![[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 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;
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 %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.
464 */
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;
477 }
479 /*
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
485 */
486 if (idx_size != 4*256 + nr * 24 + 20 + 20)
487 return error("wrong index file size in %s", path);
489 return 0;
490 }
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)
496 {
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;
505 }
506 }
507 w_l = w;
508 }
509 }
511 static int unuse_one_window(struct packed_git *current)
512 {
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;
530 }
531 }
532 free(lru_w);
533 pack_open_windows--;
534 return 1;
535 }
536 return 0;
537 }
539 void release_pack_memory(size_t need)
540 {
541 size_t cur = pack_mapped;
542 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
543 ; /* nothing */
544 }
546 void unuse_pack(struct pack_window **w_cursor)
547 {
548 struct pack_window *w = *w_cursor;
549 if (w) {
550 w->inuse_cnt--;
551 *w_cursor = NULL;
552 }
553 }
555 static void open_packed_git(struct packed_git *p)
556 {
557 struct stat st;
558 struct pack_header hdr;
559 unsigned char sha1[20];
560 unsigned char *idx_sha1;
561 long fd_flag;
563 p->pack_fd = open(p->pack_name, O_RDONLY);
564 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
565 die("packfile %s cannot be opened", p->pack_name);
567 /* If we created the struct before we had the pack we lack size. */
568 if (!p->pack_size) {
569 if (!S_ISREG(st.st_mode))
570 die("packfile %s not a regular file", p->pack_name);
571 p->pack_size = st.st_size;
572 } else if (p->pack_size != st.st_size)
573 die("packfile %s size changed", p->pack_name);
575 /* We leave these file descriptors open with sliding mmap;
576 * there is no point keeping them open across exec(), though.
577 */
578 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
579 if (fd_flag < 0)
580 die("cannot determine file descriptor flags");
581 fd_flag |= FD_CLOEXEC;
582 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
583 die("cannot set FD_CLOEXEC");
585 /* Verify we recognize this pack file format. */
586 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
587 die("file %s is far too short to be a packfile", p->pack_name);
588 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
589 die("file %s is not a GIT packfile", p->pack_name);
590 if (!pack_version_ok(hdr.hdr_version))
591 die("packfile %s is version %u and not supported"
592 " (try upgrading GIT to a newer version)",
593 p->pack_name, ntohl(hdr.hdr_version));
595 /* Verify the pack matches its index. */
596 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
597 die("packfile %s claims to have %u objects"
598 " while index size indicates %u objects",
599 p->pack_name, ntohl(hdr.hdr_entries),
600 num_packed_objects(p));
601 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
602 die("end of packfile %s is unavailable", p->pack_name);
603 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
604 die("packfile %s signature is unavailable", p->pack_name);
605 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
606 if (hashcmp(sha1, idx_sha1))
607 die("packfile %s does not match index", p->pack_name);
608 }
610 static int in_window(struct pack_window *win, unsigned long offset)
611 {
612 /* We must promise at least 20 bytes (one hash) after the
613 * offset is available from this window, otherwise the offset
614 * is not actually in this window and a different window (which
615 * has that one hash excess) must be used. This is to support
616 * the object header and delta base parsing routines below.
617 */
618 off_t win_off = win->offset;
619 return win_off <= offset
620 && (offset + 20) <= (win_off + win->len);
621 }
623 unsigned char* use_pack(struct packed_git *p,
624 struct pack_window **w_cursor,
625 unsigned long offset,
626 unsigned int *left)
627 {
628 struct pack_window *win = *w_cursor;
630 if (p->pack_fd == -1)
631 open_packed_git(p);
633 /* Since packfiles end in a hash of their content and its
634 * pointless to ask for an offset into the middle of that
635 * hash, and the in_window function above wouldn't match
636 * don't allow an offset too close to the end of the file.
637 */
638 if (offset > (p->pack_size - 20))
639 die("offset beyond end of packfile (truncated pack?)");
641 if (!win || !in_window(win, offset)) {
642 if (win)
643 win->inuse_cnt--;
644 for (win = p->windows; win; win = win->next) {
645 if (in_window(win, offset))
646 break;
647 }
648 if (!win) {
649 if (!page_size)
650 page_size = getpagesize();
651 win = xcalloc(1, sizeof(*win));
652 win->offset = (offset / page_size) * page_size;
653 win->len = p->pack_size - win->offset;
654 if (win->len > packed_git_window_size)
655 win->len = packed_git_window_size;
656 pack_mapped += win->len;
657 while (packed_git_limit < pack_mapped
658 && unuse_one_window(p))
659 ; /* nothing */
660 win->base = xmmap(NULL, win->len,
661 PROT_READ, MAP_PRIVATE,
662 p->pack_fd, win->offset);
663 if (win->base == MAP_FAILED)
664 die("packfile %s cannot be mapped: %s",
665 p->pack_name,
666 strerror(errno));
667 pack_mmap_calls++;
668 pack_open_windows++;
669 if (pack_mapped > peak_pack_mapped)
670 peak_pack_mapped = pack_mapped;
671 if (pack_open_windows > peak_pack_open_windows)
672 peak_pack_open_windows = pack_open_windows;
673 win->next = p->windows;
674 p->windows = win;
675 }
676 }
677 if (win != *w_cursor) {
678 win->last_used = pack_used_ctr++;
679 win->inuse_cnt++;
680 *w_cursor = win;
681 }
682 offset -= win->offset;
683 if (left)
684 *left = win->len - offset;
685 return win->base + offset;
686 }
688 struct packed_git *add_packed_git(char *path, int path_len, int local)
689 {
690 struct stat st;
691 struct packed_git *p;
692 unsigned long idx_size;
693 void *idx_map;
694 unsigned char sha1[20];
696 if (check_packed_git_idx(path, &idx_size, &idx_map))
697 return NULL;
699 /* do we have a corresponding .pack file? */
700 strcpy(path + path_len - 4, ".pack");
701 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
702 munmap(idx_map, idx_size);
703 return NULL;
704 }
705 /* ok, it looks sane as far as we can check without
706 * actually mapping the pack file.
