Use off_t when we really mean a file offset.
[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,
434 unsigned long *idx_size_,
435 void **idx_map_)
436 {
437 void *idx_map;
438 uint32_t *index;
439 unsigned long idx_size;
440 uint32_t 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 if (idx_size < 4 * 256 + 20 + 20) {
451 close(fd);
452 return error("index file %s is too small", path);
453 }
454 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
455 close(fd);
456
457 index = idx_map;
458 *idx_map_ = idx_map;
459 *idx_size_ = idx_size;
460
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 munmap(idx_map, idx_size);
467 return error("index file %s is a newer version"
468 " and is not supported by this binary"
469 " (try upgrading GIT to a newer version)",
470 path);
471 }
472
473 nr = 0;
474 for (i = 0; i < 256; i++) {
475 uint32_t n = ntohl(index[i]);
476 if (n < nr) {
477 munmap(idx_map, idx_size);
478 return error("non-monotonic index %s", path);
479 }
480 nr = n;
481 }
482
483 /*
484 * Total size:
485 * - 256 index entries 4 bytes each
486 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
487 * - 20-byte SHA1 of the packfile
488 * - 20-byte SHA1 file checksum
489 */
490 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
491 munmap(idx_map, idx_size);
492 return error("wrong index file size in %s", path);
493 }
494
495 return 0;
496 }
497
498 static void scan_windows(struct packed_git *p,
499 struct packed_git **lru_p,
500 struct pack_window **lru_w,
501 struct pack_window **lru_l)
502 {
503 struct pack_window *w, *w_l;
504
505 for (w_l = NULL, w = p->windows; w; w = w->next) {
506 if (!w->inuse_cnt) {
507 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
508 *lru_p = p;
509 *lru_w = w;
510 *lru_l = w_l;
511 }
512 }
513 w_l = w;
514 }
515 }
516
517 static int unuse_one_window(struct packed_git *current)
518 {
519 struct packed_git *p, *lru_p = NULL;
520 struct pack_window *lru_w = NULL, *lru_l = NULL;
521
522 if (current)
523 scan_windows(current, &lru_p, &lru_w, &lru_l);
524 for (p = packed_git; p; p = p->next)
525 scan_windows(p, &lru_p, &lru_w, &lru_l);
526 if (lru_p) {
527 munmap(lru_w->base, lru_w->len);
528 pack_mapped -= lru_w->len;
529 if (lru_l)
530 lru_l->next = lru_w->next;
531 else {
532 lru_p->windows = lru_w->next;
533 if (!lru_p->windows && lru_p != current) {
534 close(lru_p->pack_fd);
535 lru_p->pack_fd = -1;
536 }
537 }
538 free(lru_w);
539 pack_open_windows--;
540 return 1;
541 }
542 return 0;
543 }
544
545 void release_pack_memory(size_t need)
546 {
547 size_t cur = pack_mapped;
548 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
549 ; /* nothing */
550 }
551
552 void unuse_pack(struct pack_window **w_cursor)
553 {
554 struct pack_window *w = *w_cursor;
555 if (w) {
556 w->inuse_cnt--;
557 *w_cursor = NULL;
558 }
559 }
560
561 /*
562 * Do not call this directly as this leaks p->pack_fd on error return;
563 * call open_packed_git() instead.
564 */
565 static int open_packed_git_1(struct packed_git *p)
566 {
567 struct stat st;
568 struct pack_header hdr;
569 unsigned char sha1[20];
570 unsigned char *idx_sha1;
571 long fd_flag;
572
573 p->pack_fd = open(p->pack_name, O_RDONLY);
574 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
575 return -1;
576
577 /* If we created the struct before we had the pack we lack size. */
578 if (!p->pack_size) {
579 if (!S_ISREG(st.st_mode))
580 return error("packfile %s not a regular file", p->pack_name);
581 p->pack_size = st.st_size;
582 } else if (p->pack_size != st.st_size)
583 return error("packfile %s size changed", p->pack_name);
584
585 /* We leave these file descriptors open with sliding mmap;
586 * there is no point keeping them open across exec(), though.
587 */
588 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
589 if (fd_flag < 0)
590 return error("cannot determine file descriptor flags");
591 fd_flag |= FD_CLOEXEC;
592 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
593 return error("cannot set FD_CLOEXEC");
594
595 /* Verify we recognize this pack file format. */
596 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
597 return error("file %s is far too short to be a packfile", p->pack_name);
598 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
599 return error("file %s is not a GIT packfile", p->pack_name);
600 if (!pack_version_ok(hdr.hdr_version))
601 return error("packfile %s is version %u and not supported"
602 " (try upgrading GIT to a newer version)",
603 p->pack_name, ntohl(hdr.hdr_version));
604
605 /* Verify the pack matches its index. */
606 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
607 return error("packfile %s claims to have %u objects"
608 " while index size indicates %u objects",
609 p->pack_name, ntohl(hdr.hdr_entries),
610 num_packed_objects(p));
611 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
612 return error("end of packfile %s is unavailable", p->pack_name);
613 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
614 return error("packfile %s signature is unavailable", p->pack_name);
615 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
616 if (hashcmp(sha1, idx_sha1))
617 return error("packfile %s does not match index", p->pack_name);
618 return 0;
619 }
620
621 static int open_packed_git(struct packed_git *p)
622 {
623 if (!open_packed_git_1(p))
624 return 0;
625 if (p->pack_fd != -1) {
626 close(p->pack_fd);
627 p->pack_fd = -1;
628 }
629 return -1;
630 }
631
632 static int in_window(struct pack_window *win, off_t offset)
633 {
634 /* We must promise at least 20 bytes (one hash) after the
635 * offset is available from this window, otherwise the offset
636 * is not actually in this window and a different window (which
637 * has that one hash excess) must be used. This is to support
638 * the object header and delta base parsing routines below.
639 */
640 off_t win_off = win->offset;
641 return win_off <= offset
642 && (offset + 20) <= (win_off + win->len);
643 }
644
645 unsigned char* use_pack(struct packed_git *p,
646 struct pack_window **w_cursor,
647 off_t offset,
648 unsigned int *left)
649 {
650 struct pack_window *win = *w_cursor;
651
652 if (p->pack_fd == -1 && open_packed_git(p))
653 die("packfile %s cannot be accessed", p->pack_name);
654
655 /* Since packfiles end in a hash of their content and its
656 * pointless to ask for an offset into the middle of that
657 * hash, and the in_window function above wouldn't match
658 * don't allow an offset too close to the end of the file.
