convert object type handling from a string to a number
[git/git.git] / sha1_file.c
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"
16
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
24
25 #ifdef NO_C99_FORMAT
26 #define SZ_FMT "lu"
27 #else
28 #define SZ_FMT "zu"
29 #endif
30
31 const unsigned char null_sha1[20];
32
33 static unsigned int sha1_file_open_flag = O_NOATIME;
34
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 };
69
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 }
82
83 int safe_create_leading_directories(char *path)
84 {
85 char *pos = path;
86 struct stat st;
87
88 if (*pos == '/')
89 pos++;
90
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 }
115
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;
123
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';
130
131 return buffer;
132 }
133
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 }
145
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;
159
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 }
173
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;
179
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 }
187
188 buf = name;
189
190 for (i = 0; i < 20; i++) {
191 unsigned int val = *sha1++;
192 *buf++ = hex[val >> 4];
193 *buf++ = hex[val & 0xf];
194 }
195
196 return base;
197 }
198
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;
204
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 }
212
213 buf = name;
214
215 for (i = 0; i < 20; i++) {
216 unsigned int val = *sha1++;
217 *buf++ = hex[val >> 4];
218 *buf++ = hex[val & 0xf];
219 }
220
221 return base;
222 }
223
224 struct alternate_object_database *alt_odb_list;
225 static struct alternate_object_database **alt_odb_tail;
226
227 static void read_info_alternates(const char * alternates, int depth);
228
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;
254
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);
263
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);
271
272 ent->name = ent->base + pfxlen + 1;
273 ent->base[pfxlen + 3] = '/';
274 ent->base[pfxlen] = ent->base[entlen-1] = 0;
275
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 }
284
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 }
298
299 /* add the alternate entry */
300 *alt_odb_tail = ent;
301 alt_odb_tail = &(ent->next);
302 ent->next = NULL;
303
304 /* recursively add alternates */
305 read_info_alternates(ent->base, depth + 1);
306
307 ent->base[pfxlen] = '/';
308
309 return 0;
310 }
311
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;
316
317 if (depth > 5) {
318 error("%s: ignoring alternate object stores, nesting too deep.",
319 relative_base);
320 return;
321 }
322
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 }
348
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;
355
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);
366
367 link_alt_odb_entries(map, map + st.st_size, '\n', relative_base, depth);
368
369 munmap(map, st.st_size);
370 }
371
372 void prepare_alt_odb(void)
373 {
374 const char *alt;
375
376 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
377 if (!alt) alt = "";
378
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);
383
384 read_info_alternates(get_object_directory(), 0);
385 }
386
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;
391
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 }
403
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 struct packed_git *packed_git;
411
412 void pack_report()
413 {
414 fprintf(stderr,
415 "pack_report: getpagesize() = %10" SZ_FMT "\n"
416 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
417 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
418 (size_t) getpagesize(),
419 packed_git_window_size,
420 packed_git_limit);
421 fprintf(stderr,
422 "pack_report: pack_used_ctr = %10u\n"
423 "pack_report: pack_mmap_calls = %10u\n"
424 "pack_report: pack_open_windows = %10u / %10u\n"
425 "pack_report: pack_mapped = "
426 "%10" SZ_FMT " / %10" SZ_FMT "\n",
427 pack_used_ctr,
428 pack_mmap_calls,
429 pack_open_windows, peak_pack_open_windows,
430 pack_mapped, peak_pack_mapped);
431 }
432
433 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
434 void **idx_map_)
435 {
436 void *idx_map;
437 uint32_t *index;
438 unsigned long idx_size;
439 int nr, i;
440 int fd = open(path, O_RDONLY);
441 struct stat st;
442 if (fd < 0)
443 return -1;
444 if (fstat(fd, &st)) {
445 close(fd);
446 return -1;
447 }
448 idx_size = st.st_size;
449 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
450 close(fd);
451
452 index = idx_map;
453 *idx_map_ = idx_map;
454 *idx_size_ = idx_size;
455
456 /* check index map */
457 if (idx_size < 4*256 + 20 + 20)
458 return error("index file %s is too small", path);
459
460 /* a future index format would start with this, as older git
461 * binaries would fail the non-monotonic index check below.
462 * give a nicer warning to the user if we can.
463 */
464 if (index[0] == htonl(PACK_IDX_SIGNATURE))
465 return error("index file %s is a newer version"
466 " and is not supported by this binary"
467 " (try upgrading GIT to a newer version)",
468 path);
469
470 nr = 0;
471 for (i = 0; i < 256; i++) {
472 unsigned int n = ntohl(index[i]);
473 if (n < nr)
474 return error("non-monotonic index %s", path);
475 nr = n;
476 }
477
478 /*
479 * Total size:
480 * - 256 index entries 4 bytes each
481 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
482 * - 20-byte SHA1 of the packfile
483 * - 20-byte SHA1 file checksum
484 */
485 if (idx_size != 4*256 + nr * 24 + 20 + 20)
486 return error("wrong index file size in %s", path);
487
488 return 0;
489 }
490
491 static void scan_windows(struct packed_git *p,
492 struct packed_git **lru_p,
493 struct pack_window **lru_w,
494 struct pack_window **lru_l)
495 {
496 struct pack_window *w, *w_l;
497
498 for (w_l = NULL, w = p->windows; w; w = w->next) {
499 if (!w->inuse_cnt) {
500 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
501 *lru_p = p;
502 *lru_w = w;
503 *lru_l = w_l;
504 }
505 }
506 w_l = w;
507 }
508 }
509
510 static int unuse_one_window(struct packed_git *current)
511 {
512 struct packed_git *p, *lru_p = NULL;
513 struct pack_window *lru_w = NULL, *lru_l = NULL;
514
515 if (current)
516 scan_windows(current, &lru_p, &lru_w, &lru_l);
517 for (p = packed_git; p; p = p->next)
518 scan_windows(p, &lru_p, &lru_w, &lru_l);
519 if (lru_p) {
520 munmap(lru_w->base, lru_w->len);
521 pack_mapped -= lru_w->len;
522 if (lru_l)
523 lru_l->next = lru_w->next;
524 else {
525 lru_p->windows = lru_w->next;
526 if (!lru_p->windows && lru_p != current) {
527 close(lru_p->pack_fd);
528 lru_p->pack_fd = -1;
529 }
530 }
531 free(lru_w);
532 pack_open_windows--;
533 return 1;
534 }
535 return 0;
536 }
537
538 void release_pack_memory(size_t need)
539 {
540 size_t cur = pack_mapped;
541 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
542 ; /* nothing */
543 }
544
545 void unuse_pack(struct pack_window **w_cursor)
546 {
547 struct pack_window *w = *w_cursor;
548 if (w) {
549 w->inuse_cnt--;
550 *w_cursor = NULL;
551 }
552 }
553
554 /*
555 * Do not call this directly as this leaks p->pack_fd on error return;
556 * call open_packed_git() instead.
