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 /*
556 * Do not call this directly as this leaks p->pack_fd on error return;
557 * call open_packed_git() instead.
558 */
559 static int open_packed_git_1(struct packed_git *p)
560 {
561 struct stat st;
562 struct pack_header hdr;
563 unsigned char sha1[20];
564 unsigned char *idx_sha1;
565 long fd_flag;
567 p->pack_fd = open(p->pack_name, O_RDONLY);
568 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
569 return -1;
571 /* If we created the struct before we had the pack we lack size. */
572 if (!p->pack_size) {
573 if (!S_ISREG(st.st_mode))
574 return error("packfile %s not a regular file", p->pack_name);
575 p->pack_size = st.st_size;
576 } else if (p->pack_size != st.st_size)
577 return error("packfile %s size changed", p->pack_name);
579 /* We leave these file descriptors open with sliding mmap;
580 * there is no point keeping them open across exec(), though.
581 */
582 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
583 if (fd_flag < 0)
584 return error("cannot determine file descriptor flags");
585 fd_flag |= FD_CLOEXEC;
586 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
587 return error("cannot set FD_CLOEXEC");
589 /* Verify we recognize this pack file format. */
590 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
591 return error("file %s is far too short to be a packfile", p->pack_name);
592 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
593 return error("file %s is not a GIT packfile", p->pack_name);
594 if (!pack_version_ok(hdr.hdr_version))
595 return error("packfile %s is version %u and not supported"
596 " (try upgrading GIT to a newer version)",
597 p->pack_name, ntohl(hdr.hdr_version));
599 /* Verify the pack matches its index. */
600 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
601 return error("packfile %s claims to have %u objects"
602 " while index size indicates %u objects",
603 p->pack_name, ntohl(hdr.hdr_entries),
604 num_packed_objects(p));
605 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
606 return error("end of packfile %s is unavailable", p->pack_name);
607 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
608 return error("packfile %s signature is unavailable", p->pack_name);
609 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
610 if (hashcmp(sha1, idx_sha1))
611 return error("packfile %s does not match index", p->pack_name);
612 return 0;
613 }
615 static int open_packed_git(struct packed_git *p)
616 {
617 if (!open_packed_git_1(p))
618 return 0;
619 if (p->pack_fd != -1) {
620 close(p->pack_fd);
621 p->pack_fd = -1;
622 }
623 return -1;
624 }
626 static int in_window(struct pack_window *win, unsigned long offset)
627 {
628 /* We must promise at least 20 bytes (one hash) after the
629 * offset is available from this window, otherwise the offset
630 * is not actually in this window and a different window (which
631 * has that one hash excess) must be used. This is to support
632 * the object header and delta base parsing routines below.
633 */
634 off_t win_off = win->offset;
635 return win_off <= offset
636 && (offset + 20) <= (win_off + win->len);
637 }
639 unsigned char* use_pack(struct packed_git *p,
640 struct pack_window **w_cursor,
641 unsigned long offset,
642 unsigned int *left)
643 {
644 struct pack_window *win = *w_cursor;
646 if (p->pack_fd == -1 && open_packed_git(p))
647 die("packfile %s cannot be accessed", p->pack_name);
649 /* Since packfiles end in a hash of their content and its
650 * pointless to ask for an offset into the middle of that
651 * hash, and the in_window function above wouldn't match
652 * don't allow an offset too close to the end of the file.
653 */
654 if (offset > (p->pack_size - 20))
655 die("offset beyond end of packfile (truncated pack?)");
657 if (!win || !in_window(win, offset)) {
658 if (win)
659 win->inuse_cnt--;
660 for (win = p->windows; win; win = win->next) {
661 if (in_window(win, offset))
662 break;
663 }
664 if (!win) {
665 if (!page_size)
666 page_size = getpagesize();
667 win = xcalloc(1, sizeof(*win));
668 win->offset = (offset / page_size) * page_size;
669 win->len = p->pack_size - win->offset;
670 if (win->len > packed_git_window_size)
671 win->len = packed_git_window_size;
672 pack_mapped += win->len;
673 while (packed_git_limit < pack_mapped
674 && unuse_one_window(p))
675 ; /* nothing */
676 win->base = xmmap(NULL, win->len,
677 PROT_READ, MAP_PRIVATE,
678 p->pack_fd, win->offset);
679 if (win->base == MAP_FAILED)
680 die("packfile %s cannot be mapped: %s",
681 p->pack_name,
682 strerror(errno));
683 pack_mmap_calls++;
684 pack_open_windows++;
685 if (pack_mapped > peak_pack_mapped)
686 peak_pack_mapped = pack_mapped;
687 if (pack_open_windows > peak_pack_open_windows)
688 peak_pack_open_windows = pack_open_windows;
689 win->next = p->windows;
690 p->windows = win;
691 }
692 }
693 if (win != *w_cursor) {
694 win->last_used = pack_used_ctr++;
695 win->inuse_cnt++;
696 *w_cursor = win;
697 }
698 offset -= win->offset;
699 if (left)
700 *left = win->len - offset;
701 return win->base + offset;
702 }
704 struct packed_git *add_packed_git(char *path, int path_len, int local)
705 {
706 struct stat st;
707 struct packed_git *p;
708 unsigned long idx_size;
709 void *idx_map;
710 unsigned char sha1[20];
712 if (check_packed_git_idx(path, &idx_size, &idx_map))
713 return NULL;
715 /* do we have a corresponding .pack file? */
716 strcpy(path + path_len - 4, ".pack");
717 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
718 munmap(idx_map, idx_size);
719 return NULL;
720 }
721 /* ok, it looks sane as far as we can check without
722 * actually mapping the pack file.