659 */
660 if (offset > (p->pack_size - 20))
661 die("offset beyond end of packfile (truncated pack?)");
662
663 if (!win || !in_window(win, offset)) {
664 if (win)
665 win->inuse_cnt--;
666 for (win = p->windows; win; win = win->next) {
667 if (in_window(win, offset))
668 break;
669 }
670 if (!win) {
671 size_t window_align = packed_git_window_size / 2;
672 win = xcalloc(1, sizeof(*win));
673 win->offset = (offset / window_align) * window_align;
674 win->len = p->pack_size - win->offset;
675 if (win->len > packed_git_window_size)
676 win->len = packed_git_window_size;
677 pack_mapped += win->len;
678 while (packed_git_limit < pack_mapped
679 && unuse_one_window(p))
680 ; /* nothing */
681 win->base = xmmap(NULL, win->len,
682 PROT_READ, MAP_PRIVATE,
683 p->pack_fd, win->offset);
684 if (win->base == MAP_FAILED)
685 die("packfile %s cannot be mapped: %s",
686 p->pack_name,
687 strerror(errno));
688 pack_mmap_calls++;
689 pack_open_windows++;
690 if (pack_mapped > peak_pack_mapped)
691 peak_pack_mapped = pack_mapped;
692 if (pack_open_windows > peak_pack_open_windows)
693 peak_pack_open_windows = pack_open_windows;
694 win->next = p->windows;
695 p->windows = win;
696 }
697 }
698 if (win != *w_cursor) {
699 win->last_used = pack_used_ctr++;
700 win->inuse_cnt++;
701 *w_cursor = win;
702 }
703 offset -= win->offset;
704 if (left)
705 *left = win->len - offset;
706 return win->base + offset;
707 }
708
709 struct packed_git *add_packed_git(char *path, int path_len, int local)
710 {
711 struct stat st;
712 struct packed_git *p;
713 unsigned long idx_size;
714 void *idx_map;
715 unsigned char sha1[20];
716
717 if (check_packed_git_idx(path, &idx_size, &idx_map))
718 return NULL;
719
720 /* do we have a corresponding .pack file? */
721 strcpy(path + path_len - 4, ".pack");
722 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
723 munmap(idx_map, idx_size);
724 return NULL;
725 }
726 /* ok, it looks sane as far as we can check without
727 * actually mapping the pack file.
728 */
729 p = xmalloc(sizeof(*p) + path_len + 2);
730 strcpy(p->pack_name, path);
731 p->index_size = idx_size;
732 p->pack_size = st.st_size;
733 p->index_base = idx_map;
734 p->next = NULL;
735 p->windows = NULL;
736 p->pack_fd = -1;
737 p->pack_local = local;
738 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
739 hashcpy(p->sha1, sha1);
740 return p;
741 }
742
743 struct packed_git *parse_pack_index(unsigned char *sha1)
744 {
745 char *path = sha1_pack_index_name(sha1);
746 return parse_pack_index_file(sha1, path);
747 }
748
749 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
750 {
751 struct packed_git *p;
752 unsigned long idx_size;
753 void *idx_map;
754 char *path;
755
756 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
757 return NULL;
758
759 path = sha1_pack_name(sha1);
760
761 p = xmalloc(sizeof(*p) + strlen(path) + 2);
762 strcpy(p->pack_name, path);
763 p->index_size = idx_size;
764 p->pack_size = 0;
765 p->index_base = idx_map;
766 p->next = NULL;
767 p->windows = NULL;
768 p->pack_fd = -1;
769 hashcpy(p->sha1, sha1);
770 return p;
771 }
772
773 void install_packed_git(struct packed_git *pack)
774 {
775 pack->next = packed_git;
776 packed_git = pack;
777 }
778
779 static void prepare_packed_git_one(char *objdir, int local)
780 {
781 char path[PATH_MAX];
782 int len;
783 DIR *dir;
784 struct dirent *de;
785
786 sprintf(path, "%s/pack", objdir);
787 len = strlen(path);
788 dir = opendir(path);
789 if (!dir) {
790 if (errno != ENOENT)
791 error("unable to open object pack directory: %s: %s",
792 path, strerror(errno));
793 return;
794 }
795 path[len++] = '/';
796 while ((de = readdir(dir)) != NULL) {
797 int namelen = strlen(de->d_name);
798 struct packed_git *p;
799
800 if (!has_extension(de->d_name, ".idx"))
801 continue;
802
803 /* Don't reopen a pack we already have. */
804 strcpy(path + len, de->d_name);
805 for (p = packed_git; p; p = p->next) {
806 if (!memcmp(path, p->pack_name, len + namelen - 4))
807 break;
808 }
809 if (p)
810 continue;
811 /* See if it really is a valid .idx file with corresponding
812 * .pack file that we can map.
813 */
814 p = add_packed_git(path, len + namelen, local);
815 if (!p)
816 continue;
817 install_packed_git(p);
818 }
819 closedir(dir);
820 }
821
822 static int prepare_packed_git_run_once = 0;
823 void prepare_packed_git(void)
824 {
825 struct alternate_object_database *alt;
826
827 if (prepare_packed_git_run_once)
828 return;
829 prepare_packed_git_one(get_object_directory(), 1);
830 prepare_alt_odb();
831 for (alt = alt_odb_list; alt; alt = alt->next) {
832 alt->name[-1] = 0;
833 prepare_packed_git_one(alt->base, 0);
834 alt->name[-1] = '/';
835 }
836 prepare_packed_git_run_once = 1;
837 }
838
839 void reprepare_packed_git(void)
840 {
841 prepare_packed_git_run_once = 0;
842 prepare_packed_git();
843 }
844
845 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
846 {
847 unsigned char real_sha1[20];
848 hash_sha1_file(map, size, type, real_sha1);
849 return hashcmp(sha1, real_sha1) ? -1 : 0;
850 }
851
852 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
853 {
854 struct stat st;
855 void *map;
856 int fd;
857 char *filename = find_sha1_file(sha1, &st);
858
859 if (!filename) {
860 return NULL;
861 }
862
863 fd = open(filename, O_RDONLY | sha1_file_open_flag);
864 if (fd < 0) {
865 /* See if it works without O_NOATIME */
866 switch (sha1_file_open_flag) {
867 default:
868 fd = open(filename, O_RDONLY);
869 if (fd >= 0)
870 break;
871 /* Fallthrough */
872 case 0:
873 return NULL;
874 }
875
876 /* If it failed once, it will probably fail again.