707 */
708 p = xmalloc(sizeof(*p) + path_len + 2);
709 strcpy(p->pack_name, path);
710 p->index_size = idx_size;
711 p->pack_size = st.st_size;
712 p->index_base = idx_map;
713 p->next = NULL;
714 p->windows = NULL;
715 p->pack_fd = -1;
716 p->pack_local = local;
717 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
718 hashcpy(p->sha1, sha1);
719 return p;
720 }
722 struct packed_git *parse_pack_index(unsigned char *sha1)
723 {
724 char *path = sha1_pack_index_name(sha1);
725 return parse_pack_index_file(sha1, path);
726 }
728 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
729 {
730 struct packed_git *p;
731 unsigned long idx_size;
732 void *idx_map;
733 char *path;
735 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
736 return NULL;
738 path = sha1_pack_name(sha1);
740 p = xmalloc(sizeof(*p) + strlen(path) + 2);
741 strcpy(p->pack_name, path);
742 p->index_size = idx_size;
743 p->pack_size = 0;
744 p->index_base = idx_map;
745 p->next = NULL;
746 p->windows = NULL;
747 p->pack_fd = -1;
748 hashcpy(p->sha1, sha1);
749 return p;
750 }
752 void install_packed_git(struct packed_git *pack)
753 {
754 pack->next = packed_git;
755 packed_git = pack;
756 }
758 static void prepare_packed_git_one(char *objdir, int local)
759 {
760 char path[PATH_MAX];
761 int len;
762 DIR *dir;
763 struct dirent *de;
765 sprintf(path, "%s/pack", objdir);
766 len = strlen(path);
767 dir = opendir(path);
768 if (!dir) {
769 if (errno != ENOENT)
770 error("unable to open object pack directory: %s: %s",
771 path, strerror(errno));
772 return;
773 }
774 path[len++] = '/';
775 while ((de = readdir(dir)) != NULL) {
776 int namelen = strlen(de->d_name);
777 struct packed_git *p;
779 if (!has_extension(de->d_name, ".idx"))
780 continue;
782 /* we have .idx. Is it a file we can map? */
783 strcpy(path + len, de->d_name);
784 for (p = packed_git; p; p = p->next) {
785 if (!memcmp(path, p->pack_name, len + namelen - 4))
786 break;
787 }
788 if (p)
789 continue;
790 p = add_packed_git(path, len + namelen, local);
791 if (!p)
792 continue;
793 p->next = packed_git;
794 packed_git = p;
795 }
796 closedir(dir);
797 }
799 static int prepare_packed_git_run_once = 0;
800 void prepare_packed_git(void)
801 {
802 struct alternate_object_database *alt;
804 if (prepare_packed_git_run_once)
805 return;
806 prepare_packed_git_one(get_object_directory(), 1);
807 prepare_alt_odb();
808 for (alt = alt_odb_list; alt; alt = alt->next) {
809 alt->name[-1] = 0;
810 prepare_packed_git_one(alt->base, 0);
811 alt->name[-1] = '/';
812 }
813 prepare_packed_git_run_once = 1;
814 }
816 void reprepare_packed_git(void)
817 {
818 prepare_packed_git_run_once = 0;
819 prepare_packed_git();
820 }
822 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
823 {
824 unsigned char real_sha1[20];
825 hash_sha1_file(map, size, type, real_sha1);
826 return hashcmp(sha1, real_sha1) ? -1 : 0;
827 }
829 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
830 {
831 struct stat st;
832 void *map;
833 int fd;
834 char *filename = find_sha1_file(sha1, &st);
836 if (!filename) {
837 return NULL;
838 }
840 fd = open(filename, O_RDONLY | sha1_file_open_flag);
841 if (fd < 0) {
842 /* See if it works without O_NOATIME */
843 switch (sha1_file_open_flag) {
844 default:
845 fd = open(filename, O_RDONLY);
846 if (fd >= 0)
847 break;
848 /* Fallthrough */
849 case 0:
850 return NULL;
851 }
853 /* If it failed once, it will probably fail again.