557 */
558 static int open_packed_git_1(struct packed_git *p)
559 {
560 struct stat st;
561 struct pack_header hdr;
562 unsigned char sha1[20];
563 unsigned char *idx_sha1;
564 long fd_flag;
565
566 p->pack_fd = open(p->pack_name, O_RDONLY);
567 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
568 return -1;
569
570 /* If we created the struct before we had the pack we lack size. */
571 if (!p->pack_size) {
572 if (!S_ISREG(st.st_mode))
573 return error("packfile %s not a regular file", p->pack_name);
574 p->pack_size = st.st_size;
575 } else if (p->pack_size != st.st_size)
576 return error("packfile %s size changed", p->pack_name);
577
578 /* We leave these file descriptors open with sliding mmap;
579 * there is no point keeping them open across exec(), though.
580 */
581 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
582 if (fd_flag < 0)
583 return error("cannot determine file descriptor flags");
584 fd_flag |= FD_CLOEXEC;
585 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
586 return error("cannot set FD_CLOEXEC");
587
588 /* Verify we recognize this pack file format. */
589 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
590 return error("file %s is far too short to be a packfile", p->pack_name);
591 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
592 return error("file %s is not a GIT packfile", p->pack_name);
593 if (!pack_version_ok(hdr.hdr_version))
594 return error("packfile %s is version %u and not supported"
595 " (try upgrading GIT to a newer version)",
596 p->pack_name, ntohl(hdr.hdr_version));
597
598 /* Verify the pack matches its index. */
599 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
600 return error("packfile %s claims to have %u objects"
601 " while index size indicates %u objects",
602 p->pack_name, ntohl(hdr.hdr_entries),
603 num_packed_objects(p));
604 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
605 return error("end of packfile %s is unavailable", p->pack_name);
606 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
607 return error("packfile %s signature is unavailable", p->pack_name);
608 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
609 if (hashcmp(sha1, idx_sha1))
610 return error("packfile %s does not match index", p->pack_name);
611 return 0;
612 }
613
614 static int open_packed_git(struct packed_git *p)
615 {
616 if (!open_packed_git_1(p))
617 return 0;
618 if (p->pack_fd != -1) {
619 close(p->pack_fd);
620 p->pack_fd = -1;
621 }
622 return -1;
623 }
624
625 static int in_window(struct pack_window *win, unsigned long offset)
626 {
627 /* We must promise at least 20 bytes (one hash) after the
628 * offset is available from this window, otherwise the offset
629 * is not actually in this window and a different window (which
630 * has that one hash excess) must be used. This is to support
631 * the object header and delta base parsing routines below.
632 */
633 off_t win_off = win->offset;
634 return win_off <= offset
635 && (offset + 20) <= (win_off + win->len);
636 }
637
638 unsigned char* use_pack(struct packed_git *p,
639 struct pack_window **w_cursor,
640 unsigned long offset,
641 unsigned int *left)
642 {
643 struct pack_window *win = *w_cursor;
644
645 if (p->pack_fd == -1 && open_packed_git(p))
646 die("packfile %s cannot be accessed", p->pack_name);
647
648 /* Since packfiles end in a hash of their content and its
649 * pointless to ask for an offset into the middle of that
650 * hash, and the in_window function above wouldn't match
651 * don't allow an offset too close to the end of the file.
652 */
653 if (offset > (p->pack_size - 20))
654 die("offset beyond end of packfile (truncated pack?)");
655
656 if (!win || !in_window(win, offset)) {
657 if (win)
658 win->inuse_cnt--;
659 for (win = p->windows; win; win = win->next) {
660 if (in_window(win, offset))
661 break;
662 }
663 if (!win) {
664 size_t window_align = packed_git_window_size / 2;
665 win = xcalloc(1, sizeof(*win));
666 win->offset = (offset / window_align) * window_align;
667 win->len = p->pack_size - win->offset;
668 if (win->len > packed_git_window_size)
669 win->len = packed_git_window_size;
670 pack_mapped += win->len;
671 while (packed_git_limit < pack_mapped
672 && unuse_one_window(p))
673 ; /* nothing */
674 win->base = xmmap(NULL, win->len,
675 PROT_READ, MAP_PRIVATE,
676 p->pack_fd, win->offset);
677 if (win->base == MAP_FAILED)
678 die("packfile %s cannot be mapped: %s",
679 p->pack_name,
680 strerror(errno));
681 pack_mmap_calls++;
682 pack_open_windows++;
683 if (pack_mapped > peak_pack_mapped)
684 peak_pack_mapped = pack_mapped;
685 if (pack_open_windows > peak_pack_open_windows)
686 peak_pack_open_windows = pack_open_windows;
687 win->next = p->windows;
688 p->windows = win;
689 }
690 }
691 if (win != *w_cursor) {
692 win->last_used = pack_used_ctr++;
693 win->inuse_cnt++;
694 *w_cursor = win;
695 }
696 offset -= win->offset;
697 if (left)
698 *left = win->len - offset;
699 return win->base + offset;
700 }
701
702 struct packed_git *add_packed_git(char *path, int path_len, int local)
703 {
704 struct stat st;
705 struct packed_git *p;
706 unsigned long idx_size;
707 void *idx_map;
708 unsigned char sha1[20];
709
710 if (check_packed_git_idx(path, &idx_size, &idx_map))
711 return NULL;
712
713 /* do we have a corresponding .pack file? */
714 strcpy(path + path_len - 4, ".pack");
715 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
716 munmap(idx_map, idx_size);
717 return NULL;
718 }
719 /* ok, it looks sane as far as we can check without
720 * actually mapping the pack file.