877 * Stop using O_NOATIME
878 */
879 sha1_file_open_flag = 0;
880 }
881 map = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
882 close(fd);
883 *size = st.st_size;
884 return map;
885 }
886
887 int legacy_loose_object(unsigned char *map)
888 {
889 unsigned int word;
890
891 /*
892 * Is it a zlib-compressed buffer? If so, the first byte
893 * must be 0x78 (15-bit window size, deflated), and the
894 * first 16-bit word is evenly divisible by 31
895 */
896 word = (map[0] << 8) + map[1];
897 if (map[0] == 0x78 && !(word % 31))
898 return 1;
899 else
900 return 0;
901 }
902
903 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
904 {
905 unsigned shift;
906 unsigned char c;
907 unsigned long size;
908 unsigned long used = 0;
909
910 c = buf[used++];
911 *type = (c >> 4) & 7;
912 size = c & 15;
913 shift = 4;
914 while (c & 0x80) {
915 if (len <= used)
916 return 0;
917 if (sizeof(long) * 8 <= shift)
918 return 0;
919 c = buf[used++];
920 size += (c & 0x7f) << shift;
921 shift += 7;
922 }
923 *sizep = size;
924 return used;
925 }
926
927 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
928 {
929 unsigned long size, used;
930 static const char valid_loose_object_type[8] = {
931 0, /* OBJ_EXT */
932 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
933 0, /* "delta" and others are invalid in a loose object */
934 };
935 enum object_type type;
936
937 /* Get the data stream */
938 memset(stream, 0, sizeof(*stream));
939 stream->next_in = map;
940 stream->avail_in = mapsize;
941 stream->next_out = buffer;
942 stream->avail_out = bufsiz;
943
944 if (legacy_loose_object(map)) {
945 inflateInit(stream);
946 return inflate(stream, 0);
947 }
948
949 used = unpack_object_header_gently(map, mapsize, &type, &size);
950 if (!used || !valid_loose_object_type[type])
951 return -1;
952 map += used;
953 mapsize -= used;
954
955 /* Set up the stream for the rest.. */
956 stream->next_in = map;
957 stream->avail_in = mapsize;
958 inflateInit(stream);
959
960 /* And generate the fake traditional header */
961 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
962 typename(type), size);
963 return 0;
964 }
965
966 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
967 {
968 int bytes = strlen(buffer) + 1;
969 unsigned char *buf = xmalloc(1+size);
970 unsigned long n;
971
972 n = stream->total_out - bytes;
973 if (n > size)
974 n = size;
975 memcpy(buf, (char *) buffer + bytes, n);
976 bytes = n;
977 if (bytes < size) {
978 stream->next_out = buf + bytes;
979 stream->avail_out = size - bytes;
980 while (inflate(stream, Z_FINISH) == Z_OK)
981 /* nothing */;
982 }
983 buf[size] = 0;
984 inflateEnd(stream);
985 return buf;
986 }
987
988 /*
989 * We used to just use "sscanf()", but that's actually way
990 * too permissive for what we want to check. So do an anal
991 * object header parse by hand.
992 */
993 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
994 {
995 char type[10];
996 int i;
997 unsigned long size;
998
999 /*
1000 * The type can be at most ten bytes (including the
1001 * terminating '\0' that we add), and is followed by
1002 * a space.
1003 */
1004 i = 0;
1005 for (;;) {
1006 char c = *hdr++;
1007 if (c == ' ')
1008 break;
1009 type[i++] = c;
1010 if (i >= sizeof(type))
1011 return -1;
1012 }
1013 type[i] = 0;
1014
1015 /*
1016 * The length must follow immediately, and be in canonical
1017 * decimal format (ie "010" is not valid).
1018 */
1019 size = *hdr++ - '0';
1020 if (size > 9)
1021 return -1;
1022 if (size) {
1023 for (;;) {
1024 unsigned long c = *hdr - '0';
1025 if (c > 9)
1026 break;
1027 hdr++;
1028 size = size * 10 + c;
1029 }
1030 }
1031 *sizep = size;
1032
1033 /*
1034 * The length must be followed by a zero byte
1035 */
1036 return *hdr ? -1 : type_from_string(type);
1037 }
1038
1039 void * unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size)
1040 {
1041 int ret;
1042 z_stream stream;
1043 char hdr[8192];
1044
1045 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1046 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1047 return NULL;
1048
1049 return unpack_sha1_rest(&stream, hdr, *size);
1050 }
1051
1052 static off_t get_delta_base(struct packed_git *p,
1053 struct pack_window **w_curs,
1054 off_t *curpos,
1055 enum object_type type,
1056 off_t delta_obj_offset)
1057 {
1058 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1059 off_t base_offset;
1060
1061 /* use_pack() assured us we have [base_info, base_info + 20)
1062 * as a range that we can look at without walking off the
1063 * end of the mapped window. Its actually the hash size
1064 * that is assured. An OFS_DELTA longer than the hash size
1065 * is stupid, as then a REF_DELTA would be smaller to store.
1066 */
1067 if (type == OBJ_OFS_DELTA) {
1068 unsigned used = 0;
1069 unsigned char c = base_info[used++];
1070 base_offset = c & 127;
1071 while (c & 128) {
1072 base_offset += 1;
1073 if (!base_offset || base_offset & ~(~0UL >> 7))
1074 die("offset value overflow for delta base object");
1075 c = base_info[used++];
1076 base_offset = (base_offset << 7) + (c & 127);
1077 }
1078 base_offset = delta_obj_offset - base_offset;
1079 if (base_offset >= delta_obj_offset)
1080 die("delta base offset out of bound");
1081 *curpos += used;
1082 } else if (type == OBJ_REF_DELTA) {
1083 /* The base entry _must_ be in the same pack */
1084 base_offset = find_pack_entry_one(base_info, p);
1085 if (!base_offset)
1086 die("failed to find delta-pack base object %s",
1087 sha1_to_hex(base_info));
1088 *curpos += 20;
1089 } else
1090 die("I am totally screwed");
1091 return base_offset;
1092 }
1093
1094 /* forward declaration for a mutually recursive function */
1095 static int packed_object_info(struct packed_git *p, off_t offset,
1096 unsigned long *sizep);
1097
1098 static int packed_delta_info(struct packed_git *p,
1099 struct pack_window **w_curs,
1100 off_t curpos,
1101 enum object_type type,
1102 off_t obj_offset,
1103 unsigned long *sizep)
1104 {
1105 off_t base_offset;
1106
1107 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1108 type = packed_object_info(p, base_offset, NULL);
1109
1110 /* We choose to only get the type of the base object and
1111 * ignore potentially corrupt pack file that expects the delta
1112 * based on a base with a wrong size. This saves tons of
1113 * inflate() calls.