854 * Stop using O_NOATIME
855 */
856 sha1_file_open_flag = 0;
857 }
858 map = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
859 close(fd);
860 *size = st.st_size;
861 return map;
862 }
864 int legacy_loose_object(unsigned char *map)
865 {
866 unsigned int word;
868 /*
869 * Is it a zlib-compressed buffer? If so, the first byte
870 * must be 0x78 (15-bit window size, deflated), and the
871 * first 16-bit word is evenly divisible by 31
872 */
873 word = (map[0] << 8) + map[1];
874 if (map[0] == 0x78 && !(word % 31))
875 return 1;
876 else
877 return 0;
878 }
880 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
881 {
882 unsigned shift;
883 unsigned char c;
884 unsigned long size;
885 unsigned long used = 0;
887 c = buf[used++];
888 *type = (c >> 4) & 7;
889 size = c & 15;
890 shift = 4;
891 while (c & 0x80) {
892 if (len <= used)
893 return 0;
894 if (sizeof(long) * 8 <= shift)
895 return 0;
896 c = buf[used++];
897 size += (c & 0x7f) << shift;
898 shift += 7;
899 }
900 *sizep = size;
901 return used;
902 }
904 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
905 {
906 unsigned long size, used;
907 static const char valid_loose_object_type[8] = {
908 0, /* OBJ_EXT */
909 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
910 0, /* "delta" and others are invalid in a loose object */
911 };
912 enum object_type type;
914 /* Get the data stream */
915 memset(stream, 0, sizeof(*stream));
916 stream->next_in = map;
917 stream->avail_in = mapsize;
918 stream->next_out = buffer;
919 stream->avail_out = bufsiz;
921 if (legacy_loose_object(map)) {
922 inflateInit(stream);
923 return inflate(stream, 0);
924 }
926 used = unpack_object_header_gently(map, mapsize, &type, &size);
927 if (!used || !valid_loose_object_type[type])
928 return -1;
929 map += used;
930 mapsize -= used;
932 /* Set up the stream for the rest.. */
933 stream->next_in = map;
934 stream->avail_in = mapsize;
935 inflateInit(stream);
937 /* And generate the fake traditional header */
938 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
939 type_names[type], size);
940 return 0;
941 }
943 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
944 {
945 int bytes = strlen(buffer) + 1;
946 unsigned char *buf = xmalloc(1+size);
947 unsigned long n;
949 n = stream->total_out - bytes;
950 if (n > size)
951 n = size;
952 memcpy(buf, (char *) buffer + bytes, n);
953 bytes = n;
954 if (bytes < size) {
955 stream->next_out = buf + bytes;
956 stream->avail_out = size - bytes;
957 while (inflate(stream, Z_FINISH) == Z_OK)
958 /* nothing */;
959 }
960 buf[size] = 0;
961 inflateEnd(stream);
962 return buf;
963 }
965 /*
966 * We used to just use "sscanf()", but that's actually way
967 * too permissive for what we want to check. So do an anal
968 * object header parse by hand.
969 */
970 static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
971 {
972 int i;
973 unsigned long size;
975 /*
976 * The type can be at most ten bytes (including the
977 * terminating '\0' that we add), and is followed by
978 * a space.
979 */
980 i = 10;
981 for (;;) {
982 char c = *hdr++;
983 if (c == ' ')
984 break;
985 if (!--i)
986 return -1;
987 *type++ = c;
988 }
989 *type = 0;
991 /*
992 * The length must follow immediately, and be in canonical
993 * decimal format (ie "010" is not valid).
994 */
995 size = *hdr++ - '0';
996 if (size > 9)
997 return -1;
998 if (size) {
999 for (;;) {
1000 unsigned long c = *hdr - '0';
1001 if (c > 9)
1002 break;
1003 hdr++;
1004 size = size * 10 + c;
1005 }
1006 }
1007 *sizep = size;
1009 /*
1010 * The length must be followed by a zero byte
1011 */
1012 return *hdr ? -1 : 0;
1013 }
1015 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
1016 {
1017 int ret;
1018 z_stream stream;
1019 char hdr[8192];
1021 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1022 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
1023 return NULL;
1025 return unpack_sha1_rest(&stream, hdr, *size);
1026 }
1028 static unsigned long get_delta_base(struct packed_git *p,
1029 struct pack_window **w_curs,
1030 unsigned long offset,
1031 enum object_type kind,
1032 unsigned long delta_obj_offset,
1033 unsigned long *base_obj_offset)
1034 {
1035 unsigned char *base_info = use_pack(p, w_curs, offset, NULL);
1036 unsigned long base_offset;
1038 /* use_pack() assured us we have [base_info, base_info + 20)
1039 * as a range that we can look at without walking off the
1040 * end of the mapped window. Its actually the hash size
1041 * that is assured. An OFS_DELTA longer than the hash size
1042 * is stupid, as then a REF_DELTA would be smaller to store.
1043 */
1044 if (kind == OBJ_OFS_DELTA) {
1045 unsigned used = 0;
1046 unsigned char c = base_info[used++];
1047 base_offset = c & 127;
1048 while (c & 128) {
1049 base_offset += 1;
1050 if (!base_offset || base_offset & ~(~0UL >> 7))
1051 die("offset value overflow for delta base object");
1052 c = base_info[used++];
1053 base_offset = (base_offset << 7) + (c & 127);
1054 }
1055 base_offset = delta_obj_offset - base_offset;
1056 if (base_offset >= delta_obj_offset)
1057 die("delta base offset out of bound");
1058 offset += used;
1059 } else if (kind == OBJ_REF_DELTA) {
1060 /* The base entry _must_ be in the same pack */
1061 base_offset = find_pack_entry_one(base_info, p);
1062 if (!base_offset)
1063 die("failed to find delta-pack base object %s",
1064 sha1_to_hex(base_info));
1065 offset += 20;
1066 } else
1067 die("I am totally screwed");
1068 *base_obj_offset = base_offset;
1069 return offset;
1070 }
1072 /* forward declaration for a mutually recursive function */
1073 static int packed_object_info(struct packed_git *p, unsigned long offset,
1074 char *type, unsigned long *sizep);
1076 static int packed_delta_info(struct packed_git *p,
1077 struct pack_window **w_curs,
1078 unsigned long offset,
1079 enum object_type kind,
1080 unsigned long obj_offset,
1081 char *type,
1082 unsigned long *sizep)
1083 {
1084 unsigned long base_offset;
1086 offset = get_delta_base(p, w_curs, offset, kind,
1087 obj_offset, &base_offset);
1089 /* We choose to only get the type of the base object and
1090 * ignore potentially corrupt pack file that expects the delta
1091 * based on a base with a wrong size. This saves tons of
1092 * inflate() calls.