721 */
722 p = xmalloc(sizeof(*p) + path_len + 2);
723 strcpy(p->pack_name, path);
724 p->index_size = idx_size;
725 p->pack_size = st.st_size;
726 p->index_base = idx_map;
727 p->next = NULL;
728 p->windows = NULL;
729 p->pack_fd = -1;
730 p->pack_local = local;
731 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
732 hashcpy(p->sha1, sha1);
733 return p;
734 }
735
736 struct packed_git *parse_pack_index(unsigned char *sha1)
737 {
738 char *path = sha1_pack_index_name(sha1);
739 return parse_pack_index_file(sha1, path);
740 }
741
742 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
743 {
744 struct packed_git *p;
745 unsigned long idx_size;
746 void *idx_map;
747 char *path;
748
749 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
750 return NULL;
751
752 path = sha1_pack_name(sha1);
753
754 p = xmalloc(sizeof(*p) + strlen(path) + 2);
755 strcpy(p->pack_name, path);
756 p->index_size = idx_size;
757 p->pack_size = 0;
758 p->index_base = idx_map;
759 p->next = NULL;
760 p->windows = NULL;
761 p->pack_fd = -1;
762 hashcpy(p->sha1, sha1);
763 return p;
764 }
765
766 void install_packed_git(struct packed_git *pack)
767 {
768 pack->next = packed_git;
769 packed_git = pack;
770 }
771
772 static void prepare_packed_git_one(char *objdir, int local)
773 {
774 char path[PATH_MAX];
775 int len;
776 DIR *dir;
777 struct dirent *de;
778
779 sprintf(path, "%s/pack", objdir);
780 len = strlen(path);
781 dir = opendir(path);
782 if (!dir) {
783 if (errno != ENOENT)
784 error("unable to open object pack directory: %s: %s",
785 path, strerror(errno));
786 return;
787 }
788 path[len++] = '/';
789 while ((de = readdir(dir)) != NULL) {
790 int namelen = strlen(de->d_name);
791 struct packed_git *p;
792
793 if (!has_extension(de->d_name, ".idx"))
794 continue;
795
796 /* Don't reopen a pack we already have. */
797 strcpy(path + len, de->d_name);
798 for (p = packed_git; p; p = p->next) {
799 if (!memcmp(path, p->pack_name, len + namelen - 4))
800 break;
801 }
802 if (p)
803 continue;
804 /* See if it really is a valid .idx file with corresponding
805 * .pack file that we can map.
806 */
807 p = add_packed_git(path, len + namelen, local);
808 if (!p)
809 continue;
810 install_packed_git(p);
811 }
812 closedir(dir);
813 }
814
815 static int prepare_packed_git_run_once = 0;
816 void prepare_packed_git(void)
817 {
818 struct alternate_object_database *alt;
819
820 if (prepare_packed_git_run_once)
821 return;
822 prepare_packed_git_one(get_object_directory(), 1);
823 prepare_alt_odb();
824 for (alt = alt_odb_list; alt; alt = alt->next) {
825 alt->name[-1] = 0;
826 prepare_packed_git_one(alt->base, 0);
827 alt->name[-1] = '/';
828 }
829 prepare_packed_git_run_once = 1;
830 }
831
832 void reprepare_packed_git(void)
833 {
834 prepare_packed_git_run_once = 0;
835 prepare_packed_git();
836 }
837
838 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
839 {
840 unsigned char real_sha1[20];
841 hash_sha1_file(map, size, type, real_sha1);
842 return hashcmp(sha1, real_sha1) ? -1 : 0;
843 }
844
845 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
846 {
847 struct stat st;
848 void *map;
849 int fd;
850 char *filename = find_sha1_file(sha1, &st);
851
852 if (!filename) {
853 return NULL;
854 }
855
856 fd = open(filename, O_RDONLY | sha1_file_open_flag);
857 if (fd < 0) {
858 /* See if it works without O_NOATIME */
859 switch (sha1_file_open_flag) {
860 default:
861 fd = open(filename, O_RDONLY);
862 if (fd >= 0)
863 break;
864 /* Fallthrough */
865 case 0:
866 return NULL;
867 }
868
869 /* If it failed once, it will probably fail again.
870 * Stop using O_NOATIME
871 */
872 sha1_file_open_flag = 0;
873 }
874 map = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
875 close(fd);
876 *size = st.st_size;
877 return map;
878 }
879
880 int legacy_loose_object(unsigned char *map)
881 {
882 unsigned int word;
883
884 /*
885 * Is it a zlib-compressed buffer? If so, the first byte
886 * must be 0x78 (15-bit window size, deflated), and the
887 * first 16-bit word is evenly divisible by 31
888 */
889 word = (map[0] << 8) + map[1];
890 if (map[0] == 0x78 && !(word % 31))
891 return 1;
892 else
893 return 0;
894 }
895
896 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
897 {
898 unsigned shift;
899 unsigned char c;
900 unsigned long size;
901 unsigned long used = 0;
902
903 c = buf[used++];
904 *type = (c >> 4) & 7;
905 size = c & 15;
906 shift = 4;
907 while (c & 0x80) {
908 if (len <= used)
909 return 0;
910 if (sizeof(long) * 8 <= shift)
911 return 0;
912 c = buf[used++];
913 size += (c & 0x7f) << shift;
914 shift += 7;
915 }
916 *sizep = size;
917 return used;
918 }
919
920 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
921 {
922 unsigned long size, used;
923 static const char valid_loose_object_type[8] = {
924 0, /* OBJ_EXT */
925 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
926 0, /* "delta" and others are invalid in a loose object */
927 };
928 enum object_type type;
929
930 /* Get the data stream */
931 memset(stream, 0, sizeof(*stream));
932 stream->next_in = map;
933 stream->avail_in = mapsize;
934 stream->next_out = buffer;
935 stream->avail_out = bufsiz;
936
937 if (legacy_loose_object(map)) {
938 inflateInit(stream);
939 return inflate(stream, 0);
940 }
941
942 used = unpack_object_header_gently(map, mapsize, &type, &size);
943 if (!used || !valid_loose_object_type[type])
944 return -1;
945 map += used;
946 mapsize -= used;
947
948 /* Set up the stream for the rest.. */
949 stream->next_in = map;
950 stream->avail_in = mapsize;
951 inflateInit(stream);
952
953 /* And generate the fake traditional header */
954 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
955 typename(type), size);
956 return 0;
957 }
958
959 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
960 {
961 int bytes = strlen(buffer) + 1;
962 unsigned char *buf = xmalloc(1+size);
963 unsigned long n;
964
965 n = stream->total_out - bytes;
966 if (n > size)
967 n = size;
968 memcpy(buf, (char *) buffer + bytes, n);
969 bytes = n;
970 if (bytes < size) {
971 stream->next_out = buf + bytes;
972 stream->avail_out = size - bytes;
973 while (inflate(stream, Z_FINISH) == Z_OK)
974 /* nothing */;
975 }
976 buf[size] = 0;
977 inflateEnd(stream);
978 return buf;
979 }
980
981 /*
982 * We used to just use "sscanf()", but that's actually way
983 * too permissive for what we want to check. So do an anal
984 * object header parse by hand.
985 */
986 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
987 {
988 char type[10];
989 int i;
990 unsigned long size;
991
992 /*
993 * The type can be at most ten bytes (including the
994 * terminating '\0' that we add), and is followed by
995 * a space.
996 */
997 i = 0;
998 for (;;) {
999 char c = *hdr++;
1000 if (c == ' ')
1001 break;
1002 type[i++] = c;
1003 if (i >= sizeof(type))
1004 return -1;
1005 }
1006 type[i] = 0;
1007
1008 /*
1009 * The length must follow immediately, and be in canonical
1010 * decimal format (ie "010" is not valid).