1114 */
1115 if (sizep) {
1116 const unsigned char *data;
1117 unsigned char delta_head[20], *in;
1118 z_stream stream;
1119 int st;
1120
1121 memset(&stream, 0, sizeof(stream));
1122 stream.next_out = delta_head;
1123 stream.avail_out = sizeof(delta_head);
1124
1125 inflateInit(&stream);
1126 do {
1127 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1128 stream.next_in = in;
1129 st = inflate(&stream, Z_FINISH);
1130 curpos += stream.next_in - in;
1131 } while ((st == Z_OK || st == Z_BUF_ERROR)
1132 && stream.total_out < sizeof(delta_head));
1133 inflateEnd(&stream);
1134 if ((st != Z_STREAM_END) &&
1135 stream.total_out != sizeof(delta_head))
1136 die("delta data unpack-initial failed");
1137
1138 /* Examine the initial part of the delta to figure out
1139 * the result size.
1140 */
1141 data = delta_head;
1142
1143 /* ignore base size */
1144 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1145
1146 /* Read the result size */
1147 *sizep = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1148 }
1149
1150 return type;
1151 }
1152
1153 static int unpack_object_header(struct packed_git *p,
1154 struct pack_window **w_curs,
1155 off_t *curpos,
1156 unsigned long *sizep)
1157 {
1158 unsigned char *base;
1159 unsigned int left;
1160 unsigned long used;
1161 enum object_type type;
1162
1163 /* use_pack() assures us we have [base, base + 20) available
1164 * as a range that we can look at at. (Its actually the hash
1165 * size that is assured.) With our object header encoding
1166 * the maximum deflated object size is 2^137, which is just
1167 * insane, so we know won't exceed what we have been given.
1168 */
1169 base = use_pack(p, w_curs, *curpos, &left);
1170 used = unpack_object_header_gently(base, left, &type, sizep);
1171 if (!used)
1172 die("object offset outside of pack file");
1173 *curpos += used;
1174
1175 return type;
1176 }
1177
1178 const char *packed_object_info_detail(struct packed_git *p,
1179 off_t obj_offset,
1180 unsigned long *size,
1181 unsigned long *store_size,
1182 unsigned int *delta_chain_length,
1183 unsigned char *base_sha1)
1184 {
1185 struct pack_window *w_curs = NULL;
1186 off_t curpos;
1187 unsigned long dummy;
1188 unsigned char *next_sha1;
1189 enum object_type type;
1190
1191 *delta_chain_length = 0;
1192 curpos = obj_offset;
1193 type = unpack_object_header(p, &w_curs, &curpos, size);
1194
1195 for (;;) {
1196 switch (type) {
1197 default:
1198 die("pack %s contains unknown object type %d",
1199 p->pack_name, type);
1200 case OBJ_COMMIT:
1201 case OBJ_TREE:
1202 case OBJ_BLOB:
1203 case OBJ_TAG:
1204 *store_size = 0; /* notyet */
1205 unuse_pack(&w_curs);
1206 return typename(type);
1207 case OBJ_OFS_DELTA:
1208 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1209 if (*delta_chain_length == 0) {
1210 /* TODO: find base_sha1 as pointed by curpos */
1211 hashclr(base_sha1);
1212 }
1213 break;
1214 case OBJ_REF_DELTA:
1215 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1216 if (*delta_chain_length == 0)
1217 hashcpy(base_sha1, next_sha1);
1218 obj_offset = find_pack_entry_one(next_sha1, p);
1219 break;
1220 }
1221 (*delta_chain_length)++;
1222 curpos = obj_offset;
1223 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1224 }
1225 }
1226
1227 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1228 unsigned long *sizep)
1229 {
1230 struct pack_window *w_curs = NULL;
1231 unsigned long size;
1232 off_t curpos = obj_offset;
1233 enum object_type type;
1234
1235 type = unpack_object_header(p, &w_curs, &curpos, &size);
1236
1237 switch (type) {
1238 case OBJ_OFS_DELTA:
1239 case OBJ_REF_DELTA:
1240 type = packed_delta_info(p, &w_curs, curpos,
1241 type, obj_offset, sizep);
1242 break;
1243 case OBJ_COMMIT:
1244 case OBJ_TREE:
1245 case OBJ_BLOB:
1246 case OBJ_TAG:
1247 if (sizep)
1248 *sizep = size;
1249 break;
1250 default:
1251 die("pack %s contains unknown object type %d",
1252 p->pack_name, type);
1253 }
1254 unuse_pack(&w_curs);
1255 return type;
1256 }
1257
1258 static void *unpack_compressed_entry(struct packed_git *p,
1259 struct pack_window **w_curs,
1260 off_t curpos,
1261 unsigned long size)
1262 {
1263 int st;
1264 z_stream stream;
1265 unsigned char *buffer, *in;
1266
1267 buffer = xmalloc(size + 1);
1268 buffer[size] = 0;
1269 memset(&stream, 0, sizeof(stream));
1270 stream.next_out = buffer;
1271 stream.avail_out = size;
1272
1273 inflateInit(&stream);
1274 do {
1275 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1276 stream.next_in = in;
1277 st = inflate(&stream, Z_FINISH);
1278 curpos += stream.next_in - in;
1279 } while (st == Z_OK || st == Z_BUF_ERROR);
1280 inflateEnd(&stream);
1281 if ((st != Z_STREAM_END) || stream.total_out != size) {
1282 free(buffer);
1283 return NULL;
1284 }
1285
1286 return buffer;
1287 }
1288
1289 static void *unpack_delta_entry(struct packed_git *p,
1290 struct pack_window **w_curs,
1291 off_t curpos,
1292 unsigned long delta_size,
1293 off_t obj_offset,
1294 enum object_type *type,
1295 unsigned long *sizep)
1296 {
1297 void *delta_data, *result, *base;
1298 unsigned long base_size;
1299 off_t base_offset;
1300
1301 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1302 base = unpack_entry(p, base_offset, type, &base_size);
1303 if (!base)