1093 */
1094 if (packed_object_info(p, base_offset, type, NULL))
1095 die("cannot get info for delta-pack base");
1097 if (sizep) {
1098 const unsigned char *data;
1099 unsigned char delta_head[20], *in;
1100 unsigned long result_size;
1101 z_stream stream;
1102 int st;
1104 memset(&stream, 0, sizeof(stream));
1105 stream.next_out = delta_head;
1106 stream.avail_out = sizeof(delta_head);
1108 inflateInit(&stream);
1109 do {
1110 in = use_pack(p, w_curs, offset, &stream.avail_in);
1111 stream.next_in = in;
1112 st = inflate(&stream, Z_FINISH);
1113 offset += stream.next_in - in;
1114 } while ((st == Z_OK || st == Z_BUF_ERROR)
1115 && stream.total_out < sizeof(delta_head));
1116 inflateEnd(&stream);
1117 if ((st != Z_STREAM_END) &&
1118 stream.total_out != sizeof(delta_head))
1119 die("delta data unpack-initial failed");
1121 /* Examine the initial part of the delta to figure out
1122 * the result size.
1123 */
1124 data = delta_head;
1126 /* ignore base size */
1127 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1129 /* Read the result size */
1130 result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1131 *sizep = result_size;
1132 }
1133 return 0;
1134 }
1136 static unsigned long unpack_object_header(struct packed_git *p,
1137 struct pack_window **w_curs,
1138 unsigned long offset,
1139 enum object_type *type,
1140 unsigned long *sizep)
1141 {
1142 unsigned char *base;
1143 unsigned int left;
1144 unsigned long used;
1146 /* use_pack() assures us we have [base, base + 20) available
1147 * as a range that we can look at at. (Its actually the hash
1148 * size that is assurred.) With our object header encoding
1149 * the maximum deflated object size is 2^137, which is just
1150 * insane, so we know won't exceed what we have been given.
1151 */
1152 base = use_pack(p, w_curs, offset, &left);
1153 used = unpack_object_header_gently(base, left, type, sizep);
1154 if (!used)
1155 die("object offset outside of pack file");
1157 return offset + used;
1158 }
1160 void packed_object_info_detail(struct packed_git *p,
1161 unsigned long offset,
1162 char *type,
1163 unsigned long *size,
1164 unsigned long *store_size,
1165 unsigned int *delta_chain_length,
1166 unsigned char *base_sha1)
1167 {
1168 struct pack_window *w_curs = NULL;
1169 unsigned long obj_offset, val;
1170 unsigned char *next_sha1;
1171 enum object_type kind;
1173 *delta_chain_length = 0;
1174 obj_offset = offset;
1175 offset = unpack_object_header(p, &w_curs, offset, &kind, size);
1177 for (;;) {
1178 switch (kind) {
1179 default:
1180 die("pack %s contains unknown object type %d",
1181 p->pack_name, kind);
1182 case OBJ_COMMIT:
1183 case OBJ_TREE:
1184 case OBJ_BLOB:
1185 case OBJ_TAG:
1186 strcpy(type, type_names[kind]);
1187 *store_size = 0; /* notyet */
1188 unuse_pack(&w_curs);
1189 return;
1190 case OBJ_OFS_DELTA:
1191 get_delta_base(p, &w_curs, offset, kind,
1192 obj_offset, &offset);
1193 if (*delta_chain_length == 0) {
1194 /* TODO: find base_sha1 as pointed by offset */
1195 }
1196 break;
1197 case OBJ_REF_DELTA:
1198 next_sha1 = use_pack(p, &w_curs, offset, NULL);
1199 if (*delta_chain_length == 0)
1200 hashcpy(base_sha1, next_sha1);
1201 offset = find_pack_entry_one(next_sha1, p);
1202 break;
1203 }
1204 obj_offset = offset;
1205 offset = unpack_object_header(p, &w_curs, offset, &kind, &val);
1206 (*delta_chain_length)++;
1207 }
1208 }
1210 static int packed_object_info(struct packed_git *p, unsigned long offset,
1211 char *type, unsigned long *sizep)
1212 {
1213 struct pack_window *w_curs = NULL;
1214 unsigned long size, obj_offset = offset;
1215 enum object_type kind;
1216 int r;
1218 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1220 switch (kind) {
1221 case OBJ_OFS_DELTA:
1222 case OBJ_REF_DELTA:
1223 r = packed_delta_info(p, &w_curs, offset, kind,
1224 obj_offset, type, sizep);
1225 unuse_pack(&w_curs);
1226 return r;
1227 case OBJ_COMMIT:
1228 case OBJ_TREE:
1229 case OBJ_BLOB:
1230 case OBJ_TAG:
1231 strcpy(type, type_names[kind]);
1232 unuse_pack(&w_curs);
1233 break;
1234 default:
1235 die("pack %s contains unknown object type %d",
1236 p->pack_name, kind);
1237 }
1238 if (sizep)
1239 *sizep = size;
1240 return 0;
1241 }
1243 static void *unpack_compressed_entry(struct packed_git *p,
1244 struct pack_window **w_curs,
1245 unsigned long offset,
1246 unsigned long size)
1247 {
1248 int st;
1249 z_stream stream;
1250 unsigned char *buffer, *in;
1252 buffer = xmalloc(size + 1);
1253 buffer[size] = 0;
1254 memset(&stream, 0, sizeof(stream));
1255 stream.next_out = buffer;
1256 stream.avail_out = size;