723 */
724 p = xmalloc(sizeof(*p) + path_len + 2);
725 strcpy(p->pack_name, path);
726 p->index_size = idx_size;
727 p->pack_size = st.st_size;
728 p->index_base = idx_map;
729 p->next = NULL;
730 p->windows = NULL;
731 p->pack_fd = -1;
732 p->pack_local = local;
733 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
734 hashcpy(p->sha1, sha1);
735 return p;
736 }
738 struct packed_git *parse_pack_index(unsigned char *sha1)
739 {
740 char *path = sha1_pack_index_name(sha1);
741 return parse_pack_index_file(sha1, path);
742 }
744 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
745 {
746 struct packed_git *p;
747 unsigned long idx_size;
748 void *idx_map;
749 char *path;
751 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
752 return NULL;
754 path = sha1_pack_name(sha1);
756 p = xmalloc(sizeof(*p) + strlen(path) + 2);
757 strcpy(p->pack_name, path);
758 p->index_size = idx_size;
759 p->pack_size = 0;
760 p->index_base = idx_map;
761 p->next = NULL;
762 p->windows = NULL;
763 p->pack_fd = -1;
764 hashcpy(p->sha1, sha1);
765 return p;
766 }
768 void install_packed_git(struct packed_git *pack)
769 {
770 pack->next = packed_git;
771 packed_git = pack;
772 }
774 static void prepare_packed_git_one(char *objdir, int local)
775 {
776 char path[PATH_MAX];
777 int len;
778 DIR *dir;
779 struct dirent *de;
781 sprintf(path, "%s/pack", objdir);
782 len = strlen(path);
783 dir = opendir(path);
784 if (!dir) {
785 if (errno != ENOENT)
786 error("unable to open object pack directory: %s: %s",
787 path, strerror(errno));
788 return;
789 }
790 path[len++] = '/';
791 while ((de = readdir(dir)) != NULL) {
792 int namelen = strlen(de->d_name);
793 struct packed_git *p;
795 if (!has_extension(de->d_name, ".idx"))
796 continue;
798 /* Don't reopen a pack we already have. */
799 strcpy(path + len, de->d_name);
800 for (p = packed_git; p; p = p->next) {
801 if (!memcmp(path, p->pack_name, len + namelen - 4))
802 break;
803 }
804 if (p)
805 continue;
806 /* See if it really is a valid .idx file with corresponding
807 * .pack file that we can map.
808 */
809 p = add_packed_git(path, len + namelen, local);
810 if (!p)
811 continue;
812 install_packed_git(p);
813 }
814 closedir(dir);
815 }
817 static int prepare_packed_git_run_once = 0;
818 void prepare_packed_git(void)
819 {
820 struct alternate_object_database *alt;
822 if (prepare_packed_git_run_once)
823 return;
824 prepare_packed_git_one(get_object_directory(), 1);
825 prepare_alt_odb();
826 for (alt = alt_odb_list; alt; alt = alt->next) {
827 alt->name[-1] = 0;
828 prepare_packed_git_one(alt->base, 0);
829 alt->name[-1] = '/';
830 }
831 prepare_packed_git_run_once = 1;
832 }
834 void reprepare_packed_git(void)
835 {
836 prepare_packed_git_run_once = 0;
837 prepare_packed_git();
838 }
840 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
841 {
842 unsigned char real_sha1[20];
843 hash_sha1_file(map, size, type, real_sha1);
844 return hashcmp(sha1, real_sha1) ? -1 : 0;
845 }
847 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
848 {
849 struct stat st;
850 void *map;
851 int fd;
852 char *filename = find_sha1_file(sha1, &st);
854 if (!filename) {
855 return NULL;
856 }
858 fd = open(filename, O_RDONLY | sha1_file_open_flag);
859 if (fd < 0) {
860 /* See if it works without O_NOATIME */
861 switch (sha1_file_open_flag) {
862 default:
863 fd = open(filename, O_RDONLY);
864 if (fd >= 0)
865 break;
866 /* Fallthrough */
867 case 0:
868 return NULL;
869 }
871 /* If it failed once, it will probably fail again.
872 * Stop using O_NOATIME
873 */
874 sha1_file_open_flag = 0;
875 }
876 map = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
877 close(fd);
878 *size = st.st_size;
879 return map;
880 }
882 int legacy_loose_object(unsigned char *map)
883 {
884 unsigned int word;
886 /*
887 * Is it a zlib-compressed buffer? If so, the first byte
888 * must be 0x78 (15-bit window size, deflated), and the
889 * first 16-bit word is evenly divisible by 31
890 */
891 word = (map[0] << 8) + map[1];
892 if (map[0] == 0x78 && !(word % 31))
893 return 1;
894 else
895 return 0;
896 }
898 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
899 {
900 unsigned shift;
901 unsigned char c;
902 unsigned long size;
903 unsigned long used = 0;
905 c = buf[used++];
906 *type = (c >> 4) & 7;
907 size = c & 15;
908 shift = 4;
909 while (c & 0x80) {
910 if (len <= used)
911 return 0;
912 if (sizeof(long) * 8 <= shift)
913 return 0;
914 c = buf[used++];
915 size += (c & 0x7f) << shift;
916 shift += 7;
917 }
918 *sizep = size;
919 return used;
920 }
922 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
923 {
924 unsigned long size, used;
925 static const char valid_loose_object_type[8] = {
926 0, /* OBJ_EXT */
927 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
928 0, /* "delta" and others are invalid in a loose object */
929 };
930 enum object_type type;
932 /* Get the data stream */
933 memset(stream, 0, sizeof(*stream));
934 stream->next_in = map;
935 stream->avail_in = mapsize;
936 stream->next_out = buffer;
937 stream->avail_out = bufsiz;
939 if (legacy_loose_object(map)) {
940 inflateInit(stream);
941 return inflate(stream, 0);
942 }
944 used = unpack_object_header_gently(map, mapsize, &type, &size);
945 if (!used || !valid_loose_object_type[type])
946 return -1;
947 map += used;
948 mapsize -= used;
950 /* Set up the stream for the rest.. */
951 stream->next_in = map;
952 stream->avail_in = mapsize;
953 inflateInit(stream);
955 /* And generate the fake traditional header */
956 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
957 type_names[type], size);
958 return 0;
959 }
961 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
962 {
963 int bytes = strlen(buffer) + 1;
964 unsigned char *buf = xmalloc(1+size);
965 unsigned long n;
967 n = stream->total_out - bytes;
968 if (n > size)
969 n = size;
970 memcpy(buf, (char *) buffer + bytes, n);
971 bytes = n;
972 if (bytes < size) {
973 stream->next_out = buf + bytes;
974 stream->avail_out = size - bytes;
975 while (inflate(stream, Z_FINISH) == Z_OK)
976 /* nothing */;
977 }
978 buf[size] = 0;
979 inflateEnd(stream);
980 return buf;
981 }
983 /*
984 * We used to just use "sscanf()", but that's actually way
985 * too permissive for what we want to check. So do an anal
986 * object header parse by hand.