1304 die("failed to read delta base object"
1305 " at %"PRIuMAX" from %s",
1306 (uintmax_t)base_offset, p->pack_name);
1307
1308 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1309 result = patch_delta(base, base_size,
1310 delta_data, delta_size,
1311 sizep);
1312 if (!result)
1313 die("failed to apply delta");
1314 free(delta_data);
1315 free(base);
1316 return result;
1317 }
1318
1319 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1320 enum object_type *type, unsigned long *sizep)
1321 {
1322 struct pack_window *w_curs = NULL;
1323 off_t curpos = obj_offset;
1324 void *data;
1325
1326 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1327 switch (*type) {
1328 case OBJ_OFS_DELTA:
1329 case OBJ_REF_DELTA:
1330 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1331 obj_offset, type, sizep);
1332 break;
1333 case OBJ_COMMIT:
1334 case OBJ_TREE:
1335 case OBJ_BLOB:
1336 case OBJ_TAG:
1337 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1338 break;
1339 default:
1340 die("unknown object type %i in %s", *type, p->pack_name);
1341 }
1342 unuse_pack(&w_curs);
1343 return data;
1344 }
1345
1346 uint32_t num_packed_objects(const struct packed_git *p)
1347 {
1348 /* See check_packed_git_idx() */
1349 return (p->index_size - 20 - 20 - 4*256) / 24;
1350 }
1351
1352 int nth_packed_object_sha1(const struct packed_git *p, uint32_t n,
1353 unsigned char* sha1)
1354 {
1355 void *index = p->index_base + 256;
1356 if (num_packed_objects(p) <= n)
1357 return -1;
1358 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1359 return 0;
1360 }
1361
1362 off_t find_pack_entry_one(const unsigned char *sha1,
1363 struct packed_git *p)
1364 {
1365 uint32_t *level1_ofs = p->index_base;
1366 int hi = ntohl(level1_ofs[*sha1]);
1367 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1368 void *index = p->index_base + 256;
1369
1370 do {
1371 int mi = (lo + hi) / 2;
1372 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1373 if (!cmp)
1374 return ntohl(*((uint32_t *)((char *)index + (24 * mi))));
1375 if (cmp > 0)
1376 hi = mi;
1377 else
1378 lo = mi+1;
1379 } while (lo < hi);
1380 return 0;
1381 }
1382
1383 static int matches_pack_name(struct packed_git *p, const char *ig)
1384 {
1385 const char *last_c, *c;
1386
1387 if (!strcmp(p->pack_name, ig))
1388 return 0;
1389
1390 for (c = p->pack_name, last_c = c; *c;)
1391 if (*c == '/')
1392 last_c = ++c;
1393 else
1394 ++c;
1395 if (!strcmp(last_c, ig))
1396 return 0;
1397
1398 return 1;
1399 }
1400
1401 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1402 {
1403 struct packed_git *p;
1404 off_t offset;
1405
1406 prepare_packed_git();
1407
1408 for (p = packed_git; p; p = p->next) {
1409 if (ignore_packed) {
1410 const char **ig;
1411 for (ig = ignore_packed; *ig; ig++)
1412 if (!matches_pack_name(p, *ig))
1413 break;
1414 if (*ig)
1415 continue;
1416 }
1417 offset = find_pack_entry_one(sha1, p);
1418 if (offset) {
1419 /*
1420 * We are about to tell the caller where they can
1421 * locate the requested object. We better make
1422 * sure the packfile is still here and can be
1423 * accessed before supplying that answer, as
1424 * it may have been deleted since the index
1425 * was loaded!
1426 */
1427 if (p->pack_fd == -1 && open_packed_git(p)) {
1428 error("packfile %s cannot be accessed", p->pack_name);
1429 continue;
1430 }
1431 e->offset = offset;
1432 e->p = p;
1433 hashcpy(e->sha1, sha1);
1434 return 1;
1435 }
1436 }
1437 return 0;
1438 }
1439
1440 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1441 struct packed_git *packs)
1442 {
1443 struct packed_git *p;
1444
1445 for (p = packs; p; p = p->next) {
1446 if (find_pack_entry_one(sha1, p))
1447 return p;
1448 }
1449 return NULL;
1450
1451 }
1452
1453 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1454 {
1455 int status;
1456 unsigned long mapsize, size;
1457 void *map;
1458 z_stream stream;
1459 char hdr[32];
1460
1461 map = map_sha1_file(sha1, &mapsize);
1462 if (!map)
1463 return error("unable to find %s", sha1_to_hex(sha1));
1464 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1465 status = error("unable to unpack %s header",
1466 sha1_to_hex(sha1));
1467 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1468 status = error("unable to parse %s header", sha1_to_hex(sha1));
1469 else if (sizep)
1470 *sizep = size;
1471 inflateEnd(&stream);
1472 munmap(map, mapsize);
1473 return status;
1474 }
1475
1476 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1477 {
1478 struct pack_entry e;
1479
1480 if (!find_pack_entry(sha1, &e, NULL)) {
1481 reprepare_packed_git();
1482 if (!find_pack_entry(sha1, &e, NULL))
1483 return sha1_loose_object_info(sha1, sizep);
1484 }
1485 return packed_object_info(e.p, e.offset, sizep);
1486 }
1487
1488 static void *read_packed_sha1(const unsigned char *sha1,
1489 enum object_type *type, unsigned long *size)
1490 {
1491 struct pack_entry e;
1492
1493 if (!find_pack_entry(sha1, &e, NULL))
1494 return NULL;
1495 else
1496 return unpack_entry(e.p, e.offset, type, size);
1497 }
1498
1499 /*
1500 * This is meant to hold a *small* number of objects that you would
1501 * want read_sha1_file() to be able to return, but yet you do not want
1502 * to write them into the object store (e.g. a browse-only
1503 * application).