1258 inflateInit(&stream);
1259 do {
1260 in = use_pack(p, w_curs, offset, &stream.avail_in);
1261 stream.next_in = in;
1262 st = inflate(&stream, Z_FINISH);
1263 offset += stream.next_in - in;
1264 } while (st == Z_OK || st == Z_BUF_ERROR);
1265 inflateEnd(&stream);
1266 if ((st != Z_STREAM_END) || stream.total_out != size) {
1267 free(buffer);
1268 return NULL;
1269 }
1271 return buffer;
1272 }
1274 static void *unpack_delta_entry(struct packed_git *p,
1275 struct pack_window **w_curs,
1276 unsigned long offset,
1277 unsigned long delta_size,
1278 enum object_type kind,
1279 unsigned long obj_offset,
1280 char *type,
1281 unsigned long *sizep)
1282 {
1283 void *delta_data, *result, *base;
1284 unsigned long result_size, base_size, base_offset;
1286 offset = get_delta_base(p, w_curs, offset, kind,
1287 obj_offset, &base_offset);
1288 base = unpack_entry(p, base_offset, type, &base_size);
1289 if (!base)
1290 die("failed to read delta base object at %lu from %s",
1291 base_offset, p->pack_name);
1293 delta_data = unpack_compressed_entry(p, w_curs, offset, delta_size);
1294 result = patch_delta(base, base_size,
1295 delta_data, delta_size,
1296 &result_size);
1297 if (!result)
1298 die("failed to apply delta");
1299 free(delta_data);
1300 free(base);
1301 *sizep = result_size;
1302 return result;
1303 }
1305 void *unpack_entry(struct packed_git *p, unsigned long offset,
1306 char *type, unsigned long *sizep)
1307 {
1308 struct pack_window *w_curs = NULL;
1309 unsigned long size, obj_offset = offset;
1310 enum object_type kind;
1311 void *retval;
1313 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1314 switch (kind) {
1315 case OBJ_OFS_DELTA:
1316 case OBJ_REF_DELTA:
1317 retval = unpack_delta_entry(p, &w_curs, offset, size,
1318 kind, obj_offset, type, sizep);
1319 break;
1320 case OBJ_COMMIT:
1321 case OBJ_TREE:
1322 case OBJ_BLOB:
1323 case OBJ_TAG:
1324 strcpy(type, type_names[kind]);
1325 *sizep = size;
1326 retval = unpack_compressed_entry(p, &w_curs, offset, size);
1327 break;
1328 default:
1329 die("unknown object type %i in %s", kind, p->pack_name);
1330 }
1331 unuse_pack(&w_curs);
1332 return retval;
1333 }
1335 int num_packed_objects(const struct packed_git *p)
1336 {
1337 /* See check_packed_git_idx() */
1338 return (p->index_size - 20 - 20 - 4*256) / 24;
1339 }
1341 int nth_packed_object_sha1(const struct packed_git *p, int n,
1342 unsigned char* sha1)
1343 {
1344 void *index = p->index_base + 256;
1345 if (n < 0 || num_packed_objects(p) <= n)
1346 return -1;
1347 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1348 return 0;
1349 }
1351 unsigned long find_pack_entry_one(const unsigned char *sha1,
1352 struct packed_git *p)
1353 {
1354 uint32_t *level1_ofs = p->index_base;
1355 int hi = ntohl(level1_ofs[*sha1]);
1356 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1357 void *index = p->index_base + 256;
1359 do {
1360 int mi = (lo + hi) / 2;
1361 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1362 if (!cmp)
1363 return ntohl(*((uint32_t *)((char *)index + (24 * mi))));
1364 if (cmp > 0)
1365 hi = mi;
1366 else
1367 lo = mi+1;
1368 } while (lo < hi);
1369 return 0;
1370 }
1372 static int matches_pack_name(struct packed_git *p, const char *ig)
1373 {
1374 const char *last_c, *c;
1376 if (!strcmp(p->pack_name, ig))
1377 return 0;
1379 for (c = p->pack_name, last_c = c; *c;)
1380 if (*c == '/')
1381 last_c = ++c;
1382 else
1383 ++c;
1384 if (!strcmp(last_c, ig))
1385 return 0;
1387 return 1;
1388 }
1390 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1391 {
1392 struct packed_git *p;
1393 unsigned long offset;
1395 prepare_packed_git();
1397 for (p = packed_git; p; p = p->next) {
1398 if (ignore_packed) {
1399 const char **ig;
1400 for (ig = ignore_packed; *ig; ig++)
1401 if (!matches_pack_name(p, *ig))
1402 break;
1403 if (*ig)
1404 continue;
1405 }
1406 offset = find_pack_entry_one(sha1, p);
1407 if (offset) {
1408 e->offset = offset;
1409 e->p = p;
1410 hashcpy(e->sha1, sha1);
1411 return 1;
1412 }
1413 }
1414 return 0;
1415 }
1417 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1418 struct packed_git *packs)
1419 {
1420 struct packed_git *p;
1422 for (p = packs; p; p = p->next) {
1423 if (find_pack_entry_one(sha1, p))
1424 return p;
1425 }
1426 return NULL;
1428 }
1430 static int sha1_loose_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1431 {
1432 int status;
1433 unsigned long mapsize, size;
1434 void *map;
1435 z_stream stream;
1436 char hdr[128];
1438 map = map_sha1_file(sha1, &mapsize);
1439 if (!map)
1440 return error("unable to find %s", sha1_to_hex(sha1));