987 */
988 static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
989 {
990 int i;
991 unsigned long size;
993 /*
994 * The type can be at most ten bytes (including the
995 * terminating '\0' that we add), and is followed by
996 * a space.
997 */
998 i = 10;
999 for (;;) {
1000 char c = *hdr++;
1001 if (c == ' ')
1002 break;
1003 if (!--i)
1004 return -1;
1005 *type++ = c;
1006 }
1007 *type = 0;
1009 /*
1010 * The length must follow immediately, and be in canonical
1011 * decimal format (ie "010" is not valid).
1012 */
1013 size = *hdr++ - '0';
1014 if (size > 9)
1015 return -1;
1016 if (size) {
1017 for (;;) {
1018 unsigned long c = *hdr - '0';
1019 if (c > 9)
1020 break;
1021 hdr++;
1022 size = size * 10 + c;
1023 }
1024 }
1025 *sizep = size;
1027 /*
1028 * The length must be followed by a zero byte
1029 */
1030 return *hdr ? -1 : 0;
1031 }
1033 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
1034 {
1035 int ret;
1036 z_stream stream;
1037 char hdr[8192];
1039 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1040 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
1041 return NULL;
1043 return unpack_sha1_rest(&stream, hdr, *size);
1044 }
1046 static unsigned long get_delta_base(struct packed_git *p,
1047 struct pack_window **w_curs,
1048 unsigned long offset,
1049 enum object_type kind,
1050 unsigned long delta_obj_offset,
1051 unsigned long *base_obj_offset)
1052 {
1053 unsigned char *base_info = use_pack(p, w_curs, offset, NULL);
1054 unsigned long base_offset;
1056 /* use_pack() assured us we have [base_info, base_info + 20)
1057 * as a range that we can look at without walking off the
1058 * end of the mapped window. Its actually the hash size
1059 * that is assured. An OFS_DELTA longer than the hash size
1060 * is stupid, as then a REF_DELTA would be smaller to store.
1061 */
1062 if (kind == OBJ_OFS_DELTA) {
1063 unsigned used = 0;
1064 unsigned char c = base_info[used++];
1065 base_offset = c & 127;
1066 while (c & 128) {
1067 base_offset += 1;
1068 if (!base_offset || base_offset & ~(~0UL >> 7))
1069 die("offset value overflow for delta base object");
1070 c = base_info[used++];
1071 base_offset = (base_offset << 7) + (c & 127);
1072 }
1073 base_offset = delta_obj_offset - base_offset;
1074 if (base_offset >= delta_obj_offset)
1075 die("delta base offset out of bound");
1076 offset += used;
1077 } else if (kind == OBJ_REF_DELTA) {
1078 /* The base entry _must_ be in the same pack */
1079 base_offset = find_pack_entry_one(base_info, p);
1080 if (!base_offset)
1081 die("failed to find delta-pack base object %s",
1082 sha1_to_hex(base_info));
1083 offset += 20;
1084 } else
1085 die("I am totally screwed");
1086 *base_obj_offset = base_offset;
1087 return offset;
1088 }
1090 /* forward declaration for a mutually recursive function */
1091 static int packed_object_info(struct packed_git *p, unsigned long offset,
1092 char *type, unsigned long *sizep);
1094 static int packed_delta_info(struct packed_git *p,
1095 struct pack_window **w_curs,
1096 unsigned long offset,
1097 enum object_type kind,
1098 unsigned long obj_offset,
1099 char *type,
1100 unsigned long *sizep)
1101 {
1102 unsigned long base_offset;
1104 offset = get_delta_base(p, w_curs, offset, kind,
1105 obj_offset, &base_offset);
1107 /* We choose to only get the type of the base object and
1108 * ignore potentially corrupt pack file that expects the delta
1109 * based on a base with a wrong size. This saves tons of
1110 * inflate() calls.
1111 */
1112 if (packed_object_info(p, base_offset, type, NULL))
1113 die("cannot get info for delta-pack base");
1115 if (sizep) {
1116 const unsigned char *data;
1117 unsigned char delta_head[20], *in;
1118 unsigned long result_size;
1119 z_stream stream;
1120 int st;
1122 memset(&stream, 0, sizeof(stream));
1123 stream.next_out = delta_head;
1124 stream.avail_out = sizeof(delta_head);
1126 inflateInit(&stream);
1127 do {
1128 in = use_pack(p, w_curs, offset, &stream.avail_in);
1129 stream.next_in = in;
1130 st = inflate(&stream, Z_FINISH);
1131 offset += stream.next_in - in;
1132 } while ((st == Z_OK || st == Z_BUF_ERROR)
1133 && stream.total_out < sizeof(delta_head));
1134 inflateEnd(&stream);
1135 if ((st != Z_STREAM_END) &&
1136 stream.total_out != sizeof(delta_head))
1137 die("delta data unpack-initial failed");
1139 /* Examine the initial part of the delta to figure out
1140 * the result size.