1504 */
1505 static struct cached_object {
1506 unsigned char sha1[20];
1507 enum object_type type;
1508 void *buf;
1509 unsigned long size;
1510 } *cached_objects;
1511 static int cached_object_nr, cached_object_alloc;
1512
1513 static struct cached_object *find_cached_object(const unsigned char *sha1)
1514 {
1515 int i;
1516 struct cached_object *co = cached_objects;
1517
1518 for (i = 0; i < cached_object_nr; i++, co++) {
1519 if (!hashcmp(co->sha1, sha1))
1520 return co;
1521 }
1522 return NULL;
1523 }
1524
1525 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1526 unsigned char *sha1)
1527 {
1528 struct cached_object *co;
1529
1530 hash_sha1_file(buf, len, typename(type), sha1);
1531 if (has_sha1_file(sha1) || find_cached_object(sha1))
1532 return 0;
1533 if (cached_object_alloc <= cached_object_nr) {
1534 cached_object_alloc = alloc_nr(cached_object_alloc);
1535 cached_objects = xrealloc(cached_objects,
1536 sizeof(*cached_objects) *
1537 cached_object_alloc);
1538 }
1539 co = &cached_objects[cached_object_nr++];
1540 co->size = len;
1541 co->type = type;
1542 co->buf = xmalloc(len);
1543 memcpy(co->buf, buf, len);
1544 hashcpy(co->sha1, sha1);
1545 return 0;
1546 }
1547
1548 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1549 unsigned long *size)
1550 {
1551 unsigned long mapsize;
1552 void *map, *buf;
1553 struct cached_object *co;
1554
1555 co = find_cached_object(sha1);
1556 if (co) {
1557 buf = xmalloc(co->size + 1);
1558 memcpy(buf, co->buf, co->size);
1559 ((char*)buf)[co->size] = 0;
1560 *type = co->type;
1561 *size = co->size;
1562 return buf;
1563 }
1564
1565 buf = read_packed_sha1(sha1, type, size);
1566 if (buf)
1567 return buf;
1568 map = map_sha1_file(sha1, &mapsize);
1569 if (map) {
1570 buf = unpack_sha1_file(map, mapsize, type, size);
1571 munmap(map, mapsize);
1572 return buf;
1573 }
1574 reprepare_packed_git();
1575 return read_packed_sha1(sha1, type, size);
1576 }
1577
1578 void *read_object_with_reference(const unsigned char *sha1,
1579 const char *required_type_name,
1580 unsigned long *size,
1581 unsigned char *actual_sha1_return)
1582 {
1583 enum object_type type, required_type;
1584 void *buffer;
1585 unsigned long isize;
1586 unsigned char actual_sha1[20];
1587
1588 required_type = type_from_string(required_type_name);
1589 hashcpy(actual_sha1, sha1);
1590 while (1) {
1591 int ref_length = -1;
1592 const char *ref_type = NULL;
1593
1594 buffer = read_sha1_file(actual_sha1, &type, &isize);
1595 if (!buffer)
1596 return NULL;
1597 if (type == required_type) {
1598 *size = isize;
1599 if (actual_sha1_return)
1600 hashcpy(actual_sha1_return, actual_sha1);
1601 return buffer;
1602 }
1603 /* Handle references */
1604 else if (type == OBJ_COMMIT)
1605 ref_type = "tree ";
1606 else if (type == OBJ_TAG)
1607 ref_type = "object ";
1608 else {
1609 free(buffer);
1610 return NULL;
1611 }
1612 ref_length = strlen(ref_type);
1613
1614 if (memcmp(buffer, ref_type, ref_length) ||
1615 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1616 free(buffer);
1617 return NULL;
1618 }
1619 free(buffer);
1620 /* Now we have the ID of the referred-to object in
1621 * actual_sha1. Check again. */
1622 }
1623 }
1624
1625 static void write_sha1_file_prepare(void *buf, unsigned long len,
1626 const char *type, unsigned char *sha1,
1627 char *hdr, int *hdrlen)
1628 {
1629 SHA_CTX c;
1630
1631 /* Generate the header */
1632 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
1633
1634 /* Sha1.. */
1635 SHA1_Init(&c);
1636 SHA1_Update(&c, hdr, *hdrlen);
1637 SHA1_Update(&c, buf, len);
1638 SHA1_Final(sha1, &c);
1639 }
1640
1641 /*
1642 * Link the tempfile to the final place, possibly creating the
1643 * last directory level as you do so.
1644 *
1645 * Returns the errno on failure, 0 on success.
1646 */
1647 static int link_temp_to_file(const char *tmpfile, const char *filename)
1648 {
1649 int ret;
1650 char *dir;
1651
1652 if (!link(tmpfile, filename))
1653 return 0;
1654
1655 /*
1656 * Try to mkdir the last path component if that failed.
1657 *
1658 * Re-try the "link()" regardless of whether the mkdir
1659 * succeeds, since a race might mean that somebody
1660 * else succeeded.
1661 */
1662 ret = errno;
1663 dir = strrchr(filename, '/');
1664 if (dir) {
1665 *dir = 0;
1666 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1667 *dir = '/';
1668 return -2;
1669 }
1670 *dir = '/';
1671 if (!link(tmpfile, filename))
1672 return 0;
1673 ret = errno;
1674 }
1675 return ret;
1676 }
1677
1678 /*
1679 * Move the just written object into its final resting place
1680 */
1681 int move_temp_to_file(const char *tmpfile, const char *filename)
1682 {
1683 int ret = link_temp_to_file(tmpfile, filename);
1684
1685 /*
1686 * Coda hack - coda doesn't like cross-directory links,
1687 * so we fall back to a rename, which will mean that it
1688 * won't be able to check collisions, but that's not a
1689 * big deal.
1690 *
1691 * The same holds for FAT formatted media.
1692 *
1693 * When this succeeds, we just return 0. We have nothing
1694 * left to unlink.