1441 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1442 status = error("unable to unpack %s header",
1443 sha1_to_hex(sha1));
1444 if (parse_sha1_header(hdr, type, &size) < 0)
1445 status = error("unable to parse %s header", sha1_to_hex(sha1));
1446 else {
1447 status = 0;
1448 if (sizep)
1449 *sizep = size;
1450 }
1451 inflateEnd(&stream);
1452 munmap(map, mapsize);
1453 return status;
1454 }
1456 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1457 {
1458 struct pack_entry e;
1460 if (!find_pack_entry(sha1, &e, NULL)) {
1461 reprepare_packed_git();
1462 if (!find_pack_entry(sha1, &e, NULL))
1463 return sha1_loose_object_info(sha1, type, sizep);
1464 }
1465 return packed_object_info(e.p, e.offset, type, sizep);
1466 }
1468 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1469 {
1470 struct pack_entry e;
1472 if (!find_pack_entry(sha1, &e, NULL)) {
1473 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1474 return NULL;
1475 }
1476 return unpack_entry(e.p, e.offset, type, size);
1477 }
1479 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1480 {
1481 unsigned long mapsize;
1482 void *map, *buf;
1483 struct pack_entry e;
1485 if (find_pack_entry(sha1, &e, NULL))
1486 return read_packed_sha1(sha1, type, size);
1487 map = map_sha1_file(sha1, &mapsize);
1488 if (map) {
1489 buf = unpack_sha1_file(map, mapsize, type, size);
1490 munmap(map, mapsize);
1491 return buf;
1492 }
1493 reprepare_packed_git();
1494 if (find_pack_entry(sha1, &e, NULL))
1495 return read_packed_sha1(sha1, type, size);
1496 return NULL;
1497 }
1499 void *read_object_with_reference(const unsigned char *sha1,
1500 const char *required_type,
1501 unsigned long *size,
1502 unsigned char *actual_sha1_return)
1503 {
1504 char type[20];
1505 void *buffer;
1506 unsigned long isize;
1507 unsigned char actual_sha1[20];
1509 hashcpy(actual_sha1, sha1);
1510 while (1) {
1511 int ref_length = -1;
1512 const char *ref_type = NULL;
1514 buffer = read_sha1_file(actual_sha1, type, &isize);
1515 if (!buffer)
1516 return NULL;
1517 if (!strcmp(type, required_type)) {
1518 *size = isize;
1519 if (actual_sha1_return)
1520 hashcpy(actual_sha1_return, actual_sha1);
1521 return buffer;
1522 }
1523 /* Handle references */
1524 else if (!strcmp(type, commit_type))
1525 ref_type = "tree ";
1526 else if (!strcmp(type, tag_type))
1527 ref_type = "object ";
1528 else {
1529 free(buffer);
1530 return NULL;
1531 }
1532 ref_length = strlen(ref_type);
1534 if (memcmp(buffer, ref_type, ref_length) ||
1535 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1536 free(buffer);
1537 return NULL;
1538 }
1539 free(buffer);
1540 /* Now we have the ID of the referred-to object in
1541 * actual_sha1. Check again. */
1542 }
1543 }
1545 static void write_sha1_file_prepare(void *buf, unsigned long len,
1546 const char *type, unsigned char *sha1,
1547 unsigned char *hdr, int *hdrlen)
1548 {
1549 SHA_CTX c;
1551 /* Generate the header */
1552 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1554 /* Sha1.. */
1555 SHA1_Init(&c);
1556 SHA1_Update(&c, hdr, *hdrlen);
1557 SHA1_Update(&c, buf, len);
1558 SHA1_Final(sha1, &c);
1559 }
1561 /*
1562 * Link the tempfile to the final place, possibly creating the
1563 * last directory level as you do so.
1564 *
1565 * Returns the errno on failure, 0 on success.
1566 */
1567 static int link_temp_to_file(const char *tmpfile, const char *filename)
1568 {
1569 int ret;
1570 char *dir;
1572 if (!link(tmpfile, filename))
1573 return 0;
1575 /*
1576 * Try to mkdir the last path component if that failed.
1577 *
1578 * Re-try the "link()" regardless of whether the mkdir
1579 * succeeds, since a race might mean that somebody
1580 * else succeeded.
1581 */
1582 ret = errno;
1583 dir = strrchr(filename, '/');
1584 if (dir) {
1585 *dir = 0;
1586 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1587 *dir = '/';
1588 return -2;
1589 }
1590 *dir = '/';
1591 if (!link(tmpfile, filename))
1592 return 0;
1593 ret = errno;
1594 }
1595 return ret;
1596 }
1598 /*
1599 * Move the just written object into its final resting place
1600 */
1601 int move_temp_to_file(const char *tmpfile, const char *filename)
1602 {
1603 int ret = link_temp_to_file(tmpfile, filename);
1605 /*
1606 * Coda hack - coda doesn't like cross-directory links,
1607 * so we fall back to a rename, which will mean that it
1608 * won't be able to check collisions, but that's not a
1609 * big deal.
1610 *
1611 * The same holds for FAT formatted media.
1612 *
1613 * When this succeeds, we just return 0. We have nothing
1614 * left to unlink.