1141 */
1142 data = delta_head;
1144 /* ignore base size */
1145 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1147 /* Read the result size */
1148 result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1149 *sizep = result_size;
1150 }
1151 return 0;
1152 }
1154 static unsigned long unpack_object_header(struct packed_git *p,
1155 struct pack_window **w_curs,
1156 unsigned long offset,
1157 enum object_type *type,
1158 unsigned long *sizep)
1159 {
1160 unsigned char *base;
1161 unsigned int left;
1162 unsigned long used;
1164 /* use_pack() assures us we have [base, base + 20) available
1165 * as a range that we can look at at. (Its actually the hash
1166 * size that is assured.) With our object header encoding
1167 * the maximum deflated object size is 2^137, which is just
1168 * insane, so we know won't exceed what we have been given.
1169 */
1170 base = use_pack(p, w_curs, offset, &left);
1171 used = unpack_object_header_gently(base, left, type, sizep);
1172 if (!used)
1173 die("object offset outside of pack file");
1175 return offset + used;
1176 }
1178 void packed_object_info_detail(struct packed_git *p,
1179 unsigned long offset,
1180 char *type,
1181 unsigned long *size,
1182 unsigned long *store_size,
1183 unsigned int *delta_chain_length,
1184 unsigned char *base_sha1)
1185 {
1186 struct pack_window *w_curs = NULL;
1187 unsigned long obj_offset, val;
1188 unsigned char *next_sha1;
1189 enum object_type kind;
1191 *delta_chain_length = 0;
1192 obj_offset = offset;
1193 offset = unpack_object_header(p, &w_curs, offset, &kind, size);
1195 for (;;) {
1196 switch (kind) {
1197 default:
1198 die("pack %s contains unknown object type %d",
1199 p->pack_name, kind);
1200 case OBJ_COMMIT:
1201 case OBJ_TREE:
1202 case OBJ_BLOB:
1203 case OBJ_TAG:
1204 strcpy(type, type_names[kind]);
1205 *store_size = 0; /* notyet */
1206 unuse_pack(&w_curs);
1207 return;
1208 case OBJ_OFS_DELTA:
1209 get_delta_base(p, &w_curs, offset, kind,
1210 obj_offset, &offset);
1211 if (*delta_chain_length == 0) {
1212 /* TODO: find base_sha1 as pointed by offset */
1213 }
1214 break;
1215 case OBJ_REF_DELTA:
1216 next_sha1 = use_pack(p, &w_curs, offset, NULL);
1217 if (*delta_chain_length == 0)
1218 hashcpy(base_sha1, next_sha1);
1219 offset = find_pack_entry_one(next_sha1, p);
1220 break;
1221 }
1222 obj_offset = offset;
1223 offset = unpack_object_header(p, &w_curs, offset, &kind, &val);
1224 (*delta_chain_length)++;
1225 }
1226 }
1228 static int packed_object_info(struct packed_git *p, unsigned long offset,
1229 char *type, unsigned long *sizep)
1230 {
1231 struct pack_window *w_curs = NULL;
1232 unsigned long size, obj_offset = offset;
1233 enum object_type kind;
1234 int r;
1236 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1238 switch (kind) {
1239 case OBJ_OFS_DELTA:
1240 case OBJ_REF_DELTA:
1241 r = packed_delta_info(p, &w_curs, offset, kind,
1242 obj_offset, type, sizep);
1243 unuse_pack(&w_curs);
1244 return r;
1245 case OBJ_COMMIT:
1246 case OBJ_TREE:
1247 case OBJ_BLOB:
1248 case OBJ_TAG:
1249 strcpy(type, type_names[kind]);
1250 unuse_pack(&w_curs);
1251 break;
1252 default:
1253 die("pack %s contains unknown object type %d",
1254 p->pack_name, kind);
1255 }
1256 if (sizep)
1257 *sizep = size;
1258 return 0;
1259 }
1261 static void *unpack_compressed_entry(struct packed_git *p,
1262 struct pack_window **w_curs,
1263 unsigned long offset,
1264 unsigned long size)
1265 {
1266 int st;
1267 z_stream stream;
1268 unsigned char *buffer, *in;
1270 buffer = xmalloc(size + 1);
1271 buffer[size] = 0;
1272 memset(&stream, 0, sizeof(stream));
1273 stream.next_out = buffer;
1274 stream.avail_out = size;
1276 inflateInit(&stream);
1277 do {
1278 in = use_pack(p, w_curs, offset, &stream.avail_in);
1279 stream.next_in = in;
1280 st = inflate(&stream, Z_FINISH);
1281 offset += stream.next_in - in;
1282 } while (st == Z_OK || st == Z_BUF_ERROR);
1283 inflateEnd(&stream);