1011 */
1012 size = *hdr++ - '0';
1013 if (size > 9)
1014 return -1;
1015 if (size) {
1016 for (;;) {
1017 unsigned long c = *hdr - '0';
1018 if (c > 9)
1019 break;
1020 hdr++;
1021 size = size * 10 + c;
1022 }
1023 }
1024 *sizep = size;
1025
1026 /*
1027 * The length must be followed by a zero byte
1028 */
1029 return *hdr ? -1 : type_from_string(type);
1030 }
1031
1032 void * unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size)
1033 {
1034 int ret;
1035 z_stream stream;
1036 char hdr[8192];
1037
1038 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1039 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1040 return NULL;
1041
1042 return unpack_sha1_rest(&stream, hdr, *size);
1043 }
1044
1045 static unsigned long get_delta_base(struct packed_git *p,
1046 struct pack_window **w_curs,
1047 unsigned long *curpos,
1048 enum object_type type,
1049 unsigned long delta_obj_offset)
1050 {
1051 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1052 unsigned long base_offset;
1053
1054 /* use_pack() assured us we have [base_info, base_info + 20)
1055 * as a range that we can look at without walking off the
1056 * end of the mapped window. Its actually the hash size
1057 * that is assured. An OFS_DELTA longer than the hash size
1058 * is stupid, as then a REF_DELTA would be smaller to store.
1059 */
1060 if (type == OBJ_OFS_DELTA) {
1061 unsigned used = 0;
1062 unsigned char c = base_info[used++];
1063 base_offset = c & 127;
1064 while (c & 128) {
1065 base_offset += 1;
1066 if (!base_offset || base_offset & ~(~0UL >> 7))
1067 die("offset value overflow for delta base object");
1068 c = base_info[used++];
1069 base_offset = (base_offset << 7) + (c & 127);
1070 }
1071 base_offset = delta_obj_offset - base_offset;
1072 if (base_offset >= delta_obj_offset)
1073 die("delta base offset out of bound");
1074 *curpos += used;
1075 } else if (type == OBJ_REF_DELTA) {
1076 /* The base entry _must_ be in the same pack */
1077 base_offset = find_pack_entry_one(base_info, p);
1078 if (!base_offset)
1079 die("failed to find delta-pack base object %s",
1080 sha1_to_hex(base_info));
1081 *curpos += 20;
1082 } else
1083 die("I am totally screwed");
1084 return base_offset;
1085 }
1086
1087 /* forward declaration for a mutually recursive function */
1088 static int packed_object_info(struct packed_git *p, unsigned long offset,
1089 unsigned long *sizep);
1090
1091 static int packed_delta_info(struct packed_git *p,
1092 struct pack_window **w_curs,
1093 unsigned long curpos,
1094 enum object_type type,
1095 unsigned long obj_offset,
1096 unsigned long *sizep)
1097 {
1098 unsigned long base_offset;
1099
1100 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1101 type = packed_object_info(p, base_offset, NULL);
1102
1103 /* We choose to only get the type of the base object and
1104 * ignore potentially corrupt pack file that expects the delta
1105 * based on a base with a wrong size. This saves tons of
1106 * inflate() calls.
1107 */
1108 if (sizep) {
1109 const unsigned char *data;
1110 unsigned char delta_head[20], *in;
1111 z_stream stream;
1112 int st;
1113
1114 memset(&stream, 0, sizeof(stream));
1115 stream.next_out = delta_head;
1116 stream.avail_out = sizeof(delta_head);
1117
1118 inflateInit(&stream);
1119 do {
1120 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1121 stream.next_in = in;
1122 st = inflate(&stream, Z_FINISH);
1123 curpos += stream.next_in - in;
1124 } while ((st == Z_OK || st == Z_BUF_ERROR)
1125 && stream.total_out < sizeof(delta_head));
1126 inflateEnd(&stream);
1127 if ((st != Z_STREAM_END) &&
1128 stream.total_out != sizeof(delta_head))
1129 die("delta data unpack-initial failed");
1130
1131 /* Examine the initial part of the delta to figure out
1132 * the result size.
1133 */
1134 data = delta_head;
1135
1136 /* ignore base size */
1137 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1138
1139 /* Read the result size */
1140 *sizep = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1141 }
1142
1143 return type;
1144 }
1145
1146 static int unpack_object_header(struct packed_git *p,
1147 struct pack_window **w_curs,
1148 unsigned long *curpos,
1149 unsigned long *sizep)
1150 {
1151 unsigned char *base;
1152 unsigned int left;
1153 unsigned long used;
1154 enum object_type type;
1155
1156 /* use_pack() assures us we have [base, base + 20) available
1157 * as a range that we can look at at. (Its actually the hash
1158 * size that is assured.) With our object header encoding
1159 * the maximum deflated object size is 2^137, which is just
1160 * insane, so we know won't exceed what we have been given.
1161 */
1162 base = use_pack(p, w_curs, *curpos, &left);
1163 used = unpack_object_header_gently(base, left, &type, sizep);
1164 if (!used)
1165 die("object offset outside of pack file");
1166 *curpos += used;
1167
1168 return type;
1169 }
1170
1171 const char *packed_object_info_detail(struct packed_git *p,
1172 unsigned long obj_offset,
1173 unsigned long *size,
1174 unsigned long *store_size,
1175 unsigned int *delta_chain_length,
1176 unsigned char *base_sha1)
1177 {
1178 struct pack_window *w_curs = NULL;
1179 unsigned long curpos, dummy;
1180 unsigned char *next_sha1;
1181 enum object_type type;
1182
1183 *delta_chain_length = 0;
1184 curpos = obj_offset;
1185 type = unpack_object_header(p, &w_curs, &curpos, size);
1186
1187 for (;;) {
1188 switch (type) {
1189 default:
1190 die("pack %s contains unknown object type %d",
1191 p->pack_name, type);
1192 case OBJ_COMMIT:
1193 case OBJ_TREE:
1194 case OBJ_BLOB:
1195 case OBJ_TAG:
1196 *store_size = 0; /* notyet */
1197 unuse_pack(&w_curs);
1198 return typename(type);
1199 case OBJ_OFS_DELTA:
1200 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1201 if (*delta_chain_length == 0) {
1202 /* TODO: find base_sha1 as pointed by curpos */
1203 }
1204 break;
1205 case OBJ_REF_DELTA:
1206 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1207 if (*delta_chain_length == 0)
1208 hashcpy(base_sha1, next_sha1);
1209 obj_offset = find_pack_entry_one(next_sha1, p);
1210 break;
1211 }
1212 (*delta_chain_length)++;
1213 curpos = obj_offset;
1214 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1215 }
1216 }
1217
1218 static int packed_object_info(struct packed_git *p, unsigned long obj_offset,
1219 unsigned long *sizep)
1220 {
1221 struct pack_window *w_curs = NULL;
1222 unsigned long size, curpos = obj_offset;
1223 enum object_type type;
1224
1225 type = unpack_object_header(p, &w_curs, &curpos, &size);
1226
1227 switch (type) {
1228 case OBJ_OFS_DELTA:
1229 case OBJ_REF_DELTA:
1230 type = packed_delta_info(p, &w_curs, curpos,
1231 type, obj_offset, sizep);
1232 break;
1233 case OBJ_COMMIT:
1234 case OBJ_TREE:
1235 case OBJ_BLOB:
1236 case OBJ_TAG:
1237 if (sizep)
1238 *sizep = size;
1239 break;
1240 default:
1241 die("pack %s contains unknown object type %d",
1242 p->pack_name, type);
1243 }
1244 unuse_pack(&w_curs);
1245 return type;
1246 }
1247
1248 static void *unpack_compressed_entry(struct packed_git *p,
1249 struct pack_window **w_curs,
1250 unsigned long curpos,
1251 unsigned long size)
1252 {
1253 int st;
1254 z_stream stream;
1255 unsigned char *buffer, *in;
1256
1257 buffer = xmalloc(size + 1);
1258 buffer[size] = 0;
1259 memset(&stream, 0, sizeof(stream));
1260 stream.next_out = buffer;
1261 stream.avail_out = size;
1262
1263 inflateInit(&stream);
1264 do {
1265 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1266 stream.next_in = in;
1267 st = inflate(&stream, Z_FINISH);
1268 curpos += stream.next_in - in;
1269 } while (st == Z_OK || st == Z_BUF_ERROR);
1270 inflateEnd(&stream);
1271 if ((st != Z_STREAM_END) || stream.total_out != size) {
1272 free(buffer);
1273 return NULL;
1274 }
1275
1276 return buffer;
1277 }
1278
1279 static void *unpack_delta_entry(struct packed_git *p,
1280 struct pack_window **w_curs,
1281 unsigned long curpos,
1282 unsigned long delta_size,
1283 unsigned long obj_offset,
1284 enum object_type *type,
1285 unsigned long *sizep)
1286 {
1287 void *delta_data, *result, *base;
1288 unsigned long base_size, base_offset;
1289
1290 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1291 base = unpack_entry(p, base_offset, type, &base_size);
1292 if (!base)
1293 die("failed to read delta base object at %lu from %s",
1294 base_offset, p->pack_name);
1295
1296 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1297 result = patch_delta(base, base_size,
1298 delta_data, delta_size,
1299 sizep);
1300 if (!result)
1301 die("failed to apply delta");
1302 free(delta_data);
1303 free(base);
1304 return result;
1305 }
1306
1307 void *unpack_entry(struct packed_git *p, unsigned long obj_offset,
1308 enum object_type *type, unsigned long *sizep)
1309 {
1310 struct pack_window *w_curs = NULL;
1311 unsigned long curpos = obj_offset;
1312 void *data;
1313
1314 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1315 switch (*type) {
1316 case OBJ_OFS_DELTA:
1317 case OBJ_REF_DELTA:
1318 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1319 obj_offset, type, sizep);
1320 break;
1321 case OBJ_COMMIT:
1322 case OBJ_TREE:
1323 case OBJ_BLOB:
1324 case OBJ_TAG:
1325 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1326 break;
1327 default:
1328 die("unknown object type %i in %s", *type, p->pack_name);
1329 }
1330 unuse_pack(&w_curs);
1331 return data;
1332 }
1333
1334 int num_packed_objects(const struct packed_git *p)
1335 {
1336 /* See check_packed_git_idx() */
1337 return (p->index_size - 20 - 20 - 4*256) / 24;
1338 }
1339
1340 int nth_packed_object_sha1(const struct packed_git *p, int n,
1341 unsigned char* sha1)
1342 {
1343 void *index = p->index_base + 256;
1344 if (n < 0 || num_packed_objects(p) <= n)
1345 return -1;
1346 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1347 return 0;
1348 }
1349
1350 unsigned long find_pack_entry_one(const unsigned char *sha1,
1351 struct packed_git *p)
1352 {
1353 uint32_t *level1_ofs = p->index_base;
1354 int hi = ntohl(level1_ofs[*sha1]);
1355 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1356 void *index = p->index_base + 256;
1357
1358 do {
1359 int mi = (lo + hi) / 2;
1360 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1361 if (!cmp)
1362 return ntohl(*((uint32_t *)((char *)index + (24 * mi))));
1363 if (cmp > 0)
1364 hi = mi;
1365 else
1366 lo = mi+1;
1367 } while (lo < hi);
1368 return 0;
1369 }
1370
1371 static int matches_pack_name(struct packed_git *p, const char *ig)
1372 {
1373 const char *last_c, *c;
1374
1375 if (!strcmp(p->pack_name, ig))
1376 return 0;
1377
1378 for (c = p->pack_name, last_c = c; *c;)
1379 if (*c == '/')
1380 last_c = ++c;
1381 else
1382 ++c;
1383 if (!strcmp(last_c, ig))
1384 return 0;
1385
1386 return 1;
1387 }
1388
1389 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1390 {
1391 struct packed_git *p;
1392 unsigned long offset;
1393
1394 prepare_packed_git();
1395
1396 for (p = packed_git; p; p = p->next) {
1397 if (ignore_packed) {
1398 const char **ig;
1399 for (ig = ignore_packed; *ig; ig++)
1400 if (!matches_pack_name(p, *ig))
1401 break;
1402 if (*ig)
1403 continue;
1404 }
1405 offset = find_pack_entry_one(sha1, p);
1406 if (offset) {
1407 /*
1408 * We are about to tell the caller where they can
1409 * locate the requested object. We better make
1410 * sure the packfile is still here and can be
1411 * accessed before supplying that answer, as
1412 * it may have been deleted since the index
1413 * was loaded!