1615 */
1616 if (ret && ret != EEXIST) {
1617 if (!rename(tmpfile, filename))
1618 return 0;
1619 ret = errno;
1620 }
1621 unlink(tmpfile);
1622 if (ret) {
1623 if (ret != EEXIST) {
1624 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1625 }
1626 /* FIXME!!! Collision check here ? */
1627 }
1629 return 0;
1630 }
1632 static int write_buffer(int fd, const void *buf, size_t len)
1633 {
1634 if (write_in_full(fd, buf, len) < 0)
1635 return error("file write error (%s)", strerror(errno));
1636 return 0;
1637 }
1639 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1640 {
1641 int hdr_len;
1642 unsigned char c;
1644 c = (type << 4) | (len & 15);
1645 len >>= 4;
1646 hdr_len = 1;
1647 while (len) {
1648 *hdr++ = c | 0x80;
1649 hdr_len++;
1650 c = (len & 0x7f);
1651 len >>= 7;
1652 }
1653 *hdr = c;
1654 return hdr_len;
1655 }
1657 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1658 {
1659 int obj_type, hdr;
1661 if (use_legacy_headers) {
1662 while (deflate(stream, 0) == Z_OK)
1663 /* nothing */;
1664 return;
1665 }
1666 if (!strcmp(type, blob_type))
1667 obj_type = OBJ_BLOB;
1668 else if (!strcmp(type, tree_type))
1669 obj_type = OBJ_TREE;
1670 else if (!strcmp(type, commit_type))
1671 obj_type = OBJ_COMMIT;
1672 else if (!strcmp(type, tag_type))
1673 obj_type = OBJ_TAG;
1674 else
1675 die("trying to generate bogus object of type '%s'", type);
1676 hdr = write_binary_header(stream->next_out, obj_type, len);
1677 stream->total_out = hdr;
1678 stream->next_out += hdr;
1679 stream->avail_out -= hdr;
1680 }
1682 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1683 unsigned char *sha1)
1684 {
1685 unsigned char hdr[50];
1686 int hdrlen;
1687 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1688 return 0;
1689 }
1691 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1692 {
1693 int size;
1694 unsigned char *compressed;
1695 z_stream stream;
1696 unsigned char sha1[20];
1697 char *filename;
1698 static char tmpfile[PATH_MAX];
1699 unsigned char hdr[50];
1700 int fd, hdrlen;
1702 /* Normally if we have it in the pack then we do not bother writing
1703 * it out into .git/objects/??/?{38} file.
1704 */
1705 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1706 filename = sha1_file_name(sha1);
1707 if (returnsha1)
1708 hashcpy(returnsha1, sha1);
1709 if (has_sha1_file(sha1))
1710 return 0;
1711 fd = open(filename, O_RDONLY);
1712 if (fd >= 0) {
1713 /*
1714 * FIXME!!! We might do collision checking here, but we'd
1715 * need to uncompress the old file and check it. Later.
1716 */
1717 close(fd);
1718 return 0;
1719 }
1721 if (errno != ENOENT) {
1722 return error("sha1 file %s: %s\n", filename, strerror(errno));
1723 }
1725 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1727 fd = mkstemp(tmpfile);
1728 if (fd < 0) {
1729 if (errno == EPERM)
1730 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1731 else
1732 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1733 }
1735 /* Set it up */
1736 memset(&stream, 0, sizeof(stream));
1737 deflateInit(&stream, zlib_compression_level);
1738 size = 8 + deflateBound(&stream, len+hdrlen);
1739 compressed = xmalloc(size);
1741 /* Compress it */
1742 stream.next_out = compressed;
1743 stream.avail_out = size;
1745 /* First header.. */
1746 stream.next_in = hdr;
1747 stream.avail_in = hdrlen;
1748 setup_object_header(&stream, type, len);
1750 /* Then the data itself.. */
1751 stream.next_in = buf;
1752 stream.avail_in = len;
1753 while (deflate(&stream, Z_FINISH) == Z_OK)
1754 /* nothing */;
1755 deflateEnd(&stream);
1756 size = stream.total_out;
1758 if (write_buffer(fd, compressed, size) < 0)
1759 die("unable to write sha1 file");
1760 fchmod(fd, 0444);
1761 close(fd);
1762 free(compressed);
1764 return move_temp_to_file(tmpfile, filename);
1765 }
1767 /*
1768 * We need to unpack and recompress the object for writing
1769 * it out to a different file.
1770 */
1771 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1772 {
1773 size_t size;
1774 z_stream stream;
1775 unsigned char *unpacked;
1776 unsigned long len;
1777 char type[20];
1778 char hdr[50];
1779 int hdrlen;
1780 void *buf;
1782 /* need to unpack and recompress it by itself */
1783 unpacked = read_packed_sha1(sha1, type, &len);
1785 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1787 /* Set it up */
1788 memset(&stream, 0, sizeof(stream));
1789 deflateInit(&stream, zlib_compression_level);
1790 size = deflateBound(&stream, len + hdrlen);
1791 buf = xmalloc(size);
1793 /* Compress it */
1794 stream.next_out = buf;
1795 stream.avail_out = size;
1797 /* First header.. */
1798 stream.next_in = (void *)hdr;
1799 stream.avail_in = hdrlen;
1800 while (deflate(&stream, 0) == Z_OK)
1801 /* nothing */;
1803 /* Then the data itself.. */
1804 stream.next_in = unpacked;
1805 stream.avail_in = len;
1806 while (deflate(&stream, Z_FINISH) == Z_OK)
1807 /* nothing */;
1808 deflateEnd(&stream);
1809 free(unpacked);
1811 *objsize = stream.total_out;
1812 return buf;
1813 }
1815 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1816 {
1817 int retval;
1818 unsigned long objsize;
1819 void *buf = map_sha1_file(sha1, &objsize);
1821 if (buf) {
1822 retval = write_buffer(fd, buf, objsize);
1823 munmap(buf, objsize);
1824 return retval;
1825 }
1827 buf = repack_object(sha1, &objsize);
1828 retval = write_buffer(fd, buf, objsize);
1829 free(buf);
1830 return retval;