1284 if ((st != Z_STREAM_END) || stream.total_out != size) {
1285 free(buffer);
1286 return NULL;
1287 }
1289 return buffer;
1290 }
1292 static void *unpack_delta_entry(struct packed_git *p,
1293 struct pack_window **w_curs,
1294 unsigned long offset,
1295 unsigned long delta_size,
1296 enum object_type kind,
1297 unsigned long obj_offset,
1298 char *type,
1299 unsigned long *sizep)
1300 {
1301 void *delta_data, *result, *base;
1302 unsigned long result_size, base_size, base_offset;
1304 offset = get_delta_base(p, w_curs, offset, kind,
1305 obj_offset, &base_offset);
1306 base = unpack_entry(p, base_offset, type, &base_size);
1307 if (!base)
1308 die("failed to read delta base object at %lu from %s",
1309 base_offset, p->pack_name);
1311 delta_data = unpack_compressed_entry(p, w_curs, offset, delta_size);
1312 result = patch_delta(base, base_size,
1313 delta_data, delta_size,
1314 &result_size);
1315 if (!result)
1316 die("failed to apply delta");
1317 free(delta_data);
1318 free(base);
1319 *sizep = result_size;
1320 return result;
1321 }
1323 void *unpack_entry(struct packed_git *p, unsigned long offset,
1324 char *type, unsigned long *sizep)
1325 {
1326 struct pack_window *w_curs = NULL;
1327 unsigned long size, obj_offset = offset;
1328 enum object_type kind;
1329 void *retval;
1331 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1332 switch (kind) {
1333 case OBJ_OFS_DELTA:
1334 case OBJ_REF_DELTA:
1335 retval = unpack_delta_entry(p, &w_curs, offset, size,
1336 kind, obj_offset, type, sizep);
1337 break;
1338 case OBJ_COMMIT:
1339 case OBJ_TREE:
1340 case OBJ_BLOB:
1341 case OBJ_TAG:
1342 strcpy(type, type_names[kind]);
1343 *sizep = size;
1344 retval = unpack_compressed_entry(p, &w_curs, offset, size);
1345 break;
1346 default:
1347 die("unknown object type %i in %s", kind, p->pack_name);
1348 }
1349 unuse_pack(&w_curs);
1350 return retval;
1351 }
1353 int num_packed_objects(const struct packed_git *p)
1354 {
1355 /* See check_packed_git_idx() */
1356 return (p->index_size - 20 - 20 - 4*256) / 24;
1357 }
1359 int nth_packed_object_sha1(const struct packed_git *p, int n,
1360 unsigned char* sha1)
1361 {
1362 void *index = p->index_base + 256;
1363 if (n < 0 || num_packed_objects(p) <= n)
1364 return -1;
1365 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1366 return 0;
1367 }
1369 unsigned long find_pack_entry_one(const unsigned char *sha1,
1370 struct packed_git *p)
1371 {
1372 uint32_t *level1_ofs = p->index_base;
1373 int hi = ntohl(level1_ofs[*sha1]);
1374 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1375 void *index = p->index_base + 256;
1377 do {
1378 int mi = (lo + hi) / 2;
1379 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1380 if (!cmp)
1381 return ntohl(*((uint32_t *)((char *)index + (24 * mi))));
1382 if (cmp > 0)
1383 hi = mi;
1384 else
1385 lo = mi+1;
1386 } while (lo < hi);
1387 return 0;
1388 }
1390 static int matches_pack_name(struct packed_git *p, const char *ig)
1391 {
1392 const char *last_c, *c;
1394 if (!strcmp(p->pack_name, ig))
1395 return 0;
1397 for (c = p->pack_name, last_c = c; *c;)
1398 if (*c == '/')
1399 last_c = ++c;
1400 else
1401 ++c;
1402 if (!strcmp(last_c, ig))
1403 return 0;
1405 return 1;
1406 }
1408 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1409 {
1410 struct packed_git *p;
1411 unsigned long offset;
1413 prepare_packed_git();
1415 for (p = packed_git; p; p = p->next) {
1416 if (ignore_packed) {
1417 const char **ig;
1418 for (ig = ignore_packed; *ig; ig++)
1419 if (!matches_pack_name(p, *ig))
1420 break;
1421 if (*ig)
1422 continue;
1423 }
1424 offset = find_pack_entry_one(sha1, p);
1425 if (offset) {
1426 /*
1427 * We are about to tell the caller where they can
1428 * locate the requested object. We better make
1429 * sure the packfile is still here and can be
1430 * accessed before supplying that answer, as
1431 * it may have been deleted since the index
1432 * was loaded!