1414 */
1415 if (p->pack_fd == -1 && open_packed_git(p)) {
1416 error("packfile %s cannot be accessed", p->pack_name);
1417 continue;
1418 }
1419 e->offset = offset;
1420 e->p = p;
1421 hashcpy(e->sha1, sha1);
1422 return 1;
1423 }
1424 }
1425 return 0;
1426 }
1427
1428 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1429 struct packed_git *packs)
1430 {
1431 struct packed_git *p;
1432
1433 for (p = packs; p; p = p->next) {
1434 if (find_pack_entry_one(sha1, p))
1435 return p;
1436 }
1437 return NULL;
1438
1439 }
1440
1441 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1442 {
1443 int status;
1444 unsigned long mapsize, size;
1445 void *map;
1446 z_stream stream;
1447 char hdr[32];
1448
1449 map = map_sha1_file(sha1, &mapsize);
1450 if (!map)
1451 return error("unable to find %s", sha1_to_hex(sha1));
1452 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1453 status = error("unable to unpack %s header",
1454 sha1_to_hex(sha1));
1455 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1456 status = error("unable to parse %s header", sha1_to_hex(sha1));
1457 else if (sizep)
1458 *sizep = size;
1459 inflateEnd(&stream);
1460 munmap(map, mapsize);
1461 return status;
1462 }
1463
1464 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1465 {
1466 struct pack_entry e;
1467
1468 if (!find_pack_entry(sha1, &e, NULL)) {
1469 reprepare_packed_git();
1470 if (!find_pack_entry(sha1, &e, NULL))
1471 return sha1_loose_object_info(sha1, sizep);
1472 }
1473 return packed_object_info(e.p, e.offset, sizep);
1474 }
1475
1476 static void *read_packed_sha1(const unsigned char *sha1,
1477 enum object_type *type, unsigned long *size)
1478 {
1479 struct pack_entry e;
1480
1481 if (!find_pack_entry(sha1, &e, NULL))
1482 return NULL;
1483 else
1484 return unpack_entry(e.p, e.offset, type, size);
1485 }
1486
1487 /*
1488 * This is meant to hold a *small* number of objects that you would
1489 * want read_sha1_file() to be able to return, but yet you do not want
1490 * to write them into the object store (e.g. a browse-only
1491 * application).
1492 */
1493 static struct cached_object {
1494 unsigned char sha1[20];
1495 enum object_type type;
1496 void *buf;
1497 unsigned long size;
1498 } *cached_objects;
1499 static int cached_object_nr, cached_object_alloc;
1500
1501 static struct cached_object *find_cached_object(const unsigned char *sha1)
1502 {
1503 int i;
1504 struct cached_object *co = cached_objects;
1505
1506 for (i = 0; i < cached_object_nr; i++, co++) {
1507 if (!hashcmp(co->sha1, sha1))
1508 return co;
1509 }
1510 return NULL;
1511 }
1512
1513 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1514 unsigned char *sha1)
1515 {
1516 struct cached_object *co;
1517
1518 hash_sha1_file(buf, len, typename(type), sha1);
1519 if (has_sha1_file(sha1) || find_cached_object(sha1))
1520 return 0;
1521 if (cached_object_alloc <= cached_object_nr) {
1522 cached_object_alloc = alloc_nr(cached_object_alloc);
1523 cached_objects = xrealloc(cached_objects,
1524 sizeof(*cached_objects) *
1525 cached_object_alloc);
1526 }
1527 co = &cached_objects[cached_object_nr++];
1528 co->size = len;
1529 co->type = type;
1530 co->buf = xmalloc(len);
1531 memcpy(co->buf, buf, len);
1532 hashcpy(co->sha1, sha1);
1533 return 0;
1534 }
1535
1536 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1537 unsigned long *size)
1538 {
1539 unsigned long mapsize;
1540 void *map, *buf;
1541 struct cached_object *co;
1542
1543 co = find_cached_object(sha1);
1544 if (co) {
1545 buf = xmalloc(co->size + 1);
1546 memcpy(buf, co->buf, co->size);
1547 ((char*)buf)[co->size] = 0;
1548 *type = co->type;
1549 *size = co->size;
1550 return buf;
1551 }
1552
1553 buf = read_packed_sha1(sha1, type, size);
1554 if (buf)
1555 return buf;
1556 map = map_sha1_file(sha1, &mapsize);
1557 if (map) {
1558 buf = unpack_sha1_file(map, mapsize, type, size);
1559 munmap(map, mapsize);
1560 return buf;
1561 }
1562 reprepare_packed_git();
1563 return read_packed_sha1(sha1, type, size);
1564 }
1565
1566 void *read_object_with_reference(const unsigned char *sha1,
1567 const char *required_type_name,
1568 unsigned long *size,
1569 unsigned char *actual_sha1_return)
1570 {
1571 enum object_type type, required_type;
1572 void *buffer;
1573 unsigned long isize;
1574 unsigned char actual_sha1[20];
1575
1576 required_type = type_from_string(required_type_name);
1577 hashcpy(actual_sha1, sha1);
1578 while (1) {
1579 int ref_length = -1;
1580 const char *ref_type = NULL;
1581
1582 buffer = read_sha1_file(actual_sha1, &type, &isize);
1583 if (!buffer)
1584 return NULL;
1585 if (type == required_type) {
1586 *size = isize;
1587 if (actual_sha1_return)
1588 hashcpy(actual_sha1_return, actual_sha1);
1589 return buffer;
1590 }
1591 /* Handle references */
1592 else if (type == OBJ_COMMIT)
1593 ref_type = "tree ";
1594 else if (type == OBJ_TAG)
1595 ref_type = "object ";
1596 else {
1597 free(buffer);
1598 return NULL;
1599 }
1600 ref_length = strlen(ref_type);
1601
1602 if (memcmp(buffer, ref_type, ref_length) ||
1603 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1604 free(buffer);
1605 return NULL;
1606 }
1607 free(buffer);
1608 /* Now we have the ID of the referred-to object in
1609 * actual_sha1. Check again. */
1610 }
1611 }
1612
1613 static void write_sha1_file_prepare(void *buf, unsigned long len,
1614 const char *type, unsigned char *sha1,
1615 char *hdr, int *hdrlen)
1616 {
1617 SHA_CTX c;
1618
1619 /* Generate the header */
1620 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
1621
1622 /* Sha1.. */
1623 SHA1_Init(&c);
1624 SHA1_Update(&c, hdr, *hdrlen);
1625 SHA1_Update(&c, buf, len);
1626 SHA1_Final(sha1, &c);
1627 }
1628
1629 /*
1630 * Link the tempfile to the final place, possibly creating the
1631 * last directory level as you do so.
1632 *
1633 * Returns the errno on failure, 0 on success.
1634 */
1635 static int link_temp_to_file(const char *tmpfile, const char *filename)
1636 {
1637 int ret;
1638 char *dir;
1639
1640 if (!link(tmpfile, filename))
1641 return 0;
1642
1643 /*
1644 * Try to mkdir the last path component if that failed.
1645 *
1646 * Re-try the "link()" regardless of whether the mkdir
1647 * succeeds, since a race might mean that somebody
1648 * else succeeded.
1649 */
1650 ret = errno;
1651 dir = strrchr(filename, '/');
1652 if (dir) {
1653 *dir = 0;
1654 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1655 *dir = '/';
1656 return -2;
1657 }
1658 *dir = '/';
1659 if (!link(tmpfile, filename))
1660 return 0;
1661 ret = errno;
1662 }
1663 return ret;
1664 }
1665
1666 /*
1667 * Move the just written object into its final resting place
1668 */
1669 int move_temp_to_file(const char *tmpfile, const char *filename)
1670 {
1671 int ret = link_temp_to_file(tmpfile, filename);
1672
1673 /*
1674 * Coda hack - coda doesn't like cross-directory links,
1675 * so we fall back to a rename, which will mean that it
1676 * won't be able to check collisions, but that's not a
1677 * big deal.