1695 */
1696 if (ret && ret != EEXIST) {
1697 if (!rename(tmpfile, filename))
1698 return 0;
1699 ret = errno;
1700 }
1701 unlink(tmpfile);
1702 if (ret) {
1703 if (ret != EEXIST) {
1704 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1705 }
1706 /* FIXME!!! Collision check here ? */
1707 }
1708
1709 return 0;
1710 }
1711
1712 static int write_buffer(int fd, const void *buf, size_t len)
1713 {
1714 if (write_in_full(fd, buf, len) < 0)
1715 return error("file write error (%s)", strerror(errno));
1716 return 0;
1717 }
1718
1719 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1720 {
1721 int hdr_len;
1722 unsigned char c;
1723
1724 c = (type << 4) | (len & 15);
1725 len >>= 4;
1726 hdr_len = 1;
1727 while (len) {
1728 *hdr++ = c | 0x80;
1729 hdr_len++;
1730 c = (len & 0x7f);
1731 len >>= 7;
1732 }
1733 *hdr = c;
1734 return hdr_len;
1735 }
1736
1737 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1738 {
1739 int obj_type, hdrlen;
1740
1741 if (use_legacy_headers) {
1742 while (deflate(stream, 0) == Z_OK)
1743 /* nothing */;
1744 return;
1745 }
1746 obj_type = type_from_string(type);
1747 hdrlen = write_binary_header(stream->next_out, obj_type, len);
1748 stream->total_out = hdrlen;
1749 stream->next_out += hdrlen;
1750 stream->avail_out -= hdrlen;
1751 }
1752
1753 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1754 unsigned char *sha1)
1755 {
1756 char hdr[32];
1757 int hdrlen;
1758 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1759 return 0;
1760 }
1761
1762 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1763 {
1764 int size;
1765 unsigned char *compressed;
1766 z_stream stream;
1767 unsigned char sha1[20];
1768 char *filename;
1769 static char tmpfile[PATH_MAX];
1770 char hdr[32];
1771 int fd, hdrlen;
1772
1773 /* Normally if we have it in the pack then we do not bother writing
1774 * it out into .git/objects/??/?{38} file.
1775 */
1776 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1777 filename = sha1_file_name(sha1);
1778 if (returnsha1)
1779 hashcpy(returnsha1, sha1);
1780 if (has_sha1_file(sha1))
1781 return 0;
1782 fd = open(filename, O_RDONLY);
1783 if (fd >= 0) {
1784 /*
1785 * FIXME!!! We might do collision checking here, but we'd
1786 * need to uncompress the old file and check it. Later.
1787 */
1788 close(fd);
1789 return 0;
1790 }
1791
1792 if (errno != ENOENT) {
1793 return error("sha1 file %s: %s\n", filename, strerror(errno));
1794 }
1795
1796 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1797
1798 fd = mkstemp(tmpfile);
1799 if (fd < 0) {
1800 if (errno == EPERM)
1801 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1802 else
1803 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1804 }
1805
1806 /* Set it up */
1807 memset(&stream, 0, sizeof(stream));
1808 deflateInit(&stream, zlib_compression_level);
1809 size = 8 + deflateBound(&stream, len+hdrlen);
1810 compressed = xmalloc(size);
1811
1812 /* Compress it */
1813 stream.next_out = compressed;
1814 stream.avail_out = size;
1815
1816 /* First header.. */
1817 stream.next_in = (unsigned char *)hdr;
1818 stream.avail_in = hdrlen;
1819 setup_object_header(&stream, type, len);
1820
1821 /* Then the data itself.. */
1822 stream.next_in = buf;
1823 stream.avail_in = len;
1824 while (deflate(&stream, Z_FINISH) == Z_OK)
1825 /* nothing */;
1826 deflateEnd(&stream);
1827 size = stream.total_out;
1828
1829 if (write_buffer(fd, compressed, size) < 0)
1830 die("unable to write sha1 file");
1831 fchmod(fd, 0444);
1832 close(fd);
1833 free(compressed);
1834
1835 return move_temp_to_file(tmpfile, filename);
1836 }
1837
1838 /*
1839 * We need to unpack and recompress the object for writing
1840 * it out to a different file.
1841 */
1842 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1843 {
1844 size_t size;
1845 z_stream stream;
1846 unsigned char *unpacked;
1847 unsigned long len;
1848 enum object_type type;
1849 char hdr[32];
1850 int hdrlen;
1851 void *buf;
1852
1853 /* need to unpack and recompress it by itself */
1854 unpacked = read_packed_sha1(sha1, &type, &len);
1855 if (!unpacked)
1856 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1857
1858 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1859
1860 /* Set it up */
1861 memset(&stream, 0, sizeof(stream));
1862 deflateInit(&stream, zlib_compression_level);
1863 size = deflateBound(&stream, len + hdrlen);
1864 buf = xmalloc(size);
1865
1866 /* Compress it */
1867 stream.next_out = buf;
1868 stream.avail_out = size;
1869
1870 /* First header.. */
1871 stream.next_in = (void *)hdr;
1872 stream.avail_in = hdrlen;
1873 while (deflate(&stream, 0) == Z_OK)
1874 /* nothing */;
1875
1876 /* Then the data itself.. */
1877 stream.next_in = unpacked;
1878 stream.avail_in = len;
1879 while (deflate(&stream, Z_FINISH) == Z_OK)
1880 /* nothing */;
1881 deflateEnd(&stream);
1882 free(unpacked);
1883
1884 *objsize = stream.total_out;
1885 return buf;
1886 }
1887
1888 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1889 {
1890 int retval;
1891 unsigned long objsize;
1892 void *buf = map_sha1_file(sha1, &objsize);
1893
1894 if (buf) {
1895 retval = write_buffer(fd, buf, objsize);
1896 munmap(buf, objsize);
1897 return retval;
1898 }
1899
1900 buf = repack_object(sha1, &objsize);
1901 retval = write_buffer(fd, buf, objsize);
1902 free(buf);
1903 return retval;
1904 }
1905
1906 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1907 size_t bufsize, size_t *bufposn)
1908 {
1909 char tmpfile[PATH_MAX];
1910 int local;
1911 z_stream stream;
1912 unsigned char real_sha1[20];
1913 unsigned char discard[4096];
1914 int ret;
1915 SHA_CTX c;
1916
1917 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1918
1919 local = mkstemp(tmpfile);
1920 if (local < 0) {
1921 if (errno == EPERM)
1922 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1923 else
1924 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1925 }
1926
1927 memset(&stream, 0, sizeof(stream));
1928
1929 inflateInit(&stream);
1930
1931 SHA1_Init(&c);
1932