1433 */
1434 if (p->pack_fd == -1 && open_packed_git(p)) {
1435 error("packfile %s cannot be accessed", p->pack_name);
1436 continue;
1437 }
1438 e->offset = offset;
1439 e->p = p;
1440 hashcpy(e->sha1, sha1);
1441 return 1;
1442 }
1443 }
1444 return 0;
1445 }
1447 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1448 struct packed_git *packs)
1449 {
1450 struct packed_git *p;
1452 for (p = packs; p; p = p->next) {
1453 if (find_pack_entry_one(sha1, p))
1454 return p;
1455 }
1456 return NULL;
1458 }
1460 static int sha1_loose_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1461 {
1462 int status;
1463 unsigned long mapsize, size;
1464 void *map;
1465 z_stream stream;
1466 char hdr[128];
1468 map = map_sha1_file(sha1, &mapsize);
1469 if (!map)
1470 return error("unable to find %s", sha1_to_hex(sha1));
1471 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1472 status = error("unable to unpack %s header",
1473 sha1_to_hex(sha1));
1474 if (parse_sha1_header(hdr, type, &size) < 0)
1475 status = error("unable to parse %s header", sha1_to_hex(sha1));
1476 else {
1477 status = 0;
1478 if (sizep)
1479 *sizep = size;
1480 }
1481 inflateEnd(&stream);
1482 munmap(map, mapsize);
1483 return status;
1484 }
1486 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1487 {
1488 struct pack_entry e;
1490 if (!find_pack_entry(sha1, &e, NULL)) {
1491 reprepare_packed_git();
1492 if (!find_pack_entry(sha1, &e, NULL))
1493 return sha1_loose_object_info(sha1, type, sizep);
1494 }
1495 return packed_object_info(e.p, e.offset, type, sizep);
1496 }
1498 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1499 {
1500 struct pack_entry e;
1502 if (!find_pack_entry(sha1, &e, NULL))
1503 return NULL;
1504 else
1505 return unpack_entry(e.p, e.offset, type, size);
1506 }
1508 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1509 {
1510 unsigned long mapsize;
1511 void *map, *buf;
1513 buf = read_packed_sha1(sha1, type, size);
1514 if (buf)
1515 return buf;
1516 map = map_sha1_file(sha1, &mapsize);
1517 if (map) {
1518 buf = unpack_sha1_file(map, mapsize, type, size);
1519 munmap(map, mapsize);
1520 return buf;
1521 }
1522 reprepare_packed_git();
1523 return read_packed_sha1(sha1, type, size);
1524 }
1526 void *read_object_with_reference(const unsigned char *sha1,
1527 const char *required_type,
1528 unsigned long *size,
1529 unsigned char *actual_sha1_return)
1530 {
1531 char type[20];
1532 void *buffer;
1533 unsigned long isize;
1534 unsigned char actual_sha1[20];
1536 hashcpy(actual_sha1, sha1);
1537 while (1) {
1538 int ref_length = -1;
1539 const char *ref_type = NULL;
1541 buffer = read_sha1_file(actual_sha1, type, &isize);
1542 if (!buffer)
1543 return NULL;
1544 if (!strcmp(type, required_type)) {
1545 *size = isize;
1546 if (actual_sha1_return)
1547 hashcpy(actual_sha1_return, actual_sha1);
1548 return buffer;
1549 }
1550 /* Handle references */
1551 else if (!strcmp(type, commit_type))
1552 ref_type = "tree ";
1553 else if (!strcmp(type, tag_type))
1554 ref_type = "object ";
1555 else {
1556 free(buffer);
1557 return NULL;
1558 }
1559 ref_length = strlen(ref_type);
1561 if (memcmp(buffer, ref_type, ref_length) ||
1562 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1563 free(buffer);
1564 return NULL;
1565 }
1566 free(buffer);
1567 /* Now we have the ID of the referred-to object in
1568 * actual_sha1. Check again. */
1569 }
1570 }
1572 static void write_sha1_file_prepare(void *buf, unsigned long len,
1573 const char *type, unsigned char *sha1,
1574 unsigned char *hdr, int *hdrlen)
1575 {
1576 SHA_CTX c;
1578 /* Generate the header */
1579 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1581 /* Sha1.. */
1582 SHA1_Init(&c);
1583 SHA1_Update(&c, hdr, *hdrlen);
1584 SHA1_Update(&c, buf, len);
1585 SHA1_Final(sha1, &c);
1586 }
1588 /*
1589 * Link the tempfile to the final place, possibly creating the
1590 * last directory level as you do so.
1591 *
1592 * Returns the errno on failure, 0 on success.
1593 */
1594 static int link_temp_to_file(const char *tmpfile, const char *filename)
1595 {
1596 int ret;
1597 char *dir;
1599 if (!link(tmpfile, filename))
1600 return 0;
1602 /*
1603 * Try to mkdir the last path component if that failed.
1604 *
1605 * Re-try the "link()" regardless of whether the mkdir
1606 * succeeds, since a race might mean that somebody
1607 * else succeeded.
1608 */
1609 ret = errno;
1610 dir = strrchr(filename, '/');
1611 if (dir) {
1612 *dir = 0;
1613 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1614 *dir = '/';
1615 return -2;
1616 }
1617 *dir = '/';
1618 if (!link(tmpfile, filename))
1619 return 0;
1620 ret = errno;
1621 }
1622 return ret;
1623 }
1625 /*
1626 * Move the just written object into its final resting place
1627 */
1628 int move_temp_to_file(const char *tmpfile, const char *filename)
1629 {
1630 int ret = link_temp_to_file(tmpfile, filename);
1632 /*
1633 * Coda hack - coda doesn't like cross-directory links,
1634 * so we fall back to a rename, which will mean that it
1635 * won't be able to check collisions, but that's not a
1636 * big deal.
1637 *
1638 * The same holds for FAT formatted media.
1639 *
1640 * When this succeeds, we just return 0. We have nothing
1641 * left to unlink.
1642 */
1643 if (ret && ret != EEXIST) {
1644 if (!rename(tmpfile, filename))
1645 return 0;
1646 ret = errno;
1647 }
1648 unlink(tmpfile);
1649 if (ret) {
1650 if (ret != EEXIST) {
1651 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1652 }
1653 /* FIXME!!! Collision check here ? */
1654 }
1656 return 0;
1657 }
1659 static int write_buffer(int fd, const void *buf, size_t len)
1660 {
1661 if (write_in_full(fd, buf, len) < 0)
1662 return error("file write error (%s)", strerror(errno));
1663 return 0;
1664 }
1666 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1667 {
1668 int hdr_len;
1669 unsigned char c;
1671 c = (type << 4) | (len & 15);
1672 len >>= 4;
1673 hdr_len = 1;
1674 while (len) {
1675 *hdr++ = c | 0x80;
1676 hdr_len++;
1677 c = (len & 0x7f);
1678 len >>= 7;
1679 }
1680 *hdr = c;
1681 return hdr_len;
1682 }
1684 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1685 {
1686 int obj_type, hdr;
1688 if (use_legacy_headers) {
1689 while (deflate(stream, 0) == Z_OK)
1690 /* nothing */;
1691 return;
1692 }
1693 if (!strcmp(type, blob_type))
1694 obj_type = OBJ_BLOB;
1695 else if (!strcmp(type, tree_type))
1696 obj_type = OBJ_TREE;
1697 else if (!strcmp(type, commit_type))
1698 obj_type = OBJ_COMMIT;
1699 else if (!strcmp(type, tag_type))
1700 obj_type = OBJ_TAG;
1701 else
1702 die("trying to generate bogus object of type '%s'", type);
1703 hdr = write_binary_header(stream->next_out, obj_type, len);
1704 stream->total_out = hdr;
1705 stream->next_out += hdr;
1706 stream->avail_out -= hdr;
1707 }
1709 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1710 unsigned char *sha1)
1711 {
1712 unsigned char hdr[50];
1713 int hdrlen;
1714 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1715 return 0;
1716 }
1718 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1719 {
1720 int size;
1721 unsigned char *compressed;
1722 z_stream stream;
1723 unsigned char sha1[20];
1724 char *filename;
1725 static char tmpfile[PATH_MAX];
1726 unsigned char hdr[50];
1727 int fd, hdrlen;
1729 /* Normally if we have it in the pack then we do not bother writing
1730 * it out into .git/objects/??/?{38} file.