1678 *
1679 * The same holds for FAT formatted media.
1680 *
1681 * When this succeeds, we just return 0. We have nothing
1682 * left to unlink.
1683 */
1684 if (ret && ret != EEXIST) {
1685 if (!rename(tmpfile, filename))
1686 return 0;
1687 ret = errno;
1688 }
1689 unlink(tmpfile);
1690 if (ret) {
1691 if (ret != EEXIST) {
1692 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1693 }
1694 /* FIXME!!! Collision check here ? */
1695 }
1696
1697 return 0;
1698 }
1699
1700 static int write_buffer(int fd, const void *buf, size_t len)
1701 {
1702 if (write_in_full(fd, buf, len) < 0)
1703 return error("file write error (%s)", strerror(errno));
1704 return 0;
1705 }
1706
1707 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1708 {
1709 int hdr_len;
1710 unsigned char c;
1711
1712 c = (type << 4) | (len & 15);
1713 len >>= 4;
1714 hdr_len = 1;
1715 while (len) {
1716 *hdr++ = c | 0x80;
1717 hdr_len++;
1718 c = (len & 0x7f);
1719 len >>= 7;
1720 }
1721 *hdr = c;
1722 return hdr_len;
1723 }
1724
1725 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1726 {
1727 int obj_type, hdrlen;
1728
1729 if (use_legacy_headers) {
1730 while (deflate(stream, 0) == Z_OK)
1731 /* nothing */;
1732 return;
1733 }
1734 obj_type = type_from_string(type);
1735 hdrlen = write_binary_header(stream->next_out, obj_type, len);
1736 stream->total_out = hdrlen;
1737 stream->next_out += hdrlen;
1738 stream->avail_out -= hdrlen;
1739 }
1740
1741 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1742 unsigned char *sha1)
1743 {
1744 char hdr[32];
1745 int hdrlen;
1746 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1747 return 0;
1748 }
1749
1750 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1751 {
1752 int size;
1753 unsigned char *compressed;
1754 z_stream stream;
1755 unsigned char sha1[20];
1756 char *filename;
1757 static char tmpfile[PATH_MAX];
1758 char hdr[32];
1759 int fd, hdrlen;
1760
1761 /* Normally if we have it in the pack then we do not bother writing
1762 * it out into .git/objects/??/?{38} file.
1763 */
1764 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1765 filename = sha1_file_name(sha1);
1766 if (returnsha1)
1767 hashcpy(returnsha1, sha1);
1768 if (has_sha1_file(sha1))
1769 return 0;
1770 fd = open(filename, O_RDONLY);
1771 if (fd >= 0) {
1772 /*
1773 * FIXME!!! We might do collision checking here, but we'd
1774 * need to uncompress the old file and check it. Later.
1775 */
1776 close(fd);
1777 return 0;
1778 }
1779
1780 if (errno != ENOENT) {
1781 return error("sha1 file %s: %s\n", filename, strerror(errno));
1782 }
1783
1784 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1785
1786 fd = mkstemp(tmpfile);
1787 if (fd < 0) {
1788 if (errno == EPERM)
1789 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1790 else
1791 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1792 }
1793
1794 /* Set it up */
1795 memset(&stream, 0, sizeof(stream));
1796 deflateInit(&stream, zlib_compression_level);
1797 size = 8 + deflateBound(&stream, len+hdrlen);
1798 compressed = xmalloc(size);
1799
1800 /* Compress it */
1801 stream.next_out = compressed;
1802 stream.avail_out = size;
1803
1804 /* First header.. */
1805 stream.next_in = (unsigned char *)hdr;
1806 stream.avail_in = hdrlen;
1807 setup_object_header(&stream, type, len);
1808
1809 /* Then the data itself.. */
1810 stream.next_in = buf;
1811 stream.avail_in = len;
1812 while (deflate(&stream, Z_FINISH) == Z_OK)
1813 /* nothing */;
1814 deflateEnd(&stream);
1815 size = stream.total_out;
1816
1817 if (write_buffer(fd, compressed, size) < 0)
1818 die("unable to write sha1 file");
1819 fchmod(fd, 0444);
1820 close(fd);
1821 free(compressed);
1822
1823 return move_temp_to_file(tmpfile, filename);
1824 }
1825
1826 /*
1827 * We need to unpack and recompress the object for writing
1828 * it out to a different file.
1829 */
1830 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1831 {
1832 size_t size;
1833 z_stream stream;
1834 unsigned char *unpacked;
1835 unsigned long len;
1836 enum object_type type;
1837 char hdr[32];
1838 int hdrlen;
1839 void *buf;
1840
1841 /* need to unpack and recompress it by itself */
1842 unpacked = read_packed_sha1(sha1, &type, &len);
1843 if (!unpacked)
1844 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1845
1846 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1847
1848 /* Set it up */
1849 memset(&stream, 0, sizeof(stream));
1850 deflateInit(&stream, zlib_compression_level);
1851 size = deflateBound(&stream, len + hdrlen);
1852 buf = xmalloc(size);
1853
1854 /* Compress it */
1855 stream.next_out = buf;
1856 stream.avail_out = size;
1857
1858 /* First header.. */
1859 stream.next_in = (void *)hdr;
1860 stream.avail_in = hdrlen;
1861 while (deflate(&stream, 0) == Z_OK)
1862 /* nothing */;
1863
1864 /* Then the data itself.. */
1865 stream.next_in = unpacked;
1866 stream.avail_in = len;
1867 while (deflate(&stream, Z_FINISH) == Z_OK)
1868 /* nothing */;
1869 deflateEnd(&stream);
1870 free(unpacked);
1871
1872 *objsize = stream.total_out;
1873 return buf;
1874 }
1875
1876 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1877 {
1878 int retval;
1879 unsigned long objsize;
1880 void *buf = map_sha1_file(sha1, &objsize);
1881
1882 if (buf) {
1883 retval = write_buffer(fd, buf, objsize);
1884 munmap(buf, objsize);
1885 return retval;
1886 }
1887
1888 buf = repack_object(sha1, &objsize);
1889 retval = write_buffer(fd, buf, objsize);
1890 free(buf);
1891 return retval;
1892 }
1893
1894 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1895 size_t bufsize, size_t *bufposn)
1896 {
1897 char tmpfile[PATH_MAX];
1898 int local;
1899 z_stream stream;
1900 unsigned char real_sha1[20];
1901 unsigned char discard[4096];
1902 int ret;
1903 SHA_CTX c;
1904
1905 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1906
1907 local = mkstemp(tmpfile);
1908 if (local < 0) {
1909 if (errno == EPERM)
1910 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1911 else
1912 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1913 }
