019 path);
2020 break;
2021 case S_IFLNK:
2022 target = xmalloc(st->st_size+1);
2023 if (readlink(path, target, st->st_size+1) != st->st_size) {
2024 char *errstr = strerror(errno);
2025 free(target);
2026 return error("readlink(\"%s\"): %s", path,
2027 errstr);
2028 }
2029 if (!write_object)
2030 hash_sha1_file(target, st->st_size, blob_type, sha1);
2031 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
2032 return error("%s: failed to insert into database",
2033 path);
2034 free(target);
2035 break;
2036 default:
2037 return error("%s: unsupported file type", path);
2038 }
2039 return 0;
2040 }