1731 */
1732 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1733 filename = sha1_file_name(sha1);
1734 if (returnsha1)
1735 hashcpy(returnsha1, sha1);
1736 if (has_sha1_file(sha1))
1737 return 0;
1738 fd = open(filename, O_RDONLY);
1739 if (fd >= 0) {
1740 /*
1741 * FIXME!!! We might do collision checking here, but we'd
1742 * need to uncompress the old file and check it. Later.
1743 */
1744 close(fd);
1745 return 0;
1746 }
1748 if (errno != ENOENT) {
1749 return error("sha1 file %s: %s\n", filename, strerror(errno));
1750 }
1752 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1754 fd = mkstemp(tmpfile);
1755 if (fd < 0) {
1756 if (errno == EPERM)
1757 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1758 else
1759 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1760 }
1762 /* Set it up */
1763 memset(&stream, 0, sizeof(stream));
1764 deflateInit(&stream, zlib_compression_level);
1765 size = 8 + deflateBound(&stream, len+hdrlen);
1766 compressed = xmalloc(size);
1768 /* Compress it */
1769 stream.next_out = compressed;
1770 stream.avail_out = size;
1772 /* First header.. */
1773 stream.next_in = hdr;
1774 stream.avail_in = hdrlen;
1775 setup_object_header(&stream, type, len);
1777 /* Then the data itself.. */
1778 stream.next_in = buf;
1779 stream.avail_in = len;
1780 while (deflate(&stream, Z_FINISH) == Z_OK)
1781 /* nothing */;
1782 deflateEnd(&stream);
1783 size = stream.total_out;
1785 if (write_buffer(fd, compressed, size) < 0)
1786 die("unable to write sha1 file");
1787 fchmod(fd, 0444);
1788 close(fd);
1789 free(compressed);
1791 return move_temp_to_file(tmpfile, filename);
1792 }
1794 /*
1795 * We need to unpack and recompress the object for writing
1796 * it out to a different file.
1797 */
1798 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1799 {
1800 size_t size;
1801 z_stream stream;
1802 unsigned char *unpacked;
1803 unsigned long len;
1804 char type[20];
1805 char hdr[50];
1806 int hdrlen;
1807 void *buf;
1809 /* need to unpack and recompress it by itself */
1810 unpacked = read_packed_sha1(sha1, type, &len);
1811 if (!unpacked)
1812 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1814 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1816 /* Set it up */
1817 memset(&stream, 0, sizeof(stream));
1818 deflateInit(&stream, zlib_compression_level);
1819 size = deflateBound(&stream, len + hdrlen);
1820 buf = xmalloc(size);
1822 /* Compress it */
1823 stream.next_out = buf;
1824 stream.avail_out = size;
1826 /* First header.. */
1827 stream.next_in = (void *)hdr;
1828 stream.avail_in = hdrlen;
1829 while (deflate(&stream, 0) == Z_OK)
1830 /* nothing */;
1832 /* Then the data itself.. */
1833 stream.next_in = unpacked;
1834 stream.avail_in = len;
1835 while (deflate(&stream, Z_FINISH) == Z_OK)
1836 /* nothing */;
1837 deflateEnd(&stream);
1838 free(unpacked);
1840 *objsize = stream.total_out;
1841 return buf;
1842 }
1844 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1845 {
1846 int retval;
1847 unsigned long objsize;
1848 void *buf = map_sha1_file(sha1, &objsize);
1850 if (buf) {
1851 retval = write_buffer(fd, buf, objsize);
1852 munmap(buf, objsize);
1853 return retval;
1854 }
1856 buf = repack_object(sha1, &objsize);
1857 retval = write_buffer(fd, buf, objsize);
1858 free(buf);
1859 return retval;
1860 }
1862 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1863 size_t bufsize, size_t *bufposn)
1864 {
1865 char tmpfile[PATH_MAX];
1866 int local;
1867 z_stream stream;
1868 unsigned char real_sha1[20];
1869 unsigned char discard[4096];
1870 int ret;
1871 SHA_CTX c;
1873 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1875 local = mkstemp(tmpfile);
1876 if (local < 0) {
1877 if (errno == EPERM)
1878 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1879 else
1880 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1881 }
1883 memset(&stream, 0, sizeof(stream));
1885 inflateInit(&stream);
1887 SHA1_Init(&c);
1889 do {
1890 ssize_t size;
1891 if (*bufposn) {
1892 stream.avail_in = *bufposn;
1893 stream.next_in = (unsigned char *) buffer;
1894 do {
1895 stream.next_out = discard;
1896 stream.avail_out = sizeof(discard);
1897 ret = inflate(&stream, Z_SYNC_FLUSH);
1898 SHA1_Update(&c, discard, sizeof(discard) -
1899 stream.avail_out);
1900 } while (stream.avail_in && ret == Z_OK);
1901 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1902 die("unable to write sha1 file");
1903 memmove(buffer, buffer + *bufposn - stream.avail_in,
1904 stream.avail_in);
1905 *bufposn = stream.avail_in;
1906 if (ret != Z_OK)
1907 break;
1908 }
1909 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
1910 if (size <= 0) {
1911 close(local);