1914
1915 memset(&stream, 0, sizeof(stream));
1916
1917 inflateInit(&stream);
1918
1919 SHA1_Init(&c);
1920
1921 do {
1922 ssize_t size;
1923 if (*bufposn) {
1924 stream.avail_in = *bufposn;
1925 stream.next_in = (unsigned char *) buffer;
1926 do {
1927 stream.next_out = discard;
1928 stream.avail_out = sizeof(discard);
1929 ret = inflate(&stream, Z_SYNC_FLUSH);
1930 SHA1_Update(&c, discard, sizeof(discard) -
1931 stream.avail_out);
1932 } while (stream.avail_in && ret == Z_OK);
1933 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1934 die("unable to write sha1 file");
1935 memmove(buffer, buffer + *bufposn - stream.avail_in,
1936 stream.avail_in);
1937 *bufposn = stream.avail_in;
1938 if (ret != Z_OK)
1939 break;
1940 }
1941 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
1942 if (size <= 0) {
1943 close(local);
1944 unlink(tmpfile);
1945 if (!size)
1946 return error("Connection closed?");
1947 perror("Reading from connection");
1948 return -1;
1949 }
1950 *bufposn += size;
1951 } while (1);
1952 inflateEnd(&stream);
1953
1954 close(local);
1955 SHA1_Final(real_sha1, &c);
1956 if (ret != Z_STREAM_END) {
1957 unlink(tmpfile);
1958 return error("File %s corrupted", sha1_to_hex(sha1));
1959 }
1960 if (hashcmp(sha1, real_sha1)) {
1961 unlink(tmpfile);
1962 return error("File %s has bad hash", sha1_to_hex(sha1));
1963 }
1964
1965 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1966 }
1967
1968 int has_pack_index(const unsigned char *sha1)
1969 {
1970 struct stat st;
1971 if (stat(sha1_pack_index_name(sha1), &st))
1972 return 0;
1973 return 1;
1974 }
1975
1976 int has_pack_file(const unsigned char *sha1)
1977 {
1978 struct stat st;
1979 if (stat(sha1_pack_name(sha1), &st))
1980 return 0;
1981 return 1;
1982 }
1983
1984 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1985 {
1986 struct pack_entry e;
1987 return find_pack_entry(sha1, &e, ignore_packed);
1988 }
1989
1990 int has_sha1_file(const unsigned char *sha1)
1991 {
1992 struct stat st;
1993 struct pack_entry e;
1994
1995 if (find_pack_entry(sha1, &e, NULL))
1996 return 1;
1997 return find_sha1_file(sha1, &st) ? 1 : 0;
1998 }
1999
2000 /*
2001 * reads from fd as long as possible into a supplied buffer of size bytes.
2002 * If necessary the buffer's size is increased using realloc()
2003 *
2004 * returns 0 if anything went fine and -1 otherwise
2005 *
2006 * NOTE: both buf and size may change, but even when -1 is returned
2007 * you still have to free() it yourself.
2008 */
2009 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
2010 {
2011 char* buf = *return_buf;
2012 unsigned long size = *return_size;
2013 int iret;
2014 unsigned long off = 0;
2015
2016 do {
2017 iret = xread(fd, buf + off, size - off);
2018 if (iret > 0) {
2019 off += iret;
2020 if (off == size) {
2021 size *= 2;
2022 buf = xrealloc(buf, size);
2023 }
2024 }
2025 } while (iret > 0);
2026
2027 *return_buf = buf;
2028 *return_size = off;
2029
2030 if (iret < 0)
2031 return -1;
2032 return 0;
2033 }
2034
2035 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2036 {
2037 unsigned long size = 4096;
2038 char *buf = xmalloc(size);
2039 int ret;
2040
2041 if (read_pipe(fd, &buf, &size)) {
2042 free(buf);
2043 return -1;
2044 }
2045
2046 if (!type)
2047 type = blob_type;
2048 if (write_object)
2049 ret = write_sha1_file(buf, size, type, sha1);
2050 else
2051 ret = hash_sha1_file(buf, size, type, sha1);
2052 free(buf);
2053 return ret;
2054 }
2055
2056 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
2057 {
2058 unsigned long size = st->st_size;
2059 void *buf;
2060 int ret, re_allocated = 0;
2061
2062 buf = "";
2063 if (size)
2064 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2065 close(fd);
2066
2067 if (!type)
2068 type = blob_type;
2069
2070 /*
2071 * Convert blobs to git internal format
2072 */
2073 if (!strcmp(type, blob_type)) {
2074 unsigned long nsize = size;
2075 char *nbuf = buf;
2076 if (convert_to_git(NULL, &nbuf, &nsize)) {
2077 if (size)
2078 munmap(buf, size);
2079 size = nsize;
2080 buf = nbuf;
2081 re_allocated = 1;
2082 }
2083 }
2084
2085 if (write_object)
2086 ret = write_sha1_file(buf, size, type, sha1);
2087 else
2088 ret = hash_sha1_file(buf, size, type, sha1);
2089 if (re_allocated) {
2090 free(buf);
2091 return ret;
2092 }
2093 if (size)
2094 munmap(buf, size);
2095 return ret;
2096 }
2097
2098 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2099 {
2100 int fd;
2101 char *target;
2102
2103 switch (st->st_mode & S_IFMT) {
2104 case S_IFREG:
2105 fd = open(path, O_RDONLY);
2106 if (fd < 0)
2107 return error("open(\"%s\"): %s", path,
2108 strerror(errno));
2109 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
2110 return error("%s: failed to insert into database",
2111 path);
2112 break;
2113 case S_IFLNK:
2114 target = xmalloc(st->st_size+1);
2115 if (readlink(path, target, st->st_size+1) != st->st_size) {
2116 char *errstr = strerror(errno);
2117 free(target);
2118 return error("readlink(\"%s\"): %s", path,
2119 errstr);
2120 }
2121 if (!write_object)
2122 hash_sha1_file(target, st->st_size, blob_type, sha1);
2123 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
2124 return error("%s: failed to insert into database",
2125 path);
2126 free(target);
2127 break;
2128 default:
2129 return error("%s: unsupported file type", path);
2130 }
2131 return 0;
2132 }
2133
2134 int read_pack_header(int fd, struct pack_header *header)
2135 {
2136 char *c = (char*)header;
2137 ssize_t remaining = sizeof(struct pack_header);
2138 do {
2139 ssize_t r = xread(fd, c, remaining);
2140 if (r <= 0)
2141 /* "eof before pack header was fully read" */
2142 return PH_ERROR_EOF;
2143 remaining -= r;
2144 c += r;
2145 } while (remaining > 0);
2146 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2147 /* "protocol error (pack signature mismatch detected)" */
2148 return PH_ERROR_PACK_SIGNATURE;
2149 if (!pack_version_ok(header->hdr_version))
2150 /* "protocol error (pack version unsupported)" */
2151 return PH_ERROR_PROTOCOL;
2152 return 0;
2153 }