1933 do {
1934 ssize_t size;
1935 if (*bufposn) {
1936 stream.avail_in = *bufposn;
1937 stream.next_in = (unsigned char *) buffer;
1938 do {
1939 stream.next_out = discard;
1940 stream.avail_out = sizeof(discard);
1941 ret = inflate(&stream, Z_SYNC_FLUSH);
1942 SHA1_Update(&c, discard, sizeof(discard) -
1943 stream.avail_out);
1944 } while (stream.avail_in && ret == Z_OK);
1945 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1946 die("unable to write sha1 file");
1947 memmove(buffer, buffer + *bufposn - stream.avail_in,
1948 stream.avail_in);
1949 *bufposn = stream.avail_in;
1950 if (ret != Z_OK)
1951 break;
1952 }
1953 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
1954 if (size <= 0) {
1955 close(local);
1956 unlink(tmpfile);
1957 if (!size)
1958 return error("Connection closed?");
1959 perror("Reading from connection");
1960 return -1;
1961 }
1962 *bufposn += size;
1963 } while (1);
1964 inflateEnd(&stream);
1965
1966 close(local);
1967 SHA1_Final(real_sha1, &c);
1968 if (ret != Z_STREAM_END) {
1969 unlink(tmpfile);
1970 return error("File %s corrupted", sha1_to_hex(sha1));
1971 }
1972 if (hashcmp(sha1, real_sha1)) {
1973 unlink(tmpfile);
1974 return error("File %s has bad hash", sha1_to_hex(sha1));
1975 }
1976
1977 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1978 }
1979
1980 int has_pack_index(const unsigned char *sha1)
1981 {
1982 struct stat st;
1983 if (stat(sha1_pack_index_name(sha1), &st))
1984 return 0;
1985 return 1;
1986 }
1987
1988 int has_pack_file(const unsigned char *sha1)
1989 {
1990 struct stat st;
1991 if (stat(sha1_pack_name(sha1), &st))
1992 return 0;
1993 return 1;
1994 }
1995
1996 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1997 {
1998 struct pack_entry e;
1999 return find_pack_entry(sha1, &e, ignore_packed);
2000 }
2001
2002 int has_sha1_file(const unsigned char *sha1)
2003 {
2004 struct stat st;
2005 struct pack_entry e;
2006
2007 if (find_pack_entry(sha1, &e, NULL))
2008 return 1;
2009 return find_sha1_file(sha1, &st) ? 1 : 0;
2010 }
2011
2012 /*
2013 * reads from fd as long as possible into a supplied buffer of size bytes.
2014 * If necessary the buffer's size is increased using realloc()
2015 *
2016 * returns 0 if anything went fine and -1 otherwise
2017 *
2018 * NOTE: both buf and size may change, but even when -1 is returned
2019 * you still have to free() it yourself.
2020 */
2021 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
2022 {
2023 char* buf = *return_buf;
2024 unsigned long size = *return_size;
2025 int iret;
2026 unsigned long off = 0;
2027
2028 do {
2029 iret = xread(fd, buf + off, size - off);
2030 if (iret > 0) {
2031 off += iret;
2032 if (off == size) {
2033 size *= 2;
2034 buf = xrealloc(buf, size);
2035 }
2036 }
2037 } while (iret > 0);
2038
2039 *return_buf = buf;
2040 *return_size = off;
2041
2042 if (iret < 0)
2043 return -1;
2044 return 0;
2045 }
2046
2047 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2048 {
2049 unsigned long size = 4096;
2050 char *buf = xmalloc(size);
2051 int ret;
2052
2053 if (read_pipe(fd, &buf, &size)) {
2054 free(buf);
2055 return -1;
2056 }
2057
2058 if (!type)
2059 type = blob_type;
2060 if (write_object)
2061 ret = write_sha1_file(buf, size, type, sha1);
2062 else
2063 ret = hash_sha1_file(buf, size, type, sha1);
2064 free(buf);
2065 return ret;
2066 }
2067
2068 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2069 enum object_type type, const char *path)
2070 {
2071 unsigned long size = st->st_size;
2072 void *buf = NULL;
2073 int ret, re_allocated = 0;
2074
2075 if (size)
2076 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2077 close(fd);
2078
2079 if (!type)
2080 type = OBJ_BLOB;
2081
2082 /*
2083 * Convert blobs to git internal format
2084 */
2085 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2086 unsigned long nsize = size;
2087 char *nbuf = buf;
2088 if (convert_to_git(path, &nbuf, &nsize)) {
2089 if (size)
2090 munmap(buf, size);
2091 size = nsize;
2092 buf = nbuf;
2093 re_allocated = 1;
2094 }
2095 }
2096
2097 if (write_object)
2098 ret = write_sha1_file(buf, size, typename(type), sha1);
2099 else
2100 ret = hash_sha1_file(buf, size, typename(type), sha1);
2101 if (re_allocated) {
2102 free(buf);
2103 return ret;
2104 }
2105 if (size)
2106 munmap(buf, size);
2107 return ret;
2108 }
2109
2110 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2111 {
2112 int fd;
2113 char *target;
2114
2115 switch (st->st_mode & S_IFMT) {
2116 case S_IFREG:
2117 fd = open(path, O_RDONLY);
2118 if (fd < 0)
2119 return error("open(\"%s\"): %s", path,
2120 strerror(errno));
2121 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2122 return error("%s: failed to insert into database",
2123 path);
2124 break;
2125 case S_IFLNK:
2126 target = xmalloc(st->st_size+1);
2127 if (readlink(path, target, st->st_size+1) != st->st_size) {
2128 char *errstr = strerror(errno);
2129 free(target);
2130 return error("readlink(\"%s\"): %s", path,
2131 errstr);
2132 }
2133 if (!write_object)
2134 hash_sha1_file(target, st->st_size, blob_type, sha1);
2135 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
2136 return error("%s: failed to insert into database",
2137 path);
2138 free(target);
2139 break;
2140 default:
2141 return error("%s: unsupported file type", path);
2142 }
2143 return 0;
2144 }
2145
2146 int read_pack_header(int fd, struct pack_header *header)
2147 {
2148 char *c = (char*)header;
2149 ssize_t remaining = sizeof(struct pack_header);
2150 do {
2151 ssize_t r = xread(fd, c, remaining);
2152 if (r <= 0)
2153 /* "eof before pack header was fully read" */
2154 return PH_ERROR_EOF;
2155 remaining -= r;
2156 c += r;
2157 } while (remaining > 0);
2158 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2159 /* "protocol error (pack signature mismatch detected)" */
2160 return PH_ERROR_PACK_SIGNATURE;
2161 if (!pack_version_ok(header->hdr_version))
2162 /* "protocol error (pack version unsupported)" */
2163 return PH_ERROR_PROTOCOL;
2164 return 0;
2165 }