1912 unlink(tmpfile);
1913 if (!size)
1914 return error("Connection closed?");
1915 perror("Reading from connection");
1916 return -1;
1917 }
1918 *bufposn += size;
1919 } while (1);
1920 inflateEnd(&stream);
1922 close(local);
1923 SHA1_Final(real_sha1, &c);
1924 if (ret != Z_STREAM_END) {
1925 unlink(tmpfile);
1926 return error("File %s corrupted", sha1_to_hex(sha1));
1927 }
1928 if (hashcmp(sha1, real_sha1)) {
1929 unlink(tmpfile);
1930 return error("File %s has bad hash", sha1_to_hex(sha1));
1931 }
1933 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1934 }
1936 int has_pack_index(const unsigned char *sha1)
1937 {
1938 struct stat st;
1939 if (stat(sha1_pack_index_name(sha1), &st))
1940 return 0;
1941 return 1;
1942 }
1944 int has_pack_file(const unsigned char *sha1)
1945 {
1946 struct stat st;
1947 if (stat(sha1_pack_name(sha1), &st))
1948 return 0;
1949 return 1;
1950 }
1952 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1953 {
1954 struct pack_entry e;
1955 return find_pack_entry(sha1, &e, ignore_packed);
1956 }
1958 int has_sha1_file(const unsigned char *sha1)
1959 {
1960 struct stat st;
1961 struct pack_entry e;
1963 if (find_pack_entry(sha1, &e, NULL))
1964 return 1;
1965 return find_sha1_file(sha1, &st) ? 1 : 0;
1966 }
1968 /*
1969 * reads from fd as long as possible into a supplied buffer of size bytes.
1970 * If necessary the buffer's size is increased using realloc()
1971 *
1972 * returns 0 if anything went fine and -1 otherwise
1973 *
1974 * NOTE: both buf and size may change, but even when -1 is returned
1975 * you still have to free() it yourself.
1976 */
1977 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
1978 {
1979 char* buf = *return_buf;
1980 unsigned long size = *return_size;
1981 int iret;
1982 unsigned long off = 0;
1984 do {
1985 iret = xread(fd, buf + off, size - off);
1986 if (iret > 0) {
1987 off += iret;
1988 if (off == size) {
1989 size *= 2;
1990 buf = xrealloc(buf, size);
1991 }
1992 }
1993 } while (iret > 0);
1995 *return_buf = buf;
1996 *return_size = off;
1998 if (iret < 0)
1999 return -1;
2000 return 0;
2001 }
2003 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2004 {
2005 unsigned long size = 4096;
2006 char *buf = xmalloc(size);
2007 int ret;
2009 if (read_pipe(fd, &buf, &size)) {
2010 free(buf);
2011 return -1;
2012 }
2014 if (!type)
2015 type = blob_type;
2016 if (write_object)
2017 ret = write_sha1_file(buf, size, type, sha1);
2018 else
2019 ret = hash_sha1_file(buf, size, type, sha1);
2020 free(buf);
2021 return ret;
2022 }
2024 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
2025 {
2026 unsigned long size = st->st_size;
2027 void *buf;
2028 int ret;
2030 buf = "";
2031 if (size)
2032 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2033 close(fd);
2035 if (!type)
2036 type = blob_type;
2037 if (write_object)
2038 ret = write_sha1_file(buf, size, type, sha1);
2039 else
2040 ret = hash_sha1_file(buf, size, type, sha1);
2041 if (size)
2042 munmap(buf, size);
2043 return ret;
2044 }
2046 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2047 {
2048 int fd;
2049 char *target;
2051 switch (st->st_mode & S_IFMT) {
2052 case S_IFREG:
2053 fd = open(path, O_RDONLY);
2054 if (fd < 0)
2055 return error("open(\"%s\"): %s", path,
2056 strerror(errno));
2057 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
2058 return error("%s: failed to insert into database",
2059 path);
2060 break;
2061 case S_IFLNK:
2062 target = xmalloc(st->st_size+1);
2063 if (readlink(path, target, st->st_size+1) != st->st_size) {
2064 char *errstr = strerror(errno);
2065 free(target);
2066 return error("readlink(\"%s\"): %s", path,
2067 errstr);
2068 }
2069 if (!write_object)
2070 hash_sha1_file(target, st->st_size, blob_type, sha1);
2071 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
2072 return error("%s: failed to insert into database",
2073 path);
2074 free(target);
2075 break;
2076 default:
2077 return error("%s: unsupported file type", path);
2078 }
2079 return 0;
2080 }
2082 int read_pack_header(int fd, struct pack_header *header)
2083 {
2084 char *c = (char*)header;
2085 ssize_t remaining = sizeof(struct pack_header);
2086 do {
2087 ssize_t r = xread(fd, c, remaining);
2088 if (r <= 0)
2089 /* "eof before pack header was fully read" */
2090 return PH_ERROR_EOF;
2091 remaining -= r;
2092 c += r;
2093 } while (remaining > 0);
2094 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2095 /* "protocol error (pack signature mismatch detected)" */
2096 return PH_ERROR_PACK_SIGNATURE;
2097 if (!pack_version_ok(header->hdr_version))
2098 /* "protocol error (pack version unsupported)" */
2099 return PH_ERROR_PROTOCOL;
2100 return 0;
2101 }