1831 }
1833 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1834 size_t bufsize, size_t *bufposn)
1835 {
1836 char tmpfile[PATH_MAX];
1837 int local;
1838 z_stream stream;
1839 unsigned char real_sha1[20];
1840 unsigned char discard[4096];
1841 int ret;
1842 SHA_CTX c;
1844 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1846 local = mkstemp(tmpfile);
1847 if (local < 0) {
1848 if (errno == EPERM)
1849 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1850 else
1851 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1852 }
1854 memset(&stream, 0, sizeof(stream));
1856 inflateInit(&stream);
1858 SHA1_Init(&c);
1860 do {
1861 ssize_t size;
1862 if (*bufposn) {
1863 stream.avail_in = *bufposn;
1864 stream.next_in = (unsigned char *) buffer;
1865 do {
1866 stream.next_out = discard;
1867 stream.avail_out = sizeof(discard);
1868 ret = inflate(&stream, Z_SYNC_FLUSH);
1869 SHA1_Update(&c, discard, sizeof(discard) -
1870 stream.avail_out);
1871 } while (stream.avail_in && ret == Z_OK);
1872 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1873 die("unable to write sha1 file");
1874 memmove(buffer, buffer + *bufposn - stream.avail_in,
1875 stream.avail_in);
1876 *bufposn = stream.avail_in;
1877 if (ret != Z_OK)
1878 break;
1879 }
1880 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
1881 if (size <= 0) {
1882 close(local);
1883 unlink(tmpfile);
1884 if (!size)
1885 return error("Connection closed?");
1886 perror("Reading from connection");
1887 return -1;
1888 }
1889 *bufposn += size;
1890 } while (1);
1891 inflateEnd(&stream);
1893 close(local);
1894 SHA1_Final(real_sha1, &c);
1895 if (ret != Z_STREAM_END) {
1896 unlink(tmpfile);
1897 return error("File %s corrupted", sha1_to_hex(sha1));
1898 }
1899 if (hashcmp(sha1, real_sha1)) {
1900 unlink(tmpfile);
1901 return error("File %s has bad hash", sha1_to_hex(sha1));
1902 }
1904 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1905 }
1907 int has_pack_index(const unsigned char *sha1)
1908 {
1909 struct stat st;
1910 if (stat(sha1_pack_index_name(sha1), &st))
1911 return 0;
1912 return 1;
1913 }
1915 int has_pack_file(const unsigned char *sha1)
1916 {
1917 struct stat st;
1918 if (stat(sha1_pack_name(sha1), &st))
1919 return 0;
1920 return 1;
1921 }
1923 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1924 {
1925 struct pack_entry e;
1926 return find_pack_entry(sha1, &e, ignore_packed);
1927 }
1929 int has_sha1_file(const unsigned char *sha1)
1930 {
1931 struct stat st;
1932 struct pack_entry e;
1934 if (find_pack_entry(sha1, &e, NULL))
1935 return 1;
1936 return find_sha1_file(sha1, &st) ? 1 : 0;
1937 }
1939 /*
1940 * reads from fd as long as possible into a supplied buffer of size bytes.
1941 * If necessary the buffer's size is increased using realloc()
1942 *
1943 * returns 0 if anything went fine and -1 otherwise
1944 *
1945 * NOTE: both buf and size may change, but even when -1 is returned
1946 * you still have to free() it yourself.
1947 */
1948 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
1949 {
1950 char* buf = *return_buf;
1951 unsigned long size = *return_size;
1952 int iret;
1953 unsigned long off = 0;
1955 do {
1956 iret = xread(fd, buf + off, size - off);
1957 if (iret > 0) {
1958 off += iret;
1959 if (off == size) {
1960 size *= 2;
1961 buf = xrealloc(buf, size);
1962 }
1963 }
1964 } while (iret > 0);
1966 *return_buf = buf;
1967 *return_size = off;
1969 if (iret < 0)
1970 return -1;
1971 return 0;
1972 }
1974 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
1975 {
1976 unsigned long size = 4096;
1977 char *buf = xmalloc(size);
1978 int ret;
1980 if (read_pipe(fd, &buf, &size)) {
1981 free(buf);
1982 return -1;
1983 }
1985 if (!type)
1986 type = blob_type;
1987 if (write_object)
1988 ret = write_sha1_file(buf, size, type, sha1);
1989 else
1990 ret = hash_sha1_file(buf, size, type, sha1);
1991 free(buf);
1992 return ret;
1993 }
1995 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1996 {
1997 unsigned long size = st->st_size;
1998 void *buf;
1999 int ret;
2001 buf = "";
2002 if (size)
2003 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2004 close(fd);
2006 if (!type)
2007 type = blob_type;
2008 if (write_object)
2009 ret = write_sha1_file(buf, size, type, sha1);
2010 else
2011 ret = hash_sha1_file(buf, size, type, sha1);
2012 if (size)
2013 munmap(buf, size);
2014 return ret;
2015 }
2017 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2018 {
2019 int fd;
2020 char *target;
2022 switch (st->st_mode & S_IFMT) {
2023 case S_IFREG:
2024 fd = open(path, O_RDONLY);
2025 if (fd < 0)
2026 return error("open(\"%s\"): %s", path,
2027 strerror(errno));
2028 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
2029 return error("%s: failed to insert into database",
2030 path);
2031 break;
2032 case S_IFLNK:
2033 target = xmalloc(st->st_size+1);
2034 if (readlink(path, target, st->st_size+1) != st->st_size) {
2035 char *errstr = strerror(errno);
2036 free(target);
2037 return error("readlink(\"%s\"): %s", path,
2038 errstr);
2039 }
2040 if (!write_object)
2041 hash_sha1_file(target, st->st_size, blob_type, sha1);
2042 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
2043 return error("%s: failed to insert into database",
2044 path);
2045 free(target);
2046 break;
2047 default:
2048 return error("%s: unsupported file type", path);
2049 }
2050 return 0;
2051 }